* [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library
@ 2019-02-28 19:20 Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 1/6] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (14 more replies)
0 siblings, 15 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:20 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
The first patch in this series introduce new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-5% depending on the test-case) for ipsec library.
Konstantin Ananyev (6):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 406 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1016 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 71 ++-
10 files changed, 1170 insertions(+), 995 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 1/6] mbuf: new function to generate raw Tx offload value
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-02-28 19:20 ` Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 2/6] ipsec: add Tx offload template into SA Konstantin Ananyev
` (13 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:20 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
1 file changed, 64 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..b967ad17e 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,26 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/** enum for the tx_offload bit-fields lenghts and offsets. */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OL3_LEN_BITS = 9,
+ RTE_MBUF_OL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +660,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @return
+ * raw tx_offload value.
+ */
+static inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OL3_LEN_OFS |
+ ol2 << RTE_MBUF_OL2_LEN_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 2/6] ipsec: add Tx offload template into SA
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 1/6] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-02-28 19:20 ` Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 3/6] ipsec: change the order in filling crypto op Konstantin Ananyev
` (12 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:20 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 5f55c2a4e..31548e968 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -243,6 +246,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -285,6 +293,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -486,7 +499,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -513,8 +526,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1029,8 +1042,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 392e8fd7b..133a35d83 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -64,6 +64,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t proto; /* next proto */
uint8_t aad_len;
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 3/6] ipsec: change the order in filling crypto op
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 1/6] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 2/6] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-02-28 19:20 ` Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 4/6] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (11 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:20 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 31548e968..74ed3b41e 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -425,20 +425,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -603,11 +600,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -625,10 +624,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -636,9 +635,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -739,11 +735,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -765,10 +763,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -776,9 +774,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -926,11 +921,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -938,9 +935,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -952,9 +951,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 4/6] ipsec: change the way unprocessed mbufs are accounted
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (2 preceding siblings ...)
2019-02-28 19:20 ` [dpdk-dev] [PATCH 3/6] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-02-28 19:20 ` Konstantin Ananyev
2019-02-28 19:21 ` [dpdk-dev] [PATCH 5/6] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (10 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:20 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 74ed3b41e..e4e3d7fc3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -411,14 +411,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -603,7 +620,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -627,17 +644,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -738,7 +755,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -766,17 +783,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -924,7 +941,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -941,10 +958,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -953,7 +969,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1105,7 +1121,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1115,9 +1131,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1131,10 +1147,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1144,23 +1160,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1170,10 +1190,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1183,23 +1203,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1215,7 +1239,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1228,16 +1252,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1257,23 +1281,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1314,7 +1338,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1332,22 +1356,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1366,7 +1388,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1388,22 +1410,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 5/6] ipsec: move inbound and outbound code into different files
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (3 preceding siblings ...)
2019-02-28 19:20 ` [dpdk-dev] [PATCH 4/6] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-02-28 19:21 ` Konstantin Ananyev
2019-02-28 19:21 ` [dpdk-dev] [PATCH 6/6] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (9 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:21 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 408 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1011 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1086 insertions(+), 1000 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index b5f264831..5a7b11f56 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -145,4 +145,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..562185ebe
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,408 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ uint64_t *ivc, *ivp;
+ uint32_t clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ }
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..dce0fe051
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,535 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e4e3d7fc3..be6109768 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -411,569 +442,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- }
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1008,265 +476,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1303,142 +512,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1456,23 +529,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1498,10 +571,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1542,7 +615,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 133a35d83..93ff7bdf1 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -108,4 +108,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH 6/6] ipsec: reorder packet check for esp inbound
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (4 preceding siblings ...)
2019-02-28 19:21 ` [dpdk-dev] [PATCH 5/6] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-02-28 19:21 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (8 subsequent siblings)
14 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-02-28 19:21 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 30 ++++++++++++++----------------
1 file changed, 14 insertions(+), 16 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 562185ebe..d479af8ea 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -26,11 +26,6 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct rte_crypto_sym_op *sop;
struct aead_gcm_iv *gcm;
uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
/* fill sym op fields */
sop = cop->sym;
@@ -38,7 +33,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* AEAD (AES_GCM) case */
if (sa->aad_len != 0) {
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -53,7 +48,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* CRYPT+AUTH case */
} else {
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -101,7 +96,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -128,6 +123,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -170,21 +170,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (5 preceding siblings ...)
2019-02-28 19:21 ` [dpdk-dev] [PATCH 6/6] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (7 subsequent siblings)
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduce new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (7):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 509 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1016 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 71 ++-
10 files changed, 1273 insertions(+), 999 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduce new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (7):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 509 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1016 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 71 ++-
10 files changed, 1273 insertions(+), 999 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (6 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:53 ` Wiles, Keith
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
` (6 subsequent siblings)
14 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
1 file changed, 64 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..b967ad17e 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,26 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/** enum for the tx_offload bit-fields lenghts and offsets. */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OL3_LEN_BITS = 9,
+ RTE_MBUF_OL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +660,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @return
+ * raw tx_offload value.
+ */
+static inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OL3_LEN_OFS |
+ ol2 << RTE_MBUF_OL2_LEN_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:53 ` Wiles, Keith
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
1 file changed, 64 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..b967ad17e 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,26 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/** enum for the tx_offload bit-fields lenghts and offsets. */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OL3_LEN_BITS = 9,
+ RTE_MBUF_OL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +660,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @return
+ * raw tx_offload value.
+ */
+static inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OL3_LEN_OFS |
+ ol2 << RTE_MBUF_OL2_LEN_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (7 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
` (5 subsequent siblings)
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 5f55c2a4e..31548e968 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -243,6 +246,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -285,6 +293,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -486,7 +499,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -513,8 +526,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1029,8 +1042,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 392e8fd7b..133a35d83 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -64,6 +64,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t proto; /* next proto */
uint8_t aad_len;
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 5f55c2a4e..31548e968 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -243,6 +246,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -285,6 +293,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -486,7 +499,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -513,8 +526,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1029,8 +1042,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 392e8fd7b..133a35d83 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -64,6 +64,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t proto; /* next proto */
uint8_t aad_len;
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (8 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (4 subsequent siblings)
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 31548e968..74ed3b41e 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -425,20 +425,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -603,11 +600,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -625,10 +624,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -636,9 +635,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -739,11 +735,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -765,10 +763,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -776,9 +774,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -926,11 +921,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -938,9 +935,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -952,9 +951,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 31548e968..74ed3b41e 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -425,20 +425,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -603,11 +600,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -625,10 +624,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -636,9 +635,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -739,11 +735,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -765,10 +763,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -776,9 +774,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -926,11 +921,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -938,9 +935,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -952,9 +951,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (9 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (3 subsequent siblings)
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 74ed3b41e..e4e3d7fc3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -411,14 +411,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -603,7 +620,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -627,17 +644,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -738,7 +755,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -766,17 +783,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -924,7 +941,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -941,10 +958,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -953,7 +969,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1105,7 +1121,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1115,9 +1131,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1131,10 +1147,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1144,23 +1160,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1170,10 +1190,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1183,23 +1203,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1215,7 +1239,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1228,16 +1252,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1257,23 +1281,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1314,7 +1338,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1332,22 +1356,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1366,7 +1388,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1388,22 +1410,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 74ed3b41e..e4e3d7fc3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -411,14 +411,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -603,7 +620,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -627,17 +644,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -738,7 +755,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -766,17 +783,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -924,7 +941,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -941,10 +958,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -953,7 +969,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1105,7 +1121,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1115,9 +1131,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1131,10 +1147,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1144,23 +1160,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1170,10 +1190,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1183,23 +1203,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1215,7 +1239,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1228,16 +1252,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1257,23 +1281,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1314,7 +1338,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1332,22 +1356,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1366,7 +1388,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1388,22 +1410,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (10 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (2 subsequent siblings)
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 408 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1011 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1086 insertions(+), 1000 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index b5f264831..5a7b11f56 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -145,4 +145,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..562185ebe
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,408 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ uint64_t *ivc, *ivp;
+ uint32_t clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ }
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..dce0fe051
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,535 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e4e3d7fc3..be6109768 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -411,569 +442,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- }
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1008,265 +476,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1303,142 +512,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1456,23 +529,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1498,10 +571,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1542,7 +615,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 133a35d83..93ff7bdf1 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -108,4 +108,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 408 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1011 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1086 insertions(+), 1000 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index b5f264831..5a7b11f56 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -145,4 +145,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..562185ebe
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,408 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ uint64_t *ivc, *ivp;
+ uint32_t clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ }
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..dce0fe051
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,535 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e4e3d7fc3..be6109768 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -411,569 +442,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- }
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1008,265 +476,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1303,142 +512,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1456,23 +529,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1498,10 +571,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1542,7 +615,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 133a35d83..93ff7bdf1 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -108,4 +108,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (11 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 30 ++++++++++++++----------------
1 file changed, 14 insertions(+), 16 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 562185ebe..d479af8ea 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -26,11 +26,6 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct rte_crypto_sym_op *sop;
struct aead_gcm_iv *gcm;
uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
/* fill sym op fields */
sop = cop->sym;
@@ -38,7 +33,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* AEAD (AES_GCM) case */
if (sa->aad_len != 0) {
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -53,7 +48,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* CRYPT+AUTH case */
} else {
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -101,7 +96,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -128,6 +123,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -170,21 +170,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 30 ++++++++++++++----------------
1 file changed, 14 insertions(+), 16 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 562185ebe..d479af8ea 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -26,11 +26,6 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct rte_crypto_sym_op *sop;
struct aead_gcm_iv *gcm;
uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
/* fill sym op fields */
sop = cop->sym;
@@ -38,7 +33,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* AEAD (AES_GCM) case */
if (sa->aad_len != 0) {
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -53,7 +48,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* CRYPT+AUTH case */
} else {
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -101,7 +96,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -128,6 +123,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -170,21 +170,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process for esp inbound
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (12 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
14 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index d479af8ea..86a95e2f4 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline int32_t
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -188,111 +191,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
+
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -305,11 +436,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -320,13 +455,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -334,16 +469,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -362,6 +488,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -369,38 +505,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process for esp inbound
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-20 17:24 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 17:24 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index d479af8ea..86a95e2f4 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline int32_t
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -188,111 +191,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
+
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -305,11 +436,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -320,13 +455,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -334,16 +469,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -362,6 +488,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -369,38 +505,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
@ 2019-03-20 17:53 ` Wiles, Keith
2019-03-20 17:53 ` Wiles, Keith
2019-03-22 17:37 ` Ananyev, Konstantin
1 sibling, 2 replies; 163+ messages in thread
From: Wiles, Keith @ 2019-03-20 17:53 UTC (permalink / raw)
To: Ananyev, Konstantin; +Cc: dpdk-dev, akhil.goyal, olivier.matz
> On Mar 20, 2019, at 10:24 AM, Konstantin Ananyev <konstantin.ananyev@intel.com> wrote:
>
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 64 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..b967ad17e 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/** enum for the tx_offload bit-fields lenghts and offsets. */
> +enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OL3_LEN_BITS = 9,
> + RTE_MBUF_OL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
The naming of these fields is a bit unreadable and why do we need to add RTE_MBUF_ to everything especially adding RTE_ to everything.
Then we have to shorten words like OFFSET to OFS, it would be much better for readability to use something like this
RTE_MBUF_OL3_LEN_OFS to MBUF_OL3_LEN_OFFSET, then we have OL3 or OL2, which does not jump out at me as well.
What does OL3 or OL2 mean? OffLoad-3 or OffLoad-2 this also needs to be clearly defined to improve the readability of the macros.
Also for TXOFLD needs to be clearer here to help the reader understand the mean of the defines.
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +660,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @return
> + * raw tx_offload value.
> + */
> +static inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
> 2.17.1
>
Regards,
Keith
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 17:53 ` Wiles, Keith
@ 2019-03-20 17:53 ` Wiles, Keith
2019-03-22 17:37 ` Ananyev, Konstantin
1 sibling, 0 replies; 163+ messages in thread
From: Wiles, Keith @ 2019-03-20 17:53 UTC (permalink / raw)
To: Ananyev, Konstantin; +Cc: dpdk-dev, akhil.goyal, olivier.matz
> On Mar 20, 2019, at 10:24 AM, Konstantin Ananyev <konstantin.ananyev@intel.com> wrote:
>
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 64 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..b967ad17e 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/** enum for the tx_offload bit-fields lenghts and offsets. */
> +enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OL3_LEN_BITS = 9,
> + RTE_MBUF_OL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
The naming of these fields is a bit unreadable and why do we need to add RTE_MBUF_ to everything especially adding RTE_ to everything.
Then we have to shorten words like OFFSET to OFS, it would be much better for readability to use something like this
RTE_MBUF_OL3_LEN_OFS to MBUF_OL3_LEN_OFFSET, then we have OL3 or OL2, which does not jump out at me as well.
What does OL3 or OL2 mean? OffLoad-3 or OffLoad-2 this also needs to be clearly defined to improve the readability of the macros.
Also for TXOFLD needs to be clearer here to help the reader understand the mean of the defines.
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +660,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @return
> + * raw tx_offload value.
> + */
> +static inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
> 2.17.1
>
Regards,
Keith
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
` (13 preceding siblings ...)
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
` (7 more replies)
14 siblings, 8 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduce new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (7):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 509 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1016 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 71 ++-
10 files changed, 1273 insertions(+), 999 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (6 subsequent siblings)
7 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduce new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (7):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 509 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1016 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 71 ++-
10 files changed, 1273 insertions(+), 999 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
` (3 more replies)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
` (5 subsequent siblings)
7 siblings, 4 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
1 file changed, 64 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..b967ad17e 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,26 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/** enum for the tx_offload bit-fields lenghts and offsets. */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OL3_LEN_BITS = 9,
+ RTE_MBUF_OL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +660,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @return
+ * raw tx_offload value.
+ */
+static inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OL3_LEN_OFS |
+ ol2 << RTE_MBUF_OL2_LEN_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
` (2 subsequent siblings)
3 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
1 file changed, 64 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..b967ad17e 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,26 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/** enum for the tx_offload bit-fields lenghts and offsets. */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OL3_LEN_BITS = 9,
+ RTE_MBUF_OL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +660,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @return
+ * raw tx_offload value.
+ */
+static inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OL3_LEN_OFS |
+ ol2 << RTE_MBUF_OL2_LEN_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
` (4 subsequent siblings)
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 5f55c2a4e..31548e968 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -243,6 +246,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -285,6 +293,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -486,7 +499,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -513,8 +526,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1029,8 +1042,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 392e8fd7b..133a35d83 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -64,6 +64,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t proto; /* next proto */
uint8_t aad_len;
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 5f55c2a4e..31548e968 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -243,6 +246,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -285,6 +293,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -486,7 +499,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -513,8 +526,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1029,8 +1042,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 392e8fd7b..133a35d83 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -64,6 +64,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t proto; /* next proto */
uint8_t aad_len;
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (2 preceding siblings ...)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (3 subsequent siblings)
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 31548e968..74ed3b41e 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -425,20 +425,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -603,11 +600,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -625,10 +624,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -636,9 +635,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -739,11 +735,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -765,10 +763,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -776,9 +774,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -926,11 +921,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -938,9 +935,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -952,9 +951,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 31548e968..74ed3b41e 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -425,20 +425,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -603,11 +600,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -625,10 +624,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -636,9 +635,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -739,11 +735,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -765,10 +763,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -776,9 +774,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -926,11 +921,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -938,9 +935,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -952,9 +951,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (3 preceding siblings ...)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (2 subsequent siblings)
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 74ed3b41e..e4e3d7fc3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -411,14 +411,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -603,7 +620,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -627,17 +644,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -738,7 +755,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -766,17 +783,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -924,7 +941,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -941,10 +958,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -953,7 +969,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1105,7 +1121,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1115,9 +1131,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1131,10 +1147,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1144,23 +1160,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1170,10 +1190,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1183,23 +1203,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1215,7 +1239,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1228,16 +1252,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1257,23 +1281,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1314,7 +1338,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1332,22 +1356,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1366,7 +1388,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1388,22 +1410,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 74ed3b41e..e4e3d7fc3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -411,14 +411,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -603,7 +620,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -627,17 +644,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -738,7 +755,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -766,17 +783,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -924,7 +941,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -941,10 +958,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -953,7 +969,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1105,7 +1121,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1115,9 +1131,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1131,10 +1147,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1144,23 +1160,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1170,10 +1190,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1183,23 +1203,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1215,7 +1239,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1228,16 +1252,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1257,23 +1281,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1314,7 +1338,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1332,22 +1356,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1366,7 +1388,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1388,22 +1410,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (4 preceding siblings ...)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 408 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1011 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1086 insertions(+), 1000 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index b5f264831..5a7b11f56 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -145,4 +145,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..562185ebe
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,408 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ uint64_t *ivc, *ivp;
+ uint32_t clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ }
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..dce0fe051
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,535 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e4e3d7fc3..be6109768 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -411,569 +442,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- }
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1008,265 +476,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1303,142 +512,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1456,23 +529,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1498,10 +571,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1542,7 +615,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 133a35d83..93ff7bdf1 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -108,4 +108,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 408 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 535 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1011 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1086 insertions(+), 1000 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index b5f264831..5a7b11f56 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -145,4 +145,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..562185ebe
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,408 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ uint64_t *ivc, *ivp;
+ uint32_t clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ }
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..dce0fe051
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,535 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ /* AEAD (AES_GCM) case */
+ if (sa->aad_len != 0) {
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ /* CRYPT+AUTH case */
+ } else {
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e4e3d7fc3..be6109768 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -411,569 +442,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- /* AEAD (AES_GCM) case */
- if (sa->aad_len != 0) {
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- /* CRYPT+AUTH case */
- } else {
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- }
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1008,265 +476,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1303,142 +512,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1456,23 +529,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1498,10 +571,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1542,7 +615,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 133a35d83..93ff7bdf1 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -108,4 +108,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (5 preceding siblings ...)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 30 ++++++++++++++----------------
1 file changed, 14 insertions(+), 16 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 562185ebe..d479af8ea 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -26,11 +26,6 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct rte_crypto_sym_op *sop;
struct aead_gcm_iv *gcm;
uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
/* fill sym op fields */
sop = cop->sym;
@@ -38,7 +33,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* AEAD (AES_GCM) case */
if (sa->aad_len != 0) {
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -53,7 +48,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* CRYPT+AUTH case */
} else {
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -101,7 +96,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -128,6 +123,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -170,21 +170,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 30 ++++++++++++++----------------
1 file changed, 14 insertions(+), 16 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 562185ebe..d479af8ea 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -26,11 +26,6 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct rte_crypto_sym_op *sop;
struct aead_gcm_iv *gcm;
uint64_t *ivc, *ivp;
- uint32_t clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
/* fill sym op fields */
sop = cop->sym;
@@ -38,7 +33,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* AEAD (AES_GCM) case */
if (sa->aad_len != 0) {
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -53,7 +48,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
/* CRYPT+AUTH case */
} else {
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -101,7 +96,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -128,6 +123,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -170,21 +170,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process for esp inbound
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
` (6 preceding siblings ...)
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
7 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index d479af8ea..86a95e2f4 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline int32_t
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -188,111 +191,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
+
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -305,11 +436,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -320,13 +455,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -334,16 +469,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -362,6 +488,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -369,38 +505,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process for esp inbound
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-20 18:46 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-20 18:46 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index d479af8ea..86a95e2f4 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline int32_t
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -188,111 +191,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
+
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -305,11 +436,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -320,13 +455,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -334,16 +469,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -362,6 +488,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -369,38 +505,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
@ 2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
2019-03-21 6:04 ` Shahaf Shuler
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
3 siblings, 1 reply; 163+ messages in thread
From: Jerin Jacob Kollanukkaran @ 2019-03-21 3:33 UTC (permalink / raw)
To: konstantin.ananyev, dev; +Cc: olivier.matz, akhil.goyal
>
> +static inline uint64_t
I like this idea. I prefer to add __rte_always_inline for this
function.
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4,
> uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
@ 2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
0 siblings, 0 replies; 163+ messages in thread
From: Jerin Jacob Kollanukkaran @ 2019-03-21 3:33 UTC (permalink / raw)
To: konstantin.ananyev, dev; +Cc: olivier.matz, akhil.goyal
>
> +static inline uint64_t
I like this idea. I prefer to add __rte_always_inline for this
function.
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4,
> uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
@ 2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
3 siblings, 2 replies; 163+ messages in thread
From: Shahaf Shuler @ 2019-03-21 6:04 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: akhil.goyal, olivier.matz
Hi Konstantin,
Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> Subject: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx
> offload value
>
> Operations to set/update bit-fields often cause compilers to generate
> suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets, and new
> function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 71
> ++++++++++++++++++++++++++++++++++----
> 1 file changed, 64 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h index
> d961ccaf6..b967ad17e 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/** enum for the tx_offload bit-fields lenghts and offsets. */ enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OL3_LEN_BITS = 9,
> + RTE_MBUF_OL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS +
> RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS +
> RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS +
> RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS +
> RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS +
> RTE_MBUF_OL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT -
> RTE_MBUF_TXOFLD_UNUSED_OFS, };
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +660,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling
> pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t
> unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
I am not sure we want to decide now that this remaining byte is for another Tx offloads.
Maybe it is better to change a bit the rte_mbuf_tx_offload API to receive also the mbuf and to assign the offload values w/ mask.
> };
> };
>
> @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct
> rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @return
> + * raw tx_offload value.
> + */
> +static inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
> 2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-21 6:04 ` Shahaf Shuler
@ 2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 13:51 ` Ananyev, Konstantin
1 sibling, 0 replies; 163+ messages in thread
From: Shahaf Shuler @ 2019-03-21 6:04 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: akhil.goyal, olivier.matz
Hi Konstantin,
Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> Subject: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx
> offload value
>
> Operations to set/update bit-fields often cause compilers to generate
> suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets, and new
> function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 71
> ++++++++++++++++++++++++++++++++++----
> 1 file changed, 64 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h index
> d961ccaf6..b967ad17e 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/** enum for the tx_offload bit-fields lenghts and offsets. */ enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OL3_LEN_BITS = 9,
> + RTE_MBUF_OL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS +
> RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS +
> RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS +
> RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS +
> RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS +
> RTE_MBUF_OL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT -
> RTE_MBUF_TXOFLD_UNUSED_OFS, };
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +660,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling
> pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t
> unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
I am not sure we want to decide now that this remaining byte is for another Tx offloads.
Maybe it is better to change a bit the rte_mbuf_tx_offload API to receive also the mbuf and to assign the offload values w/ mask.
> };
> };
>
> @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct
> rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @return
> + * raw tx_offload value.
> + */
> +static inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OL3_LEN_OFS |
> + ol2 << RTE_MBUF_OL2_LEN_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
> 2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 6:04 ` Shahaf Shuler
@ 2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-24 8:00 ` Shahaf Shuler
1 sibling, 2 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-21 13:51 UTC (permalink / raw)
To: Shahaf Shuler, dev; +Cc: akhil.goyal, olivier.matz
Hi Shahaf,
> Hi Konstantin,
>
> Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> > Subject: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx
> > offload value
> >
> > Operations to set/update bit-fields often cause compilers to generate
> > suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets, and new
> > function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 71
> > ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 64 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h index
> > d961ccaf6..b967ad17e 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/** enum for the tx_offload bit-fields lenghts and offsets. */ enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OL3_LEN_BITS = 9,
> > + RTE_MBUF_OL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS +
> > RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS +
> > RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS +
> > RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS +
> > RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS +
> > RTE_MBUF_OL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT -
> > RTE_MBUF_TXOFLD_UNUSED_OFS, };
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling
> > pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> > */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> > Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> > Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t
> > unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
>
> I am not sure we want to decide now that this remaining byte is for another Tx offloads.
Not sure, I understand what is your concern here...
tx_offload field is 8B long, see 'uint64_t tx_offload' above.
All I did replaced raw number '8' with enum value ' RTE_MBUF_TXOFLD_UNUSED_BITS' which also evalueate to 8.
> Maybe it is better to change a bit the rte_mbuf_tx_offload API to receive also the mbuf and to assign the offload values w/ mask.
Could you explain a bit more what do you suggest here?
Konstantin
>
> > };
> > };
> >
> > @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct
> > rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OL2_LEN_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
> > 2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-21 13:51 ` Ananyev, Konstantin
@ 2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-24 8:00 ` Shahaf Shuler
1 sibling, 0 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-21 13:51 UTC (permalink / raw)
To: Shahaf Shuler, dev; +Cc: akhil.goyal, olivier.matz
Hi Shahaf,
> Hi Konstantin,
>
> Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> > Subject: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx
> > offload value
> >
> > Operations to set/update bit-fields often cause compilers to generate
> > suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets, and new
> > function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 71
> > ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 64 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h index
> > d961ccaf6..b967ad17e 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/** enum for the tx_offload bit-fields lenghts and offsets. */ enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OL3_LEN_BITS = 9,
> > + RTE_MBUF_OL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS +
> > RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS +
> > RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS +
> > RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS +
> > RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS +
> > RTE_MBUF_OL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT -
> > RTE_MBUF_TXOFLD_UNUSED_OFS, };
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling
> > pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> > */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> > Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> > Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t
> > unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
>
> I am not sure we want to decide now that this remaining byte is for another Tx offloads.
Not sure, I understand what is your concern here...
tx_offload field is 8B long, see 'uint64_t tx_offload' above.
All I did replaced raw number '8' with enum value ' RTE_MBUF_TXOFLD_UNUSED_BITS' which also evalueate to 8.
> Maybe it is better to change a bit the rte_mbuf_tx_offload API to receive also the mbuf and to assign the offload values w/ mask.
Could you explain a bit more what do you suggest here?
Konstantin
>
> > };
> > };
> >
> > @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct
> > rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OL2_LEN_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
> > 2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-20 17:53 ` Wiles, Keith
2019-03-20 17:53 ` Wiles, Keith
@ 2019-03-22 17:37 ` Ananyev, Konstantin
2019-03-22 17:37 ` Ananyev, Konstantin
1 sibling, 1 reply; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-22 17:37 UTC (permalink / raw)
To: Wiles, Keith; +Cc: dpdk-dev, akhil.goyal, olivier.matz
> -----Original Message-----
> From: Wiles, Keith
> Sent: Wednesday, March 20, 2019 5:53 PM
> To: Ananyev, Konstantin <konstantin.ananyev@intel.com>
> Cc: dpdk-dev <dev@dpdk.org>; akhil.goyal@nxp.com; olivier.matz@6wind.com
> Subject: Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
>
>
>
> > On Mar 20, 2019, at 10:24 AM, Konstantin Ananyev <konstantin.ananyev@intel.com> wrote:
> >
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 64 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..b967ad17e 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/** enum for the tx_offload bit-fields lenghts and offsets. */
> > +enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OL3_LEN_BITS = 9,
> > + RTE_MBUF_OL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
>
> The naming of these fields is a bit unreadable and why do we need to add RTE_MBUF_ to everything especially adding RTE_ to everything.
It is part of public API, so I think we'll have to follow DPDK naming convention here.
> Then we have to shorten words like OFFSET to OFS, it would be much better for readability to use something like this
>
> RTE_MBUF_OL3_LEN_OFS to MBUF_OL3_LEN_OFFSET, then we have OL3 or OL2, which does not jump out at me as well.
> What does OL3 or OL2 mean? OffLoad-3 or OffLoad-2 this also needs to be clearly defined to improve the readability of the macros.
outer_l3_len, outer_l2_len.
These enum values follows tx_offload bit-fields definitions - see below.
> Also for TXOFLD needs to be clearer here to help the reader understand the mean of the defines.
>
>
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OL2_LEN_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
> > 2.17.1
> >
>
> Regards,
> Keith
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-22 17:37 ` Ananyev, Konstantin
@ 2019-03-22 17:37 ` Ananyev, Konstantin
0 siblings, 0 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-22 17:37 UTC (permalink / raw)
To: Wiles, Keith; +Cc: dpdk-dev, akhil.goyal, olivier.matz
> -----Original Message-----
> From: Wiles, Keith
> Sent: Wednesday, March 20, 2019 5:53 PM
> To: Ananyev, Konstantin <konstantin.ananyev@intel.com>
> Cc: dpdk-dev <dev@dpdk.org>; akhil.goyal@nxp.com; olivier.matz@6wind.com
> Subject: Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
>
>
>
> > On Mar 20, 2019, at 10:24 AM, Konstantin Ananyev <konstantin.ananyev@intel.com> wrote:
> >
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 71 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 64 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..b967ad17e 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,26 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/** enum for the tx_offload bit-fields lenghts and offsets. */
> > +enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OL3_LEN_BITS = 9,
> > + RTE_MBUF_OL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OL3_LEN_OFS = RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OL2_LEN_OFS = RTE_MBUF_OL3_LEN_OFS + RTE_MBUF_OL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OL2_LEN_OFS + RTE_MBUF_OL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
>
> The naming of these fields is a bit unreadable and why do we need to add RTE_MBUF_ to everything especially adding RTE_ to everything.
It is part of public API, so I think we'll have to follow DPDK naming convention here.
> Then we have to shorten words like OFFSET to OFS, it would be much better for readability to use something like this
>
> RTE_MBUF_OL3_LEN_OFS to MBUF_OL3_LEN_OFFSET, then we have OL3 or OL2, which does not jump out at me as well.
> What does OL3 or OL2 mean? OffLoad-3 or OffLoad-2 this also needs to be clearly defined to improve the readability of the macros.
outer_l3_len, outer_l2_len.
These enum values follows tx_offload bit-fields definitions - see below.
> Also for TXOFLD needs to be clearer here to help the reader understand the mean of the defines.
>
>
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2268,38 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OL2_LEN_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
> > 2.17.1
> >
>
> Regards,
> Keith
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-21 13:51 ` Ananyev, Konstantin
@ 2019-03-24 8:00 ` Shahaf Shuler
2019-03-24 8:00 ` Shahaf Shuler
1 sibling, 1 reply; 163+ messages in thread
From: Shahaf Shuler @ 2019-03-24 8:00 UTC (permalink / raw)
To: Ananyev, Konstantin, dev; +Cc: akhil.goyal, olivier.matz
Thursday, March 21, 2019 3:52 PM, Ananyev, Konstantin:
> Subject: RE: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw
> Tx offload value
>
> Hi Shahaf,
>
> > Hi Konstantin,
> >
> > Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> > > /**
> > > * The generic rte_mbuf, containing a packet mbuf.
> > > */
> > > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > > uint64_t tx_offload; /**< combined for easy fetch */
> > > __extension__
> > > struct {
> > > - uint64_t l2_len:7;
> > > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > > * Outer_L4_len + ... + Inner_L2_len for tunneling
> pkt.
> > > */
> > > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> > > */
> > > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > > + /**< L3 (IP) Header Length. */
> > > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > > + /**< L4 (TCP/UDP) Header Length. */
> > > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > > + /**< TCP TSO segment size */
> > >
> > > /* fields for TX offloading of tunnels */
> > > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> > > Length. */
> > > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> > > Length. */
> > > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > > + /**< Outer L3 (IP) Hdr Length. */
> > > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > > + /**< Outer L2 (MAC) Hdr Length. */
> > >
> > > - /* uint64_t unused:8; */
> > > + /* uint64_t
> > > unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> >
> > I am not sure we want to decide now that this remaining byte is for another
> Tx offloads.
>
> Not sure, I understand what is your concern here...
> tx_offload field is 8B long, see 'uint64_t tx_offload' above.
> All I did replaced raw number '8' with enum value '
> RTE_MBUF_TXOFLD_UNUSED_BITS' which also evalueate to 8.
Sorry, my bad, missed this union w/ the tx_offload. No further issues 😊.
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value
2019-03-24 8:00 ` Shahaf Shuler
@ 2019-03-24 8:00 ` Shahaf Shuler
0 siblings, 0 replies; 163+ messages in thread
From: Shahaf Shuler @ 2019-03-24 8:00 UTC (permalink / raw)
To: Ananyev, Konstantin, dev; +Cc: akhil.goyal, olivier.matz
Thursday, March 21, 2019 3:52 PM, Ananyev, Konstantin:
> Subject: RE: [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw
> Tx offload value
>
> Hi Shahaf,
>
> > Hi Konstantin,
> >
> > Wednesday, March 20, 2019 8:47 PM, Konstantin Ananyev:
> > > /**
> > > * The generic rte_mbuf, containing a packet mbuf.
> > > */
> > > @@ -640,19 +660,24 @@ struct rte_mbuf {
> > > uint64_t tx_offload; /**< combined for easy fetch */
> > > __extension__
> > > struct {
> > > - uint64_t l2_len:7;
> > > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > > * Outer_L4_len + ... + Inner_L2_len for tunneling
> pkt.
> > > */
> > > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length.
> > > */
> > > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > > + /**< L3 (IP) Header Length. */
> > > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > > + /**< L4 (TCP/UDP) Header Length. */
> > > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > > + /**< TCP TSO segment size */
> > >
> > > /* fields for TX offloading of tunnels */
> > > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr
> > > Length. */
> > > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr
> > > Length. */
> > > + uint64_t outer_l3_len:RTE_MBUF_OL3_LEN_BITS;
> > > + /**< Outer L3 (IP) Hdr Length. */
> > > + uint64_t outer_l2_len:RTE_MBUF_OL2_LEN_BITS;
> > > + /**< Outer L2 (MAC) Hdr Length. */
> > >
> > > - /* uint64_t unused:8; */
> > > + /* uint64_t
> > > unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> >
> > I am not sure we want to decide now that this remaining byte is for another
> Tx offloads.
>
> Not sure, I understand what is your concern here...
> tx_offload field is 8B long, see 'uint64_t tx_offload' above.
> All I did replaced raw number '8' with enum value '
> RTE_MBUF_TXOFLD_UNUSED_BITS' which also evalueate to 8.
Sorry, my bad, missed this union w/ the tx_offload. No further issues 😊.
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (2 preceding siblings ...)
2019-03-21 6:04 ` Shahaf Shuler
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
` (10 more replies)
3 siblings, 11 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (8):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 580 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1072 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 79 ++-
10 files changed, 1364 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (9 subsequent siblings)
10 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (8):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 +++++++++++++++++
lib/librte_ipsec/esp_outb.c | 580 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 --
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1072 ++--------------------------------
lib/librte_ipsec/sa.h | 45 ++
lib/librte_mbuf/rte_mbuf.h | 79 ++-
10 files changed, 1364 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:16 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA Konstantin Ananyev
` (8 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
1 file changed, 72 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..0b197e8ce 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:16 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
1 file changed, 72 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..0b197e8ce 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:52 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op Konstantin Ananyev
` (7 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:52 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (2 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 9:02 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (6 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 9:02 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (3 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 10:52 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (5 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..009efd1d3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 10:52 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
1 file changed, 93 insertions(+), 73 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..009efd1d3 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ mbuf_bad_move(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ mbuf_bad_move(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (4 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:20 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (4 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..2a2b1be82
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..7ac020fa3
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 009efd1d3..3068e1b28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:20 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..2a2b1be82
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ mbuf_bad_move(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..7ac020fa3
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ mbuf_bad_move(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ mbuf_bad_move(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..8988b4903
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * dr[] contains the indexes of bad mbufs insinde the mb[].
+ */
+static inline void
+mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
+ uint32_t drn)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[drn];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != num; i++) {
+ if (j != drn && i == dr[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != drn; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 009efd1d3..3068e1b28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * dr[] contains the indexes of bad mbufs insinde the mb[].
- */
-static void
-mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
- uint32_t drn)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[drn];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != num; i++) {
- if (j != drn && i == dr[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != drn; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- mbuf_bad_move(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (5 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:27 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process " Konstantin Ananyev
` (3 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 2a2b1be82..ea95e21e1 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:27 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 2a2b1be82..ea95e21e1 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
mbuf_bad_move(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process for esp inbound
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (6 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
` (2 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index ea95e21e1..299c9d55d 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process for esp inbound
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index ea95e21e1..299c9d55d 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- mbuf_bad_move(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- mbuf_bad_move(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (7 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:35 ` Akhil Goyal
2019-03-28 11:21 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Akhil Goyal
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 299c9d55d..3e717e550 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 7ac020fa3..a37561eba 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:35 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-26 15:43 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 299c9d55d..3e717e550 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 7ac020fa3..a37561eba 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 8:16 ` Akhil Goyal
2019-03-28 8:16 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 8:16 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 72 insertions(+), 7 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value
2019-03-28 8:16 ` Akhil Goyal
@ 2019-03-28 8:16 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 8:16 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 72 insertions(+), 7 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 8:52 ` Akhil Goyal
2019-03-28 8:52 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 8:52 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code. To avoid such negative effect,
> use tx_offload raw value and mask to update l2_len and l3_len
> fields within mbufs.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA
2019-03-28 8:52 ` Akhil Goyal
@ 2019-03-28 8:52 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 8:52 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code. To avoid such negative effect,
> use tx_offload raw value and mask to update l2_len and l3_len
> fields within mbufs.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 9:02 ` Akhil Goyal
2019-03-28 9:02 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 9:02 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Right now we first fill crypto_sym_op part of crypto_op,
> then in a separate cycle we fill crypto op fields.
> It makes more sense to fill whole crypto-op in one go instead.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op
2019-03-28 9:02 ` Akhil Goyal
@ 2019-03-28 9:02 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 9:02 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Right now we first fill crypto_sym_op part of crypto_op,
> then in a separate cycle we fill crypto op fields.
> It makes more sense to fill whole crypto-op in one go instead.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 10:52 ` Akhil Goyal
2019-03-28 10:52 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 10:52 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> As was pointed in one of previous reviews - we can avoid updating
> contents of mbuf array for successfully processed packets.
> Instead store indexes of failed packets, to move them beyond the good
> ones later.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
> 1 file changed, 93 insertions(+), 73 deletions(-)
>
> diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
> index 97c0f8c61..009efd1d3 100644
> --- a/lib/librte_ipsec/sa.c
> +++ b/lib/librte_ipsec/sa.c
> @@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
> return sz;
> }
>
> -static inline void
> -mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
> - uint32_t num)
> +/*
> + * Move bad (unprocessed) mbufs beyond the good (processed) ones.
> + * dr[] contains the indexes of bad mbufs insinde the mb[].
spell check.
dr[],drb,drn looks cryptic can we have better naming which is more readable.
> + */
> +static void
> +mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
move_bad_mbufs() would be better.
> + uint32_t drn)
> {
> - uint32_t i;
> + uint32_t i, j, k;
> + struct rte_mbuf *drb[drn];
> +
> + j = 0;
> + k = 0;
>
> - for (i = 0; i != num; i++)
> - dst[i] = src[i];
> + /* copy bad ones into a temp place */
> + for (i = 0; i != num; i++) {
> + if (j != drn && i == dr[j])
> + drb[j++] = mb[i];
> + else
> + mb[k++] = mb[i];
> + }
> +
> + /* copy bad ones after the good ones */
> + for (i = 0; i != drn; i++)
> + mb[k + i] = drb[i];
> }
>
Apart from that
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted
2019-03-28 10:52 ` Akhil Goyal
@ 2019-03-28 10:52 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 10:52 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> As was pointed in one of previous reviews - we can avoid updating
> contents of mbuf array for successfully processed packets.
> Instead store indexes of failed packets, to move them beyond the good
> ones later.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/sa.c | 166 +++++++++++++++++++++++-------------------
> 1 file changed, 93 insertions(+), 73 deletions(-)
>
> diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
> index 97c0f8c61..009efd1d3 100644
> --- a/lib/librte_ipsec/sa.c
> +++ b/lib/librte_ipsec/sa.c
> @@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
> return sz;
> }
>
> -static inline void
> -mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
> - uint32_t num)
> +/*
> + * Move bad (unprocessed) mbufs beyond the good (processed) ones.
> + * dr[] contains the indexes of bad mbufs insinde the mb[].
spell check.
dr[],drb,drn looks cryptic can we have better naming which is more readable.
> + */
> +static void
> +mbuf_bad_move(struct rte_mbuf *mb[], const uint32_t dr[], uint32_t num,
move_bad_mbufs() would be better.
> + uint32_t drn)
> {
> - uint32_t i;
> + uint32_t i, j, k;
> + struct rte_mbuf *drb[drn];
> +
> + j = 0;
> + k = 0;
>
> - for (i = 0; i != num; i++)
> - dst[i] = src[i];
> + /* copy bad ones into a temp place */
> + for (i = 0; i != num; i++) {
> + if (j != drn && i == dr[j])
> + drb[j++] = mb[i];
> + else
> + mb[k++] = mb[i];
> + }
> +
> + /* copy bad ones after the good ones */
> + for (i = 0; i != drn; i++)
> + mb[k + i] = drb[i];
> }
>
Apart from that
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 11:20 ` Akhil Goyal
2019-03-28 11:20 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:20 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> sa.c becomes too big, so decided to split it into 3 chunks:
> - sa.c - control path related functions (init/fini, etc.)
> - esp_inb.c - ESP inbound packet processing
> - esp_outb.c - ESP outbound packet processing
>
> Plus few changes in internal function names to follow the same
> code convention.
> No functional changes introduced.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
> lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 30 -
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 ++
> lib/librte_ipsec/sa.c | 1067 ++--------------------------------
> lib/librte_ipsec/sa.h | 40 ++
> 9 files changed, 1141 insertions(+), 1056 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files
2019-03-28 11:20 ` Akhil Goyal
@ 2019-03-28 11:20 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:20 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> sa.c becomes too big, so decided to split it into 3 chunks:
> - sa.c - control path related functions (init/fini, etc.)
> - esp_inb.c - ESP inbound packet processing
> - esp_outb.c - ESP outbound packet processing
>
> Plus few changes in internal function names to follow the same
> code convention.
> No functional changes introduced.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
> lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 30 -
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 ++
> lib/librte_ipsec/sa.c | 1067 ++--------------------------------
> lib/librte_ipsec/sa.h | 40 ++
> 9 files changed, 1141 insertions(+), 1056 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (8 preceding siblings ...)
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-03-28 11:21 ` Akhil Goyal
2019-03-28 11:21 ` Akhil Goyal
2019-03-28 11:49 ` Ananyev, Konstantin
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
10 siblings, 2 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:21 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
Hi Konstantin,
I do not see ipsec lib in Shared Library Versions in release notes.
Could you also send that change.
Thanks,
Akhil
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> v3 -> v2
> Address comments:
> - make rte_mbuf_tx_offload() always inline
> - rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
> Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
> Some code de-duplication
>
> v1 -> v2
> Add patch to improve a bit esp inbound pkt_process code
>
> The first patch in this series introduces new function to
> set raw value for rte_mbuf tx_offload.
> Remaining ones are cleanups and small performance improvements
> (on IA ~3-6% depending on the test-case) for ipsec library.
>
> Konstantin Ananyev (8):
> mbuf: new function to generate raw Tx offload value
> ipsec: add Tx offload template into SA
> ipsec: change the order in filling crypto op
> ipsec: change the way unprocessed mbufs are accounted
> ipsec: move inbound and outbound code into different files
> ipsec: reorder packet check for esp inbound
> ipsec: reorder packet process for esp inbound
> ipsec: de-duplicate crypto op prepare code-path
>
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 547 +++++++++++++++++
> lib/librte_ipsec/esp_outb.c | 580 ++++++++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 34 --
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 ++
> lib/librte_ipsec/sa.c | 1072 ++--------------------------------
> lib/librte_ipsec/sa.h | 45 ++
> lib/librte_mbuf/rte_mbuf.h | 79 ++-
> 10 files changed, 1364 insertions(+), 1055 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-28 11:21 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Akhil Goyal
@ 2019-03-28 11:21 ` Akhil Goyal
2019-03-28 11:49 ` Ananyev, Konstantin
1 sibling, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:21 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
Hi Konstantin,
I do not see ipsec lib in Shared Library Versions in release notes.
Could you also send that change.
Thanks,
Akhil
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> v3 -> v2
> Address comments:
> - make rte_mbuf_tx_offload() always inline
> - rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
> Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
> Some code de-duplication
>
> v1 -> v2
> Add patch to improve a bit esp inbound pkt_process code
>
> The first patch in this series introduces new function to
> set raw value for rte_mbuf tx_offload.
> Remaining ones are cleanups and small performance improvements
> (on IA ~3-6% depending on the test-case) for ipsec library.
>
> Konstantin Ananyev (8):
> mbuf: new function to generate raw Tx offload value
> ipsec: add Tx offload template into SA
> ipsec: change the order in filling crypto op
> ipsec: change the way unprocessed mbufs are accounted
> ipsec: move inbound and outbound code into different files
> ipsec: reorder packet check for esp inbound
> ipsec: reorder packet process for esp inbound
> ipsec: de-duplicate crypto op prepare code-path
>
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 547 +++++++++++++++++
> lib/librte_ipsec/esp_outb.c | 580 ++++++++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 34 --
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 ++
> lib/librte_ipsec/sa.c | 1072 ++--------------------------------
> lib/librte_ipsec/sa.h | 45 ++
> lib/librte_mbuf/rte_mbuf.h | 79 ++-
> 10 files changed, 1364 insertions(+), 1055 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 11:27 ` Akhil Goyal
2019-03-28 11:27 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:27 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Right now check for packet length and padding is done inside cop_prepare().
> It makes sense to have all necessary checks in one place at early stage:
> inside pkt_prepare().
> That allows to simplify (and later hopefully) optimize cop_prepare() part.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
> 1 file changed, 18 insertions(+), 23 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound
2019-03-28 11:27 ` Akhil Goyal
@ 2019-03-28 11:27 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:27 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> Right now check for packet length and padding is done inside cop_prepare().
> It makes sense to have all necessary checks in one place at early stage:
> inside pkt_prepare().
> That allows to simplify (and later hopefully) optimize cop_prepare() part.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
> 1 file changed, 18 insertions(+), 23 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
@ 2019-03-28 11:35 ` Akhil Goyal
2019-03-28 11:35 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:35 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> For sym_crypto_op prepare move common code into a separate function(s).
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
> lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
> 2 files changed, 80 insertions(+), 49 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path
2019-03-28 11:35 ` Akhil Goyal
@ 2019-03-28 11:35 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-28 11:35 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/26/2019 9:13 PM, Konstantin Ananyev wrote:
> For sym_crypto_op prepare move common code into a separate function(s).
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
> lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
> 2 files changed, 80 insertions(+), 49 deletions(-)
>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-28 11:21 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Akhil Goyal
2019-03-28 11:21 ` Akhil Goyal
@ 2019-03-28 11:49 ` Ananyev, Konstantin
2019-03-28 11:49 ` Ananyev, Konstantin
1 sibling, 1 reply; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-28 11:49 UTC (permalink / raw)
To: Akhil Goyal, dev; +Cc: olivier.matz
Hi Akhil,
>
> Hi Konstantin,
>
> I do not see ipsec lib in Shared Library Versions in release notes.
> Could you also send that change.
Oops, completely forgot about that one.
Will do in v4.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library
2019-03-28 11:49 ` Ananyev, Konstantin
@ 2019-03-28 11:49 ` Ananyev, Konstantin
0 siblings, 0 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-28 11:49 UTC (permalink / raw)
To: Akhil Goyal, dev; +Cc: olivier.matz
Hi Akhil,
>
> Hi Konstantin,
>
> I do not see ipsec lib in Shared Library Versions in release notes.
> Could you also send that change.
Oops, completely forgot about that one.
Will do in v4.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 0/9] Few small improvements for ipsec library
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
` (9 preceding siblings ...)
2019-03-28 11:21 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Akhil Goyal
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
` (10 more replies)
10 siblings, 11 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
V4 -> V3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 79 +-
11 files changed, 1365 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 0/9] Few small improvements for ipsec library
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (9 subsequent siblings)
10 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
V4 -> V3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new function to
set raw value for rte_mbuf tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 79 +-
11 files changed, 1365 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 12:54 ` Olivier Matz
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
` (8 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
1 file changed, 72 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..0b197e8ce 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 12:54 ` Olivier Matz
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
1 file changed, 72 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..0b197e8ce 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
` (7 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (2 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (6 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (3 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (5 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (4 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (4 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (5 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process " Konstantin Ananyev
` (3 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process for esp inbound
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (6 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
` (2 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process for esp inbound
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (7 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (8 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 16:03 ` Akhil Goyal
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index d11bb5a2b..f41d1104e 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -199,6 +199,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 16:03 ` Akhil Goyal
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-03-29 10:27 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index d11bb5a2b..f41d1104e 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -199,6 +199,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
@ 2019-03-29 12:54 ` Olivier Matz
2019-03-29 12:54 ` Olivier Matz
2019-03-30 14:20 ` Ananyev, Konstantin
1 sibling, 2 replies; 163+ messages in thread
From: Olivier Matz @ 2019-03-29 12:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
Hi Konstantin,
On Fri, Mar 29, 2019 at 10:27:18AM +0000, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
I understand the need. Out of curiosity, do you have any performance
numbers to share?
Few cosmetic questions below.
> ---
> lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 72 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..0b197e8ce 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/**
> + * enum for the tx_offload bit-fields lenghts and offsets.
> + * defines the layout of rte_mbuf tx_offload field.
> + */
> +enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OUTL3_LEN_BITS = 9,
> + RTE_MBUF_OUTL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OUTL3_LEN_OFS =
> + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OUTL2_LEN_OFS =
> + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
> +
What is the advantage of defining an enum instead of #defines?
In any case, I wonder if it wouldn't be clearer to change the order like
this:
enum {
RTE_MBUF_L2_LEN_OFS = 0,
RTE_MBUF_L2_LEN_BITS = 7,
RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
RTE_MBUF_L3_LEN_BITS = 9,
RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
RTE_MBUF_L4_LEN_BITS = 8,
...
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +665,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @param unused
> + * unused value.
> + * @return
> + * raw tx_offload value.
> + */
> +static __rte_always_inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2, uint64_t unused)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> +}
> +
> /**
>From what I see, the problem is quite similar to what was done with
rte_mbuf_sched_set() recently. So I wondered if it was possible to
declare a structure like this:
struct rte_mbuf_ol_len {
uint64_t l2_len:7;
uint64_t l3_len:9; /**< L3 (IP) Header Length. */
uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
...
}
And have the set function like this:
m->l = (struct rte_mbuf_ol_len) {
.l2_len = l2_len,
.l3_len = l3_len,
.l4_len = l4_len,
...
This would avoid the definition of the offsets and bits, but I didn't
find any way to declare these fields as anonymous in the mbuf structure.
Did you tried that way too?
Thanks,
Olivier
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-29 12:54 ` Olivier Matz
@ 2019-03-29 12:54 ` Olivier Matz
2019-03-30 14:20 ` Ananyev, Konstantin
1 sibling, 0 replies; 163+ messages in thread
From: Olivier Matz @ 2019-03-29 12:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
Hi Konstantin,
On Fri, Mar 29, 2019 at 10:27:18AM +0000, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
I understand the need. Out of curiosity, do you have any performance
numbers to share?
Few cosmetic questions below.
> ---
> lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 72 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..0b197e8ce 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/**
> + * enum for the tx_offload bit-fields lenghts and offsets.
> + * defines the layout of rte_mbuf tx_offload field.
> + */
> +enum {
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OUTL3_LEN_BITS = 9,
> + RTE_MBUF_OUTL2_LEN_BITS = 7,
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_OUTL3_LEN_OFS =
> + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OUTL2_LEN_OFS =
> + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
> +
What is the advantage of defining an enum instead of #defines?
In any case, I wonder if it wouldn't be clearer to change the order like
this:
enum {
RTE_MBUF_L2_LEN_OFS = 0,
RTE_MBUF_L2_LEN_BITS = 7,
RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
RTE_MBUF_L3_LEN_BITS = 9,
RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
RTE_MBUF_L4_LEN_BITS = 8,
...
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +665,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @param unused
> + * unused value.
> + * @return
> + * raw tx_offload value.
> + */
> +static __rte_always_inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2, uint64_t unused)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> +}
> +
> /**
From what I see, the problem is quite similar to what was done with
rte_mbuf_sched_set() recently. So I wondered if it was possible to
declare a structure like this:
struct rte_mbuf_ol_len {
uint64_t l2_len:7;
uint64_t l3_len:9; /**< L3 (IP) Header Length. */
uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
...
}
And have the set function like this:
m->l = (struct rte_mbuf_ol_len) {
.l2_len = l2_len,
.l3_len = l3_len,
.l4_len = l4_len,
...
This would avoid the definition of the offsets and bits, but I didn't
find any way to declare these fields as anonymous in the mbuf structure.
Did you tried that way too?
Thanks,
Olivier
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
@ 2019-03-29 16:03 ` Akhil Goyal
2019-03-29 16:03 ` Akhil Goyal
1 sibling, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-03-29 16:03 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/29/2019 3:57 PM, Konstantin Ananyev wrote:
> Add librte_ipsec into 'Shared Library Versions' list in the release notes.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> doc/guides/rel_notes/release_19_05.rst | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
> index d11bb5a2b..f41d1104e 100644
> --- a/doc/guides/rel_notes/release_19_05.rst
> +++ b/doc/guides/rel_notes/release_19_05.rst
> @@ -199,6 +199,7 @@ The libraries prepended with a plus sign were incremented in this version.
> librte_gso.so.1
> librte_hash.so.2
> librte_ip_frag.so.1
> + librte_ipsec.so.1
> librte_jobstats.so.1
> librte_kni.so.2
> librte_kvargs.so.1
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list
2019-03-29 16:03 ` Akhil Goyal
@ 2019-03-29 16:03 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-03-29 16:03 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 3/29/2019 3:57 PM, Konstantin Ananyev wrote:
> Add librte_ipsec into 'Shared Library Versions' list in the release notes.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
> doc/guides/rel_notes/release_19_05.rst | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
> index d11bb5a2b..f41d1104e 100644
> --- a/doc/guides/rel_notes/release_19_05.rst
> +++ b/doc/guides/rel_notes/release_19_05.rst
> @@ -199,6 +199,7 @@ The libraries prepended with a plus sign were incremented in this version.
> librte_gso.so.1
> librte_hash.so.2
> librte_ip_frag.so.1
> + librte_ipsec.so.1
> librte_jobstats.so.1
> librte_kni.so.2
> librte_kvargs.so.1
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-29 12:54 ` Olivier Matz
2019-03-29 12:54 ` Olivier Matz
@ 2019-03-30 14:20 ` Ananyev, Konstantin
2019-03-30 14:20 ` Ananyev, Konstantin
1 sibling, 1 reply; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-30 14:20 UTC (permalink / raw)
To: Olivier Matz; +Cc: dev, akhil.goyal
Hi Olivier,
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
>
> I understand the need. Out of curiosity, do you have any performance
> numbers to share?
On my board (SKX):
for micro-benchmark (doing nothing but setting tx_offload for 1M mbufs in a loop)
the difference is more than 150% - from ~55 cycles to ~20 cycles per iteration.
For ipsec-secgw - ~3% improvement for tunneled outbound packets.
>
> Few cosmetic questions below.
>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 72 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..0b197e8ce 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/**
> > + * enum for the tx_offload bit-fields lenghts and offsets.
> > + * defines the layout of rte_mbuf tx_offload field.
> > + */
> > +enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OUTL3_LEN_BITS = 9,
> > + RTE_MBUF_OUTL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OUTL3_LEN_OFS =
> > + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OUTL2_LEN_OFS =
> > + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
> > +
>
> What is the advantage of defining an enum instead of #defines?
No big difference here, just looks nicer to me.
>
> In any case, I wonder if it wouldn't be clearer to change the order like
> this:
>
> enum {
> RTE_MBUF_L2_LEN_OFS = 0,
> RTE_MBUF_L2_LEN_BITS = 7,
> RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> RTE_MBUF_L3_LEN_BITS = 9,
> RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> RTE_MBUF_L4_LEN_BITS = 8,
> ...
NP, can do this way.
>
>
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +665,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @param unused
> > + * unused value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static __rte_always_inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2, uint64_t unused)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> > + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> > +}
> > +
> > /**
>
>
> From what I see, the problem is quite similar to what was done with
> rte_mbuf_sched_set() recently. So I wondered if it was possible to
> declare a structure like this:
>
> struct rte_mbuf_ol_len {
> uint64_t l2_len:7;
> uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> ...
> }
>
> And have the set function like this:
>
> m->l = (struct rte_mbuf_ol_len) {
> .l2_len = l2_len,
> .l3_len = l3_len,
> .l4_len = l4_len,
> ...
>
> This would avoid the definition of the offsets and bits, but I didn't
> find any way to declare these fields as anonymous in the mbuf structure.
> Did you tried that way too?
I thought about such approach, but as you said above it would change
from unnamed struct to named one.
Which, as I understand, means API breakage.
So don't think the hassle will be worth the benefit.
Also the code wouldn't be totally identical - that approach will generate few
extra 'AND' instructions.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value
2019-03-30 14:20 ` Ananyev, Konstantin
@ 2019-03-30 14:20 ` Ananyev, Konstantin
0 siblings, 0 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-03-30 14:20 UTC (permalink / raw)
To: Olivier Matz; +Cc: dev, akhil.goyal
Hi Olivier,
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
>
> I understand the need. Out of curiosity, do you have any performance
> numbers to share?
On my board (SKX):
for micro-benchmark (doing nothing but setting tx_offload for 1M mbufs in a loop)
the difference is more than 150% - from ~55 cycles to ~20 cycles per iteration.
For ipsec-secgw - ~3% improvement for tunneled outbound packets.
>
> Few cosmetic questions below.
>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 79 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 72 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..0b197e8ce 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/**
> > + * enum for the tx_offload bit-fields lenghts and offsets.
> > + * defines the layout of rte_mbuf tx_offload field.
> > + */
> > +enum {
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OUTL3_LEN_BITS = 9,
> > + RTE_MBUF_OUTL2_LEN_BITS = 7,
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_OUTL3_LEN_OFS =
> > + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OUTL2_LEN_OFS =
> > + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
> > +
>
> What is the advantage of defining an enum instead of #defines?
No big difference here, just looks nicer to me.
>
> In any case, I wonder if it wouldn't be clearer to change the order like
> this:
>
> enum {
> RTE_MBUF_L2_LEN_OFS = 0,
> RTE_MBUF_L2_LEN_BITS = 7,
> RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> RTE_MBUF_L3_LEN_BITS = 9,
> RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> RTE_MBUF_L4_LEN_BITS = 8,
> ...
NP, can do this way.
>
>
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +665,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2273,41 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @param unused
> > + * unused value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static __rte_always_inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2, uint64_t unused)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> > + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> > +}
> > +
> > /**
>
>
> From what I see, the problem is quite similar to what was done with
> rte_mbuf_sched_set() recently. So I wondered if it was possible to
> declare a structure like this:
>
> struct rte_mbuf_ol_len {
> uint64_t l2_len:7;
> uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> ...
> }
>
> And have the set function like this:
>
> m->l = (struct rte_mbuf_ol_len) {
> .l2_len = l2_len,
> .l3_len = l3_len,
> .l4_len = l4_len,
> ...
>
> This would avoid the definition of the offsets and bits, but I didn't
> find any way to declare these fields as anonymous in the mbuf structure.
> Did you tried that way too?
I thought about such approach, but as you said above it would change
from unnamed struct to named one.
Which, as I understand, means API breakage.
So don't think the hassle will be worth the benefit.
Also the code wouldn't be totally identical - that approach will generate few
extra 'AND' instructions.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
` (9 preceding siblings ...)
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
` (10 more replies)
10 siblings, 11 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v5 -> v4
Few cosmetic changes in mbuf tx_offload patch requested by Olivier
v4 -> v3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new enum for rte_mbuf
tx_offload layout and new function to set raw value for rte_mbuf
tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 81 +-
11 files changed, 1367 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (9 subsequent siblings)
10 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v5 -> v4
Few cosmetic changes in mbuf tx_offload patch requested by Olivier
v4 -> v3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new enum for rte_mbuf
tx_offload layout and new function to set raw value for rte_mbuf
tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 81 +-
11 files changed, 1367 insertions(+), 1055 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
` (2 more replies)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
` (8 subsequent siblings)
10 siblings, 3 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
1 file changed, 74 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..c3d6c4d78 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * Note that it is caller responsibility to make sure that input parameters
+ * don't exceed maximum bit-field values.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 13:18 ` Akhil Goyal
2019-04-01 13:22 ` Olivier Matz
2 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
1 file changed, 74 insertions(+), 7 deletions(-)
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..c3d6c4d78 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -479,6 +479,31 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_BITS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +665,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * Note that it is caller responsibility to make sure that input parameters
+ * don't exceed maximum bit-field values.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
` (7 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (2 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (6 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (3 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (5 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (4 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (4 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (5 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process " Konstantin Ananyev
` (3 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process for esp inbound
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (6 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
` (2 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process for esp inbound
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (7 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (8 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index 3aad8d6a5..321a8b48b 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -234,6 +234,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-04-01 12:56 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-01 12:56 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index 3aad8d6a5..321a8b48b 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -234,6 +234,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
@ 2019-04-01 13:18 ` Akhil Goyal
2019-04-01 13:18 ` Akhil Goyal
2019-04-01 13:22 ` Olivier Matz
2 siblings, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-04-01 13:18 UTC (permalink / raw)
To: olivier.matz; +Cc: Konstantin Ananyev, dev
Hi Olivier,
Could you please review this. We have an RC1 deadline.
If no further comments, could you please Ack?
Thanks,
Akhil
On 4/1/2019 6:26 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 74 insertions(+), 7 deletions(-)
>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 13:18 ` Akhil Goyal
@ 2019-04-01 13:18 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-04-01 13:18 UTC (permalink / raw)
To: olivier.matz; +Cc: Konstantin Ananyev, dev
Hi Olivier,
Could you please review this. We have an RC1 deadline.
If no further comments, could you please Ack?
Thanks,
Akhil
On 4/1/2019 6:26 PM, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 74 insertions(+), 7 deletions(-)
>
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 13:18 ` Akhil Goyal
@ 2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:55 ` Ananyev, Konstantin
2 siblings, 2 replies; 163+ messages in thread
From: Olivier Matz @ 2019-04-01 13:22 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
Hi Konstantin,
On Mon, Apr 01, 2019 at 01:56:48PM +0100, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 74 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..c3d6c4d78 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/**
> + * enum for the tx_offload bit-fields lenghts and offsets.
> + * defines the layout of rte_mbuf tx_offload field.
> + */
> +enum {
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OUTL3_LEN_OFS =
> + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OUTL3_LEN_BITS = 9,
> + RTE_MBUF_OUTL2_LEN_OFS =
> + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> + RTE_MBUF_OUTL2_LEN_BITS = 7,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +665,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * Note that it is caller responsibility to make sure that input parameters
> + * don't exceed maximum bit-field values.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @param unused
> + * unused value.
> + * @return
> + * raw tx_offload value.
> + */
> +static __rte_always_inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2, uint64_t unused)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
Thank you for the update.
I missed something important at the first review: will it work on any
endianess?
Olivier
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 13:22 ` Olivier Matz
@ 2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:55 ` Ananyev, Konstantin
1 sibling, 0 replies; 163+ messages in thread
From: Olivier Matz @ 2019-04-01 13:22 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
Hi Konstantin,
On Mon, Apr 01, 2019 at 01:56:48PM +0100, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> ---
> lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> 1 file changed, 74 insertions(+), 7 deletions(-)
>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index d961ccaf6..c3d6c4d78 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> uint16_t reserved; /**< Reserved. */
> }; /**< Hierarchical scheduler */
>
> +/**
> + * enum for the tx_offload bit-fields lenghts and offsets.
> + * defines the layout of rte_mbuf tx_offload field.
> + */
> +enum {
> + RTE_MBUF_L2_LEN_OFS = 0,
> + RTE_MBUF_L2_LEN_BITS = 7,
> + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> + RTE_MBUF_L3_LEN_BITS = 9,
> + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> + RTE_MBUF_L4_LEN_BITS = 8,
> + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> + RTE_MBUF_OUTL3_LEN_OFS =
> + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> + RTE_MBUF_OUTL3_LEN_BITS = 9,
> + RTE_MBUF_OUTL2_LEN_OFS =
> + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> + RTE_MBUF_OUTL2_LEN_BITS = 7,
> + RTE_MBUF_TXOFLD_UNUSED_OFS =
> + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> + RTE_MBUF_TXOFLD_UNUSED_BITS =
> + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> +};
> +
> /**
> * The generic rte_mbuf, containing a packet mbuf.
> */
> @@ -640,19 +665,24 @@ struct rte_mbuf {
> uint64_t tx_offload; /**< combined for easy fetch */
> __extension__
> struct {
> - uint64_t l2_len:7;
> + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> /**< L2 (MAC) Header Length for non-tunneling pkt.
> * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> */
> - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> + /**< L3 (IP) Header Length. */
> + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> + /**< L4 (TCP/UDP) Header Length. */
> + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> + /**< TCP TSO segment size */
>
> /* fields for TX offloading of tunnels */
> - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> + /**< Outer L3 (IP) Hdr Length. */
> + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> + /**< Outer L2 (MAC) Hdr Length. */
>
> - /* uint64_t unused:8; */
> + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> };
> };
>
> @@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> return 0;
> }
>
> +/*
> + * @warning
> + * @b EXPERIMENTAL: This API may change without prior notice.
> + *
> + * For given input values generate raw tx_offload value.
> + * Note that it is caller responsibility to make sure that input parameters
> + * don't exceed maximum bit-field values.
> + * @param il2
> + * l2_len value.
> + * @param il3
> + * l3_len value.
> + * @param il4
> + * l4_len value.
> + * @param tso
> + * tso_segsz value.
> + * @param ol3
> + * outer_l3_len value.
> + * @param ol2
> + * outer_l2_len value.
> + * @param unused
> + * unused value.
> + * @return
> + * raw tx_offload value.
> + */
> +static __rte_always_inline uint64_t
> +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> + uint64_t ol3, uint64_t ol2, uint64_t unused)
> +{
> + return il2 << RTE_MBUF_L2_LEN_OFS |
> + il3 << RTE_MBUF_L3_LEN_OFS |
> + il4 << RTE_MBUF_L4_LEN_OFS |
> + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> +}
> +
> /**
> * Validate general requirements for Tx offload in mbuf.
> *
> --
Thank you for the update.
I missed something important at the first review: will it work on any
endianess?
Olivier
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:22 ` Olivier Matz
@ 2019-04-01 13:55 ` Ananyev, Konstantin
2019-04-01 13:55 ` Ananyev, Konstantin
1 sibling, 1 reply; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-04-01 13:55 UTC (permalink / raw)
To: Olivier Matz; +Cc: dev, akhil.goyal
Hi Olivier,
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 74 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..c3d6c4d78 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/**
> > + * enum for the tx_offload bit-fields lenghts and offsets.
> > + * defines the layout of rte_mbuf tx_offload field.
> > + */
> > +enum {
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OUTL3_LEN_OFS =
> > + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OUTL3_LEN_BITS = 9,
> > + RTE_MBUF_OUTL2_LEN_OFS =
> > + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> > + RTE_MBUF_OUTL2_LEN_BITS = 7,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +665,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * Note that it is caller responsibility to make sure that input parameters
> > + * don't exceed maximum bit-field values.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @param unused
> > + * unused value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static __rte_always_inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2, uint64_t unused)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> > + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
>
>
> Thank you for the update.
>
> I missed something important at the first review: will it work on any
> endianess?
Good question, I think no (need different offsets for BE).
Will need to submit v6 here, I suppose.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value
2019-04-01 13:55 ` Ananyev, Konstantin
@ 2019-04-01 13:55 ` Ananyev, Konstantin
0 siblings, 0 replies; 163+ messages in thread
From: Ananyev, Konstantin @ 2019-04-01 13:55 UTC (permalink / raw)
To: Olivier Matz; +Cc: dev, akhil.goyal
Hi Olivier,
> > Operations to set/update bit-fields often cause compilers
> > to generate suboptimal code.
> > To help avoid such situation for tx_offload fields:
> > introduce new enum for tx_offload bit-fields lengths and offsets,
> > and new function to generate raw tx_offload value.
> >
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> > Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
> > ---
> > lib/librte_mbuf/rte_mbuf.h | 81 ++++++++++++++++++++++++++++++++++----
> > 1 file changed, 74 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> > index d961ccaf6..c3d6c4d78 100644
> > --- a/lib/librte_mbuf/rte_mbuf.h
> > +++ b/lib/librte_mbuf/rte_mbuf.h
> > @@ -479,6 +479,31 @@ struct rte_mbuf_sched {
> > uint16_t reserved; /**< Reserved. */
> > }; /**< Hierarchical scheduler */
> >
> > +/**
> > + * enum for the tx_offload bit-fields lenghts and offsets.
> > + * defines the layout of rte_mbuf tx_offload field.
> > + */
> > +enum {
> > + RTE_MBUF_L2_LEN_OFS = 0,
> > + RTE_MBUF_L2_LEN_BITS = 7,
> > + RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
> > + RTE_MBUF_L3_LEN_BITS = 9,
> > + RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
> > + RTE_MBUF_L4_LEN_BITS = 8,
> > + RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
> > + RTE_MBUF_TSO_SEGSZ_BITS = 16,
> > + RTE_MBUF_OUTL3_LEN_OFS =
> > + RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
> > + RTE_MBUF_OUTL3_LEN_BITS = 9,
> > + RTE_MBUF_OUTL2_LEN_OFS =
> > + RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
> > + RTE_MBUF_OUTL2_LEN_BITS = 7,
> > + RTE_MBUF_TXOFLD_UNUSED_OFS =
> > + RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
> > + RTE_MBUF_TXOFLD_UNUSED_BITS =
> > + sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_TXOFLD_UNUSED_OFS,
> > +};
> > +
> > /**
> > * The generic rte_mbuf, containing a packet mbuf.
> > */
> > @@ -640,19 +665,24 @@ struct rte_mbuf {
> > uint64_t tx_offload; /**< combined for easy fetch */
> > __extension__
> > struct {
> > - uint64_t l2_len:7;
> > + uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
> > /**< L2 (MAC) Header Length for non-tunneling pkt.
> > * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
> > */
> > - uint64_t l3_len:9; /**< L3 (IP) Header Length. */
> > - uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
> > - uint64_t tso_segsz:16; /**< TCP TSO segment size */
> > + uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
> > + /**< L3 (IP) Header Length. */
> > + uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
> > + /**< L4 (TCP/UDP) Header Length. */
> > + uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
> > + /**< TCP TSO segment size */
> >
> > /* fields for TX offloading of tunnels */
> > - uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
> > - uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
> > + uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
> > + /**< Outer L3 (IP) Hdr Length. */
> > + uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
> > + /**< Outer L2 (MAC) Hdr Length. */
> >
> > - /* uint64_t unused:8; */
> > + /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
> > };
> > };
> >
> > @@ -2243,6 +2273,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
> > return 0;
> > }
> >
> > +/*
> > + * @warning
> > + * @b EXPERIMENTAL: This API may change without prior notice.
> > + *
> > + * For given input values generate raw tx_offload value.
> > + * Note that it is caller responsibility to make sure that input parameters
> > + * don't exceed maximum bit-field values.
> > + * @param il2
> > + * l2_len value.
> > + * @param il3
> > + * l3_len value.
> > + * @param il4
> > + * l4_len value.
> > + * @param tso
> > + * tso_segsz value.
> > + * @param ol3
> > + * outer_l3_len value.
> > + * @param ol2
> > + * outer_l2_len value.
> > + * @param unused
> > + * unused value.
> > + * @return
> > + * raw tx_offload value.
> > + */
> > +static __rte_always_inline uint64_t
> > +rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
> > + uint64_t ol3, uint64_t ol2, uint64_t unused)
> > +{
> > + return il2 << RTE_MBUF_L2_LEN_OFS |
> > + il3 << RTE_MBUF_L3_LEN_OFS |
> > + il4 << RTE_MBUF_L4_LEN_OFS |
> > + tso << RTE_MBUF_TSO_SEGSZ_OFS |
> > + ol3 << RTE_MBUF_OUTL3_LEN_OFS |
> > + ol2 << RTE_MBUF_OUTL2_LEN_OFS |
> > + unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
> > +}
> > +
> > /**
> > * Validate general requirements for Tx offload in mbuf.
> > *
> > --
>
>
> Thank you for the update.
>
> I missed something important at the first review: will it work on any
> endianess?
Good question, I think no (need different offsets for BE).
Will need to submit v6 here, I suppose.
Konstantin
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (9 preceding siblings ...)
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
` (10 more replies)
10 siblings, 11 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v6 -> v5
Fix endianness issue spotted by Olivier
Added UT test-case for rte_mbuf_tx_offload()
v5 -> v4
Few cosmetic changes in mbuf tx_offload patch requested by Olivier
v4 -> v3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new enum for rte_mbuf
tx_offload layout and new function to set raw value for rte_mbuf
tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
app/test/test_mbuf.c | 101 ++-
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 101 ++-
12 files changed, 1487 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
` (9 subsequent siblings)
10 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
v6 -> v5
Fix endianness issue spotted by Olivier
Added UT test-case for rte_mbuf_tx_offload()
v5 -> v4
Few cosmetic changes in mbuf tx_offload patch requested by Olivier
v4 -> v3
Address Akhil comments:
- function/paramers names
- add ipsec lib into shared libraries list
v3 -> v2
Address comments:
- make rte_mbuf_tx_offload() always inline
- rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
Some code de-duplication
Rebase against the latest next-crypto
v1 -> v2
Add patch to improve a bit esp inbound pkt_process code
The first patch in this series introduces new enum for rte_mbuf
tx_offload layout and new function to set raw value for rte_mbuf
tx_offload.
Remaining ones are cleanups and small performance improvements
(on IA ~3-6% depending on the test-case) for ipsec library.
Konstantin Ananyev (9):
mbuf: new function to generate raw Tx offload value
ipsec: add Tx offload template into SA
ipsec: change the order in filling crypto op
ipsec: change the way unprocessed mbufs are accounted
ipsec: move inbound and outbound code into different files
ipsec: reorder packet check for esp inbound
ipsec: reorder packet process for esp inbound
ipsec: de-duplicate crypto op prepare code-path
doc: add ipsec lib into shared libraries list
app/test/test_mbuf.c | 101 ++-
doc/guides/rel_notes/release_19_05.rst | 1 +
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 34 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 +
lib/librte_ipsec/sa.c | 1072 ++----------------------
lib/librte_ipsec/sa.h | 45 +
lib/librte_mbuf/rte_mbuf.h | 101 ++-
12 files changed, 1487 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:49 ` Olivier Matz
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
` (8 subsequent siblings)
10 siblings, 2 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Add new test-case into UT for introduced function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
app/test/test_mbuf.c | 101 ++++++++++++++++++++++++++++++++++++-
lib/librte_mbuf/rte_mbuf.h | 101 ++++++++++++++++++++++++++++++++++---
2 files changed, 194 insertions(+), 8 deletions(-)
diff --git a/app/test/test_mbuf.c b/app/test/test_mbuf.c
index 9e82a20be..030385ec5 100644
--- a/app/test/test_mbuf.c
+++ b/app/test/test_mbuf.c
@@ -27,6 +27,7 @@
#include <rte_mbuf.h>
#include <rte_random.h>
#include <rte_cycles.h>
+#include <rte_malloc.h>
#include "test.h"
@@ -1028,6 +1029,99 @@ test_mbuf_linearize_check(struct rte_mempool *pktmbuf_pool)
return 0;
}
+/*
+ * Helper function for test_tx_ofload
+ */
+static inline void
+set_tx_offload(struct rte_mbuf *mb, uint64_t il2, uint64_t il3, uint64_t il4,
+ uint64_t tso, uint64_t ol3, uint64_t ol2)
+{
+ mb->l2_len = il2;
+ mb->l3_len = il3;
+ mb->l4_len = il4;
+ mb->tso_segsz = tso;
+ mb->outer_l3_len = ol3;
+ mb->outer_l2_len = ol2;
+}
+
+static int
+test_tx_offload(void)
+{
+ struct rte_mbuf *mb;
+ uint64_t tm, v1, v2;
+ size_t sz;
+ uint32_t i;
+
+ static volatile struct {
+ uint16_t l2;
+ uint16_t l3;
+ uint16_t l4;
+ uint16_t tso;
+ } txof;
+
+ const uint32_t num = 0x10000;
+
+ txof.l2 = rte_rand() % (1 << RTE_MBUF_L2_LEN_BITS);
+ txof.l3 = rte_rand() % (1 << RTE_MBUF_L3_LEN_BITS);
+ txof.l4 = rte_rand() % (1 << RTE_MBUF_L4_LEN_BITS);
+ txof.tso = rte_rand() % (1 << RTE_MBUF_TSO_SEGSZ_BITS);
+
+ printf("%s started, tx_offload = {\n"
+ "\tl2_len=%#hx,\n"
+ "\tl3_len=%#hx,\n"
+ "\tl4_len=%#hx,\n"
+ "\ttso_segsz=%#hx,\n"
+ "\touter_l3_len=%#x,\n"
+ "\touter_l2_len=%#x,\n"
+ "};\n",
+ __func__,
+ txof.l2, txof.l3, txof.l4, txof.tso, txof.l3, txof.l2);
+
+ sz = sizeof(*mb) * num;
+ mb = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (mb == NULL) {
+ printf("%s failed, out of memory\n", __func__);
+ return -ENOMEM;
+ }
+
+ memset(mb, 0, sz);
+ tm = rte_rdtsc_precise();
+
+ for (i = 0; i != num; i++)
+ set_tx_offload(mb + i, txof.l2, txof.l3, txof.l4,
+ txof.tso, txof.l3, txof.l2);
+
+ tm = rte_rdtsc_precise() - tm;
+ printf("%s set tx_offload by bit-fields: %u iterations, %"
+ PRIu64 " cycles, %#Lf cycles/iter\n",
+ __func__, num, tm, (long double)tm / num);
+
+ v1 = mb[rte_rand() % num].tx_offload;
+
+ memset(mb, 0, sz);
+ tm = rte_rdtsc_precise();
+
+ for (i = 0; i != num; i++)
+ mb[i].tx_offload = rte_mbuf_tx_offload(txof.l2, txof.l3,
+ txof.l4, txof.tso, txof.l3, txof.l2, 0);
+
+ tm = rte_rdtsc_precise() - tm;
+ printf("%s set raw tx_offload: %u iterations, %"
+ PRIu64 " cycles, %#Lf cycles/iter\n",
+ __func__, num, tm, (long double)tm / num);
+
+ v2 = mb[rte_rand() % num].tx_offload;
+
+ rte_free(mb);
+
+ printf("%s finished\n"
+ "expected tx_offload value: 0x%" PRIx64 ";\n"
+ "rte_mbuf_tx_offload value: 0x%" PRIx64 ";\n",
+ __func__, v1, v2);
+
+ return (v1 == v2) ? 0 : -EINVAL;
+}
+
static int
test_mbuf(void)
{
@@ -1126,8 +1220,13 @@ test_mbuf(void)
printf("test_mbuf_linearize_check() failed\n");
goto err;
}
- ret = 0;
+ if (test_tx_offload() < 0) {
+ printf("test_tx_offload() failed\n");
+ goto err;
+ }
+
+ ret = 0;
err:
rte_mempool_free(pktmbuf_pool);
rte_mempool_free(pktmbuf_pool2);
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..d7439e889 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -40,6 +40,7 @@
#include <rte_atomic.h>
#include <rte_prefetch.h>
#include <rte_branch_prediction.h>
+#include <rte_byteorder.h>
#include <rte_mbuf_ptype.h>
#ifdef __cplusplus
@@ -479,6 +480,50 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_TXOFLD_UNUSED_BITS = sizeof(uint64_t) * CHAR_BIT -
+ RTE_MBUF_L2_LEN_BITS -
+ RTE_MBUF_L3_LEN_BITS -
+ RTE_MBUF_L4_LEN_BITS -
+ RTE_MBUF_TSO_SEGSZ_BITS -
+ RTE_MBUF_OUTL3_LEN_BITS -
+ RTE_MBUF_OUTL2_LEN_BITS,
+#if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
+ RTE_MBUF_L2_LEN_OFS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_L2_LEN_BITS
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS - RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS - RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS - RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS - RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS - RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS - RTE_MBUF_TXOFLD_UNUSED_BITS,
+#else
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+#endif
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +685,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2293,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * Note that it is caller responsibility to make sure that input parameters
+ * don't exceed maximum bit-field values.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:49 ` Olivier Matz
1 sibling, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code.
To help avoid such situation for tx_offload fields:
introduce new enum for tx_offload bit-fields lengths and offsets,
and new function to generate raw tx_offload value.
Add new test-case into UT for introduced function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
app/test/test_mbuf.c | 101 ++++++++++++++++++++++++++++++++++++-
lib/librte_mbuf/rte_mbuf.h | 101 ++++++++++++++++++++++++++++++++++---
2 files changed, 194 insertions(+), 8 deletions(-)
diff --git a/app/test/test_mbuf.c b/app/test/test_mbuf.c
index 9e82a20be..030385ec5 100644
--- a/app/test/test_mbuf.c
+++ b/app/test/test_mbuf.c
@@ -27,6 +27,7 @@
#include <rte_mbuf.h>
#include <rte_random.h>
#include <rte_cycles.h>
+#include <rte_malloc.h>
#include "test.h"
@@ -1028,6 +1029,99 @@ test_mbuf_linearize_check(struct rte_mempool *pktmbuf_pool)
return 0;
}
+/*
+ * Helper function for test_tx_ofload
+ */
+static inline void
+set_tx_offload(struct rte_mbuf *mb, uint64_t il2, uint64_t il3, uint64_t il4,
+ uint64_t tso, uint64_t ol3, uint64_t ol2)
+{
+ mb->l2_len = il2;
+ mb->l3_len = il3;
+ mb->l4_len = il4;
+ mb->tso_segsz = tso;
+ mb->outer_l3_len = ol3;
+ mb->outer_l2_len = ol2;
+}
+
+static int
+test_tx_offload(void)
+{
+ struct rte_mbuf *mb;
+ uint64_t tm, v1, v2;
+ size_t sz;
+ uint32_t i;
+
+ static volatile struct {
+ uint16_t l2;
+ uint16_t l3;
+ uint16_t l4;
+ uint16_t tso;
+ } txof;
+
+ const uint32_t num = 0x10000;
+
+ txof.l2 = rte_rand() % (1 << RTE_MBUF_L2_LEN_BITS);
+ txof.l3 = rte_rand() % (1 << RTE_MBUF_L3_LEN_BITS);
+ txof.l4 = rte_rand() % (1 << RTE_MBUF_L4_LEN_BITS);
+ txof.tso = rte_rand() % (1 << RTE_MBUF_TSO_SEGSZ_BITS);
+
+ printf("%s started, tx_offload = {\n"
+ "\tl2_len=%#hx,\n"
+ "\tl3_len=%#hx,\n"
+ "\tl4_len=%#hx,\n"
+ "\ttso_segsz=%#hx,\n"
+ "\touter_l3_len=%#x,\n"
+ "\touter_l2_len=%#x,\n"
+ "};\n",
+ __func__,
+ txof.l2, txof.l3, txof.l4, txof.tso, txof.l3, txof.l2);
+
+ sz = sizeof(*mb) * num;
+ mb = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (mb == NULL) {
+ printf("%s failed, out of memory\n", __func__);
+ return -ENOMEM;
+ }
+
+ memset(mb, 0, sz);
+ tm = rte_rdtsc_precise();
+
+ for (i = 0; i != num; i++)
+ set_tx_offload(mb + i, txof.l2, txof.l3, txof.l4,
+ txof.tso, txof.l3, txof.l2);
+
+ tm = rte_rdtsc_precise() - tm;
+ printf("%s set tx_offload by bit-fields: %u iterations, %"
+ PRIu64 " cycles, %#Lf cycles/iter\n",
+ __func__, num, tm, (long double)tm / num);
+
+ v1 = mb[rte_rand() % num].tx_offload;
+
+ memset(mb, 0, sz);
+ tm = rte_rdtsc_precise();
+
+ for (i = 0; i != num; i++)
+ mb[i].tx_offload = rte_mbuf_tx_offload(txof.l2, txof.l3,
+ txof.l4, txof.tso, txof.l3, txof.l2, 0);
+
+ tm = rte_rdtsc_precise() - tm;
+ printf("%s set raw tx_offload: %u iterations, %"
+ PRIu64 " cycles, %#Lf cycles/iter\n",
+ __func__, num, tm, (long double)tm / num);
+
+ v2 = mb[rte_rand() % num].tx_offload;
+
+ rte_free(mb);
+
+ printf("%s finished\n"
+ "expected tx_offload value: 0x%" PRIx64 ";\n"
+ "rte_mbuf_tx_offload value: 0x%" PRIx64 ";\n",
+ __func__, v1, v2);
+
+ return (v1 == v2) ? 0 : -EINVAL;
+}
+
static int
test_mbuf(void)
{
@@ -1126,8 +1220,13 @@ test_mbuf(void)
printf("test_mbuf_linearize_check() failed\n");
goto err;
}
- ret = 0;
+ if (test_tx_offload() < 0) {
+ printf("test_tx_offload() failed\n");
+ goto err;
+ }
+
+ ret = 0;
err:
rte_mempool_free(pktmbuf_pool);
rte_mempool_free(pktmbuf_pool2);
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index d961ccaf6..d7439e889 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -40,6 +40,7 @@
#include <rte_atomic.h>
#include <rte_prefetch.h>
#include <rte_branch_prediction.h>
+#include <rte_byteorder.h>
#include <rte_mbuf_ptype.h>
#ifdef __cplusplus
@@ -479,6 +480,50 @@ struct rte_mbuf_sched {
uint16_t reserved; /**< Reserved. */
}; /**< Hierarchical scheduler */
+/**
+ * enum for the tx_offload bit-fields lenghts and offsets.
+ * defines the layout of rte_mbuf tx_offload field.
+ */
+enum {
+ RTE_MBUF_L2_LEN_BITS = 7,
+ RTE_MBUF_L3_LEN_BITS = 9,
+ RTE_MBUF_L4_LEN_BITS = 8,
+ RTE_MBUF_TSO_SEGSZ_BITS = 16,
+ RTE_MBUF_OUTL3_LEN_BITS = 9,
+ RTE_MBUF_OUTL2_LEN_BITS = 7,
+ RTE_MBUF_TXOFLD_UNUSED_BITS = sizeof(uint64_t) * CHAR_BIT -
+ RTE_MBUF_L2_LEN_BITS -
+ RTE_MBUF_L3_LEN_BITS -
+ RTE_MBUF_L4_LEN_BITS -
+ RTE_MBUF_TSO_SEGSZ_BITS -
+ RTE_MBUF_OUTL3_LEN_BITS -
+ RTE_MBUF_OUTL2_LEN_BITS,
+#if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
+ RTE_MBUF_L2_LEN_OFS =
+ sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_L2_LEN_BITS
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS - RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS - RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS - RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS - RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS - RTE_MBUF_OUTL2_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS - RTE_MBUF_TXOFLD_UNUSED_BITS,
+#else
+ RTE_MBUF_L2_LEN_OFS = 0,
+ RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
+ RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
+ RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
+ RTE_MBUF_OUTL3_LEN_OFS =
+ RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
+ RTE_MBUF_OUTL2_LEN_OFS =
+ RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
+ RTE_MBUF_TXOFLD_UNUSED_OFS =
+ RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
+#endif
+};
+
/**
* The generic rte_mbuf, containing a packet mbuf.
*/
@@ -640,19 +685,24 @@ struct rte_mbuf {
uint64_t tx_offload; /**< combined for easy fetch */
__extension__
struct {
- uint64_t l2_len:7;
+ uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
/**< L2 (MAC) Header Length for non-tunneling pkt.
* Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
*/
- uint64_t l3_len:9; /**< L3 (IP) Header Length. */
- uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
- uint64_t tso_segsz:16; /**< TCP TSO segment size */
+ uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
+ /**< L3 (IP) Header Length. */
+ uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
+ /**< L4 (TCP/UDP) Header Length. */
+ uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
+ /**< TCP TSO segment size */
/* fields for TX offloading of tunnels */
- uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
- uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
+ uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
+ /**< Outer L3 (IP) Hdr Length. */
+ uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
+ /**< Outer L2 (MAC) Hdr Length. */
- /* uint64_t unused:8; */
+ /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
};
};
@@ -2243,6 +2293,43 @@ static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail
return 0;
}
+/*
+ * @warning
+ * @b EXPERIMENTAL: This API may change without prior notice.
+ *
+ * For given input values generate raw tx_offload value.
+ * Note that it is caller responsibility to make sure that input parameters
+ * don't exceed maximum bit-field values.
+ * @param il2
+ * l2_len value.
+ * @param il3
+ * l3_len value.
+ * @param il4
+ * l4_len value.
+ * @param tso
+ * tso_segsz value.
+ * @param ol3
+ * outer_l3_len value.
+ * @param ol2
+ * outer_l2_len value.
+ * @param unused
+ * unused value.
+ * @return
+ * raw tx_offload value.
+ */
+static __rte_always_inline uint64_t
+rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
+ uint64_t ol3, uint64_t ol2, uint64_t unused)
+{
+ return il2 << RTE_MBUF_L2_LEN_OFS |
+ il3 << RTE_MBUF_L3_LEN_OFS |
+ il4 << RTE_MBUF_L4_LEN_OFS |
+ tso << RTE_MBUF_TSO_SEGSZ_OFS |
+ ol3 << RTE_MBUF_OUTL3_LEN_OFS |
+ ol2 << RTE_MBUF_OUTL2_LEN_OFS |
+ unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
+}
+
/**
* Validate general requirements for Tx offload in mbuf.
*
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
` (7 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Operations to set/update bit-fields often cause compilers
to generate suboptimal code. To avoid such negative effect,
use tx_offload raw value and mask to update l2_len and l3_len
fields within mbufs.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 23 ++++++++++++++++++-----
lib/librte_ipsec/sa.h | 5 +++++
2 files changed, 23 insertions(+), 5 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 2eb6bae07..747c37002 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -14,6 +14,9 @@
#include "iph.h"
#include "pad.h"
+#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
+#define MBUF_MAX_L3_LEN RTE_LEN2MASK(RTE_MBUF_L3_LEN_BITS, uint64_t)
+
/* some helper structures */
struct crypto_xform {
struct rte_crypto_auth_xform *auth;
@@ -254,6 +257,11 @@ esp_outb_tun_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm)
sa->proto = prm->tun.next_proto;
sa->hdr_len = prm->tun.hdr_len;
sa->hdr_l3_off = prm->tun.hdr_l3_off;
+
+ /* update l2_len and l3_len fields for outbound mbuf */
+ sa->tx_offload.val = rte_mbuf_tx_offload(sa->hdr_l3_off,
+ sa->hdr_len - sa->hdr_l3_off, 0, 0, 0, 0, 0);
+
memcpy(sa->hdr, prm->tun.hdr, sa->hdr_len);
esp_outb_init(sa, sa->hdr_len);
@@ -324,6 +332,11 @@ esp_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
sa->spi = rte_cpu_to_be_32(prm->ipsec_xform.spi);
sa->salt = prm->ipsec_xform.salt;
+ /* preserve all values except l2_len and l3_len */
+ sa->tx_offload.msk =
+ ~rte_mbuf_tx_offload(MBUF_MAX_L2_LEN, MBUF_MAX_L3_LEN,
+ 0, 0, 0, 0, 0);
+
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -549,7 +562,7 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
/* size of ipsec protected data */
l2len = mb->l2_len;
- plen = mb->pkt_len - mb->l2_len;
+ plen = mb->pkt_len - l2len;
/* number of bytes to encrypt */
clen = plen + sizeof(*espt);
@@ -576,8 +589,8 @@ esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
/* update pkt l2/l3 len */
- mb->l2_len = sa->hdr_l3_off;
- mb->l3_len = sa->hdr_len - sa->hdr_l3_off;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* copy tunnel pkt header */
rte_memcpy(ph, sa->hdr, sa->hdr_len);
@@ -1124,8 +1137,8 @@ esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->l2_len = 0;
- mb->l3_len = 0;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index c3a0d84bc..48c3c103a 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -87,6 +87,11 @@ struct rte_ipsec_sa {
union sym_op_ofslen cipher;
union sym_op_ofslen auth;
} ctp;
+ /* tx_offload template for tunnel mbuf */
+ struct {
+ uint64_t msk;
+ uint64_t val;
+ } tx_offload;
uint32_t salt;
uint8_t algo_type;
uint8_t proto; /* next proto */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (2 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
` (6 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now we first fill crypto_sym_op part of crypto_op,
then in a separate cycle we fill crypto op fields.
It makes more sense to fill whole crypto-op in one go instead.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 46 ++++++++++++++++++++-----------------------
1 file changed, 21 insertions(+), 25 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 747c37002..97c0f8c61 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -464,20 +464,17 @@ mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
* setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
*/
static inline void
-lksd_none_cop_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], struct rte_crypto_op *cop[], uint16_t num)
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
{
- uint32_t i;
struct rte_crypto_sym_op *sop;
- for (i = 0; i != num; i++) {
- sop = cop[i]->sym;
- cop[i]->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop[i]->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop[i]->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb[i];
- __rte_crypto_sym_op_attach_sym_session(sop, ss->crypto.ses);
- }
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
}
/*
@@ -667,11 +664,13 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -689,10 +688,10 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -700,9 +699,6 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -803,11 +799,13 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint64_t sqn;
rte_be64_t sqc;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
n = num;
sqn = esn_outb_update_sqn(sa, &n);
@@ -829,10 +827,10 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* success, setup crypto op */
if (rc >= 0) {
- mb[k] = mb[i];
outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
+ mb[k++] = mb[i];
/* failure, put packet into the death-row */
} else {
dr[i - k] = mb[i];
@@ -840,9 +838,6 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
}
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
mbuf_bulk_copy(mb + k, dr, n - k);
@@ -1021,11 +1016,13 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
int32_t rc;
uint32_t i, k, hl;
struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
struct rte_mbuf *dr[num];
sa = ss->sa;
+ cs = ss->crypto.ses;
rsn = rsn_acquire(sa);
k = 0;
@@ -1033,9 +1030,11 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl = mb[i]->l2_len + mb[i]->l3_len;
rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0)
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
hl, rc);
+ }
if (rc == 0)
mb[k++] = mb[i];
@@ -1047,9 +1046,6 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rsn_release(sa, rsn);
- /* update cops */
- lksd_none_cop_prepare(ss, mb, cop, k);
-
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
mbuf_bulk_copy(mb + k, dr, num - k);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (3 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
` (5 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
As was pointed in one of previous reviews - we can avoid updating
contents of mbuf array for successfully processed packets.
Instead store indexes of failed packets, to move them beyond the good
ones later.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/sa.c | 164 +++++++++++++++++++++++-------------------
1 file changed, 92 insertions(+), 72 deletions(-)
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index 97c0f8c61..e728c6a28 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -450,14 +450,31 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
static inline void
-mbuf_bulk_copy(struct rte_mbuf *dst[], struct rte_mbuf * const src[],
- uint32_t num)
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
{
- uint32_t i;
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
- for (i = 0; i != num; i++)
- dst[i] = src[i];
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
}
/*
@@ -667,7 +684,7 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -691,17 +708,17 @@ outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -802,7 +819,7 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -830,17 +847,17 @@ outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
outb_pkt_xprepare(sa, sqc, &icv);
lksd_none_cop_prepare(cop[k], cs, mb[i]);
esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- mb[k++] = mb[i];
+ k++;
/* failure, put packet into the death-row */
} else {
- dr[i - k] = mb[i];
+ dr[i - k] = i;
rte_errno = -rc;
}
}
/* copy not prepared mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
return k;
}
@@ -1019,7 +1036,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
struct rte_cryptodev_sym_session *cs;
struct replay_sqn *rsn;
union sym_op_data icv;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
cs = ss->crypto.ses;
@@ -1036,10 +1053,9 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
hl, rc);
}
- if (rc == 0)
- mb[k++] = mb[i];
- else {
- dr[i - k] = mb[i];
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
@@ -1048,7 +1064,7 @@ inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
/* copy not prepared mbufs beyond good ones */
if (k != num && k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
return k;
}
@@ -1200,7 +1216,7 @@ esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
*/
static inline uint16_t
esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- struct rte_mbuf *mb[], struct rte_mbuf *dr[], uint16_t num)
+ uint32_t dr[], uint16_t num)
{
uint32_t i, k;
struct replay_sqn *rsn;
@@ -1210,9 +1226,9 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
k = 0;
for (i = 0; i != num; i++) {
if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
rsn_update_finish(sa, rsn);
@@ -1226,10 +1242,10 @@ static uint16_t
inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1239,23 +1255,27 @@ inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1265,10 +1285,10 @@ static uint16_t
inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
- uint32_t i, k;
+ uint32_t i, k, n;
uint32_t sqn[num];
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1278,23 +1298,27 @@ inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
for (i = 0; i != num; i++) {
/* good packet */
if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- mb[k++] = mb[i];
+ k++;
/* bad packet, will drop from furhter processing */
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
- /* update seq # and replay winow */
- k = esp_inb_rsn_update(sa, sqn, mb, dr + i - k, k);
-
/* handle unprocessed mbufs */
- if (k != num) {
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
rte_errno = EBADMSG;
- if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
- }
- return k;
+ return n;
}
/*
@@ -1310,7 +1334,7 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint32_t i, k, icv_len, *icv;
struct rte_mbuf *ml;
struct rte_ipsec_sa *sa;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1323,16 +1347,16 @@ outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
icv = rte_pktmbuf_mtod_offset(ml, void *,
ml->data_len - icv_len);
remove_sqh(icv, icv_len);
- mb[k++] = mb[i];
+ k++;
} else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1352,23 +1376,23 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
uint16_t num)
{
uint32_t i, k;
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
RTE_SET_USED(ss);
k = 0;
for (i = 0; i != num; i++) {
if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0)
- mb[k++] = mb[i];
+ k++;
else
- dr[i - k] = mb[i];
+ dr[i - k] = i;
}
/* handle unprocessed mbufs */
if (k != num) {
rte_errno = EBADMSG;
if (k != 0)
- mbuf_bulk_copy(mb + k, dr, num - k);
+ move_bad_mbufs(mb, dr, num, num - k);
}
return k;
@@ -1409,7 +1433,7 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1427,22 +1451,20 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
/* try to update the packet itself */
rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
@@ -1461,7 +1483,7 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_ipsec_sa *sa;
union sym_op_data icv;
uint64_t iv[IPSEC_MAX_IV_QWORD];
- struct rte_mbuf *dr[num];
+ uint32_t dr[num];
sa = ss->sa;
@@ -1483,22 +1505,20 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
l2, l3, &icv);
- /* success, update mbuf fields */
- if (rc >= 0)
- mb[k++] = mb[i];
+ k += (rc >= 0);
+
/* failure, put packet into the death-row */
- else {
- dr[i - k] = mb[i];
+ if (rc < 0) {
+ dr[i - k] = i;
rte_errno = -rc;
}
}
- inline_outb_mbuf_prepare(ss, mb, k);
-
/* copy not processed mbufs beyond good ones */
if (k != n && k != 0)
- mbuf_bulk_copy(mb + k, dr, n - k);
+ move_bad_mbufs(mb, dr, n, n - k);
+ inline_outb_mbuf_prepare(ss, mb, k);
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (4 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
` (4 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
sa.c becomes too big, so decided to split it into 3 chunks:
- sa.c - control path related functions (init/fini, etc.)
- esp_inb.c - ESP inbound packet processing
- esp_outb.c - ESP outbound packet processing
Plus few changes in internal function names to follow the same
code convention.
No functional changes introduced.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/Makefile | 2 +
lib/librte_ipsec/crypto.h | 17 +
lib/librte_ipsec/esp_inb.c | 439 ++++++++++++++
lib/librte_ipsec/esp_outb.c | 559 ++++++++++++++++++
lib/librte_ipsec/ipsec_sqn.h | 30 -
lib/librte_ipsec/meson.build | 2 +-
lib/librte_ipsec/misc.h | 41 ++
lib/librte_ipsec/sa.c | 1067 ++--------------------------------
lib/librte_ipsec/sa.h | 40 ++
9 files changed, 1141 insertions(+), 1056 deletions(-)
create mode 100644 lib/librte_ipsec/esp_inb.c
create mode 100644 lib/librte_ipsec/esp_outb.c
create mode 100644 lib/librte_ipsec/misc.h
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index 77506d6ad..e80926baa 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -16,6 +16,8 @@ EXPORT_MAP := rte_ipsec_version.map
LIBABIVER := 1
# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_inb.c
+SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += esp_outb.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += sa.c
SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += ses.c
diff --git a/lib/librte_ipsec/crypto.h b/lib/librte_ipsec/crypto.h
index 4f551e39c..3e9a8f41c 100644
--- a/lib/librte_ipsec/crypto.h
+++ b/lib/librte_ipsec/crypto.h
@@ -162,4 +162,21 @@ remove_sqh(void *picv, uint32_t icv_len)
icv[i] = icv[i + 1];
}
+/*
+ * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
+ */
+static inline void
+lksd_none_cop_prepare(struct rte_crypto_op *cop,
+ struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
+{
+ struct rte_crypto_sym_op *sop;
+
+ sop = cop->sym;
+ cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+ cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
+ sop->m_src = mb;
+ __rte_crypto_sym_op_attach_sym_session(sop, cs);
+}
+
#endif /* _CRYPTO_H_ */
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
new file mode 100644
index 000000000..a775c7b0b
--- /dev/null
+++ b/lib/librte_ipsec/esp_inb.c
@@ -0,0 +1,439 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ */
+static inline int32_t
+inb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint64_t *ivc, *ivp;
+ uint32_t algo, clen;
+
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EINVAL;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_GCM:
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = clen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CBC:
+ case ALGO_TYPE_3DES_CBC:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ copy_iv(ivc, ivp, sa->iv_len);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ /* copy iv from the input packet to the cop */
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_NULL:
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = clen;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0)
+ insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
+
+ /*
+ * fill AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM.
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP inbound tunnel case.
+ */
+static inline int32_t
+inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
+ struct rte_mbuf *mb, uint32_t hlen, union sym_op_data *icv)
+{
+ int32_t rc;
+ uint64_t sqn;
+ uint32_t icv_ofs, plen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+
+ /*
+ * retrieve and reconstruct SQN, then check it, then
+ * convert it back into network byte order.
+ */
+ sqn = rte_be_to_cpu_32(esph->seq);
+ if (IS_ESN(sa))
+ sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
+
+ rc = esn_inb_check_sqn(rsn, sa, sqn);
+ if (rc != 0)
+ return rc;
+
+ sqn = rte_cpu_to_be_64(sqn);
+
+ /* start packet manipulation */
+ plen = mb->pkt_len;
+ plen = plen - hlen;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+
+ /* we have to allocate space for AAD somewhere,
+ * right now - just use free trailing space at the last segment.
+ * Would probably be more convenient to reserve space for AAD
+ * inside rte_crypto_op itself
+ * (again for IV space is already reserved inside cop).
+ */
+ if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
+
+ inb_pkt_xprepare(sa, sqn, icv);
+ return plen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP inbound case.
+ */
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, hl;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ struct replay_sqn *rsn;
+ union sym_op_data icv;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+ rsn = rsn_acquire(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ hl = mb[i]->l2_len + mb[i]->l3_len;
+ rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
+ if (rc >= 0) {
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ }
+
+ k += (rc == 0);
+ if (rc != 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ rsn_release(sa, rsn);
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ return k;
+}
+
+/*
+ * process ESP inbound tunnel packet.
+ */
+static inline int
+inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /*
+ * check padding and next proto.
+ * return an error if something is wrong.
+ */
+ pd = (char *)espt - espt->pad_len;
+ if (espt->next_proto != sa->proto ||
+ memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* cut of L2/L3 headers, ESP header and IV */
+ hlen = mb->l2_len + mb->l3_len;
+ esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
+ rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+
+ /* retrieve SQN for later check */
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* reset mbuf metatdata: L2/L3 len, packet type */
+ mb->packet_type = RTE_PTYPE_UNKNOWN;
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * process ESP inbound transport packet.
+ */
+static inline int
+inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
+ uint32_t *sqn)
+{
+ uint32_t hlen, icv_len, l2len, l3len, tlen;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ struct rte_mbuf *ml;
+ char *np, *op, *pd;
+
+ if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
+ return -EBADMSG;
+
+ icv_len = sa->icv_len;
+
+ ml = rte_pktmbuf_lastseg(mb);
+ espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
+ ml->data_len - icv_len - sizeof(*espt));
+
+ /* check padding, return an error if something is wrong. */
+ pd = (char *)espt - espt->pad_len;
+ if (memcmp(pd, esp_pad_bytes, espt->pad_len))
+ return -EINVAL;
+
+ /* cut of ICV, ESP tail and padding bytes */
+ tlen = icv_len + sizeof(*espt) + espt->pad_len;
+ ml->data_len -= tlen;
+ mb->pkt_len -= tlen;
+
+ /* retrieve SQN for later check */
+ l2len = mb->l2_len;
+ l3len = mb->l3_len;
+ hlen = l2len + l3len;
+ op = rte_pktmbuf_mtod(mb, char *);
+ esph = (struct esp_hdr *)(op + hlen);
+ *sqn = rte_be_to_cpu_32(esph->seq);
+
+ /* cut off ESP header and IV, update L3 header */
+ np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
+ remove_esph(np, op, hlen);
+ update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
+ espt->next_proto);
+
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
+ return 0;
+}
+
+/*
+ * for group of ESP inbound packets perform SQN check and update.
+ */
+static inline uint16_t
+esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
+ uint32_t dr[], uint16_t num)
+{
+ uint32_t i, k;
+ struct replay_sqn *rsn;
+
+ rsn = rsn_update_start(sa);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ k++;
+ else
+ dr[i - k] = i;
+ }
+
+ rsn_update_finish(sa, rsn);
+ return k;
+}
+
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ struct rte_ipsec_sa *sa;
+ uint32_t sqn[num];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
+
+/*
+ * process group of ESP inbound transport packets.
+ */
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, k, n;
+ uint32_t sqn[num];
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ /* process packets, extract seq numbers */
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+ /* good packet */
+ if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
+ k++;
+ /* bad packet, will drop from furhter processing */
+ else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num && k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+
+ /* update SQN and replay winow */
+ n = esp_inb_rsn_update(sa, sqn, dr, k);
+
+ /* handle mbufs with wrong SQN */
+ if (n != k && n != 0)
+ move_bad_mbufs(mb, dr, k, k - n);
+
+ if (n != num)
+ rte_errno = EBADMSG;
+
+ return n;
+}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
new file mode 100644
index 000000000..09bfb8658
--- /dev/null
+++ b/lib/librte_ipsec/esp_outb.c
@@ -0,0 +1,559 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ipsec.h>
+#include <rte_esp.h>
+#include <rte_ip.h>
+#include <rte_errno.h>
+#include <rte_cryptodev.h>
+
+#include "sa.h"
+#include "ipsec_sqn.h"
+#include "crypto.h"
+#include "iph.h"
+#include "misc.h"
+#include "pad.h"
+
+/*
+ * setup crypto op and crypto sym op for ESP outbound packet.
+ */
+static inline void
+outb_cop_prepare(struct rte_crypto_op *cop,
+ const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
+ const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
+{
+ struct rte_crypto_sym_op *sop;
+ struct aead_gcm_iv *gcm;
+ struct aesctr_cnt_blk *ctr;
+ uint32_t algo;
+
+ algo = sa->algo_type;
+
+ /* fill sym op fields */
+ sop = cop->sym;
+
+ switch (algo) {
+ case ALGO_TYPE_AES_CBC:
+ /* Cipher-Auth (AES-CBC *) case */
+ case ALGO_TYPE_3DES_CBC:
+ /* Cipher-Auth (3DES-CBC *) case */
+ case ALGO_TYPE_NULL:
+ /* NULL case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+ break;
+ case ALGO_TYPE_AES_GCM:
+ /* AEAD (AES_GCM) case */
+ sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+
+ /* fill AAD IV (located inside crypto op) */
+ gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
+ sa->iv_ofs);
+ aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
+ break;
+ case ALGO_TYPE_AES_CTR:
+ /* Cipher-Auth (AES-CTR *) case */
+ sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + hlen;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+
+ ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
+ sa->iv_ofs);
+ aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
+ break;
+ }
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound tunnel case.
+ */
+static inline int32_t
+outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ union sym_op_data *icv)
+{
+ uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ /* calculate extra header space required */
+ hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
+
+ /* size of ipsec protected data */
+ l2len = mb->l2_len;
+ plen = mb->pkt_len - l2len;
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and prepend */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend header */
+ ph = rte_pktmbuf_prepend(mb, hlen - l2len);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* update pkt l2/l3 len */
+ mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
+ sa->tx_offload.val;
+
+ /* copy tunnel pkt header */
+ rte_memcpy(ph, sa->hdr, sa->hdr_len);
+
+ /* update original and new ip header fields */
+ update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
+ sqn_low16(sqc));
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + sa->hdr_len);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = sa->proto;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * for pure cryptodev (lookaside none) depending on SA settings,
+ * we might have to write some extra data to the packet.
+ */
+static inline void
+outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const union sym_op_data *icv)
+{
+ uint32_t *psqh;
+ struct aead_gcm_aad *aad;
+
+ /* insert SQN.hi between ESP trailer and ICV */
+ if (sa->sqh_len != 0) {
+ psqh = (uint32_t *)(icv->va - sa->sqh_len);
+ psqh[0] = sqn_hi32(sqc);
+ }
+
+ /*
+ * fill IV and AAD fields, if any (aad fields are placed after icv),
+ * right now we support only one AEAD algorithm: AES-GCM .
+ */
+ if (sa->aad_len != 0) {
+ aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
+ aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
+ }
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound tunnel case.
+ */
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * setup/update packet data and metadata for ESP outbound transport case.
+ */
+static inline int32_t
+outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
+ const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
+ uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
+{
+ uint8_t np;
+ uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
+ struct rte_mbuf *ml;
+ struct esp_hdr *esph;
+ struct esp_tail *espt;
+ char *ph, *pt;
+ uint64_t *iv;
+
+ uhlen = l2len + l3len;
+ plen = mb->pkt_len - uhlen;
+
+ /* calculate extra header space required */
+ hlen = sa->iv_len + sizeof(*esph);
+
+ /* number of bytes to encrypt */
+ clen = plen + sizeof(*espt);
+ clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
+
+ /* pad length + esp tail */
+ pdlen = clen - plen;
+ tlen = pdlen + sa->icv_len;
+
+ /* do append and insert */
+ ml = rte_pktmbuf_lastseg(mb);
+ if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
+ return -ENOSPC;
+
+ /* prepend space for ESP header */
+ ph = rte_pktmbuf_prepend(mb, hlen);
+ if (ph == NULL)
+ return -ENOSPC;
+
+ /* append tail */
+ pdofs = ml->data_len;
+ ml->data_len += tlen;
+ mb->pkt_len += tlen;
+ pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
+
+ /* shift L2/L3 headers */
+ insert_esph(ph, ph + hlen, uhlen);
+
+ /* update ip header fields */
+ np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
+ IPPROTO_ESP);
+
+ /* update spi, seqn and iv */
+ esph = (struct esp_hdr *)(ph + uhlen);
+ iv = (uint64_t *)(esph + 1);
+ copy_iv(iv, ivp, sa->iv_len);
+
+ esph->spi = sa->spi;
+ esph->seq = sqn_low32(sqc);
+
+ /* offset for ICV */
+ pdofs += pdlen + sa->sqh_len;
+
+ /* pad length */
+ pdlen -= sizeof(*espt);
+
+ /* copy padding data */
+ rte_memcpy(pt, esp_pad_bytes, pdlen);
+
+ /* update esp trailer */
+ espt = (struct esp_tail *)(pt + pdlen);
+ espt->pad_len = pdlen;
+ espt->next_proto = np;
+
+ icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
+ icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
+
+ return clen;
+}
+
+/*
+ * setup/update packets and crypto ops for ESP outbound transport case.
+ */
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ struct rte_cryptodev_sym_session *cs;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+ cs = ss->crypto.ses;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i], l2, l3, &icv);
+
+ /* success, setup crypto op */
+ if (rc >= 0) {
+ outb_pkt_xprepare(sa, sqc, &icv);
+ lksd_none_cop_prepare(cop[k], cs, mb[i]);
+ outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
+ k++;
+ /* failure, put packet into the death-row */
+ } else {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not prepared mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ return k;
+}
+
+/*
+ * process outbound packets for SA with ESN support,
+ * for algorithms that require SQN.hibits to be implictly included
+ * into digest computation.
+ * In that case we have to move ICV bytes back to their proper place.
+ */
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num)
+{
+ uint32_t i, k, icv_len, *icv;
+ struct rte_mbuf *ml;
+ struct rte_ipsec_sa *sa;
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ k = 0;
+ icv_len = sa->icv_len;
+
+ for (i = 0; i != num; i++) {
+ if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
+ ml = rte_pktmbuf_lastseg(mb[i]);
+ icv = rte_pktmbuf_mtod_offset(ml, void *,
+ ml->data_len - icv_len);
+ remove_sqh(icv, icv_len);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ /* handle unprocessed mbufs */
+ if (k != num) {
+ rte_errno = EBADMSG;
+ if (k != 0)
+ move_bad_mbufs(mb, dr, num, num - k);
+ }
+
+ return k;
+}
+
+/*
+ * prepare packets for inline ipsec processing:
+ * set ol_flags and attach metadata.
+ */
+static inline void
+inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ uint32_t i, ol_flags;
+
+ ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
+ for (i = 0; i != num; i++) {
+
+ mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
+ if (ol_flags != 0)
+ rte_security_set_pkt_metadata(ss->security.ctx,
+ ss->security.ses, mb[i], NULL);
+ }
+}
+
+/*
+ * process group of ESP outbound tunnel packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * process group of ESP outbound transport packets destined for
+ * INLINE_CRYPTO type of device.
+ */
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ int32_t rc;
+ uint32_t i, k, n, l2, l3;
+ uint64_t sqn;
+ rte_be64_t sqc;
+ struct rte_ipsec_sa *sa;
+ union sym_op_data icv;
+ uint64_t iv[IPSEC_MAX_IV_QWORD];
+ uint32_t dr[num];
+
+ sa = ss->sa;
+
+ n = num;
+ sqn = esn_outb_update_sqn(sa, &n);
+ if (n != num)
+ rte_errno = EOVERFLOW;
+
+ k = 0;
+ for (i = 0; i != n; i++) {
+
+ l2 = mb[i]->l2_len;
+ l3 = mb[i]->l3_len;
+
+ sqc = rte_cpu_to_be_64(sqn + i);
+ gen_iv(iv, sqc);
+
+ /* try to update the packet itself */
+ rc = outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
+ l2, l3, &icv);
+
+ k += (rc >= 0);
+
+ /* failure, put packet into the death-row */
+ if (rc < 0) {
+ dr[i - k] = i;
+ rte_errno = -rc;
+ }
+ }
+
+ /* copy not processed mbufs beyond good ones */
+ if (k != n && k != 0)
+ move_bad_mbufs(mb, dr, n, n - k);
+
+ inline_outb_mbuf_prepare(ss, mb, k);
+ return k;
+}
+
+/*
+ * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+ * actual processing is done by HW/PMD, just set flags and metadata.
+ */
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ inline_outb_mbuf_prepare(ss, mb, num);
+ return num;
+}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index a3ae7e2de..4ba079d75 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -56,21 +56,6 @@ sqn_low16(rte_be64_t sqn)
#endif
}
-/*
- * for given size, calculate required number of buckets.
- */
-static uint32_t
-replay_num_bucket(uint32_t wsz)
-{
- uint32_t nb;
-
- nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
- WINDOW_BUCKET_SIZE);
- nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
-
- return nb;
-}
-
/*
* According to RFC4303 A2.1, determine the high-order bit of sequence number.
* use 32bit arithmetic inside, return uint64_t.
@@ -222,21 +207,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
* between writer and readers.
*/
-/**
- * Based on number of buckets calculated required size for the
- * structure that holds replay window and sequence number (RSN) information.
- */
-static size_t
-rsn_size(uint32_t nb_bucket)
-{
- size_t sz;
- struct replay_sqn *rsn;
-
- sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
- sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
- return sz;
-}
-
/**
* Copy replay window and SQN.
*/
diff --git a/lib/librte_ipsec/meson.build b/lib/librte_ipsec/meson.build
index d2427b809..18fb2a143 100644
--- a/lib/librte_ipsec/meson.build
+++ b/lib/librte_ipsec/meson.build
@@ -3,7 +3,7 @@
allow_experimental_apis = true
-sources=files('sa.c', 'ses.c')
+sources=files('esp_inb.c', 'esp_outb.c', 'sa.c', 'ses.c')
install_headers = files('rte_ipsec.h', 'rte_ipsec_group.h', 'rte_ipsec_sa.h')
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
new file mode 100644
index 000000000..67a6be2aa
--- /dev/null
+++ b/lib/librte_ipsec/misc.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _MISC_H_
+#define _MISC_H_
+
+/**
+ * @file misc.h
+ * Contains miscelaneous functions/structures/macros used internally
+ * by ipsec library.
+ */
+
+/*
+ * Move bad (unprocessed) mbufs beyond the good (processed) ones.
+ * bad_idx[] contains the indexes of bad mbufs inside the mb[].
+ */
+static inline void
+move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
+ uint32_t nb_bad)
+{
+ uint32_t i, j, k;
+ struct rte_mbuf *drb[nb_bad];
+
+ j = 0;
+ k = 0;
+
+ /* copy bad ones into a temp place */
+ for (i = 0; i != nb_mb; i++) {
+ if (j != nb_bad && i == bad_idx[j])
+ drb[j++] = mb[i];
+ else
+ mb[k++] = mb[i];
+ }
+
+ /* copy bad ones after the good ones */
+ for (i = 0; i != nb_bad; i++)
+ mb[k + i] = drb[i];
+}
+
+#endif /* _MISC_H_ */
diff --git a/lib/librte_ipsec/sa.c b/lib/librte_ipsec/sa.c
index e728c6a28..846e317fe 100644
--- a/lib/librte_ipsec/sa.c
+++ b/lib/librte_ipsec/sa.c
@@ -12,6 +12,7 @@
#include "ipsec_sqn.h"
#include "crypto.h"
#include "iph.h"
+#include "misc.h"
#include "pad.h"
#define MBUF_MAX_L2_LEN RTE_LEN2MASK(RTE_MBUF_L2_LEN_BITS, uint64_t)
@@ -82,6 +83,36 @@ rte_ipsec_sa_type(const struct rte_ipsec_sa *sa)
return sa->type;
}
+/**
+ * Based on number of buckets calculated required size for the
+ * structure that holds replay window and sequence number (RSN) information.
+ */
+static size_t
+rsn_size(uint32_t nb_bucket)
+{
+ size_t sz;
+ struct replay_sqn *rsn;
+
+ sz = sizeof(*rsn) + nb_bucket * sizeof(rsn->window[0]);
+ sz = RTE_ALIGN_CEIL(sz, RTE_CACHE_LINE_SIZE);
+ return sz;
+}
+
+/*
+ * for given size, calculate required number of buckets.
+ */
+static uint32_t
+replay_num_bucket(uint32_t wsz)
+{
+ uint32_t nb;
+
+ nb = rte_align32pow2(RTE_ALIGN_MUL_CEIL(wsz, WINDOW_BUCKET_SIZE) /
+ WINDOW_BUCKET_SIZE);
+ nb = RTE_MAX(nb, (uint32_t)WINDOW_BUCKET_MIN);
+
+ return nb;
+}
+
static int32_t
ipsec_sa_size(uint64_t type, uint32_t *wnd_sz, uint32_t *nb_bucket)
{
@@ -450,625 +481,6 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
return sz;
}
-/*
- * Move bad (unprocessed) mbufs beyond the good (processed) ones.
- * bad_idx[] contains the indexes of bad mbufs inside the mb[].
- */
-static inline void
-move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
- uint32_t nb_bad)
-{
- uint32_t i, j, k;
- struct rte_mbuf *drb[nb_bad];
-
- j = 0;
- k = 0;
-
- /* copy bad ones into a temp place */
- for (i = 0; i != nb_mb; i++) {
- if (j != nb_bad && i == bad_idx[j])
- drb[j++] = mb[i];
- else
- mb[k++] = mb[i];
- }
-
- /* copy bad ones after the good ones */
- for (i = 0; i != nb_bad; i++)
- mb[k + i] = drb[i];
-}
-
-/*
- * setup crypto ops for LOOKASIDE_NONE (pure crypto) type of devices.
- */
-static inline void
-lksd_none_cop_prepare(struct rte_crypto_op *cop,
- struct rte_cryptodev_sym_session *cs, struct rte_mbuf *mb)
-{
- struct rte_crypto_sym_op *sop;
-
- sop = cop->sym;
- cop->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
- cop->status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
- cop->sess_type = RTE_CRYPTO_OP_WITH_SESSION;
- sop->m_src = mb;
- __rte_crypto_sym_op_attach_sym_session(sop, cs);
-}
-
-/*
- * setup crypto op and crypto sym op for ESP outbound packet.
- */
-static inline void
-esp_outb_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, const uint64_t ivp[IPSEC_MAX_IV_QWORD],
- const union sym_op_data *icv, uint32_t hlen, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint8_t algo_type = sa->algo_type;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_CBC:
- /* Cipher-Auth (AES-CBC *) case */
- case ALGO_TYPE_3DES_CBC:
- /* Cipher-Auth (3DES-CBC *) case */
- case ALGO_TYPE_NULL:
- /* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
- case ALGO_TYPE_AES_GCM:
- /* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CTR:
- /* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- default:
- break;
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound tunnel case.
- */
-static inline int32_t
-esp_outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- union sym_op_data *icv)
-{
- uint32_t clen, hlen, l2len, pdlen, pdofs, plen, tlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- /* calculate extra header space required */
- hlen = sa->hdr_len + sa->iv_len + sizeof(*esph);
-
- /* size of ipsec protected data */
- l2len = mb->l2_len;
- plen = mb->pkt_len - l2len;
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and prepend */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend header */
- ph = rte_pktmbuf_prepend(mb, hlen - l2len);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* update pkt l2/l3 len */
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* copy tunnel pkt header */
- rte_memcpy(ph, sa->hdr, sa->hdr_len);
-
- /* update original and new ip header fields */
- update_tun_l3hdr(sa, ph + sa->hdr_l3_off, mb->pkt_len, sa->hdr_l3_off,
- sqn_low16(sqc));
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + sa->hdr_len);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = sa->proto;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-outb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- uint32_t *psqh;
- struct aead_gcm_aad *aad;
- uint8_t algo_type = sa->algo_type;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0) {
- psqh = (uint32_t *)(icv->va - sa->sqh_len);
- psqh[0] = sqn_hi32(sqc);
- }
-
- /*
- * fill IV and AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM .
- */
- if (algo_type == ALGO_TYPE_AES_GCM) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound tunnel case.
- */
-static uint16_t
-outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, 0, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup/update packet data and metadata for ESP outbound transport case.
- */
-static inline int32_t
-esp_outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const uint64_t ivp[IPSEC_MAX_IV_QWORD], struct rte_mbuf *mb,
- uint32_t l2len, uint32_t l3len, union sym_op_data *icv)
-{
- uint8_t np;
- uint32_t clen, hlen, pdlen, pdofs, plen, tlen, uhlen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- char *ph, *pt;
- uint64_t *iv;
-
- uhlen = l2len + l3len;
- plen = mb->pkt_len - uhlen;
-
- /* calculate extra header space required */
- hlen = sa->iv_len + sizeof(*esph);
-
- /* number of bytes to encrypt */
- clen = plen + sizeof(*espt);
- clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
-
- /* pad length + esp tail */
- pdlen = clen - plen;
- tlen = pdlen + sa->icv_len;
-
- /* do append and insert */
- ml = rte_pktmbuf_lastseg(mb);
- if (tlen + sa->sqh_len + sa->aad_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- /* prepend space for ESP header */
- ph = rte_pktmbuf_prepend(mb, hlen);
- if (ph == NULL)
- return -ENOSPC;
-
- /* append tail */
- pdofs = ml->data_len;
- ml->data_len += tlen;
- mb->pkt_len += tlen;
- pt = rte_pktmbuf_mtod_offset(ml, typeof(pt), pdofs);
-
- /* shift L2/L3 headers */
- insert_esph(ph, ph + hlen, uhlen);
-
- /* update ip header fields */
- np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len, l2len, l3len,
- IPPROTO_ESP);
-
- /* update spi, seqn and iv */
- esph = (struct esp_hdr *)(ph + uhlen);
- iv = (uint64_t *)(esph + 1);
- copy_iv(iv, ivp, sa->iv_len);
-
- esph->spi = sa->spi;
- esph->seq = sqn_low32(sqc);
-
- /* offset for ICV */
- pdofs += pdlen + sa->sqh_len;
-
- /* pad length */
- pdlen -= sizeof(*espt);
-
- /* copy padding data */
- rte_memcpy(pt, esp_pad_bytes, pdlen);
-
- /* update esp trailer */
- espt = (struct esp_tail *)(pt + pdlen);
- espt->pad_len = pdlen;
- espt->next_proto = np;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, pdofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, pdofs);
-
- return clen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP outbound transport case.
- */
-static uint16_t
-outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- /* success, setup crypto op */
- if (rc >= 0) {
- outb_pkt_xprepare(sa, sqc, &icv);
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- esp_outb_cop_prepare(cop[k], sa, iv, &icv, l2 + l3, rc);
- k++;
- /* failure, put packet into the death-row */
- } else {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not prepared mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- return k;
-}
-
-/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
- */
-static inline int32_t
-esp_inb_tun_cop_prepare(struct rte_crypto_op *cop,
- const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
-{
- struct rte_crypto_sym_op *sop;
- struct aead_gcm_iv *gcm;
- struct aesctr_cnt_blk *ctr;
- uint64_t *ivc, *ivp;
- uint32_t clen;
- uint8_t algo_type = sa->algo_type;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
-
- /* fill sym op fields */
- sop = cop->sym;
-
- switch (algo_type) {
- case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
-
- /* fill AAD IV (located inside crypto op) */
- gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_AES_CBC:
- case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- copy_iv(ivc, ivp, sa->iv_len);
- break;
- case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
-
- /* copy iv from the input packet to the cop */
- ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
- sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
- aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
- break;
- case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
- break;
-
- default:
- return -EINVAL;
- }
-
- return 0;
-}
-
-/*
- * for pure cryptodev (lookaside none) depending on SA settings,
- * we might have to write some extra data to the packet.
- */
-static inline void
-inb_pkt_xprepare(const struct rte_ipsec_sa *sa, rte_be64_t sqc,
- const union sym_op_data *icv)
-{
- struct aead_gcm_aad *aad;
-
- /* insert SQN.hi between ESP trailer and ICV */
- if (sa->sqh_len != 0)
- insert_sqh(sqn_hi32(sqc), icv->va, sa->icv_len);
-
- /*
- * fill AAD fields, if any (aad fields are placed after icv),
- * right now we support only one AEAD algorithm: AES-GCM.
- */
- if (sa->aad_len != 0) {
- aad = (struct aead_gcm_aad *)(icv->va + sa->icv_len);
- aead_gcm_aad_fill(aad, sa->spi, sqc, IS_ESN(sa));
- }
-}
-
-/*
- * setup/update packet data and metadata for ESP inbound tunnel case.
- */
-static inline int32_t
-esp_inb_tun_pkt_prepare(const struct rte_ipsec_sa *sa,
- const struct replay_sqn *rsn, struct rte_mbuf *mb,
- uint32_t hlen, union sym_op_data *icv)
-{
- int32_t rc;
- uint64_t sqn;
- uint32_t icv_ofs, plen;
- struct rte_mbuf *ml;
- struct esp_hdr *esph;
-
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
-
- /*
- * retrieve and reconstruct SQN, then check it, then
- * convert it back into network byte order.
- */
- sqn = rte_be_to_cpu_32(esph->seq);
- if (IS_ESN(sa))
- sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
-
- rc = esn_inb_check_sqn(rsn, sa, sqn);
- if (rc != 0)
- return rc;
-
- sqn = rte_cpu_to_be_64(sqn);
-
- /* start packet manipulation */
- plen = mb->pkt_len;
- plen = plen - hlen;
-
- ml = rte_pktmbuf_lastseg(mb);
- icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
-
- /* we have to allocate space for AAD somewhere,
- * right now - just use free trailing space at the last segment.
- * Would probably be more convenient to reserve space for AAD
- * inside rte_crypto_op itself
- * (again for IV space is already reserved inside cop).
- */
- if (sa->aad_len + sa->sqh_len > rte_pktmbuf_tailroom(ml))
- return -ENOSPC;
-
- icv->va = rte_pktmbuf_mtod_offset(ml, void *, icv_ofs);
- icv->pa = rte_pktmbuf_iova_offset(ml, icv_ofs);
-
- inb_pkt_xprepare(sa, sqn, icv);
- return plen;
-}
-
-/*
- * setup/update packets and crypto ops for ESP inbound case.
- */
-static uint16_t
-inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- struct rte_crypto_op *cop[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, hl;
- struct rte_ipsec_sa *sa;
- struct rte_cryptodev_sym_session *cs;
- struct replay_sqn *rsn;
- union sym_op_data icv;
- uint32_t dr[num];
-
- sa = ss->sa;
- cs = ss->crypto.ses;
- rsn = rsn_acquire(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
-
- hl = mb[i]->l2_len + mb[i]->l3_len;
- rc = esp_inb_tun_pkt_prepare(sa, rsn, mb[i], hl, &icv);
- if (rc >= 0) {
- lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = esp_inb_tun_cop_prepare(cop[k], sa, mb[i], &icv,
- hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- rsn_release(sa, rsn);
-
- /* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- return k;
-}
-
/*
* setup crypto ops for LOOKASIDE_PROTO type of devices.
*/
@@ -1103,265 +515,6 @@ lksd_proto_prepare(const struct rte_ipsec_session *ss,
return num;
}
-/*
- * process ESP inbound tunnel packet.
- */
-static inline int
-esp_inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
-
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* reset mbuf metatdata: L2/L3 len, packet type */
- mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * process ESP inbound transport packet.
- */
-static inline int
-esp_inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
-{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
-
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
-
- icv_len = sa->icv_len;
-
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
-
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
-
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
-
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
-
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
-
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
-
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
-}
-
-/*
- * for group of ESP inbound packets perform SQN check and update.
- */
-static inline uint16_t
-esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
- uint32_t dr[], uint16_t num)
-{
- uint32_t i, k;
- struct replay_sqn *rsn;
-
- rsn = rsn_update_start(sa);
-
- k = 0;
- for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
- k++;
- else
- dr[i - k] = i;
- }
-
- rsn_update_finish(sa, rsn);
- return k;
-}
-
-/*
- * process group of ESP inbound tunnel packets.
- */
-static uint16_t
-inb_tun_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- struct rte_ipsec_sa *sa;
- uint32_t sqn[num];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process group of ESP inbound transport packets.
- */
-static uint16_t
-inb_trs_pkt_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (esp_inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
-}
-
-/*
- * process outbound packets for SA with ESN support,
- * for algorithms that require SQN.hibits to be implictly included
- * into digest computation.
- * In that case we have to move ICV bytes back to their proper place.
- */
-static uint16_t
-outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
- uint16_t num)
-{
- uint32_t i, k, icv_len, *icv;
- struct rte_mbuf *ml;
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- k = 0;
- icv_len = sa->icv_len;
-
- for (i = 0; i != num; i++) {
- if ((mb[i]->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) == 0) {
- ml = rte_pktmbuf_lastseg(mb[i]);
- icv = rte_pktmbuf_mtod_offset(ml, void *,
- ml->data_len - icv_len);
- remove_sqh(icv, icv_len);
- k++;
- } else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num) {
- rte_errno = EBADMSG;
- if (k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
- }
-
- return k;
-}
-
/*
* simplest pkt process routine:
* all actual processing is already done by HW/PMD,
@@ -1398,142 +551,6 @@ pkt_flag_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
return k;
}
-/*
- * prepare packets for inline ipsec processing:
- * set ol_flags and attach metadata.
- */
-static inline void
-inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- uint32_t i, ol_flags;
-
- ol_flags = ss->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA;
- for (i = 0; i != num; i++) {
-
- mb[i]->ol_flags |= PKT_TX_SEC_OFFLOAD;
- if (ol_flags != 0)
- rte_security_set_pkt_metadata(ss->security.ctx,
- ss->security.ses, mb[i], NULL);
- }
-}
-
-/*
- * process group of ESP outbound tunnel packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_tun_pkt_prepare(sa, sqc, iv, mb[i], &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * process group of ESP outbound transport packets destined for
- * INLINE_CRYPTO type of device.
- */
-static uint16_t
-inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- int32_t rc;
- uint32_t i, k, n, l2, l3;
- uint64_t sqn;
- rte_be64_t sqc;
- struct rte_ipsec_sa *sa;
- union sym_op_data icv;
- uint64_t iv[IPSEC_MAX_IV_QWORD];
- uint32_t dr[num];
-
- sa = ss->sa;
-
- n = num;
- sqn = esn_outb_update_sqn(sa, &n);
- if (n != num)
- rte_errno = EOVERFLOW;
-
- k = 0;
- for (i = 0; i != n; i++) {
-
- l2 = mb[i]->l2_len;
- l3 = mb[i]->l3_len;
-
- sqc = rte_cpu_to_be_64(sqn + i);
- gen_iv(iv, sqc);
-
- /* try to update the packet itself */
- rc = esp_outb_trs_pkt_prepare(sa, sqc, iv, mb[i],
- l2, l3, &icv);
-
- k += (rc >= 0);
-
- /* failure, put packet into the death-row */
- if (rc < 0) {
- dr[i - k] = i;
- rte_errno = -rc;
- }
- }
-
- /* copy not processed mbufs beyond good ones */
- if (k != n && k != 0)
- move_bad_mbufs(mb, dr, n, n - k);
-
- inline_outb_mbuf_prepare(ss, mb, k);
- return k;
-}
-
-/*
- * outbound for RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
- * actual processing is done by HW/PMD, just set flags and metadata.
- */
-static uint16_t
-outb_inline_proto_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
-{
- inline_outb_mbuf_prepare(ss, mb, num);
- return num;
-}
-
/*
* Select packet processing function for session on LOOKASIDE_NONE
* type of device.
@@ -1551,23 +568,23 @@ lksd_none_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_tun_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = inb_pkt_prepare;
- pf->process = inb_trs_pkt_process;
+ pf->prepare = esp_inb_pkt_prepare;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->prepare = outb_tun_prepare;
+ pf->prepare = esp_outb_tun_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->prepare = outb_trs_prepare;
+ pf->prepare = esp_outb_trs_prepare;
pf->process = (sa->sqh_len != 0) ?
- outb_sqh_process : pkt_flag_process;
+ esp_outb_sqh_process : pkt_flag_process;
break;
default:
rc = -ENOTSUP;
@@ -1593,10 +610,10 @@ inline_crypto_pkt_func_select(const struct rte_ipsec_sa *sa,
switch (sa->type & msk) {
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TUNLV6):
- pf->process = inb_tun_pkt_process;
+ pf->process = esp_inb_tun_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_IB | RTE_IPSEC_SATP_MODE_TRANS):
- pf->process = inb_trs_pkt_process;
+ pf->process = esp_inb_trs_pkt_process;
break;
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV4):
case (RTE_IPSEC_SATP_DIR_OB | RTE_IPSEC_SATP_MODE_TUNLV6):
@@ -1637,7 +654,7 @@ ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
RTE_IPSEC_SATP_DIR_IB)
pf->process = pkt_flag_process;
else
- pf->process = outb_inline_proto_process;
+ pf->process = inline_proto_outb_pkt_process;
break;
case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
pf->prepare = lksd_proto_prepare;
diff --git a/lib/librte_ipsec/sa.h b/lib/librte_ipsec/sa.h
index 48c3c103a..ffb5fb4f8 100644
--- a/lib/librte_ipsec/sa.h
+++ b/lib/librte_ipsec/sa.h
@@ -132,4 +132,44 @@ int
ipsec_sa_pkt_func_select(const struct rte_ipsec_session *ss,
const struct rte_ipsec_sa *sa, struct rte_ipsec_sa_pkt_func *pf);
+/* inbound processing */
+
+uint16_t
+esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+/* outbound processing */
+
+uint16_t
+esp_outb_tun_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_trs_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ struct rte_crypto_op *cop[], uint16_t num);
+
+uint16_t
+esp_outb_sqh_process(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
+ uint16_t num);
+
+uint16_t
+inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
+uint16_t
+inline_proto_outb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num);
+
#endif /* _SA_H_ */
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (5 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process " Konstantin Ananyev
` (3 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Right now check for packet length and padding is done inside cop_prepare().
It makes sense to have all necessary checks in one place at early stage:
inside pkt_prepare().
That allows to simplify (and later hopefully) optimize cop_prepare() part.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 41 +++++++++++++++++---------------------
1 file changed, 18 insertions(+), 23 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index a775c7b0b..8d1171556 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,7 +18,7 @@
/*
* setup crypto op and crypto sym op for ESP inbound tunnel packet.
*/
-static inline int32_t
+static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
const struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
const union sym_op_data *icv, uint32_t pofs, uint32_t plen)
@@ -27,11 +27,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
struct aead_gcm_iv *gcm;
struct aesctr_cnt_blk *ctr;
uint64_t *ivc, *ivp;
- uint32_t algo, clen;
-
- clen = plen - sa->ctp.cipher.length;
- if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
- return -EINVAL;
+ uint32_t algo;
algo = sa->algo_type;
@@ -41,7 +37,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
switch (algo) {
case ALGO_TYPE_AES_GCM:
sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = clen;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
sop->aead.digest.data = icv->va;
sop->aead.digest.phys_addr = icv->pa;
sop->aead.aad.data = icv->va + sa->icv_len;
@@ -57,7 +53,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -71,7 +67,7 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
@@ -86,17 +82,13 @@ inb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_NULL:
sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = clen;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
sop->auth.data.offset = pofs + sa->ctp.auth.offset;
sop->auth.data.length = plen - sa->ctp.auth.length;
sop->auth.digest.data = icv->va;
sop->auth.digest.phys_addr = icv->pa;
break;
- default:
- return -EINVAL;
}
-
- return 0;
}
/*
@@ -132,7 +124,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
{
int32_t rc;
uint64_t sqn;
- uint32_t icv_ofs, plen;
+ uint32_t clen, icv_ofs, plen;
struct rte_mbuf *ml;
struct esp_hdr *esph;
@@ -159,6 +151,11 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
ml = rte_pktmbuf_lastseg(mb);
icv_ofs = ml->data_len - sa->icv_len + sa->sqh_len;
+ /* check that packet has a valid length */
+ clen = plen - sa->ctp.cipher.length;
+ if ((int32_t)clen < 0 || (clen & (sa->pad_align - 1)) != 0)
+ return -EBADMSG;
+
/* we have to allocate space for AAD somewhere,
* right now - just use free trailing space at the last segment.
* Would probably be more convenient to reserve space for AAD
@@ -201,21 +198,19 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
rc = inb_pkt_prepare(sa, rsn, mb[i], hl, &icv);
if (rc >= 0) {
lksd_none_cop_prepare(cop[k], cs, mb[i]);
- rc = inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
- }
-
- k += (rc == 0);
- if (rc != 0) {
+ inb_cop_prepare(cop[k], sa, mb[i], &icv, hl, rc);
+ k++;
+ } else
dr[i - k] = i;
- rte_errno = -rc;
- }
}
rsn_release(sa, rsn);
/* copy not prepared mbufs beyond good ones */
- if (k != num && k != 0)
+ if (k != num && k != 0) {
move_bad_mbufs(mb, dr, num, num - k);
+ rte_errno = EBADMSG;
+ }
return k;
}
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process for esp inbound
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (6 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
` (2 subsequent siblings)
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process for esp inbound
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Change the order of operations for esp inbound post-process:
- read mbuf metadata and esp tail first for all packets in the burst
first to minimize stalls due to load latency.
- move code that is common for both transport and tunnel modes into
separate functions to reduce code duplication.
- add extra check for packet consitency
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
lib/librte_ipsec/esp_inb.c | 351 ++++++++++++++++++++++-------------
lib/librte_ipsec/ipsec_sqn.h | 4 -
2 files changed, 227 insertions(+), 128 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 8d1171556..138ed0450 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -15,8 +15,11 @@
#include "misc.h"
#include "pad.h"
+typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
+ struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+
/*
- * setup crypto op and crypto sym op for ESP inbound tunnel packet.
+ * setup crypto op and crypto sym op for ESP inbound packet.
*/
static inline void
inb_cop_prepare(struct rte_crypto_op *cop,
@@ -216,111 +219,239 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
}
/*
- * process ESP inbound tunnel packet.
+ * Start with processing inbound packet.
+ * This is common part for both tunnel and transport mode.
+ * Extract information that will be needed later from mbuf metadata and
+ * actual packet data:
+ * - mbuf for packet's last segment
+ * - length of the L2/L3 headers
+ * - esp tail structure
*/
-static inline int
-inb_tun_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline void
+process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
+ struct esp_tail *espt, uint32_t *hlen)
{
- uint32_t hlen, icv_len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *pd;
+ const struct esp_tail *pt;
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ ml[0] = rte_pktmbuf_lastseg(mb);
+ hlen[0] = mb->l2_len + mb->l3_len;
+ pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
+ ml[0]->data_len - tlen);
+ espt[0] = pt[0];
+}
- icv_len = sa->icv_len;
+/*
+ * packet checks for transport mode:
+ * - no reported IPsec related failures in ol_flags
+ * - tail length is valid
+ * - padding bytes are valid
+ */
+static inline int32_t
+trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, uint32_t tlen)
+{
+ const uint8_t *pd;
+ int32_t ofs;
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ ofs = ml->data_len - tlen;
+ pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
- /*
- * check padding and next proto.
- * return an error if something is wrong.
- */
- pd = (char *)espt - espt->pad_len;
- if (espt->next_proto != sa->proto ||
- memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ return ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+ ofs < 0 || tlen + hlen > mb->pkt_len ||
+ (espt.pad_len != 0 && memcmp(pd, esp_pad_bytes, espt.pad_len)));
+}
+
+/*
+ * packet checks for tunnel mode:
+ * - same as for trasnport mode
+ * - esp tail next proto contains expected for that SA value
+ */
+static inline int32_t
+tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
+ struct esp_tail espt, uint32_t hlen, const uint32_t tlen, uint8_t proto)
+{
+ return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+ espt.next_proto != proto);
+}
+
+/*
+ * step two for tunnel mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV, also if needed - L2/L3 headers
+ * (controlled by *adj* value)
+ */
+static inline void *
+tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ const struct esp_hdr *ph;
+
+ /* read SQN value */
+ ph = rte_pktmbuf_mtod_offset(mb, const struct esp_hdr *, hlen);
+ sqn[0] = ph->seq;
/* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
ml->data_len -= tlen;
mb->pkt_len -= tlen;
/* cut of L2/L3 headers, ESP header and IV */
- hlen = mb->l2_len + mb->l3_len;
- esph = rte_pktmbuf_mtod_offset(mb, struct esp_hdr *, hlen);
- rte_pktmbuf_adj(mb, hlen + sa->ctp.cipher.offset);
+ return rte_pktmbuf_adj(mb, adj);
+}
+
+/*
+ * step two for transport mode:
+ * - read SQN value (for future use)
+ * - cut of ICV, ESP tail and padding bytes
+ * - cut of ESP header and IV
+ * - move L2/L3 header to fill the gap after ESP header removal
+ */
+static inline void *
+trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
+ uint32_t adj, uint32_t tlen, uint32_t *sqn)
+{
+ char *np, *op;
+
+ /* get start of the packet before modifications */
+ op = rte_pktmbuf_mtod(mb, char *);
+
+ /* cut off ESP header and IV */
+ np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+
+ /* move header bytes to fill the gap after ESP header removal */
+ remove_esph(np, op, hlen);
+ return np;
+}
- /* retrieve SQN for later check */
- *sqn = rte_be_to_cpu_32(esph->seq);
+/*
+ * step three for transport mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ */
+static inline void
+trs_process_step3(struct rte_mbuf *mb)
+{
+ /* reset mbuf packet type */
+ mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+ /* clear the PKT_RX_SEC_OFFLOAD flag if set */
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
+}
+
+/*
+ * step three for tunnel mode:
+ * update mbuf metadata:
+ * - packet_type
+ * - ol_flags
+ * - tx_offload
+ */
+static inline void
+tun_process_step3(struct rte_mbuf *mb, uint64_t txof_msk, uint64_t txof_val)
+{
/* reset mbuf metatdata: L2/L3 len, packet type */
mb->packet_type = RTE_PTYPE_UNKNOWN;
- mb->tx_offload = (mb->tx_offload & sa->tx_offload.msk) |
- sa->tx_offload.val;
+ mb->tx_offload = (mb->tx_offload & txof_msk) | txof_val;
/* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+ mb->ol_flags &= ~PKT_RX_SEC_OFFLOAD;
}
+
/*
- * process ESP inbound transport packet.
+ * *process* function for tunnel packets
*/
-static inline int
-inb_trs_single_pkt_process(struct rte_ipsec_sa *sa, struct rte_mbuf *mb,
- uint32_t *sqn)
+static inline uint16_t
+tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
{
- uint32_t hlen, icv_len, l2len, l3len, tlen;
- struct esp_hdr *esph;
- struct esp_tail *espt;
- struct rte_mbuf *ml;
- char *np, *op, *pd;
+ uint32_t adj, i, k, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
- if (mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED)
- return -EBADMSG;
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
- icv_len = sa->icv_len;
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
- ml = rte_pktmbuf_lastseg(mb);
- espt = rte_pktmbuf_mtod_offset(ml, struct esp_tail *,
- ml->data_len - icv_len - sizeof(*espt));
+ k = 0;
+ for (i = 0; i != num; i++) {
- /* check padding, return an error if something is wrong. */
- pd = (char *)espt - espt->pad_len;
- if (memcmp(pd, esp_pad_bytes, espt->pad_len))
- return -EINVAL;
+ adj = hl[i] + cofs;
+ tl = tlen + espt[i].pad_len;
- /* cut of ICV, ESP tail and padding bytes */
- tlen = icv_len + sizeof(*espt) + espt->pad_len;
- ml->data_len -= tlen;
- mb->pkt_len -= tlen;
+ /* check that packet is valid */
+ if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+ sa->proto) == 0) {
- /* retrieve SQN for later check */
- l2len = mb->l2_len;
- l3len = mb->l3_len;
- hlen = l2len + l3len;
- op = rte_pktmbuf_mtod(mb, char *);
- esph = (struct esp_hdr *)(op + hlen);
- *sqn = rte_be_to_cpu_32(esph->seq);
+ /* modify packet's layout */
+ tun_process_step2(mb[i], ml[i], hl[i], adj,
+ tl, sqn + k);
+ /* update mbuf's metadata */
+ tun_process_step3(mb[i], sa->tx_offload.msk,
+ sa->tx_offload.val);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
- /* cut off ESP header and IV, update L3 header */
- np = rte_pktmbuf_adj(mb, sa->ctp.cipher.offset);
- remove_esph(np, op, hlen);
- update_trs_l3hdr(sa, np + l2len, mb->pkt_len, l2len, l3len,
- espt->next_proto);
+ return k;
+}
- /* reset mbuf packet type */
- mb->packet_type &= (RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
- /* clear the PKT_RX_SEC_OFFLOAD flag if set */
- mb->ol_flags &= ~(mb->ol_flags & PKT_RX_SEC_OFFLOAD);
- return 0;
+/*
+ * *process* function for tunnel packets
+ */
+static inline uint16_t
+trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
+ uint32_t sqn[], uint32_t dr[], uint16_t num)
+{
+ char *np;
+ uint32_t i, k, l2, tl;
+ uint32_t hl[num];
+ struct esp_tail espt[num];
+ struct rte_mbuf *ml[num];
+
+ const uint32_t tlen = sa->icv_len + sizeof(espt[0]);
+ const uint32_t cofs = sa->ctp.cipher.offset;
+
+ /*
+ * to minimize stalls due to load latency,
+ * read mbufs metadata and esp tail first.
+ */
+ for (i = 0; i != num; i++)
+ process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+
+ k = 0;
+ for (i = 0; i != num; i++) {
+
+ tl = tlen + espt[i].pad_len;
+ l2 = mb[i]->l2_len;
+
+ /* check that packet is valid */
+ if (trs_process_check(mb[i], ml[i], espt[i], hl[i] + cofs,
+ tl) == 0) {
+
+ /* modify packet's layout */
+ np = trs_process_step2(mb[i], ml[i], hl[i], cofs, tl,
+ sqn + k);
+ update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
+ l2, hl[i] - l2, espt[i].next_proto);
+
+ /* update mbuf's metadata */
+ trs_process_step3(mb[i]);
+ k++;
+ } else
+ dr[i - k] = i;
+ }
+
+ return k;
}
/*
@@ -333,11 +464,15 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
uint32_t i, k;
struct replay_sqn *rsn;
+ /* replay not enabled */
+ if (sa->replay.win_sz == 0)
+ return num;
+
rsn = rsn_update_start(sa);
k = 0;
for (i = 0; i != num; i++) {
- if (esn_inb_update_sqn(rsn, sa, sqn[i]) == 0)
+ if (esn_inb_update_sqn(rsn, sa, rte_be_to_cpu_32(sqn[i])) == 0)
k++;
else
dr[i - k] = i;
@@ -348,13 +483,13 @@ esp_inb_rsn_update(struct rte_ipsec_sa *sa, const uint32_t sqn[],
}
/*
- * process group of ESP inbound tunnel packets.
+ * process group of ESP inbound packets.
*/
-uint16_t
-esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
- struct rte_mbuf *mb[], uint16_t num)
+static inline uint16_t
+esp_inb_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num, esp_inb_process_t process)
{
- uint32_t i, k, n;
+ uint32_t k, n;
struct rte_ipsec_sa *sa;
uint32_t sqn[num];
uint32_t dr[num];
@@ -362,16 +497,7 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
sa = ss->sa;
/* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_tun_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
+ k = process(sa, mb, sqn, dr, num);
/* handle unprocessed mbufs */
if (k != num && k != 0)
@@ -390,6 +516,16 @@ esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
return n;
}
+/*
+ * process group of ESP inbound tunnel packets.
+ */
+uint16_t
+esp_inb_tun_pkt_process(const struct rte_ipsec_session *ss,
+ struct rte_mbuf *mb[], uint16_t num)
+{
+ return esp_inb_pkt_process(ss, mb, num, tun_process);
+}
+
/*
* process group of ESP inbound transport packets.
*/
@@ -397,38 +533,5 @@ uint16_t
esp_inb_trs_pkt_process(const struct rte_ipsec_session *ss,
struct rte_mbuf *mb[], uint16_t num)
{
- uint32_t i, k, n;
- uint32_t sqn[num];
- struct rte_ipsec_sa *sa;
- uint32_t dr[num];
-
- sa = ss->sa;
-
- /* process packets, extract seq numbers */
-
- k = 0;
- for (i = 0; i != num; i++) {
- /* good packet */
- if (inb_trs_single_pkt_process(sa, mb[i], sqn + k) == 0)
- k++;
- /* bad packet, will drop from furhter processing */
- else
- dr[i - k] = i;
- }
-
- /* handle unprocessed mbufs */
- if (k != num && k != 0)
- move_bad_mbufs(mb, dr, num, num - k);
-
- /* update SQN and replay winow */
- n = esp_inb_rsn_update(sa, sqn, dr, k);
-
- /* handle mbufs with wrong SQN */
- if (n != k && n != 0)
- move_bad_mbufs(mb, dr, k, k - n);
-
- if (n != num)
- rte_errno = EBADMSG;
-
- return n;
+ return esp_inb_pkt_process(ss, mb, num, trs_process);
}
diff --git a/lib/librte_ipsec/ipsec_sqn.h b/lib/librte_ipsec/ipsec_sqn.h
index 4ba079d75..0c2f76a7a 100644
--- a/lib/librte_ipsec/ipsec_sqn.h
+++ b/lib/librte_ipsec/ipsec_sqn.h
@@ -152,10 +152,6 @@ esn_inb_update_sqn(struct replay_sqn *rsn, const struct rte_ipsec_sa *sa,
{
uint32_t bit, bucket, last_bucket, new_bucket, diff, i;
- /* replay not enabled */
- if (sa->replay.win_sz == 0)
- return 0;
-
/* handle ESN */
if (IS_ESN(sa))
sqn = reconstruct_esn(rsn->sqn, sqn, sa->replay.win_sz);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (7 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process " Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-04-02 15:36 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Akhil Goyal
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
For sym_crypto_op prepare move common code into a separate function(s).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
lib/librte_ipsec/esp_inb.c | 72 +++++++++++++++++++++----------------
lib/librte_ipsec/esp_outb.c | 57 +++++++++++++++++++----------
2 files changed, 80 insertions(+), 49 deletions(-)
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 138ed0450..4e0e12a85 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -18,6 +18,40 @@
typedef uint16_t (*esp_inb_process_t)(const struct rte_ipsec_sa *sa,
struct rte_mbuf *mb[], uint32_t sqn[], uint32_t dr[], uint16_t num);
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->cipher.data.length = plen - sa->ctp.cipher.length;
+ sop->auth.data.offset = pofs + sa->ctp.auth.offset;
+ sop->auth.data.length = plen - sa->ctp.auth.length;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for aead algorithms
+ * used by inb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
+ sop->aead.data.length = plen - sa->ctp.cipher.length;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP inbound packet.
*/
@@ -33,63 +67,39 @@ inb_cop_prepare(struct rte_crypto_op *cop,
uint32_t algo;
algo = sa->algo_type;
+ ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
+ pofs + sizeof(struct esp_hdr));
/* fill sym op fields */
sop = cop->sym;
switch (algo) {
case ALGO_TYPE_AES_GCM:
- sop->aead.data.offset = pofs + sa->ctp.cipher.offset;
- sop->aead.data.length = plen - sa->ctp.cipher.length;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, pofs, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aead_gcm_iv_fill(gcm, ivp[0], sa->salt);
break;
case ALGO_TYPE_AES_CBC:
case ALGO_TYPE_3DES_CBC:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
/* copy iv from the input packet to the cop */
ivc = rte_crypto_op_ctod_offset(cop, uint64_t *, sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
copy_iv(ivc, ivp, sa->iv_len);
break;
case ALGO_TYPE_AES_CTR:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
- /* copy iv from the input packet to the cop */
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
- ivp = rte_pktmbuf_mtod_offset(mb, uint64_t *,
- pofs + sizeof(struct esp_hdr));
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
break;
case ALGO_TYPE_NULL:
- sop->cipher.data.offset = pofs + sa->ctp.cipher.offset;
- sop->cipher.data.length = plen - sa->ctp.cipher.length;
- sop->auth.data.offset = pofs + sa->ctp.auth.offset;
- sop->auth.data.length = plen - sa->ctp.auth.length;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, pofs, plen);
break;
}
}
diff --git a/lib/librte_ipsec/esp_outb.c b/lib/librte_ipsec/esp_outb.c
index 09bfb8658..c798bc4c4 100644
--- a/lib/librte_ipsec/esp_outb.c
+++ b/lib/librte_ipsec/esp_outb.c
@@ -15,6 +15,41 @@
#include "misc.h"
#include "pad.h"
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_ciph_auth_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->cipher.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->cipher.data.length = sa->ctp.cipher.length + plen;
+ sop->auth.data.offset = sa->ctp.auth.offset + pofs;
+ sop->auth.data.length = sa->ctp.auth.length + plen;
+ sop->auth.digest.data = icv->va;
+ sop->auth.digest.phys_addr = icv->pa;
+}
+
+/*
+ * helper function to fill crypto_sym op for cipher+auth algorithms.
+ * used by outb_cop_prepare(), see below.
+ */
+static inline void
+sop_aead_prepare(struct rte_crypto_sym_op *sop,
+ const struct rte_ipsec_sa *sa, const union sym_op_data *icv,
+ uint32_t pofs, uint32_t plen)
+{
+ sop->aead.data.offset = sa->ctp.cipher.offset + pofs;
+ sop->aead.data.length = sa->ctp.cipher.length + plen;
+ sop->aead.digest.data = icv->va;
+ sop->aead.digest.phys_addr = icv->pa;
+ sop->aead.aad.data = icv->va + sa->icv_len;
+ sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+}
+
/*
* setup crypto op and crypto sym op for ESP outbound packet.
*/
@@ -40,21 +75,11 @@ outb_cop_prepare(struct rte_crypto_op *cop,
/* Cipher-Auth (3DES-CBC *) case */
case ALGO_TYPE_NULL:
/* NULL case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
break;
case ALGO_TYPE_AES_GCM:
/* AEAD (AES_GCM) case */
- sop->aead.data.offset = sa->ctp.cipher.offset + hlen;
- sop->aead.data.length = sa->ctp.cipher.length + plen;
- sop->aead.digest.data = icv->va;
- sop->aead.digest.phys_addr = icv->pa;
- sop->aead.aad.data = icv->va + sa->icv_len;
- sop->aead.aad.phys_addr = icv->pa + sa->icv_len;
+ sop_aead_prepare(sop, sa, icv, hlen, plen);
/* fill AAD IV (located inside crypto op) */
gcm = rte_crypto_op_ctod_offset(cop, struct aead_gcm_iv *,
@@ -63,13 +88,9 @@ outb_cop_prepare(struct rte_crypto_op *cop,
break;
case ALGO_TYPE_AES_CTR:
/* Cipher-Auth (AES-CTR *) case */
- sop->cipher.data.offset = sa->ctp.cipher.offset + hlen;
- sop->cipher.data.length = sa->ctp.cipher.length + plen;
- sop->auth.data.offset = sa->ctp.auth.offset + hlen;
- sop->auth.data.length = sa->ctp.auth.length + plen;
- sop->auth.digest.data = icv->va;
- sop->auth.digest.phys_addr = icv->pa;
+ sop_ciph_auth_prepare(sop, sa, icv, hlen, plen);
+ /* fill CTR block (located inside crypto op) */
ctr = rte_crypto_op_ctod_offset(cop, struct aesctr_cnt_blk *,
sa->iv_ofs);
aes_ctr_cnt_blk_fill(ctr, ivp[0], sa->salt);
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (8 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 15:36 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Akhil Goyal
10 siblings, 1 reply; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index 3aad8d6a5..321a8b48b 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -234,6 +234,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-04-02 8:34 ` Konstantin Ananyev
0 siblings, 0 replies; 163+ messages in thread
From: Konstantin Ananyev @ 2019-04-02 8:34 UTC (permalink / raw)
To: dev; +Cc: akhil.goyal, olivier.matz, Konstantin Ananyev
Add librte_ipsec into 'Shared Library Versions' list in the release notes.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
---
doc/guides/rel_notes/release_19_05.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index 3aad8d6a5..321a8b48b 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -234,6 +234,7 @@ The libraries prepended with a plus sign were incremented in this version.
librte_gso.so.1
librte_hash.so.2
librte_ip_frag.so.1
+ librte_ipsec.so.1
librte_jobstats.so.1
librte_kni.so.2
librte_kvargs.so.1
--
2.17.1
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
@ 2019-04-02 8:49 ` Olivier Matz
2019-04-02 8:49 ` Olivier Matz
1 sibling, 1 reply; 163+ messages in thread
From: Olivier Matz @ 2019-04-02 8:49 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
On Tue, Apr 02, 2019 at 09:34:36AM +0100, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
> Add new test-case into UT for introduced function.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Thanks
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value
2019-04-02 8:49 ` Olivier Matz
@ 2019-04-02 8:49 ` Olivier Matz
0 siblings, 0 replies; 163+ messages in thread
From: Olivier Matz @ 2019-04-02 8:49 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev, akhil.goyal
On Tue, Apr 02, 2019 at 09:34:36AM +0100, Konstantin Ananyev wrote:
> Operations to set/update bit-fields often cause compilers
> to generate suboptimal code.
> To help avoid such situation for tx_offload fields:
> introduce new enum for tx_offload bit-fields lengths and offsets,
> and new function to generate raw tx_offload value.
> Add new test-case into UT for introduced function.
>
> Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Thanks
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
` (9 preceding siblings ...)
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
@ 2019-04-02 15:36 ` Akhil Goyal
2019-04-02 15:36 ` Akhil Goyal
10 siblings, 1 reply; 163+ messages in thread
From: Akhil Goyal @ 2019-04-02 15:36 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 4/2/2019 2:04 PM, Konstantin Ananyev wrote:
> v6 -> v5
> Fix endianness issue spotted by Olivier
> Added UT test-case for rte_mbuf_tx_offload()
>
> v5 -> v4
> Few cosmetic changes in mbuf tx_offload patch requested by Olivier
>
> v4 -> v3
> Address Akhil comments:
> - function/paramers names
> - add ipsec lib into shared libraries list
>
> v3 -> v2
> Address comments:
> - make rte_mbuf_tx_offload() always inline
> - rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
> Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
> Some code de-duplication
> Rebase against the latest next-crypto
>
> v1 -> v2
> Add patch to improve a bit esp inbound pkt_process code
>
> The first patch in this series introduces new enum for rte_mbuf
> tx_offload layout and new function to set raw value for rte_mbuf
> tx_offload.
> Remaining ones are cleanups and small performance improvements
> (on IA ~3-6% depending on the test-case) for ipsec library.
>
> Konstantin Ananyev (9):
> mbuf: new function to generate raw Tx offload value
> ipsec: add Tx offload template into SA
> ipsec: change the order in filling crypto op
> ipsec: change the way unprocessed mbufs are accounted
> ipsec: move inbound and outbound code into different files
> ipsec: reorder packet check for esp inbound
> ipsec: reorder packet process for esp inbound
> ipsec: de-duplicate crypto op prepare code-path
> doc: add ipsec lib into shared libraries list
>
> app/test/test_mbuf.c | 101 ++-
> doc/guides/rel_notes/release_19_05.rst | 1 +
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
> lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 34 -
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 +
> lib/librte_ipsec/sa.c | 1072 ++----------------------
> lib/librte_ipsec/sa.h | 45 +
> lib/librte_mbuf/rte_mbuf.h | 101 ++-
> 12 files changed, 1487 insertions(+), 1056 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
Applied to dpdk-next-crypto
Thanks.
^ permalink raw reply [flat|nested] 163+ messages in thread
* Re: [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library
2019-04-02 15:36 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Akhil Goyal
@ 2019-04-02 15:36 ` Akhil Goyal
0 siblings, 0 replies; 163+ messages in thread
From: Akhil Goyal @ 2019-04-02 15:36 UTC (permalink / raw)
To: Konstantin Ananyev, dev; +Cc: olivier.matz
On 4/2/2019 2:04 PM, Konstantin Ananyev wrote:
> v6 -> v5
> Fix endianness issue spotted by Olivier
> Added UT test-case for rte_mbuf_tx_offload()
>
> v5 -> v4
> Few cosmetic changes in mbuf tx_offload patch requested by Olivier
>
> v4 -> v3
> Address Akhil comments:
> - function/paramers names
> - add ipsec lib into shared libraries list
>
> v3 -> v2
> Address comments:
> - make rte_mbuf_tx_offload() always inline
> - rename OL[2,3]_LEN to OUTL[2,3]_LEN for better naming
> Add 'unused' parameter into rte_mbuf_tx_offload() for consistency
> Some code de-duplication
> Rebase against the latest next-crypto
>
> v1 -> v2
> Add patch to improve a bit esp inbound pkt_process code
>
> The first patch in this series introduces new enum for rte_mbuf
> tx_offload layout and new function to set raw value for rte_mbuf
> tx_offload.
> Remaining ones are cleanups and small performance improvements
> (on IA ~3-6% depending on the test-case) for ipsec library.
>
> Konstantin Ananyev (9):
> mbuf: new function to generate raw Tx offload value
> ipsec: add Tx offload template into SA
> ipsec: change the order in filling crypto op
> ipsec: change the way unprocessed mbufs are accounted
> ipsec: move inbound and outbound code into different files
> ipsec: reorder packet check for esp inbound
> ipsec: reorder packet process for esp inbound
> ipsec: de-duplicate crypto op prepare code-path
> doc: add ipsec lib into shared libraries list
>
> app/test/test_mbuf.c | 101 ++-
> doc/guides/rel_notes/release_19_05.rst | 1 +
> lib/librte_ipsec/Makefile | 2 +
> lib/librte_ipsec/crypto.h | 17 +
> lib/librte_ipsec/esp_inb.c | 547 ++++++++++++
> lib/librte_ipsec/esp_outb.c | 580 +++++++++++++
> lib/librte_ipsec/ipsec_sqn.h | 34 -
> lib/librte_ipsec/meson.build | 2 +-
> lib/librte_ipsec/misc.h | 41 +
> lib/librte_ipsec/sa.c | 1072 ++----------------------
> lib/librte_ipsec/sa.h | 45 +
> lib/librte_mbuf/rte_mbuf.h | 101 ++-
> 12 files changed, 1487 insertions(+), 1056 deletions(-)
> create mode 100644 lib/librte_ipsec/esp_inb.c
> create mode 100644 lib/librte_ipsec/esp_outb.c
> create mode 100644 lib/librte_ipsec/misc.h
>
Applied to dpdk-next-crypto
Thanks.
^ permalink raw reply [flat|nested] 163+ messages in thread
end of thread, other threads:[~2019-04-02 15:36 UTC | newest]
Thread overview: 163+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-02-28 19:20 [dpdk-dev] [PATCH 0/6] Few small improvements for ipsec library Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 1/6] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 2/6] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 3/6] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-02-28 19:20 ` [dpdk-dev] [PATCH 4/6] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-02-28 19:21 ` [dpdk-dev] [PATCH 5/6] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-02-28 19:21 ` [dpdk-dev] [PATCH 6/6] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:53 ` Wiles, Keith
2019-03-20 17:53 ` Wiles, Keith
2019-03-22 17:37 ` Ananyev, Konstantin
2019-03-22 17:37 ` Ananyev, Konstantin
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 17:24 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
2019-03-20 17:24 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 0/7] Few small improvements for ipsec library Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 1/7] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
2019-03-21 3:33 ` Jerin Jacob Kollanukkaran
2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 6:04 ` Shahaf Shuler
2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-21 13:51 ` Ananyev, Konstantin
2019-03-24 8:00 ` Shahaf Shuler
2019-03-24 8:00 ` Shahaf Shuler
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 1/8] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:16 ` Akhil Goyal
2019-03-28 8:16 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 2/8] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 8:52 ` Akhil Goyal
2019-03-28 8:52 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 3/8] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 9:02 ` Akhil Goyal
2019-03-28 9:02 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 4/8] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 10:52 ` Akhil Goyal
2019-03-28 10:52 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 5/8] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:20 ` Akhil Goyal
2019-03-28 11:20 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 6/8] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:27 ` Akhil Goyal
2019-03-28 11:27 ` Akhil Goyal
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 7/8] ipsec: reorder packet process " Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-26 15:43 ` [dpdk-dev] [PATCH v3 8/8] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
2019-03-26 15:43 ` Konstantin Ananyev
2019-03-28 11:35 ` Akhil Goyal
2019-03-28 11:35 ` Akhil Goyal
2019-03-28 11:21 ` [dpdk-dev] [PATCH v3 0/8] Few small improvements for ipsec library Akhil Goyal
2019-03-28 11:21 ` Akhil Goyal
2019-03-28 11:49 ` Ananyev, Konstantin
2019-03-28 11:49 ` Ananyev, Konstantin
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 0/9] " Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 12:54 ` Olivier Matz
2019-03-29 12:54 ` Olivier Matz
2019-03-30 14:20 ` Ananyev, Konstantin
2019-03-30 14:20 ` Ananyev, Konstantin
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 7/9] ipsec: reorder packet process " Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 10:27 ` [dpdk-dev] [PATCH v4 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-03-29 10:27 ` Konstantin Ananyev
2019-03-29 16:03 ` Akhil Goyal
2019-03-29 16:03 ` Akhil Goyal
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 13:18 ` Akhil Goyal
2019-04-01 13:18 ` Akhil Goyal
2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:22 ` Olivier Matz
2019-04-01 13:55 ` Ananyev, Konstantin
2019-04-01 13:55 ` Ananyev, Konstantin
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 7/9] ipsec: reorder packet process " Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-01 12:56 ` [dpdk-dev] [PATCH v5 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-04-01 12:56 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 1/9] mbuf: new function to generate raw Tx offload value Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:49 ` Olivier Matz
2019-04-02 8:49 ` Olivier Matz
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 2/9] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 3/9] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 4/9] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 5/9] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 6/9] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 7/9] ipsec: reorder packet process " Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 8/9] ipsec: de-duplicate crypto op prepare code-path Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 8:34 ` [dpdk-dev] [PATCH v6 9/9] doc: add ipsec lib into shared libraries list Konstantin Ananyev
2019-04-02 8:34 ` Konstantin Ananyev
2019-04-02 15:36 ` [dpdk-dev] [PATCH v6 0/9] Few small improvements for ipsec library Akhil Goyal
2019-04-02 15:36 ` Akhil Goyal
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 2/7] ipsec: add Tx offload template into SA Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 3/7] ipsec: change the order in filling crypto op Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 4/7] ipsec: change the way unprocessed mbufs are accounted Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 5/7] ipsec: move inbound and outbound code into different files Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 6/7] ipsec: reorder packet check for esp inbound Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
2019-03-20 18:46 ` [dpdk-dev] [PATCH v2 7/7] ipsec: reorder packet process " Konstantin Ananyev
2019-03-20 18:46 ` Konstantin Ananyev
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).