[dpdk-dev] [PATCH v2] ipsec: support multi-segment packets

[Konstantin Ananyev <konstantin.ananyev@intel.com>]

Add support for packets that consist of multiple segments.
Take into account that trailer bytes (padding, ESP tail, ICV)
can spawn across multiple segments.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---

v1 -> v2:
 merge with latest mainline
 fix build problem for RTE_BUILD_SHARED_LIB=y
 update programmer's guide

 doc/guides/prog_guide/ipsec_lib.rst |   1 -
 lib/librte_ipsec/Makefile           |   3 +-
 lib/librte_ipsec/esp_inb.c          | 148 +++++++++++++++++++++-------
 lib/librte_ipsec/misc.h             |  61 ++++++++++++
 4 files changed, 175 insertions(+), 38 deletions(-)

diff --git a/doc/guides/prog_guide/ipsec_lib.rst b/doc/guides/prog_guide/ipsec_lib.rst
index 6fc08886f..63b75b652 100644
--- a/doc/guides/prog_guide/ipsec_lib.rst
+++ b/doc/guides/prog_guide/ipsec_lib.rst
@@ -162,7 +162,6 @@ Limitations
 The following features are not properly supported in the current version:
 
 *  ESP transport mode for IPv6 packets with extension headers.
-*  Multi-segment packets.
 *  Updates of the fields in inner IP header for tunnel mode
    (as described in RFC 4301, section 5.1.2).
 *  Hard/soft limit for SA lifetime (time interval/byte count).
diff --git a/lib/librte_ipsec/Makefile b/lib/librte_ipsec/Makefile
index e80926baa..22f29d98a 100644
--- a/lib/librte_ipsec/Makefile
+++ b/lib/librte_ipsec/Makefile
@@ -9,7 +9,8 @@ LIB = librte_ipsec.a
 CFLAGS += -O3
 CFLAGS += $(WERROR_FLAGS) -I$(SRCDIR)
 CFLAGS += -DALLOW_EXPERIMENTAL_API
-LDLIBS += -lrte_eal -lrte_mbuf -lrte_net -lrte_cryptodev -lrte_security
+LDLIBS += -lrte_eal -lrte_mempool -lrte_mbuf -lrte_net
+LDLIBS += -lrte_cryptodev -lrte_security
 
 EXPORT_MAP := rte_ipsec_version.map
 
diff --git a/lib/librte_ipsec/esp_inb.c b/lib/librte_ipsec/esp_inb.c
index 3e12ca103..819d2bf25 100644
--- a/lib/librte_ipsec/esp_inb.c
+++ b/lib/librte_ipsec/esp_inb.c
@@ -104,6 +104,34 @@ inb_cop_prepare(struct rte_crypto_op *cop,
 	}
 }
 
+/*
+ * Helper function for prepare() to deal with situation when
+ * ICV is spread by two segments. Tries to move ICV completely into the
+ * last segment.
+ */
+static struct rte_mbuf *
+move_icv(struct rte_mbuf *ml, uint32_t ofs)
+{
+	uint32_t n;
+	struct rte_mbuf *ms;
+	const void *prev;
+	void *new;
+
+	ms = ml->next;
+	n = ml->data_len - ofs;
+
+	prev = rte_pktmbuf_mtod_offset(ml, const void *, ofs);
+	new = rte_pktmbuf_prepend(ms, n);
+	if (new == NULL)
+		return NULL;
+
+	/* move n ICV bytes from ml into ms */
+	rte_memcpy(new, prev, n);
+	ml->data_len -= n;
+
+	return ms;
+}
+
 /*
  * for pure cryptodev (lookaside none) depending on SA settings,
  * we might have to write some extra data to the packet.
@@ -137,7 +165,7 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
 {
 	int32_t rc;
 	uint64_t sqn;
-	uint32_t clen, icv_ofs, plen;
+	uint32_t clen, icv_len, icv_ofs, plen;
 	struct rte_mbuf *ml;
 	struct rte_esp_hdr *esph;
 
@@ -161,14 +189,33 @@ inb_pkt_prepare(const struct rte_ipsec_sa *sa, const struct replay_sqn *rsn,
 	plen = mb->pkt_len;
 	plen = plen - hlen;
 
-	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;
 
+	/* find ICV location */
+	icv_len = sa->icv_len;
+	icv_ofs = mb->pkt_len - icv_len;
+
+	ml = mbuf_get_seg_ofs(mb, &icv_ofs);
+
+	/*
+	 * if ICV is spread by two segments, then try to
+	 * move ICV completely into the last segment.
+	 */
+	if (ml->data_len < icv_ofs + icv_len) {
+
+		ml = move_icv(ml, icv_ofs);
+		if (ml == NULL)
+			return -ENOSPC;
+
+		/* new ICV location */
+		icv_ofs = 0;
+	}
+
+	icv_ofs += 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
@@ -239,36 +286,65 @@ esp_inb_pkt_prepare(const struct rte_ipsec_session *ss, struct rte_mbuf *mb[],
  */
 static inline void
 process_step1(struct rte_mbuf *mb, uint32_t tlen, struct rte_mbuf **ml,
-	struct esp_tail *espt, uint32_t *hlen)
+	struct esp_tail *espt, uint32_t *hlen, uint32_t *tofs)
 {
 	const struct esp_tail *pt;
+	uint32_t ofs;
 
-	ml[0] = rte_pktmbuf_lastseg(mb);
+	ofs = mb->pkt_len - tlen;
 	hlen[0] = mb->l2_len + mb->l3_len;
-	pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *,
-		ml[0]->data_len - tlen);
+	ml[0] = mbuf_get_seg_ofs(mb, &ofs);
+	pt = rte_pktmbuf_mtod_offset(ml[0], const struct esp_tail *, ofs);
+	tofs[0] = ofs;
 	espt[0] = pt[0];
 }
 
+/*
+ * Helper function to check pad bytes values.
+ * Note that pad bytes can be spread across multiple segments.
+ */
+static inline int
+check_pad_bytes(struct rte_mbuf *mb, uint32_t ofs, uint32_t len)
+{
+	const uint8_t *pd;
+	uint32_t k, n;
+
+	for (n = 0; n != len; n += k, mb = mb->next) {
+		k = mb->data_len - ofs;
+		k = RTE_MIN(k, len - n);
+		pd = rte_pktmbuf_mtod_offset(mb, const uint8_t *, ofs);
+		if (memcmp(pd, esp_pad_bytes + n, k) != 0)
+			break;
+		ofs = 0;
+	}
+
+	return len - n;
+}
+
 /*
  * packet checks for transport mode:
  * - no reported IPsec related failures in ol_flags
- * - tail length is valid
+ * - tail and header lengths are valid
  * - padding bytes are valid
+ * apart from checks, function also updates tail offset (and segment)
+ * by taking into account pad length.
  */
 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)
+trs_process_check(struct rte_mbuf *mb, struct rte_mbuf **ml,
+	uint32_t *tofs, struct esp_tail espt, uint32_t hlen, uint32_t tlen)
 {
-	const uint8_t *pd;
-	int32_t ofs;
+	if ((mb->ol_flags & PKT_RX_SEC_OFFLOAD_FAILED) != 0 ||
+			tlen + hlen > mb->pkt_len)
+		return -EBADMSG;
 
-	ofs = ml->data_len - tlen;
-	pd = rte_pktmbuf_mtod_offset(ml, const uint8_t *, ofs);
+	/* padding bytes are spread over multiple segments */
+	if (tofs[0] < espt.pad_len) {
+		tofs[0] = mb->pkt_len - tlen;
+		ml[0] = mbuf_get_seg_ofs(mb, tofs);
+	} else
+		tofs[0] -= espt.pad_len;
 
-	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)));
+	return check_pad_bytes(ml[0], tofs[0], espt.pad_len);
 }
 
 /*
@@ -277,10 +353,11 @@ trs_process_check(const struct rte_mbuf *mb, const struct rte_mbuf *ml,
  * - 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)
+tun_process_check(struct rte_mbuf *mb, struct rte_mbuf **ml,
+	uint32_t *tofs, struct esp_tail espt, uint32_t hlen, uint32_t tlen,
+	uint8_t proto)
 {
-	return (trs_process_check(mb, ml, espt, hlen, tlen) ||
+	return (trs_process_check(mb, ml, tofs, espt, hlen, tlen) ||
 		espt.next_proto != proto);
 }
 
@@ -293,7 +370,7 @@ tun_process_check(const struct rte_mbuf *mb, struct rte_mbuf *ml,
  */
 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)
+	uint32_t adj, uint32_t tofs, uint32_t tlen, uint32_t *sqn)
 {
 	const struct rte_esp_hdr *ph;
 
@@ -302,8 +379,7 @@ tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
 	sqn[0] = ph->seq;
 
 	/* cut of ICV, ESP tail and padding bytes */
-	ml->data_len -= tlen;
-	mb->pkt_len -= tlen;
+	mbuf_cut_seg_ofs(mb, ml, tofs, tlen);
 
 	/* cut of L2/L3 headers, ESP header and IV */
 	return rte_pktmbuf_adj(mb, adj);
@@ -318,7 +394,7 @@ tun_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
  */
 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)
+	uint32_t adj, uint32_t tofs, uint32_t tlen, uint32_t *sqn)
 {
 	char *np, *op;
 
@@ -326,7 +402,7 @@ trs_process_step2(struct rte_mbuf *mb, struct rte_mbuf *ml, uint32_t hlen,
 	op = rte_pktmbuf_mtod(mb, char *);
 
 	/* cut off ESP header and IV */
-	np = tun_process_step2(mb, ml, hlen, adj, tlen, sqn);
+	np = tun_process_step2(mb, ml, hlen, adj, tofs, tlen, sqn);
 
 	/* move header bytes to fill the gap after ESP header removal */
 	remove_esph(np, op, hlen);
@@ -376,7 +452,7 @@ tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 	uint32_t sqn[], uint32_t dr[], uint16_t num)
 {
 	uint32_t adj, i, k, tl;
-	uint32_t hl[num];
+	uint32_t hl[num], to[num];
 	struct esp_tail espt[num];
 	struct rte_mbuf *ml[num];
 
@@ -388,7 +464,7 @@ tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 	 * read mbufs metadata and esp tail first.
 	 */
 	for (i = 0; i != num; i++)
-		process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+		process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i], &to[i]);
 
 	k = 0;
 	for (i = 0; i != num; i++) {
@@ -397,11 +473,11 @@ tun_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 		tl = tlen + espt[i].pad_len;
 
 		/* check that packet is valid */
-		if (tun_process_check(mb[i], ml[i], espt[i], adj, tl,
+		if (tun_process_check(mb[i], &ml[i], &to[i], espt[i], adj, tl,
 					sa->proto) == 0) {
 
 			/* modify packet's layout */
-			tun_process_step2(mb[i], ml[i], hl[i], adj,
+			tun_process_step2(mb[i], ml[i], hl[i], adj, to[i],
 				tl, sqn + k);
 			/* update mbuf's metadata */
 			tun_process_step3(mb[i], sa->tx_offload.msk,
@@ -424,7 +500,7 @@ trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 {
 	char *np;
 	uint32_t i, k, l2, tl;
-	uint32_t hl[num];
+	uint32_t hl[num], to[num];
 	struct esp_tail espt[num];
 	struct rte_mbuf *ml[num];
 
@@ -436,7 +512,7 @@ trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 	 * read mbufs metadata and esp tail first.
 	 */
 	for (i = 0; i != num; i++)
-		process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i]);
+		process_step1(mb[i], tlen, &ml[i], &espt[i], &hl[i], &to[i]);
 
 	k = 0;
 	for (i = 0; i != num; i++) {
@@ -445,12 +521,12 @@ trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
 		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) {
+		if (trs_process_check(mb[i], &ml[i], &to[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);
+			np = trs_process_step2(mb[i], ml[i], hl[i], cofs,
+				to[i], tl, sqn + k);
 			update_trs_l3hdr(sa, np + l2, mb[i]->pkt_len,
 				l2, hl[i] - l2, espt[i].next_proto);
 
diff --git a/lib/librte_ipsec/misc.h b/lib/librte_ipsec/misc.h
index 693a4afdd..b0cafef4e 100644
--- a/lib/librte_ipsec/misc.h
+++ b/lib/librte_ipsec/misc.h
@@ -38,4 +38,65 @@ move_bad_mbufs(struct rte_mbuf *mb[], const uint32_t bad_idx[], uint32_t nb_mb,
 		mb[k + i] = drb[i];
 }
 
+/*
+ * Find packet's segment for the specified offset.
+ * ofs - at input should contain required offset, at output would contain
+ * offset value within the segment.
+ */
+static inline struct rte_mbuf *
+mbuf_get_seg_ofs(struct rte_mbuf *mb, uint32_t *ofs)
+{
+	uint32_t k, n;
+	struct rte_mbuf *ms;
+
+	ms = mb;
+	n = *ofs;
+
+	for (k = rte_pktmbuf_data_len(ms); n >= k;
+			k = rte_pktmbuf_data_len(ms)) {
+		ms = ms->next;
+		n -= k;
+	}
+
+	*ofs = n;
+	return ms;
+}
+
+/*
+ * Trim multi-segment packet at the specified offset, and free
+ * all unused segments.
+ * mb - input packet
+ * ms - segment where to cut
+ * ofs - offset within the *ms*
+ * len - length to cut (from given offset to the end of the packet)
+ * Can be used in conjunction with mbuf_get_seg_ofs():
+ * ofs = new_len;
+ * ms = mbuf_get_seg_ofs(mb, &ofs);
+ * mbuf_cut_seg_ofs(mb, ms, ofs, mb->pkt_len - new_len);
+ */
+static inline void
+mbuf_cut_seg_ofs(struct rte_mbuf *mb, struct rte_mbuf *ms, uint32_t ofs,
+	uint32_t len)
+{
+	uint32_t n, slen;
+	struct rte_mbuf *mn;
+
+	slen = ms->data_len;
+	ms->data_len = ofs;
+
+	/* tail spawns through multiple segments */
+	if (slen < ofs + len) {
+		mn = ms->next;
+		ms->next = NULL;
+		for (n = 0; mn != NULL; n++) {
+			ms = mn->next;
+			rte_pktmbuf_free_seg(mn);
+			mn = ms;
+		}
+		mb->nb_segs -= n;
+	}
+
+	mb->pkt_len -= len;
+}
+
 #endif /* _MISC_H_ */
-- 
2.17.1