Re: [dpdk-dev] [PATCH v6 06/10] ipsec: add transmit segmentation offload support

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

> Add support for transmit segmentation offload to inline crypto processing
> mode. This offload is not supported by other offload modes, as at a
> minimum it requires inline crypto for IPsec to be supported on the
> network interface.
> 
> Signed-off-by: Declan Doherty <declan.doherty@intel.com>
> Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
> Signed-off-by: Abhijit Sinha <abhijit.sinha@intel.com>
> Signed-off-by: Daniel Martin Buckley <daniel.m.buckley@intel.com>
> Acked-by: Fan Zhang <roy.fan.zhang@intel.com>
> ---
>  lib/ipsec/esp_inb.c  |   4 +-
>  lib/ipsec/esp_outb.c | 115 +++++++++++++++++++++++++++++++++++--------
>  lib/ipsec/iph.h      |  10 +++-
>  lib/ipsec/sa.c       |   6 +++
>  lib/ipsec/sa.h       |   4 ++
>  5 files changed, 114 insertions(+), 25 deletions(-)
> 
> diff --git a/lib/ipsec/esp_inb.c b/lib/ipsec/esp_inb.c
> index d66c88f05d..a6ab8fbdd5 100644
> --- a/lib/ipsec/esp_inb.c
> +++ b/lib/ipsec/esp_inb.c
> @@ -668,8 +668,8 @@ trs_process(const struct rte_ipsec_sa *sa, struct rte_mbuf *mb[],
>  			/* modify packet's layout */
>  			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);
> +			update_trs_l34hdrs(sa, np + l2, mb[i]->pkt_len,
> +				l2, hl[i] - l2, espt[i].next_proto, 0);
> 
>  			/* update mbuf's metadata */
>  			trs_process_step3(mb[i]);
> diff --git a/lib/ipsec/esp_outb.c b/lib/ipsec/esp_outb.c
> index a3f77469c3..9fc7075796 100644
> --- a/lib/ipsec/esp_outb.c
> +++ b/lib/ipsec/esp_outb.c
> @@ -2,6 +2,8 @@
>   * Copyright(c) 2018-2020 Intel Corporation
>   */
> 
> +#include <math.h>
> +
>  #include <rte_ipsec.h>
>  #include <rte_esp.h>
>  #include <rte_ip.h>
> @@ -156,11 +158,20 @@ outb_tun_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
> 
>  	/* number of bytes to encrypt */
>  	clen = plen + sizeof(*espt);
> -	clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
> +
> +	/* We don't need to pad/ailgn packet when using TSO offload */
> +	if (likely(!(mb->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_UDP_SEG))))
> +		clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
> +

Here and everywhere:
It doesn't look nice that we have to pollute generic functions with
checking TSO specific flags all over the place.
Can we probably have a specific prepare/process function for inline+tso case?
As we do have for cpu and inline cases right now.
Or just update inline version?

> 
>  	/* pad length + esp tail */
>  	pdlen = clen - plen;
> -	tlen = pdlen + sa->icv_len + sqh_len;
> +
> +	/* We don't append ICV length when using TSO offload */
> +	if (likely(!(mb->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_UDP_SEG))))
> +		tlen = pdlen + sa->icv_len + sqh_len;
> +	else
> +		tlen = pdlen + sqh_len;
> 
>  	/* do append and prepend */
>  	ml = rte_pktmbuf_lastseg(mb);
> @@ -337,6 +348,7 @@ outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
>  	char *ph, *pt;
>  	uint64_t *iv;
>  	uint32_t l2len, l3len;
> +	uint8_t tso = mb->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_UDP_SEG) ? 1 : 0;
> 
>  	l2len = mb->l2_len;
>  	l3len = mb->l3_len;
> @@ -349,11 +361,19 @@ outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
> 
>  	/* number of bytes to encrypt */
>  	clen = plen + sizeof(*espt);
> -	clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
> +
> +	/* We don't need to pad/ailgn packet when using TSO offload */
> +	if (likely(!tso))
> +		clen = RTE_ALIGN_CEIL(clen, sa->pad_align);
> 
>  	/* pad length + esp tail */
>  	pdlen = clen - plen;
> -	tlen = pdlen + sa->icv_len + sqh_len;
> +
> +	/* We don't append ICV length when using TSO offload */
> +	if (likely(!tso))
> +		tlen = pdlen + sa->icv_len + sqh_len;
> +	else
> +		tlen = pdlen + sqh_len;
> 
>  	/* do append and insert */
>  	ml = rte_pktmbuf_lastseg(mb);
> @@ -375,8 +395,8 @@ outb_trs_pkt_prepare(struct rte_ipsec_sa *sa, rte_be64_t sqc,
>  	insert_esph(ph, ph + hlen, uhlen);
> 
>  	/* update ip  header fields */
> -	np = update_trs_l3hdr(sa, ph + l2len, mb->pkt_len - sqh_len, l2len,
> -			l3len, IPPROTO_ESP);
> +	np = update_trs_l34hdrs(sa, ph + l2len, mb->pkt_len - sqh_len, l2len,
> +			l3len, IPPROTO_ESP, tso);
> 
>  	/* update spi, seqn and iv */
>  	esph = (struct rte_esp_hdr *)(ph + uhlen);
> @@ -651,6 +671,33 @@ inline_outb_mbuf_prepare(const struct rte_ipsec_session *ss,
>  	}
>  }
> 
> +/* check if packet will exceed MSS and segmentation is required */
> +static inline int
> +esn_outb_nb_segments(const struct rte_ipsec_sa *sa, struct rte_mbuf *m) {
> +	uint16_t segments = 1;
> +	uint16_t pkt_l3len = m->pkt_len - m->l2_len;
> +
> +	/* Only support segmentation for UDP/TCP flows */
> +	if (!(m->packet_type & (RTE_PTYPE_L4_UDP | RTE_PTYPE_L4_TCP)))
> +		return segments;
> +
> +	if (sa->tso.enabled && pkt_l3len > sa->tso.mss) {
> +		segments = ceil((float)pkt_l3len / sa->tso.mss);

Float calculations in the middle of data-path?
Just to calculate roundup?
Doesn't look good to me at all.

> +
> +		if  (m->packet_type & RTE_PTYPE_L4_TCP) {
> +			m->ol_flags |= (PKT_TX_TCP_SEG | PKT_TX_TCP_CKSUM);

That's really strange - why ipsec library will set PKT_TX_TCP_SEG unconditionally?
That should be responsibility of the upper layer, I think.
In the lib we should only check was tso requested for that packet or not.
Same for UDP.

> +			m->l4_len = sizeof(struct rte_tcp_hdr);

Hmm, how do we know there are no TCP options present for that packet?
Wouldn't it be better to expect user to provide proper l4_len for such packets?

> +		} else {
> +			m->ol_flags |= (PKT_TX_UDP_SEG | PKT_TX_UDP_CKSUM);
> +			m->l4_len = sizeof(struct rte_udp_hdr);
> +		}
> +
> +		m->tso_segsz = sa->tso.mss;
> +	}
> +
> +	return segments;
> +}
> +
>  /*
>   * process group of ESP outbound tunnel packets destined for
>   * INLINE_CRYPTO type of device.
> @@ -660,24 +707,29 @@ 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;
> +	uint32_t i, k, nb_sqn = 0, nb_sqn_alloc;
>  	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];
> +	uint16_t nb_segs[num];
> 
>  	sa = ss->sa;
> 
> -	n = num;
> -	sqn = esn_outb_update_sqn(sa, &n);
> -	if (n != num)
> +	for (i = 0; i != num; i++) {
> +		nb_segs[i] = esn_outb_nb_segments(sa, mb[i]);
> +		nb_sqn += nb_segs[i];
> +	}
> +
> +	nb_sqn_alloc = nb_sqn;
> +	sqn = esn_outb_update_sqn(sa, &nb_sqn_alloc);
> +	if (nb_sqn_alloc != nb_sqn)
>  		rte_errno = EOVERFLOW;
> 
>  	k = 0;
> -	for (i = 0; i != n; i++) {
> -
> +	for (i = 0; i != num; i++) {
>  		sqc = rte_cpu_to_be_64(sqn + i);
>  		gen_iv(iv, sqc);
> 
> @@ -691,11 +743,18 @@ inline_outb_tun_pkt_process(const struct rte_ipsec_session *ss,
>  			dr[i - k] = i;
>  			rte_errno = -rc;
>  		}
> +
> +		/**
> +		 * If packet is using tso, increment sqn by the number of
> +		 * segments for	packet
> +		 */
> +		if  (mb[i]->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_UDP_SEG))
> +			sqn += nb_segs[i] - 1;
>  	}
> 
>  	/* copy not processed mbufs beyond good ones */
> -	if (k != n && k != 0)
> -		move_bad_mbufs(mb, dr, n, n - k);
> +	if (k != num && k != 0)
> +		move_bad_mbufs(mb, dr, num, num - k);
> 
>  	inline_outb_mbuf_prepare(ss, mb, k);
>  	return k;
> @@ -710,23 +769,30 @@ 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;
> +	uint32_t i, k, nb_sqn, nb_sqn_alloc;
>  	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];
> +	uint16_t nb_segs[num];
> 
>  	sa = ss->sa;
> 
> -	n = num;
> -	sqn = esn_outb_update_sqn(sa, &n);
> -	if (n != num)
> +	/* Calculate number of sequence numbers required */
> +	for (i = 0, nb_sqn = 0; i != num; i++) {
> +		nb_segs[i] = esn_outb_nb_segments(sa, mb[i]);
> +		nb_sqn += nb_segs[i];
> +	}
> +
> +	nb_sqn_alloc = nb_sqn;
> +	sqn = esn_outb_update_sqn(sa, &nb_sqn_alloc);
> +	if (nb_sqn_alloc != nb_sqn)
>  		rte_errno = EOVERFLOW;
> 
>  	k = 0;
> -	for (i = 0; i != n; i++) {
> +	for (i = 0; i != num; i++) {
> 
>  		sqc = rte_cpu_to_be_64(sqn + i);
>  		gen_iv(iv, sqc);
> @@ -741,11 +807,18 @@ inline_outb_trs_pkt_process(const struct rte_ipsec_session *ss,
>  			dr[i - k] = i;
>  			rte_errno = -rc;
>  		}
> +
> +		/**
> +		 * If packet is using tso, increment sqn by the number of
> +		 * segments for	packet
> +		 */
> +		if  (mb[i]->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_UDP_SEG))
> +			sqn += nb_segs[i] - 1;
>  	}
> 
>  	/* copy not processed mbufs beyond good ones */
> -	if (k != n && k != 0)
> -		move_bad_mbufs(mb, dr, n, n - k);
> +	if (k != num && k != 0)
> +		move_bad_mbufs(mb, dr, num, num - k);
> 
>  	inline_outb_mbuf_prepare(ss, mb, k);
>  	return k;
> diff --git a/lib/ipsec/iph.h b/lib/ipsec/iph.h
> index 861f16905a..2d223199ac 100644
> --- a/lib/ipsec/iph.h
> +++ b/lib/ipsec/iph.h
> @@ -6,6 +6,8 @@
>  #define _IPH_H_
> 
>  #include <rte_ip.h>
> +#include <rte_udp.h>
> +#include <rte_tcp.h>
> 
>  /**
>   * @file iph.h
> @@ -39,8 +41,8 @@ insert_esph(char *np, char *op, uint32_t hlen)
> 
>  /* update original ip header fields for transport case */
>  static inline int
> -update_trs_l3hdr(const struct rte_ipsec_sa *sa, void *p, uint32_t plen,
> -		uint32_t l2len, uint32_t l3len, uint8_t proto)
> +update_trs_l34hdrs(const struct rte_ipsec_sa *sa, void *p, uint32_t plen,
> +		uint32_t l2len, uint32_t l3len, uint8_t proto, uint8_t tso)

Hmm... why to change name of the function?

>  {
>  	int32_t rc;
> 
> @@ -51,6 +53,10 @@ update_trs_l3hdr(const struct rte_ipsec_sa *sa, void *p, uint32_t plen,
>  		v4h = p;
>  		rc = v4h->next_proto_id;
>  		v4h->next_proto_id = proto;
> +		if (tso) {
> +			v4h->hdr_checksum = 0;
> +			v4h->total_length = 0;

total_len will be overwritten unconditionally at next line below.

Another question - why it is necessary?
Is it HW specific requirment or ... ?


> +		}
>  		v4h->total_length = rte_cpu_to_be_16(plen - l2len);


>  	/* IPv6 */
>  	} else {
> diff --git a/lib/ipsec/sa.c b/lib/ipsec/sa.c
> index 720e0f365b..2ecbbce0a4 100644
> --- a/lib/ipsec/sa.c
> +++ b/lib/ipsec/sa.c
> @@ -565,6 +565,12 @@ rte_ipsec_sa_init(struct rte_ipsec_sa *sa, const struct rte_ipsec_sa_prm *prm,
>  	sa->type = type;
>  	sa->size = sz;
> 
> +
> +	if (prm->ipsec_xform.options.tso == 1) {
> +		sa->tso.enabled = 1;
> +		sa->tso.mss = prm->ipsec_xform.mss;
> +	}
> +
>  	/* check for ESN flag */
>  	sa->sqn_mask = (prm->ipsec_xform.options.esn == 0) ?
>  		UINT32_MAX : UINT64_MAX;
> diff --git a/lib/ipsec/sa.h b/lib/ipsec/sa.h
> index 107ebd1519..5e237f3525 100644
> --- a/lib/ipsec/sa.h
> +++ b/lib/ipsec/sa.h
> @@ -113,6 +113,10 @@ struct rte_ipsec_sa {
>  	uint8_t iv_len;
>  	uint8_t pad_align;
>  	uint8_t tos_mask;
> +	struct {
> +		uint8_t enabled:1;
> +		uint16_t mss;
> +	} tso;

Wouldn't one field be enough?
uint16_t tso_mss;
And it it is zero, then tso is disabled. 
In fact, do we need it at all?
Wouldn't it be better to request user to fill mbuf->tso_segsz properly for us?

> 
>  	/* template for tunnel header */
>  	uint8_t hdr[IPSEC_MAX_HDR_SIZE];
> --
> 2.25.1