examples/ipsec-secgw: fix buffer freeing in vector mode
[dpdk.git] / examples / ipsec-secgw / ipsec_worker.c
index 647e22d..8639426 100644 (file)
 #include "ipsec-secgw.h"
 #include "ipsec_worker.h"
 
+struct port_drv_mode_data {
+       struct rte_security_session *sess;
+       struct rte_security_ctx *ctx;
+};
+
 static inline enum pkt_type
 process_ipsec_get_pkt_type(struct rte_mbuf *pkt, uint8_t **nlp)
 {
        struct rte_ether_hdr *eth;
+       uint32_t ptype = pkt->packet_type;
 
        eth = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
-       if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
+       rte_prefetch0(eth);
+
+       if (RTE_ETH_IS_IPV4_HDR(ptype)) {
                *nlp = RTE_PTR_ADD(eth, RTE_ETHER_HDR_LEN +
                                offsetof(struct ip, ip_p));
-               if (**nlp == IPPROTO_ESP)
+               if ((ptype & RTE_PTYPE_TUNNEL_MASK) == RTE_PTYPE_TUNNEL_ESP)
                        return PKT_TYPE_IPSEC_IPV4;
                else
                        return PKT_TYPE_PLAIN_IPV4;
-       } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
+       } else if (RTE_ETH_IS_IPV6_HDR(ptype)) {
                *nlp = RTE_PTR_ADD(eth, RTE_ETHER_HDR_LEN +
                                offsetof(struct ip6_hdr, ip6_nxt));
-               if (**nlp == IPPROTO_ESP)
+               if ((ptype & RTE_PTYPE_TUNNEL_MASK) == RTE_PTYPE_TUNNEL_ESP)
                        return PKT_TYPE_IPSEC_IPV6;
                else
                        return PKT_TYPE_PLAIN_IPV6;
@@ -44,8 +52,8 @@ update_mac_addrs(struct rte_mbuf *pkt, uint16_t portid)
        struct rte_ether_hdr *ethhdr;
 
        ethhdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
-       memcpy(&ethhdr->s_addr, &ethaddr_tbl[portid].src, RTE_ETHER_ADDR_LEN);
-       memcpy(&ethhdr->d_addr, &ethaddr_tbl[portid].dst, RTE_ETHER_ADDR_LEN);
+       memcpy(&ethhdr->src_addr, &ethaddr_tbl[portid].src, RTE_ETHER_ADDR_LEN);
+       memcpy(&ethhdr->dst_addr, &ethaddr_tbl[portid].dst, RTE_ETHER_ADDR_LEN);
 }
 
 static inline void
@@ -58,9 +66,29 @@ ipsec_event_pre_forward(struct rte_mbuf *m, unsigned int port_id)
        rte_event_eth_tx_adapter_txq_set(m, 0);
 }
 
+static inline void
+ev_vector_attr_init(struct rte_event_vector *vec)
+{
+       vec->attr_valid = 1;
+       vec->port = 0xFFFF;
+       vec->queue = 0;
+}
+
+static inline void
+ev_vector_attr_update(struct rte_event_vector *vec, struct rte_mbuf *pkt)
+{
+       if (vec->port == 0xFFFF) {
+               vec->port = pkt->port;
+               return;
+       }
+       if (vec->attr_valid && (vec->port != pkt->port))
+               vec->attr_valid = 0;
+}
+
 static inline void
 prepare_out_sessions_tbl(struct sa_ctx *sa_out,
-               struct rte_security_session **sess_tbl, uint16_t size)
+                        struct port_drv_mode_data *data,
+                        uint16_t size)
 {
        struct rte_ipsec_session *pri_sess;
        struct ipsec_sa *sa;
@@ -95,9 +123,10 @@ prepare_out_sessions_tbl(struct sa_ctx *sa_out,
                }
 
                /* Use only first inline session found for a given port */
-               if (sess_tbl[sa->portid])
+               if (data[sa->portid].sess)
                        continue;
-               sess_tbl[sa->portid] = pri_sess->security.ses;
+               data[sa->portid].sess = pri_sess->security.ses;
+               data[sa->portid].ctx = pri_sess->security.ctx;
        }
 }
 
@@ -123,6 +152,76 @@ check_sp(struct sp_ctx *sp, const uint8_t *nlp, uint32_t *sa_idx)
        return 1;
 }
 
+static inline void
+check_sp_bulk(struct sp_ctx *sp, struct traffic_type *ip,
+             struct traffic_type *ipsec)
+{
+       uint32_t i, j, res;
+       struct rte_mbuf *m;
+
+       if (unlikely(sp == NULL || ip->num == 0))
+               return;
+
+       rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res, ip->num,
+                        DEFAULT_MAX_CATEGORIES);
+
+       j = 0;
+       for (i = 0; i < ip->num; i++) {
+               m = ip->pkts[i];
+               res = ip->res[i];
+               if (unlikely(res == DISCARD))
+                       free_pkts(&m, 1);
+               else if (res == BYPASS)
+                       ip->pkts[j++] = m;
+               else {
+                       ipsec->res[ipsec->num] = res - 1;
+                       ipsec->pkts[ipsec->num++] = m;
+               }
+       }
+       ip->num = j;
+}
+
+static inline void
+check_sp_sa_bulk(struct sp_ctx *sp, struct sa_ctx *sa_ctx,
+                struct traffic_type *ip)
+{
+       struct ipsec_sa *sa;
+       uint32_t i, j, res;
+       struct rte_mbuf *m;
+
+       if (unlikely(sp == NULL || ip->num == 0))
+               return;
+
+       rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res, ip->num,
+                        DEFAULT_MAX_CATEGORIES);
+
+       j = 0;
+       for (i = 0; i < ip->num; i++) {
+               m = ip->pkts[i];
+               res = ip->res[i];
+               if (unlikely(res == DISCARD))
+                       free_pkts(&m, 1);
+               else if (res == BYPASS)
+                       ip->pkts[j++] = m;
+               else {
+                       sa = *(struct ipsec_sa **)rte_security_dynfield(m);
+                       if (sa == NULL) {
+                               free_pkts(&m, 1);
+                               continue;
+                       }
+
+                       /* SPI on the packet should match with the one in SA */
+                       if (unlikely(sa->spi != sa_ctx->sa[res - 1].spi)) {
+                               free_pkts(&m, 1);
+                               continue;
+                       }
+
+                       ip->pkts[j++] = m;
+               }
+       }
+       ip->num = j;
+}
+
 static inline uint16_t
 route4_pkt(struct rte_mbuf *pkt, struct rt_ctx *rt_ctx)
 {
@@ -201,9 +300,9 @@ process_ipsec_ev_inbound(struct ipsec_ctx *ctx, struct route_table *rt,
 
        switch (type) {
        case PKT_TYPE_PLAIN_IPV4:
-               if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
+               if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) {
                        if (unlikely(pkt->ol_flags &
-                                    PKT_RX_SEC_OFFLOAD_FAILED)) {
+                                    RTE_MBUF_F_RX_SEC_OFFLOAD_FAILED)) {
                                RTE_LOG(ERR, IPSEC,
                                        "Inbound security offload failed\n");
                                goto drop_pkt_and_exit;
@@ -219,9 +318,9 @@ process_ipsec_ev_inbound(struct ipsec_ctx *ctx, struct route_table *rt,
                break;
 
        case PKT_TYPE_PLAIN_IPV6:
-               if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
+               if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) {
                        if (unlikely(pkt->ol_flags &
-                                    PKT_RX_SEC_OFFLOAD_FAILED)) {
+                                    RTE_MBUF_F_RX_SEC_OFFLOAD_FAILED)) {
                                RTE_LOG(ERR, IPSEC,
                                        "Inbound security offload failed\n");
                                goto drop_pkt_and_exit;
@@ -237,7 +336,8 @@ process_ipsec_ev_inbound(struct ipsec_ctx *ctx, struct route_table *rt,
                break;
 
        default:
-               RTE_LOG(ERR, IPSEC, "Unsupported packet type = %d\n", type);
+               RTE_LOG_DP(DEBUG, IPSEC_ESP, "Unsupported packet type = %d\n",
+                          type);
                goto drop_pkt_and_exit;
        }
 
@@ -336,7 +436,7 @@ process_ipsec_ev_outbound(struct ipsec_ctx *ctx, struct route_table *rt,
        }
 
        /* Validate sa_idx */
-       if (sa_idx >= ctx->sa_ctx->nb_sa)
+       if (unlikely(sa_idx >= ctx->sa_ctx->nb_sa))
                goto drop_pkt_and_exit;
 
        /* Else the packet has to be protected */
@@ -351,22 +451,24 @@ process_ipsec_ev_outbound(struct ipsec_ctx *ctx, struct route_table *rt,
        sess = ipsec_get_primary_session(sa);
 
        /* Allow only inline protocol for now */
-       if (sess->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL) {
+       if (unlikely(sess->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)) {
                RTE_LOG(ERR, IPSEC, "SA type not supported\n");
                goto drop_pkt_and_exit;
        }
 
-       if (sess->security.ol_flags & RTE_SECURITY_TX_OLOAD_NEED_MDATA)
-               *(struct rte_security_session **)rte_security_dynfield(pkt) =
-                               sess->security.ses;
+       rte_security_set_pkt_metadata(sess->security.ctx,
+                                     sess->security.ses, pkt, NULL);
 
        /* Mark the packet for Tx security offload */
-       pkt->ol_flags |= PKT_TX_SEC_OFFLOAD;
+       pkt->ol_flags |= RTE_MBUF_F_TX_SEC_OFFLOAD;
 
        /* Get the port to which this pkt need to be submitted */
        port_id = sa->portid;
 
 send_pkt:
+       /* Provide L2 len for Outbound processing */
+       pkt->l2_len = RTE_ETHER_HDR_LEN;
+
        /* Update mac addresses */
        update_mac_addrs(pkt, port_id);
 
@@ -381,6 +483,254 @@ drop_pkt_and_exit:
        return PKT_DROPPED;
 }
 
+static inline int
+ipsec_ev_route_pkts(struct rte_event_vector *vec, struct route_table *rt,
+                   struct ipsec_traffic *t, struct sa_ctx *sa_ctx)
+{
+       struct rte_ipsec_session *sess;
+       uint32_t sa_idx, i, j = 0;
+       uint16_t port_id = 0;
+       struct rte_mbuf *pkt;
+       struct ipsec_sa *sa;
+
+       /* Route IPv4 packets */
+       for (i = 0; i < t->ip4.num; i++) {
+               pkt = t->ip4.pkts[i];
+               port_id = route4_pkt(pkt, rt->rt4_ctx);
+               if (port_id != RTE_MAX_ETHPORTS) {
+                       /* Update mac addresses */
+                       update_mac_addrs(pkt, port_id);
+                       /* Update the event with the dest port */
+                       ipsec_event_pre_forward(pkt, port_id);
+                       ev_vector_attr_update(vec, pkt);
+                       vec->mbufs[j++] = pkt;
+               } else
+                       free_pkts(&pkt, 1);
+       }
+
+       /* Route IPv6 packets */
+       for (i = 0; i < t->ip6.num; i++) {
+               pkt = t->ip6.pkts[i];
+               port_id = route6_pkt(pkt, rt->rt6_ctx);
+               if (port_id != RTE_MAX_ETHPORTS) {
+                       /* Update mac addresses */
+                       update_mac_addrs(pkt, port_id);
+                       /* Update the event with the dest port */
+                       ipsec_event_pre_forward(pkt, port_id);
+                       ev_vector_attr_update(vec, pkt);
+                       vec->mbufs[j++] = pkt;
+               } else
+                       free_pkts(&pkt, 1);
+       }
+
+       /* Route ESP packets */
+       for (i = 0; i < t->ipsec.num; i++) {
+               /* Validate sa_idx */
+               sa_idx = t->ipsec.res[i];
+               pkt = t->ipsec.pkts[i];
+               if (unlikely(sa_idx >= sa_ctx->nb_sa))
+                       free_pkts(&pkt, 1);
+               else {
+                       /* Else the packet has to be protected */
+                       sa = &(sa_ctx->sa[sa_idx]);
+                       /* Get IPsec session */
+                       sess = ipsec_get_primary_session(sa);
+                       /* Allow only inline protocol for now */
+                       if (unlikely(sess->type !=
+                               RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)) {
+                               RTE_LOG(ERR, IPSEC, "SA type not supported\n");
+                               free_pkts(&pkt, 1);
+                               continue;
+                       }
+                       rte_security_set_pkt_metadata(sess->security.ctx,
+                                               sess->security.ses, pkt, NULL);
+
+                       pkt->ol_flags |= RTE_MBUF_F_TX_SEC_OFFLOAD;
+                       port_id = sa->portid;
+                       update_mac_addrs(pkt, port_id);
+                       ipsec_event_pre_forward(pkt, port_id);
+                       ev_vector_attr_update(vec, pkt);
+                       vec->mbufs[j++] = pkt;
+               }
+       }
+
+       return j;
+}
+
+static inline void
+classify_pkt(struct rte_mbuf *pkt, struct ipsec_traffic *t)
+{
+       enum pkt_type type;
+       uint8_t *nlp;
+
+       /* Check the packet type */
+       type = process_ipsec_get_pkt_type(pkt, &nlp);
+
+       switch (type) {
+       case PKT_TYPE_PLAIN_IPV4:
+               t->ip4.data[t->ip4.num] = nlp;
+               t->ip4.pkts[(t->ip4.num)++] = pkt;
+               break;
+       case PKT_TYPE_PLAIN_IPV6:
+               t->ip6.data[t->ip6.num] = nlp;
+               t->ip6.pkts[(t->ip6.num)++] = pkt;
+               break;
+       default:
+               RTE_LOG_DP(DEBUG, IPSEC_ESP, "Unsupported packet type = %d\n",
+                          type);
+               free_pkts(&pkt, 1);
+               break;
+       }
+}
+
+static inline int
+process_ipsec_ev_inbound_vector(struct ipsec_ctx *ctx, struct route_table *rt,
+                               struct rte_event_vector *vec)
+{
+       struct ipsec_traffic t;
+       struct rte_mbuf *pkt;
+       int i;
+
+       t.ip4.num = 0;
+       t.ip6.num = 0;
+       t.ipsec.num = 0;
+
+       for (i = 0; i < vec->nb_elem; i++) {
+               /* Get pkt from event */
+               pkt = vec->mbufs[i];
+
+               if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) {
+                       if (unlikely(pkt->ol_flags &
+                                    RTE_MBUF_F_RX_SEC_OFFLOAD_FAILED)) {
+                               RTE_LOG(ERR, IPSEC,
+                                       "Inbound security offload failed\n");
+                               free_pkts(&pkt, 1);
+                               continue;
+                       }
+               }
+
+               classify_pkt(pkt, &t);
+       }
+
+       check_sp_sa_bulk(ctx->sp4_ctx, ctx->sa_ctx, &t.ip4);
+       check_sp_sa_bulk(ctx->sp6_ctx, ctx->sa_ctx, &t.ip6);
+
+       return ipsec_ev_route_pkts(vec, rt, &t, ctx->sa_ctx);
+}
+
+static inline int
+process_ipsec_ev_outbound_vector(struct ipsec_ctx *ctx, struct route_table *rt,
+                                struct rte_event_vector *vec)
+{
+       struct ipsec_traffic t;
+       struct rte_mbuf *pkt;
+       uint32_t i;
+
+       t.ip4.num = 0;
+       t.ip6.num = 0;
+       t.ipsec.num = 0;
+
+       for (i = 0; i < vec->nb_elem; i++) {
+               /* Get pkt from event */
+               pkt = vec->mbufs[i];
+
+               classify_pkt(pkt, &t);
+
+               /* Provide L2 len for Outbound processing */
+               pkt->l2_len = RTE_ETHER_HDR_LEN;
+       }
+
+       check_sp_bulk(ctx->sp4_ctx, &t.ip4, &t.ipsec);
+       check_sp_bulk(ctx->sp6_ctx, &t.ip6, &t.ipsec);
+
+       return ipsec_ev_route_pkts(vec, rt, &t, ctx->sa_ctx);
+}
+
+static inline int
+process_ipsec_ev_drv_mode_outbound_vector(struct rte_event_vector *vec,
+                                         struct port_drv_mode_data *data)
+{
+       struct rte_mbuf *pkt;
+       int16_t port_id;
+       uint32_t i;
+       int j = 0;
+
+       for (i = 0; i < vec->nb_elem; i++) {
+               pkt = vec->mbufs[i];
+               port_id = pkt->port;
+
+               if (unlikely(!data[port_id].sess)) {
+                       free_pkts(&pkt, 1);
+                       continue;
+               }
+               ipsec_event_pre_forward(pkt, port_id);
+               /* Save security session */
+               rte_security_set_pkt_metadata(data[port_id].ctx,
+                                             data[port_id].sess, pkt,
+                                             NULL);
+
+               /* Mark the packet for Tx security offload */
+               pkt->ol_flags |= RTE_MBUF_F_TX_SEC_OFFLOAD;
+
+               /* Provide L2 len for Outbound processing */
+               pkt->l2_len = RTE_ETHER_HDR_LEN;
+
+               vec->mbufs[j++] = pkt;
+       }
+
+       return j;
+}
+
+static inline void
+ipsec_ev_vector_process(struct lcore_conf_ev_tx_int_port_wrkr *lconf,
+                       struct eh_event_link_info *links,
+                       struct rte_event *ev)
+{
+       struct rte_event_vector *vec = ev->vec;
+       struct rte_mbuf *pkt;
+       int ret;
+
+       pkt = vec->mbufs[0];
+
+       ev_vector_attr_init(vec);
+       if (is_unprotected_port(pkt->port))
+               ret = process_ipsec_ev_inbound_vector(&lconf->inbound,
+                                                     &lconf->rt, vec);
+       else
+               ret = process_ipsec_ev_outbound_vector(&lconf->outbound,
+                                                      &lconf->rt, vec);
+
+       if (likely(ret > 0)) {
+               vec->nb_elem = ret;
+               rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
+                                                links[0].event_port_id,
+                                                ev, 1, 0);
+       } else {
+               rte_mempool_put(rte_mempool_from_obj(vec), vec);
+       }
+}
+
+static inline void
+ipsec_ev_vector_drv_mode_process(struct eh_event_link_info *links,
+                                struct rte_event *ev,
+                                struct port_drv_mode_data *data)
+{
+       struct rte_event_vector *vec = ev->vec;
+       struct rte_mbuf *pkt;
+
+       pkt = vec->mbufs[0];
+
+       if (!is_unprotected_port(pkt->port))
+               vec->nb_elem = process_ipsec_ev_drv_mode_outbound_vector(vec,
+                                                                        data);
+       if (vec->nb_elem > 0)
+               rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
+                                                links[0].event_port_id,
+                                                ev, 1, 0);
+       else
+               rte_mempool_put(rte_mempool_from_obj(vec), vec);
+}
+
 /*
  * Event mode exposes various operating modes depending on the
  * capabilities of the event device and the operating mode
@@ -398,7 +748,7 @@ static void
 ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
                uint8_t nb_links)
 {
-       struct rte_security_session *sess_tbl[RTE_MAX_ETHPORTS] = { NULL };
+       struct port_drv_mode_data data[RTE_MAX_ETHPORTS];
        unsigned int nb_rx = 0;
        struct rte_mbuf *pkt;
        struct rte_event ev;
@@ -412,6 +762,8 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
                return;
        }
 
+       memset(&data, 0, sizeof(struct port_drv_mode_data));
+
        /* Get core ID */
        lcore_id = rte_lcore_id();
 
@@ -422,8 +774,8 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
         * Prepare security sessions table. In outbound driver mode
         * we always use first session configured for a given port
         */
-       prepare_out_sessions_tbl(socket_ctx[socket_id].sa_out, sess_tbl,
-                       RTE_MAX_ETHPORTS);
+       prepare_out_sessions_tbl(socket_ctx[socket_id].sa_out, data,
+                                RTE_MAX_ETHPORTS);
 
        RTE_LOG(INFO, IPSEC,
                "Launching event mode worker (non-burst - Tx internal port - "
@@ -450,6 +802,19 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
                if (nb_rx == 0)
                        continue;
 
+               switch (ev.event_type) {
+               case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+               case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                       ipsec_ev_vector_drv_mode_process(links, &ev, data);
+                       continue;
+               case RTE_EVENT_TYPE_ETHDEV:
+                       break;
+               default:
+                       RTE_LOG(ERR, IPSEC, "Invalid event type %u",
+                               ev.event_type);
+                       continue;
+               }
+
                pkt = ev.mbuf;
                port_id = pkt->port;
 
@@ -460,19 +825,21 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
 
                if (!is_unprotected_port(port_id)) {
 
-                       if (unlikely(!sess_tbl[port_id])) {
+                       if (unlikely(!data[port_id].sess)) {
                                rte_pktmbuf_free(pkt);
                                continue;
                        }
 
                        /* Save security session */
-                       if (rte_security_dynfield_is_registered())
-                               *(struct rte_security_session **)
-                                       rte_security_dynfield(pkt) =
-                                               sess_tbl[port_id];
+                       rte_security_set_pkt_metadata(data[port_id].ctx,
+                                                     data[port_id].sess, pkt,
+                                                     NULL);
 
                        /* Mark the packet for Tx security offload */
-                       pkt->ol_flags |= PKT_TX_SEC_OFFLOAD;
+                       pkt->ol_flags |= RTE_MBUF_F_TX_SEC_OFFLOAD;
+
+                       /* Provide L2 len for Outbound processing */
+                       pkt->l2_len = RTE_ETHER_HDR_LEN;
                }
 
                /*
@@ -557,10 +924,16 @@ ipsec_wrkr_non_burst_int_port_app_mode(struct eh_event_link_info *links,
                if (nb_rx == 0)
                        continue;
 
-               if (unlikely(ev.event_type != RTE_EVENT_TYPE_ETHDEV)) {
+               switch (ev.event_type) {
+               case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+               case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                       ipsec_ev_vector_process(&lconf, links, &ev);
+                       continue;
+               case RTE_EVENT_TYPE_ETHDEV:
+                       break;
+               default:
                        RTE_LOG(ERR, IPSEC, "Invalid event type %u",
                                ev.event_type);
-
                        continue;
                }