examples/ipsec-secgw: add poll mode worker for inline proto
[dpdk.git] / examples / ipsec-secgw / ipsec_worker.c
index e9493c5..a6305e0 100644 (file)
@@ -17,6 +17,8 @@ struct port_drv_mode_data {
        struct rte_security_ctx *ctx;
 };
 
+typedef void (*ipsec_worker_fn_t)(void);
+
 static inline enum pkt_type
 process_ipsec_get_pkt_type(struct rte_mbuf *pkt, uint8_t **nlp)
 {
@@ -205,12 +207,16 @@ check_sp_sa_bulk(struct sp_ctx *sp, struct sa_ctx *sa_ctx,
                        ip->pkts[j++] = m;
                else {
                        sa = *(struct ipsec_sa **)rte_security_dynfield(m);
-                       if (sa == NULL)
+                       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))
+                       if (unlikely(sa->spi != sa_ctx->sa[res - 1].spi)) {
                                free_pkts(&m, 1);
+                               continue;
+                       }
 
                        ip->pkts[j++] = m;
                }
@@ -536,6 +542,7 @@ ipsec_ev_route_pkts(struct rte_event_vector *vec, struct route_table *rt,
                                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);
@@ -695,11 +702,13 @@ ipsec_ev_vector_process(struct lcore_conf_ev_tx_int_port_wrkr *lconf,
                ret = process_ipsec_ev_outbound_vector(&lconf->outbound,
                                                       &lconf->rt, vec);
 
-       if (ret > 0) {
+       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);
        }
 }
 
@@ -720,6 +729,8 @@ ipsec_ev_vector_drv_mode_process(struct eh_event_link_info *links,
                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);
 }
 
 /*
@@ -728,6 +739,13 @@ ipsec_ev_vector_drv_mode_process(struct eh_event_link_info *links,
  * selected.
  */
 
+static void
+ipsec_event_port_flush(uint8_t eventdev_id __rte_unused, struct rte_event ev,
+                      void *args __rte_unused)
+{
+       rte_pktmbuf_free(ev.mbuf);
+}
+
 /* Workers registered */
 #define IPSEC_EVENTMODE_WORKERS                2
 
@@ -740,7 +758,7 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
                uint8_t nb_links)
 {
        struct port_drv_mode_data data[RTE_MAX_ETHPORTS];
-       unsigned int nb_rx = 0;
+       unsigned int nb_rx = 0, nb_tx;
        struct rte_mbuf *pkt;
        struct rte_event ev;
        uint32_t lcore_id;
@@ -838,12 +856,23 @@ ipsec_wrkr_non_burst_int_port_drv_mode(struct eh_event_link_info *links,
                 * directly enqueued to the adapter and it would be
                 * internally submitted to the eth device.
                 */
-               rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
-                               links[0].event_port_id,
-                               &ev,    /* events */
-                               1,      /* nb_events */
-                               0       /* flags */);
+               nb_tx = rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
+                                                        links[0].event_port_id,
+                                                        &ev, /* events */
+                                                        1,   /* nb_events */
+                                                        0 /* flags */);
+               if (!nb_tx)
+                       rte_pktmbuf_free(ev.mbuf);
+       }
+
+       if (ev.u64) {
+               ev.op = RTE_EVENT_OP_RELEASE;
+               rte_event_enqueue_burst(links[0].eventdev_id,
+                                       links[0].event_port_id, &ev, 1);
        }
+
+       rte_event_port_quiesce(links[0].eventdev_id, links[0].event_port_id,
+                              ipsec_event_port_flush, NULL);
 }
 
 /*
@@ -855,7 +884,7 @@ ipsec_wrkr_non_burst_int_port_app_mode(struct eh_event_link_info *links,
                uint8_t nb_links)
 {
        struct lcore_conf_ev_tx_int_port_wrkr lconf;
-       unsigned int nb_rx = 0;
+       unsigned int nb_rx = 0, nb_tx;
        struct rte_event ev;
        uint32_t lcore_id;
        int32_t socket_id;
@@ -943,12 +972,23 @@ ipsec_wrkr_non_burst_int_port_app_mode(struct eh_event_link_info *links,
                 * directly enqueued to the adapter and it would be
                 * internally submitted to the eth device.
                 */
-               rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
-                               links[0].event_port_id,
-                               &ev,    /* events */
-                               1,      /* nb_events */
-                               0       /* flags */);
+               nb_tx = rte_event_eth_tx_adapter_enqueue(links[0].eventdev_id,
+                                                        links[0].event_port_id,
+                                                        &ev, /* events */
+                                                        1,   /* nb_events */
+                                                        0 /* flags */);
+               if (!nb_tx)
+                       rte_pktmbuf_free(ev.mbuf);
+       }
+
+       if (ev.u64) {
+               ev.op = RTE_EVENT_OP_RELEASE;
+               rte_event_enqueue_burst(links[0].eventdev_id,
+                                       links[0].event_port_id, &ev, 1);
        }
+
+       rte_event_port_quiesce(links[0].eventdev_id, links[0].event_port_id,
+                              ipsec_event_port_flush, NULL);
 }
 
 static uint8_t
@@ -995,6 +1035,374 @@ ipsec_eventmode_worker(struct eh_conf *conf)
        eh_launch_worker(conf, ipsec_wrkr, nb_wrkr_param);
 }
 
+static __rte_always_inline void
+outb_inl_pro_spd_process(struct sp_ctx *sp,
+                        struct sa_ctx *sa_ctx,
+                        struct traffic_type *ip,
+                        struct traffic_type *match,
+                        struct traffic_type *mismatch,
+                        bool match_flag,
+                        struct ipsec_spd_stats *stats)
+{
+       uint32_t prev_sa_idx = UINT32_MAX;
+       struct rte_mbuf *ipsec[MAX_PKT_BURST];
+       struct rte_ipsec_session *ips;
+       uint32_t i, j, j_mis, sa_idx;
+       struct ipsec_sa *sa = NULL;
+       uint32_t ipsec_num = 0;
+       struct rte_mbuf *m;
+       uint64_t satp;
+
+       if (ip->num == 0 || sp == NULL)
+               return;
+
+       rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
+                       ip->num, DEFAULT_MAX_CATEGORIES);
+
+       j = match->num;
+       j_mis = mismatch->num;
+
+       for (i = 0; i < ip->num; i++) {
+               m = ip->pkts[i];
+               sa_idx = ip->res[i] - 1;
+
+               if (unlikely(ip->res[i] == DISCARD)) {
+                       free_pkts(&m, 1);
+
+                       stats->discard++;
+               } else if (unlikely(ip->res[i] == BYPASS)) {
+                       match->pkts[j++] = m;
+
+                       stats->bypass++;
+               } else {
+                       if (prev_sa_idx == UINT32_MAX) {
+                               prev_sa_idx = sa_idx;
+                               sa = &sa_ctx->sa[sa_idx];
+                               ips = ipsec_get_primary_session(sa);
+                               satp = rte_ipsec_sa_type(ips->sa);
+                       }
+
+                       if (sa_idx != prev_sa_idx) {
+                               prep_process_group(sa, ipsec, ipsec_num);
+
+                               /* Prepare packets for outbound */
+                               rte_ipsec_pkt_process(ips, ipsec, ipsec_num);
+
+                               /* Copy to current tr or a different tr */
+                               if (SATP_OUT_IPV4(satp) == match_flag) {
+                                       memcpy(&match->pkts[j], ipsec,
+                                              ipsec_num * sizeof(void *));
+                                       j += ipsec_num;
+                               } else {
+                                       memcpy(&mismatch->pkts[j_mis], ipsec,
+                                              ipsec_num * sizeof(void *));
+                                       j_mis += ipsec_num;
+                               }
+
+                               /* Update to new SA */
+                               sa = &sa_ctx->sa[sa_idx];
+                               ips = ipsec_get_primary_session(sa);
+                               satp = rte_ipsec_sa_type(ips->sa);
+                               ipsec_num = 0;
+                       }
+
+                       ipsec[ipsec_num++] = m;
+                       stats->protect++;
+               }
+       }
+
+       if (ipsec_num) {
+               prep_process_group(sa, ipsec, ipsec_num);
+
+               /* Prepare pacekts for outbound */
+               rte_ipsec_pkt_process(ips, ipsec, ipsec_num);
+
+               /* Copy to current tr or a different tr */
+               if (SATP_OUT_IPV4(satp) == match_flag) {
+                       memcpy(&match->pkts[j], ipsec,
+                              ipsec_num * sizeof(void *));
+                       j += ipsec_num;
+               } else {
+                       memcpy(&mismatch->pkts[j_mis], ipsec,
+                              ipsec_num * sizeof(void *));
+                       j_mis += ipsec_num;
+               }
+       }
+       match->num = j;
+       mismatch->num = j_mis;
+}
+
+/* Poll mode worker when all SA's are of type inline protocol */
+void
+ipsec_poll_mode_wrkr_inl_pr(void)
+{
+       const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
+                       / US_PER_S * BURST_TX_DRAIN_US;
+       struct sp_ctx *sp4_in, *sp6_in, *sp4_out, *sp6_out;
+       struct rte_mbuf *pkts[MAX_PKT_BURST];
+       uint64_t prev_tsc, diff_tsc, cur_tsc;
+       struct ipsec_core_statistics *stats;
+       struct rt_ctx *rt4_ctx, *rt6_ctx;
+       struct sa_ctx *sa_in, *sa_out;
+       struct traffic_type ip4, ip6;
+       struct lcore_rx_queue *rxql;
+       struct rte_mbuf **v4, **v6;
+       struct ipsec_traffic trf;
+       struct lcore_conf *qconf;
+       uint16_t v4_num, v6_num;
+       int32_t socket_id;
+       uint32_t lcore_id;
+       int32_t i, nb_rx;
+       uint16_t portid;
+       uint8_t queueid;
+
+       prev_tsc = 0;
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_conf[lcore_id];
+       rxql = qconf->rx_queue_list;
+       socket_id = rte_lcore_to_socket_id(lcore_id);
+       stats = &core_statistics[lcore_id];
+
+       rt4_ctx = socket_ctx[socket_id].rt_ip4;
+       rt6_ctx = socket_ctx[socket_id].rt_ip6;
+
+       sp4_in = socket_ctx[socket_id].sp_ip4_in;
+       sp6_in = socket_ctx[socket_id].sp_ip6_in;
+       sa_in = socket_ctx[socket_id].sa_in;
+
+       sp4_out = socket_ctx[socket_id].sp_ip4_out;
+       sp6_out = socket_ctx[socket_id].sp_ip6_out;
+       sa_out = socket_ctx[socket_id].sa_out;
+
+       qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
+
+       if (qconf->nb_rx_queue == 0) {
+               RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
+                       lcore_id);
+               return;
+       }
+
+       RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
+
+       for (i = 0; i < qconf->nb_rx_queue; i++) {
+               portid = rxql[i].port_id;
+               queueid = rxql[i].queue_id;
+               RTE_LOG(INFO, IPSEC,
+                       " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
+                       lcore_id, portid, queueid);
+       }
+
+       while (!force_quit) {
+               cur_tsc = rte_rdtsc();
+
+               /* TX queue buffer drain */
+               diff_tsc = cur_tsc - prev_tsc;
+
+               if (unlikely(diff_tsc > drain_tsc)) {
+                       drain_tx_buffers(qconf);
+                       prev_tsc = cur_tsc;
+               }
+
+               for (i = 0; i < qconf->nb_rx_queue; ++i) {
+                       /* Read packets from RX queues */
+                       portid = rxql[i].port_id;
+                       queueid = rxql[i].queue_id;
+                       nb_rx = rte_eth_rx_burst(portid, queueid,
+                                       pkts, MAX_PKT_BURST);
+
+                       if (nb_rx <= 0)
+                               continue;
+
+                       core_stats_update_rx(nb_rx);
+
+                       prepare_traffic(rxql[i].sec_ctx, pkts, &trf, nb_rx);
+
+                       /* Drop any IPsec traffic */
+                       free_pkts(trf.ipsec.pkts, trf.ipsec.num);
+
+                       if (is_unprotected_port(portid)) {
+                               inbound_sp_sa(sp4_in, sa_in, &trf.ip4,
+                                             trf.ip4.num,
+                                             &stats->inbound.spd4);
+
+                               inbound_sp_sa(sp6_in, sa_in, &trf.ip6,
+                                             trf.ip6.num,
+                                             &stats->inbound.spd6);
+
+                               v4 = trf.ip4.pkts;
+                               v4_num = trf.ip4.num;
+                               v6 = trf.ip6.pkts;
+                               v6_num = trf.ip6.num;
+                       } else {
+                               ip4.num = 0;
+                               ip6.num = 0;
+
+                               outb_inl_pro_spd_process(sp4_out, sa_out,
+                                                        &trf.ip4, &ip4, &ip6,
+                                                        true,
+                                                        &stats->outbound.spd4);
+
+                               outb_inl_pro_spd_process(sp6_out, sa_out,
+                                                        &trf.ip6, &ip6, &ip4,
+                                                        false,
+                                                        &stats->outbound.spd6);
+                               v4 = ip4.pkts;
+                               v4_num = ip4.num;
+                               v6 = ip6.pkts;
+                               v6_num = ip6.num;
+                       }
+
+                       route4_pkts(rt4_ctx, v4, v4_num, 0, false);
+                       route6_pkts(rt6_ctx, v6, v6_num);
+               }
+       }
+}
+
+/* Poll mode worker when all SA's are of type inline protocol
+ * and single sa mode is enabled.
+ */
+void
+ipsec_poll_mode_wrkr_inl_pr_ss(void)
+{
+       const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
+                       / US_PER_S * BURST_TX_DRAIN_US;
+       uint16_t sa_out_portid = 0, sa_out_proto = 0;
+       struct rte_mbuf *pkts[MAX_PKT_BURST], *pkt;
+       uint64_t prev_tsc, diff_tsc, cur_tsc;
+       struct rte_ipsec_session *ips = NULL;
+       struct lcore_rx_queue *rxql;
+       struct ipsec_sa *sa = NULL;
+       struct lcore_conf *qconf;
+       struct sa_ctx *sa_out;
+       uint32_t i, nb_rx, j;
+       int32_t socket_id;
+       uint32_t lcore_id;
+       uint16_t portid;
+       uint8_t queueid;
+
+       prev_tsc = 0;
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_conf[lcore_id];
+       rxql = qconf->rx_queue_list;
+       socket_id = rte_lcore_to_socket_id(lcore_id);
+
+       /* Get SA info */
+       sa_out = socket_ctx[socket_id].sa_out;
+       if (sa_out && single_sa_idx < sa_out->nb_sa) {
+               sa = &sa_out->sa[single_sa_idx];
+               ips = ipsec_get_primary_session(sa);
+               sa_out_portid = sa->portid;
+               if (sa->flags & IP6_TUNNEL)
+                       sa_out_proto = IPPROTO_IPV6;
+               else
+                       sa_out_proto = IPPROTO_IP;
+       }
+
+       qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
+
+       if (qconf->nb_rx_queue == 0) {
+               RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
+                       lcore_id);
+               return;
+       }
+
+       RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
+
+       for (i = 0; i < qconf->nb_rx_queue; i++) {
+               portid = rxql[i].port_id;
+               queueid = rxql[i].queue_id;
+               RTE_LOG(INFO, IPSEC,
+                       " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
+                       lcore_id, portid, queueid);
+       }
+
+       while (!force_quit) {
+               cur_tsc = rte_rdtsc();
+
+               /* TX queue buffer drain */
+               diff_tsc = cur_tsc - prev_tsc;
+
+               if (unlikely(diff_tsc > drain_tsc)) {
+                       drain_tx_buffers(qconf);
+                       prev_tsc = cur_tsc;
+               }
+
+               for (i = 0; i < qconf->nb_rx_queue; ++i) {
+                       /* Read packets from RX queues */
+                       portid = rxql[i].port_id;
+                       queueid = rxql[i].queue_id;
+                       nb_rx = rte_eth_rx_burst(portid, queueid,
+                                                pkts, MAX_PKT_BURST);
+
+                       if (nb_rx <= 0)
+                               continue;
+
+                       core_stats_update_rx(nb_rx);
+
+                       if (is_unprotected_port(portid)) {
+                               /* Nothing much to do for inbound inline
+                                * decrypted traffic.
+                                */
+                               for (j = 0; j < nb_rx; j++) {
+                                       uint32_t ptype, proto;
+
+                                       pkt = pkts[j];
+                                       ptype = pkt->packet_type &
+                                               RTE_PTYPE_L3_MASK;
+                                       if (ptype == RTE_PTYPE_L3_IPV4)
+                                               proto = IPPROTO_IP;
+                                       else
+                                               proto = IPPROTO_IPV6;
+
+                                       send_single_packet(pkt, portid, proto);
+                               }
+
+                               continue;
+                       }
+
+                       /* Free packets if there are no outbound sessions */
+                       if (unlikely(!ips)) {
+                               rte_pktmbuf_free_bulk(pkts, nb_rx);
+                               continue;
+                       }
+
+                       rte_ipsec_pkt_process(ips, pkts, nb_rx);
+
+                       /* Send pkts out */
+                       for (j = 0; j < nb_rx; j++) {
+                               pkt = pkts[j];
+
+                               pkt->l2_len = RTE_ETHER_HDR_LEN;
+                               send_single_packet(pkt, sa_out_portid,
+                                                  sa_out_proto);
+                       }
+               }
+       }
+}
+
+static void
+ipsec_poll_mode_wrkr_launch(void)
+{
+       static ipsec_worker_fn_t poll_mode_wrkrs[MAX_F] = {
+               [INL_PR_F]        = ipsec_poll_mode_wrkr_inl_pr,
+               [INL_PR_F | SS_F] = ipsec_poll_mode_wrkr_inl_pr_ss,
+       };
+       ipsec_worker_fn_t fn;
+
+       if (!app_sa_prm.enable) {
+               fn = ipsec_poll_mode_worker;
+       } else {
+               fn = poll_mode_wrkrs[wrkr_flags];
+
+               /* Always default to all mode worker */
+               if (!fn)
+                       fn = ipsec_poll_mode_worker;
+       }
+
+       /* Launch worker */
+       (*fn)();
+}
+
 int ipsec_launch_one_lcore(void *args)
 {
        struct eh_conf *conf;
@@ -1003,7 +1411,7 @@ int ipsec_launch_one_lcore(void *args)
 
        if (conf->mode == EH_PKT_TRANSFER_MODE_POLL) {
                /* Run in poll mode */
-               ipsec_poll_mode_worker();
+               ipsec_poll_mode_wrkr_launch();
        } else if (conf->mode == EH_PKT_TRANSFER_MODE_EVENT) {
                /* Run in event mode */
                ipsec_eventmode_worker(conf);