examples/ipsec-secgw: support flow director
[dpdk.git] / examples / ipsec-secgw / ipsec.c
index 144f0aa..bf88d80 100644 (file)
@@ -1,34 +1,5 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2016 Intel Corporation. All rights reserved.
- *   All rights reserved.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Intel Corporation nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2016-2020 Intel Corporation
  */
 #include <sys/types.h>
 #include <netinet/in.h>
 #include <rte_branch_prediction.h>
 #include <rte_log.h>
 #include <rte_crypto.h>
+#include <rte_security.h>
 #include <rte_cryptodev.h>
+#include <rte_ipsec.h>
+#include <rte_ethdev.h>
 #include <rte_mbuf.h>
 #include <rte_hash.h>
 
 #include "ipsec.h"
 #include "esp.h"
 
-static inline int
-create_session(struct ipsec_ctx *ipsec_ctx __rte_unused, struct ipsec_sa *sa)
+static inline void
+set_ipsec_conf(struct ipsec_sa *sa, struct rte_security_ipsec_xform *ipsec)
+{
+       if (ipsec->mode == RTE_SECURITY_IPSEC_SA_MODE_TUNNEL) {
+               struct rte_security_ipsec_tunnel_param *tunnel =
+                               &ipsec->tunnel;
+               if (IS_IP4_TUNNEL(sa->flags)) {
+                       tunnel->type =
+                               RTE_SECURITY_IPSEC_TUNNEL_IPV4;
+                       tunnel->ipv4.ttl = IPDEFTTL;
+
+                       memcpy((uint8_t *)&tunnel->ipv4.src_ip,
+                               (uint8_t *)&sa->src.ip.ip4, 4);
+
+                       memcpy((uint8_t *)&tunnel->ipv4.dst_ip,
+                               (uint8_t *)&sa->dst.ip.ip4, 4);
+               } else if (IS_IP6_TUNNEL(sa->flags)) {
+                       tunnel->type =
+                               RTE_SECURITY_IPSEC_TUNNEL_IPV6;
+                       tunnel->ipv6.hlimit = IPDEFTTL;
+                       tunnel->ipv6.dscp = 0;
+                       tunnel->ipv6.flabel = 0;
+
+                       memcpy((uint8_t *)&tunnel->ipv6.src_addr,
+                               (uint8_t *)&sa->src.ip.ip6.ip6_b, 16);
+
+                       memcpy((uint8_t *)&tunnel->ipv6.dst_addr,
+                               (uint8_t *)&sa->dst.ip.ip6.ip6_b, 16);
+               }
+               /* TODO support for Transport */
+       }
+       ipsec->esn_soft_limit = IPSEC_OFFLOAD_ESN_SOFTLIMIT;
+       ipsec->replay_win_sz = app_sa_prm.window_size;
+       ipsec->options.esn = app_sa_prm.enable_esn;
+}
+
+int
+create_lookaside_session(struct ipsec_ctx *ipsec_ctx, struct ipsec_sa *sa,
+               struct rte_ipsec_session *ips)
 {
+       struct rte_cryptodev_info cdev_info;
        unsigned long cdev_id_qp = 0;
-       int32_t ret;
+       int32_t ret = 0;
        struct cdev_key key = { 0 };
 
        key.lcore_id = (uint8_t)rte_lcore_id();
 
        key.cipher_algo = (uint8_t)sa->cipher_algo;
        key.auth_algo = (uint8_t)sa->auth_algo;
+       key.aead_algo = (uint8_t)sa->aead_algo;
 
        ret = rte_hash_lookup_data(ipsec_ctx->cdev_map, &key,
                        (void **)&cdev_id_qp);
        if (ret < 0) {
-               RTE_LOG(ERR, IPSEC, "No cryptodev: core %u, cipher_algo %u, "
-                               "auth_algo %u\n", key.lcore_id, key.cipher_algo,
-                               key.auth_algo);
+               RTE_LOG(ERR, IPSEC,
+                               "No cryptodev: core %u, cipher_algo %u, "
+                               "auth_algo %u, aead_algo %u\n",
+                               key.lcore_id,
+                               key.cipher_algo,
+                               key.auth_algo,
+                               key.aead_algo);
                return -1;
        }
 
@@ -70,45 +87,444 @@ create_session(struct ipsec_ctx *ipsec_ctx __rte_unused, struct ipsec_sa *sa)
                        ipsec_ctx->tbl[cdev_id_qp].id,
                        ipsec_ctx->tbl[cdev_id_qp].qp);
 
-       sa->crypto_session = rte_cryptodev_sym_session_create(
-                       ipsec_ctx->tbl[cdev_id_qp].id, sa->xforms);
+       if (ips->type != RTE_SECURITY_ACTION_TYPE_NONE &&
+               ips->type != RTE_SECURITY_ACTION_TYPE_CPU_CRYPTO) {
+               struct rte_security_session_conf sess_conf = {
+                       .action_type = ips->type,
+                       .protocol = RTE_SECURITY_PROTOCOL_IPSEC,
+                       {.ipsec = {
+                               .spi = sa->spi,
+                               .salt = sa->salt,
+                               .options = { 0 },
+                               .replay_win_sz = 0,
+                               .direction = sa->direction,
+                               .proto = RTE_SECURITY_IPSEC_SA_PROTO_ESP,
+                               .mode = (IS_TUNNEL(sa->flags)) ?
+                                       RTE_SECURITY_IPSEC_SA_MODE_TUNNEL :
+                                       RTE_SECURITY_IPSEC_SA_MODE_TRANSPORT,
+                       } },
+                       .crypto_xform = sa->xforms,
+                       .userdata = NULL,
+
+               };
+
+               if (ips->type == RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL) {
+                       struct rte_security_ctx *ctx = (struct rte_security_ctx *)
+                                                       rte_cryptodev_get_sec_ctx(
+                                                       ipsec_ctx->tbl[cdev_id_qp].id);
+
+                       /* Set IPsec parameters in conf */
+                       set_ipsec_conf(sa, &(sess_conf.ipsec));
+
+                       ips->security.ses = rte_security_session_create(ctx,
+                                       &sess_conf, ipsec_ctx->session_priv_pool);
+                       if (ips->security.ses == NULL) {
+                               RTE_LOG(ERR, IPSEC,
+                               "SEC Session init failed: err: %d\n", ret);
+                               return -1;
+                       }
+               } else {
+                       RTE_LOG(ERR, IPSEC, "Inline not supported\n");
+                       return -1;
+               }
+       } else {
+               if (ips->type == RTE_SECURITY_ACTION_TYPE_CPU_CRYPTO) {
+                       struct rte_cryptodev_info info;
+                       uint16_t cdev_id;
+
+                       cdev_id = ipsec_ctx->tbl[cdev_id_qp].id;
+                       rte_cryptodev_info_get(cdev_id, &info);
+                       if (!(info.feature_flags &
+                               RTE_CRYPTODEV_FF_SYM_CPU_CRYPTO))
+                               return -ENOTSUP;
+
+                       ips->crypto.dev_id = cdev_id;
+               }
+               ips->crypto.ses = rte_cryptodev_sym_session_create(
+                               ipsec_ctx->session_pool);
+               rte_cryptodev_sym_session_init(ipsec_ctx->tbl[cdev_id_qp].id,
+                               ips->crypto.ses, sa->xforms,
+                               ipsec_ctx->session_priv_pool);
+
+               rte_cryptodev_info_get(ipsec_ctx->tbl[cdev_id_qp].id,
+                               &cdev_info);
+       }
 
        sa->cdev_id_qp = cdev_id_qp;
 
        return 0;
 }
 
-static inline void
-enqueue_cop(struct cdev_qp *cqp, struct rte_crypto_op *cop)
+int
+create_inline_session(struct socket_ctx *skt_ctx, struct ipsec_sa *sa,
+               struct rte_ipsec_session *ips)
 {
-       int32_t ret, i;
+       int32_t ret = 0;
+       struct rte_security_ctx *sec_ctx;
+       struct rte_security_session_conf sess_conf = {
+               .action_type = ips->type,
+               .protocol = RTE_SECURITY_PROTOCOL_IPSEC,
+               {.ipsec = {
+                       .spi = sa->spi,
+                       .salt = sa->salt,
+                       .options = { 0 },
+                       .replay_win_sz = 0,
+                       .direction = sa->direction,
+                       .proto = RTE_SECURITY_IPSEC_SA_PROTO_ESP,
+                       .mode = (sa->flags == IP4_TUNNEL ||
+                                       sa->flags == IP6_TUNNEL) ?
+                                       RTE_SECURITY_IPSEC_SA_MODE_TUNNEL :
+                                       RTE_SECURITY_IPSEC_SA_MODE_TRANSPORT,
+               } },
+               .crypto_xform = sa->xforms,
+               .userdata = NULL,
+       };
+
+       RTE_LOG_DP(DEBUG, IPSEC, "Create session for SA spi %u on port %u\n",
+               sa->spi, sa->portid);
+
+       if (ips->type == RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO) {
+               struct rte_flow_error err;
+               const struct rte_security_capability *sec_cap;
+               int ret = 0;
+
+               sec_ctx = (struct rte_security_ctx *)
+                                       rte_eth_dev_get_sec_ctx(
+                                       sa->portid);
+               if (sec_ctx == NULL) {
+                       RTE_LOG(ERR, IPSEC,
+                               " rte_eth_dev_get_sec_ctx failed\n");
+                       return -1;
+               }
 
-       cqp->buf[cqp->len++] = cop;
+               ips->security.ses = rte_security_session_create(sec_ctx,
+                               &sess_conf, skt_ctx->session_pool);
+               if (ips->security.ses == NULL) {
+                       RTE_LOG(ERR, IPSEC,
+                               "SEC Session init failed: err: %d\n", ret);
+                       return -1;
+               }
 
-       if (cqp->len == MAX_PKT_BURST) {
-               ret = rte_cryptodev_enqueue_burst(cqp->id, cqp->qp,
-                               cqp->buf, cqp->len);
-               if (ret < cqp->len) {
-                       RTE_LOG_DP(DEBUG, IPSEC, "Cryptodev %u queue %u:"
-                                       " enqueued %u crypto ops out of %u\n",
-                                        cqp->id, cqp->qp,
-                                        ret, cqp->len);
-                       for (i = ret; i < cqp->len; i++)
-                               rte_pktmbuf_free(cqp->buf[i]->sym->m_src);
+               sec_cap = rte_security_capabilities_get(sec_ctx);
+
+               /* iterate until ESP tunnel*/
+               while (sec_cap->action != RTE_SECURITY_ACTION_TYPE_NONE) {
+                       if (sec_cap->action == ips->type &&
+                           sec_cap->protocol ==
+                               RTE_SECURITY_PROTOCOL_IPSEC &&
+                           sec_cap->ipsec.mode ==
+                               RTE_SECURITY_IPSEC_SA_MODE_TUNNEL &&
+                           sec_cap->ipsec.direction == sa->direction)
+                               break;
+                       sec_cap++;
+               }
+
+               if (sec_cap->action == RTE_SECURITY_ACTION_TYPE_NONE) {
+                       RTE_LOG(ERR, IPSEC,
+                               "No suitable security capability found\n");
+                       return -1;
+               }
+
+               ips->security.ol_flags = sec_cap->ol_flags;
+               ips->security.ctx = sec_ctx;
+               sa->pattern[0].type = RTE_FLOW_ITEM_TYPE_ETH;
+
+               if (IS_IP6(sa->flags)) {
+                       sa->pattern[1].mask = &rte_flow_item_ipv6_mask;
+                       sa->pattern[1].type = RTE_FLOW_ITEM_TYPE_IPV6;
+                       sa->pattern[1].spec = &sa->ipv6_spec;
+
+                       memcpy(sa->ipv6_spec.hdr.dst_addr,
+                               sa->dst.ip.ip6.ip6_b, 16);
+                       memcpy(sa->ipv6_spec.hdr.src_addr,
+                              sa->src.ip.ip6.ip6_b, 16);
+               } else if (IS_IP4(sa->flags)) {
+                       sa->pattern[1].mask = &rte_flow_item_ipv4_mask;
+                       sa->pattern[1].type = RTE_FLOW_ITEM_TYPE_IPV4;
+                       sa->pattern[1].spec = &sa->ipv4_spec;
+
+                       sa->ipv4_spec.hdr.dst_addr = sa->dst.ip.ip4;
+                       sa->ipv4_spec.hdr.src_addr = sa->src.ip.ip4;
+               }
+
+               sa->pattern[2].type = RTE_FLOW_ITEM_TYPE_ESP;
+               sa->pattern[2].spec = &sa->esp_spec;
+               sa->pattern[2].mask = &rte_flow_item_esp_mask;
+               sa->esp_spec.hdr.spi = rte_cpu_to_be_32(sa->spi);
+
+               sa->pattern[3].type = RTE_FLOW_ITEM_TYPE_END;
+
+               sa->action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
+               sa->action[0].conf = ips->security.ses;
+
+               sa->action[1].type = RTE_FLOW_ACTION_TYPE_END;
+
+               sa->attr.egress = (sa->direction ==
+                               RTE_SECURITY_IPSEC_SA_DIR_EGRESS);
+               sa->attr.ingress = (sa->direction ==
+                               RTE_SECURITY_IPSEC_SA_DIR_INGRESS);
+               if (sa->attr.ingress) {
+                       uint8_t rss_key[40];
+                       struct rte_eth_rss_conf rss_conf = {
+                               .rss_key = rss_key,
+                               .rss_key_len = 40,
+                       };
+                       struct rte_eth_dev_info dev_info;
+                       uint16_t queue[RTE_MAX_QUEUES_PER_PORT];
+                       struct rte_flow_action_rss action_rss;
+                       unsigned int i;
+                       unsigned int j;
+
+                       /* Don't create flow if default flow is created */
+                       if (flow_info_tbl[sa->portid].rx_def_flow)
+                               return 0;
+
+                       ret = rte_eth_dev_info_get(sa->portid, &dev_info);
+                       if (ret != 0) {
+                               RTE_LOG(ERR, IPSEC,
+                                       "Error during getting device (port %u) info: %s\n",
+                                       sa->portid, strerror(-ret));
+                               return ret;
+                       }
+
+                       sa->action[2].type = RTE_FLOW_ACTION_TYPE_END;
+                       /* Try RSS. */
+                       sa->action[1].type = RTE_FLOW_ACTION_TYPE_RSS;
+                       sa->action[1].conf = &action_rss;
+                       ret = rte_eth_dev_rss_hash_conf_get(sa->portid,
+                                       &rss_conf);
+                       if (ret != 0) {
+                               RTE_LOG(ERR, IPSEC,
+                                       "rte_eth_dev_rss_hash_conf_get:ret=%d\n",
+                                       ret);
+                               return -1;
+                       }
+                       for (i = 0, j = 0; i < dev_info.nb_rx_queues; ++i)
+                               queue[j++] = i;
+
+                       action_rss = (struct rte_flow_action_rss){
+                                       .types = rss_conf.rss_hf,
+                                       .key_len = rss_conf.rss_key_len,
+                                       .queue_num = j,
+                                       .key = rss_key,
+                                       .queue = queue,
+                       };
+                       ret = rte_flow_validate(sa->portid, &sa->attr,
+                                               sa->pattern, sa->action,
+                                               &err);
+                       if (!ret)
+                               goto flow_create;
+                       /* Try Queue. */
+                       sa->action[1].type = RTE_FLOW_ACTION_TYPE_QUEUE;
+                       sa->action[1].conf =
+                               &(struct rte_flow_action_queue){
+                               .index = 0,
+                       };
+                       ret = rte_flow_validate(sa->portid, &sa->attr,
+                                               sa->pattern, sa->action,
+                                               &err);
+                       /* Try End. */
+                       sa->action[1].type = RTE_FLOW_ACTION_TYPE_END;
+                       sa->action[1].conf = NULL;
+                       ret = rte_flow_validate(sa->portid, &sa->attr,
+                                               sa->pattern, sa->action,
+                                               &err);
+                       if (ret)
+                               goto flow_create_failure;
+               } else if (sa->attr.egress &&
+                               (ips->security.ol_flags &
+                                       RTE_SECURITY_TX_HW_TRAILER_OFFLOAD)) {
+                       sa->action[1].type =
+                                       RTE_FLOW_ACTION_TYPE_PASSTHRU;
+                       sa->action[2].type =
+                                       RTE_FLOW_ACTION_TYPE_END;
+               }
+flow_create:
+               sa->flow = rte_flow_create(sa->portid,
+                               &sa->attr, sa->pattern, sa->action, &err);
+               if (sa->flow == NULL) {
+flow_create_failure:
+                       RTE_LOG(ERR, IPSEC,
+                               "Failed to create ipsec flow msg: %s\n",
+                               err.message);
+                       return -1;
+               }
+       } else if (ips->type == RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL) {
+               const struct rte_security_capability *sec_cap;
+
+               sec_ctx = (struct rte_security_ctx *)
+                               rte_eth_dev_get_sec_ctx(sa->portid);
+
+               if (sec_ctx == NULL) {
+                       RTE_LOG(ERR, IPSEC,
+                               "Ethernet device doesn't have security features registered\n");
+                       return -1;
+               }
+
+               /* Set IPsec parameters in conf */
+               set_ipsec_conf(sa, &(sess_conf.ipsec));
+
+               /* Save SA as userdata for the security session. When
+                * the packet is received, this userdata will be
+                * retrieved using the metadata from the packet.
+                *
+                * The PMD is expected to set similar metadata for other
+                * operations, like rte_eth_event, which are tied to
+                * security session. In such cases, the userdata could
+                * be obtained to uniquely identify the security
+                * parameters denoted.
+                */
+
+               sess_conf.userdata = (void *) sa;
+
+               ips->security.ses = rte_security_session_create(sec_ctx,
+                                       &sess_conf, skt_ctx->session_pool);
+               if (ips->security.ses == NULL) {
+                       RTE_LOG(ERR, IPSEC,
+                               "SEC Session init failed: err: %d\n", ret);
+                       return -1;
+               }
+
+               sec_cap = rte_security_capabilities_get(sec_ctx);
+               if (sec_cap == NULL) {
+                       RTE_LOG(ERR, IPSEC,
+                               "No capabilities registered\n");
+                       return -1;
                }
-               cqp->in_flight += ret;
-               cqp->len = 0;
+
+               /* iterate until ESP tunnel*/
+               while (sec_cap->action !=
+                               RTE_SECURITY_ACTION_TYPE_NONE) {
+                       if (sec_cap->action == ips->type &&
+                           sec_cap->protocol ==
+                               RTE_SECURITY_PROTOCOL_IPSEC &&
+                           sec_cap->ipsec.mode ==
+                               sess_conf.ipsec.mode &&
+                           sec_cap->ipsec.direction == sa->direction)
+                               break;
+                       sec_cap++;
+               }
+
+               if (sec_cap->action == RTE_SECURITY_ACTION_TYPE_NONE) {
+                       RTE_LOG(ERR, IPSEC,
+                               "No suitable security capability found\n");
+                       return -1;
+               }
+
+               ips->security.ol_flags = sec_cap->ol_flags;
+               ips->security.ctx = sec_ctx;
        }
+
+       return 0;
+}
+
+int
+create_ipsec_esp_flow(struct ipsec_sa *sa)
+{
+       int ret = 0;
+       struct rte_flow_error err;
+       if (sa->direction == RTE_SECURITY_IPSEC_SA_DIR_EGRESS) {
+               RTE_LOG(ERR, IPSEC,
+                       "No Flow director rule for Egress traffic\n");
+               return -1;
+       }
+       if (sa->flags == TRANSPORT) {
+               RTE_LOG(ERR, IPSEC,
+                       "No Flow director rule for transport mode\n");
+               return -1;
+       }
+       sa->action[0].type = RTE_FLOW_ACTION_TYPE_QUEUE;
+       sa->pattern[0].type = RTE_FLOW_ITEM_TYPE_ETH;
+       sa->action[0].conf = &(struct rte_flow_action_queue) {
+                               .index = sa->fdir_qid,
+       };
+       sa->attr.egress = 0;
+       sa->attr.ingress = 1;
+       if (IS_IP6(sa->flags)) {
+               sa->pattern[1].mask = &rte_flow_item_ipv6_mask;
+               sa->pattern[1].type = RTE_FLOW_ITEM_TYPE_IPV6;
+               sa->pattern[1].spec = &sa->ipv6_spec;
+               memcpy(sa->ipv6_spec.hdr.dst_addr,
+                       sa->dst.ip.ip6.ip6_b, sizeof(sa->dst.ip.ip6.ip6_b));
+               memcpy(sa->ipv6_spec.hdr.src_addr,
+                       sa->src.ip.ip6.ip6_b, sizeof(sa->src.ip.ip6.ip6_b));
+               sa->pattern[2].type = RTE_FLOW_ITEM_TYPE_ESP;
+               sa->pattern[2].spec = &sa->esp_spec;
+               sa->pattern[2].mask = &rte_flow_item_esp_mask;
+               sa->esp_spec.hdr.spi = rte_cpu_to_be_32(sa->spi);
+               sa->pattern[3].type = RTE_FLOW_ITEM_TYPE_END;
+       } else if (IS_IP4(sa->flags)) {
+               sa->pattern[1].mask = &rte_flow_item_ipv4_mask;
+               sa->pattern[1].type = RTE_FLOW_ITEM_TYPE_IPV4;
+               sa->pattern[1].spec = &sa->ipv4_spec;
+               sa->ipv4_spec.hdr.dst_addr = sa->dst.ip.ip4;
+               sa->ipv4_spec.hdr.src_addr = sa->src.ip.ip4;
+               sa->pattern[2].type = RTE_FLOW_ITEM_TYPE_ESP;
+               sa->pattern[2].spec = &sa->esp_spec;
+               sa->pattern[2].mask = &rte_flow_item_esp_mask;
+               sa->esp_spec.hdr.spi = rte_cpu_to_be_32(sa->spi);
+               sa->pattern[3].type = RTE_FLOW_ITEM_TYPE_END;
+       }
+       sa->action[1].type = RTE_FLOW_ACTION_TYPE_END;
+
+       ret = rte_flow_validate(sa->portid, &sa->attr, sa->pattern, sa->action,
+                               &err);
+       if (ret < 0) {
+               RTE_LOG(ERR, IPSEC, "Flow validation failed %s\n", err.message);
+               return ret;
+       }
+
+       sa->flow = rte_flow_create(sa->portid, &sa->attr, sa->pattern,
+                                       sa->action, &err);
+       if (!sa->flow) {
+               RTE_LOG(ERR, IPSEC, "Flow creation failed %s\n", err.message);
+               return -1;
+       }
+
+       return 0;
+}
+
+/*
+ * queue crypto-ops into PMD queue.
+ */
+void
+enqueue_cop_burst(struct cdev_qp *cqp)
+{
+       uint32_t i, len, ret;
+
+       len = cqp->len;
+       ret = rte_cryptodev_enqueue_burst(cqp->id, cqp->qp, cqp->buf, len);
+       if (ret < len) {
+               RTE_LOG_DP(DEBUG, IPSEC, "Cryptodev %u queue %u:"
+                       " enqueued %u crypto ops out of %u\n",
+                       cqp->id, cqp->qp, ret, len);
+                       /* drop packets that we fail to enqueue */
+                       for (i = ret; i < len; i++)
+                               rte_pktmbuf_free(cqp->buf[i]->sym->m_src);
+       }
+       cqp->in_flight += ret;
+       cqp->len = 0;
+}
+
+static inline void
+enqueue_cop(struct cdev_qp *cqp, struct rte_crypto_op *cop)
+{
+       cqp->buf[cqp->len++] = cop;
+
+       if (cqp->len == MAX_PKT_BURST)
+               enqueue_cop_burst(cqp);
 }
 
 static inline void
 ipsec_enqueue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
-               struct rte_mbuf *pkts[], struct ipsec_sa *sas[],
+               struct rte_mbuf *pkts[], void *sas[],
                uint16_t nb_pkts)
 {
        int32_t ret = 0, i;
        struct ipsec_mbuf_metadata *priv;
+       struct rte_crypto_sym_op *sym_cop;
        struct ipsec_sa *sa;
+       struct rte_ipsec_session *ips;
 
        for (i = 0; i < nb_pkts; i++) {
                if (unlikely(sas[i] == NULL)) {
@@ -120,28 +536,89 @@ ipsec_enqueue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
                rte_prefetch0(pkts[i]);
 
                priv = get_priv(pkts[i]);
-               sa = sas[i];
+               sa = ipsec_mask_saptr(sas[i]);
                priv->sa = sa;
+               ips = ipsec_get_primary_session(sa);
 
-               priv->cop.type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
-               priv->cop.status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+               switch (ips->type) {
+               case RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL:
+                       priv->cop.type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+                       priv->cop.status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
 
-               rte_prefetch0(&priv->sym_cop);
-               priv->cop.sym = &priv->sym_cop;
+                       rte_prefetch0(&priv->sym_cop);
 
-               if ((unlikely(sa->crypto_session == NULL)) &&
-                               create_session(ipsec_ctx, sa)) {
-                       rte_pktmbuf_free(pkts[i]);
-                       continue;
-               }
+                       if ((unlikely(ips->security.ses == NULL)) &&
+                               create_lookaside_session(ipsec_ctx, sa, ips)) {
+                               rte_pktmbuf_free(pkts[i]);
+                               continue;
+                       }
 
-               rte_crypto_op_attach_sym_session(&priv->cop,
-                               sa->crypto_session);
+                       sym_cop = get_sym_cop(&priv->cop);
+                       sym_cop->m_src = pkts[i];
 
-               ret = xform_func(pkts[i], sa, &priv->cop);
-               if (unlikely(ret)) {
+                       rte_security_attach_session(&priv->cop,
+                               ips->security.ses);
+                       break;
+
+               case RTE_SECURITY_ACTION_TYPE_CPU_CRYPTO:
+                       RTE_LOG(ERR, IPSEC, "CPU crypto is not supported by the"
+                                       " legacy mode.");
                        rte_pktmbuf_free(pkts[i]);
                        continue;
+
+               case RTE_SECURITY_ACTION_TYPE_NONE:
+
+                       priv->cop.type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+                       priv->cop.status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+
+                       rte_prefetch0(&priv->sym_cop);
+
+                       if ((unlikely(ips->crypto.ses == NULL)) &&
+                               create_lookaside_session(ipsec_ctx, sa, ips)) {
+                               rte_pktmbuf_free(pkts[i]);
+                               continue;
+                       }
+
+                       rte_crypto_op_attach_sym_session(&priv->cop,
+                                       ips->crypto.ses);
+
+                       ret = xform_func(pkts[i], sa, &priv->cop);
+                       if (unlikely(ret)) {
+                               rte_pktmbuf_free(pkts[i]);
+                               continue;
+                       }
+                       break;
+               case RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL:
+                       RTE_ASSERT(ips->security.ses != NULL);
+                       ipsec_ctx->ol_pkts[ipsec_ctx->ol_pkts_cnt++] = pkts[i];
+                       if (ips->security.ol_flags &
+                               RTE_SECURITY_TX_OLOAD_NEED_MDATA)
+                               rte_security_set_pkt_metadata(
+                                       ips->security.ctx, ips->security.ses,
+                                       pkts[i], NULL);
+                       continue;
+               case RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO:
+                       RTE_ASSERT(ips->security.ses != NULL);
+                       priv->cop.type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+                       priv->cop.status = RTE_CRYPTO_OP_STATUS_NOT_PROCESSED;
+
+                       rte_prefetch0(&priv->sym_cop);
+                       rte_security_attach_session(&priv->cop,
+                                       ips->security.ses);
+
+                       ret = xform_func(pkts[i], sa, &priv->cop);
+                       if (unlikely(ret)) {
+                               rte_pktmbuf_free(pkts[i]);
+                               continue;
+                       }
+
+                       ipsec_ctx->ol_pkts[ipsec_ctx->ol_pkts_cnt++] = pkts[i];
+                       if (ips->security.ol_flags &
+                               RTE_SECURITY_TX_OLOAD_NEED_MDATA)
+                               rte_security_set_pkt_metadata(
+                                       ips->security.ctx, ips->security.ses,
+                                       pkts[i], NULL);
+                       continue;
                }
 
                RTE_ASSERT(sa->cdev_id_qp < ipsec_ctx->nb_qps);
@@ -149,9 +626,35 @@ ipsec_enqueue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
        }
 }
 
+static inline int32_t
+ipsec_inline_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
+             struct rte_mbuf *pkts[], uint16_t max_pkts)
+{
+       int32_t nb_pkts, ret;
+       struct ipsec_mbuf_metadata *priv;
+       struct ipsec_sa *sa;
+       struct rte_mbuf *pkt;
+
+       nb_pkts = 0;
+       while (ipsec_ctx->ol_pkts_cnt > 0 && nb_pkts < max_pkts) {
+               pkt = ipsec_ctx->ol_pkts[--ipsec_ctx->ol_pkts_cnt];
+               rte_prefetch0(pkt);
+               priv = get_priv(pkt);
+               sa = priv->sa;
+               ret = xform_func(pkt, sa, &priv->cop);
+               if (unlikely(ret)) {
+                       rte_pktmbuf_free(pkt);
+                       continue;
+               }
+               pkts[nb_pkts++] = pkt;
+       }
+
+       return nb_pkts;
+}
+
 static inline int
 ipsec_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
-               struct rte_mbuf *pkts[], uint16_t max_pkts)
+             struct rte_mbuf *pkts[], uint16_t max_pkts)
 {
        int32_t nb_pkts = 0, ret = 0, i, j, nb_cops;
        struct ipsec_mbuf_metadata *priv;
@@ -183,11 +686,21 @@ ipsec_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
 
                        RTE_ASSERT(sa != NULL);
 
-                       ret = xform_func(pkt, sa, cops[j]);
-                       if (unlikely(ret))
-                               rte_pktmbuf_free(pkt);
-                       else
-                               pkts[nb_pkts++] = pkt;
+                       if (ipsec_get_action_type(sa) ==
+                               RTE_SECURITY_ACTION_TYPE_NONE) {
+                               ret = xform_func(pkt, sa, cops[j]);
+                               if (unlikely(ret)) {
+                                       rte_pktmbuf_free(pkt);
+                                       continue;
+                               }
+                       } else if (ipsec_get_action_type(sa) ==
+                               RTE_SECURITY_ACTION_TYPE_LOOKASIDE_PROTOCOL) {
+                               if (cops[j]->status) {
+                                       rte_pktmbuf_free(pkt);
+                                       continue;
+                               }
+                       }
+                       pkts[nb_pkts++] = pkt;
                }
        }
 
@@ -199,12 +712,19 @@ uint16_t
 ipsec_inbound(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
                uint16_t nb_pkts, uint16_t len)
 {
-       struct ipsec_sa *sas[nb_pkts];
+       void *sas[nb_pkts];
 
        inbound_sa_lookup(ctx->sa_ctx, pkts, sas, nb_pkts);
 
        ipsec_enqueue(esp_inbound, ctx, pkts, sas, nb_pkts);
 
+       return ipsec_inline_dequeue(esp_inbound_post, ctx, pkts, len);
+}
+
+uint16_t
+ipsec_inbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+               uint16_t len)
+{
        return ipsec_dequeue(esp_inbound_post, ctx, pkts, len);
 }
 
@@ -212,11 +732,18 @@ uint16_t
 ipsec_outbound(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
                uint32_t sa_idx[], uint16_t nb_pkts, uint16_t len)
 {
-       struct ipsec_sa *sas[nb_pkts];
+       void *sas[nb_pkts];
 
        outbound_sa_lookup(ctx->sa_ctx, sa_idx, sas, nb_pkts);
 
        ipsec_enqueue(esp_outbound, ctx, pkts, sas, nb_pkts);
 
+       return ipsec_inline_dequeue(esp_outbound_post, ctx, pkts, len);
+}
+
+uint16_t
+ipsec_outbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+               uint16_t len)
+{
        return ipsec_dequeue(esp_outbound_post, ctx, pkts, len);
 }