examples/ipsec-secgw: fix usage of incorrect port
authorAnoob Joseph <anoob.joseph@caviumnetworks.com>
Tue, 12 Dec 2017 08:32:25 +0000 (08:32 +0000)
committerPablo de Lara <pablo.de.lara.guarch@intel.com>
Sat, 20 Jan 2018 15:08:34 +0000 (16:08 +0100)
When security offload is enabled, the packet should be forwarded on the
port configured in the SA. Security session will be configured on that
port only, and sending the packet on other ports could result in
unencrypted packets being sent out.

This would have performance improvements too, as the per packet LPM
lookup would be avoided for IPsec packets, in inline mode.

Fixes: ec17993a145a ("examples/ipsec-secgw: support security offload")

Signed-off-by: Anoob Joseph <anoob.joseph@caviumnetworks.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Radu Nicolau <radu.nicolau@intel.com>
doc/guides/sample_app_ug/ipsec_secgw.rst
examples/ipsec-secgw/ipsec-secgw.c

index d6cfdbf..ae18acd 100644 (file)
@@ -61,6 +61,12 @@ In case of complete protocol offload, the processing of headers(ESP and outer
 IP header) is done by the hardware and the application does not need to
 add/remove them during outbound/inbound processing.
 
+For inline offloaded outbound traffic, the application will not do the LPM
+lookup for routing, as the port on which the packet has to be forwarded will be
+part of the SA. Security parameters will be configured on that port only, and
+sending the packet on other ports could result in unencrypted packets being
+sent out.
+
 The Path for IPsec Inbound traffic is:
 
 *  Read packets from the port.
@@ -543,7 +549,9 @@ where each options means:
  ``<port_id>``
 
  * Port/device ID of the ethernet/crypto accelerator for which the SA is
-   configured. This option is used when *type* is NOT *no-offload*
+   configured. For *inline-crypto-offload* and *inline-protocol-offload*, this
+   port will be used for routing. The routing table will not be referred in
+   this case.
 
  * Optional: No, if *type* is not *no-offload*
 
index 0a47af5..caa9d57 100644 (file)
@@ -558,31 +558,81 @@ process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
                traffic->ip6.num = nb_pkts_out;
 }
 
+static inline int32_t
+get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
+{
+       struct ipsec_mbuf_metadata *priv;
+       struct ipsec_sa *sa;
+
+       priv = get_priv(pkt);
+
+       sa = priv->sa;
+       if (unlikely(sa == NULL)) {
+               RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
+               goto fail;
+       }
+
+       if (is_ipv6)
+               return sa->portid;
+
+       /* else */
+       return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
+
+fail:
+       if (is_ipv6)
+               return -1;
+
+       /* else */
+       return 0;
+}
+
 static inline void
 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 {
        uint32_t hop[MAX_PKT_BURST * 2];
        uint32_t dst_ip[MAX_PKT_BURST * 2];
+       int32_t pkt_hop = 0;
        uint16_t i, offset;
+       uint16_t lpm_pkts = 0;
 
        if (nb_pkts == 0)
                return;
 
+       /* Need to do an LPM lookup for non-inline packets. Inline packets will
+        * have port ID in the SA
+        */
+
        for (i = 0; i < nb_pkts; i++) {
-               offset = offsetof(struct ip, ip_dst);
-               dst_ip[i] = *rte_pktmbuf_mtod_offset(pkts[i],
-                               uint32_t *, offset);
-               dst_ip[i] = rte_be_to_cpu_32(dst_ip[i]);
+               if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
+                       /* Security offload not enabled. So an LPM lookup is
+                        * required to get the hop
+                        */
+                       offset = offsetof(struct ip, ip_dst);
+                       dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
+                                       uint32_t *, offset);
+                       dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
+                       lpm_pkts++;
+               }
        }
 
-       rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, nb_pkts);
+       rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
+
+       lpm_pkts = 0;
 
        for (i = 0; i < nb_pkts; i++) {
-               if ((hop[i] & RTE_LPM_LOOKUP_SUCCESS) == 0) {
+               if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
+                       /* Read hop from the SA */
+                       pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
+               } else {
+                       /* Need to use hop returned by lookup */
+                       pkt_hop = hop[lpm_pkts++];
+               }
+
+               if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
                        rte_pktmbuf_free(pkts[i]);
                        continue;
                }
-               send_single_packet(pkts[i], hop[i] & 0xff);
+               send_single_packet(pkts[i], pkt_hop & 0xff);
        }
 }
 
@@ -592,26 +642,49 @@ route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
        int32_t hop[MAX_PKT_BURST * 2];
        uint8_t dst_ip[MAX_PKT_BURST * 2][16];
        uint8_t *ip6_dst;
+       int32_t pkt_hop = 0;
        uint16_t i, offset;
+       uint16_t lpm_pkts = 0;
 
        if (nb_pkts == 0)
                return;
 
+       /* Need to do an LPM lookup for non-inline packets. Inline packets will
+        * have port ID in the SA
+        */
+
        for (i = 0; i < nb_pkts; i++) {
-               offset = offsetof(struct ip6_hdr, ip6_dst);
-               ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *, offset);
-               memcpy(&dst_ip[i][0], ip6_dst, 16);
+               if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
+                       /* Security offload not enabled. So an LPM lookup is
+                        * required to get the hop
+                        */
+                       offset = offsetof(struct ip6_hdr, ip6_dst);
+                       ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
+                                       offset);
+                       memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
+                       lpm_pkts++;
+               }
        }
 
-       rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip,
-                       hop, nb_pkts);
+       rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
+                       lpm_pkts);
+
+       lpm_pkts = 0;
 
        for (i = 0; i < nb_pkts; i++) {
-               if (hop[i] == -1) {
+               if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
+                       /* Read hop from the SA */
+                       pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
+               } else {
+                       /* Need to use hop returned by lookup */
+                       pkt_hop = hop[lpm_pkts++];
+               }
+
+               if (pkt_hop == -1) {
                        rte_pktmbuf_free(pkts[i]);
                        continue;
                }
-               send_single_packet(pkts[i], hop[i] & 0xff);
+               send_single_packet(pkts[i], pkt_hop & 0xff);
        }
 }