ip6: avoid duplicate nh entry with overlapping prefix routes

Similar to ip4 fix, do not create unconditionnaly a new nexthop. Reuse an existing one if any. If the NH is associated with an existing route, makes sure that the interface id is properly populated. Signed-off-by: Christophe Fontaine <[email protected]>
DPDK · Oct 31, 2024 · 07123c0 · 07123c0
1 parent 20ff41b
commit 07123c0
Showing 1 changed file with 8 additions and 3 deletions.
diff --git a/modules/ip6/datapath/ip6_output.c b/modules/ip6/datapath/ip6_output.c
@@ -110,13 +110,18 @@ ip6_output_process(struct rte_graph *graph, struct rte_node *node, void **objs,
 			// We currently do not have an explicit entry for this destination IP.
 			// Create a new next hop and its associated /128 route so that next
 			// packets take it in priority with a single route lookup.
-			struct nexthop6 *remote = ip6_nexthop_new(
-				nh->vrf_id, nh->iface_id, &ip->dst_addr
-			);
+			struct nexthop6 *remote = ip6_nexthop_lookup(nh->vrf_id, &ip->dst_addr);
+			if (remote == NULL)
+				remote = ip6_nexthop_new(nh->vrf_id, nh->iface_id, &ip->dst_addr);
+			else if (remote->flags & GR_IP6_NH_F_GATEWAY && remote->iface_id == 0)
+				remote->iface_id = nh->iface_id;
+
 			if (remote == NULL) {
 				edge = ERROR;
 				goto next;
 			}
+
+			assert(remote->iface_id == nh->iface_id);
 			if (ip6_route_insert(nh->vrf_id, &ip->dst_addr, RTE_IPV6_MAX_DEPTH, remote)
 			    < 0) {
 				edge = ERROR;