BWAN-16636: On non-overlay route change, don't evaluate overlay rnh

Author: ns-praghunathareddy

zebra/rib.h

@@ -84,6 +84,12 @@ struct rnh {
 #ifdef ZEBRA_INFIOT_CUSTOM_NEXTHOP_CHECK
         int dest_trkr_index;
         int nh_trkr_index;
+        /* Epoch cache: stores the g_overlay_trkr_eval_seq value at which the
+         * last SHM lookup was performed and its result, so that repeated calls
+         * within the same route-change batch are free. This avoid liner lookup
+         * into trkr_client_get_trkr_by_index case unsuccesful lookup*/
+        uint32_t overlay_trkr_seq;
+        uint8_t  overlay_trkr_reachable;
 #endif
 	struct rnh_list_item rnh_list_item;
 };

zebra/zebra_rib.c

@@ -838,7 +838,30 @@ void zebra_rib_evaluate_rn_nexthops(struct route_node *rn, uint32_t seq,
 {
 	rib_dest_t *dest = rib_dest_from_rnode(rn);
 	struct rnh *rnh;
+#ifdef ZEBRA_INFIOT_CUSTOM_NEXTHOP_CHECK
+	struct route_node *trigger_rn = rn;
+	/* Determine whether this route change can affect overlay reachability.
+	 * Only the default route (which all unreachable overlay RNHs park on)
+	 * or a route within the overlay supernet (169.254.0.0/16) can change
+	 * overlay NHT state.
+	 */
+	const bool overlay_relevant =
+		is_default_prefix(&rn->p)
+		|| prefix_match(&g_infovlay_prefix, &rn->p);
+	if (overlay_relevant) {
+		g_overlay_trkr_eval_seq++;
+	}
+
+	if (overlay_relevant && dest) {
+		struct zebra_vrf *zvrf = rib_dest_vrf(dest);
+		struct rib_table_info *info = srcdest_rnode_table_info(trigger_rn);
 
+		if (zvrf && info) {
+			zebra_rnh_prescan_overlay_nht(
+				zvrf, info->afi, 0, &trigger_rn->p, info->safi);
+		}
+	}
+#endif
 	/*
 	 * We are storing the rnh's associated withb
 	 * the tracked nexthop as a list of the rn's.
@@ -904,6 +927,22 @@ void zebra_rib_evaluate_rn_nexthops(struct route_node *rn, uint32_t seq,
 			}
 
 			rnh->seqno = seq;
+#ifdef ZEBRA_INFIOT_CUSTOM_NEXTHOP_CHECK
+			/* Skip overlay RNH evaluations for non-overlay-relevant
+			 * route changes.  All unreachable overlay RNHs park at
+			 * 0.0.0.0/0's nht list; the while(rn) walk always reaches
+			 * 0.0.0.0/0, so without this guard every route install
+			 * triggers full RNH evaluations. Overlay reachability is
+			 * independent of non-overlay routes so skipping is safe. */
+			if (!overlay_relevant
+			    && p->family == AF_INET
+			    && prefix_match(&g_infovlay_prefix, p)) {
+				if (IS_ZEBRA_DEBUG_NHT) {
+					zlog_debug("skip overlay RNH %pFX non-overlay route change", p);
+				}
+				continue;
+			}
+#endif
 			zebra_evaluate_rnh(zvrf, family2afi(p->family), 0, p,
 					   rnh->safi);
 		}

zebra/zebra_rnh.c

@@ -73,6 +73,10 @@ extern struct trkr_client *g_infovlay_trkr;
 int g_inf_nhcntr_read_success = 0;
 extern int g_inf_is_controller;
 extern struct list *g_inf_ctrl_overlay_ips;
+/* Epoch counter: incremented once per overlay/default route-change event in
+ * zebra_rib_evaluate_rn_nexthops(). Per-RNH cache in check_overlay_nexthop()
+ * uses this to skip redundant SHM lookups within the same batch. */
+uint32_t g_overlay_trkr_eval_seq = 1;
 #endif
 
 static bool compare_state(struct route_entry *r1, struct route_entry *r2);
@@ -574,6 +578,17 @@ static int check_overlay_nexthop(struct prefix *pp, uint8_t *isreachable, struct
 		}
 	}
 
+	/* Epoch cache hit: result already computed for this route-change.
+	 * Return immediately without touching SHM. */
+	if (rnh->overlay_trkr_seq == g_overlay_trkr_eval_seq) {
+		*isreachable = rnh->overlay_trkr_reachable;
+		if (IS_ZEBRA_DEBUG_NHT) {
+			zlog_debug("Overlay cache HIT for %s: reachable=%d seq=%u",
+				   via, *isreachable, g_overlay_trkr_eval_seq);
+		}
+		return *isreachable;
+	}
+
 	// It is possible that when zebra starts, click has not created the
 	// SHM in which case the client initialization will fail in infnh_init.
 	// retry here
@@ -618,6 +633,9 @@ static int check_overlay_nexthop(struct prefix *pp, uint8_t *isreachable, struct
 		zlog_debug("Infiot via: %s, cntrname %s val %llu reachable %d nhindex %d destindex %d", via, cntrname,
 			trkr == NULL ? 1 : trkr->val, *isreachable, rnh->nh_trkr_index, rnh->dest_trkr_index);
 	}
+	/* Store SHM result in per-RNH epoch cache for future hits this batch. */
+	rnh->overlay_trkr_seq = g_overlay_trkr_eval_seq;
+	rnh->overlay_trkr_reachable = *isreachable;
 	return *isreachable;
 }
 #endif
@@ -961,6 +979,26 @@ static void zebra_rnh_evaluate_overlay_prefixes(struct zebra_vrf *zvrf, afi_t af
 		zebra_rnh_evaluate_entry(zvrf, afi, force, nrn);
 	}
 }
+
+void zebra_rnh_prescan_overlay_nht(struct zebra_vrf *zvrf, afi_t afi,
+							   int force, const struct prefix *p,
+							   safi_t safi)
+{
+	/* Guard against re-entrant invocation while pre-scan is in progress. */
+	static bool overlay_prescan_active = false;
+
+	if (!overlay_prescan_active && p != NULL
+	    && p->family == g_infovlay_prefix.family
+	    && (prefix_match(&g_infovlay_prefix, p) ||
+		prefix_match(&g_infovlay_prefix, p))) {
+		overlay_prescan_active = true;
+		if (IS_ZEBRA_DEBUG_NHT)
+			zlog_debug("overlay pre-scan seq=%u skip_p=%pFX",
+				   g_overlay_trkr_eval_seq, p);
+		zebra_rnh_evaluate_overlay_prefixes(zvrf, afi, force, p, safi);
+		overlay_prescan_active = false;
+	}
+}
 #endif
 
 /* Evaluate all tracked entries (nexthops or routes for import into BGP)

zebra/zebra_rnh.h

@@ -45,6 +45,10 @@ extern void zebra_deregister_rnh_pseudowire(vrf_id_t, struct zebra_pw *);
 extern void zebra_remove_rnh_client(struct rnh *rnh, struct zserv *client);
 extern void zebra_evaluate_rnh(struct zebra_vrf *zvrf, afi_t afi, int force,
 			       const struct prefix *p, safi_t safi);
+#ifdef ZEBRA_INFIOT_CUSTOM_NEXTHOP_CHECK
+extern void zebra_rnh_prescan_overlay_nht(struct zebra_vrf *zvrf, afi_t afi,
+			int force, const struct prefix *p, safi_t safi);
+#endif
 extern void zebra_print_rnh_table(vrf_id_t vrfid, afi_t afi, safi_t safi,
 				  struct vty *vty, const struct prefix *p,
 				  json_object *json);
@@ -66,6 +70,8 @@ void show_route_nexthop_helper(struct vty *vty, const struct route_entry *re,
 
 extern struct prefix g_infovlay_prefix;
 extern struct list *g_inf_ctrl_overlay_ips;
+/* Epoch counter shared with zebra_rib.c for per-RNH SHM result caching. */
+extern uint32_t g_overlay_trkr_eval_seq;
 #ifdef __cplusplus
 }
 #endif