pimd: Make the rpf scan per vrf.

We know the vrf that we are in when we need to initiate a
rescan of the rpf cache.  So pass it in and use that information.
This should help the rescan at scale with several vrf's cutting
out a lot of unnecessary work.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

pimd/pim_cmd.c

@@ -2566,8 +2566,8 @@ static void pim_show_upstream_rpf(struct pim_instance *pim, struct vty *vty,
 	}
 }
 
-static void show_rpf_refresh_stats(struct vty *vty, time_t now,
+static void show_rpf_refresh_stats(struct vty *vty, struct pim_instance *pim,
-				   json_object *json)
+				   time_t now, json_object *json)
 {
 	char refresh_uptime[10];
 
@@ -2579,7 +2579,7 @@ static void show_rpf_refresh_stats(struct vty *vty, time_t now,
 				    qpim_rpf_cache_refresh_delay_msec);
 		json_object_int_add(
 			json, "rpfCacheRefreshTimer",
-			pim_time_timer_remain_msec(qpim_rpf_cache_refresher));
+			pim_time_timer_remain_msec(pim->rpf_cache_refresher));
 		json_object_int_add(json, "rpfCacheRefreshRequests",
 				    qpim_rpf_cache_refresh_requests);
 		json_object_int_add(json, "rpfCacheRefreshEvents",
@@ -2600,7 +2600,7 @@ static void show_rpf_refresh_stats(struct vty *vty, time_t now,
 			"Nexthop Lookups:            %lld\n"
 			"Nexthop Lookups Avoided:    %lld\n",
 			qpim_rpf_cache_refresh_delay_msec,
-			pim_time_timer_remain_msec(qpim_rpf_cache_refresher),
+			pim_time_timer_remain_msec(pim->rpf_cache_refresher),
 			(long long)qpim_rpf_cache_refresh_requests,
 			(long long)qpim_rpf_cache_refresh_events,
 			refresh_uptime, (long long)qpim_nexthop_lookups,
@@ -2642,9 +2642,9 @@ static void pim_show_rpf(struct pim_instance *pim, struct vty *vty, u_char uj)
 
 	if (uj) {
 		json = json_object_new_object();
-		show_rpf_refresh_stats(vty, now, json);
+		show_rpf_refresh_stats(vty, pim, now, json);
 	} else {
-		show_rpf_refresh_stats(vty, now, json);
+		show_rpf_refresh_stats(vty, pim, now, json);
 		vty_out(vty, "\n");
 		vty_out(vty,
 			"Source          Group           RpfIface RpfAddress      RibNextHop      Metric Pref\n");
@@ -4337,7 +4337,7 @@ static void pim_cmd_show_ip_multicast_helper(struct pim_instance *pim,
 
 	vty_out(vty, "\n");
 
-	show_rpf_refresh_stats(vty, now, NULL);
+	show_rpf_refresh_stats(vty, pim, now, NULL);
 
 	vty_out(vty, "\n");
 

pimd/pim_instance.h

@@ -94,6 +94,9 @@ struct pim_instance {
 
 	unsigned int keep_alive_time;
 	unsigned int rp_keep_alive_time;
+
+	/* If we need to rescan all our upstreams */
+	struct thread *rpf_cache_refresher;
 };
 
 void pim_vrf_init(void);

pimd/pim_neighbor.c

@@ -545,7 +545,7 @@ pim_neighbor_add(struct interface *ifp, struct in_addr source_addr,
 
 	pim_rp_setup(pim_ifp->pim);
 
-	sched_rpf_cache_refresh();
+	sched_rpf_cache_refresh(pim_ifp->pim);
 	return neigh;
 }
 
@@ -667,7 +667,7 @@ void pim_neighbor_delete(struct interface *ifp, struct pim_neighbor *neigh,
 
 	pim_neighbor_free(neigh);
 
-	sched_rpf_cache_refresh();
+	sched_rpf_cache_refresh(pim_ifp->pim);
 }
 
 void pim_neighbor_delete_all(struct interface *ifp, const char *delete_message)

pimd/pim_zebra.c

@@ -394,23 +394,15 @@ static int pim_zebra_if_address_del(int command, struct zclient *client,
 	return 0;
 }
 
-static void scan_upstream_rpf_cache()
+static void scan_upstream_rpf_cache(struct pim_instance *pim)
 {
 	struct listnode *up_node;
 	struct listnode *up_nextnode;
 	struct listnode *node;
 	struct pim_upstream *up;
 	struct interface *ifp;
-	struct vrf *vrf;
-	struct pim_instance *pim;
 
-	RB_FOREACH (vrf, vrf_name_head, &vrfs_by_name) {
+	for (ALL_LIST_ELEMENTS(pim->upstream_list, up_node, up_nextnode, up)) {
-		pim = vrf->info;
-		if (!pim)
-			continue;
-
-		for (ALL_LIST_ELEMENTS(pim->upstream_list, up_node, up_nextnode,
-				       up)) {
 		enum pim_rpf_result rpf_result;
 		struct pim_rpf old;
 		struct prefix nht_p;
@@ -420,8 +412,7 @@ static void scan_upstream_rpf_cache()
 		nht_p.u.prefix4.s_addr = up->upstream_addr.s_addr;
 		pim_resolve_upstream_nh(pim, &nht_p);
 
-			old.source_nexthop.interface =
+		old.source_nexthop.interface = up->rpf.source_nexthop.interface;
-				up->rpf.source_nexthop.interface;
 		old.source_nexthop.nbr = up->rpf.source_nexthop.nbr;
 		rpf_result = pim_rpf_update(pim, up, &old, 0);
 
@@ -431,12 +422,11 @@ static void scan_upstream_rpf_cache()
 		if (rpf_result == PIM_RPF_CHANGED) {
 			struct pim_neighbor *nbr;
 
-				nbr = pim_neighbor_find(
+			nbr = pim_neighbor_find(old.source_nexthop.interface,
-					old.source_nexthop.interface,
 						old.rpf_addr.u.prefix4);
 			if (nbr)
-					pim_jp_agg_remove_group(
+				pim_jp_agg_remove_group(nbr->upstream_jp_agg,
-						nbr->upstream_jp_agg, up);
+							up);
 
 			/*
 			 * We have detected a case where we might need
@@ -444,23 +434,18 @@ static void scan_upstream_rpf_cache()
 			 * the inherited o_list so do it.
 			 */
 			if (up->channel_oil->oil_inherited_rescan) {
-					pim_upstream_inherited_olist_decide(pim,
+				pim_upstream_inherited_olist_decide(pim, up);
-									    up);
+				up->channel_oil->oil_inherited_rescan = 0;
-					up->channel_oil->oil_inherited_rescan =
-						0;
 			}
 
 			if (up->join_state == PIM_UPSTREAM_JOINED) {
 				/*
-					 * If we come up real fast we can be
+				 * If we come up real fast we can be here
-					 * here
+				 * where the mroute has not been installed
-					 * where the mroute has not been
-					 * installed
 				 * so install it.
 				 */
 				if (!up->channel_oil->installed)
-						pim_mroute_add(
+					pim_mroute_add(up->channel_oil,
-							up->channel_oil,
 						       __PRETTY_FUNCTION__);
 
 				/*
@@ -469,25 +454,19 @@ static void scan_upstream_rpf_cache()
 				 *
 				 * Transitions from Joined State
 				 *
-					 * RPF'(S,G) changes not due to an
+				 * RPF'(S,G) changes not due to an Assert
-					 * Assert
 				 *
-					 * The upstream (S,G) state machine
+				 * The upstream (S,G) state machine remains
-					 * remains in Joined
+				 * in Joined state. Send Join(S,G) to the new
-					 * state. Send Join(S,G) to the new
+				 * upstream neighbor, which is the new value
-					 * upstream neighbor, which is
+				 * of RPF'(S,G).  Send Prune(S,G) to the old
-					 * the new value of RPF'(S,G).  Send
+				 * upstream neighbor, which is the old value
-					 * Prune(S,G) to the old
+				 * of RPF'(S,G).  Set the Join Timer (JT) to
-					 * upstream neighbor, which is the old
+				 * expire after t_periodic seconds.
-					 * value of RPF'(S,G).  Set
-					 * the Join Timer (JT) to expire after
-					 * t_periodic seconds.
 				 */
-					pim_jp_agg_switch_interface(
+				pim_jp_agg_switch_interface(&old, &up->rpf, up);
-						&old, &up->rpf, up);
 
-					pim_upstream_join_timer_restart(up,
+				pim_upstream_join_timer_restart(up, &old);
-									&old);
 			} /* up->join_state == PIM_UPSTREAM_JOINED */
 
 			/* FIXME can join_desired actually be changed by
@@ -498,22 +477,16 @@ static void scan_upstream_rpf_cache()
 		} /* PIM_RPF_CHANGED */
 
 	} /* for (qpim_upstream_list) */
-	}
-
-	RB_FOREACH (vrf, vrf_name_head, &vrfs_by_name) {
-		pim = vrf->info;
-		if (!pim)
-			continue;
 
 	FOR_ALL_INTERFACES (pim->vrf, ifp)
 		if (ifp->info) {
 			struct pim_interface *pim_ifp = ifp->info;
 			struct pim_iface_upstream_switch *us;
 
-				for (ALL_LIST_ELEMENTS_RO(
+			for (ALL_LIST_ELEMENTS_RO(pim_ifp->upstream_switch_list,
-					     pim_ifp->upstream_switch_list,
 						  node, us)) {
 				struct pim_rpf rpf;
+
 				rpf.source_nexthop.interface = ifp;
 				rpf.rpf_addr.u.prefix4 = us->address;
 				pim_joinprune_send(&rpf, us->us);
@@ -521,7 +494,6 @@ static void scan_upstream_rpf_cache()
 			}
 		}
 }
-}
 
 void pim_scan_individual_oil(struct channel_oil *c_oil, int in_vif_index)
 {
@@ -657,53 +629,41 @@ void pim_scan_individual_oil(struct channel_oil *c_oil, int in_vif_index)
 	}
 }
 
-void pim_scan_oil(struct pim_instance *pim_matcher)
+void pim_scan_oil(struct pim_instance *pim)
 {
 	struct listnode *node;
 	struct listnode *nextnode;
 	struct channel_oil *c_oil;
 	ifindex_t ifindex;
 	int vif_index = 0;
-	struct vrf *vrf;
-	struct pim_instance *pim;
 
 	qpim_scan_oil_last = pim_time_monotonic_sec();
 	++qpim_scan_oil_events;
 
-	RB_FOREACH (vrf, vrf_name_head, &vrfs_by_name) {
+	for (ALL_LIST_ELEMENTS(pim->channel_oil_list, node, nextnode, c_oil)) {
-		pim = vrf->info;
+		if (c_oil->up && c_oil->up->rpf.source_nexthop.interface) {
-		if (!pim)
-			continue;
-
-		if (pim_matcher && pim != pim_matcher)
-			continue;
-
-		for (ALL_LIST_ELEMENTS(pim->channel_oil_list, node, nextnode,
-				       c_oil)) {
-			if (c_oil->up
-			    && c_oil->up->rpf.source_nexthop.interface) {
 			ifindex = c_oil->up->rpf.source_nexthop
 					  .interface->ifindex;
-				vif_index = pim_if_find_vifindex_by_ifindex(
+			vif_index =
-					pim, ifindex);
+				pim_if_find_vifindex_by_ifindex(pim, ifindex);
 			/* Pass Current selected NH vif index to mroute
 			 * download */
 			if (vif_index)
-					pim_scan_individual_oil(c_oil,
+				pim_scan_individual_oil(c_oil, vif_index);
-								vif_index);
 		} else
 			pim_scan_individual_oil(c_oil, 0);
 	}
 }
-}
 
 static int on_rpf_cache_refresh(struct thread *t)
 {
+	struct pim_instance *pim = THREAD_ARG(t);
+
 	/* update PIM protocol state */
-	scan_upstream_rpf_cache();
+	scan_upstream_rpf_cache(pim);
 
 	/* update kernel multicast forwarding cache (MFC) */
-	pim_scan_oil(NULL);
+	pim_scan_oil(pim);
 
 	qpim_rpf_cache_refresh_last = pim_time_monotonic_sec();
 	++qpim_rpf_cache_refresh_events;
@@ -713,13 +673,13 @@ static int on_rpf_cache_refresh(struct thread *t)
 	return 0;
 }
 
-void sched_rpf_cache_refresh(void)
+void sched_rpf_cache_refresh(struct pim_instance *pim)
 {
 	++qpim_rpf_cache_refresh_requests;
 
 	pim_rpf_set_refresh_time();
 
-	if (qpim_rpf_cache_refresher) {
+	if (pim->rpf_cache_refresher) {
 		/* Refresh timer is already running */
 		return;
 	}
@@ -731,9 +691,9 @@ void sched_rpf_cache_refresh(void)
 			   qpim_rpf_cache_refresh_delay_msec);
 	}
 
-	thread_add_timer_msec(master, on_rpf_cache_refresh, 0,
+	thread_add_timer_msec(master, on_rpf_cache_refresh, pim,
 			      qpim_rpf_cache_refresh_delay_msec,
-			      &qpim_rpf_cache_refresher);
+			      &pim->rpf_cache_refresher);
 }
 
 static void pim_zebra_connected(struct zclient *zclient)

pimd/pim_zebra.h

@@ -44,6 +44,6 @@ void igmp_source_forward_reevaluate_all(void);
 void pim_forward_start(struct pim_ifchannel *ch);
 void pim_forward_stop(struct pim_ifchannel *ch, bool install_it);
 
-void sched_rpf_cache_refresh(void);
+void sched_rpf_cache_refresh(struct pim_instance *pim);
 struct zclient *pim_zebra_zclient_get(void);
 #endif /* PIM_ZEBRA_H */

pimd/pimd.c

@@ -52,7 +52,6 @@ int qpim_t_periodic =
 	PIM_DEFAULT_T_PERIODIC; /* Period between Join/Prune Messages */
 struct pim_assert_metric qpim_infinite_assert_metric;
 long qpim_rpf_cache_refresh_delay_msec = 50;
-struct thread *qpim_rpf_cache_refresher = NULL;
 int64_t qpim_rpf_cache_refresh_requests = 0;
 int64_t qpim_rpf_cache_refresh_events = 0;
 int64_t qpim_rpf_cache_refresh_last = 0;

pimd/pimd.h

@@ -140,7 +140,6 @@ struct in_addr qpim_all_pim_routers_addr;
 int qpim_t_periodic; /* Period between Join/Prune Messages */
 struct pim_assert_metric qpim_infinite_assert_metric;
 long qpim_rpf_cache_refresh_delay_msec;
-struct thread *qpim_rpf_cache_refresher;
 int64_t qpim_rpf_cache_refresh_requests;
 int64_t qpim_rpf_cache_refresh_events;
 int64_t qpim_rpf_cache_refresh_last;