Merge pull request #14193 from donaldsharp/pim_vxlan_weirdness

Do not look into pim's eyes, pim gets mad

5 files changed, 98 insertions, 10 deletions

diff --git a/pimd/pim_msg.c b/pimd/pim_msg.c
index 5d1f08314b..6814798bf5 100644
--- a/pimd/pim_msg.c
+++ b/pimd/pim_msg.c
@@ -196,7 +196,32 @@ size_t pim_msg_get_jp_group_size(struct list *sources)
 				__func__, up->sg_str);
 
 		for (ALL_LIST_ELEMENTS_RO(up->sources, up_node, child)) {
-			if (!PIM_UPSTREAM_FLAG_TEST_USE_RPT(child->flags)) {
+			/*
+			 * PIM VXLAN is weird
+			 * It auto creates the S,G and populates a bunch
+			 * of flags that make it look like a SPT prune should
+			 * be sent.  But this regularly scheduled join
+			 * for the *,G in the VXLAN setup can happen at
+			 * scheduled times *before* the null register
+			 * is received by the RP to cause it to initiate
+			 * the S,G joins toward the source.  Let's just
+			 * assume that if this is a SRC VXLAN ORIG route
+			 * and no actual ifchannels( joins ) have been
+			 * created then do not send the embedded prune
+			 * Why you may ask?  Well if the prune is S,G
+			 * RPT Prune is received *before* the join
+			 * from the RP( if it flows to this routers
+			 * upstream interface ) then we'll just wisely
+			 * create a mroute with an empty oil on
+			 * the upstream intermediate router preventing
+			 * packets from flowing to the RP
+			 */
+			if (PIM_UPSTREAM_FLAG_TEST_SRC_VXLAN_ORIG(child->flags) &&
+			    listcount(child->ifchannels) == 0) {
+				if (PIM_DEBUG_PIM_PACKETS)
+					zlog_debug("%s: %s Vxlan originated S,G route with no ifchannels, not adding prune to compound message",
+						   __func__, child->sg_str);
+			} else if (!PIM_UPSTREAM_FLAG_TEST_USE_RPT(child->flags)) {
 				/* If we are using SPT and the SPT and RPT IIFs
 				 * are different we can prune the source off
 				 * of the RPT.
diff --git a/pimd/pim_nht.c b/pimd/pim_nht.c
index 5f0f2a5933..4e8e5f0df7 100644
--- a/pimd/pim_nht.c
+++ b/pimd/pim_nht.c
@@ -31,6 +31,8 @@
 #include "pim_zlookup.h"
 #include "pim_rp.h"
 #include "pim_addr.h"
+#include "pim_register.h"
+#include "pim_vxlan.h"
 
 /**
  * pim_sendmsg_zebra_rnh -- Format and send a nexthop register/Unregister
@@ -399,17 +401,28 @@ static void pim_update_rp_nh(struct pim_instance *pim,
 {
 	struct listnode *node = NULL;
 	struct rp_info *rp_info = NULL;
+	struct interface *ifp;
 
 	/*Traverse RP list and update each RP Nexthop info */
 	for (ALL_LIST_ELEMENTS_RO(pnc->rp_list, node, rp_info)) {
 		if (pim_rpf_addr_is_inaddr_any(&rp_info->rp))
 			continue;
 
+		ifp = rp_info->rp.source_nexthop.interface;
 		// Compute PIM RPF using cached nexthop
 		if (!pim_ecmp_nexthop_lookup(pim, &rp_info->rp.source_nexthop,
 					     rp_info->rp.rpf_addr,
 					     &rp_info->group, 1))
 			pim_rp_nexthop_del(rp_info);
+
+		/*
+		 * If we transition from no path to a path
+		 * we need to search through all the vxlan's
+		 * that use this rp and send NULL registers
+		 * for all the vxlan S,G streams
+		 */
+		if (!ifp && rp_info->rp.source_nexthop.interface)
+			pim_vxlan_rp_info_is_alive(pim, &rp_info->rp);
 	}
 }
 
@@ -436,17 +449,27 @@ static int pim_update_upstream_nh_helper(struct hash_bucket *bucket, void *arg)
 		(rpf_result == PIM_RPF_FAILURE && old.source_nexthop.interface))
 		pim_zebra_upstream_rpf_changed(pim, up, &old);
 
+	/*
+	 * If we are a VXLAN source and we are transitioning from not
+	 * having an outgoing interface to having an outgoing interface
+	 * let's immediately send the null pim register
+	 */
+	if (!old.source_nexthop.interface && up->rpf.source_nexthop.interface &&
+	    PIM_UPSTREAM_FLAG_TEST_SRC_VXLAN_ORIG(up->flags) &&
+	    (up->reg_state == PIM_REG_NOINFO || up->reg_state == PIM_REG_JOIN)) {
+		pim_null_register_send(up);
+	}
 
 	if (PIM_DEBUG_PIM_NHT) {
-		zlog_debug(
-			"%s: NHT upstream %s(%s) old ifp %s new ifp %s",
-			__func__, up->sg_str, pim->vrf->name,
-			old.source_nexthop.interface ? old.source_nexthop
-							       .interface->name
-						     : "Unknown",
-			up->rpf.source_nexthop.interface ? up->rpf.source_nexthop
-								   .interface->name
-							 : "Unknown");
+		zlog_debug("%s: NHT upstream %s(%s) old ifp %s new ifp %s rpf_result: %d",
+			   __func__, up->sg_str, pim->vrf->name,
+			   old.source_nexthop.interface ? old.source_nexthop
+								  .interface->name
+							: "Unknown",
+			   up->rpf.source_nexthop.interface ? up->rpf.source_nexthop
+								      .interface->name
+							    : "Unknown",
+			   rpf_result);
 	}
 
 	return HASHWALK_CONTINUE;
diff --git a/pimd/pim_upstream.c b/pimd/pim_upstream.c
index a8d087bf49..fd99e77761 100644
--- a/pimd/pim_upstream.c
+++ b/pimd/pim_upstream.c
@@ -912,6 +912,7 @@ static struct pim_upstream *pim_upstream_new(struct pim_instance *pim,
 				false /*update_mroute*/);
 		rpf_result = pim_rpf_update(pim, up, NULL, __func__);
 		if (rpf_result == PIM_RPF_FAILURE) {
+			up->channel_oil->oil_inherited_rescan = 1;
 			if (PIM_DEBUG_PIM_TRACE)
 				zlog_debug(
 					"%s: Attempting to create upstream(%s), Unable to RPF for source",
diff --git a/pimd/pim_vxlan.c b/pimd/pim_vxlan.c
index 8df3c90f00..9650da89a8 100644
--- a/pimd/pim_vxlan.c
+++ b/pimd/pim_vxlan.c
@@ -32,6 +32,41 @@ static void pim_vxlan_work_timer_setup(bool start);
 static void pim_vxlan_set_peerlink_rif(struct pim_instance *pim,
 			struct interface *ifp);
 
+/*
+ * The rp info has gone from no path to having a
+ * path.  Let's immediately send out the null pim register
+ * as that else we will be sitting for up to 60 seconds waiting
+ * for it too pop.  Which is not cool.
+ */
+void pim_vxlan_rp_info_is_alive(struct pim_instance *pim,
+				struct pim_rpf *rpg_changed)
+{
+	struct listnode *listnode;
+	struct pim_vxlan_sg *vxlan_sg;
+	struct pim_rpf *rpg;
+
+	/*
+	 * No vxlan here, move along, nothing to see
+	 */
+	if (!vxlan_info.work_list)
+		return;
+
+	for (listnode = vxlan_info.work_list->head; listnode;
+	     listnode = listnode->next) {
+		vxlan_sg = listgetdata(listnode);
+
+		rpg = RP(pim, vxlan_sg->up->sg.grp);
+
+		/*
+		 * If the rp is the same we should send
+		 */
+		if (rpg == rpg_changed) {
+			zlog_debug("VXLAN RP INFO is alive sending");
+			pim_null_register_send(vxlan_sg->up);
+		}
+	}
+}
+
 /*************************** vxlan work list **********************************
  * A work list is maintained for staggered generation of pim null register
  * messages for vxlan SG entries that are in a reg_join state.
@@ -66,6 +101,7 @@ static void pim_vxlan_do_reg_work(void)
 
 	for (; listnode; listnode = listnode->next) {
 		vxlan_sg = (struct pim_vxlan_sg *)listnode->data;
+
 		if (vxlan_sg->up && (vxlan_sg->up->reg_state == PIM_REG_JOIN)) {
 			if (PIM_DEBUG_VXLAN)
 				zlog_debug("vxlan SG %s periodic NULL register",
diff --git a/pimd/pim_vxlan.h b/pimd/pim_vxlan.h
index 9a135ca6b8..5039bf6540 100644
--- a/pimd/pim_vxlan.h
+++ b/pimd/pim_vxlan.h
@@ -135,6 +135,9 @@ extern bool pim_vxlan_do_mlag_reg(void);
 extern void pim_vxlan_inherit_mlag_flags(struct pim_instance *pim,
 		struct pim_upstream *up, bool inherit);
 
+extern void pim_vxlan_rp_info_is_alive(struct pim_instance *pim,
+				       struct pim_rpf *rpg_changed);
+
 /* Shutdown of PIM stop the thread */
 extern void pim_vxlan_terminate(void);
 #endif /* PIM_VXLAN_H */