bgpd: fix export prefixes when rt extcomm set by route-map

When exporting BGP prefixes, it is necessary to configure
the route target extended communities with the following
command:

> rt vpn export <RouteTarget>

But the customer may need to configure the route-target to
apply to bgp updates, solely based on a route-map criterium.
by using the below route-map configured like that:

> route-map vpn export <routemapname>

Fix this by allowing to export bgp updates based on the
presence of route-targets on either route-map or vpn
configured rt. the exportation process is stopped
if no route target is available in the ecommunity list.

Fixes: ddb5b4880b ("bgpd: vpn-vrf route leaking")
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>

bgpd/bgp_ecommunity.c

@@ -1042,6 +1042,34 @@ static int ecommunity_lb_str(char *buf, size_t bufsz, const uint8_t *pnt,
 	return len;
 }
 
+bool ecommunity_has_route_target(struct ecommunity *ecom)
+{
+	uint32_t i;
+	uint8_t *pnt;
+	uint8_t type = 0;
+	uint8_t sub_type = 0;
+
+	if (!ecom)
+		return false;
+	for (i = 0; i < ecom->size; i++) {
+		/* Retrieve value field */
+		pnt = ecom->val + (i * ecom->unit_size);
+
+		/* High-order octet is the type */
+		type = *pnt++;
+
+		if (type == ECOMMUNITY_ENCODE_AS ||
+		    type == ECOMMUNITY_ENCODE_IP ||
+		    type == ECOMMUNITY_ENCODE_AS4) {
+			/* Low-order octet of type. */
+			sub_type = *pnt++;
+			if (sub_type == ECOMMUNITY_ROUTE_TARGET)
+				return true;
+		}
+	}
+	return false;
+}
+
 /* Convert extended community attribute to string.
  * Due to historical reason of industry standard implementation, there
  * are three types of format:

bgpd/bgp_ecommunity.h

@@ -341,6 +341,7 @@ extern struct ecommunity *ecommunity_str2com(const char *, int, int);
 extern struct ecommunity *ecommunity_str2com_ipv6(const char *str, int type,
 						  int keyword_included);
 extern char *ecommunity_ecom2str(struct ecommunity *, int, int);
+extern bool ecommunity_has_route_target(struct ecommunity *ecom);
 extern void ecommunity_strfree(char **s);
 extern bool ecommunity_include(struct ecommunity *e1, struct ecommunity *e2);
 extern bool ecommunity_match(const struct ecommunity *,

bgpd/bgp_mplsvpn.c

@@ -300,7 +300,7 @@ void vpn_leak_zebra_vrf_label_update(struct bgp *bgp, afi_t afi)
 		return;
 	}
 
-	if (vpn_leak_to_vpn_active(bgp, afi, NULL)) {
+	if (vpn_leak_to_vpn_active(bgp, afi, NULL, false)) {
 		label = bgp->vpn_policy[afi].tovpn_label;
 	}
 
@@ -1533,6 +1533,9 @@ void vpn_leak_from_vrf_update(struct bgp *to_bgp,	     /* to */
 	struct bgp_dest *bn;
 	const char *debugmsg;
 	int nexthop_self_flag = 0;
+	struct ecommunity *old_ecom;
+	struct ecommunity *new_ecom = NULL;
+	struct ecommunity *rtlist_ecom;
 
 	if (debug)
 		zlog_debug("%s: from vrf %s", __func__, from_bgp->name_pretty);
@@ -1560,7 +1563,7 @@ void vpn_leak_from_vrf_update(struct bgp *to_bgp,	     /* to */
 	if (!is_route_injectable_into_vpn(path_vrf))
 		return;
 
-	if (!vpn_leak_to_vpn_active(from_bgp, afi, &debugmsg)) {
+	if (!vpn_leak_to_vpn_active(from_bgp, afi, &debugmsg, false)) {
 		if (debug)
 			zlog_debug("%s: %s skipping: %s", __func__,
 				   from_bgp->name, debugmsg);
@@ -1609,27 +1612,37 @@ void vpn_leak_from_vrf_update(struct bgp *to_bgp,	     /* to */
 	/*
 	 * Add the vpn-policy rt-list
 	 */
-	struct ecommunity *old_ecom;
-	struct ecommunity *new_ecom;
 
 	/* Export with the 'from' instance's export RTs. */
 	/* If doing VRF-to-VRF leaking, strip existing RTs first. */
 	old_ecom = bgp_attr_get_ecommunity(&static_attr);
+	rtlist_ecom = from_bgp->vpn_policy[afi].rtlist[BGP_VPN_POLICY_DIR_TOVPN];
 	if (old_ecom) {
 		new_ecom = ecommunity_dup(old_ecom);
 		if (CHECK_FLAG(from_bgp->af_flags[afi][SAFI_UNICAST],
 			       BGP_CONFIG_VRF_TO_VRF_EXPORT))
 			ecommunity_strip_rts(new_ecom);
-		new_ecom = ecommunity_merge(
+		if (rtlist_ecom)
-			new_ecom, from_bgp->vpn_policy[afi]
+			new_ecom = ecommunity_merge(new_ecom, rtlist_ecom);
-					  .rtlist[BGP_VPN_POLICY_DIR_TOVPN]);
 		if (!old_ecom->refcnt)
 			ecommunity_free(&old_ecom);
+	} else if (rtlist_ecom) {
+		new_ecom = ecommunity_dup(rtlist_ecom);
 	} else {
-		new_ecom = ecommunity_dup(
+		if (debug)
-			from_bgp->vpn_policy[afi]
+			zlog_debug("%s: %s skipping: waiting for a non empty export rt list.",
-				.rtlist[BGP_VPN_POLICY_DIR_TOVPN]);
+				   __func__, from_bgp->name_pretty);
+		return;
 	}
+
+	if (!ecommunity_has_route_target(new_ecom)) {
+		ecommunity_free(&new_ecom);
+		if (debug)
+			zlog_debug("%s: %s skipping: waiting for a valid export rt list.",
+				   __func__, from_bgp->name_pretty);
+		return;
+	}
+
 	bgp_attr_set_ecommunity(&static_attr, new_ecom);
 
 	if (debug && bgp_attr_get_ecommunity(&static_attr)) {
@@ -1885,7 +1898,7 @@ void vpn_leak_from_vrf_withdraw(struct bgp *to_bgp,		/* to */
 	if (!is_route_injectable_into_vpn(path_vrf))
 		return;
 
-	if (!vpn_leak_to_vpn_active(from_bgp, afi, &debugmsg)) {
+	if (!vpn_leak_to_vpn_active(from_bgp, afi, &debugmsg, true)) {
 		if (debug)
 			zlog_debug("%s: skipping: %s", __func__, debugmsg);
 		return;
@@ -2666,7 +2679,7 @@ void vpn_handle_router_id_update(struct bgp *bgp, bool withdraw,
 	edir = BGP_VPN_POLICY_DIR_TOVPN;
 
 	for (afi = 0; afi < AFI_MAX; ++afi) {
-		if (!vpn_leak_to_vpn_active(bgp, afi, NULL))
+		if (!vpn_leak_to_vpn_active(bgp, afi, NULL, false))
 			continue;
 
 		if (withdraw) {

bgpd/bgp_mplsvpn.h

@@ -113,7 +113,8 @@ static inline bool is_bgp_vrf_mplsvpn(struct bgp *bgp)
 }
 
 static inline int vpn_leak_to_vpn_active(struct bgp *bgp_vrf, afi_t afi,
-					 const char **pmsg)
+					 const char **pmsg,
+					 bool ignore_export_rt_list)
 {
 	if (bgp_vrf->inst_type != BGP_INSTANCE_TYPE_VRF
 		&& bgp_vrf->inst_type != BGP_INSTANCE_TYPE_DEFAULT) {
@@ -133,8 +134,21 @@ static inline int vpn_leak_to_vpn_active(struct bgp *bgp_vrf, afi_t afi,
 		return 0;
 	}
 
-	/* Is there an RT list set? */
+	/* Before performing withdrawal, VPN activation is checked; however,
-	if (!bgp_vrf->vpn_policy[afi].rtlist[BGP_VPN_POLICY_DIR_TOVPN]) {
+	 * when the route-map modifies the export route-target (RT) list, it
+	 * becomes challenging to determine if VPN prefixes were previously
+	 * present, or not. The 'ignore_export_rt_list' parameter will be
+	 * used to force the withdraw operation by not checking the possible
+	 * route-map changes.
+	 * Of the 'ignore_export_rt_list' is set to false, check the following:
+	 * - Is there an RT list set?
+	 * - Is there a route-map that sets RT communities
+	 */
+	if (!ignore_export_rt_list &&
+	    !bgp_vrf->vpn_policy[afi].rtlist[BGP_VPN_POLICY_DIR_TOVPN] &&
+	    (!bgp_vrf->vpn_policy[afi].rmap[BGP_VPN_POLICY_DIR_TOVPN] ||
+	     !bgp_route_map_has_extcommunity_rt(
+		     bgp_vrf->vpn_policy[afi].rmap[BGP_VPN_POLICY_DIR_TOVPN]))) {
 		if (pmsg)
 			*pmsg = "rtlist tovpn not defined";
 		return 0;
@@ -246,8 +260,7 @@ static inline void vpn_leak_prechange(enum vpn_policy_direction direction,
 		vpn_leak_to_vrf_withdraw_all(bgp_vrf, afi);
 	}
 	if ((direction == BGP_VPN_POLICY_DIR_TOVPN) &&
-		vpn_leak_to_vpn_active(bgp_vrf, afi, NULL)) {
+	    vpn_leak_to_vpn_active(bgp_vrf, afi, NULL, true)) {
-
 		vpn_leak_from_vrf_withdraw_all(bgp_vpn, bgp_vrf, afi);
 	}
 }

bgpd/bgp_routemap.c

@@ -4743,6 +4743,24 @@ static void bgp_route_map_delete(const char *rmap_name)
 	route_map_notify_dependencies(rmap_name, RMAP_EVENT_MATCH_DELETED);
 }
 
+bool bgp_route_map_has_extcommunity_rt(const struct route_map *map)
+{
+	struct route_map_index *index = NULL;
+	struct route_map_rule *set = NULL;
+
+	assert(map);
+
+	for (index = map->head; index; index = index->next) {
+		for (set = index->set_list.head; set; set = set->next) {
+			if (set->cmd && set->cmd->str &&
+			    (strmatch(set->cmd->str, "extcommunity rt") ||
+			     strmatch(set->cmd->str, "extended-comm-list")))
+				return true;
+		}
+	}
+	return false;
+}
+
 static void bgp_route_map_event(const char *rmap_name)
 {
 	if (route_map_mark_updated(rmap_name) == 0)

bgpd/bgpd.h

@@ -2446,6 +2446,8 @@ extern enum asnotation_mode bgp_get_asnotation(struct bgp *bgp);
 
 extern void bgp_route_map_terminate(void);
 
+extern bool bgp_route_map_has_extcommunity_rt(const struct route_map *map);
+
 extern int peer_cmp(struct peer *p1, struct peer *p2);
 
 extern int bgp_map_afi_safi_iana2int(iana_afi_t pkt_afi, iana_safi_t pkt_safi,