bgpd: Link L2VNI to L3VNI only when it really exists

When a L2VNI is made known to bgpd, the current linking of it to L3VNI
does not quite check if the L3VNI is known to bgpd. This is incorrect
and can lead to problematic behavior in some situations. In one
specific instance, if type-1 routes from a remote MH pair already
exist, the ES-VRF association gets prematurely setup leading to the
corresponding L3NHG getting setup with incorrect information.

Note: Only additional checks are added in the linkage function. Code
already exists to link all relevant L2VNIs to associated L3VNI when the
L3VNI is learnt by bgpd.

Signed-off-by: Vivek Venkatraman <vivek@nvidia.com>

Author: vivek-cumulus

bgpd/bgp_evpn_private.h

@@ -280,20 +280,23 @@ static inline void bgpevpn_link_to_l3vni(struct bgpevpn *vpn)
 	if (vpn->bgp_vrf)
 		return;
 
+	/* bail if VRF still doesn't exist */
 	bgp_vrf = bgp_lookup_by_vrf_id(vpn->tenant_vrf_id);
 	if (!bgp_vrf)
 		return;
 
+	/* or if there is no l3vni */
+	if (!bgp_vrf->l3vni)
+		return;
+
 	/* associate the vpn to the bgp_vrf instance */
 	vpn->bgp_vrf = bgp_lock(bgp_vrf);
 	listnode_add_sort(bgp_vrf->l2vnis, vpn);
 
 	/*
-	 * If L3VNI is configured,
 	 * check if we are advertising two labels for this vpn
 	 */
-	if (bgp_vrf->l3vni &&
-	    !CHECK_FLAG(bgp_vrf->vrf_flags, BGP_VRF_L3VNI_PREFIX_ROUTES_ONLY))
+	if (!CHECK_FLAG(bgp_vrf->vrf_flags, BGP_VRF_L3VNI_PREFIX_ROUTES_ONLY))
 		SET_FLAG(vpn->flags, VNI_FLAG_USE_TWO_LABELS);
 
 	bgp_evpn_es_evi_vrf_ref(vpn);