Merge pull request FRRouting#20302 from Pdoijode/pdoijode/bgp-err-codes

Add error codes and improve operator messaging

Author: mjstapp

bgpd/bgp_attr.c

@@ -470,9 +470,8 @@ static bool bgp_attr_aigp_get_tlv_metric(uint8_t *pnt, int length,
 			 * and SHOULD be discarded as specified in this section.
 			 */
 			if (*aigp == BGP_AIGP_TLV_METRIC_MAX) {
-				zlog_err("Bad AIGP TLV (%s) length: %llu",
-					 BGP_AIGP_TLV_METRIC_DESC,
-					 BGP_AIGP_TLV_METRIC_MAX);
+				flog_err(EC_BGP_ATTR_AIGP, "Bad AIGP TLV (%s) length: %llu",
+					 BGP_AIGP_TLV_METRIC_DESC, BGP_AIGP_TLV_METRIC_MAX);
 				return false;
 			}
 
@@ -558,7 +557,7 @@ static bool bgp_attr_aigp_valid(uint8_t *pnt, int length)
 	uint8_t *end = data + length;
 
 	if (length < 3) {
-		zlog_err("Bad AIGP attribute length (MUST be minimum 3): %u",
+		flog_err(EC_BGP_ATTR_AIGP, "Bad AIGP attribute length (MUST be minimum 3): %u",
 			 length);
 		return false;
 	}
@@ -575,22 +574,22 @@ static bool bgp_attr_aigp_valid(uint8_t *pnt, int length)
 		(void)data;
 
 		if (length < tlv_length) {
-			zlog_err(
-				"Bad AIGP attribute length: %u, but TLV length: %u",
-				length, tlv_length);
+			flog_err(EC_BGP_ATTR_AIGP,
+				 "Bad AIGP attribute length: %u, but TLV length: %u", length,
+				 tlv_length);
 			return false;
 		}
 
 		if (tlv_length < 3) {
-			zlog_err("Bad AIGP TLV length (MUST be minimum 3): %u",
+			flog_err(EC_BGP_ATTR_AIGP, "Bad AIGP TLV length (MUST be minimum 3): %u",
 				 tlv_length);
 			return false;
 		}
 
 		/* AIGP TLV, Length: 11 */
 		if (tlv_type == BGP_AIGP_TLV_METRIC &&
 		    tlv_length != BGP_AIGP_TLV_METRIC_LEN) {
-			zlog_err("Bad AIGP TLV (%s) length: %u",
+			flog_err(EC_BGP_ATTR_AIGP, "Bad AIGP TLV (%s) length: %u",
 				 BGP_AIGP_TLV_METRIC_DESC, tlv_length);
 			return false;
 		}
@@ -3089,8 +3088,8 @@ static int bgp_attr_encap(struct bgp_attr_parser_args *args)
 
 	if (!CHECK_FLAG(flag, BGP_ATTR_FLAG_TRANS)
 	    || !CHECK_FLAG(flag, BGP_ATTR_FLAG_OPTIONAL)) {
-		zlog_err("Tunnel Encap attribute flag isn't optional and transitive %d",
-			 flag);
+		flog_err(EC_BGP_ATTR_FLAG,
+			 "Tunnel Encap attribute flag isn't optional and transitive %d", flag);
 		return bgp_attr_malformed(args, BGP_NOTIFY_UPDATE_OPT_ATTR_ERR,
 					  args->total);
 	}
@@ -3100,8 +3099,8 @@ static int bgp_attr_encap(struct bgp_attr_parser_args *args)
 		uint16_t tlv_length;
 
 		if (length < 4) {
-			zlog_err(
-				"Tunnel Encap attribute not long enough to contain outer T,L");
+			flog_err(EC_BGP_ATTR_LEN,
+				 "Tunnel Encap attribute not long enough to contain outer T,L");
 			return bgp_attr_malformed(args,
 						  BGP_NOTIFY_UPDATE_OPT_ATTR_ERR,
 						  args->total);
@@ -3139,15 +3138,17 @@ static int bgp_attr_encap(struct bgp_attr_parser_args *args)
 		}
 
 		if (sublength > length) {
-			zlog_err("Tunnel Encap attribute sub-tlv length %d exceeds remaining length %d",
+			flog_err(EC_BGP_ATTR_LEN,
+				 "Tunnel Encap attribute sub-tlv length %d exceeds remaining length %d",
 				 sublength, length);
 			return bgp_attr_malformed(args,
 						  BGP_NOTIFY_UPDATE_OPT_ATTR_ERR,
 						  args->total);
 		}
 
 		if (STREAM_READABLE(BGP_INPUT(connection)) < sublength) {
-			zlog_err("Tunnel Encap attribute sub-tlv length %d exceeds remaining stream length %zu",
+			flog_err(EC_BGP_ATTR_LEN,
+				 "Tunnel Encap attribute sub-tlv length %d exceeds remaining stream length %zu",
 				 sublength, STREAM_READABLE(BGP_INPUT(connection)));
 			return bgp_attr_malformed(args,
 						  BGP_NOTIFY_UPDATE_OPT_ATTR_ERR,
@@ -3199,8 +3200,8 @@ static int bgp_attr_encap(struct bgp_attr_parser_args *args)
 
 	if (length) {
 		/* spurious leftover data */
-		zlog_err("Tunnel Encap attribute length is bad: %d leftover octets",
-			 length);
+		flog_err(EC_BGP_ATTR_LEN,
+			 "Tunnel Encap attribute length is bad: %d leftover octets", length);
 		return bgp_attr_malformed(args, BGP_NOTIFY_UPDATE_OPT_ATTR_ERR,
 					  args->total);
 	}

bgpd/bgp_errors.c

@@ -491,6 +491,84 @@ static struct log_ref ferr_bgp_err[] = {
 		.description = "BGP could not find the label for the specified path_info",
 		.suggestion = "Most likely a bug. This should never happen. Please try to build a reproducer and report.",
 	},
+	{
+		.code = EC_BGP_ROUTE_REFRESH_INVALID,
+		.title = "BGP route refresh message format error",
+		.description = "BGP received an invalid or malformed route refresh message from peer (enhanced route refresh length error, invalid subtype, or EoRR without BoRR)",
+		.suggestion = "Check BGP capability compatibility with peer. Verify enhanced route refresh configuration and ensure proper BoRR/EoRR sequence.",
+	},
+	{
+		.code = EC_BGP_PER_SRC_NHG,
+		.title = "BGP per-source nexthop group resolution failed",
+		.description = "BGP unable to resolve IPv6 nexthop for per-source nexthop group",
+		.suggestion = "Ensure the nexthop group and advertised paths include IPv6 information. Verify per-source nexthop group configuration.",
+	},
+	{
+		.code = EC_BGP_ROUTE_ATTRIBUTE_MALFORMED,
+		.title = "BGP route attribute is malformed or invalid",
+		.description = "BGP received or processed a route with malformed attributes (prefix, RD, ESI, MAC, gateway IP, L2VPN prefix data, address family mismatches, route-map/label-index mismatches) or referenced route not found",
+		.suggestion = "Check route definitions and EVPN/L2VPN configuration. Verify peer configuration for attribute consistency, address family matching, and static route configuration alignment. Ensure all referenced routes exist and route data format is correct.",
+	},
+	{
+		.code = EC_BGP_CONFIG_CONSTRAINT_VIOLATION,
+		.title = "BGP configuration constraint violated",
+		.description = "BGP configuration attempted to change an immutable parameter (e.g., label-index)",
+		.suggestion = "Review configuration changes to ensure immutable parameters are not modified after initial setup.",
+	},
+	{
+		.code = EC_BGP_ROUTE_MAP_SCRIPT,
+		.title = "BGP route-map script execution failed",
+		.description = "BGP route-map script or function failed to load or execute properly",
+		.suggestion = "Debug the route-map script and ensure it returns successful status. Verify script accessibility and correct any errors in the script file.",
+	},
+	{
+		.code = EC_BGP_VRF_NOT_FOUND,
+		.title = "BGP VRF not found",
+		.description = "BGP requested VRF does not exist or cannot be found",
+		.suggestion = "Create the required VRF or correct the configured VRF name.",
+	},
+	{
+		.code = EC_BGP_ZEBRA_MSG_DECODE,
+		.title = "BGP failed to decode message from Zebra",
+		.description = "BGP received a malformed or incompatible message from Zebra daemon",
+		.suggestion = "Ensure BGP and Zebra daemons are version-synchronized and IPC communication is healthy. Check for version mismatches between daemons.",
+	},
+	{
+		.code = EC_BGP_SRV6_LOCATOR_MISMATCH,
+		.title = "BGP SRv6 locator name mismatch",
+		.description = "BGP SRv6 locator advertised by BGP doesn't match the locator name in the SRv6 chunk from Zebra",
+		.suggestion = "Fix the inconsistent SRv6 locator configuration between BGP and Zebra.",
+	},
+	{
+		.code = EC_BGP_INVALID_BGP_INSTANCE_ID,
+		.title = "BGP instance ID out of range",
+		.description = "BGP instance ID specified is outside the valid range",
+		.suggestion = "Set a valid instance ID in the configuration within the allowed range.",
+	},
+	{
+		.code = EC_BGP_SRV6_SID_MODE_CONFLICT,
+		.title = "BGP SRv6 SID mode configuration conflict",
+		.description = "BGP SRv6 configuration has both index-mode and auto-mode enabled simultaneously",
+		.suggestion = "Choose one SRv6 SID mode (typically auto), remove the other from the config, and reapply the import/export settings.",
+	},
+	{
+		.code = EC_BGP_CONFIG_TIMEOUT,
+		.title = "BGP configuration timeout",
+		.description = "BGP configuration end timer expired during config application",
+		.suggestion = "Review provisioning speed or increase the configuration timeout value.",
+	},
+	{
+		.code = EC_BGP_LABEL_POOL_INSERT_FAIL,
+		.title = "BGP label pool insertion failed",
+		.description = "BGP failed to insert a label into the label pool tracking structure",
+		.suggestion = "Restart BGP service. If issues persist, gather log files and open an issue.",
+	},
+	{
+		.code = EC_BGP_TTL_SECURITY_FAIL,
+		.title = "BGP TTL security configuration failed",
+		.description = "BGP failed to configure TTL security parameters (min/max TTL) on peer connection",
+		.suggestion = "Ensure both peers support TTL security feature and verify kernel supports TTL security socket options. Check peer TTL security configuration for compatibility.",
+	},
 	{
 		.code = END_FERR,
 	}

bgpd/bgp_errors.h

@@ -93,6 +93,19 @@ enum bgp_log_refs {
 	EC_BGP_SENDQ_STUCK_WARN,
 	EC_BGP_SENDQ_STUCK_PROPER,
 	EC_BGP_PATH_WITHOUT_LABEL,
+	EC_BGP_ROUTE_REFRESH_INVALID,
+	EC_BGP_PER_SRC_NHG,
+	EC_BGP_ROUTE_ATTRIBUTE_MALFORMED,
+	EC_BGP_CONFIG_CONSTRAINT_VIOLATION,
+	EC_BGP_ROUTE_MAP_SCRIPT,
+	EC_BGP_VRF_NOT_FOUND,
+	EC_BGP_ZEBRA_MSG_DECODE,
+	EC_BGP_SRV6_LOCATOR_MISMATCH,
+	EC_BGP_INVALID_BGP_INSTANCE_ID,
+	EC_BGP_SRV6_SID_MODE_CONFLICT,
+	EC_BGP_CONFIG_TIMEOUT,
+	EC_BGP_LABEL_POOL_INSERT_FAIL,
+	EC_BGP_TTL_SECURITY_FAIL,
 };
 
 extern void bgp_error_init(void);

bgpd/bgp_fsm.c

@@ -1927,8 +1927,8 @@ static void bgp_connect_check(struct event *event)
 
 	/* If getsockopt is fail, this is fatal error. */
 	if (ret < 0) {
-		zlog_err("can't get sockopt for nonblocking connect: %d(%s)",
-			  errno, safe_strerror(errno));
+		flog_err(EC_LIB_SOCKET, "can't get sockopt for nonblocking connect: %d(%s)", errno,
+			 safe_strerror(errno));
 		BGP_EVENT_ADD(connection, TCP_fatal_error);
 		return;
 	}
@@ -2936,7 +2936,7 @@ bgp_peer_inherit_global_gr_mode(struct peer *peer,
 		break;
 	case GLOBAL_INVALID:
 	default:
-		zlog_err("Unexpected Global GR mode %d", global_gr_mode);
+		flog_err(EC_LIB_DEVELOPMENT, "Unexpected Global GR mode %d", global_gr_mode);
 	}
 }
 
@@ -3240,8 +3240,8 @@ void bgp_peer_move_to_gr_mode(struct peer *peer, enum peer_mode new_state)
 		break;
 	case PEER_INVALID:
 	default:
-		zlog_err(
-			"[BGP_GR] Default switch mode ::: SOMETHING IS WRONG !!!");
+		flog_err(EC_LIB_DEVELOPMENT,
+			 "[BGP_GR] Default switch mode ::: SOMETHING IS WRONG !!!");
 		break;
 	}
 	bgp_peer_gr_flags_update(peer);

bgpd/bgp_labelpool.c

@@ -296,8 +296,9 @@ static mpls_label_t get_label_from_pool(void *labelid)
 		lbl = chunk->first + index;
 		if (skiplist_insert(lp->inuse, (void *)lbl, labelid)) {
 			/* something is very wrong */
-			zlog_err("%s: unable to insert inuse label %u (id %p)",
-				 __func__, (uint32_t)lbl, labelid);
+			flog_err(EC_BGP_LABEL_POOL_INSERT_FAIL,
+				 "%s: unable to insert inuse label %u (id %p)", __func__,
+				 (uint32_t)lbl, labelid);
 			return MPLS_LABEL_NONE;
 		}
 

bgpd/bgp_main.c

@@ -484,8 +484,9 @@ int main(int argc, char **argv)
 		case 'I':
 			instance = atoi(optarg);
 			if (instance > (unsigned short)-1)
-				zlog_err("Instance %i out of range (0..%u)",
-					 instance, (unsigned short)-1);
+				flog_err(EC_BGP_INVALID_BGP_INSTANCE_ID,
+					 "Instance %i out of range (0..%u)", instance,
+					 (unsigned short)-1);
 			break;
 		case 's':
 			buffer_size = atoi(optarg);

bgpd/bgp_mplsvpn.c

@@ -827,7 +827,8 @@ void ensure_vrf_tovpn_sid_per_af(struct bgp *bgp_vpn, struct bgp *bgp_vrf,
 	if ((tovpn_sid_index != 0 && tovpn_sid_auto) ||
 	    (tovpn_sid_index != 0 && is_tovpn_sid_explicit) ||
 	    (tovpn_sid_auto && is_tovpn_sid_explicit)) {
-		zlog_err("%s: more than one mode selected among index-mode, auto-mode and explicit-mode. ignored.",
+		flog_err(EC_BGP_SRV6_SID_MODE_CONFLICT,
+			 "%s: more than one mode selected among index-mode, auto-mode and explicit-mode. ignored.",
 			 __func__);
 		return;
 	}
@@ -915,7 +916,8 @@ void ensure_vrf_tovpn_sid_per_vrf(struct bgp *bgp_vpn, struct bgp *bgp_vrf)
 	if ((tovpn_sid_index != 0 && tovpn_sid_auto) ||
 	    (tovpn_sid_index != 0 && is_tovpn_sid_explicit) ||
 	    (tovpn_sid_auto && is_tovpn_sid_explicit)) {
-		zlog_err("%s: more than one mode selected among index-mode, auto-mode and explicit-mode. ignored.",
+		flog_err(EC_BGP_SRV6_SID_MODE_CONFLICT,
+			 "%s: more than one mode selected among index-mode, auto-mode and explicit-mode. ignored.",
 			 __func__);
 		return;
 	}

bgpd/bgp_network.c

@@ -217,11 +217,10 @@ static void bgp_update_setsockopt_tcp_keepalive(struct bgp *bgp, int fd)
 					       bgp->tcp_keepalive_intvl,
 					       bgp->tcp_keepalive_probes);
 		if (ret < 0)
-			zlog_err(
-				"Can't set TCP keepalive on socket %d, idle %u intvl %u probes %u",
-				fd, bgp->tcp_keepalive_idle,
-				bgp->tcp_keepalive_intvl,
-				bgp->tcp_keepalive_probes);
+			flog_err(EC_LIB_SOCKET,
+				 "Can't set TCP keepalive on socket %d, idle %u intvl %u probes %u",
+				 fd, bgp->tcp_keepalive_idle, bgp->tcp_keepalive_intvl,
+				 bgp->tcp_keepalive_probes);
 	}
 }
 
@@ -522,7 +521,8 @@ static void bgp_accept(struct event *event)
 
 			if (bgp_set_socket_ttl(incoming) < 0) {
 				peer_set_last_reset(dynamic_peer, PEER_DOWN_SOCKET_ERROR);
-				zlog_err("%s: Unable to set min/max TTL on peer %s (dynamic), error received: %s(%d)",
+				flog_err(EC_BGP_TTL_SECURITY_FAIL,
+					 "%s: Unable to set min/max TTL on peer %s (dynamic), error received: %s(%d)",
 					 __func__, dynamic_peer->host, safe_strerror(errno), errno);
 				frrtrace(3, frr_bgp, bgp_err_str, dynamic_peer->host,
 					 dynamic_peer->flags, 1);