bgpd: Ignore peer GR commands that are effectively not operable

The code used to treat a repeated GR configuration on a peer or some
other inappropriate command (e.g., trying to remove 'helper' configuration
when it is not present) as errors. Instead, just ignore these. This is
more in line with other behavior.

Signed-off-by: Vivek Venkatraman <[email protected]>

Ticket: #2736244, #2736249
Testing Done:
1. Manual testing - documented in the RM tickets
2. Precommit - user job #15 - 1 failure seen is existing failure

Author: vivek-cumulus

bgpd/bgp_fsm.c

@@ -2744,7 +2744,11 @@ int bgp_gr_update_all(struct bgp *bgp, int global_gr_cmd)
 				"%s [BGP_GR] global_new_state == global_old_state :%s",
 				__func__,
 				print_global_gr_mode(global_new_state));
-		return BGP_GR_NO_OPERATION;
+        /* Next state 'invalid' is actually an 'ignore' */
+        if (global_new_state == GLOBAL_INVALID)
+            return BGP_GR_NO_OPERATION;
+
+        return BGP_GR_NO_OPERATION;
 	}
 
 	return bgp_gr_lookup_n_update_all_peer(bgp, global_new_state,
@@ -2892,6 +2896,10 @@ int bgp_neighbor_graceful_restart(struct peer *peer, int peer_gr_cmd)
 		return BGP_ERR_GR_INVALID_CMD;
 	}
 
+    /* Next state 'invalid' is actually an 'ignore' */
+	if (peer_new_state == PEER_INVALID)
+		return BGP_GR_NO_OPERATION;
+
 	if (peer_new_state != peer_old_state) {
 		result = peer_state.action_fun(peer, peer_old_state,
 					       peer_new_state);
@@ -2940,6 +2948,9 @@ unsigned int bgp_peer_gr_action(struct peer *peer, int old_peer_state,
 		return BGP_ERR_GR_INVALID_CMD;
 	}
 
+	if (new_peer_state == PEER_INVALID)
+		return BGP_GR_NO_OPERATION;
+
 	bgp_gr_global_mode = bgp_global_gr_mode_get(peer->bgp);
 
 	if ((old_peer_state == PEER_GLOBAL_INHERIT)