Fix: pacemaker-attrd: prevent segfault if a peer leaves when its name is unknown yet

If nodes are not configured with names in corosync, when a node joins
the cluster, it takes a while for it to learn the names of all the other
online peers. Previously, if any of the peers happened to leave when its
`name` was unknown to us yet, pacemaker-attrd would trigger assertion in
`attrd_peer_remove()` and segfault in
`attrd_remove_peer_protocol_ver()`.

Author: gao-yan

daemons/attrd/attrd_corosync.c

@@ -195,14 +195,16 @@ attrd_peer_change_cb(enum pcmk__node_update kind, pcmk__node_status_t *peer,
                 }
             } else {
                 // Remove all attribute values associated with lost nodes
-                attrd_peer_remove(peer->name, false, "loss");
+                if (peer->name != NULL) {
+                    attrd_peer_remove(peer->name, false, "loss");
+                }
                 gone = true;
             }
             break;
     }
 
     // Remove votes from cluster nodes that leave, in case election in progress
-    if (gone && !is_remote) {
+    if (gone && !is_remote && peer->name != NULL) {
         attrd_remove_voter(peer);
         attrd_remove_peer_protocol_ver(peer->name);
         attrd_do_not_expect_from_peer(peer->name);