ctdb-tests: Do not attempt to unregister the join handler multiple times

MSG_ID_SYNC is broadcast to each node when a MSG_ID_JOIN has been
received from all nodes.  After MSG_ID_SYNC is successfully broadcast,
the join handler is unregistered.  However, if another MSG_ID_JOIN is
received before the join handler is unregistered then MSG_ID_SYNC is
re-broadcast.  This results in multiple attempts to unregister the
join handler.

Once all MSG_ID_JOIN messages are received, unregister the join handler
to ignore any extra MSG_ID_JOIN messages.  Also, make sure that while
join handler is being unregistered, MSG_ID_JOIN messages are ignored.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12500

Identified-by: Martin Schwenke <martin@meltin.net>
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Fri Jan  6 12:27:23 CET 2017 on sn-devel-144

(cherry picked from commit 4635c22411a7864dd70703f854ec9844816e0294)

1 files changed, 24 insertions, 16 deletions

diff --git a/ctdb/tests/src/cluster_wait.c b/ctdb/tests/src/cluster_wait.c
index ddc3e02cbff..1405738ac47 100644
--- a/ctdb/tests/src/cluster_wait.c
+++ b/ctdb/tests/src/cluster_wait.c
@@ -36,6 +36,7 @@ struct cluster_wait_state {
 	struct ctdb_client_context *client;
 	int num_nodes;
 	bool *ready;
+	bool join_done;
 };
 
 static void cluster_wait_join_registered(struct tevent_req *subreq);
@@ -44,8 +45,8 @@ static void cluster_wait_join(struct tevent_req *subreq);
 static void cluster_wait_join_sent(struct tevent_req *subreq);
 static void cluster_wait_join_handler(uint64_t srvid, TDB_DATA data,
 				      void *private_data);
-static void cluster_wait_sync_sent(struct tevent_req *subreq);
 static void cluster_wait_join_unregistered(struct tevent_req *subreq);
+static void cluster_wait_sync_sent(struct tevent_req *subreq);
 static void cluster_wait_sync_handler(uint64_t srvid, TDB_DATA data,
 				      void *private_data);
 static void cluster_wait_sync_unregistered(struct tevent_req *subreq);
@@ -67,6 +68,8 @@ struct tevent_req *cluster_wait_send(TALLOC_CTX *mem_ctx,
 	state->client = client;
 	state->num_nodes = num_nodes;
 
+	state->join_done = false;
+
 	if (ctdb_client_pnn(client) == 0) {
 		state->ready = talloc_zero_array(state, bool, num_nodes);
 		if (tevent_req_nomem(state->ready, req)) {
@@ -201,7 +204,6 @@ static void cluster_wait_join_handler(uint64_t srvid, TDB_DATA data,
 		private_data, struct tevent_req);
 	struct cluster_wait_state *state = tevent_req_data(
 		req, struct cluster_wait_state);
-	struct ctdb_req_message msg;
 	struct tevent_req *subreq;
 	uint32_t pnn;
 	int i;
@@ -228,50 +230,56 @@ static void cluster_wait_join_handler(uint64_t srvid, TDB_DATA data,
 		}
 	}
 
-	msg.srvid = MSG_ID_SYNC;
-	msg.data.data = tdb_null;
+	if (state->join_done) {
+		return;
+	}
 
-	subreq = ctdb_client_message_send(state, state->ev, state->client,
-					  CTDB_BROADCAST_ALL, &msg);
+	state->join_done = true;
+	subreq = ctdb_client_remove_message_handler_send(
+					state, state->ev, state->client,
+					MSG_ID_JOIN, req);
 	if (tevent_req_nomem(subreq, req)) {
 		return;
 	}
-	tevent_req_set_callback(subreq, cluster_wait_sync_sent, req);
+	tevent_req_set_callback(subreq, cluster_wait_join_unregistered, req);
 }
 
-static void cluster_wait_sync_sent(struct tevent_req *subreq)
+static void cluster_wait_join_unregistered(struct tevent_req *subreq)
 {
 	struct tevent_req *req = tevent_req_callback_data(
 		subreq, struct tevent_req);
 	struct cluster_wait_state *state = tevent_req_data(
 		req, struct cluster_wait_state);
+	struct ctdb_req_message msg;
 	bool status;
 	int ret;
 
-	status = ctdb_client_message_recv(subreq, &ret);
-	TALLOC_FREE(subreq);
+	status = ctdb_client_remove_message_handler_recv(subreq, &ret);
 	if (! status) {
 		tevent_req_error(req, ret);
 		return;
 	}
 
-	subreq = ctdb_client_remove_message_handler_send(
-					state, state->ev, state->client,
-					MSG_ID_JOIN, req);
+	msg.srvid = MSG_ID_SYNC;
+	msg.data.data = tdb_null;
+
+	subreq = ctdb_client_message_send(state, state->ev, state->client,
+					  CTDB_BROADCAST_ALL, &msg);
 	if (tevent_req_nomem(subreq, req)) {
 		return;
 	}
-	tevent_req_set_callback(subreq, cluster_wait_join_unregistered, req);
+	tevent_req_set_callback(subreq, cluster_wait_sync_sent, req);
 }
 
-static void cluster_wait_join_unregistered(struct tevent_req *subreq)
+static void cluster_wait_sync_sent(struct tevent_req *subreq)
 {
 	struct tevent_req *req = tevent_req_callback_data(
 		subreq, struct tevent_req);
 	bool status;
 	int ret;
 
-	status = ctdb_client_remove_message_handler_recv(subreq, &ret);
+	status = ctdb_client_message_recv(subreq, &ret);
+	TALLOC_FREE(subreq);
 	if (! status) {
 		tevent_req_error(req, ret);
 		return;