clusterer: fix possible deadlocks when discovering a new node

Use only standard RW locking operations for the global list lock instead of the "switchable" mechanism (aquiring the lock for writing when necessary).
OpenSIPS · Jul 14, 2022 · e0e9e2f · e0e9e2f
1 parent 4ab5c63
commit e0e9e2f
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 13 deletions.
diff --git a/modules/clusterer/clusterer.c b/modules/clusterer/clusterer.c
@@ -958,7 +958,6 @@ static void handle_cl_mi_msg(bin_packet_t *packet)
 static void handle_remove_node(bin_packet_t *packet, cluster_info_t *cl)
 {
 	int target_node;
-	int lock_old_flag;
 	node_info_t *node;
 	int ev_actions_cl = 1;
 
@@ -995,9 +994,7 @@ static void handle_remove_node(bin_packet_t *packet, cluster_info_t *cl)
 		return;
 	}
 
-	lock_switch_write(cl_list_lock, lock_old_flag);
 	remove_node(cl, node);
-	lock_switch_read(cl_list_lock, lock_old_flag);
 }
 
 void bin_rcv_cl_extra_packets(bin_packet_t *packet, int packet_type,
@@ -1024,7 +1021,7 @@ void bin_rcv_cl_extra_packets(bin_packet_t *packet, int packet_type,
 	}
 
 	if (!db_mode && packet_type == CLUSTERER_REMOVE_NODE)
-		lock_start_sw_read(cl_list_lock);
+		lock_start_write(cl_list_lock);
 	else
 		lock_start_read(cl_list_lock);
 
@@ -1119,7 +1116,7 @@ void bin_rcv_cl_extra_packets(bin_packet_t *packet, int packet_type,
 
 exit:
 	if (!db_mode && packet_type == CLUSTERER_REMOVE_NODE)
-		lock_stop_sw_read(cl_list_lock);
+		lock_stop_write(cl_list_lock);
 	else
 		lock_stop_read(cl_list_lock);
 }
@@ -1151,7 +1148,7 @@ void bin_rcv_cl_packets(bin_packet_t *packet, int packet_type,
 
 	if (!db_mode && (packet_type == CLUSTERER_NODE_DESCRIPTION ||
 		packet_type == CLUSTERER_FULL_TOP_UPDATE))
-		lock_start_sw_read(cl_list_lock);
+		lock_start_write(cl_list_lock);
 	else
 		lock_start_read(cl_list_lock);
 
@@ -1199,7 +1196,7 @@ void bin_rcv_cl_packets(bin_packet_t *packet, int packet_type,
 exit:
 	if (!db_mode && (packet_type == CLUSTERER_NODE_DESCRIPTION ||
 		packet_type == CLUSTERER_FULL_TOP_UPDATE))
-		lock_stop_sw_read(cl_list_lock);
+		lock_stop_write(cl_list_lock);
 	else
 		lock_stop_read(cl_list_lock);
 }

diff --git a/modules/clusterer/topology.c b/modules/clusterer/topology.c
@@ -682,7 +682,6 @@ static node_info_t *add_node(bin_packet_t *received, cluster_info_t *cl,
 								int src_node_id, str *str_vals, int *int_vals)
 {
 	node_info_t *new_node = NULL;
-	int lock_old_flag;
 
 	str_vals[STR_VALS_FLAGS_COL].s = 0;
 	str_vals[STR_VALS_DESCRIPTION_COL].s = 0;
@@ -691,16 +690,11 @@ static node_info_t *add_node(bin_packet_t *received, cluster_info_t *cl,
 	int_vals[INT_VALS_NODE_ID_COL] = src_node_id;
 	int_vals[INT_VALS_STATE_COL] = 1;	/* enabled */
 
-	lock_switch_write(cl_list_lock, lock_old_flag);
-
 	if (add_node_info(&new_node, &cl, int_vals, str_vals) != 0) {
 		LM_ERR("Unable to add node info to backing list\n");
-		lock_switch_read(cl_list_lock, lock_old_flag);
 		return NULL;
 	}
 
-	lock_switch_read(cl_list_lock, lock_old_flag);
-
 	return new_node;
 }