MT#55283 work around LWS thread non-safety

While LWS explicitly allows usage of lws_callback_on_writable() from
other threads, for some reason there is no internal locking in place,
and so a concurrently running lws_service() can interfere with internal
structures, in particular if lws_service() is closing connections at the
same time as lws_callback_on_writable() is invoked.

The suggested approach of using lws_cancel_service() in combination with
the LWS_CALLBACK_EVENT_WAIT_CANCELLED callback and a user-kept queue is
not feasible, as we need to support LWS 2.x, which doesn't have
LWS_CALLBACK_EVENT_WAIT_CANCELLED.

Closes #1624

Change-Id: Ia3ddeda66fd553c87f99404e0816d97ecbd4cdfe

daemon/websocket.c

@@ -54,6 +54,8 @@ struct websocket_ng_buf {
 static GQueue websocket_vhost_configs;
 static struct lws_context *websocket_context;
 static GThreadPool *websocket_threads;
+static mutex_t websocket_callback_lock = MUTEX_STATIC_INIT;
+static mutex_t websocket_service_lock = MUTEX_STATIC_INIT;
 
 
 static struct websocket_message *websocket_message_new(struct websocket_conn *wc) {
@@ -135,14 +137,42 @@ void websocket_write_raw(struct websocket_conn *wc, const char *msg, size_t len,
 	wo->protocol = protocol;
 	g_queue_push_tail(&wc->output_q, websocket_output_new());
 
+	mutex_unlock(&wc->lock);
+
 	if (done) {
-		lws_callback_on_writable(wc->wsi);
+		// Sadly lws_callback_on_writable() doesn't do any internal
+		// locking, therefore we must protect it against a concurrently
+		// running lws_service(), as well as against other threads
+		// invoking lws_callback_on_writable().
+		//
+		// Acquire the callback lock first, which is normally unlocked,
+		// then wake up the service thread and try to break out of
+		// lws_service().  The service thread holds the service lock
+		// while lws_service() is executing and releases it as soon as
+		// lws_service() is done. We therefore try to acquire the
+		// service lock here, which blocks us until lws_service() is
+		// actually done. At this point the service thread will try to
+		// acquire the callback lock, which is still held by us here,
+		// and so the service thread will block until we are done
+		// calling lws_callback_on_writable(). Finally we release both
+		// locks, which allows the service thread to resume
+		// lws_service().
+		//
+		// The suggested approach of using
+		// LWS_CALLBACK_EVENT_WAIT_CANCELLED together with a queue and
+		// then calling lws_callback_on_writable() from the service
+		// thread is not usable as libwebsockets 2.0 doesn't support
+		// LWS_CALLBACK_EVENT_WAIT_CANCELLED.
+
+		mutex_lock(&websocket_callback_lock);
 		lws_cancel_service(websocket_context);
-	}
 
-	mutex_unlock(&wc->lock);
+		mutex_lock(&websocket_service_lock);
+		lws_callback_on_writable(wc->wsi);
 
-	return 0;
+		mutex_unlock(&websocket_service_lock);
+		mutex_unlock(&websocket_callback_lock);
+	}
 }
 
 
@@ -1124,8 +1154,16 @@ err:
 
 static void websocket_loop(void *p) {
 	ilogs(http, LOG_INFO, "Websocket listener thread running");
-	while (!rtpe_shutdown)
+	while (!rtpe_shutdown) {
+		// see websocket_write_raw() for locking logic
+
+		mutex_lock(&websocket_service_lock);
 		lws_service(websocket_context, 100);
+		mutex_unlock(&websocket_service_lock);
+
+		mutex_lock(&websocket_callback_lock);
+		mutex_unlock(&websocket_callback_lock);
+	}
 
 	websocket_cleanup();
 }