network-stack: Fix deadlock in perform_on_lwip_thread()

Where in the case of already being on the lwIP thread, we would enqueue
the request and then perform whichever is the oldest pending request. So
when there was already one or more requests pending, we would perform
that request instead. Then next we would join() our request, assuming it
had completed, and end up deadlocking the lwIP thread.

Kudos to @mrmacete for reporting and helping track this one down.

Co-authored-by: Håvard Sørbø <[email protected]>

src/fruity/network-stack.vala

@@ -288,13 +288,13 @@ namespace Frida.Fruity {
 		internal LWIP.ErrorCode perform_on_lwip_thread (owned WorkFunc work) {
 			var req = new Request ((owned) work);
 
-			lock (requests)
-				requests.offer (req);
-
-			if (Thread.self<bool> () != lwip_thread)
+			if (Thread.self<bool> () == lwip_thread) {
+				perform_request (req);
+			} else {
+				lock (requests)
+					requests.offer (req);
 				LWIP.Runtime.schedule (perform_next_request);
-			else
-				perform_next_request ();
+			}
 
 			return req.join ();
 		}
@@ -307,6 +307,10 @@ namespace Frida.Fruity {
 			lock (requests)
 				req = requests.poll ();
 
+			perform_request (res);
+		}
+
+		private void perform_request (Request req) {
 			LWIP.ErrorCode err = req.work ();
 			req.complete (err);
 		}