cgrates: handle inter-mixing of requests and events

In case when OpenSIPS sends a request, but before receiving the reply on
the same connection, it receives an event, we need to make sure the
reply is not marked as completed.
Before this fix, when the actual reply was coming back, we no longer had
a reply handler, therefore OpenSIPS was crashing.

Reported by Vasilios Tzanoudakis (vtzan on GitHub) in ticket #1713

modules/cgrates/cgrates_common.c

@@ -597,7 +597,8 @@ static int cgrates_async_resume_repl(int fd,
 
 	ret = cgrc_async_read(c, cp->reply_f, cp->reply_p);
 
-	if (async_status == ASYNC_DONE) {
+	switch (async_status) {
+	case ASYNC_DONE:
 		/* processing done - remove the FD and replace the handler */
 		async_status = ASYNC_DONE_NO_IO;
 		reactor_del_reader(c->fd, -1, 0);
@@ -606,6 +607,9 @@ static int cgrates_async_resume_repl(int fd,
 			ret = -1;
 			goto end;
 		}
+		break;
+	case ASYNC_CONTINUE:
+		return 1;
 	}
 end:
 	/* done with this connection */
@@ -759,6 +763,14 @@ int cgrc_async_read(struct cgr_conn *c,
 	}
 	/* all done */
 	json_tokener_reset(c->jtok);
+	/* if a request was processed and we were waiting for a reply, indicate
+	 * that we shold continue reading from this socket, otherwise we will
+	 * wrongfully waiting for a reply
+	 */
+	if (ret == 0 && f) {
+		async_status = ASYNC_CONTINUE;
+		return 1;
+	}
 done:
 	async_status = ASYNC_DONE;
 	return final_ret;