re-main-tmr-thread-check-002.patch

diff --git a/src/main/main.c b/src/main/main.c
index 78a47e3..7858aa9 100644
--- a/src/main/main.c
+++ b/src/main/main.c
@@ -115,6 +115,7 @@ struct re {
 #ifdef HAVE_PTHREAD
 	pthread_mutex_t mutex;       /**< Mutex for thread synchronization  */
 	pthread_mutex_t *mutexp;     /**< Pointer to active mutex           */
+	pthread_t tid;
 #endif
 };
 
@@ -145,12 +146,16 @@ static struct re global_re = {
 	PTHREAD_MUTEX_INITIALIZER,
 #endif
 	&global_re.mutex,
+	0
 #endif
 };
 
 
 #ifdef HAVE_PTHREAD
 
+extern pthread_t re_thread_get(void);
+
+
 static void poll_close(struct re *re);
 
 static pthread_once_t pt_once = PTHREAD_ONCE_INIT;
@@ -181,6 +186,11 @@ static struct re *re_get(void)
 		re = &global_re;
 	}
 
+	if (!re->tid) {
+		re->tid = pthread_self();
+		re_printf("  *** RE MAIN THREAD INITED: tid=%p\n", re->tid);
+	}
+
 	return re;
 }
 
@@ -599,6 +609,15 @@ int fd_listen(int fd, int flags, fd_h *fh, void *arg)
 
 	DEBUG_INFO("fd_listen: fd=%d flags=0x%02x\n", fd, flags);
 
+#ifdef HAVE_PTHREAD
+	if (!pthread_equal(pthread_self(), re_thread_get())) {
+		DEBUG_WARNING("fd_listen: called from a NON-RE thread"
+			      " (fd=%d, flags=0x%x, handler=%p, arg=%p)\n",
+			      fd, flags, fh, arg);
+		return EPERM;
+	}
+#endif
+
 	if (fd < 0) {
 		DEBUG_WARNING("fd_listen: corrupt fd %d\n", fd);
 		return EBADF;
@@ -1259,3 +1278,11 @@ struct list *tmrl_get(void)
 {
 	return &re_get()->tmrl;
 }
+
+
+#ifdef HAVE_PTHREAD
+pthread_t re_thread_get(void)
+{
+	return re_get()->tid;
+}
+#endif
diff --git a/src/tmr/tmr.c b/src/tmr/tmr.c
index 5e9022c..a6ce86c 100644
--- a/src/tmr/tmr.c
+++ b/src/tmr/tmr.c
@@ -268,6 +268,11 @@ void tmr_init(struct tmr *tmr)
 }
 
 
+#ifdef HAVE_PTHREAD
+	extern pthread_t re_thread_get(void);
+#endif
+
+
 /**
  * Start a timer
  *
@@ -284,6 +289,16 @@ void tmr_start(struct tmr *tmr, uint64_t delay, tmr_h *th, void *arg)
 	if (!tmr)
 		return;
 
+#ifdef HAVE_PTHREAD
+
+	if (!pthread_equal(pthread_self(), re_thread_get())) {
+		DEBUG_WARNING("tmr_start: called from a NON-RE thread"
+			      " (tmr=%p, delay=%llu, handler=%p, arg=%p)\n",
+			      tmr, delay, th, arg);
+		BREAKPOINT;
+	}
+#endif
+
 	if (tmr->th) {
 		list_unlink(&tmr->le);
 	}

I like the idea. Should we implement this in baresip/re by default? Maybe for non RELEASE only:

#if defined(HAVE_PTHREAD) && !defined(RELEASE)

that would be great.

it would catch many bugs, that is hard for application developers to find.

hi @juha-h here is the patch, please apply locally and test

Alfred E. Heggestad writes:

hi @juha-h here is the patch, please apply locally and test

Thanks, I'll try. The patch needs some editing since now, for example, #ifdef HAVE_PTHREAD is gone.

juha-h writes:

Alfred E. Heggestad writes: > hi @juha-h here is the patch, please apply locally and test Thanks, I'll try. The patch needs some editing since now, for example, #ifdef HAVE_PTHREAD is gone.

The patch is too much out of sync for me to apply correctly. For example here @@ -145,12 +146,16 @@ static struct re global_re = { PTHREAD_MUTEX_INITIALIZER, #endif &global_re.mutex, + 0 #endif }; global_re does not exist anymore. Can you create new one patch that can be applied to current version of re?

I think it would always show a different thread id since juha is using re_thread_enter/re_thread_leave from another thread https://github.com/baresip/baresip/wiki/Using-baresip-as-a-library#examples-using-re-lock
So it needs maybe a slight modification for this use case anyway. Will try to bring this patch upstream with re_thread_enter detection, since its really useful for finding these kinds of bugs.

Ok you can give this PR a try: baresip/re#389

Sebastian Reimers writes:

Ok you can give this PR a try: baresip/re#389

Tried and the delay is still there. Now I started to get lots of these: 06-06 18:42:51.712 16953 17131 D Baresip Lib: main: thread check: called from a NON-RE thread without thread_enter()! They disappeared when I added re_thread_enter()/leave() around this call: JNIEXPORT void JNICALL Java_com_tutpro_baresip_Api_net_1debug(JNIEnv *env, jobject thiz) { re_thread_enter(); net_debug_log(); re_thread_leave(); } But as I wrote, the long delay when I make calls, is still there even after fixing net_debug.

I noticed also, that sometimes explicit registering has long delay: 06-06 18:56:45.448 17867 17867 D Baresip Lib: ua event (REGISTERING) 06-06 18:57:08.284 17867 18041 D Baresip Lib: ***@***.***: (prio 0) {1/TLS/v4} 200 OK (OpenSIPg SIP Proxy (5.6.0-0b31 (x86_64/linux))) [1 binding] 06-06 18:57:08.286 17867 18040 D Baresip Lib: ua event (REGISTER_OK) 200 OK It too needs to resolve outbound address.

…

-- Juha