From: Anton Lindqvist <anton@basename.se>
Date: Fri, 3 Dec 2021 19:45:51 +0000 (+0100)
Subject: avoid recursive mutex acquisition in sndio backend
X-Git-Tag: 4.1-rc1~24^2~365^2^2
X-Git-Url: http://git.ipfire.org/gitweb/gitweb.cgi?a=commitdiff_plain;h=refs%2Fpull%2F1362%2Fhead;p=thirdparty%2Fshairport-sync.git

avoid recursive mutex acquisition in sndio backend

Ending a RTSP session while running on OpenBSD using sndio backend causes the
following crash:

	#0  thrkill ()
	#1  0x000005208224403e in _libc_abort
	#2  0x00000520821b77be in _rthread_mutex_trylock
	#3  _rthread_mutex_timedlock
	#4  0x0000051e0d54e2c0 in stop ()
	#5  0x0000051e0d544e85 in player_thread_cleanup_handler
	#6  0x0000052082243126 in _libc_pthread_exit
	#7  0x000005209a158700 in sigthr_handler
	#8  <signal handler called>
	#9  _thread_sys_poll ()
	#10 0x000005208223533e in _libc_poll_cancel
	#11 0x00000520df54c9a0 in sio_psleep
	#12 0x00000520df54cc1f in sio_write
	#13 0x0000051e0d54e27a in play
	#14 0x0000051e0d547fc0 in player_thread_func
	#15 0x000005209a158cc1 in _rthread_start
	#16 0x000005208223565a in __tfork_thread

The player thread is blocking inside sio_write() -> poll(2) while the thread is
being terminated. The stop routine tied to the same backend is invoked through
player_thread_cleanup_handler() which tries to acquire the mutex which it
already acquired before invoking sio_write(). Avoiding blocking writes would
require switching to async I/O which is quite an undertaking.

The fact that there's only one `struct sio_hdl *' instance in the compilation
unit sort of implies there can only be one player thread at a time. Therefore
fix the crash by only trying to acquire the mutex and continue as usual if it's
already acquired.
---

diff --git a/audio_sndio.c b/audio_sndio.c
index f8582c8a..6ada40ad 100644
--- a/audio_sndio.c
+++ b/audio_sndio.c
@@ -239,11 +239,19 @@ static int play(void *buf, int frames) {
 }
 
 static void stop() {
-  pthread_mutex_lock(&sndio_mutex);
+  int gotlock = 1;
+
+  // The player thread could already be waiting in sio_write() during
+  // termination implying that the same thread already have acquired the mutex.
+  if (pthread_mutex_trylock(&sndio_mutex))
+    gotlock = 0;
+
   if (!sio_stop(hdl))
     die("sndio: unable to stop");
   written = played = 0;
-  pthread_mutex_unlock(&sndio_mutex);
+
+  if (gotlock)
+    pthread_mutex_unlock(&sndio_mutex);
 }
 
 static void onmove_cb(__attribute__((unused)) void *arg, int delta) {