Don't hold mutex until release cv in cv_wait

If a thread is holding mutex when doing cv_destroy, it might end up waiting a
thread in cv_wait. The waiter would wake up trying to aquire the same mutex
and cause deadlock.

We solve this by move the mutex_enter to the bottom of cv_wait, so that
the waiter will release the cv first, allowing cv_destroy to succeed and have
a chance to free the mutex.

This would create race condition on the cv_mutex. We use xchg to set and check
it to ensure we won't be harmed by the race. This would result in the cv_mutex
debugging becomes best-effort.

Also, the change reveals a race, which was unlikely before, where we call
mutex_destroy while test threads are still holding the mutex. We use
kthread_stop to make sure the threads are exit before mutex_destroy.

Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Issue zfsonlinux/zfs#4166
Issue zfsonlinux/zfs#4106

module/spl/spl-condvar.c

@@ -80,6 +80,7 @@ static void
 cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io)
 {
 	DEFINE_WAIT(wait);
+	kmutex_t *m;
 
 	ASSERT(cvp);
 	ASSERT(mp);
@@ -87,11 +88,11 @@ cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io)
 	ASSERT(mutex_owned(mp));
 	atomic_inc(&cvp->cv_refs);
 
-	if (cvp->cv_mutex == NULL)
+	m = ACCESS_ONCE(cvp->cv_mutex);
-		cvp->cv_mutex = mp;
+	if (!m)
-
+		m = xchg(&cvp->cv_mutex, mp);
 	/* Ensure the same mutex is used by all callers */
-	ASSERT(cvp->cv_mutex == mp);
+	ASSERT(m == NULL || m == mp);
 
 	prepare_to_wait_exclusive(&cvp->cv_event, &wait, state);
 	atomic_inc(&cvp->cv_waiters);
@@ -106,16 +107,25 @@ cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io)
 		io_schedule();
 	else
 		schedule();
-	mutex_enter(mp);
 
 	/* No more waiters a different mutex could be used */
 	if (atomic_dec_and_test(&cvp->cv_waiters)) {
+		/*
+		 * This is set without any lock, so it's racy. But this is
+		 * just for debug anyway, so make it best-effort
+		 */
 		cvp->cv_mutex = NULL;
 		wake_up(&cvp->cv_destroy);
 	}
 
 	finish_wait(&cvp->cv_event, &wait);
 	atomic_dec(&cvp->cv_refs);
+
+	/*
+	 * Hold mutex after we release the cvp, otherwise we could dead lock
+	 * with a thread holding the mutex and call cv_destroy.
+	 */
+	mutex_enter(mp);
 }
 
 void
@@ -148,6 +158,7 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
     int state)
 {
 	DEFINE_WAIT(wait);
+	kmutex_t *m;
 	clock_t time_left;
 
 	ASSERT(cvp);
@@ -156,15 +167,16 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
 	ASSERT(mutex_owned(mp));
 	atomic_inc(&cvp->cv_refs);
 
-	if (cvp->cv_mutex == NULL)
+	m = ACCESS_ONCE(cvp->cv_mutex);
-		cvp->cv_mutex = mp;
+	if (!m)
-
+		m = xchg(&cvp->cv_mutex, mp);
 	/* Ensure the same mutex is used by all callers */
-	ASSERT(cvp->cv_mutex == mp);
+	ASSERT(m == NULL || m == mp);
 
 	/* XXX - Does not handle jiffie wrap properly */
 	time_left = expire_time - jiffies;
 	if (time_left <= 0) {
+		/* XXX - doesn't reset cv_mutex */
 		atomic_dec(&cvp->cv_refs);
 		return (-1);
 	}
@@ -179,10 +191,13 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
 	 */
 	mutex_exit(mp);
 	time_left = schedule_timeout(time_left);
-	mutex_enter(mp);
 
 	/* No more waiters a different mutex could be used */
 	if (atomic_dec_and_test(&cvp->cv_waiters)) {
+		/*
+		 * This is set without any lock, so it's racy. But this is
+		 * just for debug anyway, so make it best-effort
+		 */
 		cvp->cv_mutex = NULL;
 		wake_up(&cvp->cv_destroy);
 	}
@@ -190,6 +205,11 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
 	finish_wait(&cvp->cv_event, &wait);
 	atomic_dec(&cvp->cv_refs);
 
+	/*
+	 * Hold mutex after we release the cvp, otherwise we could dead lock
+	 * with a thread holding the mutex and call cv_destroy.
+	 */
+	mutex_enter(mp);
 	return (time_left > 0 ? time_left : -1);
 }
 
@@ -216,6 +236,7 @@ __cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t expire_time,
     int state)
 {
 	DEFINE_WAIT(wait);
+	kmutex_t *m;
 	hrtime_t time_left, now;
 	unsigned long time_left_us;
 
@@ -225,11 +246,11 @@ __cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t expire_time,
 	ASSERT(mutex_owned(mp));
 	atomic_inc(&cvp->cv_refs);
 
-	if (cvp->cv_mutex == NULL)
+	m = ACCESS_ONCE(cvp->cv_mutex);
-		cvp->cv_mutex = mp;
+	if (!m)
-
+		m = xchg(&cvp->cv_mutex, mp);
 	/* Ensure the same mutex is used by all callers */
-	ASSERT(cvp->cv_mutex == mp);
+	ASSERT(m == NULL || m == mp);
 
 	now = gethrtime();
 	time_left = expire_time - now;
@@ -253,10 +274,13 @@ __cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t expire_time,
 	 * interrupts
 	 */
 	usleep_range(time_left_us, time_left_us + 100);
-	mutex_enter(mp);
 
 	/* No more waiters a different mutex could be used */
 	if (atomic_dec_and_test(&cvp->cv_waiters)) {
+		/*
+		 * This is set without any lock, so it's racy. But this is
+		 * just for debug anyway, so make it best-effort
+		 */
 		cvp->cv_mutex = NULL;
 		wake_up(&cvp->cv_destroy);
 	}
@@ -264,6 +288,7 @@ __cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t expire_time,
 	finish_wait(&cvp->cv_event, &wait);
 	atomic_dec(&cvp->cv_refs);
 
+	mutex_enter(mp);
 	time_left = expire_time - gethrtime();
 	return (time_left > 0 ? time_left : -1);
 }

module/splat/splat-condvar.c

@@ -88,6 +88,9 @@ splat_condvar_test12_thread(void *arg)
 	    ct->ct_thread->comm, atomic_read(&cv->cv_condvar.cv_waiters));
 	mutex_exit(&cv->cv_mtx);
 
+	/* wait for main thread reap us */
+	while (!kthread_should_stop())
+		schedule();
 	return 0;
 }
 
@@ -151,6 +154,12 @@ splat_condvar_test1(struct file *file, void *arg)
 	/* Wake everything for the failure case */
 	cv_broadcast(&cv.cv_condvar);
 	cv_destroy(&cv.cv_condvar);
+
+	/* wait for threads to exit */
+	for (i = 0; i < SPLAT_CONDVAR_TEST_COUNT; i++) {
+		if (!IS_ERR(ct[i].ct_thread))
+			kthread_stop(ct[i].ct_thread);
+	}
 	mutex_destroy(&cv.cv_mtx);
 
 	return rc;
@@ -199,6 +208,12 @@ splat_condvar_test2(struct file *file, void *arg)
 
 	/* Wake everything for the failure case */
 	cv_destroy(&cv.cv_condvar);
+
+	/* wait for threads to exit */
+	for (i = 0; i < SPLAT_CONDVAR_TEST_COUNT; i++) {
+		if (!IS_ERR(ct[i].ct_thread))
+			kthread_stop(ct[i].ct_thread);
+	}
 	mutex_destroy(&cv.cv_mtx);
 
 	return rc;
@@ -234,6 +249,9 @@ splat_condvar_test34_thread(void *arg)
 
 	mutex_exit(&cv->cv_mtx);
 
+	/* wait for main thread reap us */
+	while (!kthread_should_stop())
+		schedule();
 	return 0;
 }
 
@@ -302,6 +320,12 @@ splat_condvar_test3(struct file *file, void *arg)
 	/* Wake everything for the failure case */
 	cv_broadcast(&cv.cv_condvar);
 	cv_destroy(&cv.cv_condvar);
+
+	/* wait for threads to exit */
+	for (i = 0; i < SPLAT_CONDVAR_TEST_COUNT; i++) {
+		if (!IS_ERR(ct[i].ct_thread))
+			kthread_stop(ct[i].ct_thread);
+	}
 	mutex_destroy(&cv.cv_mtx);
 
 	return rc;
@@ -372,6 +396,12 @@ splat_condvar_test4(struct file *file, void *arg)
 	/* Wake everything for the failure case */
 	cv_broadcast(&cv.cv_condvar);
 	cv_destroy(&cv.cv_condvar);
+
+	/* wait for threads to exit */
+	for (i = 0; i < SPLAT_CONDVAR_TEST_COUNT; i++) {
+		if (!IS_ERR(ct[i].ct_thread))
+			kthread_stop(ct[i].ct_thread);
+	}
 	mutex_destroy(&cv.cv_mtx);
 
 	return rc;