fix pthread_cond_wait cancellation race

it's possible that signaling a waiter races with cancellation of that
same waiter. previously, cancellation was acted upon, causing the
signal to be consumed with no waiter returning. by using the new
masked cancellation state, it's possible to refuse to act on the
cancellation request and instead leave it pending.

to ease review and understanding of the changes made, this commit
leaves the unwait function, which was previously the cancellation
cleanup handler, in place. additional simplifications could be made by
removing it.

src/thread/pthread_cond_timedwait.c

@@ -32,7 +32,7 @@ struct waiter {
 	int *notify;
 	pthread_mutex_t *mutex;
 	pthread_cond_t *cond;
-	int shared;
+	int shared, err;
 };
 
 /* Self-synchronized-destruction-safe lock functions */
@@ -73,6 +73,11 @@ static void unwait(void *arg)
 	if (node->shared) {
 		pthread_cond_t *c = node->cond;
 		pthread_mutex_t *m = node->mutex;
+		/* Suppress cancellation if a signal was potentially
+		 * consumed; this is a legitimate form of spurious
+		 * wake even if not. */
+		if (node->err == ECANCELED && c->_c_seq != node->state)
+			node->err = 0;
 		if (a_fetch_add(&c->_c_waiters, -1) == -0x7fffffff)
 			__wake(&c->_c_waiters, 1, 0);
 		node->mutex_ret = pthread_mutex_lock(m);
@@ -121,12 +126,23 @@ static void unwait(void *arg)
 	} else {
 		a_dec(&node->mutex->_m_waiters);
 	}
+
+	/* Since a signal was consumed, acting on cancellation is not
+	 * permitted. The only other error possible at this stage,
+	 * ETIMEDOUT, is permitted even if a signal was consumed. */
+	if (node->err = ECANCELED) node->err = 0;
 }
 
+static void dummy(void *arg)
+{
+}
+
+int __pthread_setcancelstate(int, int *);
+
 int __pthread_cond_timedwait(pthread_cond_t *restrict c, pthread_mutex_t *restrict m, const struct timespec *restrict ts)
 {
 	struct waiter node = { .cond = c, .mutex = m };
-	int e, seq, *fut, clock = c->_c_clock;
+	int e, seq, *fut, clock = c->_c_clock, cs;
 
 	if ((m->_m_type&15) && (m->_m_lock&INT_MAX) != __pthread_self()->tid)
 		return EPERM;
@@ -139,7 +155,7 @@ int __pthread_cond_timedwait(pthread_cond_t *restrict c, pthread_mutex_t *restri
 	if (c->_c_shared) {
 		node.shared = 1;
 		fut = &c->_c_seq;
-		seq = c->_c_seq;
+		seq = node.state = c->_c_seq;
 		a_inc(&c->_c_waiters);
 	} else {
 		lock(&c->_c_lock);
@@ -157,13 +173,30 @@ int __pthread_cond_timedwait(pthread_cond_t *restrict c, pthread_mutex_t *restri
 
 	__pthread_mutex_unlock(m);
 
-	do e = __timedwait(fut, seq, clock, ts, unwait, &node, !node.shared);
+	__pthread_setcancelstate(PTHREAD_CANCEL_MASKED, &cs);
+
+	do e = __timedwait(fut, seq, clock, ts, dummy, 0, !node.shared);
 	while (*fut==seq && (!e || e==EINTR));
 	if (e == EINTR) e = 0;
 
+	node.err = e;
 	unwait(&node);
+	e = node.err;
+
+	/* Suppress cancellation if there was an error locking the mutex,
+	 * since the contract for cancellation requires the mutex to be
+	 * locked when the cleanup handler is called, and there is no
+	 * way to report an error. */
+	if (node.mutex_ret) e = node.mutex_ret;
+
+	__pthread_setcancelstate(cs, 0);
+
+	if (e == ECANCELED) {
+		__pthread_testcancel();
+		__pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, 0);
+	}
 
-	return node.mutex_ret ? node.mutex_ret : e;
+	return e;
 }
 
 int __private_cond_signal(pthread_cond_t *c, int n)