fix pthread_detach inadvertently acting as cancellation point in race case

disabling cancellation around the pthread_join call seems to be the
safest and logically simplest fix. i believe it would also be possible
to just perform the unmap directly here after __tl_sync, removing the
dependency on pthread_join, but such an approach duplicately encodes a
lot more implementation assumptions.

1 files changed, 6 insertions, 2 deletions

diff --git a/src/thread/pthread_detach.c b/src/thread/pthread_detach.c
index 77772af2..d73a500e 100644
--- a/src/thread/pthread_detach.c
+++ b/src/thread/pthread_detach.c
@@ -5,8 +5,12 @@ static int __pthread_detach(pthread_t t)
 {
 	/* If the cas fails, detach state is either already-detached
 	 * or exiting/exited, and pthread_join will trap or cleanup. */
-	if (a_cas(&t->detach_state, DT_JOINABLE, DT_DETACHED) != DT_JOINABLE)
-		return __pthread_join(t, 0);
+	if (a_cas(&t->detach_state, DT_JOINABLE, DT_DETACHED) != DT_JOINABLE) {
+		int cs;
+		__pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cs);
+		__pthread_join(t, 0);
+		__pthread_setcancelstate(cs, 0);
+	}
 	return 0;
 }