[PATCH] Fix setxid race handing exiting threads

[Andreas Schwab <schwab at redhat dot com>]

If two threads are calling setxid simultanously a race exists between
releasing setxid_futex in sighandler_setxid and locking it again in
setxid_mark_thread when the thread exited in the mean time.  Make sure
the exiting thread is eventually woken up.

Andreas.

2010-06-28  Andreas Schwab  <schwab@redhat.com>

	* allocatestack.c (setxid_mark_thread): Ensure that the exiting
	thread is woken up.

diff --git a/nptl/allocatestack.c b/nptl/allocatestack.c
index 831e98e..1ce9ed5 100644
--- a/nptl/allocatestack.c
+++ b/nptl/allocatestack.c
@@ -994,7 +994,16 @@ setxid_mark_thread (struct xid_command *cmdp, struct pthread *t)
 
       /* If the thread is exiting right now, ignore it.  */
       if ((ch & EXITING_BITMASK) != 0)
-	return;
+	{
+	  /* Release the futex if there is no other setxid in
+	     progress.  */
+	  if ((ch & SETXID_BITMASK) == 0)
+	    {
+	      t->setxid_futex = 1;
+	      lll_futex_wake (&t->setxid_futex, 1, LLL_PRIVATE);
+	    }
+	  return;
+	}
     }
   while (atomic_compare_and_exchange_bool_acq (&t->cancelhandling,
 					       ch | SETXID_BITMASK, ch));
-- 
1.7.1


-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."