View | Details | Raw Unified | Return to bug 91434
Collapse All | Expand All

(-)file_not_specified_in_diff (-2 / +13 lines)
Line  Link Here 
 [PATCH] reset real_timer target on exec leader change
 
 [PATCH] reset real_timer target on exec leader change
1
1
2 
 When a noninitial thread does exec, it becomes the new group leader.  If
 2 
 When a noninitial thread does exec, it becomes the new group leader.  If
3 
 there is a ITIMER_REAL timer running, it points at the old group leader and
 3 
 there is a ITIMER_REAL timer running, it points at the old group leader and
4 
 when it fires it can follow a stale pointer.  The timer data needs to be
 4 
 when it fires it can follow a stale pointer.  The timer data needs to be
5 
 reset to point at the exec'ing thread that is becoming the group leader.
 5 
 reset to point at the exec'ing thread that is becoming the group leader.
6 
 This has to synchronize with any concurrent firing of the timer to make
 6 
 This has to synchronize with any concurrent firing of the timer to make
7 
 sure that it_real_fn can never run when the data points to a thread that
 7 
 sure that it_real_fn can never run when the data points to a thread that
8 
 might have been reaped already.
 8 
 might have been reaped already.
9
9
10 
 Signed-off-by: Roland McGrath <roland@redhat.com>
 10 
 Signed-off-by: Roland McGrath <roland@redhat.com>
11 
 Signed-off-by: Andrew Morton <akpm@osdl.org>
 11 
 Signed-off-by: Andrew Morton <akpm@osdl.org>
12 
 Signed-off-by: Linus Torvalds <torvalds@osdl.org>
 12 
 Signed-off-by: Linus Torvalds <torvalds@osdl.org>
13 
--
 13 
++ b/fs/exec.c
14 
-- a/fs/exec.c
Lines 642-647 static inline int de_thread(struct task_ Link Here
642 
	count = 2;
 642 
	count = 2;
643 
	if (thread_group_leader(current))
 643 
	if (thread_group_leader(current))
644 
		count = 1;
 644 
		count = 1;
645 
	else {
646 
		/*
647 
		 * The SIGALRM timer survives the exec, but needs to point
648 
		 * at us as the new group leader now.  We have a race with
649 
		 * a timer firing now getting the old leader, so we need to
650 
		 * synchronize with any firing (by calling del_timer_sync)
651 
		 * before we can safely let the old group leader die.
652 
		 */
653 
		sig->real_timer.data = (unsigned long)current;
654 
		if (del_timer_sync(&sig->real_timer))
655 
			add_timer(&sig->real_timer);
656 
	}
645 
	while (atomic_read(&sig->count) > count) {
 657 
	while (atomic_read(&sig->count) > count) {
646 
		sig->group_exit_task = current;
 658 
		sig->group_exit_task = current;
647 
		sig->notify_count = count;
 659 
		sig->notify_count = count;

Return to bug 91434