On 09/02/11 11:02 -0800, Sukadev Bhattiprolu wrote:
> Louis Rilling [Louis.Rilling-aw0BnHfMbSpBDgjK7y7TUQ@public.gmane.org] wrote:
> | > | Are we still getting it with 2.6.37 ?
> | > 
> | > I am not currently getting the crash on 2.6.37 - I thought it was due to
> | > the following commit which removed the check for task_detached() in
> | > do_wait_thread().
> | > 
> | > 	commit 9cd80bbb07fcd6d4d037fad4297496d3b132ac6b
> | > 	Author: Oleg Nesterov <oleg-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> | > 	Date:   Thu Dec 17 15:27:15 2009 -0800
> | 
> | I don't think that this introduced the bug. The bug triggers with EXIT_DEAD
> | tasks, for which wait() must ignore (see below). So, the bug looks still there
> | in 2.6.37.
> 
> Sorry, I did not mean to imply that the above commit caused the crash
> you saw in Jun 2010.
> 
> I can reproduce a crash with 2.6.32 - where if container-init terminates
> before a detached child, we get a crash when the detached child calls
> proc_flush_mnt(). I suspected it was because do_wait_thread() skipped
> over detached tasks (in 2.6.32).
> 
> The same test case does not crash on 2.6.37 - which includes the above commit.
> The removes the check for detached tasks, my initial guess is that the above
> commit, may have contributed to _fixing_ the crash in 2.6.37.

Hm, I don't see how this commit changed things for detached tasks, unless ptrace
is involved. Detached tasks go atomically
from ->exit_state == 0 to ->exit_state == EXIT_DEAD in exit_notify(),
because tracehook_notify_death() returns DEATH_REAP for all not ptraced detached
tasks.

What do you think has changed precisely?

Thanks,

Louis

-- 
Dr Louis Rilling			Kerlabs
Skype: louis.rilling			Batiment Germanium
Phone: (+33|0) 6 80 89 08 23		80 avenue des Buttes de Coesmes
http://www.kerlabs.com/			35700 Rennes