Re: Possible race between cgroup_attach_proc and de_thread, and questionable code in de_thread.

[NeilBrown <neilb-l3A5Bk7waGM@public.gmane.org>]

On Sun, 14 Aug 2011 19:51:19 +0200 Oleg Nesterov <oleg-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:

> On 07/28, Paul E. McKenney wrote:
> >
> > On Thu, Jul 28, 2011 at 11:08:13AM +1000, NeilBrown wrote:
> > >
> > > I disagree.  It also requires - by virtue of the use of while_each_thread() -
> > > that 'g' remains on the list that 't' is walking along.
> >
> > Doesn't the following code in the loop body deal with this possibilty?
> >
> > 	/* Exit if t or g was unhashed during refresh. */
> > 	if (t->state == TASK_DEAD || g->state == TASK_DEAD)
> > 		goto unlock;
> 
> This code is completely wrong even if while_each_thread() was fine.
> 
> I sent the patch but it was ignored.
> 
> 	[PATCH] fix the racy check_hung_uninterruptible_tasks()->rcu_lock_break()
> 	http://marc.info/?l=linux-kernel&m=127688790019041
> 
> Oleg.


I agree with that patch.
RCU only protects a task_struct until release_task() is called (which
removes it from the task list).

So holding rcu_lock doesn't stop put_task_struct from freeing the memory
unless we *know* that release_task hasn't been called.  This is exactly that
pid_alive() tests.


I must say that handling of task_struct seems to violate the law of least
surprise a little to often for my taste.  Maybe it is just a difficult
problem and it needs a complex solution - but it would be really nice if it
were a bit simpler :-(

NeilBrown