Re: [RFC][PATCH] ->signal->tty locking

[Oleg Nesterov <oleg@tv-sign.ru>]

On 10/17, Peter Zijlstra wrote:
>
> On Tue, 2006-10-17 at 12:10 +0400, Oleg Nesterov wrote:
> > 
> > We don't need lock_task_sighand() here, we can use spin_lock_irq(->siglock).
> > 
> > We are holding tasklist_lock. This means that all tasks found by
> > do_each_task_pid() have a valid ->signal/->sighand != NULL.
> > tasklist_lock protects against release_task()->__exit_signal() and
> > from changing ->sighand by de_thread().
> 
> I think sys_unshare() spoils the game here; it changes ->sighand in
> midair without holding tasklist_lock. So any ->sighand but current's is
> fair game.
> 
> Hmm, either sys_unshare() is broken in that it doesn't take the
> tasklist_lock or a lot of other code is broken.

Yes, it is broken, please look at
	http://marc.theaimsgroup.com/?t=114253118100003

I sent a patch,
	http://marc.theaimsgroup.com/?l=linux-kernel&m=114268787415193

but it was ignored. Probably I should re-send it.

> Right, use tty_mutex when using the tty, use ->sighand when changing
> signal->tty.

I think that things like do_task_stat()/do_acct_process() do not need
global tty_mutex, they can use ->siglock.

Oleg.