Re: TASK_WAKEKILL && /sbin/init (was: [PATCH 1/2] schedule: fix TASK_WAKEKILL vs SIGKILL race)

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

On 06/05, Matthew Wilcox wrote:
>
> On Thu, Jun 05, 2008 at 07:23:16PM +0400, Oleg Nesterov wrote:
> > 
> > If lock_page_killable() fails because the task was killed by SIGKILL or
> > another fatal signal, do_generic_file_read() returns -EIO.
> > 
> > This seems to be OK, because in fact the userspace won't see this error, the
> > task will dequeue SIGKILL and exit.
> > 
> > However, /sbin/init is different, it will dequeue SIGKILL, ignore it, and be
> > confused by this bogus -EIO. Please note that while this bug is not likely,
> > it is _not_ theoretical. It does happen that user-space sends the unhandled
> > fatal signals to init.
> 
> Have you actually tested this?

No I didn't. And I would be happy to be wrong. But,

> I thought it was handled by:
> 
>                 /*
>                  * Global init gets no signals it doesn't want.
>                  */
>                 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
>                     !signal_group_exit(signal))
>                         continue;
> 
> in get_signal_to_deliver().

This is what I am talking about. The SIGNAL_UNKILLABLE task (init) dequeues
the pending SIGKILL and just ignores it. Then it returns to the user space
with -EIO.

But when we send SIGKILL, the sender wakes up the TASK_KILLABLE task, and
after that fatal_signal_pending() is true. Once again, it is not hard to
fix this problem in kernel/signal.c, but _perhaps_ the change in filemap.c
makes sense anyway.

Oleg.