Re: Q: check_unsafe_exec() races (Was: [PATCH 2/4] fix setuid sometimes doesn't)

[Oleg Nesterov <oleg@redhat.com>]

On 03/29, Al Viro wrote:
>
> On Sun, Mar 29, 2009 at 11:36:35PM +0200, Oleg Nesterov wrote:
> > > ... or just do that to fs_struct.  After finding that there's no outside
> > > users.  Commenst?
> >
> > This is even worse. Not only we race with our sub-threads, we race
> > with CLONE_FS processes.
> >
> > We can't mark fs_struct after finding that there's no outside users
> > lockless. Because we can't know whether this is "after" or not, we
> > can't trust "atomic_read(fs->count) <= n_fs".
>
> We can lock fs_struct in question, go through the threads, then mark
> or bail out.  With cloning a reference to fs_struct protected by the
> same lock.

Yes, this is what I meant, copy_fs() needs this lock too,

> FWIW, I'm not at all sure that we want atomic_t for refcount in that
> case...

I think you are right, because exit_fs() should take fs->lock as well.

But, again. What whould we do when check_unsafe_exec() takes fs->lock
and sees that this ->fs is already marked?

In that case -EWHATEVER is not very good, it could be another process
(not sub-thread) doing exec.

Oleg.