Re: [PATCH] taskstats: fix data-race

[Christian Brauner]

On Mon, Oct 07, 2019 at 12:40:39PM +0200, Andrea Parri wrote:
> Hi Christian,
> 
> On Mon, Oct 07, 2019 at 01:52:16AM +0200, Christian Brauner wrote:
> > When assiging and testing taskstats in taskstats_exit() there's a race
> > when writing and reading sig->stats when a thread-group with more than
> > one thread exits:
> > 
> > cpu0:
> > thread catches fatal signal and whole thread-group gets taken down
> >  do_exit()
> >  do_group_exit()
> >  taskstats_exit()
> >  taskstats_tgid_alloc()
> > The tasks reads sig->stats holding sighand lock seeing garbage.
> > 
> > cpu1:
> > task calls exit_group()
> >  do_exit()
> >  do_group_exit()
> >  taskstats_exit()
> >  taskstats_tgid_alloc()
> > The task takes sighand lock and assigns new stats to sig->stats.
> > 
> > Fix this by using READ_ONCE() and smp_store_release().
> > 
> > Reported-by: syzbot+c5d03165a1bd1dead0c1@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> > Signed-off-by: Christian Brauner <christian.brauner@xxxxxxxxxx>
> 
> FYI, checkpatch.pl says:
> 
> WARNING: memory barrier without comment
> #62: FILE: kernel/taskstats.c:568:
> +		smp_store_release(&sig->stats, stats_new);
> 
> Maybe you can make checkpatch.pl happy  ;-) and add a comment to stress
> the 'pairing' between this barrier and the added READ_ONCE() (as Dmitry
> was alluding to in a previous post)?

Of course. I totally forgot the memory barrier documentation requirement.

Christian