Re: [PATCH] workqueue: Fix UAF report by KASAN in pwq_release_workfn()

[Z qiang]

>
> Hello,
>
> On Sat, Sep 02, 2023 at 07:50:26PM +0800, Zqiang wrote:
> > diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> > index c85825e17df8..f3f9c9222070 100644
> > --- a/kernel/workqueue.c
> > +++ b/kernel/workqueue.c
> > @@ -4609,6 +4609,10 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq)
> >               free_percpu(wq->cpu_pwq);
> >               wq->cpu_pwq = NULL;
> >       }
> > +     wq_unregister_lockdep(wq);
> > +     wq_free_lockdep(wq);
> > +     free_workqueue_attrs(wq->unbound_attrs);
> > +     kfree(wq);
>
> So, this would fix the bug but it's quite confusing because we end up taking
> two separate error handling paths and alloc_and_link_pwqs() ends up freeing
> stuff allocated outside the function. Wouldn't it be enough to flush the
> pwq_release_worker after apply_workqueue_attrs() failure so that the cleanup
> operation ordering is maintained?

Hello, Tejun

Flush the pwq_release_worker is insufficient, the call_rcu() is
invoked to release wq
in pwq_release_workfn(), this is also asynchronous.


Thanks
Zqiang

>
> Thanks.
>
> --
> tejun