Re: [PATCH 1/1] Don't set mddev private to NULL in raid0 pers->free

[Song Liu <song@kernel.org>]

On Wed, May 11, 2022 at 9:14 AM Xiao Ni <xni@redhat.com> wrote:
>
> On Mon, May 9, 2022 at 2:24 PM Song Liu <song@kernel.org> wrote:
> >
> > Hi Xiao,
> >
> > Thanks for the patch!
> >
> > On Thu, Apr 28, 2022 at 10:09 PM Xiao Ni <xni@redhat.com> wrote:
> > >
> >
> > prefix the subject with "md:", and provide more details.
>
> Hi Song
>
> Thanks for pointing about this and I will add "md:" in v2.
> >
> > > It panics when reshaping from raid0 to other raid levels. raid0 sets
> > > mddev->private to NULL. It's the reason that causes the problem.
> > > Function level_store finds new pers and create new conf, then it
> > > calls oldpers->free. In oldpers->free, raid0 sets mddev->private
> > > to NULL again. And __md_stop is the right position to set
> > > mddev->private to NULL.
> >
> > We need more details here: the panic backtrace, and why it panics.
> > raid5_free also sets private to NULL. Does it has the same problem?
>
> The backtrace will be added in v2.
>
> I don't find raid5_free sets mddev->private to NULL.
> The normal process of stopping a raid should be like this:
>
> do_md_stop
>        |
> __md_stop  (pers->free(); mddev->private=NULL)
>        |
> md_free (free biosets of mddev; free mddev)
>
> personality raid doesn't set mddev->private to NULL. __md_stop does this job.
> Reshaping from raid0 to raid other raid level doesn't stop the raid.
> The new raid
> will go on working. Now raid0_free->free_conf sets mddev->private to NULL during
> reshape. The new raid can't work anymore. It will panic when dereference
> mddev->private because of NULL pointer dereference.

Got it. Thanks for the explanation.

Thanks,
Song