[dm-devel] block: fix a crash when bio_for_each_folio_all iterates over an empty bio

[Mike Snitzer]

On Tue, Apr 18 2023 at 11:20P -0400,
Mike Snitzer <snitzer at kernel.org> wrote:

> On Mon, Apr 17 2023 at  3:11P -0400,
> Mikulas Patocka <mpatocka at redhat.com> wrote:
> 
> > If we use bio_for_each_folio_all on an empty bio, it will access the first
> > bio vector unconditionally (it is uninitialized) and it may crash
> > depending on the uninitialized data.
> > 
> > This patch fixes it by checking the parameter "i" against "bio->bi_vcnt"
> > and returning NULL fi->folio if it is out of range.
> > 
> > The patch also drops the test "if (fi->_i + 1 < bio->bi_vcnt)" from
> > bio_next_folio because the same condition is already being checked in
> > bio_first_folio.
> >
> > Signed-off-by: Mikulas Patocka <mpatocka at redhat.com>
> 
> This fix is a prereq for this dm-crypt patch to use folios:
> https://patchwork.kernel.org/project/dm-devel/patch/alpine.LRH.2.21.2302161619430.5436@file01.intranet.prod.int.rdu2.redhat.com/
> Mikulas explained why an empty bio is possible here:
> https://listman.redhat.com/archives/dm-devel/2023-April/053916.html
> 
> Reviewed-by: Mike Snitzer <snitzer at kernel.org>

Hey Jens (and Matthew),

Can you please pick this up?

Without it DM needs to do the checking (by open-coding a fixed variant
of bio_for_each_folio_all); while we _could_ do that: fixing
bio_first_folio seems best.

Thanks,
Mike