Re: [PATCH] iomap: fix memory corruption when recording errors during writeback

["Darrick J. Wong" <djwong@kernel.org>]

On Thu, Sep 29, 2022 at 08:33:06PM +0100, Matthew Wilcox wrote:
> On Thu, Sep 29, 2022 at 11:44:49AM -0700, Darrick J. Wong wrote:
> > Fixes: e735c0079465 ("iomap: Convert iomap_add_to_ioend() to take a folio")
> > Probably-Fixes: 598ecfbaa742 ("iomap: lift the xfs writeback code to iomap")
> 
> I think this is a misuse of Fixes.  As I understand it, Fixes: is "Here's
> the commit that introduced the bug", not "This is the most recent change
> after which this patch will still apply".  e735c0079465 only changed
> s/page/folio/ in this line of code, so clearly didn't introduce the bug.
> 
> Any kernel containing 598ecfbaa742 has this same bug, so that should be
> the Fixes: line.  As you say though, 598ecfbaa742 merely moved the code
> from xfs_writepage_map().  bfce7d2e2d5e moved it from xfs_do_writepage(),
> but 150d5be09ce4 introduced it.  Six years ago!  Good find.  So how about:
> 
> Fixes: 150d5be09ce4 ("xfs: remove xfs_cancel_ioend")

Sounds fine to me, though if I hear complaints from AUTOSEL about how
the patch does not directly apply to old kernels, I'll forward them to
you, because that's what I do now to avoid getting even /more/ email.

> Also,
> 
> Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

However, thank you for the quick review. :)

--D

> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> >  fs/iomap/buffered-io.c |    2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> > index ca5c62901541..77d59c159248 100644
> > --- a/fs/iomap/buffered-io.c
> > +++ b/fs/iomap/buffered-io.c
> > @@ -1421,7 +1421,7 @@ iomap_writepage_map(struct iomap_writepage_ctx *wpc,
> >  	if (!count)
> >  		folio_end_writeback(folio);
> >  done:
> > -	mapping_set_error(folio->mapping, error);
> > +	mapping_set_error(inode->i_mapping, error);
> >  	return error;
> >  }
> >