Re: [PATCH v4 1/1] iomap: fix invalid folio access after folio_end_read()

[Matthew Wilcox <willy@infradead.org>]

On Sun, Jan 25, 2026 at 09:46:30PM -0800, Christoph Hellwig wrote:
> > -	if (ctx->ops->submit_read)
> > -		ctx->ops->submit_read(ctx);
> > -
> > -	iomap_read_end(folio, bytes_submitted);
> > +	iomap_read_submit_and_end(ctx, bytes_submitted);
> 
> Can you drop this cleanup for now?  I think it's actually useful,
> but it should be in a separate patch, and creates a conflict with
> my iomap PI series.
> 
> The actual fix looks great and simplified the code nicely!

I don't think it's just a cleanup -- I think it's a bug fix.  But, yes,
it should be a separate patch because it's a separate bug.  That bug
can be hit if the folio passed to iomap_read_folio() covers more than
one extent, the first call to iomap_iter() succeeds, and then the second
one fails.  Now we have a folio with a positive read_pending that will
never become zero, so we'll never unlock the folio.

Not sure how we'd write an fstest that would exercise this ...