Re: [PATCH] iomap: Make sure iomap_end is called after iomap_begin

[Matthew Wilcox <willy@infradead.org>]

On Tue, Jun 16, 2020 at 09:32:39AM +1000, Dave Chinner wrote:
> On Mon, Jun 15, 2020 at 06:02:44PM +0200, Andreas Gruenbacher wrote:
> > Make sure iomap_end is always called when iomap_begin succeeds: the
> > filesystem may take locks in iomap_begin and release them in iomap_end,
> > for example.
> 
> Ok, i get that from the patch, but I don't know anything else about
> this problem, and nor will anyone else trying to determine if this
> is a fix they need to backport to other kernels. Can you add some
> more information to the commit message, such as how was this found
> and what filesystems it affects? It would also be good to know what
> commit introduced this issue and whether it need stable back ports
> (i.e. a Fixes tag).

I'd assume Andreas is looking at converting a filesystem to use iomap,
since this problem only occurs for filesystems which have returned an
invalid extent.

I almost wonder if this should return -EFSCORRUPTED rather than -EIO.
Um, except that's currently a per-fs define.  Is it time to move that
up to errno.h?