Re: removing the remaining blockdev_direct_IO users

[Viacheslav Dubeyko <slava@dubeyko.com>]

Hi Christoph,

On Mon, 2026-07-13 at 15:33 -0700, Viacheslav Dubeyko wrote:
> On Mon, 2026-07-13 at 11:58 +0200, Christoph Hellwig wrote:
> > On Fri, Jul 10, 2026 at 09:51:09AM -0700, Viacheslav Dubeyko wrote:
> > > On Fri, 2026-07-10 at 07:07 +0200, Christoph Hellwig wrote:
> > > > Hi all,
> > > > 
> > > > we have 7 users of blockdev_direct_IO and one of the lower-
> > > > level
> > > > __blockdev_direct_IO left, and I'd really like to get rid of
> > > > them.
> > > > 
> > > > The filesystems are affs, fat, hfs, hfsplus, jfs, nilfs2, udf
> > > > and
> > > > ocfs2,
> > > > and you are the relevant maintainers.  In 7.2 exfat converted
> > > > entirely
> > > > to iomap, which would be a really good example for a "simple"
> > > > file
> > > > system.  Another options especially for the file systems
> > > > supporting
> > > > foreign formats would be to just drop direct I/O support and
> > > > reduce
> > > > the support burden.
> > > 
> > > I like the idea. I think we can try to switch HFS/HFS+ and NILFS2
> > > on
> > > iomap approach. Do you have some initial patches for this? Which
> > > particular commit(s) converted the exfat to iomap? It will be
> > > good
> > > to
> > > take a deeper look into this conversion to estimate the
> > > complexity
> > > of
> > > HFS/HFS+, NILFS2 modifications.
> > 
> > Almost all commits in fs/exfat/ in 7.2-rc1, starting with
> > 623f0aa1eca5c2a94ca1e4e5de719d062eac3b6c
> 
> OK. Thanks for sharing this. :) Let me take a look and spend some
> time
> on this. I should have a time for checking commits during this week.
> I
> hope so. :)
> 

I have a very initial state of patchset for HFS+ file system. I am
testing it by xfstests now. As far as I can see, currently, this
patchset introduced multiple xfstests failures. If you like I can share
this patchset for discussion and review. Otherwise, let me fix the
found xfstests failures at first and, then, I'll share the patchset for
the review.

Thanks,
Slava.