Re: [PATCH] btrfs: add test for cases when a dio write has to fallback to a buffered write

[Eryu Guan <guan@eryu.me>]

On Wed, Mar 10, 2021 at 10:48:35AM +0000, Filipe Manana wrote:
> On Sun, Mar 7, 2021 at 3:24 PM Eryu Guan <guan@eryu.me> wrote:
> >
> > On Sun, Mar 07, 2021 at 03:07:43PM +0000, Filipe Manana wrote:
> > > On Sun, Mar 7, 2021 at 2:41 PM Eryu Guan <guan@eryu.me> wrote:
> > > >
> > > > On Thu, Feb 11, 2021 at 05:01:18PM +0000, fdmanana@kernel.org wrote:
> > > > > From: Filipe Manana <fdmanana@suse.com>
> > > > >
> > > > > Test cases where a direct IO write, with O_DSYNC, can not be done and has
> > > > > to fallback to a buffered write.
> > > > >
> > > > > This is motivated by a regression that was introduced in kernel 5.10 by
> > > > > commit 0eb79294dbe328 ("btrfs: dio iomap DSYNC workaround")) and was
> > > > > fixed in kernel 5.11 by commit ecfdc08b8cc65d ("btrfs: remove dio iomap
> > > > > DSYNC workaround").
> > > > >
> > > > > Signed-off-by: Filipe Manana <fdmanana@suse.com>
> > > >
> > > > Sorry for the late review..
> > > >
> > > > So this is supposed to fail with v5.10 kernel, right? But I got it
> > > > passed
> > >
> > > Because either you are testing with a patched 5.10.x kernel, or you
> > > don't have CONFIG_BTRFS_ASSERT=y in your config.
> > > The fix landed in 5.10.18:
> >
> > You're right, I don't have CONFIG_BTRFS_ASSERT=y. As the test dumps the
> > od output of the file content, so I thought the failure would be a data
> > corruption, and expected a od output diff failure.
> 
> I see the test was not merged yet, do you expect me to update anything
> in the patch?

Sorry, it was lost in my to-review queue.. But I'd be great if you could
add more descriptions about the CONFIG_BTRFS_ASSERT and the compress
case either in commit log or in test description.

Thanks,
Eryu