RWF_NOWAIT writes not failing when writing to a range with holes

RWF_NOWAIT writes not failing when writing to a range with holes

[Filipe Manana]

Hi,

I found out a bug in btrfs where a RWF_NOWRITE does not fail if we
write to a range that starts with an extent followed by holes (since
it requires allocating extent(s)).

When writing a test case for fstests I noticed xfs fails with -EAGAIN
as expected, but ext4 succeeds just like btrfs currently does:

mkfs.ext4 -F /dev/sdb
mount /dev/sdb /mnt

xfs_io -f -d -c "pwrite -S 0xab -b 256K 0 256K" /mnt/bar
xfs_io -c "fpunch 64K 64K" /mnt/bar
sync
xfs_io -d -c "pwrite -N -V 1 -b 128K -S 0xfe 0 128K" /mnt/bar

Is this a known bug? Or is there a technical reason that makes it too
expensive to check no extents will need to be allocated?

Thanks.

-- 
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”

Re: RWF_NOWAIT writes not failing when writing to a range with holes

[Jan Kara]

Hi!

On Mon 15-06-20 18:53:11, Filipe Manana wrote:
> I found out a bug in btrfs where a RWF_NOWRITE does not fail if we
> write to a range that starts with an extent followed by holes (since
> it requires allocating extent(s)).
> 
> When writing a test case for fstests I noticed xfs fails with -EAGAIN
> as expected, but ext4 succeeds just like btrfs currently does:
> 
> mkfs.ext4 -F /dev/sdb
> mount /dev/sdb /mnt
> 
> xfs_io -f -d -c "pwrite -S 0xab -b 256K 0 256K" /mnt/bar
> xfs_io -c "fpunch 64K 64K" /mnt/bar
> sync
> xfs_io -d -c "pwrite -N -V 1 -b 128K -S 0xfe 0 128K" /mnt/bar
> 
> Is this a known bug? Or is there a technical reason that makes it too
> expensive to check no extents will need to be allocated?

Thanks for report! This is actually a fallout of the conversion of ext4
direct IO code to iomap (commit 378f32bab37 "ext4: introduce direct I/O
write using iomap infrastructure"). I'll send a fix.

								Honza

-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: RWF_NOWAIT writes not failing when writing to a range with holes

[Filipe Manana]

On Wed, Jul 8, 2020 at 3:17 PM Jan Kara <jack@suse.cz> wrote:
>
> Hi!
>
> On Mon 15-06-20 18:53:11, Filipe Manana wrote:
> > I found out a bug in btrfs where a RWF_NOWRITE does not fail if we
> > write to a range that starts with an extent followed by holes (since
> > it requires allocating extent(s)).
> >
> > When writing a test case for fstests I noticed xfs fails with -EAGAIN
> > as expected, but ext4 succeeds just like btrfs currently does:
> >
> > mkfs.ext4 -F /dev/sdb
> > mount /dev/sdb /mnt
> >
> > xfs_io -f -d -c "pwrite -S 0xab -b 256K 0 256K" /mnt/bar
> > xfs_io -c "fpunch 64K 64K" /mnt/bar
> > sync
> > xfs_io -d -c "pwrite -N -V 1 -b 128K -S 0xfe 0 128K" /mnt/bar
> >
> > Is this a known bug? Or is there a technical reason that makes it too
> > expensive to check no extents will need to be allocated?
>
> Thanks for report! This is actually a fallout of the conversion of ext4
> direct IO code to iomap (commit 378f32bab37 "ext4: introduce direct I/O
> write using iomap infrastructure"). I'll send a fix.

Thanks for looking into it Jan.

I just wrote a test case for fstests that exercises that case and
others where btrfs used to fail.
And I think I found some regression happened in ext4 in the meanwhile.
Basically a write into a fallocated extent that starts at eof now
fails with -EAGAIN in ext4 on a 5.9-rc6 kernel at least, but it used
to work when I found the bug in btrfs [1] (with a 5.7 kernel iirc).

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4b1946284dd6641afdb9457101056d9e6ee6204c

That new test case:
https://patchwork.kernel.org/project/fstests/patch/aa8318c5beb380a9e99142d1b5e776b739d04bdb.1602774113.git.fdmanana@suse.com/

Thanks.

>
>                                                                 Honza
>
> --
> Jan Kara <jack@suse.com>
> SUSE Labs, CR



-- 
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”