Re: [PATCH 0/3 v2] xfs: Fix SEEK_HOLE implementation

[Jan Kara <jack@suse.cz>]

On Thu 18-05-17 17:47:53, Eryu Guan wrote:
> On Thu, May 18, 2017 at 11:03:46AM +0200, Jan Kara wrote:
> > On Wed 17-05-17 16:57:46, Jan Kara wrote:
> > > On Wed 17-05-17 20:31:15, Eryu Guan wrote:
> > > > Hi Jan,
> > > > 
> > > > On Wed, May 17, 2017 at 02:10:43PM +0200, Jan Kara wrote:
> > > > > Hello,
> > > > > 
> > > > > this is the second revision of the patches to fix bugs in XFS's SEEK_HOLE
> > > > > implementation and cleanup the code a bit.
> > > > > 
> > > > > Changes since v1:
> > > > > * Fixed some more buggy cases
> > > > > * Simplified code a bit as suggested by Darrick
> > > > > * Fixed range check as spotted by Brian
> > > > 
> > > > I applied this patchset on top of 4.12-rc1 kernel to test your v4 test
> > > > case, your new test passed all my tests, but I found generic/285
> > > > regressed with sub-page block size XFS, 285.full showed that failure was
> > > > from subtest 7
> > > > 
> > > > 07. Test file with unwritten extents, only have dirty pages
> > > > 07.01 SEEK_HOLE expected 0 or 11264, got 0.                       succ
> > > > 07.02 SEEK_HOLE expected 1 or 11264, got 1.                       succ
> > > > 07.03 SEEK_DATA expected 10240 or 10240, got -1.                  FAIL
> > > > 07.04 SEEK_DATA expected 10240 or 10240, got -1.                  FAIL
> > > > 
> > > > And manual test showed subtest 8 failed too
> > > > 
> > > > # ./src/seek_sanity_test -s 8 -e 8 /mnt/xfs/testfile
> > > > File system magic#: 0x58465342
> > > > Allocation size: 4096
> > > > 
> > > > 08. Test file with unwritten extents, only have unwritten pages
> > > > 08.01 SEEK_HOLE expected 0 or 5632, got 0.                        succ
> > > > 08.02 SEEK_HOLE expected 1 or 5632, got 1.                        succ
> > > > 08.03 SEEK_DATA expected 5120 or 5120, got -1.                    FAIL
> > > > 08.04 SEEK_DATA expected 5120 or 5120, got -1.                    FAIL
> > > > 
> > > > Other subtests all passed with sub-page block size XFS.
> > > 
> > > Strange. It doesn't fail for me this way even with 1k blocksize. I'll
> > > investigate more tomorrow.
> > 
> > So I've been trying quite hard to reproduce the failure but I failed. Since
> > you are apparently getting some error out of lseek can you find out which
> > error it is (likely ENXIO but I'd like to confirm) and where it gets
> > generated? I don't see how it could possibly happen that SEEK_DATA would
> > miss that single page generated by this test and how any of my patches
> > would influence this particular situation. Thanks!

<snip>

> Seems that's because the do {} while() loop in xfs_find_get_desired_pgoff() was
> broken out earlier due to patch 2.
> 
>                         /* Searching done if the page index is out of range. */
>                         if (page->index >= end) {
>                                 goto out;
>                         }
> 
> In my case, it returned earlier because page->index == end == 2.

Ah! That's it. I'm not sure why you get so short unwritten extent but it's
certainly possible. I can now reproduce the issue with:

xfs_io -f -c "falloc 0 10k" -c "pwrite 9k 512" -c "seek -d 0" /mnt/file
wrote 512/512 bytes at offset 9216
512.000000 bytes, 1 ops; 0.0000 sec (6.975 MiB/sec and 14285.7143 ops/sec)
Whence	Result
DATA	EOF

on 1k blocksize filesystem. And the problem is indeed that in this case I
have screwed up the condition due to rounding. I'll fix the second patch in
both series for ext4 & xfs. Thanks for debugging this!

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