Random xfs_check errors

Random xfs_check errors

[Amir Goldstein]

Hi guys,

I've been running the latest xfstests with xfsprogs v5.9 and stable
kernel v5.9.y past week (including v5.9.2 today) and I'm getting
reports like below after random tests.

This one was after generic/466, but already got similar reports
(with more blocks) after generic/511,518,519 on different test runs.

Is anyone else seeing this?
Any advice on what to look for?

Did not get to test v5.10-rc1 yet. I thought I would ask here first
in case it's a known issue.

Sasha, I am wondering which xfsprogs version are you using
when verifying stable kernel patches, because I have just recently
upgraded xfsprogs to v5.9 in my stable test environment.

Thanks,
Amir.


_check_xfs_filesystem: filesystem on
/dev/mapper/0fce4d2d--779b--44e8--91d0--2cb1b252f68e-xfs_scratch is
inconsistent (c)
*** xfs_check output ***
block 0/10 expected type unknown got free1
agf_freeblks 1310702, counted 1310701 in ag 0
agi_freecount 61, counted 58 in ag 0
sb_ifree 61, counted 58
sb_fdblocks 5240288, counted 5240287
sb_fdblocks 5240288, aggregate AGF count 5240287
*** end xfs_check output

Re: Random xfs_check errors

[Darrick J. Wong]

On Fri, Oct 30, 2020 at 01:36:30PM +0200, Amir Goldstein wrote:
> Hi guys,
> 
> I've been running the latest xfstests with xfsprogs v5.9 and stable
> kernel v5.9.y past week (including v5.9.2 today) and I'm getting
> reports like below after random tests.
> 
> This one was after generic/466, but already got similar reports
> (with more blocks) after generic/511,518,519 on different test runs.
> 
> Is anyone else seeing this?
> Any advice on what to look for?
> 
> Did not get to test v5.10-rc1 yet. I thought I would ask here first
> in case it's a known issue.
> 
> Sasha, I am wondering which xfsprogs version are you using
> when verifying stable kernel patches, because I have just recently
> upgraded xfsprogs to v5.9 in my stable test environment.

FWIW I patched xfs_check out of fstests years ago.

Once Eric merges the large(ish) pile of xfs_repair patches I sent him
for 5.10 I think we'll be at the point where there aren't any
corruptions caught by xfs_check but not by xfs_repair.

Soon I'll also submit my rewrite of the fstests fuzzer code to Eryu, and
when that lands upstream then the rest of you will be able to check my
assertion.

Assuming none of you find new coverage holes that I didn't, we will
finally be able to drop xfs_check (which is bitrotting) from fstests
entirely.

--D

> Thanks,
> Amir.
> 
> 
> _check_xfs_filesystem: filesystem on
> /dev/mapper/0fce4d2d--779b--44e8--91d0--2cb1b252f68e-xfs_scratch is
> inconsistent (c)
> *** xfs_check output ***
> block 0/10 expected type unknown got free1
> agf_freeblks 1310702, counted 1310701 in ag 0
> agi_freecount 61, counted 58 in ag 0
> sb_ifree 61, counted 58
> sb_fdblocks 5240288, counted 5240287
> sb_fdblocks 5240288, aggregate AGF count 5240287
> *** end xfs_check output