Re: [xfs-masters] filesystem corruption on xfs after 2.6.25-rc1 (bisected, powerpc related?)

Re: [xfs-masters] filesystem corruption on xfs after 2.6.25-rc1 (bisected, powerpc related?)

[Eric Sandeen]

Gaudenz Steinlin wrote:
> Hi
> 
> Since upgrading to 2.6.25-rc1 I see filesystem corruption on my XFS
> filesystem. I can reproduce this by doing "git reset --hard v2.6.25-rc1"
> on a git checkout which is on some other revision. Git outputs strange
> error messages (like file xxx is a directory when xxx really is a file)
> and sometimes the filesystem "hangs" (I can no longer do any operations
> on it even from another shell). If I reboot with a working kernel and
> check the filesystem xfs_check reports many errors. I also see the
> problem when doing other (not related to git) operations on the
> filesystem. Git reset is just the easiest way to reproduce it.
> 
> I was able to track this corruption down to commit
> a69b176df246d59626e6a9c640b44c0921fa4566 ([XFS] Use the generic bitops
> rather than implementing them ourselves.) using git bisect.
> 
> Reverting edd319dc527733e61eec5bdc9ce20c94634b6482 ([XFS] Fix
> xfs_lowbit64) to avoid merge conflicts and the faulty commit on top of
> 2.6.25-rc3 fixes the problem.

If you're feeling motivated, maybe you can narrow it down to which of
the changes - xfs_highbit32, xfs_highbit64, xfs_lowbit32, or
xfs_lowbit64 - is causing the problem?  (or maybe they all are ...)

Or maybe someone looking at the commit can immediately see the
problem... but I can't :)

-Eric

Re: [xfs-masters] Re: filesystem corruption on xfs after 2.6.25-rc1 (bisected, powerpc related?)

[Eric Sandeen]

Johannes Berg wrote:
>> I debuged this a bit further by testing the 4 changed functions
>> individually. The problem only occurs with the new version of
>> xfs_lowbit64. 
...

> You really need to keep xfs_lowbit64 defined as it was before, or, maybe
> even better, define ffs64 in parallel to fls64.

Yep, I agree.

-Eric