> On Mon, Aug 22, 2011 at 11:27:05AM +1000, Dave Chinner wrote:
> > On Sun, Aug 21, 2011 at 03:59:34PM -0700, Robert Jacobs wrote:
> > > > On Mittwoch, 17. August 2011 Robert Jacobs wrote:
> > > > > I've attached the outputs of xfs_check and three different runs of
> > > > > xfs_repair (with backtrace, when available).
> > > >
> > > > Can you provide a XFS metadump? That should help devs to reproduce the
> > > > error.
> > >
> > > I've uploaded it to <http://eamp.org/metadumpf.lzma>. It's 9MB but I
> > > have somewhat limited upload quota. I noticed that some filenames
> > > weren't obscured so I obscured them by hand -- if this is a problem,
> > > please tell me.
> >
> > The metadump image you have provided does not trigger problems on
> > either 3.1.5 on a current xfs_repair build out of the xfsprogs-dev
> > repository on my machines. Can you restore the image and run
> > xfs_repair on the image and confirm it still crashes?
>
> I stand corrected - it doesn't crash on x86-64, but does crash on
> i686. I get several different random crashes, so it looks like some
> kind of heap corruption is occurring. Running under valgrind shows
> that the inode extent blkmaps are getting out of bound reads and
> writes. I think the problem is this:
[...]
> As it is, I'd recommend upgrading to x86-64 anyway - XFS on i686
> doesn't receive a lot of developer testing...
Sadly, I can't upgrade to x86-64 because this is a i686 machine that I
have not yet outgrown. Someday soon...
But thank you very much! I now have a functioning filesystem again.
- Robert Jacobs
|