On Wed, Aug 25, 2004 at 12:35:55PM -0700, Wenzel Jakob wrote:
> I am experiencing XFS file system corruption on a vanilla kernel.org
> kernel 2.6.7 (the only additional module is the nvidia driver which
> really shouldn't be the cause of the problems).
Ick, as pointed out the nvidia driver *might* (not necessarily is, but
*might* be causing problems --- we just can't tell). How repeatable
are the problems and can you make them occur w/o the nvidia driver
having been loaded?
> The problem is that, all of a sudden, files have the content of an
> entirely different file. For example one second I am editing a .cpp
> file, the next moment I open it up again, it is a jpeg file from a
> directory containing my digital camera pictures. The original .cpp
> file was lost, so the files were not exchanged.
That's really weird. If you reboot and check the files again do you
see the wrong content? It would be good to know if this is on-disk or
in-memory (page-cache corruption).
Also, for the files involved, do you know *which* two files got messed
up? I'm wondering if they have a similar location on disk, similiar
inode numbers or something that a bit-flip would explain.
> This has happened to me before with previous kernels from 2.6 - what
> I usually did is run knoppix and fsck the filesystem which seemed to
> stop it for a while.
Did it find any errors? What output do you typically get here?
On Wed, Aug 25, 2004 at 04:10:07PM -0500, Net Llama! wrote:
> There is no fsck for XFS. What exactly were you running?
xfs_repair I'm guessing. It *is* fsck for XFS really. fsck.xfs isn't
a NOP to make init-scripts work.
> Have you checked your logs (particularly messages) for errors? This
> sounds like a hardware problem, or the nvidia driver is eating you
> disk.
It sounds more like in-core corruption to me (which could get flushed
out so persist on disk at a later stage).
--cw
|