On Fri, 2002-03-01 at 17:07, Stuart Levy wrote:
I think that translates into a Jan 23rd kernel ....
>
> The whole thing worked well for a few days' fairly heavy use.
> But for the last couple of days -- the file system has become
> more than about half full -- I've been getting kernel oopses
> like those below, every 2-4 hours' use.
>
> After each crash I try mounting/unmounting the filesystem;
> xfs_check says all's well, and
> xfs_repair -l /dev/hda5 /dev/sda1
> also does. (I had to tweak xfs_repair.c to accept a value for -l --
> I see the same change was made in CVS recently too.)
>
> Files don't seem to be corrupted, except sometimes for
> 0-filled ones that were being written at crash time,
> i.e. all seems well.
>
> Does anything seem especially promising to try?
> Should I build from the live CVS copy (if so, which tag?)?
> Go back to 2.4.14 and use the 1.0.2 XFS release?
> Try another compiler?
>
Hmm, difficult to detangle that one, but there is a chance you got a
buffer with no memory in it down there. Could you possibly try the
current cvs tree - or there will be new split patches for 2.4.18
in a day or so (probably Monday). I put some memory related fixes
in - and a whole boat load of other stuff. If you are oopsing this
frequently then a newer kernel cannot do any harm.
You will need new commands if you care about acls, they are all
available on the ftp site.
Let us know if this helps and if not we will take it from there.
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|