Been running 2.4.2-XFS for a while. Been telling myself
to upgrade that thing for a while--never did, and now I've
got a blown up filesystem to show for it. Oops.
80 gig drive, filled to under 1.5 gigs free, about 5-10
active FTPs running at once. Hard crash, kernel panic
("tried to kill init"), mount fails on reboot with "I/O
error" while doing the mount recovery, but nothing more
useful than that (it kernel panics immediately afterward,
having no root FS.)
After some effort (mostly due to not being able to find
a boot CD with XFS on it), I managed to get xfs_repair
running from a floppy. Ran it; noticed far too many
errors, so stopped it. Good thing I did--running it
again (this time with -n--which I should have done to
begin with) shows that, all told, it probably would have
wiped out anything that remains ("would have junked"
each of my root directories, after a massive number of
other directories.) A quick check showed the boot CD
was a 2.2 kernel (XFS-less, for the sake of xfs_repairing);
I don't know if this would cause problems for what appears
to be a fully userspace program. (In any event, it probably
made the situation worse.)
I've logged a full xfs_repair -n. One thing missing from it
that the real (partial) run started out with was "error setting
block size: I/O error" (related to old kernel?) The log
is at
ftp://bscwulf.bridgew.edu/out.txt.gz
It's listing a lot of files that havn't been touched in
days or weeks, not just files that were being written
at the crash.
I did a dry run with both XFS trees; they were visibly identical
(though I didn't bother to diff them.)
xfs_logprint shows one entry: "Unmount filesystem".
Any options short of "complete data loss"? Oh, and if anyone
could point me to a reasonably stable CD image for XFS
recovery, I'd appreciate it.
--
Glenn Maynard
|