[Top] [All Lists]

Re: Serious XFS trouble: umount filesystem -> corrupt (on IDE drive)

To: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Subject: Re: Serious XFS trouble: umount filesystem -> corrupt (on IDE drive)
From: Michael Sinz <msinz@xxxxxxxxx>
Date: Mon, 25 Nov 2002 07:32:57 -0500
Cc: "Ralf G. R. Bergs" <rabe@xxxxxxxxxxxxxx>, Linux XFS Mailing List <linux-xfs@xxxxxxxxxxx>, "Maciej.BOROWKA@xxxxxxxxxxxxxxxxxxx" <Maciej.BOROWKA@xxxxxxxxxxxxxxxxxxx>
References: <E18FXQB-0002j5-00@xxxxxxxxxxxxxxxxxxxxxxxxxxxx> <20021123155213.A29887@xxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.1) Gecko/20021111
Christoph Hellwig wrote:
On Sat, Nov 23, 2002 at 11:21:54AM +0100, Ralf G. R. Bergs wrote:

Hi there,

yesterday I replaced the motherboard of my home server (running Debian/GNU Linux 3.0) for a CPU upgrade (486-133 -> K6-III+ 400) and the network card. Therefore (to support the card) I had to recompile the kernel, and so I upgraded the kernel source from 2.4.19-xfs to yesterday's 2.4.20-rc2-xfs ("SGI XFS CVS-2002-11-22_06:00_UTC with ACLs, quota, no debug enabled.")

Suddenly I noticed strange filesystem corruption. Even after a clean "halt" or "reboot" upon mounting the XFS filesystems the system would give me error messages like the following:

This was a rgeression I introduced in the xfs tree on thursday.  it's fixed
now.  let me apologize for the pain it caused to you.

Well, it did force me to do some cleanup on my recovery tools.
I also found out that it is very bad when you have a small disk with
XFS.  (Say 33meg or so...)  As my /boot is XFS since my cluster
kernels are compiled down to only support XFS and NFS, it makes for
a problem recovering such small partitions.  (xfs_restore does not
even begin is there is only on allocation group - not even if
it may be something simple :-()

Anyway, I would like to submit the following tool that I have written
up to install into /boot a special recovery support system.  It
assumes lilo as the boot loader and, if you let it update your lilo.conf,
it assumes that there is a linux.lastchance kernel (which is a kernel
that I keep around that is known to be good).  This is all also devfs
based.  (I know, I could fix that, but my direct need was to support
my environment...)

Anyway, it installs a set of tools to get you a minimal shell that
can be used to recover and an auto-recover mode (which is not that
great but good enough to cover the simple cases)

Read the script for more details...

(Oh, and the copying of kernel modules will get more than you need
but it was the easy way to manage this.  On my laptop, the recovery
stuff added just under 4meg of files to /boot)

REMINDER - this requires DEVFS - I am sure it would not be too much
work to make a non-DEVFS version.  It does handle SCSI and IDE.  It
also handles full disk volumes and partitions that the kernel supports.

Michael Sinz -- Director, Systems Engineering -- Worldgate Communications
A master's secrets are only as good as
        the master's ability to explain them to others.

Attachment: InstallRecoveryBoot
Description: application/java-vm

<Prev in Thread] Current Thread [Next in Thread>