From: David Coulson <david@da...> - 2002-12-09 18:07:34
This may interest those who are using ReiserFS for UML filesystems.
-------- Original Message --------
Subject: Re: Damaged filesystems
Date: Mon, 9 Dec 2002 09:21:53 +0300
From: Oleg Drokin <green@...>
To: David Coulson <david@...>
On Sun, Dec 08, 2002 at 10:07:45PM -0500, David Coulson wrote:
> I know there are reiserfs lists, but I already searched on google and
> groups.google, coming up blank, and since you use UML I figured you
> might have come across this.
Actually not yet. (But I must tell you I've seen plently of reiserfs
on LVM after system crashes (and not only reiserfs to be precise, but I am
mostly interested in reiserfs to the nature of my job ;) ), seems that
LVM have some deep problems syncing data to disk and this kind of stuff.
We've yet been able to reproduce something like that reliably on our boxes.
> I've a ~420Gb ReiserFS filesystem, which resides on a LVM LV. Within
> this filesystem I have a number of sparse files containing ReiserFS
> filesystems for UMLs. The filesystem on the LVM had a number of errors
> which became apparent during 'updatedb', so I ran '--rebuild-tree' on it
> using the current 3.6.4 reiserfs tools, after running --check on it. It
> took a while, but eventually it finished and said all the errors had
> been repaired. The filesystem mounted without problems and appears to be
Have you copy of all the messages reiserfsck --rebuild-tree printed to
(by using e.g. -l switch to reiserfsck).
Alternatively if you can supply kernel messages from reiserfs that indicate
on-disk corruptions, that might also help.
> Today, a number of UML users have said their homefs image files are
> useless and I can't reiserfsck them:
> Pass 0:
> ####### Pass 0 #######
> Loading on-disk bitmap .. ok, 8242 blocks marked used
> Skipping 8242 blocks (super block, journal, bitmaps) 0 blocks will be read
Hm, this "0 blocks will be read" is a most probably explanation to
"no reiserfs metadata found".
> The filesystem still appears to exist, but I can't do anything useful
> with it. Have you any suggestions how to recover from this, or should I
> just put it down to experience and never use --rebuild-tree again?
Actually --rebuild-tree is mostly safe these days. This is a first first
in a long time of reiserfsck --rebuild-tree corrupting valid stuff.
Were filesystems accessible before you did reiserfsck --rebuild-tree?
Are these damaged FS images big?
If they are really big, I still want at least metadata snapshot.
You can obtain that with debugreiserfs -p fs_image | bzip2 -9c >metadata.bz2
If that produces "no reiserfs metadata found", then try to add -S switch
to debugreiserfs (of course I imply you use latest version of debugreiserfs
too). If -S switch won't help either, then we probably need a shell
the box to look at the fs image.
David Coulson email: d@...
Linux Developer / web: http://davidcoulson.net/
Network Engineer phone: (216) 533-6967