[Top] [All Lists]

Re: Filesystem Consistency Issues

To: Ethan Benson <erbenson@xxxxxxxxxx>
Subject: Re: Filesystem Consistency Issues
From: Eric Sandeen <sandeen@xxxxxxx>
Date: Tue, 06 Dec 2005 08:11:34 -0600
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <20051206054953.GP14319@plato.local.lan>
References: <20051204142506.GE2605@free.net.ph> <20051205014137.GA7685@old.davidb.org> <p734q5nsre6.fsf@verdi.suse.de> <43943F3E.5080804@xfs.org> <20051206054953.GP14319@plato.local.lan>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 1.0.6 (Macintosh/20050716)
Ethan Benson wrote:
On Mon, Dec 05, 2005 at 07:23:10AM -0600, Stephen Lord wrote:

XFS runs recovery on a readonly filesystem. It has been doing that
for years. There are log messages which come out when recovery runs,
you should see them in dmesg or the syslog after boot up.

I believe there may be a slight problem with how this works in certain
circumstances.  I have observed the following problem, but have
not had time to do proper tests to confirm it.

Basically the situation is as follows:

1) /etc/fstab contains a filesystem mounted read-only by default.

2) said filesystem is remounted read-write, and files which are in use
   (running executables for example) are unlinked (such that link
   count becomes 0).

3) time passes.  (more then enough for everything to be synced).

4) system crashes or is rebooted uncleanly.

5) filesystem is mounted read-only, log recovery occurs anyway.

6) xfs_check or xfs_repair will report orphaned inodes which need to
   be moved to lost+found, if repair is performed inodes are indeed moved
   to lost+found, the inodes in question are the previously deleted files.

I have not had the chance to prove this, but I have seen enough
instances close enough to this that I believe it to be true.

There is something called "unlinked list processing" which should be handling this... bears investigation, I guess.



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