Re: OCFS2 Filesystem inconsistency across nodes

[Jan Kara <jack@suse.cz>]

> On Sat, Feb 11, 2006 at 05:40:57AM +0000, Claudio Martins wrote:
> > This is my /etc/ocfs2/cluster.conf on every node:
> 
> Hi Claudio,
> 	Thanks for sending me your config files. Everything seems in order.
> I was easily able to reproduce your problem on my cluster and was able to
> git-bisect my way to some JBD changes which seem to be causing the issue.
> Reverting those patches fixes things. Can you apply the attached patch and
> confirm that it also fixes this particular problem for you? You'll have to
> apply to all kernels in your cluster and either run fsck.ocfs2 or create a
> new file system before testing again.
> 
> Linus, Andrew, Jan,
> 	OCFS2 uses journal_flush() to sync metadata out to disk when another
> node wants to obtain a lock on an inode which has pending journaled changes.
> Something in Jan Kara's patch titled "jbd: split checkpoint lists" broke
> this for OCFS2 (and I suspect for other users of JBD as well). As a result
> metadata is not always completely flushed to disk by the end of the
> journal_flush() call.
> 
> One easy way to reproduce is to create files from one node and list the
> directory from another. Switching the listing and creating nodes around
> makes things reproduce more quickly -- eventually the listing node will
> start missing new files.
  Ok, I'll have a look at the problem. Probably something in
log_do_checkpoint() is not waiting for all the data or something like
that. I'll try to reproduce with ext3 - it uses journal_flush() in
ext3_bmap() so if journal_flush() is not flushing all the data we should
be able to see that... Thanks for spotting the problem.

							Bye
								Honza

			
-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/