[Top] [All Lists]

Re: Issue with xfs on dm snapshots

To: sdrb <sdrb@xxxxxxx>
Subject: Re: Issue with xfs on dm snapshots
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Thu, 19 May 2011 09:52:23 +1000
Cc: xfs@xxxxxxxxxxx
In-reply-to: <4DD3876F.2080805@xxxxxxx>
References: <4DD3876F.2080805@xxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
On Wed, May 18, 2011 at 10:46:39AM +0200, sdrb wrote:
> Hello
> I encountered some problems using XFS on LVM and snapshots on linux 2.6.35.
> The problem is that when there are several snapshots and some of
> them reached theirs maximum space and disks are under heavy load -
> it causes that the tasks that rely somehow on device mapper - just
> hung.

Yup. Running your shapshot target out of disk space causes Bad
Things To Happen. Typical problem is that metadata write IO's fail
and XFS shutѕ the filesytem down in response to the corruption that
the failing IO has caused.

Looking at your log, i see that is exactly what XFS has done.

The lockdep reports are about dm-snapshot locks and are not XFs
related, so that's probably where the hang is occurring. I'm sure
you could produce the same hang using ext4 instead of XFS. You should
report this to the dm-devel list.

The take away message: Don't let your snapshot volumes run out of
space - you'll corrupt data and/or your filesystem.


Dave Chinner

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