xfs
[Top] [All Lists]

Re: 2.6.26.3 / loop-aes/xfs corruption?

To: Justin Piszcz <jpiszcz@xxxxxxxxxxxxxxx>
Subject: Re: 2.6.26.3 / loop-aes/xfs corruption?
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Sun, 23 Aug 2009 18:48:38 -0400
Cc: xfs@xxxxxxxxxxx
In-reply-to: <alpine.DEB.2.00.0908221040220.30321@xxxxxxxxxxxxxxxx>
References: <alpine.DEB.2.00.0908221040220.30321@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.19 (2009-01-05)
On Sat, Aug 22, 2009 at 10:44:31AM -0400, Justin Piszcz wrote:
> Hi,
>
> On an older machine:
>
> [2340962.903497] XFS internal error XFS_WANT_CORRUPTED_GOTO at line 1545 of 
> file
>  fs/xfs/xfs_alloc.c.  Caller 0xc021ccd8
> [2340962.903497] Pid: 24649, comm: mv Not tainted 2.6.26.3 #1

This is a very typical sign for corruption cause by the lack of write
barriers on enabled write caches.

> [2340962.911473] Filesystem "loop0": Corruption of in-memory data detected.  
> Shu

And using the loop device is kindof a worsened version of hanving write
caches as we don't just have small disk caches that can fill up but
major amounts of the main memory.

The loop device before 2.6.30 is unfortunately not safe in case of
system crashes.  And given your other report we'll have to see if maybe
the barrier support in 2.6.30 cause other kinds of problems to, although
it could be something else.

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