[Top] [All Lists]

Re: XFS filesystem shutting down on linux (xfs_rename)

To: Chris Samuel <csamuel@xxxxxxxx>
Subject: Re: XFS filesystem shutting down on linux (xfs_rename)
From: Eric Sandeen <sandeen@xxxxxxxxxxx>
Date: Tue, 04 Aug 2009 23:26:49 -0500
Cc: xfs@xxxxxxxxxxx
In-reply-to: <1367391532.793061249444829356.JavaMail.root@xxxxxxxxxxxxx>
References: <1367391532.793061249444829356.JavaMail.root@xxxxxxxxxxxxx>
User-agent: Thunderbird (Macintosh/20090605)
Chris Samuel wrote:
> Hi folks,
> I believe we've been hitting the same issue that
> Gabriel Barazer reported in on the 22nd
> of July on our NFS server for our HPC Linux clusters.


> This kernel is built with XFS as a kernel module so I've
> been able to attach the objdump output that Eric Sandeen
> had originally requested from Gabriel.

I'll take a look.  The reason for the objdump was to try to work out
where in xfs_rename we were erroring out, but in your case:


I guess it's the trans_cancel at the very end of the function on the
goto <blah>: error cases.

Is there any chance you could instrument xfs_rename so that for every
goto abort_return and goto error_return you add a printk so we know
which one you hit?

Maybe something like:

printk("line %s error %d\n", __LINE__, error);

just before each goto, with appropriate braces {} sprinkled in as necessary?

I can send a patch if that helps.

Sorry for the brute-force debugging but not at all sure yet what this
may be.


> Like Gabriel we're stuck on 2.6.28.x as the last working
> NFS exporting XFS kernel due to kernel bug #13375 (the
> radix bug), so I hope this helps!
> cheers,
> Chris
> ------------------------------------------------------------------------
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs

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