xfs
[Top] [All Lists]

Re: XFS on 2.4 Tib raid

To: Dave Chinner <dgc@xxxxxxx>, naude@xxxxxxxxxxxxxxxxxxxxx
Subject: Re: XFS on 2.4 Tib raid
From: Leon Vismer <lvismer@xxxxxxxxxxxxxxxx>
Date: Wed, 11 May 2005 07:27:34 +0200
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <20050511084450.A91712@xxxxxxxxxxxxxxxxx>
References: <200505092210.25667.lvismer@xxxxxxxxxxxxxxxx> <1115712231.42806ae75b4d1@xxxxxxxxxxxxxxxx> <20050511084450.A91712@xxxxxxxxxxxxxxxxx>
Reply-to: lvismer@xxxxxxxxxxxxxxxx
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: KMail/1.8
Hi Dave

On Wednesday 11 May 2005 00:44, Dave Chinner wrote:
> On Tue, May 10, 2005 at 10:03:51AM +0200, lvismer@xxxxxxxxxxxxxxxx wrote:
> > > > I am using an ATTO scsi card speaking to an Infortrend 2.4 Tib raid
> > > > system.
> > > >
> > > > I created a filesystem using:
> > > > mkfs.xfs -L iftraid -f /dev/sdc
>
> So you haven't created a partition table on the device at all?
>
No, however the device mounted 100%

> > > What's the actual error? This is the stack trace that's part of the
> > > error message, but does not inidcate what the error was. Can you
> > > include the error messages that were emitted before the stack trace?
> >
> > I include the full error from /var/log/messages. As one copies
> > files it stops at some point and the following error loops and the
> > machine needs to be rebooted.
> >
> > 0x0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> > Filesystem "sdc": XFS internal error xfs_alloc_read_agf at line 2195 of
> > file fs/ xfs/xfs_alloc.c.  Caller 0xd0b638ca
> >  [<d0b63ced>] xfs_alloc_read_agf+0x12d/0x1f0 [xfs]
> >  [<d0b638ca>] xfs_alloc_fix_freelist+0x47a/0x490 [xfs]
> >  [<d0b6407d>] xfs_alloc_vextent+0x2cd/0x500 [xfs]
>
> So we're trying to allocate an extent and the AGF read from disk
> is full of zeros.
>
> Can you do the following:
>
>  - write a known pattern to the disk before making the
>    filesystem (0xa5 is a good one)
>
The reason we did not originally create a partition is that the fdisk utils 
could not address the 2.4 Tib device. It returned a size of 200Mb or 
something similar.

>  - make the filesystem
>  - run xfs_check on the device before mounting to validate
>    the filesystem was made properly
>  - mount the filesystem
>  - run your copy until it breaks
>  - run xfs_check on the filesystem (if you needed to reboot
>    to get here, you should mount and unmount the filesystem
>    to replay the log first)
>  - run xfs_repair on the filesystem if xfs_check finds
>    errors
>
> And post the output of any errors that are found?
>
Dave, unfortunately we were under presure to get the device going as the 
clients current storage was not sufficient. We in the end created two 
partitions on the 2,4 Tib InforTrend (2048Gib and 234Gib). We create XFS on 
both sdc and sdd filesystems, mounted them and managed to copy 400Gig onto 
sdc without a single suspicious message. It would have been nice to use the 
full size.

> If there really is a AGF full of zeros (or 0xa5!) in the filesystem,
> xfs_check will find it.
>
> BTW, did you build the kernel with CONFIG_LBD=y (i.e. support block
> devices larger than 2TiB)?
>
I am using the 2.6.11 kernel from Debian sid and according to 
the /boot/config-2.6.11-1-i386 file (the .config used to build the shipped 
kernel) CONFIG_LBD is set to Y.

Many thanks for all the help
--
Leon

> Cheers,
>
> Dave.


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