Problems with XFS on iSCSI in RHEL4

Problems with XFS on iSCSI in RHEL4

[Andrew Daviel]

A while back (2004) there was some discussion about problems with XFS
and iSCSI.

I wondered whether the issues had been fixed in later kernels, or whether 
there is an underlying issue that will not go away.

We have been using XFS for our mailstore both on local disks and on ATA 
Over Ethernet to a CORAID box with no problem.
We want to replace the CORAID with a Sun Storage 7410 box, using iSCSI 
instead of AOE.

When I create an ext3 volume on the Sun, things are OK.
When I create an XFS volume, after doing mkdir, I get e.g.
   kernel: Bad page state at free_hot_cold_page (in process 'iscsi-rx',
   page c1069a20)
   kernel: Trying to fix it up, but a reboot is needed
   kernel: Bad page state at free_hot_cold_page (in process 'syslogd',
   page c1069a20)

Scientific Linux SL release 4.6
kernel-smp-2.6.9-78.0.17.EL from kernel-2.6.9-78.0.17.EL.src.rpm
   includes scsi_transport_iscsi.ko,iscsi_sfnet.ko
kernel-module-xfs-2.6.9-78.0.17.ELsmp-0.4-1 from xfs-0.4-1.src.rpm
iscsi-initiator-utils-4.0.3.0-6

Also seen on:
Scientific Linux SL release 4.5
kernel-smp-2.6.9-55.0.6.EL.i686
kernel-module-xfs-2.6.9-55.0.6.ELsmp-0.1-1.i686

SL is basically the same as CENTOS (compiled from RHEL GPL'd sources)


-- 
Andrew Daviel, TRIUMF, Canada
Tel. +1 (604) 222-7376  (Pacific Time)
Network Security Manager

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Problems with XFS on iSCSI in RHEL4

[Tim Sesow]

For what it is worth:  I run three systems, one 32-bit and two 64-bit, 
with 2.6.24-rc3 kernels on
them, using iSCSI storage running under OpenFiler as storage for
XFS volumes.  The systems started out as CentOS 4.6 distributions.

Two of the volumes are less than 1TByte in size (300GB and 800GB).
The third volume under LVM and is 45TBytes in size (several 
multi-terabyte volumes
merged into one LVM volume group).  The smaller volumes have been running
for several months, the large volume only for about a month now.
 
 I haven't encountered any problems creating or using those volumes with 
XFS.
The 45TByte volume also uses DMAPI as a mount option. 

Andrew Daviel wrote:
>
> A while back (2004) there was some discussion about problems with XFS
> and iSCSI.
>
> I wondered whether the issues had been fixed in later kernels, or 
> whether there is an underlying issue that will not go away.
>
> We have been using XFS for our mailstore both on local disks and on 
> ATA Over Ethernet to a CORAID box with no problem.
> We want to replace the CORAID with a Sun Storage 7410 box, using iSCSI 
> instead of AOE.
>
> When I create an ext3 volume on the Sun, things are OK.
> When I create an XFS volume, after doing mkdir, I get e.g.
>   kernel: Bad page state at free_hot_cold_page (in process 'iscsi-rx',
>   page c1069a20)
>   kernel: Trying to fix it up, but a reboot is needed
>   kernel: Bad page state at free_hot_cold_page (in process 'syslogd',
>   page c1069a20)
>
> Scientific Linux SL release 4.6
> kernel-smp-2.6.9-78.0.17.EL from kernel-2.6.9-78.0.17.EL.src.rpm
>   includes scsi_transport_iscsi.ko,iscsi_sfnet.ko
> kernel-module-xfs-2.6.9-78.0.17.ELsmp-0.4-1 from xfs-0.4-1.src.rpm
> iscsi-initiator-utils-4.0.3.0-6
>
> Also seen on:
> Scientific Linux SL release 4.5
> kernel-smp-2.6.9-55.0.6.EL.i686
> kernel-module-xfs-2.6.9-55.0.6.ELsmp-0.1-1.i686
>
> SL is basically the same as CENTOS (compiled from RHEL GPL'd sources)
>
>

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Problems with XFS on iSCSI in RHEL4

[Christoph Hellwig]

On Wed, May 06, 2009 at 12:32:25PM -0700, Andrew Daviel wrote:
> When I create an ext3 volume on the Sun, things are OK.
> When I create an XFS volume, after doing mkdir, I get e.g.
>   kernel: Bad page state at free_hot_cold_page (in process 'iscsi-rx',
>   page c1069a20)
>   kernel: Trying to fix it up, but a reboot is needed
>   kernel: Bad page state at free_hot_cold_page (in process 'syslogd',
>   page c1069a20)

That looks like the old disagreement on whether we could send
non-refcountable pages to the block layer.  XFS used to do this and
iscsi didn't like it.  Until today it's still not specified if it
is allowed or not and most drivers don't care but XFS stopped doing
that.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs