In recent domU kernels, we need to mount the filesystems with barrier=0, because the old dom0 kernels lie about their barrier support or something. Maybe barriers are really obsolete anyway though? http://lkml.indiana.edu/hypermail/linux/kernel/1105.0/02123.html http://lists.xen.org/archives/html/xen-users/2011-12/msg00017.html
I forget where I read now, about the old dom0s lying about the barrier support which is how the filesystem can become out of whack. Is that really true?
added by lsc:
So this is a problem with the 2.6.32 dom0 kernels, but not the 2.6.18-308.4.1.el5xen (CentOS-xen) dom0 kernels? if so, perhaps we should upgrade everything to the centos-xen kernels?
We've also had the problem on hamper, which is running 184.108.40.206-xen. I will also try to test this on the centos dom0 kernel. --Nick 22:29, 29 May 2012 (EDT)
For reference this is the bug:
What it boils down to is the blk drivers in 3.x were changed significantly and it introduced a bug that occurs when the dom0 is still using a 2.6 kernel. It basically causes significant filesystem errors and typically prevents booting. It can even corrupt the guests's filesystem entirely.
This bug was found and patched in the 2.6 series. The Debian package kernel received a point release. Any Debian Squeeze kernel beyond 2.6.32-41 should contain the patch. The dom0 host has to be upgraded for this, obviously. This was patched relatively recently, barely even a year if that.
There is a workaround that the guest can use by disabling barriers (where the bug lies), using barriers=0 in the /etc/fstab options for mounting their filesystems. This workaround is mentioned in the bug report linked above.
-- Best, Dan Linehan