Xen users - Mar 2008 - Save/Restore/Reboot FS Corruption

I''ve been looking for information about this matter, but
didn''t found
too much.

Saving and restoring works pretty good, but the problem comes when I  
reboot after
restoring. FileSystem becomes corrupt after the restoration, so it  
enters "read only"
status.

Can''t provide more info, since I''m far from the computer, only
thing I
can say, its a Debian Etch debootstraped VM, with a ext3-fs.
The disk image of the VM is stored on a remote storage device, so the  
directory containing the image is mounted via NFS.

Is anyone else experiencing this issue?



_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hrrrmm.  Sounds nasty :-(
> I''ve been looking for information about this matter, but
didn''t found
> too much.
>
> Saving and restoring works pretty good, but the problem comes when I
> reboot after
> restoring. FileSystem becomes corrupt after the restoration, so it
> enters "read only"
> status.
>
> Can''t provide more info, since I''m far from the computer,
only thing I
> can say, its a Debian Etch debootstraped VM, with a ext3-fs.
> The disk image of the VM is stored on a remote storage device, so the
> directory containing the image is mounted via NFS.
>
> Is anyone else experiencing this issue?
What was the sequence of operations that you went through when this happened?  
Something like:

xm save <domain>
... time passes ...
xm restore <domain>
... reboot domain ...
... corruption!

Is it possible that anything modified the domain''s virtual disks during
the "... time passes ..." section above?  Like maybe somebody mounted
a
filesystem in the virtual disk from dom0 for some reason?  Or if somebody 
accidentally did an xm create in the meantime?

Domains will, unfortunately, typically corrupt their virtual disks if their 
filesystem changes whilst they''re suspended.  You might not notice this
until
the filesystem gets mounted at reboot time.  Linux doesn''t generally
expect
data to change on a disk it thinks it has mounted - even if it''s
actually xm
saved at the time - and it can get confused if changes do occur.

Cheers,
Mark

-- 
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

> 
> Is it possible that anything modified the domain''s virtual disks
during
> the "... time passes ..." section above?  Like maybe somebody
mounted
a
> filesystem in the virtual disk from dom0 for some reason?  Or if
somebody
> accidentally did an xm create in the meantime?
> 
Just thinking out loud here, but how much work do you think it would be
to stop this happening for at least lvm backed devices?

lvchange has the ability to change an lv to not-available (basically
offline), or to read-only. Assuming these changes are persistent across
reboots of Dom0, an ''xm save'' could set the lv to
not-available after
the save is complete, and resume could check that it is in a
not-available state before changing it back to available and starting
the domain again. That would make sure nobody has inadvertently touched
the lv in the meantime. An ''xm create'' would fail because the
lv was not
available, as would any attempt to mount it. ''xm create'' would
need to
have a ''force'' option to set the lv back to available again.

For file backed devices, you could just rename the backing file on save
and rename it back again on resume.

For physical disk backed devices I don''t think there is such an easy
solution...

James

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users