Fedora xen - Jan 2008 - How to backup a Domu filesystem (on LVM) from Dom0 on Fedora 8 ?

Hello

I use XEN on Fedora 8 with Domu on a LVM partition.

Before, on Fedora Core 6, I could backup each Domu by directly mounting 
the filesystem of the Domu (mount /dev/vg00/lvfoo /mnt) and then running 
dump on it.

On Fedora 8, I can''t mount directly the filesystem of the Domu, so I 
backup the filesystem of the Domu using the dd command (dd 
if=/dev/vg00/lvfoo of=/var/xen/lvfoo.img)
But the backup with dd is longer and takes more disk space than the 
backup with dump, as dd copies unused blocks.

Is there a better method to backup a Domu filesystem from Dom0 on Fedora 8 ?

Thanks in advance

Anne Facq

Anne Facq wrote:
> Hello
> 
> I use XEN on Fedora 8 with Domu on a LVM partition.
> 
> Before, on Fedora Core 6, I could backup each Domu by directly mounting 
> the filesystem of the Domu (mount /dev/vg00/lvfoo /mnt) and then running 
> dump on it.
> 
> On Fedora 8, I can''t mount directly the filesystem of the Domu, so
I
Pardon if this is a dumb question, but why not?
> backup the filesystem of the Domu using the dd command (dd 
> if=/dev/vg00/lvfoo of=/var/xen/lvfoo.img)
> But the backup with dd is longer and takes more disk space than the 
> backup with dump, as dd copies unused blocks.
> 
> Is there a better method to backup a Domu filesystem from Dom0 on Fedora 
> 8 ?
I suppose you could use the img2qcow tool (possibly directly on 
/dev/vg00/lvfoo) to get a qcow file instead of a raw image.  However, I 
vaguely remember seeing somewhere that the qcow created by the xen tools 
is not necessarily 100% compatible with other qcow tools.

HTH,
Aaron

Aaron Clark wrote:
> Anne Facq wrote:
>> Hello
>>
>> I use XEN on Fedora 8 with Domu on a LVM partition.
>>
>> Before, on Fedora Core 6, I could backup each Domu by directly 
>> mounting the filesystem of the Domu (mount /dev/vg00/lvfoo /mnt) and 
>> then running dump on it.
>>
>> On Fedora 8, I can''t mount directly the filesystem of the
Domu, so I
> 
> Pardon if this is a dumb question, but why not?
Because in Dom0 the device for a paravirt Domu (/dev/vg00/lvfoo in my
case) is mapped to a virtual disk /dev/xvda (I created the Domu with 
virt-install), and the only way I found to mount this Domu filesystem, 
is to :
- get the size of the boot sector of /dev/vg00/lvfoo
	(with fdisk -ul /dev/vg00/lvfoo)
	= 63
- compute the offset (multiply by block size)
- specify this offset to the mount command :
mount -o offset=32256 /dev/vg00/lvfoo /mnt/

As I can''t specify an offset to dump, the command "dump 0 -f
lvfoo.dump
/mnt" doesn''t work.
>> backup the filesystem of the Domu using the dd command (dd 
>> if=/dev/vg00/lvfoo of=/var/xen/lvfoo.img)
>> But the backup with dd is longer and takes more disk space than the 
>> backup with dump, as dd copies unused blocks.
>>
>> Is there a better method to backup a Domu filesystem from Dom0 on 
>> Fedora 8 ?
> 
> I suppose you could use the img2qcow tool (possibly directly on 
> /dev/vg00/lvfoo) to get a qcow file instead of a raw image.  However, I 
> vaguely remember seeing somewhere that the qcow created by the xen tools 
> is not necessarily 100% compatible with other qcow tools.
Thanks for you help, I''m going to have a look at it.

Regards,

Anne Facq

Anne Facq wrote:
> Because in Dom0 the device for a paravirt Domu (/dev/vg00/lvfoo in my
> case) is mapped to a virtual disk /dev/xvda (I created the Domu with
> virt-install), and the only way I found to mount this Domu filesystem,
> is to :
> - get the size of the boot sector of /dev/vg00/lvfoo
>     (with fdisk -ul /dev/vg00/lvfoo)
>     = 63
> - compute the offset (multiply by block size)
> - specify this offset to the mount command :
> mount -o offset=32256 /dev/vg00/lvfoo /mnt/
Uh, oh, quite complicated and error prone.  Can be done much easier.

Option (1): Use kpartx, it will create device mapper mappings for your
partitions.  Try "kpartx -v -a /dev/vg00/lvfoo", should give you
/dev/vg00/lvfoop[1234].  "kpartx -d ... " removes the mappings.

Option (2): (works only with xen kernel): Configure the device as
virtual disk *in Domain-0*: "xm block-attach 0 phy:/dev/vg00/lvfoo xvda
w".  Gives you /dev/xvda[1234].  "xm block-detach ..." removes
it.

HTH,
  Gerd

Anne Facq <annefacq@crpp-bordeaux.cnrs.fr> writes:
> Aaron Clark wrote:
>> Anne Facq wrote:
>>> Hello
>>>
>>> I use XEN on Fedora 8 with Domu on a LVM partition.
>>>
>>> Before, on Fedora Core 6, I could backup each Domu by directly
>>> mounting the filesystem of the Domu (mount /dev/vg00/lvfoo /mnt)
>>> and then running dump on it.
>>>
>>> On Fedora 8, I can''t mount directly the filesystem of the
Domu, so
>>> I 
>>
>> Pardon if this is a dumb question, but why not?
>
> Because in Dom0 the device for a paravirt Domu (/dev/vg00/lvfoo in my
> case) is mapped to a virtual disk /dev/xvda (I created the Domu with
> virt-install), and the only way I found to mount this Domu filesystem,
> is to :
> - get the size of the boot sector of /dev/vg00/lvfoo
> 	(with fdisk -ul /dev/vg00/lvfoo)
> 	= 63
> - compute the offset (multiply by block size)
> - specify this offset to the mount command :
> mount -o offset=32256 /dev/vg00/lvfoo /mnt/
>
> As I can''t specify an offset to dump, the command "dump 0 -f
lvfoo.dump
> /mnt" doesn''t work.
Shot from the hip: losetup with the offset?

[...]

Gerd Hoffmann wrote:
> Anne Facq wrote:
>> Because in Dom0 the device for a paravirt Domu (/dev/vg00/lvfoo in my
>> case) is mapped to a virtual disk /dev/xvda (I created the Domu with
>> virt-install), and the only way I found to mount this Domu filesystem,
>> is to :
>> - get the size of the boot sector of /dev/vg00/lvfoo
>>     (with fdisk -ul /dev/vg00/lvfoo)
>>     = 63
>> - compute the offset (multiply by block size)
>> - specify this offset to the mount command :
>> mount -o offset=32256 /dev/vg00/lvfoo /mnt/
> 
> Uh, oh, quite complicated and error prone.  Can be done much easier.
> 
> Option (1): Use kpartx, it will create device mapper mappings for your
> partitions.  Try "kpartx -v -a /dev/vg00/lvfoo", should give you
> /dev/vg00/lvfoop[1234].  "kpartx -d ... " removes the mappings.
> 
> Option (2): (works only with xen kernel): Configure the device as
> virtual disk *in Domain-0*: "xm block-attach 0 phy:/dev/vg00/lvfoo
xvda
> w".  Gives you /dev/xvda[1234].  "xm block-detach ..."
removes it.
Many thanks for this solution, it helps me a lot.

If this can help someone, here are the steps to backup a Domu 
(filesystem on LVM in backend and on xvda in frontend) with the dump 
command from Dom0 (Fedora 8), using one of the 2 solutions you advised, 
kpartx :

----------------------------------------------------------------

# The DomU named "foo" must be stopped before dump
xm shutdown -w foo

# Creates device mapper mappings /dev/mapper/lvfoop[1234] for 
/dev/vg00/lvfoo
kpartx -v -a /dev/vg00/lvfoo

# Mounts temporarly the filesystem to backup
mount /dev/mapper/lvfoop1 /mnt/lvfoo/

# Launches the dump
/sbin/dump 0 -L FULL_foo -f /BACKUP-XEN/foo.colddump /mnt/lvfoo/

# Umount the filesystem
umount /mnt/lvfoo/

# Removes the mappings
kpartx -d /dev/vg00/lvfoo

# Unmount the filesystem
umount /mnt/lvfoo

# The DomU can now be restarted
xm create --config /var/lib/xend/....

----------------------------------------------------------------

Regards,

Anne Facq