This script allows you to backup Virtual Machines managed by libvirt.
It has only be tested with KVM based VM
This script will dump (or mount as a set of chunks):
 * each block devices
 * optionnally the memory (if --state flag is given)
 * the XML description of the VM

These files are writen in a temporary backup dir. Everything is done
in order to minimize donwtime of the guest. For example, it takes
a snapshot of the block devices (if backed with LVM) so the guest is
just paused for a couple of seconds. Once this is done, the guest is
resumed, and the script starts to dump the snapshot.

Once a backup is finished, you'll have several files in the backup
directory. Let's take an example with a VM called my_vm which has
two virtual disks: hda and hdb. You have passed the --state flag:
 * my_vm.lock: lock file to prevent another backup to run at the same time
 * my_vm.xml: this file is the XML description of the VM (for libvirt configuraiton)
 * my_vm_hda.img: this file is an image of the hda drive of the guest
 * my_vm_hdb.img: this file is an image of the hdb drive of the guest
 * my_vm.state: this is a dump of the memory (result of virsh save my_vm my_vm.state)

This script was made to be ran with BackupPC pre/post commands.
In pre-backup, you dump everything (or mount as a set of chunks), then, backuppc backups,
compress, pools etc... the dumped file. Eventually, when the backup is finished
The script is called with the --action=cleanup flag, which cleanups everything.
(remove the temporary files, umount the fuse mount points if any etc.)

Some examples:

Backup the VM named mail01 and devsrv. Also dump the memory.
Exclude any virtual disk attached as vdb or hdb and on the fly
compress the dumped disks (uses gzip by default)

virt-backup --dump --vm=mail01,devsrv --state --exclude=vdb,hdb --compress


Remove all the files related to mail01 VM in the backup directory

virt-backup --cleanup --vm=mail01


Backup devsrv, use 10G for LVM snapshots (if available), do not dump the memory
(the guest will just be paused while we take a snapshot)
Keep the lock file present after the dump

virt-backup --dump --vm=devsrv --snapsize=10G --keep-lock


Backup devsrv, and disable LVM snapshots

virt-backup --dump --vm=devsrv --no-snapshot


Backup mail01, and enable debug (verbose output)

virt-backup --dump --vm=mail01 --debug


Backup winprd, shutdown before taking the backup
Dont wait more than 5 minutes for the shutdown to complete
restart the VM once the backup is finished

virt-backup --action=dump --shutdown --shutdown-timeout=300 --vm=winprd


Don't dump, but mount as a set of chunks the disks of vm mail01

virt-backup --action=chunkmount --vm=mail01

The idea here is to expose the big blocks/files which represent the VM disks
as small chunks (default is 256kB), then, you can use your favorite backup script/software
to backup /var/lib/libvirt/backup/vm_name/ where you want
This lets you create incremential backups of VM disks, which can save
a lot of space, a lot of bandwidth, and will also be much more efficient
with rsync based backup scripts (because rsync doesn't handle huge files very well
but if very efficient with a lot of small files)
The cleanup routine (--cleanup or --action=cleanup) will unmount all
the chunkfs mount points