>> what do you think of "qvm-copy-to-vm backupvm ." followed by rdiff-backup on the backupvm to luks encrypted disks?
>
> It's better, but personally I wouldn't do that either.
>
>> if you were using qubes-backup, how would you restore a single file or folder?
>
> Restore selected VM (under another name - it's done automatically),
> copy that single file to original VM, then remove restored VM.
How large would the attack surface be if I create a huge .img container
(50% of diskspace), mount it in dom0, do an rsync of all app-vm data
onto it, then mount it in my backup-vm for the actual remote backup?
Even if the backup-vm was compromised, all malicious changes _in_ the
.img container would be overwritten by the next rsync.
I am unsure if "sharing" the blockdevice-metadata (partitiontable etc)
is such a high risk?
Also, as dom0 and the backup-vm don't see any userdata, but only the
other vms .img files, this should be pretty safe?
For me, it would be nice as the backup-vm handles all backup-logic, can
do incremental backups, and there is almost no backchannel from
backup-vm to dom0.
Of course, as soon as my backup-vm or remote backup target is
compromised, I have a huge problem anyway. At least some (vault) data
would always be encrypted (by the regular qubes procedure), and would
necessarily be full-backupped every time.
In general, availability of my data is more important to me than
privacy. I'm still trying to achieve both, though :-)
N2
p.s.:
Please let me know if generally I should leave single emailadresses in
CC, I removed all but the list itself.