[pve-devel] bug in backup/restore for non-64K aligned disks + host mem leak to backups

[Roland Kammerer]

On Wed, Jun 17, 2020 at 06:03:25PM +0200, Dietmar Maurer wrote:
> > I initially noticed that backup/restore has a bug when the disk size is
> > not a multiple of 64K for DRBD devices. 
> 
> But qemu block size is 64K, so disk size should/must always be a multiple of 64K?

hm, I think this depends a bit from which standpoint you are coming from.

>From the qemu point of view what I saw in qemu itself - with a very very
limited knowledge - I don't see such a limitation. Looks like there are
qemu/block drivers that handle subsector/single byte level.

And one, in general, can always pass a host device as oddly shaped as I
want to qemu.

> AFAIK we never generate such disk sizes?

Again, point of view I guess and what you define as okay for a user to
do and what "we" is. Is it okay for a user to use the linstor-proxmox
(== DRBD) plugin?
Then this bug happens.
Is it okay for a user to use another plugin that generates "oddly shaped
devices"?
Then this bug happens.
Is it okay that the user does this weird operations on plain LVM I
showed as a non-DRBD reproducer?
Then this bug happens.

Don't get me wrong, I just wanted to report what IMO is a bug in a
backup tool (an actually pretty clever one looking at the format), and
that to me is the point of view that really matters here. There is a
backup tool. Let's compare it to "dd", another backup tool. "dd" is
perfectly fine to backup and restore such "weird" (IMO they aren't)
disks. And then there is the one on Proxmox that fails on restore, so I
as an admin don't even know until it is too late, and leaks host memory
to backups. Wouldn't it be easier to just fix this tool?

Best, rck