Re: [PATCH-SERIES storage/qemu-server v3 00/11] fix #7094: external snapshot delete: ensure commit target is large enough

["Lukas Sichert" <l.sichert@proxmox.com>]

On 2026-05-19 13:06, Fiona Ebner <f.ebner@proxmox.com> wrote:

> Changes in v3:
> * rebase on latest master, need to bump APIAGE+VER one more
>
> Changes in v2:
> * rebase on latest master.
>
> For LVM storages using 'snapshot-as-volume-chain', it's necessary to
> ensure that the size for the LV of the target snapshot volume is large
> enough before doing a commit operation. This affects both the offline
> and the online case, which have different code paths.
>
> To make this possible via the storage (plugin) API, the
> volume_resize() method gains a new 'snapname' parameter and the
> 'virtual-size' of each snapshot is returned in the result of
> volume_snapshot_info().
>
> The series also adds documentation for volume_resize() and
> volume_snapshot_info(), the latter of which is a bit hairy, since it
> was previously only used for ZFS replication and then re-used for
> snapshot-as-volume-chain which returns quite different information.
> It's not too bad though.
>
> Lastly, there's some cleanups for the volume_snapshot_info() method.
>
>
> Storage plugin API age and version are bumped in the series.
> Dependency bump from qemu-server to libpve-storage-perl needed!


I reproduced the old offline failure on an lvm storage with
snapshot-as-volume-chain 1, backed by a thinly provisioned SAN LUN. The
repro was: create a first external snapshot, resize the active disk,
write data into the newly added region, then delete that first snapshot.
With the old code, the offline delete path fails with qemu-img commit
returning No space left on device.

I also tested the live path with the old code using the same general
setup. I did not get the same No space left on device failure there;
instead the block job was cancelled in my environment, so I could not
reproduce the live failure cleanly.

With the patch installed, the offline case succeeds as expected because
the target snapshot LV is resized before commit. I also checked lvmthin
for regressions, and normal qm snapshot, qm resize, and qm delsnapshot
behavior still works there.

Tested-by: Lukas Sichert <l.sichert@proxmox.com>
Reviewed-by: Lukas Sichert <l.sichert@proxmox.com>