Am 18.06.25 um 12:25 schrieb Fiona Ebner:
> Same rationale as with upstream QEMU commit 5aaac46793 ("migration:
> savevm: consult migration blockers"), migration and (async) snapshot
> are essentially the same operation and thus snapshot also needs to
> check for migration blockers. For example, this catches passed-through
> PCI devices, where the driver does not support migration and VirtIO-GL
> display, which also does not support migration yet.
> 
> In the case of VirtIO-GL, there were crashes [0].
> 
> However, the commit notes:
> 
>> There is really no difference between live migration and savevm, except
>> that savevm does not require bdrv_invalidate_cache to be implemented
>> by all disks.  However, it is unlikely that savevm is used with anything
>> except qcow2 disks, so the penalty is small and worth the improvement
>> in catching bad usage of savevm.
> 
> and for Proxmox VE, suspend-to-disk with VMDK does use savevm-async
> and would be broken by simply using migration_is_blocked(). To keep
> this working, introduce a new helper that filters blockers with the
> prefix used by the VMDK migration blocker.
> 
> The function qemu_savevm_state_blocked() is called as part of
> savevm_async_is_blocked() so no check is lost with this
> patch. The helper is declared in migration/migration.c to be able to
> access the 'migration_blockers'.
> 
> The VMDK blocker message is declared via a '#define', because using a
> 'const char*' led to the linker to complain about multiple
> declarations. The message does not include the reference to the block
> node anymore, but users can still easily find a VMDK disk in the VM
> configuration.
> 
> Note, this also "breaks" snapshot and hibernate with VNC clipboard by
> preventing it. Previously, this would "work", because the Proxmox VE
> API has no check yet, but the clipboard will be broken after rollback,
> in the sense that it cannot be used anymore, not just lost contents.
> So some users might consider adding the check here a breaking change
> even if it's technically correct to prevent snapshot and hibernate
> with VNC clipboard. But other users might rightfully complain about
> broken clipboard. And again, the check also prevents blockers from
> passed-through PCI devices, etc. so it seems worth tolerating that
> breakage.
> 
> [0]: https://forum.proxmox.com/threads/136976/
> 
> Signed-off-by: Fiona Ebner <f.eb...@proxmox.com>
> ---
> 
> It's still worth adding a front-end check for more things to catch
> failure early, but there should be a full check in the backend to
> prevent crashes and for future-proofing.
> 
> Previous iteration:
> https://lore.proxmox.com/pve-devel/20240911140910.190670-1-f.eb...@proxmox.com/
> 
> Changes in v3:
> * Rebase.

Ping


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

Reply via email to