> On 20 Jan 2026, at 8:45 PM, Peter Krempa <[email protected]> wrote:
> 
> !-------------------------------------------------------------------|
>  CAUTION: External Email
> 
> |-------------------------------------------------------------------!
> 
> On Mon, Jan 19, 2026 at 14:08:29 +0000, Tejus GK wrote:
>>> On 14 Jan 2026, at 9:00 PM, Peter Krempa <[email protected]> wrote:
>>> 
>>> !-------------------------------------------------------------------|
>>> CAUTION: External Email
>>> 
>>> |-------------------------------------------------------------------!
> 
> Oh no!
> 
 Hah!

> 
>>> 
>>>> 
>>>> Hence, virDomainFDStore is there to store FDs associated with the “name” 
>>>> key, rather than the domain.
>>> 
>>> Okay but that will make things much more complex due to our ACL stuff.
>>> 
>>> Allowing to pass any FD means that any user (even those who can't see
>>> the VM) could pass a FD and theoretically siphon data off from it.
>>> 
>>> I don't think we want to have a 'global' FD pool to pass. In case of
>>> migration I'd either suggest to define the VM (with possibly a dummy
>>> definition) just to have the domain object available. Alternatively we
>>> should have a migration API allowing FD passing.
>> 
>> Hi Peter, 
>> Thank you for the suggestion! I wanted to clarify that my design of 
>> maintaining a global pool of FDs, was so to seamlessly support the use case 
>> for managed P2P migrations. 
>> 
>>> I'd either suggest to define the VM (with possibly a dummy definition)
>> Is this workflow supported in P2P migrations? Also, we would have to at some 
>> point of time, blow up the VM definition to match the one on the source, 
>> does libvirt have support for that?
> 
> Yes, ...
> 
>>> Alternatively we should have a migration API allowing FD passing.
>> Hmm, even if we go ahead and implement that, I’m wondering how to make it 
>> work for P2P managed migrations. Since, source libvirt won’t be able to pass 
>> on FDs to the destination host. Thoughts?
> 
> you need to open another new client connection on the destination and
> pass the FDs. In fact that's requrired since FD passing works only
> locally and the remote daemon is thus unable to do that.
> 
> For now you can test it for FD-passed disks, where prior to migration
> you open a new connection and pass the FDs then kick off the migration
> 
> This is a complex deployment where you do need to set up stuff on the
> destination manually so IMO this is a reasonable trade-off.
> 

Hi Peter, thanks for the dummy VM suggestion, I gave it a spin, and it seems to 
work! I’ll rework the patch-series keeping the approach in mind, and share it 
on the list soon. 

Regards,
Tejus

Reply via email to