Re: [PATCH v2] drivers/virt: vmgenid: add vm generation id driver

[Dmitry Safonov]

Hello,

+Cc Eric, Adrian

On 11/19/20 6:36 PM, Alexander Graf wrote:
> On 19.11.20 18:38, Mike Rapoport wrote:
>> On Thu, Nov 19, 2020 at 01:51:18PM +0100, Alexander Graf wrote:
>>> On 19.11.20 13:02, Christian Borntraeger wrote:
>>>> On 16.11.20 16:34, Catangiu, Adrian Costin wrote:
>>>>> - Background
>>>>>
>>>>> The VM Generation ID is a feature defined by Microsoft (paper:
>>>>> http://go.microsoft.com/fwlink/?LinkId=260709) and supported by
>>>>> multiple hypervisor vendors.
>>>>>
>>>>> The feature is required in virtualized environments by apps that work
>>>>> with local copies/caches of world-unique data such as random values,
>>>>> uuids, monotonically increasing counters, etc.
>>>>> Such apps can be negatively affected by VM snapshotting when the VM
>>>>> is either cloned or returned to an earlier point in time.
>>>>>
>>>>> The VM Generation ID is a simple concept meant to alleviate the issue
>>>>> by providing a unique ID that changes each time the VM is restored
>>>>> from a snapshot. The hw provided UUID value can be used to
>>>>> differentiate between VMs or different generations of the same VM.
>>>>>
>>>>> - Problem
>>>>>
>>>>> The VM Generation ID is exposed through an ACPI device by multiple
>>>>> hypervisor vendors but neither the vendors or upstream Linux have no
>>>>> default driver for it leaving users to fend for themselves.
[..]

>>> The only piece where I'm unsure is how this will interact with CRIU.
>>
>> To C/R applications that use /dev/vmgenid CRIU need to be aware of it.
>> Checkpointing and restoring withing the same "VM generation" shouldn't be
>> a problem, but IMHO, making restore work after genid bump could be
>> challenging.
>>
>> Alex, what scenario involving CRIU did you have in mind?
> 
> You can in theory run into the same situation with containers that this
> patch is solving for virtual machines. You could for example do a
> snapshot of a prewarmed Java runtime with CRIU to get full JIT speeds
> starting from the first request.
> 
> That however means you run into the problem of predictable randomness
> again.
> 
>>
>>> Can containers emulate ioctls and device nodes?
>>
>> Containers do not emulate ioctls but they can have /dev/vmgenid inside
>> the container, so applications can use it the same way as outside the
>> container.
> 
> Hm. I suppose we could add a CAP_ADMIN ioctl interface to /dev/vmgenid
> (when container people get to the point of needing it) that sets the
> generation to "at least X". That way on restore, you could just call
> that with "generation at snapshot"+1.
> 
> That also means we need to have this interface available without virtual
> machines then though, right?

Sounds like a good idea.
I guess, genvmid can be global on host, rather than per-userns or
per-process for simplicity. Later if somebody will have a bottleneck on
restore when every process on the machine wakes up from read() it could
be virtualized, but doing it now sounds too early.

ioctl() probably should go under
checkpoint_restore_ns_capable(current_user_ns()), rather than
CAP_SYS_ADMIN (I believe it should be safe from DOS as only CRIU should
run with this capability, but worth to document this).

Thanks,
         Dmitry