Re: [PATCH v2 02/17] xsm/silo: Support hardware & xenstore domains

[Jan Beulich <jbeulich@xxxxxxxx>]

On 30.07.2025 23:05, Jason Andryuk wrote:
> On 2025-07-30 06:17, Jan Beulich wrote:
>> On 16.07.2025 23:14, Jason Andryuk wrote:
>>> In a disaggregated environment, dom0 is split into Control, Hardware,
>>> and Xenstore domains, along with domUs.
>>
>> Here we are with terminology again. In a truly disaggregated env, yet
>> more (service) domains would come into play. What you mean here is
>> only coarse disaggregation, as you're trying to get away without using
>> Flask.
> 
> If disaggregation only means, fine grain disaggregation, then I'm not 
> sure how to differentiate coarse.  I could write "split control and 
> hardware domain".

IMO that would be better here.

>>>   The is_control_domain() check
>>> is not sufficient to handle all these cases.  Add is_priv_domain() to
>>> support allowing for the various domains.
>>>
>>> The purpose of SILO mode is to prevent domUs from interacting with each
>>> other.  But dom0 was allowed to communicate with domUs to provide
>>> services.
>>>
>>> To provide xenstore connections, the Xenstore domain must be allowed to
>>> connect via grants and event channels.  Xenstore domain must also be
>>> allowed to connect to Control and Hardware to provide xenstore to them.
>>>
>>> Hardware domain will provide PV devices to domains, so it must be
>>> allowed to connect to domains.
>>>
>>> That leaves Control.  Xenstore and Hardware would already allow access
>>> to Control, so it can obtain services that way.  Control should be
>>> "privileged", which would mean it can make the connections.  But with
>>> Xenstore and Hardware providing their services to domUs, there may not
>>> be a reason to allow Control to use grants or event channels with domUs.
>>
>> "may not be" is too weak for my taste to forbid such.
> 
> I can't come up with a concrete example of why Control needs to directly 
> communicate with a domU.  Originally I allowed it, but it was your 
> previous feedback which made me remove Control.
> 
> I don't have a strong opinion on the handling of Control.  I can see it 
> argued either way.

My comment was merely on the wording: If Control is to be excluded, the
wording imo wants to be firm.

>>> This silo check is for grants, event channels and argo.  The dummy
>>> policy handles other calls, so Hardware is prevented from foreign
>>> mapping Control's memory with that.
>>
>> By "foreign mapping" you only mean what would result in p2m_foreign
>> entries? But grant mapping is okay?
> 
> Yes, "foreign mapping" = p2m_foreign.
> 
> Using grants requires explicit actions on both sides. 
> silo_mode_dom_check() passing still requires action by both sides to 
> establish a communication channel.
> 
> This is different from a foreign mapping, which is a unilateral action 
> by the privileged side.

Right, yet protecting from interference may want to extend to Control
e.g. having some forgotten code path installing a grant entry, which
some other domain may then map.

>>> @@ -29,8 +40,8 @@ static bool silo_mode_dom_check(const struct domain *ldom,
>>>   {
>>>       const struct domain *currd = current->domain;
>>>   
>>> -    return (is_control_domain(currd) || is_control_domain(ldom) ||
>>> -            is_control_domain(rdom) || ldom == rdom);
>>> +    return (is_priv_domain(currd) || is_priv_domain(ldom) ||
>>> +            is_priv_domain(rdom) || ldom == rdom);
>>>   }
>>
>> IOW we're turning by 180°? Interesting ...
> 
> While the previous code is written "is_control_domain()" its use in silo 
> mode is really "is_dom0()".
> 
> (Working on this, I've been thinking about how dom0 is like the god 
> Janus with two faces.  A single entity with two faces.  We use 
> is_control_domain() and is_hardware_domain(), but often it is just a 
> different name for the single dom0)
> 
> As stated above, domU <-> Xenstore and domU <-> Hardware (for PV 
> devices) are needed to allow those services.  So while it looks like a 
> 180°, maybe is_control_domain() was the wrong name for dom0?

The mix was - imo - a direct result of the boundaries never having been
drawn properly. That's necessary now, first in an abstract way, then
carried through to actual code. (Right here my remark really was just a
remark anyway, because the move looked particularly heavy on the
surface.)

Jan