Re: [Ace] Keeping the same key identifier for groups

[Jim Schaad <[e-mail]>]

-----Original Message-----
From: Ace <[e-mail]> On Behalf Of Ludwig Seitz
Sent: Tuesday, August 20, 2019 2:09 AM
To: [e-mail]
Subject: Re: [Ace] Keeping the same key identifier for groups

On 19/08/2019 22:40, Jim Schaad wrote:
> As Ludwig pointed out during the F2F, it makes far more sense to try 
> and keep an entity using the same key identifier for as long as 
> possible.  This is in part to make sure that signing keys do not need 
> to be retrieved if they can be easily cached.  In looking at this 
> deeper during my implementation I ended up with the following question:
> 
> The way that I have set things up in my implementation it is simple to 
> ensure that the same kid value is going to be used with the same CWT, 
> however it might make more sense to use the signing key as the 
> continuity identifier instead.  The issue that arises in this case is 
> that there might be two different active CWT objects that are 
> associated with the same signing key.  That is there are two CWTs but the same signing key was used
> while doing a join operation.   I already do some matching between different
> CWTs by assuming that if the bearer key in the CWT is the same then 
> they are sufficiently equivalent to threat them as the same.  This 
> lead to some interesting discussions in Montreal about if this meant 
> just the "secret" or if it meant all of the elements provided by the 
> AS which are used in the key derivation process.  (I have gone back 
> and forth on this and currently am sitting on the "just the secret" 
> side of the fence.)
> 
> Does anyone have any opinions?
> 
Could you please expand the explanation of your use case a bit?

Are you thinking of a scenario where someone would be using the counter-signature key from group-OSCORE as a proof-of-possession (pop) key in serveral CWTs?

What would these CWT authorize?

Why would an entity hold several CWTs for the same audience?

[JLS]  Trying to get a bit more detailed on the situations that I am thinking about.   At the moment I am working on group key manager code so both of these are going to be couched in those terms and not in general terms.

I have a CWT that says group X is fine for this entity.  The CWT expires or is going to expire soon.  The AS does not use the same pop-key in a newly issued CWT.  The CWT is posted to the Group KDC and the entity performs a join under that new CWT, but with the same signing key for group messaging.   Should this operation preserve the group kid assigned to that entity based on the signing key or should it be a new group kid based on the new CWT pop key.  That is Entity E1 using CWT1 and public key S1 received a GK-kid value for group G1.  E1 then posts CWT2, with a different pop-key, does a join request for G1 using the new pop-key and S1.  Does the AS return GK-kid or GK-kid2?

This can happen in any number of different situations:  Adding or removing group Y from the token, token renewal, reboot of somebody so state is lost, key exhaustion or merely a requirement from the AS that symmetric pop keys be rotated based on duration.



Side-note:
My stance on multiple CWTs linked to the same pop-key and for the same audience is that the latter one should supersede the previous ones.
Example: If you have a CWT authorizing A for audience Z and you now also need authorization B for audience Z, you should request a CWT for A+B for audience Z, that replaces your previous one.

[JLS]  I agree that this is the preferred way for these things to happen, that is either the request says use the same pop-key or the AS just uses the same pop-key because it knows that a similar CWT exists already.  However there are reasons why that cannot be done.  Easy example, my AS does not currently have any memory so it can never issue a new CWT with the same symmetric pop-key.  It can do it with the same asymmetric pop key.


jim

[JLS]  

/Ludwig




--
Ludwig Seitz, PhD
Security Lab, RISE
Phone [phone]