On 2021-07-10, at 12:07, Ludwig Seitz <[email protected]> wrote:
> 
> I can remove the text entirely, since we don't seem to agree on the details. 
> Would that be acceptable?

I can’t answer that question, but it seems to me that we both have a 
requirement in mind.  The text that resulted from processing my version is:

> Profiles are expected to prepare for being combined with others by clearly 
> specifying their security requirements.

While you had:

> The security of a profile MUST NOT depend on the assumption that this profile 
> is used in all steps of the authorization flow (C-AS, C-RS, RS-AS). 

The difference is that you want to impose a requirement on all profiles, a 
requirement that I believe is hard to provably fulfill in general.
While I was mainly putting out a requirement that profiles document their 
properties in this space, which is both more practical and more general.

Maybe we can combine these two into one sentence that covers a common 
requirement?

Grüße, Carsten



> /Ludwig
> 
> Sent from my smartphone
> 
> 
> ---- Carsten Bormann wrote ----
> 
> How do we get this done before Monday’s I-D deadline?
> 
> On 2021-07-06, at 08:22, Ludwig Seitz <[email protected]> wrote:
> > 
> > Hello Francesca, Carsten,
> > 
> > Sorry but I do not like what you did in the first sentence. Combining 
> > profiles does not necessarily equate to creating a new one, and I still 
> > don't see why we should needlessly request that it be so.
> 
> The devil is in the details. If we have pairs profiles that combine, the 
> component profiles should say so and say how.
> 
> > Given an example use case where a client talks dtls-profile to the AS and 
> > gets token and parameters for an oscore-profile back for the client-RS leg, 
> > why should there be a need for a new profile to support this?
> 
> Because the interaction e.g. with the DTLS and OSCORE/LAKE security setup has 
> details that need to be covered in the component profiles.
> 
> > I also do not like that you removed the requirement to design profiles so 
> > that the security for the different legs of the communication (C-AS, C-RS, 
> > RS-AS) stands on its own.
> 
> Some of the security can stand on its own, but the overall security derives 
> from the properties of the legs combining.
> 
> > What could happen now is that someone designs clever protocol foo that has 
> > a dependency between the C-AS and the C-RS communication for its security, 
> > and thus breaks when it is used on only one leg of this communication. I 
> > don't think you need to know all possible future profiles to design yours 
> > to be secure in that way. Note that the framework puts requirements on the 
> > security of future profiles, so you can assume e.g., that communication 
> > will be secure.
> 
> We may be better off writing a separate document that explains how to exactly 
> do this mixing-and-matching. 
> 
> I’d like to hear from others how they see this issue.
> 
> Grüße, Carsten
> 
> _______________________________________________
> Ace mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/ace

_______________________________________________
Ace mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/ace

Reply via email to