On Mon, Apr 17, 2023 at 05:23:54AM +0900, AJITOMI Daisuke wrote:
> 
> > I think c) is a really bad idea.
> 
> I understand your preference. However, as I mentioned in the previous post
> to Orie, I think defining all of alg values is not such a bad idea.
> All of the other HPKE modes are well-defined in RFC9180 and I think that
> they don't affect the definition of KEM key representation and hkc
> structure.

The problem with this is that any alg registered SHALL be defined. And
this includes any security considerations.

>From the relevant BCP laying out the requirements:

"<...> the values and their meanings must be documented in <...>
specification, in sufficient detail so that interoperability between
independent implementations is possible."

For COSE, what this means is that in order to define the alg values,
COSE-HPKE MUST be extended to all four modes. Which is something that
COSE-HPKE draft itself seems unwilling to do.

For JOSE, the implications are even worse. It means that the entiere
JOSE-HPKE MUST be defined, a task harder than defining COSE-HPKE. One
can not assume that the alg definitions will be the same. 

As examples of differences between use of algorithms in COSE-HPKE and
JOSE-HPKE, I think that (after the breakthrough in JOSE-HPKE I was
waiting for):

- JOSE-HPKE requires two algorithms for the base mode, in contrast to
  COSE-HPKE, that only has one.
- Auth mode can not be used at all in JOSE-HPKE, due to vulnerability
  that is unfixable without extending JWE in highly nontrivial ways.


And there are very much nontrivial security considerations. For
example, I think the same vulnerability as above requires restricting
auth modes in COSE-HPKE to just COSE_Encrypt0.




-Ilari

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

Reply via email to