On Wed, May 31, 2023 at 05:44:02PM -0500, Orie Steele wrote:
> 
> On Wed, May 31, 2023 at 2:02 PM Hannes Tschofenig <[email protected]>
> wrote:
> 
> > Hi Orie,
> >
> > Thanks for your contribution. I am trying to understand your suggestion
> > better.
> >
> > When it comes to the algorithm registry there are only two choices:
> >
> >    - A la carte, or
> >    - Ciphersuites
> >
> > From your text below I believe you are arguing for a ciphersuite as Apple
> > uses it.
> >
> > For the initial versions of the COSE-HPKE we used the ciphersuite approach
> > and then we switched to the a la carte approach.
> >
> > Is it correct that you want to move back to the ciphersuite approach?
> >
> 
> Yes, I think "alg" should work the way it has previously worked, for
> both keys and envelopes.
> 
> I don't think a new "alg" system should be invented in COSE-HPKE.

It does work like it worked before.

Restricting bulk encryption using "alg" has never worked.


> > Once this decision has been taken, there is the question about where to
> > take the registered algorithm values from.
> >
> 
> I suggest requesting registrations only for the algorithm suites that
> people want to use in COSE-HPKE, here:
> 
> https://www.iana.org/assignments/cose/cose.xhtml

14 as of HPKE 2023-05-18, vulnerable to combinatorial blowup.


> This follows the conventions we have for ECDH alg's, which is what
> many people will be familiar with, and will potentially be upgrading
> from when considering COSE-HPKE.
> 
> This will also result in simpler keys and envelopes.

ECDH algs do not work that way. Which is why there are only 2
fundamental ECDH-ES algorithms, plus 3 size optimizations.

And as of currently, it is able to match all the pre-quantum stuff.


> > Version -05 of the draft takes the values directly from the IANA
> > HPKE registry. In an attempt to avoid creating conflicts with the
> > COSE algorithm registry, they had to be used in the newly designed
> > structure rather than in the “usual” COSE algorithm parameter.
> 
> Yes, I disagree with this design choice.

As note, there is need to define new stuff because of the ciphertext-
is-not-key problem.


> It is more contentious and requires more work, because it has to
> define this "new way of handling alg with an external registry".

I don't think it will be more work up-front. And it will definitely not
be more work in the long run.

Up-front, it is encoding three integers explicitly versus defining how
to extract those from alg values.

Long-run, it is doing nothing versus registering all the easily dozens
of new alg values standing for various triples (and hoping nobody gets
any "smart" ideas).


> This complexity is not coming from HPKE, it is coming from wanting
> to establish new conventions in COSE Keys and Envelopes that avoid
> registry updates related to new algorithms.

The complexity is coming from HPKE. And what I think is the simplest
way to handle it avoids registry updates.


> If it is going to be attempted, it would be better done in a
> separate document, as an update to COSE.

Attempted? I consider "attempted" to mean it might not work. This
stuff is proven to work.

And I do not see any need to update COSE.


> However, I would still object because as Chris said, it seems to go
> against the previous conventions, and current trends in security best
> practices.

It does not go against the previous conventions. 

And many "best practices" are actually just fads. Only some will truly
stand the test.
 



-Ilari

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

Reply via email to