Re: [Cfrg] Requirements for curve candidate evaluation update

On Tue, Aug 12, 2014 at 2:58 PM, Benjamin Black <b@b3k.us> wrote:
> The CFRG, while part of IRTF, cannot separate itself from the practical
> demands of IETF. The CFRG brings a different set of skills to bear, but the
> outcome of its work must be pragmatic. The original list of requirements in
> the request forwarded by Kenny reflected the required pragmatism in asking
> only for new curves based on the 15 years of research and deployment
> experience with the NIST curves. Every other aspect of the problem was and
> is constrained by the existing RFCs, existing installed base, and existing
> customer expectations.
>
> Returning to the engineering requirements for new curves, the curves must
> take into consideration the well-established fact that many implementations
> will not be authored by professional cryptographers and almost no users of
> systems relying on them will be professional cryptographers. If we require,
> rather than allow, multiple curve models be implemented to support different
> algorithms or for different security levels, we are increasing complexity
> and hence opportunities for security and interoperability errors. If we
> require new security levels, we are again working against the existing
> protocols, implementation and deployment experience, customer expectations,
> even user interfaces. None of these can simply be dismissed because they are
> inconvenient. They are the facts on the ground.
>
> 1. Curves must be generated using a rigid process, and the more rigid the
> better.
>
> 2.  The curves we recommend should only require implementation of a single
> curve model, which implies:
>
>   a) ECDHE and ECDSA must be supported with a single curve model.
>
>   b) The same curve model must be used for all security levels.

This doesn't distinguish among curves. Read DJB's email "curves,
coordinates and computations" email. It's a question about what
protocols should do: why should we prohibit x-coordinate only
exchanges, if protocols see a security or bandwidth benefit to them?

>
> 3. For the purposes of evaluating candidates, ephemeral in ECDHE means used
> exactly once per key exchange.

How does this help distinguish between curves?

>
> 4. The security levels are 128, 192, and 256 bits and each curve will only
> be evaluated at one of those levels.

Well, this is not what the NIST curves are (P521, not P512). Nor does
Curve25519 fit this model, and yet it continues to be adopted, despite
being against "existing  protocols, implementation and deployment
experience, customer expectations, and even user interfaces". Of
course, what does evaluation at a level mean? Will people get upset if
we give them more security for the same speed? What about the
cofactors? Should we complain about those also?

>
> 5. No changes to coordinate wire formats.

Does this mean just Weierstrass? Or is a change to Edwards somehow not
a change in the wire format? (Of course this doesn't constrain the
curves either)
>
> The outcome of these is that we need 3 rigidly generated  curves in either
> short Weierstrass or twisted Edwards form. Anything beyond that should be
> addressed at a later point, and should not block progress on this specific
> request for new curves.

Except the Microsoft curves send points on a twisted Edwards curve
with a=-1, and q=3 (4), so aren't complete without using an isogeny
and a slower formula. This is a much more serious problem, as it makes
implementation quite a bit more complicated. If we assume implementors
don't know crypto, then complete formulas are very necessary.

Lastly, the requests on the table for TLS and SSH do change the point
format and use new algorithms: clearly these facts aren't actually
facts.

Sincerely,
Watson Ladd
>
>
> b
>
> _______________________________________________
> Cfrg mailing list
> Cfrg@irtf.org
> http://www.irtf.org/mailman/listinfo/cfrg
>

-- 
"Those who would give up Essential Liberty to purchase a little
Temporary Safety deserve neither  Liberty nor Safety."
-- Benjamin Franklin