Re: [TLS] Adding a new signature scheme

[Simon Josefsson <[e-mail]>]

Ilari Liusvaara <[e-mail]> writes:

> This is about adding a new signature primitive (such as the
> (eventual) CFRG scheme).

I believe these aspects were already discussed in the EdDSA threads.  It
would be useful to establish consensus around these topics.

> There are basically two issues:
>
> 1) Do we allocate new ciphersuite codepoints or not? So far
> each certificate algorithm in ciphersuite has corresponded
> to only one signature algorithm.
>
> Implications of new codepoints:
> - More ciphersuites (about 11).
> - Needs TLS 1.2+ anyway, because the ciphers are presumably
>   AEAD mode and those need TLS 1.2+.
> - Keep existing semantics.
>
> Implication of reusing ECDSA codepoints.
> - No new ciphersuites
> - Needs TLS 1.2+
> - Redefines existing semantics a bit.

The feedback I got on draft-josefsson-tls-eddsa from several people was
to reuse the ECDSA codepoints instead of defining new ciphersuites.  I
am currently (mildly) in favor of that, as it appears to imply less work
for everyone.  I have since published draft-josefsson-tls-eddsa2 that
adopts that approach.  I decided to fork the draft so both are available
and can be evaluated on its own.  They are available here:

https://tools.ietf.org/html/draft-josefsson-tls-eddsa
https://tools.ietf.org/html/draft-josefsson-tls-eddsa2

> 2) What does the SignatureAndHashAlgorithm.hash mean exactly? The
> TLS 1.2 RFC isn't specific enough.

I believe the document introducing a new signature primitive must be
clear about this.  I don't believe TLS 1.2 mandates any particular
interpretation here, so it is up to the new primitive to define
semantics.

I believe the choice is not particular interesting, as long as the
specification is clear about what is intended.  For example,
draft-josefsson-tls-eddsa uses "none", and draft-josefsson-tls-eddsa2
uses "sha512", and both are in their own way a bit redundant.

> I see two choices:
> a) The field sets hash (prehash) to perform before passing the
>    hash to be signed (the description of "none" and description of
>    Certificate Verify message hints at this interpretation).
> b) The field parameterizes the signature scheme in some scheme-
>    specific way (but what would "none" mean in context of this,
>    the scheme having no hash parameters?).
>
> All the existing schemes are compatible with interpratation a).
>
> But for proposals for CFRG scheme: All proposed schemes have a
> hash parameter (the only one or one of two) that is incompatible
> with interpretation a):
>
> 1) Only one hash that does not prehash the message.
> 2) Two hash functions, one prehash (can be identity) and
>    one internal hash. Prehash can be chosen per-signature.
> 3) Two hash functions, one prehash (can be identity) and
>    one internal hash. Prehash is not indicated, so one
>    has to be careful changing it.
>
> So if interpretation a) is the correct one, one presumably has to
> fix the internal hash in signature scheme codepoint (E.g. to
> SHA-512 for signature scheme #4).

I don't think we can come any closer to my discussion above until we
know what the CFRG outcome is.

/Simon