[openpgp] Ed25519 and digest choices (issue 31)

[Daniel Kahn Gillmor <[e-mail]>]

Over on https://gitlab.com/openpgp-wg/rfc4880bis/-/issues/31, jethrogb
writes:

> Appendix A contains an example for EdDSA. The example states that the
> hash function used is SHA2-256. The example also states that the curve
> used is 2b06010401da470f01, which is defined as “Ed25519” elsewhere in
> the draft. However, RFC 8032 specifies Ed25519 as an instantiation of
> EdDSA with specific parameters, one of which is that H is SHA2-512 and
> PH (in the ph case) is SHA2-512. Is it the intention that OpenPGP
> implements not Ed25519 but some other form of EdDSA? If yes, this
> should be called out explicitly in the text and it shouldn't be called
> Ed25519. If no, the example needs to be updated and it would probably
> be good to explicitly call out Ed25519ph in section 14.8.

How does the WG think this should be resolved?

I intend to sign this message with an EdDSA signature from a Curve25519
key, but it will likely use SHA2-256 as the OpenPGP digest choice (in
the EdDSA RFC 8032 framing, that would be the pre-hash "PH" parameter to
EdDSA).  This would mean that we are *not* using Ed25519ph, since
OpenPGP permits variance of the PH parameter.

One approach would be to clarify that OpenPGP signatures made with
Ed25519 SHOULD use SHA2-512 as the OpenPGP digest, which I believe would
align it with Ed25519ph.  But there would still be existing signatures
out there (like the one signing this message) which would use SHA2-256,
and it's hard to say that signature verifiers should reject those
signatures.

Alternately, maybe we should instead reframe OpenPGP's use of Ed25519 as
a "PureEdDSA" scheme that signs only the OpenPGP digest (not the signed
data directly).  That bypasses the "PH" parameter, but it also means
that any cryptanalsis that is applied to EdDSA isn't necessarily
applicable to OpenPGP, because we have this additional step involved.

Either way, it seems that we need to clarify the standard.

       --dkg