Re: [TLS] Updated EdDSA/Ed25519 PKIX document

[Ilari Liusvaara <[e-mail]>]

On Wed, Sep 23, 2015 at 10:33:29AM +0200, Simon Josefsson wrote:
> Hi all,
> 
> I have pushed out a new version of the document describing EdDSA public
> keys, signatures and certificates for PKIX.  The change in -03 include
> the addition of the prehash mode, test vectors generated by GnuTLS, and
> a section recommending certain human readable names.
> 
> https://tools.ietf.org/html/draft-josefsson-pkix-eddsa-03
> 
> I've started a thread to discuss whether it is wortwhile to be able to
> use the same Ed25519 key for both PureEdDSA mode and HashEdDSA signing,
> and I'd appreciate feedback on whether people are interested in this and
> generally if it is a good idea or not.  The complexity involved make me
> shy away a bit from it, but it is fun to consider.  The thread is here:
> https://moderncrypto.org/mail-archive/curves/2015/000630.html

Some comments:

1) Taking private key example and decoding the ASN.1 yields:

  0  37: SEQUENCE {
  2   1:   ENUMERATED 1
  5  32:   OCTET STRING
       :     D4 EE 72 DB F9 13 58 4A D5 B6 D8 F1 F7 69 F8 AD
       :     3A FE 7C 28 CB F1 D4 FB E0 97 A8 8F 44 75 58 42
       :   }

What is that ENUMERATED 1 for? The text description has algorithm of 
SHA512-Ed25519, which would AFAICT be ENUMERATED 2 if private keys use
the same kind enumeration as public keys.


2) AFAICT, EdDSA-20150704 calls the signature without prehash, with
25519 curve and SHA-512 hash "Ed25519-SHA-512" (and the corresponding
version with SHA-512 prehash "SHA-512-Ed25519-SHA-512").

The first of those is AFAIK the same as the old "Ed25519" signature
primitive.

One problem with those names is that those things might get rather
nasty to read when '-' both appears inside hash names and component
separator.


3) Using the same key for both PureEdDSA and HashEdDSA sounds like a
bad idea. However, it would be technically possible to alter the
internal subkey derivation so that different prehashes give logically
separate keys.

Such change would only affect signing, not verification (since verifier
sees the change as different public key).


4) For TLS PoP signatures, does it make sense to use HashEdDSA at all?
Another way would to always use PureEdDSA and perform hash separtion
from TLS side (e.g. sign(privkey, hash_func_id|H(tbs_data))).

The certificate signatures are different matter tho, since CAs use
HSMs for signing (those HSMs tend to be rather beefy, but still).

Then there are CRLs. I would imagine those are so large that one rather
uses prehashing.


-Ilari