Re: [TLS] rm Hello.random w/ (EC)DHE [was OPTLS: Signature-less TLS 1.3]

[Nico Williams <[e-mail]>]

On Mon, Nov 10, 2014 at 05:58:12PM +0000, Dan Brown wrote:
> When using (EC)DHE (inc. OPTLS below), the TLS random nonces
> (ClientHello.random and ServerHello.random) seem redundant (unless one
> re-uses ephemeral secret keys (but then maybe a counter or secure time
> would be almost as good a nonce as a random in this case), or perhaps
> for some other reason I’m missing), and perhaps even a point of
> fragility (e.g. if a common generator for nonces and keys is weak, or
> corrupted). So, does it make sense to remove these Hello.randoms, when
> using (EC)DHE, or to replace them with something milder, like
> Hello.time, Hello.counter, or just Hello.nonce?

The main problem with the hello randoms today is that they are big.
This is what made Dual_EC deadly for TLS.  Besides, large nonces also
create a large covert channel.

Larger nonces make sense for plain PSK, static DH, and session
resumption.

For RSA key transport I don't see the point of larger nonces.

For (EC)DHE small nonces help the briefly reused server-side ephemeral
private key case, which is nice to have.

64-bit nonces should suffice for the key exchanges that need larger
nonces, no?

32-bit nonces should suffice for the key exchanges that need smaller or
no nonces.

The client should send a nonce of size matching the largest needed by
any of the ciphersuites it offers.

The server should send a nonce whose size is given by the selected
ciphersuite.

If variable size nonces are a problem (e.g., fingreprinting), then send
large nonces but set unneeded bits to one of a small set of suitable
constant patterns.

Nico
--