Re: [Curdle] State of draft-ietf-curdle-ssh-kex-sha2?

[denis bider <[e-mail]>]

I think Ron's suggestions are all reasonable.

On Sun, Jul 12, 2020 at 11:17 AM Ron Frederick <[e-mail]> wrote:

>
> On Jul 12, 2020, at 5:05 AM, denis bider <[e-mail]>
> wrote:
>
> It looks like I found the key thing I missed: Google did not show me RFC
> 8268 when I searched for "diffie-hellman-group14-sha256". It finds it now
> when I search for "rfc diffie-hellman-group14-sha256".
>
> I could have found it in the IANA registry but I trusted the lack of
> Google results in my first search. Alas.
>
> That resolves the key concern for me: that names like "diffie-hellman-group14-sha256"
> are authoritatively defined. I was looking for this reference for the SSH
> QUIC draft.
>
> With regard to MUST / MAY / SHOULD / etc assignments, I agree with all of
> these:
>
> > curve25519-sha256                     SHOULD
> > curve448-sha512                       MAY
> > diffie-hellman-group-exchange-sha1    SHOULD NOT
> > diffie-hellman-group-exchange-sha256  MAY
> > diffie-hellman-group1-sha1            SHOULD NOT
> > diffie-hellman-group14-sha1           SHOULD NOT
> > diffie-hellman-group14-sha256         SHOULD
> > diffie-hellman-group15-sha256         MAY
>
> I do not agree with this one:
>
> > diffie-hellman-group16-sha512   MUST
>
> I find this too computationally expensive to justify "MUST" for servers.
> Last time I checked, this costs about 100 ms in server CPU time, more on
> weaker CPUs, and makes it trivial to DoS a resource-constrained server - no
> DDoS needed.
>
>
> I’m good with group16-sha512 remaining a “MAY” along with group17-sha512
> and group18-sha512, What about group15-sha256, though? Should that perhaps
> be a “SHOULD”, and also group14-sha256 changed to a “MUST”? I think we want
> at least one DH algorithm to be listed as a “MUST”.
>
>
>
> I agree with all of these:
>
> > diffie-hellman-group17-sha512  MAY
> > diffie-hellman-group18-sha512  MAY
> > ecdh-sha2-*                    MAY
> > ecdh-sha2-nistp256             SHOULD
> > ecdh-sha2-nistp384             SHOULD
> > ecmqv-sha2                     MAY
> > ext-info-c                     SHOULD
> > ext-info-s                     SHOULD
> > gss-*                          MAY
>
> I conditionally agree on these:
>
> > gss-curve25519-sha256-*  SHOULD
> > gss-curve448-sha512-*    MAY
> > gss-gex-sha1-*           SHOULD NOT
> > gss-group1-sha1-*        SHOULD NOT
> > gss-group14-sha256-*     SHOULD
> > gss-group15-sha512-*     MAY
> > gss-group16-sha512-*     SHOULD
> > gss-group17-sha512-*     MAY
> > gss-group18-sha512-*     MAY
> > gss-nistp256-sha256-*    SHOULD
> > gss-nistp384-sha384-*    MAY
> > gss-nistp521-sha512-*    MAY
>
> The condition is that the reader understands that "SHOULD" for these
> methods applies only if you implement any GSS-API key exchange methods at
> all. In other words, the whole category of GSS-API needs to be "MAY", but
> within the category the above SHOULDs are appropriate as long as the
> decision is made to implement any of the GSS-API key exchange methods in
> the first place.
>
>
> With that condition in mind, I think all of these should match their
> non-GSS equivalents in terms of MAY/SHOULD/MUST. So, if changes are made to
> group14/15/16, they should be reflected here as well.
>
>
> I agree with these:
>
> > rsa1024-sha1             MUST NOT
> > rsa2048-sha256           MAY
>
> In summary - my main complaint is that "diffie-hellman-group16-sha512" is
> too expensive to be a MUST. If that leaves no key exchange method that is a
> "MUST", I'm fine with it.
>
> If there needs to be a method that's a "MUST", I vote for either
> "curve25519-sha256" or "diffie-hellman-group14-sha256" as meeting
> security requirements while being the most widely compatible.
>
>
> I like having curve25519-sha256 as a strongly encouraged SHOULD, but I
> think it might be better to stick with group14-sha256 as the MUST.
> --
> Ron Frederick
> [e-mail]
>
>
>
>