Re: [TLS] DSA should die

[Nico Williams <[e-mail]>]

On Thursday, April 2, 2015, Watson Ladd <[e-mail]> wrote:

> On Wed, Apr 1, 2015 at 7:17 PM, Martin Rex <[e-mail] <javascript:;>>
> wrote:
> > Dave Garrett wrote:
> >> On Wednesday, April 01, 2015 08:40:40 pm Stephen Farrell wrote:
> >>> <no-hats-except-the-330+-ciphersuites-is-crap-hat>
> >>>
> >>> Here's a suggestion: why pick 'em off one by one? How about
> >>> creating a new registry that only includes stuff we think is
> >>> really good for TLS1.3?
> >>
> >> I don't think a whole new registry is a good idea. The ClientHello
> >> has to stay the same for backwards compatibility, and cipher suites
> >> will need to be listed for TLS 1.2 compatibility at least.
> >> Creating a whole new system is only good if you can completely
> >> ignore the old one.
> >>
> >> There's plenty of space in the current registry if you want to
> >> start over. Just say 0xD000 and up is for TLS2 suites, and all
> >> below are TLS1. (at this point it's definitely time for 2.0)
> >> Define TLS2 codepoints for valid combinations for TLS2 and the TLS1
> >> would be invalid to negotiate for TLS2 but in there for TLS1.
> >
> > I agree that ClientHello needs to remain backwards-compatible.
> >
> > But how about using the Cipher Suites registry in a more creative
> fashion.
> >
> > For TLSv1.3, we could do the negotiation through he cipher suites list
> > more like this:
> >
> >    0x10,0xXX    specifies a key exchange algorithm (XX) 256 codepoints
> >    0x12,0xYY    specifies an authentication algorithm (YY) 256 codepoints
> >    0x14,0xZZ    specifies a symmetric encryption scheme (ZZ) 256
> codepoints
> >    0x16,0xQQ    specifies a mac algorithm (QQ) 256 codepoints
> >    0x18,0xPP    specifies a PRF algorithm (PP) 256 codepoints
> >
> > (with a little room in between if we ever exceed the 256 codepoints)
>
> I don't understand why we want a different mechanism for TLS 1.3 and
> TLS 1.2 ciphersuite negotiation. Even worse, I don't see how something
> like SRP fits cleanly into the scheme above. Yes, SRP is little-used,
> but people seem to think it's important.
>
> What's wrong with the existing mechanism, if we remove things we don't
> want?


For one: some combinations that are useful get lost in the shuffle because
there are so many ciphersuites.  It has happened before and it will happen
again.  A la carte combination is a useful compression function for: us,
operators, and the network.  It's a compression function that leaves us
with the essential information that we really need.

There are two separate proposals here: a la carte registration, and a la
carte negotiation on the wire.  The latter is only an on-the-wire
compression function -- perhaps it's too disruptive, though now is the only
time it could happen.  There is no excuse for not doing the former.

Nico
--