Re: [TLS] DTLS 1.3

[Eric Rescorla <[e-mail]>]

On Mon, Jul 4, 2016 at 1:46 PM, Ilari Liusvaara <[e-mail]>
wrote:

> On Mon, Jul 04, 2016 at 09:45:38PM +0200, Hannes Tschofenig wrote:
> > Hi Ilari,
> >
> > thanks for your super quick response.
> >
> > On 07/04/2016 04:03 PM, Ilari Liusvaara wrote:
> > > On Mon, Jul 04, 2016 at 12:21:22PM +0200, Hannes Tschofenig wrote:
> > >> Hi all,
> > >>
> > >> I have made an attempt to integrate DTLS 1.3 into the TLS 1.3 document
> > >> and you can find the result at
> https://github.com/tlswg/tls13-spec/pull/512
> > >>
> > >> I have worked on a prototype implementation of DTLS 1.3 and if someone
> > >> else has something working by the time of the Hackathon in Berlin
> please
> > >> let me know.
> > >
> > > Taking a look:
> > >
> > > - The version field is described as identical to TLS 1.3 version field.
> > >   What value is there actually? FCFE (wasn't that how "DTLS 1.0" was
> > >   encoded on wire)?
> >
> > For the record layer TLS 1.3 says that the version field is deprecated
> > and ignored. I guess the same approach applies to DTLS 1.3. So, it
> > probably doesn't matter what is really is. In DTLS 1.2 the version value
> > was 254.253.
>
> Oh, it was that. Then I presume DTLS 1.0 was 254,255 (yup, checked the
> RFC).
>
> > For the client_version and server_version in the handshake I thought it
> > would be fine to just re-use the TLS 1.3 version number.
>
> I think that is not a good idea, and one should follow the pattern (which
> would give 254,252 for ClientHello/ServerHello).
>
> But for record version, reusing previous versions (either 254,255 or
> 254,253) should be fine.
>
> > > - KeyUpdate does not work in DTLS. Might just use epochs for similar
> > >   purpose, and reserve first few epochs for special purposes.
> >
> > Could you explain in more detail why you think a key update does not
> > work with DTLS?
>
> That message is not idempotent (nor has it timely ACK).
>
> > > - The full handshake protocol is only run once. After that, there's
> > >   only rekeying, new tickets and dynamic reauth.
> >
> > How TLS 1.3/DTLS 1.3 is used in specific environments remains to be seen.
>
> Well, certainly the main handshake only runs once per connection.
>
> > > - There's special case with cookies: DTLS 1.3 ClientHello getting
> > >   rejected with HelloVerifyRequest. I think the correct reaction
> > >   is for client to send a DTLS 1.3 ClientHello without 0-RTT data,
> > >   containing the cookie sent now in legacy cookie field (NOT the
> > >   extension).
> > > - According to TLS 1.3 rules, handshakes rejected using
> > >   HelloRetryRequest use the Cookie extension for cookie
> > >   transmitback, not the legacy cookie field (the cookie might not
> > >   even fit into that field!)
> >
> > Regarding the issue of where to send the cookie in the ClientHello there
> > are two options. The first option is to use the cookie extension, and
> > the second option is to use the DTLS 1.2 style where the cookie is
> > placed in the ClientHello directly (without using an extension).
> >
> > My (weak) preference is to use the cookie extension also in the
> > ClientHello. My main reason was the length mismatch between the DTLS 1.2
> > cookie extension (which is now in the legacy_cookie filed) and the
> > cookie extension. The cookie extension has a max length of 2^16-1 bytes
> > and the legacy_cookie extension only 2^8-1 bytes long. If the design of
> > the ClientHello uses the legacy_cookie extension then the server has to
> > take care that the cookie size does not exceed the 2^8-1 bytes.
>
> I think the obvious way is:
>
> - Cookies from HelloVerifyRequest go to legacy cookie field.
> - Cookies from HelloRetryRequest go to extension cookie field.
>
> If you don't do the first, you can't downnegotiate successfully. And
> if you don't do the second, you would have to have special case with
> cookie sizes (since HRR can transmit >255 byte cookie, which would be
> too big for legacy cookie).
>

IMO we should just forbid HVR for DTLS 1.3. I.e., you should just send
HRR.


> > - The handshake retransmit scheme doesn't seem to work that
> > >   well with post-handshake auth, and even less well with
> > >   session tickets.
> > Why do you think so? Of course, unreliable transports creates
> > inconvenience; is it that what you are referring to?
>
> DTLS assumes handshake messages are reliable, and that reliability is
> implemented via handshake messages ACKing one another.
>
> - Session tickets have no ACK at all.
>

DTLS 1.3 should add an ACK, IMO.



> - CertificateRequest can have very slow ACK.
> - KeyUpdate has no real ACK (and isn't idempotent either).
>

Yes, I think we should remove KeyUpdate for DTLS 1.3 and just use epoch
instead.

-Ekr



>
> -Ilari
>
> _______________________________________________
> TLS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/tls
>