Re: [TLS] A flags extension

[Hubert Kario <[e-mail]>]

On Tuesday, 26 March 2019 16:38:11 CET Yoav Nir wrote:
> > On 26 Mar 2019, at 14:45, Hubert Kario <[e-mail]> wrote:
> > 
> > On Monday, 25 March 2019 22:09:35 CET Yoav Nir wrote:
> >> Hi.  Today at the TLS meeting, there was a discussion at the mic about
> >> 1-bit extensions that only serve to indicate support for an optional
> >> feature. EKR commented that such extensions take 4 bytes each and that
> >> maybe we need to replace them with a flags extension.
> >> 
> >> So I threw together a quick -00 draft with an extension that does just
> >> that
> >> [1].
> >> 
> >> Comments are welcome.
> > 
> > I don't think that "penny-pinching" the 4 bytes necessary to send a flag
> > is
> > worth the interoperability problems, and increased complexing of parsing
> > Client Hello. Especially if we go the route of actual bit flags.
> 
> Right. Which is why I went with a 1-byte encoding rather than a bitstring.
> 
> > I think the likelihood of bugs in that code over the possible bytes saved
> > makes it a net negative.
> 
> I don’t think so. My encoding is not all that complex.
> 
> > yes, TLS is quite chatty protocol, it could encode values much more
> > tightly, but I think we all remember the bugs related to ASN.1 parsing
> > from inside of PKCS#1 v1.5 signatures
> 
> Complexity is on a spectrum.  DER encoding is pretty far on this spectrum. 
> A list of 1-octet identifiers is on the other end. A bitstring is more
> complex than the identifier list, but not anywhere near DER.

1-octet identifiers may not be considered extensible enough
(yes, you can add another extension, but the first extension to use it will be 
paying an additional price of 2 bytes on top of the extension overhead; same 
if you just need to use only one flag, then you are paying the same price for 
every connection)

2-octet identifiers asymptotically approach 2-octet saved per flag, which is 
about 50% saved per flag, I don't see it as much

to approach it from another way: while I think we will, sometime in the 
future, reach a situation when we have few hundred flag extensions *defined* , 
I do not see a future in which we will need to *use* more than few dozen flag 
extensions in any real world client. So we are talking about a possible saving 
of around 100 bytes in ClientHello (36 extensions * 3 bytes saved) in this 
proposal

won't this be completely erased by any post-quantum key share?

> I don’t think we should project the failings of DER parsing to the parsing
> of much simpler structures.

yes, that was an extreme example; I was replying to all sent messages so far

-- 
Regards,
Hubert Kario
Senior Quality Engineer, QE BaseOS Security team
Web: www.cz.redhat.com
Red Hat Czech s.r.o., Purkyňova 115, 612 00  Brno, Czech Republic