Deploying TLS 1.3: the great, the good and the bad [video]

https://media.ccc.de/v/33c3-8348-deploying_tls_1_3_the_great_the_good_and_the_bad

by jgrahamc • 10 years ago

78 6 10 years ago

media.ccc.de

FiloSottile2• 10 years ago

Hi, co-speaker here.

Please feel free to ask any questions you have here, or by emailing tls13 at cloudflare.

The slide deck is at https://speakerdeck.com/filosottile/tls-1-dot-3-at-33c3 (but it's not really made for standalone consumption).

cesarb• 10 years ago

I have one question.

What prevents a Sufficiently Incompetent Implementer from manually ignoring all known GREASE values, while still doing something crazy otherwise?

davidben• 10 years ago

Ultimately we can't do anything about a server that is near-maliciously determined to be wrong. The hope is we can at least defend against some of the common genuine mistakes at fairly low cost.

The implementer that tries to special-case GREASE values must have first hit an interop problem and thus had a second or third opportunity to think about it. I would hope, at that point, they realize they're meant to ignore unknown ones!

kpcyrd• 10 years ago

Do you know of 0-RTT implementation efforts on reverse proxies, say, nginx?

Let's say I run a static website and I'm 100% sure unauthenticated GET won't break anything, is there anything special I need to do or is it going to be "enable this flag"?

Another question: Can I set 0-RTT per vhost (sni) or do I need dedicated 0-RTT IPs?

FiloSottile2• 10 years ago

There is no 0-RTT implementation in servers that I know of, except our internal one at Cloudflare, which currently waits for the Finished confirmation until we decide on a policy.

The following two questions are of course implementation-specific, but there is nothing in the protocol blocking a "just do 0-RTT" flag or vhost configs.

FullyFunctional• 10 years ago

I'm a bit late, but it was a great talk and TLS 1.3 is looking nice.

I'm not sure who was the first to propose 0-RTT like this, but it appears that both MinimaLT and QUIC did at about the same time.

TLS 1.3 does nothing to improve TCP deficiencies, but there are alternatives that does, including QUIC. What I can't quite tell is how the the two combine, if at all. Does a QUIC connection mean that TLS 1.3 doesn't apply/is redundant?

Deploying TLS 1.3: the great, the good and the bad [video]

Related Stories