4 ms·
My only problem is why force TLS when we have better protocols nowadays. We did some research on using noise (in addition to TLS) but I’m not sure anyone really
by baby 6y ago
My only problem is why force TLS when we have better protocols nowadays. We did some research on using noise (in addition to TLS) but I’m not sure anyone really pushed for this to be considered officially (cf nQUIC)
- Skunkleton 6y agoThere is nothing precluding HTTP/3.1 from switching up some aspect of the protocol. TLS was probably selected here because it is mature and well supported.
- baby 6y agoThe problem is in how the specs are deeply mixed with TLS. We advocated for the specs to be more flexible so that we could extend QUIC to be specified with noise as well but I don’t believe that has happened. Also TLS was selected because of regulations. Which is damning to say the least.
- Skunkleton 6y ago> Also TLS was selected because of regulations. I would like to read more about this. Link?
- microcolonel 6y agoThe benefit of TLS is widely deployed certificate infrastructure. If you could find a way to keep the server certificate infrastructure intact while cutting out most of the ugliness of TLS, that'd probably be a workable solution. FWIW though, TLS 1.3 (the only currently supported TLS in QUIC, AFAIK) is a lot less hairy than TLS 1.2 and earlier; and most of the primitives are the same as you would use for noise (i.e. standard AEADs like ChaCha20-Poly1305).
- baby 6y agoX509/DER parsing has been known as the most dangerous code in the world (https://crypto.stanford.edu/~dabo/pubs/abstracts/ssl-client-bugs.html https://crypto.stanford.edu/~dabo/pubs/abstracts/ssl-client-...) Sure support is good but so is PGP’s
- tialaramex 6y agoWe have actual security proofs for TLS 1.3 (which is the only version offered in QUIC) Of course such proofs come with caveats - in particular the proofs assume your cryptographic primitives work as intended (e.g. that AES isn't broken) and that you've implemented the specification and not something else - but our experience has been that it's valuable to get those proofs, and where proof turns out to be difficult it's a valuable pointer to weaknesses in your design. As far as I know nobody has done such a proof for the Noise design, and so we only have our intuition that it looks safe.
- luizfelberti 6y agoAFAIK Noise has end-to-end proofs of correctness using CryptoVerif, the work having mostly been done by the Prosecco team at INRIA. You can explore Noise protocols on NoiseExplorer [0][1] Also I may be wrong again, but it is for TLS, that even with all the simplifications and major changes on TLS 1.3, that we only have proofs of correctness for certain subsets of the spec, and specific modes of operation (theres a version of the protocol targeting IoT for which this work has been done, but I can't remember the name) [0] https://noiseexplorer.com/ https://noiseexplorer.com/ [1] https://rwc.iacr.org/2019/slides/NoiseExplorer.pdf https://rwc.iacr.org/2019/slides/NoiseExplorer.pdf
- baby 6y agoNot sure why you are getting downvoted because what you say is true.
- baby 6y agoIt’s actually the other way around. There are no up-to-date and complete proofs of tls, but there are several proofs for noise.
- luizfelberti 6y agoI loved the nQUIC paper. It's really sad that QUIC has not been thought of as a "meta" protocol of sorts, where you can bring your own crypto (or none at all). I find it kind of annoying that I can't use QUIC for intra-cluster communication with the crypto disabled, and leaving those preocupations for a service-mesh layer, or even just overlaying the entire cluster with a WireGuard VPN. There are legitimate use cases for "running naked", but now we're doomed to jerry-rigging some BS ephemeral self-signed certificates in order to use any QUIC client that follows the official spec.