Skip to content

Comment on Hypertext Transfer Protocol Version 2

Comments

To those saying HTTP2 should require HTTPS: do you really want to be stuck with our shitty CA model for even longer?

TLS is actually not glued to x.509 certificates, and certainly not to ones chained to the public CA system. DANE TLSA <https://tools.ietf.org/html/rfc6698> for example provides for a way to pin them to DNSSEC, whether in addition to or instead of the CA system.

Other ways of keying may become available in the future, especially as we go through TLS 1.3 to future versions.

There's an argument I've heard that DANE TLSA ties crypto closer to nation-state control than the CAs do, since the DNS is ultimately managed through a governmental framework.

This was presented to me as a showstopper, on the grounds of government = bogeyman. Personally I'm much more equivocal about the dangers. But I do see DANE as a useful alternative in many circumstances, particularly those where defending against a major government isn't considered a worthwhile aspiration.

Or even better as a trust multiplier or CA-issued certs.

Jurisdiction is a complex question, and one I can't answer, but people should think about the jurisdictional consequences of using ccTLDs in particular. You are at the mercy of whomever's higher in the DNS hierarchy - that's why the root control is such a big deal and needs to come out of the US into some sort of non-governmental or international treaty org.

However, certainly DANE 2 & 3 are way better than nothing, and DANE 0 & 1 let you limit the subset of "people who can spoof your site" from "people who have pwned any CA they want trusted by a major browser, including intermediates" (do we even know how many intermediates are out there? Answer: No, we don't!! It's an open research question to scan to try to find out: please save us, Certificate Transparency!) to "people who have pwned the correct CA, or the government with the most obvious jurisdiction over my site".

If that isn't enough for you, and I can understand why it may not, then you need to be considering using Tor or something similar to meet your threat model...

Certificates are absolutely not required to have useful opportunistic encryption.

Alternative? Non-https is really harmful to users who browse via open wifi.

Alternative?

There exist several semi-viable alternatives at the moment. The most promising is Namecoin, which essentially allows for cryptographically authenticated public key/value storage, and so can function as both a DNS system and a system for announcing SSL pubkeys (without any trusted parties).

can be really harmful, in some applications

I don't need to know that my connection to lol cats is secure and uninterrupted.

Let's assume a simple video site and html5 video.

Do you want to bet there are no vulnerabilities in your browser's mp4/vp8 decoder? If there are, anyone can MITM your non-SSL connection and compromise the video decoding process, limited only by their cleverness and by whatever sandboxing your browser employs.

Why does it have to be video?

Some of the simplest sites around are plain text or text and images. Why do I need SSL or TLS to view either of these?

I am very aware that anything authenticated should be carried over an encrypted connection - I'm not arguing against that. I'm saying there are use-cases that do not require it.

Because otherwise nation-state adversaries can see everything you read.

IETF position on this is pretty clear after the technical plenary last year.

Again, why does it matter if you or Barack Obama or Kimg Jung Un knows that I like looking at Captain Picard meme pictures? How can you tell me that REQUIRES an encrypted connection?

You know what might actually make some fucking sense. Enforce SSL/TLS where it is needed - disable HTTP Auth, Cookies, Location Services, Local Storage, etc over plain HTTP connections.

Do you want your location along with your credentials to your lol cats forum in the wild ?

Depending on what site you're going to, you can be screwed pretty easily.

I didn't mention credentials. Plenty of web use is anonymous, and has 0 need for SSL/TLS

You don't need credentials to be identified, your browser can be identified quite accurately (see https://panopticlick.eff.org/).

Counterexample: If you want it to be anonymous, you need TLS (and Tor!).

Well I'm glad you're able to miss the fucking point with such efficiency

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.