[anti-censorship-team] Azure domain fronting, meek ESNI

[David Fifield]

On Thu, Apr 01, 2021 at 11:15:47PM -0400, Roger Dingledine wrote:
> On Mon, Mar 29, 2021 at 01:19:33PM -0600, David Fifield wrote:
> > One possible alternative is ESNI with Cloudflare, using the mainline
> > meek code and its support for a headless (ESNI-supporting) Firefox.
> > However, this will require a lot of Tor Browser work to swap meek
> > implementations and re-wire the headless browser support files.
> 
> One huge advantage of routing via Cloudflare is that it's free (gratis),
> right? That is, we could move the (currently hugely rate limited and
> thus very slow) meek-azure traffic over to this future meek-cloudflare
> service, and open up the rate limits a lot more?

I think that's right. I don't know how the paid features break down, but
you can use the CDN free of charge.

> > One problem with the headless Firefox model is that the TLS fingerprint
> > of the ESR release used by Tor Browser would rapidly become uncommon
> > (because most people don't run ESRs). See Section V of
> > https://tlsfingerprint.io/static/frolov2019.pdf. But we currently have
> > that problem anyway, as the version of uTLS we are using is two years
> > old (Chrome 72, Firefox 65, and even the dev branch is 9 months old).
> 
> How far is the current utls from being able to do ESNI? That approach
> might be more work in the short term, but provide the "easier to maintain"
> feature in the long term?

I don't think it's close. uTLS is patches on top of the Go standard
library tls/crypto, and the Go maintainers don't have plans to support
it until after browsers do.
https://github.com/golang/go/issues/9671#issuecomment-439561672
ESNI itself is a dead end now; any development work now would go toward
ECH instead.

> I hear ESNI won't work so well in China, but there are plenty of other
> censored situations where it would be really useful to offer users a
> higher-bandwidth domain-fronted option.

There's a secondary risk, though. ESNI/ECH are not deployed by default
in any clients. If we're the only ones using it, then far from being
covert, ESNI/ECH becomes a signal for traffic censors want to block.
Worst case, if we act incautiously, is that we get a protocol blocked
before it catches on and undo a lot of hard work. With ESNI it's maybe
not so bad (compared to ECH), as it's on the way out anyway.