Re: SSL SNI

On Mon, 2021-02-15 at 15:09 +0100, Peter Eisentraut wrote:
> The question I had was whether this should be an optional behavior, or
> conversely a behavior that can be turned off, or whether it should just
> be turned on all the time.

Personally I think there should be a toggle, so that any users for whom
hostnames are potentially sensitive don't have to make that information
available on the wire. Opt-in, to avoid having any new information
disclosure after a version upgrade?

> The Wikipedia page[1] discusses some privacy concerns in the context of
> web browsing, but it seems there is no principled solution to those.

I think Encrypted Client Hello is the new-and-improved Encrypted SNI,
and it's on the very bleeding edge. You'd need to load a public key
into the client using some out-of-band communication -- e.g. browsers
would use DNS-over-TLS, but it might not make sense for a Postgres
client to use that same system.

NSS will probably be receiving any final implementation before OpenSSL,
if I had to guess, since Mozilla is driving pieces of the
implementation.

--Jacob