S. Bortzmeyer ✅ utilise mastodon.gougere.fr. Vous pouvez læ suivre et interagir si vous possédez un compte quelque part dans le "fediverse". Si ce n’est pas le cas, vous pouvez en créer un ici.

Snowden was nice enough to publish the NSA documents just one month before the publication of 6973 "Privacy considerations in Internet protocols".

The Internet is a leaky boat. Information comes out by many means (DNS, SNI, timing, metadata…)

Many people can see your DNS requests: resolver sysadmin, authoritative sysadmins, and of course the sniffers (and the DNS requests can travel far).

Some extensions (client subnet…) may make things much worse, -wise.

To summarize: operators can (and do) see and log your queries, and one can extract a lot of info from it (reidentification).

Solutions against leakage: DPRIVE working group, its 7626 (problem analysis, rebuts the "alleged public nature of DNS data" myth).

Several technical solutions were considered at the . The choice was DNS-over-TLS-over-new-port. Does not solve everything: timing and sizes are still there.

As my grand-mother used to say, "encrypting is easy, authenticating is difficult". Sara Dickinson now explains the possible choices for DNS-over-TLS auth. Strict or opportunistic?

Encryption also protects against sniffers, not against server operators. For that, you also need QNAME minimisation (RFC 7816). Less data, less issues.

S. Bortzmeyer ✅ @bortzmeyer

And for the cases where port 853 is blocked, I want you to meet DNS-over-HTTP(S)…

There is also a DNS-over- project. (QUIC can encrypt, and validate the source IP address.)