ODNS: Oblivious DNS | Hacker News Reader

ODNS: Oblivious DNS

odns.cs.princeton.edu

Previous discussion: https://news.ycombinator.com/item?id=16741031

I was hoping this would be DNS with oblivious transfer, but its just a silly proxy.

If you can snoop on an authoritative dns server then you can probably just snoop on the authoritative odns server and break the scheme with no real cost.

For the odns server to scale it would probably be implemented by the same people implementing dns servers, like ISPs and governments, the same people doing the snooping.

It just looks weak in theory and broken in reality.

It still seems to have the same problem as DNS over TLS - that is that someone in that pipeline sees your query.

In the case of 'standard' DNS (unencrypted port 53 traffic) - anyone can 'snoop' (ISP, internet routers, etc) and the resolver can retain that information.

In the case of DNS over TLS - it's the box the performs the decryption that can retain that information.

In this case, the ODNS Stub can retain that information.

The only question is who do you want to be able to view your queries. AKA who do you trust the most (or possibly, who do you trust the least).

That someone in the case of the stub could be your local machine or a trusted router though. If your operating system environment knows ODNS, it doesn't even have to be an extra piece of software.

Let's take the case that it's a local machine - on your local network - acting as the ODNS Stub. The DNS request is then encrypted from your machine to the trusted ODNS Stub - but then from that ODNS Stub out to some resolver on the internet, there exists the DNS query (whether encrypted or unencrypted) - sourced from your same public IP that your machine would have sourced it from in the first place, correct?

I'm certain I don't understand the entirety of ODNS - but the basics still have to exist - something still needs to make a query on behalf of the initial user to some authoritative server (unless the answer is already cached). I guess, what I'm picturing in this case, is that if the trusted resource (ODNS Stub) exists in your local infrastructure, then the source outgoing to the internet might as well just be your local machine in the first place - because that's how 'the internet' sees it anyway.

Anyway, I'm all for making DNS more secure...and this seems to be one way to change where your trust lies, but not a definitive solution, to me.

> The DNS request is then encrypted from your machine to the trusted ODNS Stub

No. The stub is doing the encryption and sending the encrypted query to a resolver in the internet.

> but then from that ODNS Stub out to some resolver on the internet, there exists the DNS query (whether encrypted or unencrypted)

Yes. But since it's encrypted, the resolver only knows that I have made a DNS request, not what the query is in it. The resolver can't decrypt it, it can only pass it on to a server that can decrypt it, but won't be able to see your source IP then.

I think they overstate the usefulness against the kind of attacker they describe though, since such a powerful adversary could correlate across servers.

Thats what I like about oblivious transfer and/or private information retrieval, the server is oblivious to what information is being requested.

The problem with oblivious transfer is the computational time and resources to do it would make it pretty slow.

[1] https://en.wikipedia.org/wiki/Oblivious_transfer

Reply on news.ycombinator.com