Re: [quicwg/base-drafts] Long Header Packets and Routing Connection IDs (#2834)

[Nick Banks <[e-mail]>]

The attack vector would only be for designs without an auth tag in the CID. I'm personally not a big fan of the really large CID designs, nor am I convinced that they are truly necessary for a complete and secure system. So, I continue to look into how it would be possible to route with smaller CIDs. For that kind of design, the "server ID" or SID would be encoded somehow in the CID, but not encrypted. If you applied this design to all CIDs, including client chosen CIDs, then it would allow an attacker to directly create new connections to a particular server, by specifying the correct SID. That is definitely not what we want in a design though. So, instead of extracting the SID out of the CID for client chosen CIDs, we'd just hash the CID with some private key instead, and use the hash to pick a SID. BUT that only works if you can differentiate client chosen and server chosen.

As you indicated, an auth tag would help with that, but require additional complexity for the overall design. I'm looking for solutions that don't necessarily increase the length of the CID.

Being able to route based on the CID consistently just by looking at the long header bit and then interpreting the destination CID accordingly would definitely make for simpler load balancing. It could even be version independent. I find the idea of a version independent LB very attractive.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/2834#issuecomment-504807474