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

[MikkelFJ <[e-mail]>]

I'm a bit of a loss here because I thought the OCID was only in the first packet and not in all long headers (and perhaps in 0-RTT/HelloRetry - we can discuss these separately). But assuming you have OCID in anything but the very first packet:

The first user created CID is name OCID for original CID. This one is random, or supposed to be. It cannot be trusted, and it does not contain routing information nor reliable entropy.

If the LB is to use CID for routing, OCID cannot be used, but it is important to know that it is an OCID. This is the information that I want in the invariants.

If the OCID only appears in the first packet there is no state. If it can be used in multiple packets (and not a retransmitted first packet), then routing must be consistent. There is help in the OCID so you must either use another CID if possible, or some other mechanism.

If that other mechanism is 5-tuple, you need to store state, or you need to route consistently. And you have to do it until you know that you have a useful CID.

If instead you can trust a server assigned CID on anything but the first packet things start to improve. Since we don't have a full handshake, any CID is a bit dubious still, but then CID's are always dubious to external observers. The important part is that the receiver can reject. The server can reject a server assigned CID.


-- 
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-505164170