Cleartext integrity as version independent

[mcmanus]

making the cleartext integrity checksum version independent would prevent a lot of version negotiation packets from being generated based on receipt of non-quic (and make it harder to use that as an amplification vector?)

[martinthomson]

If version negotiation is amplifying, something is badly wrong.

I believe that version negotiation should only be generated for packets that exceed a certain size (because the only packet that it makes sense to send version negotiation in response to is an initial client packet and those are guaranteed* to be pretty big*). That is, for some version of a guarantee and some version of big. Note however that we don't have that written down (or even agreed).

As for being able to distinguish noise from QUIC, that's a good reason. We should discuss that.

[mcmanus]

re: amp - that's a versioning guarantee thing again.. a v1 initial client packet is pretty big.. but a >1 packet received by a v1 server? we have very few rules about the future.. maybe we need a few.

[martinthomson]

Yes, some constraints on future flexibility seem wise, as long as we can keep it light.

[martinthomson]

I think that we resolved this with invariants and #724. @mcmanus, explain and reopen if you disagree.