Re: [quicwg/base-drafts] Compatible version upgrade (#1901)

[martinduke <[e-mail]>]

After discussing this with @kazuho, @janaiyengar, and @ianswett in Bangkok, I'm not sure the server upgrade problem has been correctly reasoned.

Downgrade attack detection happens in the context of a single connection. After receiving VN, the client uses the same CID and increments the packet number. Non-pathological load balancers will direct this to the same server. People have repeatedly pointed out to me that the LB config could flip over in that moment, but that is an edge case of an edge case, and the only consequence is a fatal connection error and a possible security alert.

It might we worth adding somewhere that in the event of a downgrade alert, the client SHOULD NOT cache the contents of VN for future use. Also, if the server's TPs don't match the contents of a client version cache from a different connection, that is not a downgrade attack.

Furthermore, @ianswett points out that downgrade attack detection traditionally happens at the client. We haven't fully thought through the incentives and other implications of moving this responsibility to the server.

In summary, I no longer support moving downgrade detection to the server.  I am open to seamless upgrade in v2 after we do some more analysis to make sure there are no gotchas.

-- 
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/pull/1901#issuecomment-435774516