Hi Kyle, Thanks for the review! There's a PR with the resulting changes: https://github.com/quicwg/quic-v2/pull/75
replies inline On Tue, Oct 4, 2022 at 8:18 AM Kyle Rose via Datatracker <[email protected]> wrote: > Reviewer: Kyle Rose > Review result: Ready > > I have only three additional questions/comments: > > - What are the implications of the server not encoding the version in its > Retry > message and subsequently checking that the client didn't change versions > upon > retrying? > AFAICT there are no security implications here. The spec is restrictive to reduce the complexity of the code, and gives the server the option of enforcing the rule in order to discourage clients from violating it. > > - Is there any optimization possible if the server keeps the Initial > receive > keys slightly longer than the first instance of processing a packet using > keys > generated for the negotiated version? I'm guessing not, but I just wanted > to > confirm. > No. It might want to keep around the 0-RTT keys from the original version, but once it receives the negotiated version there are no outstanding client Initial packets with the old version. > > - In "Endpoints have no need to generate the keying material that would > allow > them to decrypt or authenticate these packets", I would s/these/such/. > > >
