Downgrade is a problem even if v2 offers no security improvements. For instance, a middlebox might want to inspect v1 Initials and force downgrades, which would prevent deployment of new versions.
On Fri, Jan 8, 2021 at 11:15 AM Spencer Dawkins at IETF < [email protected]> wrote: > Hi, Ben, > > Top-posting here - I think one other point that is worth remembering, is > that the QUIC working group also has > https://datatracker.ietf.org/doc/draft-ietf-quic-manageability/, which > seems like a fine place for recommendations about deployment of specific > versions and withdrawal of other versions. > > In my mind, there are two cases. Either > > - QUICv1 is "secure enough", so using it is OK even if QUICv2 would be > better, or > - QUIVv1 is not "secure enough", so downgrading will be a problem. > > People can deploy QUIC using a variety of implementation strategies, but > given that the QUIC implementation is likely at least a library, and may be > a library bound to a specific application, it would be reasonable to say > "QUICv1 is not secure enough, so stop using QUICv1 as soon as possible", > and let implementers and deployers put out versions of applications that > aren't bound to QUICv1 at all. > > (This discussion is slightly weird to me, because the last time I asked > about "QUICv2", the answer I got was that we're more likely to run for some > time with QUIv1 + extensions, but even then, my intention when QUIC was > chartered, was that deploying new versions should be orders of magnitude > than the universal deployment of a new version of TCP, for instance, and > withdrawing QUICv1 should be a lot easier than withdrawing TCPv4). > > I'm sure everyone will Do The Right Thing, of course. > > Best, > > Spencer > > On Thu, Jan 7, 2021 at 5:18 PM Benjamin Kaduk <[email protected]> wrote: > >> Thanks everyone for the productive discussion. It's clear that there's >> a lot of background available to those who participated in the previous >> WG discussions but (understandably!) did not make it into the document >> itself, and I appreciate the effort that was put in to help share that >> with >> me. >> >> Just to state it clearly, at no point has my position been that QUIC v1 >> needs to be delayed until a complete version negotiation story exists. >> As this was a "discuss discuss", my goal was to obtain more information >> about the actual situation in order to confirm that there are no >> significant issues, since my interpretation of the text in the document >> itself left that possibility open. >> >> Attempting to summarize salient points: >> >> - the IETF is only currently defining bindings for HTTP over QUIC, >> though other entities are free to define their own protocol over QUIC >> at any time. >> - the only way currently defined to discover a QUIC endpoint to use as >> server for a given HTTP service is the Alt-Svc header field, which >> uses an ALPN value to indicate the protocol to use; it is perhaps not >> fully nailed down that the ALPN value will be specific to a particular >> version of QUIC but the ALPN vlaue probably will be specific to a >> particular version of QUIC. >> - (SVCB is in the works, too, but may not be able to meet all the needs >> for this purpose.) >> - Anyone doing non-HTTP or non-Alt-Svc is presumed to be configuring it >> out of band and thus can provision the QUIC version to use along with >> other provisioned information; in-band version negotiation is not >> needed in that case. If needed (e.g., we cannot build a secure >> downgrade protection mechanism), this or similar techniques could be >> used generically. >> - A downgrade protection mechanism solely in-band at the QUIC layer will >> not be a complete solution for existing protocols that may also fall >> back to a TCP binding (or new protocols that need to traverse networks >> like the Internet that don't reliably pass UDP in the ways QUIC >> needs). New protocols over QUIC that are berift of such legacy would >> have a complete solution, though. >> - There seems to be a desire to have only zero or one functional downgrade >> prortection/version negotiation mechanism, globally. >> - (There is a corresponding desire to have zero non-functional downgrade >> protection/version negotiation mechanisms.) >> - In accordance with the previous two points, it's expected that a >> downgrade protection/version negotiation scheme, when specified, will >> be in an IETF standards-track protocol specification. (This document >> does not necessarily have to be a new QUIC version, as I understand it, >> though is not a blocking dependency until there is such a new version.) >> - In particular, we do *not* expect non-IETF QUIC versions to define >> their own downgrade protection scheme. They are expected to either >> pick up the IETF one (when it exists) or just only use a single >> version at a time, possibly with out of band configuration. >> >> >> I've attempted to update the text in the document to reflect my >> understanding of the current WG expectations (as summarized above), in a >> PR at https://github.com/quicwg/base-drafts/pull/4697 . >> Obviously, if my summary above is incorrect, that PR is not expected to >> be useful. >> >> In particular, since we do *not* expect or want non-IETF QUIC versions >> to be attempting to specify a downgrade protection scheme, the scope of >> the problem space seems sufficiently restricted that we have ample time >> to come up with something good and not find ourselves reacting to events >> out of our control. The phrasing in the -33 suggests, at least to me, >> that *any* future version of QUIC, including one developed outside the >> IETF, might update version negotiation handling, which is where my >> perception of risk arose. >> >> I've tried to refrain from expounding on topics that are not actually >> relevant, but since I'm prone to doing so I may have let some sneak in >> anyway... >> >> Thanks again, >> >> Ben >> >>
