- From: Demi Marie Obenour <demiobenour@gmail.com>
- Date: Wed, 1 Jul 2026 18:23:35 -0400
- To: Ben Schwartz <bemasc@meta.com>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>
- Message-ID: <7d032cfb-bd42-42ba-b3e6-3614731807ca@gmail.com>
On 7/1/26 16:58, Ben Schwartz wrote: > "Upgrade: HTTP/2.0" is a messy situation. My view is that it does not > exist, and the IANA registry entry that claims otherwise is incorrect, > due to an oversight during development of the HTTP/2 RFCs. I would > support writing a new RFC to remove that entry from the registry (or > marking it deprecated). Definitely a good idea. > See https://github.com/httpwg/http-extensions/issues/2737 for some > earlier discussion of this question. We also discussed it in HTTPBIS > at IETF 120. > > That said, I believe the original theory of "Upgrade: HTTP/*" was that > the server's HTTP/X.Y data stream would start after the HTTP/1.1 > Upgrade response header, and would include the response to the POST > request. > > --Ben Should there be a new RFC (or at least BCP) that updates processing of the Upgrade header? I suspect that the only useful Upgrade tokens used in practice are either "websocket" or start with "connect-". If one of these tokens is present, I'd be in favor of rejecting the whole message in all of these situations: - The Upgrade header appears more than once or contains ",". - The token is in a response, the status is 101 Switching Protocols, and the token is different from the value in the request. - The token is in a request, but the method is not GET. - The token is in a request, but there is a Transfer-Encoding or Content-Length header. - There is no "Upgrade" connection token. - A request has "Upgrade: websocket" but no valid "Sec-WebSocket-Key" header. - A response has "Upgrade: websocket" but no valid "Sec-WebSocket-Accept" header. Otherwise, I think the whole Upgrade header should be ignored by endpoints and stripped by intermediaries. If the response was 101 Switching Protocols, the response ought to be rejected. Thoughts on this? > On Wed, Jul 1, 2026 at 4:32 PM Demi Marie Obenour <demiobenour@gmail.com> wrote: >> >> Suppose I send the following request: >> >> POST /something HTTP/1.1 >> Upgrade: HTTP/2.0 >> Connection: Upgrade >> Content-Length: 8 >> >> Example >> >> and get the following response: >> >> HTTP/1.1 101 Switching Protocols >> Upgrade: HTTP/2.0 >> Connection: Upgrade >> >> How is the server supposed to send the response to the client? >> -- >> Sincerely, >> Demi Marie Obenour (she/her/hers) -- Sincerely, Demi Marie Obenour (she/her/hers)
Attachments
- application/pgp-keys attachment: OpenPGP public key
Received on Wednesday, 1 July 2026 22:23:45 UTC