- From: Mark Nottingham <mnot@mnot.net>
- Date: Thu, 15 Sep 2016 09:21:19 +1000
- To: Kazuho Oku <kazuhooku@gmail.com>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>
The rules in <http://httpwg.org/specs/rfc7230.html#message.body.length> still apply: - Any response to a HEAD has no body - Any 1xx, 204 and 304 response has no body - A 2xx response to a CONNECT has no body (because it's no longer HTTP after the header fields) - Otherwise, the message has a body which might be 0-length. We intentionally made the set of messages without a body unable to be extended; see <http://httpwg.org/specs/rfc7231.html#considerations.for.new.methods>: """ Since message parsing (Section 3.3 of [RFC7230]) needs to be independent of method semantics (aside from responses to HEAD), definitions of new methods cannot change the parsing algorithm or prohibit the presence of a message body on either the request or the response message. Definitions of new methods can specify that only a zero-length message body is allowed by requiring a Content-Length header field with a value of "0". """ and <http://httpwg.org/specs/rfc7231.html#considerations.for.new.status.codes>: """ To allow existing parsers to process the response message, new status codes cannot disallow a payload, although they can mandate a zero-length payload body. """ Cheers, > On 15 Sep 2016, at 7:17 AM, Kazuho Oku <kazuhooku@gmail.com> wrote: > > Hi, > > Is there any way for a H2 server to distinguish between a request > without a body and a request with 0-byte body? > > In HTTP/1, the distinction has been possible by looking for a > content-length or a transfer-encoding header. And H1 applications have > been actually looking for the headers to see if a request is > accompanied by a body by checking the existence of these headers. > > OTOH, HTTP/2 does not seem to provide a method to distinguish between the two. > > A HTTP/2 client is allowed to send a request accompanied by a body > without using the content-length header. It is also allowed to send a > HEADERS frame with END_STREAM flag set in case the size of the body is > zero-byte, omitting the DATA frame. > > In such case, a request with zero-byte body becomes indistinguishable > from a request without a body. > > The fact becomes an issue when we need to transcode a HTTP/2 request > to a HTTP/1 request (e.g. when a H2 proxy transmits a request to an H1 > server running upstream), because, some applications try to see if a > POST request is accompanied by a body by checking the existence of > content-length or transfer-encoding header, or to assert that a GET > request is _not_ accompanied by a body by checking the non-existence > of the headers. > > As a mitigation, it is certainly possible for a H2 proxy transcoding > to H1 to use the method of the request to see if content-length or > transfer-encoding header should be set in such case. But my > understanding is that generally speaking in HTTP whether if a request > is accompanied by a body is orthogonal to which method is being used. > > Could somebody clarify what I am missing, or provide a method I should > use to accommodate the issue? > > -- > Kazuho Oku > -- Mark Nottingham https://www.mnot.net/
Received on Wednesday, 14 September 2016 23:21:48 UTC