Internet-Draft | CONNECT-UDP | December 2020 |
Schinazi | Expires 15 June 2021 | [Page] |
This document describes the CONNECT-UDP HTTP method. CONNECT-UDP is similar to the HTTP CONNECT method, but it uses UDP instead of TCP.¶
Discussion of this work is encouraged to happen on the MASQUE IETF mailing list masque@ietf.org or on the GitHub repository which contains the draft: https://github.com/ietf-wg-masque/draft-ietf-masque-connect-udp.¶
This note is to be removed before publishing as an RFC.¶
Source for this draft and an issue tracker can be found at https://github.com/ietf-wg-masque/draft-ietf-masque-connect-udp.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 15 June 2021.¶
Copyright (c) 2020 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
This document describes the CONNECT-UDP HTTP method. CONNECT-UDP is similar to the HTTP CONNECT method (see section 4.3.6 of [RFC7231]), but it uses UDP [UDP] instead of TCP [TCP].¶
Discussion of this work is encouraged to happen on the MASQUE IETF mailing list masque@ietf.org or on the GitHub repository which contains the draft: https://github.com/ietf-wg-masque/draft-ietf-masque-connect-udp.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
In this document, we use the term "proxy" to refer to the HTTP server that opens the UDP socket and responds to the CONNECT-UDP request. If there are HTTP intermediaries (as defined in Section 2.3 of [RFC7230]) between the client and the proxy, those are referred to as "intermediaries" in this document.¶
The CONNECT-UDP method is defined for all versions of HTTP. When the HTTP version used runs over QUIC [QUIC], UDP payloads can be sent over QUIC DATAGRAM frames [DGRAM]. Otherwise they are sent on the stream where the CONNECT-UDP request was made. Note that, when the HTTP version in use does not support multiplexing streams (such as HTTP/1.1), then any reference to "stream" in this document is meant to represent the entire connection.¶
The CONNECT-UDP method requests that the recipient establish a tunnel over a single HTTP stream to the destination origin server identified by the request-target and, if successful, thereafter restrict its behavior to blind forwarding of packets, in both directions, until the tunnel is closed. Tunnels are commonly used to create an end-to-end virtual connection, which can then be secured using QUIC or another protocol running over UDP.¶
The request-target of a CONNECT-UDP request is a URI [RFC3986] which uses the "masque" scheme and an immutable path of "/". For example:¶
CONNECT-UDP masque://target.example.com:443/ HTTP/1.1 Host: target.example.com:443¶
When using HTTP/2 [H2] or later, CONNECT-UDP requests use HTTP pseudo-headers with the following requirements:¶
A CONNECT-UDP request that does not conform to these restrictions is malformed (see [H2], Section 8.1.2.6).¶
The recipient proxy establishes a tunnel by directly opening a UDP socket to the request-target. Any 2xx (Successful) response indicates that the proxy has opened a socket to the request-target and is willing to proxy UDP payloads. Any response other than a successful response indicates that the tunnel has not yet been formed.¶
A proxy MUST NOT send any Transfer-Encoding or Content-Length header fields in a 2xx (Successful) response to CONNECT-UDP. A client MUST treat a response to CONNECT-UDP containing any Content-Length or Transfer-Encoding header fields as malformed.¶
A payload within a CONNECT-UDP request message has no defined semantics; a CONNECT-UDP request with a non-empty payload is malformed. Note that the CONNECT-UDP stream is used to convey UDP packets, but they are not semantically part of the request or response themselves.¶
Responses to the CONNECT-UDP method are not cacheable.¶
When the HTTP connection supports HTTP/3 datagrams [H3DGRAM], UDP packets can be encoded using QUIC DATAGRAM frames. This support is ascertained by checking the received value of the H3_DATAGRAM SETTINGS Parameter.¶
If the client has both sent and received the H3_DATAGRAM SETTINGS Parameter with value 1 on this connection, it SHOULD attempt to use HTTP/3 datagrams. This is accomplished by requesting a datagram flow identifier from the flow identifier allocation service [H3DGRAM]. That service generates an even flow identifier, and the client sends it to the proxy by using the "Datagram-Flow-Id" header; see [H3DGRAM]. A CONNECT-UDP request with an odd flow identifier is malformed.¶
The proxy that is creating the UDP socket to the destination responds to the CONNECT-UDP request with a 2xx (Successful) response, and indicates it supports datagram encoding by echoing the "Datagram-Flow-Id" header. Once the client has received the "Datagram-Flow-Id" header on the successful response, it knows that it can use the HTTP/3 datagram encoding to send proxied UDP packets for this particular request. It then encodes the payload of UDP datagrams into the payload of HTTP/3 datagrams. Is the CONNECT-UDP response does not carry the "Datagram-Flow-Id" header, then the datagram encoding is not available for this request. A CONNECT-UDP response that carries the "Datagram-Flow-Id" header but with a different flow identifier than the one sent on the request is malformed.¶
When the proxy processes a new CONNECT-UDP request, it MUST ensure that the datagram flow identifier is not equal to flow identifiers from other requests: if it is, the proxy MUST reject the request with a 4xx (Client Error) status code. Extensions MAY weaken or remove this requirement.¶
Clients MAY optimistically start sending proxied UDP packets before receiving the response to its CONNECT-UDP request, noting however that those may not be processed by the proxy if it responds to the CONNECT-UDP request with a failure or without echoing the "Datagram-Flow-Id" header, or if the datagrams arrive before the CONNECT-UDP request.¶
Note that a proxy can send the H3_DATAGRAM SETTINGS Parameter with a value of 1 while disabling datagrams on a particular request by not echoing the "Datagram-Flow-Id" header. If the proxy does this, it MUST NOT treat receipt of datagrams as an error, because the client could have sent them optimistically before receiving the response. In this scenario, the proxy MUST discard those datagrams.¶
Extensions to CONNECT-UDP MAY leverage parameters on the "Datagram-Flow-Id" header (parameters are defined in Section 3.1.2 of [STRUCT-HDR]). Proxies MUST NOT echo parameters on the "Datagram-Flow-Id" header if it does not understand their semantics.¶
If HTTP/3 datagrams are not supported, the stream is used to convey UDP payloads, by using the following format (using the notation from the "Notational Conventions" section of [QUIC]):¶
The bidirectional stream that the CONNECT-UDP request was sent on is a sequence of CONNECT-UDP Stream Chunks. The CONNECT-UDP Stream Chunk Type is designed to allow future extensibility. Endpoints that receive a chunk with an unknown CONNECT-UDP Stream Chunk Type MUST silently skip over that chunk.¶
Unlike TCP, UDP is connection-less. The proxy that opens the UDP socket has no way of knowing whether the destination is reachable. Therefore it needs to respond to the CONNECT-UDP request without waiting for a TCP SYN-ACK.¶
Proxies can use connected UDP sockets if their operating system supports them, as that allows the proxy to rely on the kernel to only send it UDP packets that match the correct 5-tuple. If the proxy uses a non-connected socket, it MUST validate the IP source address and UDP source port on received packets to ensure they match the client's CONNECT-UDP request. Packets that do not match MUST be discarded by the proxy.¶
The lifetime of the socket is tied to the CONNECT-UDP stream. The proxy MUST keep the socket open while the CONNECT-UDP stream is open. Proxies MAY choose to close sockets due to a period of inactivity, but they MUST close the CONNECT-UDP stream before closing the socket.¶
HTTP/3 DATAGRAM flow identifiers are specific to a given HTTP/3 connection. However, in some cases, an HTTP request may travel across multiple HTTP connections if there are HTTP intermediaries involved; see Section 2.3 of [RFC7230].¶
Intermediaries that support both CONNECT-UDP and HTTP/3 datagrams MUST negotiate flow identifiers separately on the client-facing and server-facing connections. This is accomplished by having the intermediary parse the "Datagram-Flow-Id" header on all CONNECT-UDP requests it receives, and sending the same value in the "Datagram-Flow-Id" header on the response. The intermediary then ascertains whether it can use datagrams on the server-facing connection. If they are supported (as indicated by the H3_DATAGRAM SETTINGS parameter), the intermediary uses its own flow identifier allocation service to allocate a flow identifier for the server-facing connection, and waits for the server's reply to see if the server sent the "Datagram-Flow-Id" header on the response. The intermediary then translates datagrams between the two connections by using the flow identifier specific to that connection. An intermediary MAY also choose to use datagrams on only one of the two connections, and translate between datagrams and streams.¶
Proxies SHOULD strive to avoid increasing burstiness of UDP traffic: they SHOULD NOT queue packets in order to increase batching.¶
When the protocol running over UDP that is being proxied uses congestion control (e.g., [QUIC]), the proxied traffic will incur at least two nested congestion controllers. This can reduce performance but the underlying HTTP connection MUST NOT disable congestion control unless it has an out-of-band way of knowing with absolute certainty that the inner traffic is congestion-controlled.¶
When the protocol running over UDP that is being proxied uses loss recovery (e.g., [QUIC]), and the underlying HTTP connection runs over TCP, the proxied traffic will incur at least two nested loss recovery mechanisms. This can reduce performance as both can sometimes independently retransmit the same data. To avoid this, HTTP/3 datagrams SHOULD be used.¶
There are significant risks in allowing arbitrary clients to establish a tunnel to arbitrary servers, as that could allow bad actors to send traffic and have it attributed to the proxy. Proxies that support CONNECT-UDP SHOULD restrict its use to authenticated users.¶
Because the CONNECT method creates a TCP connection to the target, the target has to indicate its willingness to accept TCP connections by responding with a TCP SYN-ACK before the proxy can send it application data. UDP doesn't have this property, so a CONNECT-UDP proxy could send more data to an unwilling target than a CONNECT proxy. However, in practice denial of service attacks target open TCP ports so the TCP SYN-ACK does not offer much protection in real scenarios. Proxies MUST NOT introspect the contents of UDP payloads as that would lead to ossification of UDP-based protocols by proxies.¶
This document will request IANA to register "CONNECT-UDP" in the HTTP Method Registry (IETF review) maintained at <https://www.iana.org/assignments/http-methods>.¶
+-------------+------+------------+---------------+ | Method Name | Safe | Idempotent | Reference | +-------------+------+------------+---------------+ | CONNECT-UDP | no | no | This document | +-------------+------+------------+---------------+¶
This document will request IANA to register the URI scheme "masque".¶
The syntax definition below uses Augmented Backus-Naur Form (ABNF) [RFC5234]. The definitions of "host" and "port" are adopted from [RFC3986]. The syntax of a MASQUE URI is:¶
masque-URI = "masque:" "//" host ":" port "/"¶
The "host" and "port" component MUST NOT be empty, and the "port" component MUST NOT be 0.¶
This document will request IANA to create a "CONNECT-UDP Stream Chunk Type" registry. This registry governs a 62-bit space, and follows the registration policy for QUIC registries as defined in [QUIC]. In addition to the fields required by the QUIC policy, registrations in this registry MUST include the following fields:¶
The initial contents of this registry are:¶
+-------+------------+-----------------------+---------------+ | Value | Type | Description | Reference | +-------+------------+-----------------------+---------------+ | 0x00 | UDP_PACKET | Payload of UDP packet | This document | +-------+------------+-----------------------+---------------+¶
Each value of the format 37 * N + 23
for integer values of N (that is, 23,
60, 97, ...) are reserved; these values MUST NOT be assigned by IANA and MUST
NOT appear in the listing of assigned values.¶
This proposal was inspired directly or indirectly by prior work from many people. The author would like to thank Eric Rescorla for suggesting to use an HTTP method to proxy UDP. Thanks to Lucas Pardue for their inputs on this document.¶