RFC 9298




Internet Engineering Task Force (IETF)                       D. Schinazi
Request for Comments: 9298                                    Google LLC
Category: Standards Track                                    August 2022
ISSN: 2070-1721


                          Proxying UDP in HTTP

Abstract



   This document describes how to proxy UDP in HTTP, similar to how the
   HTTP CONNECT method allows proxying TCP in HTTP.  More specifically,
   this document defines a protocol that allows an HTTP client to create
   a tunnel for UDP communications through an HTTP server that acts as a
   proxy.

Status of This Memo



   This is an Internet Standards Track document.

   This document is a product of the Internet Engineering Task Force
   (IETF).  It represents the consensus of the IETF community.  It has
   received public review and has been approved for publication by the
   Internet Engineering Steering Group (IESG).  Further information on
   Internet Standards is available in Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   https://www.rfc-editor.org/info/rfc9298.

Copyright Notice



   Copyright (c) 2022 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 Revised BSD License text as described in Section 4.e of the
   Trust Legal Provisions and are provided without warranty as described
   in the Revised BSD License.

Table of Contents



   1.  Introduction
     1.1.  Conventions and Definitions
   2.  Client Configuration
   3.  Tunneling UDP over HTTP
     3.1.  UDP Proxy Handling
     3.2.  HTTP/1.1 Request
     3.3.  HTTP/1.1 Response
     3.4.  HTTP/2 and HTTP/3 Requests
     3.5.  HTTP/2 and HTTP/3 Responses
   4.  Context Identifiers
   5.  HTTP Datagram Payload Format
   6.  Performance Considerations
     6.1.  MTU Considerations
     6.2.  Tunneling of ECN Marks
   7.  Security Considerations
   8.  IANA Considerations
     8.1.  HTTP Upgrade Token
     8.2.  Well-Known URI
   9.  References
     9.1.  Normative References
     9.2.  Informative References
   Acknowledgments

   Author's Address



1.  Introduction



   While HTTP provides the CONNECT method (see Section 9.3.6 of [HTTP])
   for creating a TCP [TCP] tunnel to a proxy, it lacked a method for
   doing so for UDP [UDP] traffic prior to this specification.

   This document describes a protocol for tunneling UDP to a server
   acting as a UDP-specific proxy over HTTP.  UDP tunnels are commonly
   used to create an end-to-end virtual connection, which can then be
   secured using QUIC [QUIC] or another protocol running over UDP.
   Unlike the HTTP CONNECT method, the UDP proxy itself is identified
   with an absolute URL containing the traffic's destination.  Clients
   generate those URLs using a URI Template [TEMPLATE], as described in
   Section 2.

   This protocol supports all existing versions of HTTP by using HTTP
   Datagrams [HTTP-DGRAM].  When using HTTP/2 [HTTP/2] or HTTP/3
   [HTTP/3], it uses HTTP Extended CONNECT as described in
   [EXT-CONNECT2] and [EXT-CONNECT3].  When using HTTP/1.x [HTTP/1.1],
   it uses HTTP Upgrade as defined in Section 7.8 of [HTTP].

1.1.  Conventions and Definitions



   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 "UDP proxy" to refer to the HTTP
   server that acts upon the client's UDP tunneling request to open a
   UDP socket to a target server and that generates the response to this
   request.  If there are HTTP intermediaries (as defined in Section 3.7
   of [HTTP]) between the client and the UDP proxy, those are referred
   to as "intermediaries" in this document.

   Note that, when the HTTP version in use does not support multiplexing
   streams (such as HTTP/1.1), any reference to "stream" in this
   document represents the entire connection.

2.  Client Configuration



   HTTP clients are configured to use a UDP proxy with a URI Template
   [TEMPLATE] that has the variables "target_host" and "target_port".
   Examples are shown below:

 https://example.org/.well-known/masque/udp/{target_host}/{target_port}/
 https://proxy.example.org:4443/masque?h={target_host}&p={target_port}
 https://proxy.example.org:4443/masque{?target_host,target_port}

                    Figure 1: URI Template Examples

   The following requirements apply to the URI Template:

   *  The URI Template MUST be a level 3 template or lower.

   *  The URI Template MUST be in absolute form and MUST include non-
      empty scheme, authority, and path components.

   *  The path component of the URI Template MUST start with a slash
      ("/").

   *  All template variables MUST be within the path or query components
      of the URI.

   *  The URI Template MUST contain the two variables "target_host" and
      "target_port" and MAY contain other variables.

   *  The URI Template MUST NOT contain any non-ASCII Unicode characters
      and MUST only contain ASCII characters in the range 0x21-0x7E
      inclusive (note that percent-encoding is allowed; see Section 2.1
      of [URI]).

   *  The URI Template MUST NOT use Reserved Expansion ("+" operator),
      Fragment Expansion ("#" operator), Label Expansion with Dot-
      Prefix, Path Segment Expansion with Slash-Prefix, nor Path-Style
      Parameter Expansion with Semicolon-Prefix.

   Clients SHOULD validate the requirements above; however, clients MAY
   use a general-purpose URI Template implementation that lacks this
   specific validation.  If a client detects that any of the
   requirements above are not met by a URI Template, the client MUST
   reject its configuration and abort the request without sending it to
   the UDP proxy.

   The original HTTP CONNECT method allowed for the conveyance of the
   target host and port, but not the scheme, proxy authority, path, or
   query.  Thus, clients with proxy configuration interfaces that only
   allow the user to configure the proxy host and the proxy port exist.
   Client implementations of this specification that are constrained by
   such limitations MAY attempt to access UDP proxying capabilities
   using the default template, which is defined as
   "https://$PROXY_HOST:$PROXY_PORT/.well-known/masque/
   udp/{target_host}/{target_port}/", where $PROXY_HOST and $PROXY_PORT
   are the configured host and port of the UDP proxy, respectively.  UDP
   proxy deployments SHOULD offer service at this location if they need
   to interoperate with such clients.

3.  Tunneling UDP over HTTP



   To allow negotiation of a tunnel for UDP over HTTP, this document
   defines the "connect-udp" HTTP upgrade token.  The resulting UDP
   tunnels use the Capsule Protocol (see Section 3.2 of [HTTP-DGRAM])
   with HTTP Datagrams in the format defined in Section 5.

   To initiate a UDP tunnel associated with a single HTTP stream, a
   client issues a request containing the "connect-udp" upgrade token.
   The target of the tunnel is indicated by the client to the UDP proxy
   via the "target_host" and "target_port" variables of the URI
   Template; see Section 2.

   "target_host" supports using DNS names, IPv6 literals and IPv4
   literals.  Note that IPv6 scoped addressing zone identifiers are not
   supported.  Using the terms IPv6address, IPv4address, reg-name, and
   port from [URI], the "target_host" and "target_port" variables MUST
   adhere to the format in Figure 2, using notation from [ABNF].
   Additionally:

   *  both the "target_host" and "target_port" variables MUST NOT be
      empty.

   *  if "target_host" contains an IPv6 literal, the colons (":") MUST
      be percent-encoded.  For example, if the target host is
      "2001:db8::42", it will be encoded in the URI as
      "2001%3Adb8%3A%3A42".

   *  "target_port" MUST represent an integer between 1 and 65535
      inclusive.

   target_host = IPv6address / IPv4address / reg-name
   target_port = port

                   Figure 2: URI Template Variable Format

   When sending its UDP proxying request, the client SHALL perform URI
   Template expansion to determine the path and query of its request.

   If the request is successful, the UDP proxy commits to converting
   received HTTP Datagrams into UDP packets, and vice versa, until the
   tunnel is closed.

   By virtue of the definition of the Capsule Protocol (see Section 3.2
   of [HTTP-DGRAM]), UDP proxying requests do not carry any message
   content.  Similarly, successful UDP proxying responses also do not
   carry any message content.

3.1.  UDP Proxy Handling



   Upon receiving a UDP proxying request:

   *  if the recipient is configured to use another HTTP proxy, it will
      act as an intermediary by forwarding the request to another HTTP
      server.  Note that such intermediaries may need to re-encode the
      request if they forward it using a version of HTTP that is
      different from the one used to receive it, as the request encoding
      differs by version (see below).

   *  otherwise, the recipient will act as a UDP proxy.  It extracts the
      "target_host" and "target_port" variables from the URI it has
      reconstructed from the request headers, decodes their percent-
      encoding, and establishes a tunnel by directly opening a UDP
      socket to the requested target.

   Unlike TCP, UDP is connectionless.  The UDP proxy that opens the UDP
   socket has no way of knowing whether the destination is reachable.
   Therefore, it needs to respond to the request without waiting for a
   packet from the target.  However, if the "target_host" is a DNS name,
   the UDP proxy MUST perform DNS resolution before replying to the HTTP
   request.  If errors occur during this process, the UDP proxy MUST
   reject the request and SHOULD send details using an appropriate
   Proxy-Status header field [PROXY-STATUS].  For example, if DNS
   resolution returns an error, the proxy can use the dns_error Proxy
   Error Type from Section 2.3.2 of [PROXY-STATUS].

   UDP proxies can use connected UDP sockets if their operating system
   supports them, as that allows the UDP proxy to rely on the kernel to
   only send it UDP packets that match the correct 5-tuple.  If the UDP
   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 request.  Packets that do not match MUST be discarded by
   the UDP proxy.

   The lifetime of the socket is tied to the request stream.  The UDP
   proxy MUST keep the socket open while the request stream is open.  If
   a UDP proxy is notified by its operating system that its socket is no
   longer usable, it MUST close the request stream.  For example, this
   can happen when an ICMP Destination Unreachable message is received;
   see Section 3.1 of [ICMP6].  UDP proxies MAY choose to close sockets
   due to a period of inactivity, but they MUST close the request stream
   when closing the socket.  UDP proxies that close sockets after a
   period of inactivity SHOULD NOT use a period lower than two minutes;
   see Section 4.3 of [BEHAVE].

   A successful response (as defined in Sections 3.3 and 3.5) indicates
   that the UDP proxy has opened a socket to the requested target and is
   willing to proxy UDP payloads.  Any response other than a successful
   response indicates that the request has failed; thus, the client MUST
   abort the request.

   UDP proxies MUST NOT introduce fragmentation at the IP layer when
   forwarding HTTP Datagrams onto a UDP socket; overly large datagrams
   are silently dropped.  In IPv4, the Don't Fragment (DF) bit MUST be
   set, if possible, to prevent fragmentation on the path.  Future
   extensions MAY remove these requirements.

   Implementers of UDP proxies will benefit from reading the guidance in
   [UDP-USAGE].

3.2.  HTTP/1.1 Request



   When using HTTP/1.1 [HTTP/1.1], a UDP proxying request will meet the
   following requirements:

   *  the method SHALL be "GET".

   *  the request SHALL include a single Host header field containing
      the origin of the UDP proxy.

   *  the request SHALL include a Connection header field with value
      "Upgrade" (note that this requirement is case-insensitive as per
      Section 7.6.1 of [HTTP]).

   *  the request SHALL include an Upgrade header field with value
      "connect-udp".

   A UDP proxying request that does not conform to these restrictions is
   malformed.  The recipient of such a malformed request MUST respond
   with an error and SHOULD use the 400 (Bad Request) status code.

   For example, if the client is configured with URI Template
   "https://example.org/.well-known/masque/
   udp/{target_host}/{target_port}/" and wishes to open a UDP proxying
   tunnel to target 192.0.2.6:443, it could send the following request:

  GET https://example.org/.well-known/masque/udp/192.0.2.6/443/ HTTP/1.1
  Host: example.org
  Connection: Upgrade
  Upgrade: connect-udp
  Capsule-Protocol: ?1

                    Figure 3: Example HTTP/1.1 Request

   In HTTP/1.1, this protocol uses the GET method to mimic the design of
   the WebSocket Protocol [WEBSOCKET].

3.3.  HTTP/1.1 Response



   The UDP proxy SHALL indicate a successful response by replying with
   the following requirements:

   *  the HTTP status code on the response SHALL be 101 (Switching
      Protocols).

   *  the response SHALL include a Connection header field with value
      "Upgrade" (note that this requirement is case-insensitive as per
      Section 7.6.1 of [HTTP]).

   *  the response SHALL include a single Upgrade header field with
      value "connect-udp".

   *  the response SHALL meet the requirements of HTTP responses that
      start the Capsule Protocol; see Section 3.2 of [HTTP-DGRAM].

   If any of these requirements are not met, the client MUST treat this
   proxying attempt as failed and abort the connection.

   For example, the UDP proxy could respond with:

   HTTP/1.1 101 Switching Protocols
   Connection: Upgrade
   Upgrade: connect-udp
   Capsule-Protocol: ?1

                    Figure 4: Example HTTP/1.1 Response

3.4.  HTTP/2 and HTTP/3 Requests



   When using HTTP/2 [HTTP/2] or HTTP/3 [HTTP/3], UDP proxying requests
   use HTTP Extended CONNECT.  This requires that servers send an HTTP
   Setting as specified in [EXT-CONNECT2] and [EXT-CONNECT3] and that
   requests use HTTP pseudo-header fields with the following
   requirements:

   *  The :method pseudo-header field SHALL be "CONNECT".

   *  The :protocol pseudo-header field SHALL be "connect-udp".

   *  The :authority pseudo-header field SHALL contain the authority of
      the UDP proxy.

   *  The :path and :scheme pseudo-header fields SHALL NOT be empty.
      Their values SHALL contain the scheme and path from the URI
      Template after the URI Template expansion process has been
      completed.

   A UDP proxying request that does not conform to these restrictions is
   malformed (see Section 8.1.1 of [HTTP/2] and Section 4.1.2 of
   [HTTP/3]).

   For example, if the client is configured with URI Template
   "https://example.org/.well-known/masque/
   udp/{target_host}/{target_port}/" and wishes to open a UDP proxying
   tunnel to target 192.0.2.6:443, it could send the following request:

   HEADERS
   :method = CONNECT
   :protocol = connect-udp
   :scheme = https
   :path = /.well-known/masque/udp/192.0.2.6/443/
   :authority = example.org
   capsule-protocol = ?1

                      Figure 5: Example HTTP/2 Request

3.5.  HTTP/2 and HTTP/3 Responses



   The UDP proxy SHALL indicate a successful response by replying with
   the following requirements:

   *  the HTTP status code on the response SHALL be in the 2xx
      (Successful) range.

   *  the response SHALL meet the requirements of HTTP responses that
      start the Capsule Protocol; see Section 3.2 of [HTTP-DGRAM].

   If any of these requirements are not met, the client MUST treat this
   proxying attempt as failed and abort the request.

   For example, the UDP proxy could respond with:

   HEADERS
   :status = 200
   capsule-protocol = ?1

                     Figure 6: Example HTTP/2 Response

4.  Context Identifiers



   The mechanism for proxying UDP in HTTP defined in this document
   allows future extensions to exchange HTTP Datagrams that carry
   different semantics from UDP payloads.  Some of these extensions can
   augment UDP payloads with additional data, while others can exchange
   data that is completely separate from UDP payloads.  In order to
   accomplish this, all HTTP Datagrams associated with UDP Proxying
   request streams start with a Context ID field; see Section 5.

   Context IDs are 62-bit integers (0 to 2^62-1).  Context IDs are
   encoded as variable-length integers; see Section 16 of [QUIC].  The
   Context ID value of 0 is reserved for UDP payloads, while non-zero
   values are dynamically allocated.  Non-zero even-numbered Context IDs
   are client-allocated, and odd-numbered Context IDs are proxy-
   allocated.  The Context ID namespace is tied to a given HTTP request;
   it is possible for a Context ID with the same numeric value to be
   simultaneously allocated in distinct requests, potentially with
   different semantics.  Context IDs MUST NOT be re-allocated within a
   given HTTP namespace but MAY be allocated in any order.  The Context
   ID allocation restrictions to the use of even-numbered and odd-
   numbered Context IDs exist in order to avoid the need for
   synchronization between endpoints.  However, once a Context ID has
   been allocated, those restrictions do not apply to the use of the
   Context ID; it can be used by any client or UDP proxy, independent of
   which endpoint initially allocated it.

   Registration is the action by which an endpoint informs its peer of
   the semantics and format of a given Context ID.  This document does
   not define how registration occurs.  Future extensions MAY use HTTP
   header fields or capsules to register Context IDs.  Depending on the
   method being used, it is possible for datagrams to be received with
   Context IDs that have not yet been registered.  For instance, this
   can be due to reordering of the packet containing the datagram and
   the packet containing the registration message during transmission.

5.  HTTP Datagram Payload Format



   When HTTP Datagrams (see Section 2 of [HTTP-DGRAM]) are associated
   with UDP Proxying request streams, the HTTP Datagram Payload field
   has the format defined in Figure 7, using notation from Section 1.3
   of [QUIC].  Note that when HTTP Datagrams are encoded using QUIC
   DATAGRAM frames [QUIC-DGRAM], the Context ID field defined below
   directly follows the Quarter Stream ID field, which is at the start
   of the QUIC DATAGRAM frame payload; see Section 2.1 of [HTTP-DGRAM].

   UDP Proxying HTTP Datagram Payload {
     Context ID (i),
     UDP Proxying Payload (..),
   }

                Figure 7: UDP Proxying HTTP Datagram Format

   Context ID:  A variable-length integer (see Section 16 of [QUIC])
      that contains the value of the Context ID.  If an HTTP/3 Datagram
      that carries an unknown Context ID is received, the receiver SHALL
      either drop that datagram silently or buffer it temporarily (on
      the order of a round trip) while awaiting the registration of the
      corresponding Context ID.
   UDP Proxying Payload:  The payload of the datagram, whose semantics
      depend on the value of the previous field.  Note that this field
      can be empty.

   UDP packets are encoded using HTTP Datagrams with the Context ID
   field set to zero.  When the Context ID field is set to zero, the UDP
   Proxying Payload field contains the unmodified payload of a UDP
   packet (referred to as data octets in [UDP]).

   By virtue of the definition of the UDP header [UDP], it is not
   possible to encode UDP payloads longer than 65527 bytes.  Therefore,
   endpoints MUST NOT send HTTP Datagrams with a UDP Proxying Payload
   field longer than 65527 using Context ID zero.  An endpoint that
   receives an HTTP Datagram using Context ID zero whose UDP Proxying
   Payload field is longer than 65527 MUST abort the corresponding
   stream.  If a UDP proxy knows it can only send out UDP packets of a
   certain length due to its underlying link MTU, it has no choice but
   to discard incoming HTTP Datagrams using Context ID zero whose UDP
   Proxying Payload field is longer than that limit.  If the discarded
   HTTP Datagram was transported by a DATAGRAM capsule, the receiver
   SHOULD discard that capsule without buffering the capsule contents.

   If a UDP proxy receives an HTTP Datagram before it has received the
   corresponding request, it SHALL either drop that HTTP Datagram
   silently or buffer it temporarily (on the order of a round trip)
   while awaiting the corresponding request.

   Note that buffering datagrams (either because the request was not yet
   received or because the Context ID is not yet known) consumes
   resources.  Receivers that buffer datagrams SHOULD apply buffering
   limits in order to reduce the risk of resource exhaustion occurring.
   For example, receivers can limit the total number of buffered
   datagrams or the cumulative size of buffered datagrams on a per-
   stream, per-context, or per-connection basis.

   A client MAY optimistically start sending UDP packets in HTTP
   Datagrams before receiving the response to its UDP proxying request.
   However, implementers should note that such proxied packets may not
   be processed by the UDP proxy if it responds to the request with a
   failure or if the proxied packets are received by the UDP proxy
   before the request and the UDP proxy chooses to not buffer them.

6.  Performance Considerations



   Bursty traffic can often lead to temporally correlated packet losses;
   in turn, this can lead to suboptimal responses from congestion
   controllers in protocols running over UDP.  To avoid this, UDP
   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.  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.

   If a client or UDP proxy with a connection containing a UDP Proxying
   request stream disables congestion control, it MUST NOT signal
   Explicit Congestion Notification (ECN) [ECN] support on that
   connection.  That is, it MUST mark all IP headers with the Not-ECT
   codepoint.  It MAY continue to report ECN feedback via QUIC ACK_ECN
   frames or the TCP ECE bit, as the peer may not have disabled
   congestion control.

   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, UDP proxying
   SHOULD be performed over HTTP/3 to allow leveraging the QUIC DATAGRAM
   frame.

6.1.  MTU Considerations



   When using HTTP/3 with the QUIC Datagram extension [QUIC-DGRAM], UDP
   payloads are transmitted in QUIC DATAGRAM frames.  Since those cannot
   be fragmented, they can only carry payloads up to a given length
   determined by the QUIC connection configuration and the Path MTU
   (PMTU).  If a UDP proxy is using QUIC DATAGRAM frames and it receives
   a UDP payload from the target that will not fit inside a QUIC
   DATAGRAM frame, the UDP proxy SHOULD NOT send the UDP payload in a
   DATAGRAM capsule, as that defeats the end-to-end unreliability
   characteristic that methods such as Datagram Packetization Layer PMTU
   Discovery (DPLPMTUD) depend on [DPLPMTUD].  In this scenario, the UDP
   proxy SHOULD drop the UDP payload and send an ICMP Packet Too Big
   message to the target; see Section 3.2 of [ICMP6].

6.2.  Tunneling of ECN Marks



   UDP proxying does not create an IP-in-IP tunnel, so the guidance in
   [ECN-TUNNEL] about transferring ECN marks between inner and outer IP
   headers does not apply.  There is no inner IP header in UDP proxying
   tunnels.

   In this specification, note that UDP proxying clients do not have the
   ability to control the ECN codepoints on UDP packets the UDP proxy
   sends to the target, nor can UDP proxies communicate the markings of
   each UDP packet from target to UDP proxy.

   A UDP proxy MUST ignore ECN bits in the IP header of UDP packets
   received from the target, and it MUST set the ECN bits to Not-ECT on
   UDP packets it sends to the target.  These do not relate to the ECN
   markings of packets sent between client and UDP proxy in any way.

7.  Security Considerations



   There are significant risks in allowing arbitrary clients to
   establish a tunnel to arbitrary targets, as that could allow bad
   actors to send traffic and have it attributed to the UDP proxy.  HTTP
   servers that support UDP proxying ought to restrict its use to
   authenticated users.

   There exist software and network deployments that perform access
   control checks based on the source IP address of incoming requests.
   For example, some software allows unauthenticated configuration
   changes if they originated from 127.0.0.1.  Such software could be
   running on the same host as the UDP proxy or in the same broadcast
   domain.  Proxied UDP traffic would then be received with a source IP
   address belonging to the UDP proxy.  If this source address is used
   for access control, UDP proxying clients could use the UDP proxy to
   escalate their access privileges beyond those they might otherwise
   have.  This could lead to unauthorized access by UDP proxying clients
   unless the UDP proxy disallows UDP proxying requests to vulnerable
   targets, such as the UDP proxy's own addresses and localhost, link-
   local, multicast, and broadcast addresses.  UDP proxies can use the
   destination_ip_prohibited Proxy Error Type from Section 2.3.5 of
   [PROXY-STATUS] when rejecting such requests.

   UDP proxies share many similarities with TCP CONNECT proxies when
   considering them as infrastructure for abuse to enable denial-of-
   service (DoS) attacks.  Both can obfuscate the attacker's source
   address from the attack target.  In the case of a stateless
   volumetric attack (e.g., a TCP SYN flood or a UDP flood), both types
   of proxies pass the traffic to the target host.  With stateful
   volumetric attacks (e.g., HTTP flooding) being sent over a TCP
   CONNECT proxy, the proxy will only send data if the target has
   indicated its willingness to accept data by responding with a TCP
   SYN-ACK.  Once the path to the target is flooded, the TCP CONNECT
   proxy will no longer receive replies from the target and will stop
   sending data.  Since UDP does not establish shared state between the
   UDP proxy and the target, the UDP proxy could continue sending data
   to the target in such a situation.  While a UDP proxy could
   potentially limit the number of UDP packets it is willing to forward
   until it has observed a response from the target, that provides
   limited protection against DoS attacks when attacks target open UDP
   ports where the protocol running over UDP would respond and that
   would be interpreted as willingness to accept UDP by the UDP proxy.
   Such a packet limit could also cause issues for valid traffic.

   The security considerations described in Section 4 of [HTTP-DGRAM]
   also apply here.  Since it is possible to tunnel IP packets over UDP,
   the guidance in [TUNNEL-SECURITY] can apply.

8.  IANA Considerations



8.1.  HTTP Upgrade Token



   IANA has registered "connect-udp" in the "HTTP Upgrade Tokens"
   registry maintained at <https://www.iana.org/assignments/http-
   upgrade-tokens>.

   Value:  connect-udp
   Description:  Proxying of UDP Payloads
   Expected Version Tokens:  None
   Reference:  RFC 9298

8.2.  Well-Known URI



   IANA has registered "masque" in the "Well-Known URIs" registry
   maintained at <https://www.iana.org/assignments/well-known-uris>.

   URI Suffix:  masque
   Change Controller:  IETF
   Reference:  RFC 9298
   Status:  permanent
   Related Information:  Includes all resources identified with the path
      prefix "/.well-known/masque/udp/"

9.  References



9.1.  Normative References



   [ABNF]     Crocker, D., Ed. and P. Overell, "Augmented BNF for Syntax
              Specifications: ABNF", RFC 2234, DOI 10.17487/RFC2234,
              November 1997, <https://www.rfc-editor.org/info/rfc2234>.

   [ECN]      Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
              of Explicit Congestion Notification (ECN) to IP",
              RFC 3168, DOI 10.17487/RFC3168, September 2001,
              <https://www.rfc-editor.org/info/rfc3168>.

   [EXT-CONNECT2]
              McManus, P., "Bootstrapping WebSockets with HTTP/2",
              RFC 8441, DOI 10.17487/RFC8441, September 2018,
              <https://www.rfc-editor.org/info/rfc8441>.

   [EXT-CONNECT3]
              Hamilton, R., "Bootstrapping WebSockets with HTTP/3",
              RFC 9220, DOI 10.17487/RFC9220, June 2022,
              <https://www.rfc-editor.org/info/rfc9220>.

   [HTTP]     Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke,
              Ed., "HTTP Semantics", STD 97, RFC 9110,
              DOI 10.17487/RFC9110, June 2022,
              <https://www.rfc-editor.org/info/rfc9110>.

   [HTTP-DGRAM]
              Schinazi, D. and L. Pardue, "HTTP Datagrams and the
              Capsule Protocol", RFC 9297, DOI 10.17487/RFC9297, August
              2022, <https://www.rfc-editor.org/info/rfc9297>.

   [HTTP/1.1] Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke,
              Ed., "HTTP/1.1", STD 99, RFC 9112, DOI 10.17487/RFC9112,
              June 2022, <https://www.rfc-editor.org/info/rfc9112>.

   [HTTP/2]   Thomson, M., Ed. and C. Benfield, Ed., "HTTP/2", RFC 9113,
              DOI 10.17487/RFC9113, June 2022,
              <https://www.rfc-editor.org/info/rfc9113>.

   [HTTP/3]   Bishop, M., Ed., "HTTP/3", RFC 9114, DOI 10.17487/RFC9114,
              June 2022, <https://www.rfc-editor.org/info/rfc9114>.

   [PROXY-STATUS]
              Nottingham, M. and P. Sikora, "The Proxy-Status HTTP
              Response Header Field", RFC 9209, DOI 10.17487/RFC9209,
              June 2022, <https://www.rfc-editor.org/info/rfc9209>.

   [QUIC]     Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based
              Multiplexed and Secure Transport", RFC 9000,
              DOI 10.17487/RFC9000, May 2021,
              <https://www.rfc-editor.org/info/rfc9000>.

   [QUIC-DGRAM]
              Pauly, T., Kinnear, E., and D. Schinazi, "An Unreliable
              Datagram Extension to QUIC", RFC 9221,
              DOI 10.17487/RFC9221, March 2022,
              <https://www.rfc-editor.org/info/rfc9221>.

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <https://www.rfc-editor.org/info/rfc2119>.

   [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
              2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
              May 2017, <https://www.rfc-editor.org/info/rfc8174>.

   [TCP]      Eddy, W., Ed., "Transmission Control Protocol (TCP)",
              STD 7, RFC 9293, DOI 10.17487/RFC9293, August 2022,
              <https://www.rfc-editor.org/info/rfc9293>.

   [TEMPLATE] Gregorio, J., Fielding, R., Hadley, M., Nottingham, M.,
              and D. Orchard, "URI Template", RFC 6570,
              DOI 10.17487/RFC6570, March 2012,
              <https://www.rfc-editor.org/info/rfc6570>.

   [UDP]      Postel, J., "User Datagram Protocol", STD 6, RFC 768,
              DOI 10.17487/RFC0768, August 1980,
              <https://www.rfc-editor.org/info/rfc768>.

   [URI]      Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform
              Resource Identifier (URI): Generic Syntax", STD 66,
              RFC 3986, DOI 10.17487/RFC3986, January 2005,
              <https://www.rfc-editor.org/info/rfc3986>.

9.2.  Informative References



   [BEHAVE]   Audet, F., Ed. and C. Jennings, "Network Address
              Translation (NAT) Behavioral Requirements for Unicast
              UDP", BCP 127, RFC 4787, DOI 10.17487/RFC4787, January
              2007, <https://www.rfc-editor.org/info/rfc4787>.

   [DPLPMTUD] Fairhurst, G., Jones, T., Tüxen, M., Rüngeler, I., and T.
              Völker, "Packetization Layer Path MTU Discovery for
              Datagram Transports", RFC 8899, DOI 10.17487/RFC8899,
              September 2020, <https://www.rfc-editor.org/info/rfc8899>.

   [ECN-TUNNEL]
              Briscoe, B., "Tunnelling of Explicit Congestion
              Notification", RFC 6040, DOI 10.17487/RFC6040, November
              2010, <https://www.rfc-editor.org/info/rfc6040>.

   [HELIUM]   Schwartz, B. M., "Hybrid Encapsulation Layer for IP and
              UDP Messages (HELIUM)", Work in Progress, Internet-Draft,
              draft-schwartz-httpbis-helium-00, 25 June 2018,
              <https://datatracker.ietf.org/doc/html/draft-schwartz-
              httpbis-helium-00>.

   [HiNT]     Pardue, L., "HTTP-initiated Network Tunnelling (HiNT)",
              Work in Progress, Internet-Draft, draft-pardue-httpbis-
              http-network-tunnelling-00, 2 July 2018,
              <https://datatracker.ietf.org/doc/html/draft-pardue-
              httpbis-http-network-tunnelling-00>.

   [ICMP6]    Conta, A., Deering, S., and M. Gupta, Ed., "Internet
              Control Message Protocol (ICMPv6) for the Internet
              Protocol Version 6 (IPv6) Specification", STD 89,
              RFC 4443, DOI 10.17487/RFC4443, March 2006,
              <https://www.rfc-editor.org/info/rfc4443>.

   [MASQUE-ORIGINAL]
              Schinazi, D., "The MASQUE Protocol", Work in Progress,
              Internet-Draft, draft-schinazi-masque-00, 28 February
              2019, <https://datatracker.ietf.org/doc/html/draft-
              schinazi-masque-00>.

   [TUNNEL-SECURITY]
              Krishnan, S., Thaler, D., and J. Hoagland, "Security
              Concerns with IP Tunneling", RFC 6169,
              DOI 10.17487/RFC6169, April 2011,
              <https://www.rfc-editor.org/info/rfc6169>.

   [UDP-USAGE]
              Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage
              Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085,
              March 2017, <https://www.rfc-editor.org/info/rfc8085>.

   [WEBSOCKET]
              Fette, I. and A. Melnikov, "The WebSocket Protocol",
              RFC 6455, DOI 10.17487/RFC6455, December 2011,
              <https://www.rfc-editor.org/info/rfc6455>.

Acknowledgments

   This document is a product of the MASQUE Working Group, and the
   author thanks all MASQUE enthusiasts for their contributions.  This
   proposal was inspired directly or indirectly by prior work from many
   people, in particular [HELIUM] by Ben Schwartz, [HiNT] by Lucas
   Pardue, and the original MASQUE Protocol [MASQUE-ORIGINAL] by the
   author of this document.

   The author would like to thank Eric Rescorla for suggesting the use
   of an HTTP method to proxy UDP.  The author is indebted to Mark
   Nottingham and Lucas Pardue for the many improvements they
   contributed to this document.  The extensibility design in this
   document came out of the HTTP Datagrams Design Team, whose members
   were Alan Frindell, Alex Chernyakhovsky, Ben Schwartz, Eric Rescorla,
   Lucas Pardue, Marcus Ihlar, Martin Thomson, Mike Bishop, Tommy Pauly,
   Victor Vasiliev, and the author of this document.

Author's Address



   David Schinazi
   Google LLC
   1600 Amphitheatre Parkway
   Mountain View, CA 94043
   United States of America
   Email: dschinazi.ietf@gmail.com