RFC 8803




Internet Engineering Task Force (IETF)               O. Bonaventure, Ed.
Request for Comments: 8803                                      Tessares
Category: Experimental                                 M. Boucadair, Ed.
ISSN: 2070-1721                                                   Orange
                                                           S. Gundavelli
                                                                   Cisco
                                                                  S. Seo
                                                           Korea Telecom
                                                              B. Hesmans
                                                                Tessares
                                                               July 2020


                       0-RTT TCP Convert Protocol

Abstract



   This document specifies an application proxy, called Transport
   Converter, to assist the deployment of TCP extensions such as
   Multipath TCP.  A Transport Converter may provide conversion service
   for one or more TCP extensions.  The conversion service is provided
   by means of the 0-RTT TCP Convert Protocol (Convert).

   This protocol provides 0-RTT (Zero Round-Trip Time) conversion
   service since no extra delay is induced by the protocol compared to
   connections that are not proxied.  Also, the Convert Protocol does
   not require any encapsulation (no tunnels whatsoever).

   This specification assumes an explicit model, where the Transport
   Converter is explicitly configured on hosts.  As a sample
   applicability use case, this document specifies how the Convert
   Protocol applies for Multipath TCP.

Status of This Memo



   This document is not an Internet Standards Track specification; it is
   published for examination, experimental implementation, and
   evaluation.

   This document defines an Experimental Protocol for the Internet
   community.  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).  Not
   all documents approved by the IESG are candidates for any level of
   Internet Standard; see 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/rfc8803.

Copyright Notice



   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.

Table of Contents



   1.  Introduction
     1.1.  The Problem
     1.2.  Network-Assisted Connections: The Rationale
     1.3.  Applicability Scope
   2.  Conventions and Definitions
   3.  Differences with SOCKSv5
   4.  Architecture and Behaviors
     4.1.  Functional Elements
     4.2.  Theory of Operation
     4.3.  Data Processing at the Transport Converter
     4.4.  Address Preservation vs. Address Sharing
       4.4.1.  Address Preservation
       4.4.2.  Address/Prefix Sharing
   5.  Sample Examples
     5.1.  Outgoing Converter-Assisted Multipath TCP Connections
     5.2.  Incoming Converter-Assisted Multipath TCP Connection
   6.  The Convert Protocol (Convert)
     6.1.  The Convert Fixed Header
     6.2.  Convert TLVs
       6.2.1.  Generic Convert TLV Format
       6.2.2.  Summary of Supported Convert TLVs
       6.2.3.  The Info TLV
       6.2.4.  Supported TCP Extensions TLV
       6.2.5.  Connect TLV
       6.2.6.  Extended TCP Header TLV
       6.2.7.  The Cookie TLV
       6.2.8.  Error TLV
   7.  Compatibility of Specific TCP Options with the Conversion
           Service
     7.1.  Base TCP Options
     7.2.  Window Scale (WS)
     7.3.  Selective Acknowledgments
     7.4.  Timestamp
     7.5.  Multipath TCP
     7.6.  TCP Fast Open
     7.7.  TCP-AO
   8.  Interactions with Middleboxes
   9.  Security Considerations
     9.1.  Privacy & Ingress Filtering
     9.2.  Authentication and Authorization Considerations
     9.3.  Denial of Service
     9.4.  Traffic Theft
     9.5.  Logging
   10. IANA Considerations
     10.1.  Convert Service Name
     10.2.  The Convert Protocol (Convert) Parameters
       10.2.1.  Convert Versions
       10.2.2.  Convert TLVs
       10.2.3.  Convert Error Messages
   11. References
     11.1.  Normative References
     11.2.  Informative References
   Appendix A.  Example Socket API Changes to Support the 0-RTT TCP
           Convert Protocol
     A.1.  Active Open (Client Side)
     A.2.  Passive Open (Converter Side)
   Acknowledgments
   Contributors

   Authors' Addresses



1.  Introduction



1.1.  The Problem



   Transport protocols like TCP evolve regularly [RFC7414].  TCP has
   been improved in different ways.  Some improvements such as changing
   the initial window size [RFC6928] or modifying the congestion control
   scheme can be applied independently on Clients and Servers.  Other
   improvements such as Selective Acknowledgments [RFC2018] or large
   windows [RFC7323] require a new TCP option or changing the semantics
   of some fields in the TCP header.  These modifications must be
   deployed on both Clients and Servers to be actually used on the
   Internet.  Experience with the latter class of TCP extensions reveals
   that their deployment can require many years.  Fukuda reports in
   [Fukuda2011] results of a decade of measurements showing the
   deployment of Selective Acknowledgments, Window Scale, and TCP
   Timestamps.  [ANRW17] describes measurements showing that TCP Fast
   Open (TFO) [RFC7413] is still not widely deployed.

   There are some situations where the transport stack used on Clients
   (or Servers) can be upgraded at a faster pace than the transport
   stack running on Servers (or Clients).  In those situations, Clients
   (or Servers) would typically want to benefit from the features of an
   improved transport protocol even if the Servers (or Clients) have not
   yet been upgraded.  Some assistance from the network to make use of
   these features is valuable.  For example, Performance Enhancing
   Proxies [RFC3135] and other service functions have been deployed as
   solutions to improve TCP performance over links with specific
   characteristics.

   Recent examples of TCP extensions include Multipath TCP (MPTCP)
   [RFC8684] or tcpcrypt [RFC8548].  Those extensions provide features
   that are interesting for Clients such as wireless devices.  With
   Multipath TCP, those devices could seamlessly use Wireless Local Area
   Network (WLAN) and cellular networks for bonding purposes, faster
   hand-overs, or better resiliency.  Unfortunately, deploying those
   extensions on both a wide range of Clients and Servers remains
   difficult.

   More recently, 5G bonding experimentation has been conducted into
   global range of the incumbent 4G (LTE) connectivity using newly
   devised Clients and a Multipath TCP proxy.  Even if the 5G and 4G
   bonding (that relies upon Multipath TCP) increases the bandwidth, it
   is also crucial to minimize latency entirely between end hosts
   regardless of whether intermediate nodes are inside or outside of the
   mobile core.  In order to handle Ultra-Reliable Low Latency
   Communication (URLLC) for the next-generation mobile network,
   Multipath TCP and its proxy mechanism such as the one used to provide
   Access Traffic Steering, Switching, and Splitting (ATSSS) must be
   optimized to reduce latency [TS23501].

1.2.  Network-Assisted Connections: The Rationale



   This document specifies an application proxy called Transport
   Converter.  A Transport Converter is a function that is installed by
   a network operator to aid the deployment of TCP extensions and to
   provide the benefits of such extensions to Clients in particular.  A
   Transport Converter may provide conversion service for one or more
   TCP extensions.  Which TCP extensions are eligible for the conversion
   service is deployment specific.  The conversion service is provided
   by means of the 0-RTT TCP Convert Protocol (Convert), which is an
   application-layer protocol that uses a specific TCP port number on
   the Converter.

   The Convert Protocol provides Zero Round-Trip Time (0-RTT) conversion
   service since no extra delay is induced by the protocol compared to
   connections that are not proxied.  Particularly, the Convert Protocol
   does not require extra signaling setup delays before making use of
   the conversion service.  The Convert Protocol does not require any
   encapsulation (no tunnels, whatsoever).

   The Transport Converter adheres to the main steps drawn in Section 3
   of [RFC1919].  In particular, a Transport Converter achieves the
   following:

   *  Listening for Client sessions;

   *  Receiving the address of the Server from the Client;

   *  Setting up a session to the Server;

   *  Relaying control messages and data between the Client and the
      Server;

   *  Performing access controls according to local policies.

   The main advantage of network-assisted conversion services is that
   they enable new TCP extensions to be used on a subset of the path
   between endpoints, which encourages the deployment of these
   extensions.  Furthermore, the Transport Converter allows the Client
   and the Server to directly negotiate TCP extensions for the sake of
   native support along the full path.

   The Convert Protocol is a generic mechanism to provide 0-RTT
   conversion service.  As a sample applicability use case, this
   document specifies how the Convert Protocol applies for Multipath
   TCP.  It is out of scope of this document to provide a comprehensive
   list of all potential conversion services.  Applicability documents
   may be defined in the future.

   This document does not assume that all the traffic is eligible for
   the network-assisted conversion service.  Only a subset of the
   traffic will be forwarded to a Transport Converter according to a set
   of policies.  These policies, and how they are communicated to
   endpoints, are out of scope.  Furthermore, it is possible to bypass
   the Transport Converter to connect directly to the Servers that
   already support the required TCP extension(s).

   This document assumes an explicit model in which a Client is
   configured with one or a list of Transport Converters (statically or
   through protocols such as [DHC-CONVERTER]).  Configuration means are
   outside the scope of this document.

   The use of a Transport Converter means that there is no end-to-end
   transport connection between the Client and Server.  This could
   potentially create problems in some scenarios such as those discussed
   in Section 4 of [RFC3135].  Some of these problems may not be
   applicable.  For example, a Transport Converter can inform a Client
   by means of Network Failure (65) or Destination Unreachable (97)
   error messages (Section 6.2.8) that it encounters a failure problem;
   the Client can react accordingly.  An endpoint, or its network
   administrator, can assess the benefit provided by the Transport
   Converter service versus the risk.  This is one reason why the
   Transport Converter functionality has to be explicitly requested by
   an endpoint.

   This document is organized as follows:

      Section 3 provides a brief overview of the differences between the
      well-known SOCKS protocol and the 0-RTT TCP Convert Protocol.

      Section 4 provides a brief explanation of the operation of
      Transport Converters.

      Section 5 includes a set of sample examples to illustrate the
      overall behavior.

      Section 6 describes the Convert Protocol.

      Section 7 discusses how Transport Converters can be used to
      support different TCP extensions.

      Section 8 then discusses the interactions with middleboxes.

      Section 9 focuses on security considerations.

      Appendix A describes how a TCP stack would need to support the
      protocol described in this document.

1.3.  Applicability Scope



   The 0-RTT TCP Convert Protocol specified in this document MUST be
   used in a single administrative domain deployment model.  That is,
   the entity offering the connectivity service to a Client is also the
   entity that owns and operates the Transport Converter, with no
   transit over a third-party network.

   Future deployment of Transport Converters by third parties MUST
   adhere to the mutual authentication requirements in Section 9.2 to
   prevent illegitimate traffic interception (Section 9.4) in
   particular.

2.  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.

3.  Differences with SOCKSv5



   Several IETF protocols provide proxy services, the closest to the
   0-RTT TCP Convert Protocol being the SOCKSv5 protocol [RFC1928].
   This protocol is already used to deploy Multipath TCP in some
   cellular networks (Section 2.2 of [RFC8041]).

   A SOCKS Client creates a connection to a SOCKS Proxy, exchanges
   authentication information, and indicates the IP address and port
   number of the target Server.  At this point, the SOCKS Proxy creates
   a connection towards the target Server and relays all data between
   the two proxied connections.  The operation of an implementation
   based on SOCKSv5 (without authentication) is illustrated in Figure 1.

   Client                SOCKS Proxy               Server
      |                       |                       |
      | --------------------> |                       |
      |         SYN           |                       |
      | <-------------------- |                       |
      |       SYN+ACK         |                       |
      | --------------------> |                       |
      |         ACK           |                       |
      |                       |                       |
      | --------------------> |                       |
      |Version=5, Auth Methods|                       |
      | <-------------------- |                       |
      |       Method          |                       |
      | --------------------> |                       |
      |Auth Request (unless "No auth" method negotiated)
      | <-------------------- |                       |
      |     Auth Response     |                       |
      | --------------------> |                       |
      | Connect Server:Port   | --------------------> |
      |                       |          SYN          |
      |                       | <-------------------- |
      |                       |        SYN+ACK        |
      | <-------------------- |                       |
      |      Succeeded        |                       |
      | --------------------> |                       |
      |       Data1           |                       |
      |                       | --------------------> |
      |                       |         Data1         |
      |                       | <-------------------- |
      |                       |         Data2         |
      | <-------------------- |                       |
      |          Data2        |                       |
                            ...

     Figure 1: Establishment of a TCP Connection through a SOCKS Proxy
                           without Authentication

   When SOCKS is used, an "end-to-end" connection between a Client and a
   Server becomes a sequence of two TCP connections that are glued
   together on the SOCKS Proxy.  The SOCKS Client and Server exchange
   control information at the beginning of the bytestream on the Client-
   Proxy connection.  The SOCKS Proxy then creates the connection with
   the target Server and then glues the two connections together so that
   all bytes sent by the application (Client) to the SOCKS Proxy are
   relayed to the Server and vice versa.

   The Convert Protocol is also used on TCP proxies that relay data
   between an upstream and a downstream connection, but there are
   important differences with SOCKSv5.  A first difference is that the
   0-RTT TCP Convert Protocol exchanges all the control information
   during the initial RTT.  This reduces the connection establishment
   delay compared to SOCKS, which requires two or more round-trip times
   before the establishment of the downstream connection towards the
   final destination.  In today's Internet, latency is an important
   metric, and various protocols have been tuned to reduce their latency
   [LOW-LATENCY].  A recently proposed extension to SOCKS leverages the
   TCP Fast Open (TFO) option [INTAREA-SOCKS] to reduce this delay.

   A second difference is that the Convert Protocol explicitly takes the
   TCP extensions into account.  By using the Convert Protocol, the
   Client can learn whether a given TCP extension is supported by the
   destination Server.  This enables the Client to bypass the Transport
   Converter when the Server supports the required TCP extension(s).
   Neither SOCKSv5 [RFC1928] nor the proposed SOCKSv6 [INTAREA-SOCKS]
   provide such a feature.

   A third difference is that a Transport Converter will only confirm
   the establishment of the connection initiated by the Client provided
   that the downstream connection has already been accepted by the
   Server.  If the Server refuses the connection establishment attempt
   from the Transport Converter, then the upstream connection from the
   Client is rejected as well.  This feature is important for
   applications that check the availability of a Server or use the time
   to connect as a hint on the selection of a Server [RFC8305].

   A fourth difference is that the 0-RTT TCP Convert Protocol only
   allows the Client to specify the IP address/port number of the
   destination Server and not a DNS name.  We evaluated an alternate
   design that included the DNS name of the remote peer instead of its
   IP address as in SOCKS [RFC1928].  However, that design was not
   adopted because it induces both an extra load and increased delays on
   the Transport Converter to handle and manage DNS resolution requests.
   Note that the name resolution at the Converter may fail (e.g.,
   private names discussed in Section 2.1 of [RFC6731]) or may not match
   the one that would be returned by a Client's resolution library
   (e.g., Section 2.2 of [RFC6731]).

4.  Architecture and Behaviors



4.1.  Functional Elements



   The Convert Protocol considers three functional elements:

   *  Clients

   *  Transport Converters

   *  Servers

   A Transport Converter is a network function that proxies all data
   exchanged over one upstream connection to one downstream connection
   and vice versa (Figure 2).  Thus, the Transport Converter maintains
   state that associates one upstream connection to a corresponding
   downstream connection.

   A connection can be initiated from both sides of the Transport
   Converter (External realm, Internal realm).

                              |
                              :
                              |
                        +------------+
   Client <- upstream ->| Transport  |<- downstream -> Server
            connection  | Converter  |   connection
                        +------------+
                              |
               Internal realm : External realm
                              |

     Figure 2: A Transport Converter Proxies Data between Pairs of TCP
                                Connections

   "Client" refers to a software instance embedded on a host that can
   reach a Transport Converter in the internal realm.  The "Client" can
   initiate connections via a Transport Converter (referred to as
   outgoing connections).  Also, the "Client" can accept incoming
   connections via a Transport Converter (referred to as incoming
   connections).

   A Transport Converter can be embedded in a standalone device or be
   activated as a service on a router.  How such a function is enabled
   is deployment specific.

   The architecture assumes that new software will be installed on the
   Client hosts to interact with one or more Transport Converters.
   Furthermore, the architecture allows for making use of new TCP
   extensions even if those are not supported by a given Server.

   A Client is configured, through means that are outside the scope of
   this document, with the names and/or addresses of one or more
   Transport Converters and the TCP extensions that they support.  The
   procedure for selecting a Transport Converter among a list of
   configured Transport Converters is outside the scope of this
   document.

   One of the benefits of this design is that different transport
   protocol extensions can be used on the upstream and the downstream
   connections.  This encourages the deployment of new TCP extensions
   until they are widely supported, in particular, by Servers.

   The architecture does not mandate anything on the Server side.

   Similar to SOCKS, the architecture does not interfere with end-to-end
   TLS connections [RFC8446] between the Client and the Server
   (Figure 3).  In other words, end-to-end TLS is supported in the
   presence of a Converter.

       Client             Transport                Server
          |               Converter                  |
          |                   |                      |
          /==========================================\
         |            End-to-end TLS                  |
          \==========================================/

       * TLS messages exchanged between the Client
         and the Server are not shown.

             Figure 3: End-to-end TLS via a Transport Converter

   It is out of scope of this document to elaborate on specific
   considerations related to the use of TLS in the Client-Converter
   connection leg to exchange Convert messages (in addition to the end-
   to-end TLS connection).  In particular, (1) assessment of whether
   0-RTT data mode discussed in Section 2.3 of [RFC8446] is safe under
   replay and (2) specification of a profile for its use (Appendix E.5
   of [RFC8446]) are out of scope.

4.2.  Theory of Operation



   At a high level, the objective of the Transport Converter is to allow
   the use a specific extension, e.g., Multipath TCP, on a subset of the
   path even if the peer does not support this extension.  This is
   illustrated in Figure 4 where the Client initiates a Multipath TCP
   connection with the Transport Converter (packets belonging to the
   Multipath TCP connection are shown with "===") while the Transport
   Converter uses a TCP connection with the Server.

       Client             Transport                Server
          |               Converter                  |
          |                   |                      |
          |==================>|--------------------->|
          |                   |                      |
          |<==================|<---------------------|
          |                   |                      |
         Multipath TCP packets     TCP packets

       Figure 4: An Example of 0-RTT Network-Assisted Outgoing MPTCP
                                 Connection

   The packets belonging to a connection established through a Transport
   Converter may follow a different path than the packets directly
   exchanged between the Client and the Server.  Deployments should
   minimize the possible additional delay by carefully selecting the
   location of the Transport Converter used to reach a given
   destination.

   When establishing a connection, the Client can, depending on local
   policies, either contact the Server directly (e.g., by sending a TCP
   SYN towards the Server) or create the connection via a Transport
   Converter.  In the latter case (that is, the conversion service is
   used), the Client initiates a connection towards the Transport
   Converter and indicates the IP address and port number of the Server
   within the connection establishment packet.  Doing so enables the
   Transport Converter to immediately initiate a connection towards that
   Server without experiencing an extra delay.  The Transport Converter
   waits until the receipt of the confirmation that the Server agrees to
   establish the connection before confirming it to the Client.

   The Client places the destination address and port number of the
   Server in the payload of the SYN sent to the Transport Converter to
   minimize connection establishment delays.  The Transport Converter
   maintains two connections that are combined together:

   *  The upstream connection is the one between the Client and the
      Transport Converter.

   *  The downstream connection is the one between the Transport
      Converter and the Server.

   Any user data received by the Transport Converter over the upstream
   (or downstream) connection is proxied over the downstream (or
   upstream) connection.

   Figure 5 illustrates the establishment of an outgoing TCP connection
   by a Client through a Transport Converter.

      |  Note: The information shown between brackets in Figure 5 (and
      |  other figures in the document) refers to Convert Protocol
      |  messages described in Section 6.

                           Transport
       Client              Converter              Server
          |                   |                      |
          |SYN [->Server:port]|         SYN          |
          |------------------>|--------------------->|
          |<------------------|<---------------------|
          |    SYN+ACK [ ]    |        SYN+ACK       |
          |        ...        |          ...         |

      Figure 5: Establishment of an Outgoing TCP Connection through a
                            Transport Converter

   The Client sends a SYN destined to the Transport Converter.  The
   payload of this SYN contains the address and port number of the
   Server.  The Transport Converter does not reply immediately to this
   SYN.  It first tries to create a TCP connection towards the target
   Server.  If this upstream connection succeeds, the Transport
   Converter confirms the establishment of the connection to the Client
   by returning a SYN+ACK and the first bytes of the bytestream contain
   information about the TCP options that were negotiated with the
   Server.  Also, a state entry is instantiated for this connection.
   This state entry is used by the Converter to handle subsequent
   messages belonging to the connection.

   The connection can also be established from the Internet towards a
   Client via a Transport Converter (Figure 6).  This is typically the
   case when the Client hosts an application Server that listens to a
   specific port number.  When the Converter receives an incoming SYN
   from a remote host, it checks if it can provide the conversion
   service for the destination IP address and destination port number of
   that SYN.  The Transport Converter receives this SYN because it is,
   for example, on the path between the remote host and the Client or it
   provides address-sharing service for the Client (Section 2 of
   [RFC6269]).  If the check fails, the packet is silently ignored by
   the Converter.  If the check is successful, the Converter tries to
   initiate a TCP connection towards the Client from its own address and
   using its configured TCP options.  In the SYN that corresponds to
   this connection attempt, the Transport Convert inserts a TLV message
   that indicates the source address and port number of the remote host.
   A transport session entry is created by the Converter for this
   connection.  SYN+ACK and ACK will then be exchanged between the
   Client, the Converter, and remote host to confirm the establishment
   of the connection.  The Converter uses the transport session entry to
   proxy packets belonging to the connection.

     Transport              Remote
       Client              Converter             Host (RH)
          |                   |                      |
          |SYN [<-RH IP@:port]|         SYN          |
          |<------------------|<---------------------|
          |------------------>|--------------------->|
          |    SYN+ACK [ ]    |        SYN+ACK       |
          |        ...        |          ...         |

      Figure 6: Establishment of an Incoming TCP Connection through a
                            Transport Converter

   Standard TCP (Section 3.4 of [RFC0793]) allows a SYN packet to carry
   data inside its payload but forbids the receiver from delivering it
   to the application until completion of the three-way-handshake.  To
   enable applications to exchange data in a TCP handshake, this
   specification follows an approach similar to TCP Fast Open [RFC7413]
   and thus, removes the constraint by allowing data in SYN packets to
   be delivered to the Transport Converter application.

   As discussed in [RFC7413], such change to TCP semantics raises two
   issues.  First, duplicate SYNs can cause problems for applications
   that rely on TCP; whether or not a given application is affected
   depends on the details of that application protocol.  Second, TCP
   suffers from SYN flooding attacks [RFC4987].  TFO solves these two
   problems for applications that can tolerate replays by using the TCP
   Fast Open option that includes a cookie.  However, the utilization of
   this option consumes space in the limited TCP header.  Furthermore,
   there are situations, as noted in Section 7.3 of [RFC7413], where it
   is possible to accept the payload of SYN packets without creating
   additional security risks such as a network where addresses cannot be
   spoofed and the Transport Converter only serves a set of hosts that
   are identified by these addresses.

   For these reasons, this specification does not mandate the use of the
   TCP Fast Open option when the Client sends a connection establishment
   packet towards a Transport Converter.  The Convert Protocol includes
   an optional Cookie TLV that provides similar protection as the TCP
   Fast Open option without consuming space in the TCP header.
   Furthermore, this design allows for the use of longer cookies than
   [RFC7413].

   If the downstream (or upstream) connection fails for some reason
   (excessive retransmissions, reception of an RST segment, etc.), then
   the Converter reacts by forcing the teardown of the upstream (or
   downstream) connection.  In particular, if an ICMP error message that
   indicates a hard error is received on the downstream connection, the
   Converter echoes the Code field of that ICMP message in a Destination
   Unreachable Error TLV (see Section 6.2.8) that it transmits to the
   Client.  Note that if an ICMP error message that indicates a soft
   error is received on the downstream connection, the Converter will
   retransmit the corresponding data until it is acknowledged or the
   connection times out.  A classification of ICMP soft and hard errors
   is provided in Table 1 of [RFC5461].

   The same reasoning applies when the upstream connection ends with an
   exchange of FIN segments.  In this case, the Converter will also
   terminate the downstream connection by using FIN segments.  If the
   downstream connection terminates with the exchange of FIN segments,
   the Converter should initiate a graceful termination of the upstream
   connection.

4.3.  Data Processing at the Transport Converter



   As mentioned in Section 4.2, the Transport Converter acts as a TCP
   proxy between the upstream connection (i.e., between the Client and
   the Transport Converter) and the downstream connection (i.e., between
   the Transport Converter and the Server).

   The control messages (i.e., the Convert messages discussed in
   Section 6) establish state (called transport session entry) in the
   Transport Converter that will enable it to proxy between the two TCP
   connections.

   The Transport Converter uses the transport session entry to proxy
   packets belonging to the connection.  An implementation example of a
   transport session entry for TCP connections is shown in Figure 7.

                      (C,c) <--> (T,t), (S,s), Lifetime

              Figure 7: An Example of Transport Session Entry

   Where:

   *  C and c are the source IP address and source port number used by
      the Client for the upstream connection.

   *  S and s are the Server's IP address and port number.

   *  T and t are the source IP address and source port number used by
      the Transport Converter to proxy the connection.

   *  Lifetime is a timer that tracks the remaining lifetime of the
      entry as assigned by the Converter.  When the timer expires, the
      entry is deleted.

   Clients send packets bound to connections eligible for the conversion
   service to the provisioned Transport Converter and destination port
   number.  This applies for both control messages and data.  Additional
   information is supplied by Clients to the Transport Converter by
   means of Convert messages as detailed in Section 6.  User data can be
   included in SYN or non-SYN messages.  User data is unambiguously
   distinguished from Convert TLVs by a Transport Converter owing to the
   Convert Fixed Header in the Convert messages (Section 6.1).  These
   Convert TLVs are destined to the Transport Convert and are, thus,
   removed by the Transport Converter when proxying between the two
   connections.

   Upon receipt of a packet that belongs to an existing connection
   between a Client and the Transport Converter, the Converter proxies
   the user data to the Server using the information stored in the
   corresponding transport session entry.  For example, in reference to
   Figure 7, the Transport Converter proxies the data received from
   (C,c) downstream using (T,t) as source transport address and (S,s) as
   destination transport address.

   A similar process happens for data sent from the Server.  The
   Converter acts as a TCP proxy and sends the data to the Client
   relying upon the information stored in a transport session entry.
   The Converter associates a lifetime with state entries used to bind
   an upstream connection with its downstream connection.

   When Multipath TCP is used between the Client and the Transport
   Converter, the Converter maintains more state (e.g., information
   about the subflows) for each Multipath TCP connection.  The procedure
   described above continues to apply except that the Converter needs to
   manage the establishment/termination of subflows and schedule packets
   among the established ones.  These operations are part of the
   Multipath TCP implementation.  They are independent of the Convert
   Protocol that only processes the Convert messages in the beginning of
   the bytestream.

   A Transport Converter may operate in address preservation mode (that
   is, the Converter does not rewrite the source IP address (i.e.,
   C==T)) or address-sharing mode (that is, an address pool is shared
   among all Clients serviced by the Converter (i.e., C!=T)); refer to
   Section 4.4 for more details.  Which behavior to use by a Transport
   Converter is deployment specific.  If address-sharing mode is
   enabled, the Transport Converter MUST adhere to REQ-2 of [RFC6888],
   which implies a default "IP address pooling" behavior of "Paired" (as
   defined in Section 4.1 of [RFC4787]) MUST be supported.  This
   behavior is meant to avoid breaking applications that depend on the
   source address remaining constant.

4.4.  Address Preservation vs. Address Sharing



   The Transport Converter is provided with instructions about the
   behavior to adopt with regard to the processing of source addresses
   of outgoing packets.  The following subsections discuss two
   deployment models for illustration purposes.  It is out of the scope
   of this document to make a recommendation.

4.4.1.  Address Preservation



   In this model, the visible source IP address of a packet proxied by a
   Transport Converter to a Server is an IP address of the end host
   (Client).  No dedicated IP address pool is provisioned to the
   Transport Converter, but the Transport Converter is located on the
   path between the Client and the Server.

   For Multipath TCP, the Transport Converter preserves the source IP
   address used by the Client when establishing the initial subflow.
   Data conveyed in secondary subflows will be proxied by the Transport
   Converter using the source IP address of the initial subflow.  An
   example of a proxied Multipath TCP connection with address
   preservation is shown in Figure 8.

                                         Transport
          Client                        Converter          Server

           @:C1,C2                        @:Tc                @:S
              ||                            |                  |
              |src:C1     SYN         dst:Tc|src:C1       dst:S|
              |-------MPC [->S:port]------->|-------SYN------->|
              ||                            |                  |
              ||dst:C1                src:Tc|dst:C1       src:S|
              |<---------SYN/ACK------------|<-----SYN/ACK-----|
              ||                            |                  |
              |src:C1                 dst:Tc|src:C1       dst:S|
              |------------ACK------------->|-------ACK------->|
              |                             |                  |
              |src:C2          ...    dst:Tc|       ...        |
              ||<-----Secondary Subflow---->|src:C1       dst:S|
              ||                            |-------data------>|
              |               ..            |    ...           |

   Legend:
     Tc: IP address used by the Transport Converter on the internal
         realm.

                 Figure 8: Example of Address Preservation

   The Transport Converter must be on the forwarding path of incoming
   traffic.  Because the same (destination) IP address is used for both
   proxied and non-proxied connections, the Transport Converter should
   not drop incoming packets it intercepts if no matching entry is found
   for the packets.  Unless explicitly configured otherwise, such
   packets are forwarded according to the instructions of a local
   forwarding table.

4.4.2.  Address/Prefix Sharing



   A pool of global IPv4 addresses is provisioned to the Transport
   Converter along with possible instructions about the address-sharing
   ratio to apply (see Appendix B of [RFC6269]).  An address is thus
   shared among multiple Clients.

   Likewise, rewriting the source IPv6 prefix [RFC6296] may be used to
   ease redirection of incoming IPv6 traffic towards the appropriate
   Transport Converter.  A pool of IPv6 prefixes is then provisioned to
   the Transport Converter for this purpose.

   Adequate forwarding policies are enforced so that traffic destined to
   an address of such a pool is intercepted by the appropriate Transport
   Converter.  Unlike Section 4.4.1, the Transport Converter drops
   incoming packets that do not match an active transport session entry.

   An example is shown in Figure 9.

                                        Transport
         Client                         Converter          Server

            @:C                        @:Tc|Te                @:S
             |                             |                  |
             |src:C                  dst:Tc|src:Te       dst:S|
             |-------SYN [->S:port]------->|-------SYN------->|
             |                             |                  |
             |dst:C                  src:Tc|dst:Te       src:S|
             |<---------SYN/ACK------------|<-----SYN/ACK-----|
             |                             |                  |
             |src:C                  dst:Tc|src:Te       dst:S|
             |------------ACK------------->|-------ACK------->|
             |                             |                  |
             |              ...            |    ...           |

   Legend:
     Tc: IP address used by the Transport Converter on the internal
         realm.
     Te: IP address used by the Transport Converter on the external
         realm.

                         Figure 9: Address Sharing

5.  Sample Examples



5.1.  Outgoing Converter-Assisted Multipath TCP Connections



   As an example, let us consider how the Convert Protocol can help the
   deployment of Multipath TCP.  We assume that both the Client and the
   Transport Converter support Multipath TCP but consider two different
   cases depending on whether or not the Server supports Multipath TCP.

   As a reminder, a Multipath TCP connection is created by placing the
   MP_CAPABLE (MPC) option in the SYN sent by the Client.

   Figure 10 describes the operation of the Transport Converter if the
   Server does not support Multipath TCP.

                           Transport
       Client              Converter              Server
          |SYN, MPC           |                      |
          |[->Server:port]    |         SYN, MPC     |
          |------------------>|--------------------->|
          |<------------------|<---------------------|
          |  SYN+ACK,MPC [.]  |      SYN+ACK         |
          |------------------>|--------------------->|
          |     ACK, MPC      |          ACK         |
          |        ...        |          ...         |

      Figure 10: Establishment of a Multipath TCP Connection through a
         Transport Converter towards a Server That Does Not support
                               Multipath TCP

   The Client tries to initiate a Multipath TCP connection by sending a
   SYN with the MP_CAPABLE option (MPC in Figure 10).  The SYN includes
   the address and port number of the target Server, that are extracted
   and used by the Transport Converter to initiate a Multipath TCP
   connection towards this Server.  Since the Server does not support
   Multipath TCP, it replies with a SYN+ACK that does not contain the
   MP_CAPABLE option.  The Transport Converter notes that the connection
   with the Server does not support Multipath TCP and returns the
   extended TCP header received from the Server to the Client.

   Note that, if the TCP connection is reset for some reason, the
   Converter tears down the Multipath TCP connection by transmitting an
   MP_FASTCLOSE.  Likewise, if the Multipath TCP connection ends with
   the transmission of DATA_FINs, the Converter terminates the TCP
   connection by using FIN segments.  As a side note, given that with
   Multipath TCP, RST only has the scope of the subflow and will only
   close the concerned subflow but not affect the remaining subflows,
   the Converter does not terminate the downstream TCP connection upon
   receipt of an RST over a Multipath subflow.

   Figure 11 considers a Server that supports Multipath TCP.  In this
   case, it replies to the SYN sent by the Transport Converter with the
   MP_CAPABLE option.  Upon reception of this SYN+ACK, the Transport
   Converter confirms the establishment of the connection to the Client
   and indicates to the Client that the Server supports Multipath TCP.
   With this information, the Client has discovered that the Server
   supports Multipath TCP.  This will enable the Client to bypass the
   Transport Converter for the subsequent Multipath TCP connections that
   it will initiate towards this Server.

                           Transport
       Client              Converter              Server
          |SYN, MPC           |                      |
          |[->Server:port]    |         SYN, MPC     |
          |------------------>|--------------------->|
          |<------------------|<---------------------|
          |SYN+ACK, MPC       |      SYN+ACK, MPC    |
          |[MPC supported]    |                      |
          |------------------>|--------------------->|
          |     ACK, MPC      |        ACK, MPC      |
          |        ...        |          ...         |

      Figure 11: Establishment of a Multipath TCP Connection through a
                 Converter towards an MPTCP-Capable Server

5.2.  Incoming Converter-Assisted Multipath TCP Connection



   An example of an incoming Converter-assisted Multipath TCP connection
   is depicted in Figure 12.  In order to support incoming connections
   from remote hosts, the Client may use the Port Control Protocol (PCP)
   [RFC6887] to instruct the Transport Converter to create dynamic
   mappings.  Those mappings will be used by the Transport Converter to
   intercept an incoming TCP connection destined to the Client and
   convert it into a Multipath TCP connection.

   Typically, the Client sends a PCP request to the Converter asking to
   create an explicit TCP mapping for the internal IP address and
   internal port number.  The Converter accepts the request by creating
   a TCP mapping for the internal IP address, internal port number,
   external IP address, and external port number.  The external IP
   address, external port number, and assigned lifetime are returned
   back to the Client in the PCP response.  The external IP address and
   external port number will then be advertised by the Client (or the
   user) using an out-of-band mechanism so that remote hosts can
   initiate TCP connections to the Client via the Converter.  Note that
   the external and internal information may be the same.

   Then, when the Converter receives an incoming SYN, it checks its
   mapping table to verify if there is an active mapping matching the
   destination IP address and destination port of that SYN.  If no entry
   is found, the Converter silently ignores the message.  If an entry is
   found, the Converter inserts an MP_CAPABLE option and Connect TLV in
   the SYN packet, and rewrites the source IP address to one of its IP
   addresses and, eventually, the destination IP address and port number
   in accordance with the information stored in the mapping.  SYN+ACK
   and ACK will then be exchanged between the Client and the Converter
   to confirm the establishment of the initial subflow.  The Client can
   add new subflows following normal Multipath TCP procedures.

                           Transport             Remote
       Client              Converter              Host
         |                     |                    |
         |<--------------------|<-------------------|
         |SYN, MPC             |         SYN        |
         |[Remote Host:port]   |                    |
         |-------------------->|------------------->|
         |      SYN+ACK, MPC   |       SYN+ACK      |
         |<--------------------|<-------------------|
         |       ACK, MPC      |           ACK      |
         |        ...          |          ...       |

      Figure 12: Establishment of an Incoming Multipath TCP Connection
                       through a Transport Converter

   It is out of scope of this document to define specific Convert TLVs
   to manage incoming connections (that is, TLVs that mimic PCP
   messages).  These TLVs can be defined in a separate document.

6.  The Convert Protocol (Convert)



   This section defines the Convert Protocol (Convert, for short)
   messages that are exchanged between a Client and a Transport
   Converter.

   The Transport Converter listens on a specific TCP port number for
   Convert messages from Clients.  That port number is configured by an
   administrator.  Absent any policy, the Transport Converter SHOULD
   silently ignore SYNs with no Convert TLVs.

   Convert messages may appear only in SYN, SYN+ACK, or ACK.

   Convert messages MUST be included as the first bytes of the
   bytestream.  All Convert messages start with a fixed header that is
   32 bits long (Section 6.1) followed by one or more Convert TLVs
   (Type, Length, Value) (Section 6.2).

   If the initial SYN message contains user data in its payload (e.g.,
   see [RFC7413]), that data MUST be placed right after the Convert TLVs
   when generating the SYN.

   The protocol can be extended by defining new TLVs or bumping the
   version number if a different message format is needed.  If a future
   version is defined but with a different message format, the version
   negotiation procedure defined in Section 6.2.8 (see "Unsupported
   Version") is meant to agree on a version that is supported by both
   peers.

      |  Implementation note 1: Several implementers expressed concerns
      |  about the use of TFO.  As a reminder, the Fast Open Cookie
      |  protects from some attack scenarios that affect open servers
      |  like web servers.  The Convert Protocol is different and, as
      |  discussed in [RFC7413], there are different ways to protect
      |  from such attacks.  Instead of using a Fast Open Cookie inside
      |  the TCP options, which consumes precious space in the extended
      |  TCP header, the Convert Protocol supports the utilization of a
      |  Cookie that is placed in the SYN payload.  This provides the
      |  same level of protection as a Fast Open Cookie in environments
      |  were such protection is required.
      |
      |  Implementation note 2: Error messages are not included in RST
      |  but sent in the bytestream.  Implementers have indicated that
      |  processing RST on Clients was difficult on some platforms.
      |  This design simplifies Client implementations.

6.1.  The Convert Fixed Header



   The Convert Protocol uses a fixed header that is 32 bits long sent by
   both the Client and the Transport Converter over each established
   connection.  This header indicates both the version of the protocol
   used and the length of the Convert message.

   The Client and the Transport Converter MUST send the fixed-sized
   header, shown in Figure 13, as the first four bytes of the
   bytestream.

                        1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +---------------+---------------+-------------------------------+
   |  Version      |  Total Length |          Magic Number         |
   +---------------+---------------+-------------------------------+

                    Figure 13: The Convert Fixed Header

   The version is encoded as an 8-bit unsigned integer value.  This
   document specifies version 1.  Version 0 is reserved by this document
   and MUST NOT be used.

      |  Note: Early versions of this specification don't use a
      |  dedicated port number but only rely upon the IP address of the
      |  Converter.  Having a bit set in the Version field together with
      |  the Total Length field avoids misinterpreting data in a SYN as
      |  Convert TLVs.  Since the design was updated to use a specific
      |  service port, that constraint was relaxed.  Version 0 would
      |  work, but given existing implementations already use Version 1,
      |  the use of Version 0 is maintained as reserved.

   The Total Length is the number of 32-bit words, including the header,
   of the bytestream that are consumed by the Convert messages.  Since
   Total Length is also an 8-bit unsigned integer, those messages cannot
   consume more than 1020 bytes of data.  This limits the number of
   bytes that a Transport Converter needs to process.  A Total Length of
   zero is invalid and the connection MUST be reset upon reception of a
   header with such a total length.

   The Magic Number field MUST be set to 0x2263.  This field is meant to
   further strengthen the protocol to unambiguously distinguish any data
   supplied by an application from Convert TLVs.

   The Total Length field unambiguously marks the number of 32-bit words
   that carry Convert TLVs in the beginning of the bytestream.

6.2.  Convert TLVs



6.2.1.  Generic Convert TLV Format



   The Convert Protocol uses variable length messages that are encoded
   using the generic TLV format depicted in Figure 14.

   The length of all TLVs used by the Convert Protocol is always a
   multiple of four bytes.  All TLVs are aligned on 32-bit boundaries.
   All TLV fields are encoded using the network byte order.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +---------------+---------------+-------------------------------+
      |     Type      |     Length    |             Value  ...        |
      +---------------+---------------+-------------------------------+
      //              ...   (optional) Value                         //
      +---------------------------------------------------------------+

                   Figure 14: Convert Generic TLV Format

   The Length field covers Type, Length, and Value fields.  It is
   expressed in units of 32-bit words.  If necessary, Value MUST be
   padded with zeroes so that the length of the TLV is a multiple of 32
   bits.

   A given TLV MUST only appear once on a connection.  If a Client
   receives two or more instances of the same TLV over a Convert
   connection, it MUST reset the associated TCP connection.  If a
   Converter receives two or more instances of the same TLV over a
   Convert connection, it MUST return a Malformed Message Error TLV and
   close the associated TCP connection.

6.2.2.  Summary of Supported Convert TLVs



   This document specifies the following Convert TLVs:

         +======+======+==========+==============================+
         | Type | Hex  | Length   | Description                  |
         +======+======+==========+==============================+
         | 1    | 0x1  | 1        | Info TLV                     |
         +------+------+----------+------------------------------+
         | 10   | 0xA  | Variable | Connect TLV                  |
         +------+------+----------+------------------------------+
         | 20   | 0x14 | Variable | Extended TCP Header TLV      |
         +------+------+----------+------------------------------+
         | 21   | 0x15 | Variable | Supported TCP Extensions TLV |
         +------+------+----------+------------------------------+
         | 22   | 0x16 | Variable | Cookie TLV                   |
         +------+------+----------+------------------------------+
         | 30   | 0x1E | Variable | Error TLV                    |
         +------+------+----------+------------------------------+

               Table 1: The TLVs Used by the Convert Protocol

   Type 0x0 is a reserved value.  If a Client receives a TLV of type
   0x0, it MUST reset the associated TCP connection.  If a Converter
   receives a TLV of type 0x0, it MUST return an Unsupported Message
   Error TLV and close the associated TCP connection.

   The Client typically sends, in the first connection it established
   with a Transport Converter, the Info TLV (Section 6.2.3) to learn its
   capabilities.  Assuming the Client is authorized to invoke the
   Transport Converter, the latter replies with the Supported TCP
   Extensions TLV (Section 6.2.4).

   The Client can request the establishment of connections to Servers by
   using the Connect TLV (Section 6.2.5).  If the connection can be
   established with the final Server, the Transport Converter replies
   with the Extended TCP Header TLV (Section 6.2.6).  If not, the
   Transport Converter MUST return an Error TLV (Section 6.2.8) and then
   close the connection.  The Transport Converter MUST NOT send an RST
   immediately after the detection of an error to let the Error TLV
   reach the Client.  As explained later, the Client will send an RST
   regardless upon reception of the Error TLV.

6.2.3.  The Info TLV



   The Info TLV (Figure 15) is an optional TLV that can be sent by a
   Client to request the TCP extensions that are supported by a
   Transport Converter.  It is typically sent on the first connection
   that a Client establishes with a Transport Converter to learn its
   capabilities.  Assuming a Client is entitled to invoke the Transport
   Converter, the latter replies with the Supported TCP Extensions TLV
   described in Section 6.2.4.

                        1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +---------------+---------------+-------------------------------+
   |     Type=0x1  |     Length    |             Zero              |
   +---------------+---------------+-------------------------------+

                          Figure 15: The Info TLV

6.2.4.  Supported TCP Extensions TLV



   The Supported TCP Extensions TLV (Figure 16) is used by a Transport
   Converter to announce the TCP options for which it provides a
   conversion service.  A Transport Converter SHOULD include in this
   list the TCP options that it supports in outgoing SYNs.

   Each supported TCP option is encoded with its TCP option Kind listed
   in the "Transmission Control Protocol (TCP) Parameters" registry
   maintained by IANA [IANA-CONVERT].  The Unassigned field MUST be set
   to zero by the Transport Converter and ignored by the Client.

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+-------------------------------+
    |     Type=0x15 |     Length    |           Unassigned          |
    +---------------+---------------+-------------------------------+
    |     Kind #1   |     Kind #2   |           ...                 |
    +---------------+---------------+-------------------------------+
    /                              ...                              /
    /                                                               /
    +---------------------------------------------------------------+

                Figure 16: The Supported TCP Extensions TLV

   TCP option Kinds 1 and 2 defined in [RFC0793] are supported by all
   TCP implementations and thus, MUST NOT appear in this list.

   The list of Supported TCP Extensions is padded with 0 to end on a
   32-bit boundary.

   For example, if the Transport Converter supports Multipath TCP,
   Kind=30 will be present in the Supported TCP Extensions TLV that it
   returns in response to the Info TLV.

6.2.5.  Connect TLV



   The Connect TLV (Figure 17) is used to request the establishment of a
   connection via a Transport Converter.  This connection can be from or
   to a Client.

   The Remote Peer Port and Remote Peer IP Address fields contain the
   destination port number and IP address of the Server, for outgoing
   connections.  For incoming connections destined to a Client serviced
   via a Transport Converter, these fields convey the source port number
   and IP address of the SYN packet received by the Transport Converter
   from the Server.

   The Remote Peer IP Address MUST be encoded as an IPv6 address.  IPv4
   addresses MUST be encoded using the IPv4-mapped IPv6 address format
   defined in [RFC4291].  Further, the Remote Peer IP Address field MUST
   NOT
include multicast, broadcast, or host loopback addresses
   [RFC6890].  If a Converter receives a Connect TLV with such invalid
   addresses, it MUST reply with a Malformed Message Error TLV and close
   the associated TCP connection.

   We distinguish two types of Connect TLV based on their length: (1)
   the Base Connect TLV has a length set to 5 (i.e., 20 bytes) and
   contains a remote address and a remote port (Figure 17), and (2) the
   Extended Connect TLV spans more than 20 bytes and also includes the
   optional TCP Options field (Figure 18).  This field is used to
   request the advertisement of specific TCP options to the Server.

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+-------------------------------+
    |     Type=0xA  |     Length    |      Remote Peer Port         |
    +---------------+---------------+-------------------------------+
    |                                                               |
    |         Remote Peer IP Address (128 bits)                     |
    |                                                               |
    |                                                               |
    +---------------------------------------------------------------+

                      Figure 17: The Base Connect TLV

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+-------------------------------+
    |     Type=0xA  |     Length    |      Remote Peer Port         |
    +---------------+---------------+-------------------------------+
    |                                                               |
    |         Remote Peer IP Address (128 bits)                     |
    |                                                               |
    |                                                               |
    +---------------------------------------------------------------+
    /                          TCP Options (Variable)               /
    /                              ...                              /
    +---------------------------------------------------------------+

                    Figure 18: The Extended Connect TLV

   The TCP Options field is a variable length field that carries a list
   of TCP option fields (Figure 19).  Each TCP option field is encoded
   as a block of 2+n bytes where the first byte is the TCP option Kind
   and the second byte is the length of the TCP option as specified in
   [RFC0793].  The minimum value for the TCP option Length is 2.  The
   TCP options that do not include a length sub-field, i.e., option
   types 0 (EOL) and 1 (NOP) defined in [RFC0793] MUST NOT be placed
   inside the TCP options field of the Connect TLV.  The optional Value
   field contains the variable-length part of the TCP option.  A length
   of 2 indicates the absence of the Value field.  The TCP options field
   always ends on a 32-bit boundary after being padded with zeros.

                        1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+---------------+---------------+
    |  TCPOpt kind  | TCPOpt Length | Value  (opt)  |  ....         |
    +---------------+---------------+---------------+---------------+
    |                             ....                              |
    +---------------------------------------------------------------+
    |                              ...                              |
    +---------------------------------------------------------------+

                      Figure 19: The TCP Options Field

   Upon reception of a Base Connect TLV, and absent any policy (e.g.,
   rate-limit) or resource exhaustion conditions, a Transport Converter
   attempts to establish a connection to the address and port that it
   contains.  The Transport Converter MUST use by default the TCP
   options that correspond to its local policy to establish this
   connection.

   Upon reception of an Extended Connect TLV, a Transport Converter
   first checks whether or not it supports the TCP Options listed in the
   TCP Options field.  If not, it returns an error TLV set to
   "Unsupported TCP Option" (Section 6.2.8).  If the above check
   succeeded, and absent any rate-limit policy or resource exhaustion
   conditions, a Transport Converter MUST attempt to establish a
   connection to the address and port that it contains.  It MUST include
   in the SYN that it sends to the Server the options listed in the TCP
   Options subfield and the TCP options that it would have used
   according to its local policies.  For the TCP options that are
   included in the TCP Options field without an optional value, the
   Transport Converter MUST generate its own value.  For the TCP options
   that are included in the TCP Options field with an optional value, it
   MUST copy the entire option in the SYN sent to the remote Server.
   This procedure is designed with TFO in mind.  Particularly, this
   procedure allows to successfully exchange a Fast Open Cookie between
   the Client and the Server.  See Section 7 for a detailed discussion
   of the different types of TCP options.

   The Transport Converter may refuse a Connect TLV request for various
   reasons (e.g., authorization failed, out of resources, invalid
   address type, or unsupported TCP option).  An error message
   indicating the encountered error is returned to the requesting Client
   (Section 6.2.8).  In order to prevent denial-of-service attacks,
   error messages sent to a Client SHOULD be rate-limited.

6.2.6.  Extended TCP Header TLV



   The Extended TCP Header TLV (Figure 20) is used by the Transport
   Converter to return to the Client the TCP options that were returned
   by the Server in the SYN+ACK packet.  A Transport Converter MUST
   return this TLV if the Client sent an Extended Connect TLV and the
   connection was accepted by the Server.

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+-------------------------------+
    |     Type=0x14 |     Length    |           Unassigned          |
    +---------------+---------------+-------------------------------+
    /               Returned Extended TCP header                    /
    /                              ...                              /
    +---------------------------------------------------------------+

                   Figure 20: The Extended TCP Header TLV

   The Returned Extended TCP header field is a copy of the TCP Options
   that were included in the SYN+ACK received by the Transport
   Converter.

   The Unassigned field MUST be set to zero by the sender and ignored by
   the receiver.

6.2.7.  The Cookie TLV



   The Cookie TLV (Figure 21) is an optional TLV that is similar to the
   TCP Fast Open Cookie [RFC7413].  A Transport Converter may want to
   verify that a Client can receive the packets that it sends to prevent
   attacks from spoofed addresses.  This verification can be done by
   using a Cookie that is bound to, for example, the IP address(es) of
   the Client.  This Cookie can be configured on the Client by means
   that are outside of this document or provided by the Transport
   Converter.

   A Transport Converter that has been configured to use the optional
   Cookie TLV MUST verify the presence of this TLV in the payload of the
   received SYN.  If this TLV is present, the Transport Converter MUST
   validate the Cookie by means similar to those in Section 4.1.2 of
   [RFC7413] (i.e., IsCookieValid).  If the Cookie is valid, the
   connection establishment procedure can continue.  Otherwise, the
   Transport Converter MUST return an Error TLV set to "Not Authorized"
   and close the connection.

   If the received SYN did not contain a Cookie TLV, and cookie
   validation is required, the Transport Converter MAY compute a Cookie
   bound to this Client address.  In such case, the Transport Converter
   MUST return an Error TLV set to "Missing Cookie" and the computed
   Cookie and close the connection.  The Client will react to this error
   by first issuing a reset to terminate the connection.  It also stores
   the received Cookie in its cache and attempts to reestablish a new
   connection to the Transport Converter that includes the Cookie TLV.

   The format of the Cookie TLV is shown in Figure 21.

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+-------------------------------+
    |     Type=0x16 |     Length    |             Zero              |
    +---------------+---------------+-------------------------------+
    /                        Opaque  Cookie                         /
    /                              ...                              /
    +---------------------------------------------------------------+

                         Figure 21: The Cookie TLV

6.2.8.  Error TLV



   The Error TLV (Figure 22) is meant to provide information about some
   errors that occurred during the processing of a Convert message.
   This TLV has a variable length.  Upon reception of an Error TLV, a
   Client MUST reset the associated connection.

   An Error TLV can be included in the SYN+ACK or an ACK.

                         1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +---------------+---------------+----------------+--------------+
    |     Type=0x1E |     Length    |    Error Code  |  Value       |
    +---------------+---------------+----------------+--------------+
    //              ...   (optional) Value                         //
    +---------------------------------------------------------------+

                          Figure 22: The Error TLV

   Different types of errors can occur while processing Convert
   messages.  Each error is identified by an Error Code represented as
   an unsigned integer.  Four classes of error codes are defined:

   Message validation and processing errors (0-31 range):
      Returned upon reception of an invalid message (including valid
      messages but with invalid or unknown TLVs).

   Client-side errors (32-63 range):
      The Client sent a request that could not be accepted by the
      Transport Converter (e.g., unsupported operation).

   Converter-side errors (64-95 range):
      Problems encountered on the Transport Converter (e.g., lack of
      resources) that prevent it from fulfilling the Client's request.

   Errors caused by the destination Server (96-127 range):
      The final destination could not be reached or it replied with a
      reset.

   The following error codes are defined in this document:

   Unsupported Version (0):
      The version number indicated in the fixed header of a message
      received from a peer is not supported.

      This error code MUST be generated by a peer (e.g., Transport
      Converter) when it receives a request having a version number that
      it does not support.

      The Value field MUST be set to the version supported by the peer.
      When multiple versions are supported by the peer, it includes the
      list of supported versions in the Value field; each version is
      encoded in 8 bits.  The list of supported versions MUST be padded
      with zeros to end on a 32-bit boundary.

      Upon receipt of this error code, the remote peer (e.g., Client)
      checks whether it supports one of the versions returned by the
      peer.  The highest commonly supported version number MUST be used
      by the remote peer in subsequent exchanges with the peer.

   Malformed Message (1):
      This error code is sent to indicate that a message received from a
      peer cannot be successfully parsed and validated.

      Typically, this error code is sent by the Transport Converter if
      it receives a Connect TLV enclosing a multicast, broadcast, or
      loopback IP address.

      To ease troubleshooting, the Value field MUST echo the received
      message using the format depicted in Figure 23.  This format
      allows keeping the original alignment of the message that
      triggered the error.

                            1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +---------------+---------------+----------------+--------------+
       |     Type=0x1E |     Length    |    Error Code  |  Zeros       |
       +---------------+---------------+----------------+--------------+
       //        Echo the message that triggered the error            //
       +---------------------------------------------------------------+

               Figure 23: Error TLV to Ease Message Correlation

   Unsupported Message (2):
      This error code is sent to indicate that a message type received
      from a Client is not supported.

      To ease troubleshooting, the Value field MUST echo the received
      message using the format shown in Figure 23.

   Missing Cookie (3):
      If a Transport Converter requires the utilization of Cookies to
      prevent spoofing attacks and a Cookie TLV was not included in the
      Convert message, the Transport Converter MUST return this error to
      the requesting Client only if it computes a cookie for this
      Client.  The first byte of the Value field MUST be set to zero and
      the remaining bytes of the Error TLV contain the Cookie computed
      by the Transport Converter for this Client.

      A Client that receives this error code SHOULD cache the received
      Cookie and include it in subsequent Convert messages sent to that
      Transport Converter.

   Not Authorized (32):
      This error code indicates that the Transport Converter refused to
      create a connection because of a lack of authorization (e.g.,
      administratively prohibited, authorization failure, or invalid
      Cookie TLV).  The Value field MUST be set to zero.

      This error code MUST be sent by the Transport Converter when a
      request cannot be successfully processed because the authorization
      failed.

   Unsupported TCP Option (33):
      A TCP option that the Client requested to advertise to the final
      Server cannot be safely used.

      The Value field is set to the type of the unsupported TCP option.
      If several unsupported TCP options were specified in the Connect
      TLV, then the list of unsupported TCP options is returned.  The
      list of unsupported TCP options MUST be padded with zeros to end
      on a 32-bit boundary.

   Resource Exceeded (64):
      This error indicates that the Transport Converter does not have
      enough resources to perform the request.

      This error MUST be sent by the Transport Converter when it does
      not have sufficient resources to handle a new connection.  The
      Transport Converter may indicate in the Value field the suggested
      delay (in seconds) that the Client SHOULD wait before soliciting
      the Transport Converter for a new proxied connection.  A Value of
      zero corresponds to a default delay of at least 30 seconds.

   Network Failure (65):
      This error indicates that the Transport Converter is experiencing
      a network failure to proxy the request.

      The Transport Converter MUST send this error code when it
      experiences forwarding issues to proxy a connection.  The
      Transport Converter may indicate in the Value field the suggested
      delay (in seconds) that the Client SHOULD wait before soliciting
      the Transport Converter for a new proxied connection.  A Value of
      zero corresponds to a default delay of at least 30 seconds.

   Connection Reset (96):
      This error indicates that the final destination responded with an
      RST segment.  The Value field MUST be set to zero.

   Destination Unreachable (97):
      This error indicates that an ICMP message indicating a hard error
      (e.g., destination unreachable, port unreachable, or network
      unreachable) was received by the Transport Converter.  The Value
      field MUST echo the Code field of the received ICMP message.

      As a reminder, TCP implementations are supposed to act on an ICMP
      error message passed up from the IP layer, directing it to the
      connection that triggered the error using the demultiplexing
      information included in the payload of that ICMP message.  Such a
      demultiplexing issue does not apply for handling the "Destination
      Unreachable" Error TLV because the error is sent in-band.  For
      this reason, the payload of the ICMP message is not echoed in the
      Destination Unreachable Error TLV.

   Table 2 summarizes the different error codes.

                +=======+======+=========================+
                | Error | Hex  | Description             |
                +=======+======+=========================+
                | 0     | 0x00 | Unsupported Version     |
                +-------+------+-------------------------+
                | 1     | 0x01 | Malformed Message       |
                +-------+------+-------------------------+
                | 2     | 0x02 | Unsupported Message     |
                +-------+------+-------------------------+
                | 3     | 0x03 | Missing Cookie          |
                +-------+------+-------------------------+
                | 32    | 0x20 | Not Authorized          |
                +-------+------+-------------------------+
                | 33    | 0x21 | Unsupported TCP Option  |
                +-------+------+-------------------------+
                | 64    | 0x40 | Resource Exceeded       |
                +-------+------+-------------------------+
                | 65    | 0x41 | Network Failure         |
                +-------+------+-------------------------+
                | 96    | 0x60 | Connection Reset        |
                +-------+------+-------------------------+
                | 97    | 0x61 | Destination Unreachable |
                +-------+------+-------------------------+

                      Table 2: Convert Error Values

7.  Compatibility of Specific TCP Options with the Conversion Service



   In this section, we discuss how several deployed Standards Track TCP
   options can be supported through the Convert Protocol.  The other TCP
   options will be discussed in other documents.

7.1.  Base TCP Options



   Three TCP options were initially defined in [RFC0793]: End-of-Option
   List (Kind=0), No-Operation (Kind=1), and Maximum Segment Size
   (Kind=2).  The first two options are mainly used to pad the TCP
   header.  There is no reason for a Client to request a Transport
   Converter to specifically send these options towards the final
   destination.

   The Maximum Segment Size option (Kind=2) is used by a host to
   indicate the largest segment that it can receive over each
   connection.  This value is a function of the stack that terminates
   the TCP connection.  There is no reason for a Client to request a
   Transport Converter to advertise a specific Maximum Segment Size
   (MSS) value to a remote Server.

   A Transport Converter MUST ignore options with Kind=0, 1, or 2 if
   they appear in a Connect TLV.  It MUST NOT announce them in a
   Supported TCP Extensions TLV.

7.2.  Window Scale (WS)



   The Window Scale (WS) option (Kind=3) is defined in [RFC7323].  As
   for the MSS option, the window scale factor that is used for a
   connection strongly depends on the TCP stack that handles the
   connection.  When a Transport Converter opens a TCP connection
   towards a remote Server on behalf of a Client, it SHOULD use a WS
   option with a scaling factor that corresponds to the configuration of
   its stack.  A local configuration MAY allow for a WS option in the
   proxied message to be a function of the scaling factor of the
   incoming connection.

   From a deployment viewpoint, there is no benefit in enabling a Client
   of a Transport Converter to specifically request the utilization of
   the WS option (Kind=3) with a specific scaling factor towards a
   remote Server.  For this reason, a Transport Converter MUST ignore
   option Kind=3 if it appears in a Connect TLV.  The Transport
   Converter MUST NOT announce a WS option (Kind=3) in a Supported TCP
   Extensions TLV.

7.3.  Selective Acknowledgments



   Two distinct TCP options were defined to support Selective
   Acknowledgment (SACK) in [RFC2018].  This first one, SACK-Permitted
   (Kind=4), is used to negotiate the utilization of Selective
   Acknowledgments during the three-way handshake.  The second one, SACK
   (Kind=5), carries the Selective Acknowledgments inside regular
   segments.

   The SACK-Permitted option (Kind=4) MAY be advertised by a Transport
   Converter in the Supported TCP Extensions TLV.  Clients connected to
   this Transport Converter MAY include the SACK-Permitted option in the
   Connect TLV.

   The SACK option (Kind=5) cannot be used during the three-way
   handshake.  For this reason, a Transport Converter MUST ignore option
   Kind=5 if it appears in a Connect TLV.  It MUST NOT announce it in a
   TCP Supported Extensions TLV.

7.4.  Timestamp



   The Timestamp option [RFC7323] can be used during the three-way
   handshake to negotiate the utilization of timestamps during the TCP
   connection.  It is notably used to improve round-trip-time
   estimations and to provide Protection Against Wrapped Sequences
   (PAWS).  As for the WS option, the timestamps are a property of a
   connection and there is limited benefit in enabling a Client to
   request a Transport Converter to use the timestamp option when
   establishing a connection to a remote Server.  Furthermore, the
   timestamps that are used by TCP stacks are specific to each stack and
   there is no benefit in enabling a Client to specify the timestamp
   value that a Transport Converter could use to establish a connection
   to a remote Server.

   A Transport Converter MAY advertise the Timestamp option (Kind=8) in
   the TCP Supported Extensions TLV.  The Clients connected to this
   Transport Converter MAY include the Timestamp option in the Connect
   TLV but without any timestamp.

7.5.  Multipath TCP



   The Multipath TCP options are defined in [RFC8684], which defines one
   variable length TCP option (Kind=30) that includes a sub-type field
   to support several Multipath TCP options.  There are several
   operational use cases where Clients would like to use Multipath TCP
   through a Transport Converter [IETFJ16].  However, none of these use
   cases require the Client to specify the content of the Multipath TCP
   option that the Transport Converter should send to a remote Server.

   A Transport Converter that supports Multipath TCP conversion service
   MUST advertise the Multipath TCP option (Kind=30) in the Supported
   TCP Extensions TLV.  Clients serviced by this Transport Converter may
   include the Multipath TCP option in the Connect TLV but without any
   content.

7.6.  TCP Fast Open



   The TCP Fast Open Cookie option (Kind=34) is defined in [RFC7413].
   There are two different usages of this option that need to be
   supported by Transport Converters.  The first utilization of the TCP
   Fast Open Cookie option is to request a cookie from the Server.  In
   this case, the option is sent with an empty cookie by the Client, and
   the Server returns the cookie.  The second utilization of the TCP
   Fast Open Cookie option is to send a cookie to the Server.  In this
   case, the option contains a cookie.

   A Transport Converter MAY advertise the TCP Fast Open Cookie option
   (Kind=34) in the Supported TCP Extensions TLV.  If a Transport
   Converter has advertised the support for TCP Fast Open in its
   Supported TCP Extensions TLV, it needs to be able to process two
   types of Connect TLV.

   If such a Transport Converter receives a Connect TLV with the TCP
   Fast Open Cookie option that does not contain a cookie, it MUST add
   an empty TCP Fast Open Cookie option in the SYN sent to the remote
   Server.  If the remote Server supports TFO, it responds with a SYN-
   ACK according to the procedure in Section 4.1.2 of [RFC7413].  This
   SYN-ACK may contain a Fast Open option with a cookie.  Upon receipt
   of the SYN-ACK by the Converter, it relays the Fast Open option with
   the cookie to the Client.

   If such a Transport Converter receives a Connect TLV with the TCP
   Fast Open Cookie option that contains a cookie, it MUST copy the TCP
   Fast Open Cookie option in the SYN sent to the remote Server.

7.7.  TCP-AO



   The TCP Authentication Option (TCP-AO) [RFC5925] provides a technique
   to authenticate all the packets exchanged over a TCP connection.
   Given the nature of this extension, it is unlikely that the
   applications that require their packets to be authenticated end to
   end would want their connections to pass through a converter.  For
   this reason, we do not recommend the support of the TCP-AO by
   Transport Converters.  The only use cases where it could make sense
   to combine TCP-AO and the solution in this document are those where
   the TCP-AO-NAT extension [RFC6978] is in use.

   A Transport Converter MUST NOT advertise the TCP-AO (Kind=29) in the
   Supported TCP Extensions TLV.  If a Transport Converter receives a
   Connect TLV that contains the TCP-AO, it MUST reject the
   establishment of the connection with error code set to "Unsupported
   TCP Option", except if the TCP-AO-NAT option is used.  Nevertheless,
   given that TCP-AO-NAT is Experimental, its usage is not currently
   defined and must be specified by some other document before it can be
   used.

8.  Interactions with Middleboxes



   The Convert Protocol is designed to be used in networks that do not
   contain middleboxes that interfere with TCP.  Under such conditions,
   it is assumed that the network provider ensures that all involved on-
   path nodes are not breaking TCP signals (e.g., strip TCP options,
   discard some SYNs, etc.).

   Nevertheless, and in order to allow for a robust service, this
   section describes how a Client can detect middlebox interference and
   stop using the Transport Converter affected by this interference.

   Internet measurements [IMC11] have shown that middleboxes can affect
   the deployment of TCP extensions.  In this section, we focus the
   middleboxes that modify the payload since the Convert Protocol places
   its messages at the beginning of the bytestream.

   Consider a middlebox that removes the SYN payload.  The Client can
   detect this problem by looking at the acknowledgment number field of
   the SYN+ACK if returned by the Transport Converter.  The Client MUST
   stop to use this Transport Converter given the middlebox
   interference.

   Consider now a middlebox that drops SYN/ACKs with a payload.  The
   Client won't be able to establish a connection via the Transport
   Converter.  The case of a middlebox that removes the payload of
   SYN+ACKs or from the packet that follows the SYN+ACK (but not the
   payload of SYN) can be detected by a Client.  This is hinted by the
   absence of a valid Convert message in the response.

   As explained in [RFC7413], some Carrier Grade NATs (CGNs) can affect
   the operation of TFO if they assign different IP addresses to the
   same end host.  Such CGNs could affect the operation of the cookie
   validation used by the Convert Protocol.  As a reminder, CGNs that
   are enabled on the path between a Client and a Transport Converter
   must adhere to the address preservation defined in [RFC6888].  See
   also the discussion in Section 7.1 of [RFC7413].

9.  Security Considerations



   An implementation MUST check that the Convert TLVs are properly
   framed within the boundary indicated by the Total Length in the fixed
   header (Section 6.1).

   Additional security considerations are discussed in the following
   subsections.

9.1.  Privacy & Ingress Filtering



   The Transport Converter may have access to privacy-related
   information (e.g., subscriber credentials).  The Transport Converter
   is designed to not leak such sensitive information outside a local
   domain.

   Given its function and location in the network, a Transport Converter
   is in a position to observe all packets that it processes, to include
   payloads and metadata, and has the ability to profile and conduct
   some traffic analysis of user behavior.  The Transport Converter MUST
   be as protected as a core IP router (e.g., Section 10 of [RFC1812]).

   Furthermore, ingress filtering policies MUST be enforced at the
   network boundaries [RFC2827].

   This document assumes that all network attachments are managed by the
   same administrative entity.  Therefore, enforcing anti-spoofing
   filters at these networks is a guard that hosts are not sending
   traffic with spoofed source IP addresses.

9.2.  Authentication and Authorization Considerations



   The Convert Protocol is RECOMMENDED for use in a managed network
   where end hosts can be securely identified by their IP address.  If
   such control is not exerted and there is a more open network
   environment, a strong mutual authentication scheme MUST be defined to
   use the Convert Protocol.

   One possibility for mutual authentication is to use TLS to perform
   mutual authentication between the Client and the Converter.  That is,
   use TLS when a Client retrieves a Cookie from the Converter and rely
   on certificate-based, pre-shared key-based [RFC4279], or raw public
   key-based Client authentication [RFC7250] to secure this connection.
   If the authentication succeeds, the Converter returns a cookie to the
   Client.  Subsequent Connect messages will be authorized as a function
   of the content of the Cookie TLV.  An attacker from within the
   network between a Client and a Transport Converter may intercept the
   Cookie and use it to be granted access to the conversion service.
   Such an attack is only possible if the attacker spoofs the IP address
   of the Client and the network does not filter packets with source-
   spoofed IP addresses.

   The operator that manages the various network attachments (including
   the Transport Converters) has various options for enforcing
   authentication and authorization policies.  For example, a non-
   exhaustive list of methods to achieve authorization is provided
   hereafter:

   *  The network provider may enforce a policy based on the
      International Mobile Subscriber Identity (IMSI) to verify that a
      user is allowed to benefit from the TCP converter service.  If
      that authorization fails, the Packet Data Protocol (PDP) context/
      bearer will not be mounted.  This method does not require any
      interaction with the Transport Converter for authorization
      matters.

   *  The network provider may enforce a policy based upon Access
      Control Lists (ACLs), e.g., at a Broadband Network Gateway (BNG)
      to control the hosts that are authorized to communicate with a
      Transport Converter.  These ACLs may be installed as a result of
      RADIUS exchanges, e.g., [TCPM-CONVERTER].  This method does not
      require any interaction with the Transport Converter for
      authorization matters.

   *  A device that embeds a Transport Converter may also host a RADIUS
      Client that will solicit a AAA Server to check whether or not
      connections received from a given source IP address are authorized
      [TCPM-CONVERTER].

   A first safeguard against the misuse of Transport Converter resources
   by illegitimate users (e.g., users with access networks that are not
   managed by the same provider that operates the Transport Converter)
   is the Transport Converter to reject Convert connections received in
   the external realm.  Only Convert connections received in the
   internal realm of a Transport Converter will be accepted.

   In deployments where network-assisted connections are not allowed
   between hosts of a domain (i.e., hairpinning), the Converter may be
   instructed to discard such connections.  Hairpinned connections are
   thus rejected by the Transport Converter by returning an Error TLV
   set to "Not Authorized".  Otherwise, absent explicit configuration,
   hairpinning is enabled by the Converter (see Figure 24).

             <===Network Provider===>

      +----+ from X1:x1 to X2':x2'   +-----+ X1':x1'
      | C1 |>>>>>>>>>>>>>>>>>>>>>>>>>>>>>--+---
      +----+                         |  v  |
                                     |  v  |
                                     |  v  |
                                     |  v  |
      +----+ from X1':x1' to X2:x2   |  v  | X2':x2'
      | C2 |<<<<<<<<<<<<<<<<<<<<<<<<<<<<<--+---
      +----+                         +-----+
                                    Converter

      Note: X2':x2' may be equal to
            X2:x2

                       Figure 24: Hairpinning Example

9.3.  Denial of Service



   Another possible risk is amplification attacks, since a Transport
   Converter sends a SYN towards a remote Server upon reception of a SYN
   from a Client.  This could lead to amplification attacks if the SYN
   sent by the Transport Converter were larger than the SYN received
   from the Client, or if the Transport Converter retransmits the SYN.
   To mitigate such attacks, the Transport Converter SHOULD rate-limit
   the number of pending requests for a given Client.  It SHOULD also
   avoid sending SYNs that are significantly longer than the SYN
   received from the Client, to remote Servers.  Finally, the Transport
   Converter SHOULD only retransmit a SYN to a Server after having
   received a retransmitted SYN from the corresponding Client.  Means to
   protect against SYN flooding attacks should also be enabled (e.g.,
   Section 3 of [RFC4987]).

   Attacks from within the network between a Client and a Transport
   Converter (including attacks that change the protocol version) are
   yet another threat.  Means to ensure that illegitimate nodes cannot
   connect to a network should be implemented.

9.4.  Traffic Theft



   Traffic theft is a risk if an illegitimate Converter is inserted in
   the path.  Indeed, inserting an illegitimate Converter in the
   forwarding path allows traffic interception and can therefore provide
   access to sensitive data issued by or destined to a host.  Converter
   discovery and configuration are out of scope of this document.

9.5.  Logging



   If the Converter is configured to behave in the address-sharing mode
   (Section 4.4.2), the logging recommendations discussed in Section 4
   of [RFC6888] need to be considered.  Security-related issues
   encountered in address-sharing environments are documented in
   Section 13 of [RFC6269].

10.  IANA Considerations



10.1.  Convert Service Name



   IANA has assigned a service name for the Convert Protocol from the
   "Service Name and Transport Protocol Port Number Registry" available
   at <https://www.iana.org/assignments/service-names-port-numbers>.

   Service Name:            convert
   Port Number:             N/A
   Transport Protocol(s):   TCP
   Description:             0-RTT TCP Convert Protocol
   Assignee:                IESG <iesg@ietf.org>
   Contact:                 IETF Chair <chair@ietf.org>
   Reference:               RFC 8803

   Clients may use this service name to feed the procedure defined in
   [RFC2782] to discover the IP address(es) and the port number used by
   the Transport Converters of a domain.

10.2.  The Convert Protocol (Convert) Parameters



   IANA has created a new "TCP Convert Protocol (Convert) Parameters"
   registry.

   The following subsections detail new registries within the "Convert
   Protocol (Convert) Parameters" registry.

   The designated expert is expected to ascertain the existence of
   suitable documentation as described in Section 4.6 of [RFC8126] and
   to verify that the document is permanently and publicly available.
   The designated expert is also expected to check the clarity of
   purpose and use of the requested code points.

   Also, criteria that should be applied by the designated experts
   includes determining whether the proposed registration duplicates
   existing functionality, whether it is likely to be of general
   applicability or useful only for private use, and whether the
   registration description is clear.  All requests should be directed
   to the review mailing list.  For both the "Convert TLVs" and "Convert
   Errors" subregistries, IANA must only accept registry updates in the
   128-191 range from the designated experts.  It is suggested that
   multiple designated experts be appointed.  In cases where a
   registration decision could be perceived as creating a conflict of
   interest for a particular expert, that expert should defer to the
   judgment of the other experts.

10.2.1.  Convert Versions



   IANA has created the "Convert Versions" subregistry.  New values are
   assigned via IETF Review (Section 4.8 of [RFC8126]).

   The initial values of the registry are as follows:

                   +=========+=============+===========+
                   | Version | Description | Reference |
                   +=========+=============+===========+
                   | 0       | Reserved    | RFC 8803  |
                   +---------+-------------+-----------+
                   | 1       | Assigned    | RFC 8803  |
                   +---------+-------------+-----------+

                     Table 3: Current Convert Versions

10.2.2.  Convert TLVs



   IANA has created the "Convert TLVs" subregistry.  The procedures for
   assigning values from this registry are as follows:

   1-127:    IETF Review

   128-191:  Specification Required

   192-255:  Private Use

   The initial values of the registry are as follows:

            +======+=============================+===========+
            | Code | Name                        | Reference |
            +======+=============================+===========+
            | 0    | Reserved                    | RFC 8803  |
            +------+-----------------------------+-----------+
            | 1    | Info TLV                    | RFC 8803  |
            +------+-----------------------------+-----------+
            | 10   | Connect TLV                 | RFC 8803  |
            +------+-----------------------------+-----------+
            | 20   | Extended TCP Header TLV     | RFC 8803  |
            +------+-----------------------------+-----------+
            | 21   | Supported TCP Extension TLV | RFC 8803  |
            +------+-----------------------------+-----------+
            | 22   | Cookie TLV                  | RFC 8803  |
            +------+-----------------------------+-----------+
            | 30   | Error TLV                   | RFC 8803  |
            +------+-----------------------------+-----------+

                      Table 4: Initial Convert TLVs

10.2.3.  Convert Error Messages



   IANA has created the "Convert Errors" subregistry.  Codes in this
   registry are assigned as a function of the error type.  Four types
   are defined; the following ranges are reserved for each of these
   types:

   0-31:     Message validation and processing errors

   32-63:    Client-side errors

   64-95:    Transport Converter-side errors

   96-127:   Errors caused by destination Server

   The procedures for assigning values from this subregistry are as
   follows:

   0-127:    IETF Review

   128-191:  Specification Required

   192-255:  Private Use

   The initial values of the registry are as follows:

              +=======+=========================+===========+
              | Error | Description             | Reference |
              +=======+=========================+===========+
              | 0     | Unsupported Version     | RFC 8803  |
              +-------+-------------------------+-----------+
              | 1     | Malformed Message       | RFC 8803  |
              +-------+-------------------------+-----------+
              | 2     | Unsupported Message     | RFC 8803  |
              +-------+-------------------------+-----------+
              | 3     | Missing Cookie          | RFC 8803  |
              +-------+-------------------------+-----------+
              | 32    | Not Authorized          | RFC 8803  |
              +-------+-------------------------+-----------+
              | 33    | Unsupported TCP Option  | RFC 8803  |
              +-------+-------------------------+-----------+
              | 64    | Resource Exceeded       | RFC 8803  |
              +-------+-------------------------+-----------+
              | 65    | Network Failure         | RFC 8803  |
              +-------+-------------------------+-----------+
              | 96    | Connection Reset        | RFC 8803  |
              +-------+-------------------------+-----------+
              | 97    | Destination Unreachable | RFC 8803  |
              +-------+-------------------------+-----------+

                    Table 5: Initial Convert Error Codes

11.  References



11.1.  Normative References



   [RFC0793]  Postel, J., "Transmission Control Protocol", STD 7,
              RFC 793, DOI 10.17487/RFC0793, September 1981,
              <https://www.rfc-editor.org/info/rfc793>.

   [RFC2018]  Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP
              Selective Acknowledgment Options", RFC 2018,
              DOI 10.17487/RFC2018, October 1996,
              <https://www.rfc-editor.org/info/rfc2018>.

   [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>.

   [RFC2827]  Ferguson, P. and D. Senie, "Network Ingress Filtering:
              Defeating Denial of Service Attacks which employ IP Source
              Address Spoofing", BCP 38, RFC 2827, DOI 10.17487/RFC2827,
              May 2000, <https://www.rfc-editor.org/info/rfc2827>.

   [RFC4291]  Hinden, R. and S. Deering, "IP Version 6 Addressing
              Architecture", RFC 4291, DOI 10.17487/RFC4291, February
              2006, <https://www.rfc-editor.org/info/rfc4291>.

   [RFC4787]  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>.

   [RFC4987]  Eddy, W., "TCP SYN Flooding Attacks and Common
              Mitigations", RFC 4987, DOI 10.17487/RFC4987, August 2007,
              <https://www.rfc-editor.org/info/rfc4987>.

   [RFC5925]  Touch, J., Mankin, A., and R. Bonica, "The TCP
              Authentication Option", RFC 5925, DOI 10.17487/RFC5925,
              June 2010, <https://www.rfc-editor.org/info/rfc5925>.

   [RFC6888]  Perreault, S., Ed., Yamagata, I., Miyakawa, S., Nakagawa,
              A., and H. Ashida, "Common Requirements for Carrier-Grade
              NATs (CGNs)", BCP 127, RFC 6888, DOI 10.17487/RFC6888,
              April 2013, <https://www.rfc-editor.org/info/rfc6888>.

   [RFC6890]  Cotton, M., Vegoda, L., Bonica, R., Ed., and B. Haberman,
              "Special-Purpose IP Address Registries", BCP 153,
              RFC 6890, DOI 10.17487/RFC6890, April 2013,
              <https://www.rfc-editor.org/info/rfc6890>.

   [RFC7323]  Borman, D., Braden, B., Jacobson, V., and R.
              Scheffenegger, Ed., "TCP Extensions for High Performance",
              RFC 7323, DOI 10.17487/RFC7323, September 2014,
              <https://www.rfc-editor.org/info/rfc7323>.

   [RFC7413]  Cheng, Y., Chu, J., Radhakrishnan, S., and A. Jain, "TCP
              Fast Open", RFC 7413, DOI 10.17487/RFC7413, December 2014,
              <https://www.rfc-editor.org/info/rfc7413>.

   [RFC8126]  Cotton, M., Leiba, B., and T. Narten, "Guidelines for
              Writing an IANA Considerations Section in RFCs", BCP 26,
              RFC 8126, DOI 10.17487/RFC8126, June 2017,
              <https://www.rfc-editor.org/info/rfc8126>.

   [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>.

   [RFC8684]  Ford, A., Raiciu, C., Handley, M., Bonaventure, O., and C.
              Paasch, "TCP Extensions for Multipath Operation with
              Multiple Addresses", RFC 8684, DOI 10.17487/RFC8684, March
              2020, <https://www.rfc-editor.org/info/rfc8684>.

11.2.  Informative References



   [ANRW17]   Trammell, B., Kuehlewind, M., De Vaere, P., Learmonth, I.,
              and G. Fairhurst, "Tracking transport-layer evolution with
              PATHspider", Applied Networking Research Workshop 2017
              (ANRW17), July 2017.

   [DHC-CONVERTER]
              Boucadair, M., Jacquenet, C., and T. Reddy.K, "DHCP
              Options for 0-RTT TCP Converters", Work in Progress,
              Internet-Draft, draft-boucadair-tcpm-dhc-converter-03, 7
              October 2019, <https://tools.ietf.org/html/draft-
              boucadair-tcpm-dhc-converter-03>.

   [Fukuda2011]
              Fukuda, K., "An Analysis of Longitudinal TCP Passive
              Measurements (Short Paper)", Traffic Monitoring and
              Analysis, TMA 2011, Lecture Notes in Computer Science,
              vol. 6613, 2011.

   [HOT-MIDDLEBOX13]
              Detal, G., Paasch, C., and O. Bonaventure, "Multipath in
              the Middle(Box)", HotMiddlebox'13,
              DOI 10.1145/2535828.2535829, December 2013,
              <https://inl.info.ucl.ac.be/publications/multipath-
              middlebox>.

   [IANA-CONVERT]
              IANA, "TCP Convert Protocol (Convert) Parameters",
              <https://www.iana.org/assignments/tcp-convert-protocol-
              parameters>.

   [IETFJ16]  Bonaventure, O. and S. Seo, "Multipath TCP Deployments",
              IETF Journal, Vol. 12, Issue 2, November 2016.

   [IMC11]    Honda, K., Nishida, Y., Raiciu, C., Greenhalgh, A.,
              Handley, M., and T. Hideyuki, "Is it still possible to
              extend TCP?", Proceedings of the 2011 ACM SIGCOMM
              conference on Internet measurement conference,
              DOI 10.1145/2068816.2068834, November 2011,
              <https://doi.org/10.1145/2068816.2068834>.

   [INTAREA-SOCKS]
              Olteanu, V. and D. Niculescu, "SOCKS Protocol Version 6",
              Work in Progress, Internet-Draft, draft-olteanu-intarea-
              socks-6-10, 13 July 2020, <https://tools.ietf.org/html/
              draft-olteanu-intarea-socks-6-10>.

   [LOW-LATENCY]
              Arkko, J. and J. Tantsura, "Low Latency Applications and
              the Internet Architecture", Work in Progress, Internet-
              Draft, draft-arkko-arch-low-latency-02, 30 October 2017,
              <https://tools.ietf.org/html/draft-arkko-arch-low-latency-
              02>.

   [MPTCP-PLAIN]
              Boucadair, M., Jacquenet, C., Bonaventure, O., Behaghel,
              D., Secci, S., Henderickx, W., Skog, R., Vinapamula, S.,
              Seo, S., Cloetens, W., Meyer, U., Contreras, L., and B.
              Peirens, "Extensions for Network-Assisted MPTCP Deployment
              Models", Work in Progress, Internet-Draft, draft-
              boucadair-mptcp-plain-mode-10, March 2017,
              <https://tools.ietf.org/html/draft-boucadair-mptcp-plain-
              mode-10>.

   [MPTCP-TRANSPARENT]
              Peirens, B., Detal, G., Barre, S., and O. Bonaventure,
              "Link bonding with transparent Multipath TCP", Work in
              Progress, Internet-Draft, draft-peirens-mptcp-transparent-
              00, 8 July 2016, <https://tools.ietf.org/html/draft-
              peirens-mptcp-transparent-00>.

   [RFC1812]  Baker, F., Ed., "Requirements for IP Version 4 Routers",
              RFC 1812, DOI 10.17487/RFC1812, June 1995,
              <https://www.rfc-editor.org/info/rfc1812>.

   [RFC1919]  Chatel, M., "Classical versus Transparent IP Proxies",
              RFC 1919, DOI 10.17487/RFC1919, March 1996,
              <https://www.rfc-editor.org/info/rfc1919>.

   [RFC1928]  Leech, M., Ganis, M., Lee, Y., Kuris, R., Koblas, D., and
              L. Jones, "SOCKS Protocol Version 5", RFC 1928,
              DOI 10.17487/RFC1928, March 1996,
              <https://www.rfc-editor.org/info/rfc1928>.

   [RFC2782]  Gulbrandsen, A., Vixie, P., and L. Esibov, "A DNS RR for
              specifying the location of services (DNS SRV)", RFC 2782,
              DOI 10.17487/RFC2782, February 2000,
              <https://www.rfc-editor.org/info/rfc2782>.

   [RFC3135]  Border, J., Kojo, M., Griner, J., Montenegro, G., and Z.
              Shelby, "Performance Enhancing Proxies Intended to
              Mitigate Link-Related Degradations", RFC 3135,
              DOI 10.17487/RFC3135, June 2001,
              <https://www.rfc-editor.org/info/rfc3135>.

   [RFC4279]  Eronen, P., Ed. and H. Tschofenig, Ed., "Pre-Shared Key
              Ciphersuites for Transport Layer Security (TLS)",
              RFC 4279, DOI 10.17487/RFC4279, December 2005,
              <https://www.rfc-editor.org/info/rfc4279>.

   [RFC5461]  Gont, F., "TCP's Reaction to Soft Errors", RFC 5461,
              DOI 10.17487/RFC5461, February 2009,
              <https://www.rfc-editor.org/info/rfc5461>.

   [RFC6269]  Ford, M., Ed., Boucadair, M., Durand, A., Levis, P., and
              P. Roberts, "Issues with IP Address Sharing", RFC 6269,
              DOI 10.17487/RFC6269, June 2011,
              <https://www.rfc-editor.org/info/rfc6269>.

   [RFC6296]  Wasserman, M. and F. Baker, "IPv6-to-IPv6 Network Prefix
              Translation", RFC 6296, DOI 10.17487/RFC6296, June 2011,
              <https://www.rfc-editor.org/info/rfc6296>.

   [RFC6731]  Savolainen, T., Kato, J., and T. Lemon, "Improved
              Recursive DNS Server Selection for Multi-Interfaced
              Nodes", RFC 6731, DOI 10.17487/RFC6731, December 2012,
              <https://www.rfc-editor.org/info/rfc6731>.

   [RFC6887]  Wing, D., Ed., Cheshire, S., Boucadair, M., Penno, R., and
              P. Selkirk, "Port Control Protocol (PCP)", RFC 6887,
              DOI 10.17487/RFC6887, April 2013,
              <https://www.rfc-editor.org/info/rfc6887>.

   [RFC6928]  Chu, J., Dukkipati, N., Cheng, Y., and M. Mathis,
              "Increasing TCP's Initial Window", RFC 6928,
              DOI 10.17487/RFC6928, April 2013,
              <https://www.rfc-editor.org/info/rfc6928>.

   [RFC6978]  Touch, J., "A TCP Authentication Option Extension for NAT
              Traversal", RFC 6978, DOI 10.17487/RFC6978, July 2013,
              <https://www.rfc-editor.org/info/rfc6978>.

   [RFC7250]  Wouters, P., Ed., Tschofenig, H., Ed., Gilmore, J.,
              Weiler, S., and T. Kivinen, "Using Raw Public Keys in
              Transport Layer Security (TLS) and Datagram Transport
              Layer Security (DTLS)", RFC 7250, DOI 10.17487/RFC7250,
              June 2014, <https://www.rfc-editor.org/info/rfc7250>.

   [RFC7414]  Duke, M., Braden, R., Eddy, W., Blanton, E., and A.
              Zimmermann, "A Roadmap for Transmission Control Protocol
              (TCP) Specification Documents", RFC 7414,
              DOI 10.17487/RFC7414, February 2015,
              <https://www.rfc-editor.org/info/rfc7414>.

   [RFC8041]  Bonaventure, O., Paasch, C., and G. Detal, "Use Cases and
              Operational Experience with Multipath TCP", RFC 8041,
              DOI 10.17487/RFC8041, January 2017,
              <https://www.rfc-editor.org/info/rfc8041>.

   [RFC8305]  Schinazi, D. and T. Pauly, "Happy Eyeballs Version 2:
              Better Connectivity Using Concurrency", RFC 8305,
              DOI 10.17487/RFC8305, December 2017,
              <https://www.rfc-editor.org/info/rfc8305>.

   [RFC8446]  Rescorla, E., "The Transport Layer Security (TLS) Protocol
              Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018,
              <https://www.rfc-editor.org/info/rfc8446>.

   [RFC8548]  Bittau, A., Giffin, D., Handley, M., Mazieres, D., Slack,
              Q., and E. Smith, "Cryptographic Protection of TCP Streams
              (tcpcrypt)", RFC 8548, DOI 10.17487/RFC8548, May 2019,
              <https://www.rfc-editor.org/info/rfc8548>.

   [TCPM-CONVERTER]
              Boucadair, M. and C. Jacquenet, "RADIUS Extensions for
              0-RTT TCP Converters", Work in Progress, Internet-Draft,
              draft-boucadair-opsawg-tcpm-converter-01, 28 February
              2020, <https://tools.ietf.org/html/draft-boucadair-opsawg-
              tcpm-converter-01>.

   [TS23501]  3GPP (3rd Generation Partnership Project), "Technical
              Specification Group Services and System Aspects; System
              architecture for the 5G System; Stage 2 (Release 16)",
              2019, <https://www.3gpp.org/ftp/Specs/
              archive/23_series/23.501/>.

Appendix A.  Example Socket API Changes to Support the 0-RTT TCP Convert
             Protocol



A.1.  Active Open (Client Side)



   On the Client side, the support of the 0-RTT Converter protocol does
   not require any other changes than those identified in Appendix A of
   [RFC7413].  Those modifications are already supported by multiple TCP
   stacks.

   As an example, on Linux, a Client can send the 0-RTT Convert message
   inside a SYN by using sendto with the MSG_FASTOPEN flag as shown in
   the example below:

     s = socket(AF_INET, SOCK_STREAM, 0);

     sendto(s, buffer, buffer_len, MSG_FASTOPEN,
                    (struct sockaddr *) &server_addr, addr_len);

   The Client side of the Linux TFO can be used in two different modes
   depending on the host configuration (sysctl tcp_fastopen variable):

   0x1:  (client) enables sending data in the opening SYN on the Client.

   0x4:  (client) enables sending data in the opening SYN regardless of
      cookie availability and without a cookie option.

   By setting this configuration variable to 0x5, a Linux Client using
   the above code would send data inside the SYN without using a TFO
   option.

A.2.  Passive Open (Converter Side)



   The Converter needs to enable the reception of data inside the SYN
   independently of the utilization of the TFO option.  This implies
   that the Transport Converter application cannot rely on the Fast Open
   Cookies to validate the reachability of the IP address that sent the
   SYN.  It must rely on other techniques, such as the Cookie TLV
   described in this document, to verify this reachability.

   [RFC7413] suggested the utilization of a TCP_FASTOPEN socket option
   to enable the reception of SYNs containing data.  Later, Appendix A
   of [RFC7413] mentioned:

   |  Traditionally, accept() returns only after a socket is connected.
   |  But, for a Fast Open connection, accept() returns upon receiving a
   |  SYN with a valid Fast Open cookie and data, and the data is
   |  available to be read through, e.g., recvmsg(), read().

   To support the 0-RTT TCP Convert Protocol, this behavior should be
   modified as follows:

   |  Traditionally, accept() returns only after a socket is connected.
   |  But, for a Fast Open connection, accept() returns upon receiving a
   |  SYN with data, and the data is available to be read through, e.g.,
   |  recvmsg(), read().  The application that receives such SYNs with
   |  data must be able to validate the reachability of the source of
   |  the SYN and also deal with replayed SYNs.

   The Linux Server side can be configured with the following sysctls:

   0x2:  (server) enables the Server support, i.e., allowing data in a
      SYN packet to be accepted and passed to the application before a
      3-way handshake finishes.

   0x200:  (server) accepts data-in-SYN w/o any cookie option present.

   However, this configuration is system wide.  This is convenient for
   typical Transport Converter deployments where no other applications
   relying on TFO are collocated on the same device.

   Recently, the TCP_FASTOPEN_NO_COOKIE socket option has been added to
   provide the same behavior on a per-socket basis.  This enables a
   single host to support both Servers that require the Fast Open Cookie
   and Servers that do not use it.

Acknowledgments

   Although they could disagree with the contents of the document, we
   would like to thank Joe Touch and Juliusz Chroboczek, whose comments
   on the MPTCP mailing list have forced us to reconsider the design of
   the solution several times.

   We would like to thank Raphael Bauduin, Stefano Secci, Anandatirtha
   Nandugudi, and Gregory Vander Schueren for their help in preparing
   this document.  Nandini Ganesh provided valuable feedback about the
   handling of TFO and the error codes.  Yuchung Cheng and Praveen
   Balasubramanian helped to clarify the discussion on supplying data in
   SYNs.  Phil Eardley and Michael Scharf helped to clarify different
   parts of the text.  Thanks to Éric Vyncke, Roman Danyliw, Benjamin
   Kaduk, and Alexey Melnikov for the IESG review, and Christian Huitema
   for the Security Directorate review.

   Many thanks to Mirja Kühlewind for the detailed AD review.

   This document builds upon earlier documents that proposed various
   forms of Multipath TCP proxies: [MPTCP-PLAIN], [MPTCP-TRANSPARENT],
   and [HOT-MIDDLEBOX13].

   From [MPTCP-PLAIN]:

   Many thanks to Chi Dung Phung, Mingui Zhang, Rao Shoaib, Yoshifumi
   Nishida, and Christoph Paasch for their valuable comments.

   Thanks to Ian Farrer, Mikael Abrahamsson, Alan Ford, Dan Wing, and
   Sri Gundavelli for the fruitful discussions at IETF 95 (Buenos
   Aires).

   Special thanks to Pierrick Seite, Yannick Le Goff, Fred Klamm, and
   Xavier Grall for their input.

   Thanks also to Olaf Schleusing, Martin Gysi, Thomas Zasowski, Andreas
   Burkhard, Silka Simmen, Sandro Berger, Michael Melloul, Jean-Yves
   Flahaut, Adrien Desportes, Gregory Detal, Benjamin David, Arun
   Srinivasan, and Raghavendra Mallya for their input.

Contributors

   Bart Peirens contributed to an early draft version of this document.

   As noted above, this document builds on two previous documents.

   The authors of [MPTCP-PLAIN] were:

   *  Mohamed Boucadair

   *  Christian Jacquenet

   *  Olivier Bonaventure

   *  Denis Behaghel

   *  Stefano Secci

   *  Wim Henderickx

   *  Robert Skog

   *  Suresh Vinapamula

   *  SungHoon Seo

   *  Wouter Cloetens

   *  Ullrich Meyer

   *  Luis M. Contreras

   *  Bart Peirens

   The authors of [MPTCP-TRANSPARENT] were:

   *  Bart Peirens

   *  Gregory Detal

   *  Sebastien Barre

   *  Olivier Bonaventure

Authors' Addresses



   Olivier Bonaventure (editor)
   Tessares
   Avenue Jean Monnet 1
   B-1348 Louvain-la-Neuve
   Belgium

   Email: Olivier.Bonaventure@tessares.net


   Mohamed Boucadair (editor)
   Orange
   Clos Courtel
   35000 Rennes
   France

   Email: mohamed.boucadair@orange.com


   Sri Gundavelli
   Cisco
   170 West Tasman Drive
   San Jose, CA 95134
   United States of America

   Email: sgundave@cisco.com


   SungHoon Seo
   Korea Telecom
   151 Taebong-ro
   Seocho-gu, Seoul, 06763
   Republic of Korea

   Email: sh.seo@kt.com


   Benjamin Hesmans
   Tessares
   Avenue Jean Monnet 1
   B-1348 Louvain-la-Neuve
   Belgium

   Email: Benjamin.Hesmans@tessares.net