Rfc9601
TitlePropagating Explicit Congestion Notification across IP Tunnel Headers Separated by a Shim
AuthorB. Briscoe
DateAugust 2024
Format:HTML, TXT, PDF, XML
UpdatesRFC2661, RFC2784, RFC3931, RFC4380, RFC6040, RFC7450
Status:PROPOSED STANDARD





Internet Engineering Task Force (IETF)                        B. Briscoe
Request for Comments: 9601                                   Independent
Updates: 2661, 2784, 3931, 4380, 6040, 7450                  August 2024
Category: Standards Track                                               
ISSN: 2070-1721


 Propagating Explicit Congestion Notification across IP Tunnel Headers
                          Separated by a Shim

Abstract

   RFC 6040 on "Tunnelling of Explicit Congestion Notification" made the
   rules for propagation of Explicit Congestion Notification (ECN)
   consistent for all forms of IP-in-IP tunnel.  This specification
   updates RFC 6040 to clarify that its scope includes tunnels where two
   IP headers are separated by at least one shim header that is not
   sufficient on its own for wide-area packet forwarding.  It surveys
   widely deployed IP tunnelling protocols that use such shim headers
   and updates the specifications of those that do not mention ECN
   propagation (including RFCs 2661, 3931, 2784, 4380 and 7450, which
   specify L2TPv2, L2TPv3, Generic Routing Encapsulation (GRE), Teredo,
   and Automatic Multicast Tunneling (AMT), respectively).  This
   specification also updates RFC 6040 with configuration requirements
   needed to make any legacy tunnel ingress safe.

Status of This Memo

   This is an Internet Standards Track document.

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

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   https://www.rfc-editor.org/info/rfc9601.

Copyright Notice

   Copyright (c) 2024 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (https://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Revised BSD License text as described in Section 4.e of the
   Trust Legal Provisions and are provided without warranty as described
   in the Revised BSD License.

   This document may contain material from IETF Documents or IETF
   Contributions published or made publicly available before November
   10, 2008.  The person(s) controlling the copyright in some of this
   material may not have granted the IETF Trust the right to allow
   modifications of such material outside the IETF Standards Process.
   Without obtaining an adequate license from the person(s) controlling
   the copyright in such materials, this document may not be modified
   outside the IETF Standards Process, and derivative works of it may
   not be created outside the IETF Standards Process, except to format
   it for publication as an RFC or to translate it into languages other
   than English.

Table of Contents

   1.  Introduction
   2.  Terminology
   3.  Scope of RFC 6040
     3.1.  Feasibility of ECN Propagation between Tunnel Headers
     3.2.  Desirability of ECN Propagation between Tunnel Headers
   4.  Making a Non-ECN Tunnel Ingress Safe by Configuration
   5.  ECN Propagation and Fragmentation/Reassembly
   6.  IP-in-IP Tunnels with Tightly Coupled Shim Headers
     6.1.  Specific Updates to Protocols under IETF Change Control
       6.1.1.  L2TP (v2 and v3) ECN Extension
       6.1.2.  GRE
       6.1.3.  Teredo
       6.1.4.  AMT
   7.  IANA Considerations
   8.  Security Considerations
   9.  References
     9.1.  Normative References
     9.2.  Informative References
   Acknowledgements
   Author's Address

1.  Introduction

   [RFC6040] on "Tunnelling of Explicit Congestion Notification" made
   the rules for propagation of Explicit Congestion Notification (ECN)
   [RFC3168] consistent for all forms of IP-in-IP tunnel.

   A common pattern for many tunnelling protocols is to encapsulate an
   inner IP header (v4 or v6) with one or more shim headers then an
   outer IP header (v4 or v6).  Some of these shim headers are designed
   as generic encapsulations, so they do not necessarily directly
   encapsulate an inner IP header.  Instead, they can encapsulate
   headers such as link-layer (L2) protocols that, in turn, often
   encapsulate IP.  Thus, the abbreviation 'IP-shim-(L2)-IP' can be used
   for tunnels that are in scope of this document.

   To clear up confusion, this specification clarifies that the scope of
   [RFC6040] includes any IP-in-IP tunnel, including those with one or
   more shim headers and other encapsulations between the IP headers.
   Where necessary, it updates the specifications of the relevant
   encapsulation protocols with the specific text necessary to comply
   with [RFC6040].

   This specification also updates [RFC6040] to state how operators
   ought to configure a legacy tunnel ingress to avoid unsafe system
   configurations.

2.  Terminology

   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.

   This specification uses the terminology defined in [RFC6040].

3.  Scope of RFC 6040

   In Section 1.1 of [RFC6040], its scope is defined as:

   |  ...ECN field processing at encapsulation and decapsulation for any
   |  IP-in-IP tunnelling, whether IPsec or non-IPsec tunnels.  It
   |  applies irrespective of whether IPv4 or IPv6 is used for either
   |  the inner or outer headers.

   There are two problems with the above scoping statement:

   Problem 1: It was intended to include cases where one or more shim
   headers sit between the IP headers.  Many tunnelling implementers
   have interpreted the scope of [RFC6040] as it was intended, but it is
   ambiguous.  Therefore, this specification updates [RFC6040] by adding
   the following scoping text after the sentences quoted above:

   |  It applies in cases where an outer IP header encapsulates an inner
   |  IP header either directly or indirectly by encapsulating other
   |  headers that in turn encapsulate (or might encapsulate) an inner
   |  IP header.

   Problem 2: Like many IETF specifications, [RFC6040] is written as a
   specification that implementations can choose to claim compliance
   with.  This means it does not cover two important situations:

   1.  Cases where it is infeasible for an implementation to access an
       inner IP header when adding or removing an outer IP header

   2.  Cases where implementations choose not to propagate ECN between
       IP headers

   However, the ECN field is a non-optional part of the IP header (v4
   and v6), so any implementation that creates an outer IP header has to
   give the ECN field some value.  There is only one safe value a tunnel
   ingress can use if it does not know whether the egress supports
   propagation of the ECN field; it has to clear the ECN field in any
   outer IP header to 0b00.

   However, an RFC has no jurisdiction over implementations that choose
   not to comply or cannot comply with the RFC, including all
   implementations that predated it.  Therefore, it would have been
   unreasonable to add such a requirement to [RFC6040].  Nonetheless, to
   ensure safe propagation of the ECN field over tunnels, it is
   reasonable to add requirements on operators to ensure they configure
   their tunnels safely (where possible).  Before resolving 'Problem 2'
   by stating these configuration requirements (in Section 4), the
   factors that determine whether propagating ECN is feasible or
   desirable will be briefly introduced.

3.1.  Feasibility of ECN Propagation between Tunnel Headers

   In many cases, one or more shim headers and an outer IP header are
   always added to (or removed from) an inner IP packet as part of the
   same procedure.  We call these tightly coupled shim headers.
   Processing a shim and outer header together is often necessary
   because a shim is not sufficient for packet forwarding in its own
   right; not unless complemented by an outer header.  In these cases,
   it will often be feasible for an implementation to propagate the ECN
   field between the IP headers.

   In some cases, a tunnel adds an outer IP header and a tightly coupled
   shim header to an inner header that is not an IP header, but that, in
   turn, encapsulates an IP header (or might encapsulate an IP header).
   For instance, an inner Ethernet (or other link-layer) header might
   encapsulate an inner IP header as its payload.  We call this a
   tightly coupled shim over an encapsulating header.

   Digging to arbitrary depths to find an inner IP header within an
   encapsulation is strictly a layering violation, so it cannot be a
   required behaviour.  Nonetheless, some tunnel endpoints already look
   within a Layer 2 (L2) header for an IP header, for instance, to map
   the Diffserv codepoint between an encapsulated IP header and an outer
   IP header [RFC2983].  In such cases at least, it should be feasible
   to also (independently) propagate the ECN field between the same IP
   headers.  Thus, access to the ECN field within an encapsulating
   header can be a useful and benign optimization.  The guidelines in
   Section 5 of [RFC9599] give the conditions for this layering
   violation to be benign.

3.2.  Desirability of ECN Propagation between Tunnel Headers

   Developers and network operators are encouraged to implement and
   deploy tunnel endpoints compliant with [RFC6040] (as updated by the
   present specification) in order to provide the benefits of wider ECN
   deployment [RFC8087].  Nonetheless, propagation of ECN between IP
   headers, whether separated by shim headers or not, has to be optional
   to implement and to use, because:

   *  legacy implementations of tunnels without any ECN support already
      exist;

   *  a network might be designed so that there is usually no bottleneck
      within the tunnel; and

   *  if the tunnel endpoints would have to search within an L2 header
      to find an encapsulated IP header, it might not be worth the
      potential performance hit.

4.  Making a Non-ECN Tunnel Ingress Safe by Configuration

   Even when no specific attempt has been made to implement propagation
   of the ECN field at a tunnel ingress, it ought to be possible for the
   operator to render a tunnel ingress safe by configuration.  The main
   safety concern is to disable (clear to zero) the ECN capability in
   the outer IP header at the ingress if the egress of the tunnel does
   not implement ECN logic to propagate any ECN markings into the packet
   forwarded beyond the tunnel.  Otherwise, the non-ECN egress could
   discard any ECN marking introduced within the tunnel, which would
   break all the ECN-based control loops that regulate the traffic load
   over the tunnel.

   Therefore, this specification updates Section 4.3 of [RFC6040] by
   inserting the following text at the end of the section:

   |  Whether or not an ingress implementation claims compliance with
   |  [RFC6040], [RFC4301], or [RFC3168], when the outer tunnel header
   |  is IP (v4 or v6), if possible, the ingress MUST be configured to
   |  zero the outer ECN field in all of the following cases:
   |  
   |  *  if it is known that the tunnel egress does not support any of
   |     the RFCs that define propagation of the ECN field ([RFC6040],
   |     [RFC4301], or the full functionality mode of [RFC3168]);
   |  
   |  *  if the behaviour of the egress is not known or an egress with
   |     unknown behaviour might be dynamically paired with the ingress
   |     (one way for an operator of a tunnel ingress to determine the
   |     behaviour of an otherwise unknown egress is described in
   |     [decap-test]);
   |  
   |  *  if an IP header might be encapsulated within a non-IP header
   |     that the tunnel ingress is encapsulating, but the ingress does
   |     not inspect within the encapsulation.
   |  
   |  For the avoidance of doubt, the above only concerns the outer IP
   |  header.  The ingress MUST NOT alter the ECN field of the arriving
   |  IP header that will become the inner IP header.
   |  
   |  In order that the network operator can comply with the above
   |  safety rules, an implementation of a tunnel ingress:
   |  
   |  *  MUST NOT treat the former Type of Service (ToS) octet (IPv4) or
   |     the former Traffic Class octet (IPv6) as a single 8-bit field.
   |     This is because the resulting linkage of ECN and Diffserv field
   |     propagation between inner and outer headers is not consistent
   |     with the definition of the 6-bit Diffserv field in [RFC2474]
   |     and [RFC3260].
   |  
   |  *  SHOULD be able to be configured to zero the ECN field of the
   |     outer header.
   |  
   |  These last two rules apply even if an implementation of a tunnel
   |  ingress does not claim to support [RFC6040], [RFC4301], or the
   |  full functionality mode of [RFC3168]

   For instance, if a tunnel ingress with no ECN-specific logic had a
   configuration capability to refer to the last 2 bits of the old ToS
   Byte of the outer (e.g., with a 0x3 mask) and set them to zero, while
   also being able to allow the DSCP to be re-mapped independently, that
   would be sufficient to satisfy both implementation requirements
   above.

   There might be concern that the above "MUST NOT" makes compliant
   implementations non-compliant at a stroke.  However, by definition,
   it solely applies to equipment that provides Diffserv configuration.
   Any such Diffserv equipment that is configuring treatment of the
   former ToS octet (IPv4) or the former Traffic Class octet (IPv6) as a
   single 8-bit field must have always been non-compliant with the
   definition of the 6-bit Diffserv field in [RFC2474] and [RFC3260].
   If a tunnel ingress does not have any ECN logic, copying the ECN
   field as a side effect of copying the DSCP is a seriously unsafe bug
   that risks breaking the feedback loops that regulate load on a
   tunnel, because it omits to check the ECN capability of the tunnel
   egress.

   Zeroing the outer ECN field of all packets in all circumstances would
   be safe, but it would not be sufficient to claim compliance with
   [RFC6040] because it would not meet the aim of introducing ECN
   support to tunnels (see Section 4.3 of [RFC6040]).

5.  ECN Propagation and Fragmentation/Reassembly

   The following requirements update [RFC6040], which omitted handling
   of the ECN field during fragmentation or reassembly.  These changes
   might alter how many ECN-marked packets are propagated by a tunnel
   that fragments packets, but this would not raise any backward
   compatibility issues.

   If a tunnel ingress fragments a packet, it MUST set the outer ECN
   field of all the fragments to the same value as it would have set if
   it had not fragmented the packet.

   Section 5.3 of [RFC3168] specifies ECN requirements for reassembly of
   sets of 'outer fragments' into packets (in 'outer fragmentation', the
   fragmentation is visible in the outer header so that the tunnel
   egress can reassemble the fragments [INTAREA-TUNNELS]).
   Additionally, the following requirements apply at a tunnel egress:

   *  During reassembly of outer fragments, the packet MUST be discarded
      if the ECN fields of the outer headers being reassembled into a
      single packet consist of a mixture of Not ECN-Capable Transport
      (Not-ECT) and other ECN codepoints.

   *  If there is mix of ECT(0) and ECT(1) outer fragments, then the
      reassembled packet MUST be set to ECT(1).

      Reasoning: [RFC3168] originally defined ECT(0) and ECT(1) as
      equivalent, but [RFC3168] has been updated by [RFC8311] to make
      ECT(1) available for congestion marking differences.  The rule is
      independent of the current experimental use of ECT(1) for Low
      Latency, Low Loss, and Scalable throughput (L4S) [RFC9331].  The
      rule is compatible with Pre-Congestion Notification (PCN)
      [RFC6660], which uses 2 levels of congestion severity, with the
      ranking of severity from highest to lowest being Congestion
      Experienced (CE), ECT(1), ECT(0).  The decapsulation rules in
      [RFC6040] take a similar approach.

6.  IP-in-IP Tunnels with Tightly Coupled Shim Headers

   Below is a list of specifications of encapsulations with tightly
   coupled shim header(s) in rough chronological order.  This list is
   confined to Standards Track or widely deployed protocols.  So, for
   the avoidance of doubt, the updated scope of [RFC6040] is defined in
   Section 3 and is not limited to this list.

   *  Point-to-Point Tunneling Protocol (PPTP) [RFC2637]

   *  Layer Two Tunneling Protocol (L2TP), specifically L2TPv2 [RFC2661]
      and L2TPv3 [RFC3931], which not only includes all the L2-specific
      specializations of L2TP, but also derivatives such as the Keyed
      IPv6 Tunnel [RFC8159]

   *  Generic Routing Encapsulation (GRE) [RFC2784] and Network
      Virtualization using GRE (NVGRE) [RFC7637]

   *  GPRS Tunnelling Protocol (GTP), specifically GTPv1 [GTPv1], GTP v1
      User Plane [GTPv1-U], and GTP v2 Control Plane [GTPv2-C]

   *  Teredo [RFC4380]

   *  Control And Provisioning of Wireless Access Points (CAPWAP)
      [RFC5415]

   *  Locator/Identifier Separation Protocol (LISP) [RFC9300]

   *  Automatic Multicast Tunneling (AMT) [RFC7450]

   *  Virtual eXtensible Local Area Network (VXLAN) [RFC7348] and
      Generic Protocol Extensions for VXLAN (VXLAN-GPE) [NVO3-VXLAN-GPE]

   *  The Network Service Header (NSH) [RFC8300] for Service Function
      Chaining (SFC)

   *  Geneve [RFC8926]

   *  Direct tunnelling of an IP packet within a UDP/IP datagram (see
      Section 3.1.11 of [RFC8085])

   *  TCP Encapsulation of Internet Key Exchange Protocol (IKE) and
      IPsec Packets (see Section 9.5 of [RFC9329])

   Some of the listed protocols enable encapsulation of a variety of
   network layer protocols as inner and/or outer.  This specification
   applies to the cases where there is an inner and outer IP header as
   described in Section 3.  Otherwise, [RFC9599] gives guidance on how
   to design propagation of ECN into other protocols that might
   encapsulate IP.

   Where protocols in the above list need to be updated to specify ECN
   propagation and are under IETF change control, update text is given
   in the following subsections.  For those not under IETF control, it
   is RECOMMENDED that implementations of encapsulation and
   decapsulation comply with [RFC6040].  It is also RECOMMENDED that
   their specifications are updated to add a requirement to comply with
   [RFC6040] (as updated by the present document).

   PPTP is not under the change control of the IETF, but it has been
   documented in an Informational RFC [RFC2637].  However, there is no
   need for the present specification to update PPTP because L2TP has
   been developed as a standardized replacement.

   NVGRE is not under the change control of the IETF, but it has been
   documented in an Informational RFC [RFC7637].  NVGRE is a specific
   use case of GRE (it re-purposes the key field from the initial
   specification of GRE [RFC1701] as a Virtual Subnet ID).  Therefore,
   the text that updates GRE in Section 6.1.2 below is also intended to
   update NVGRE.

   Although the definition of the various GTP shim headers is under the
   control of the Third Generation Partnership Project (3GPP), it is
   hard to determine whether the 3GPP or the IETF controls
   standardization of the _process_ of adding both a GTP and an IP
   header to an inner IP header.  Nonetheless, the present specification
   is provided so that the 3GPP can refer to it from any of its own
   specifications of GTP and IP header processing.

   The specification of CAPWAP already specifies [RFC3168] ECN
   propagation and ECN capability negotiation.  Without modification,
   the CAPWAP specification already interworks with the backward-
   compatible updates to [RFC3168] in [RFC6040].

   LISP made the ECN propagation procedures in [RFC3168] mandatory from
   the start.  [RFC3168] has since been updated by [RFC6040], but the
   changes are backwards compatible, so there is still no need for LISP
   tunnel endpoints to negotiate their ECN capabilities.

   VXLAN is not under the change control of the IETF, but it has been
   documented in an Informational RFC.  It is RECOMMENDED that VXLAN
   implementations comply with [RFC6040] when the VXLAN header is
   inserted between (or removed from between) IP headers.  The authors
   of any future update of the VXLAN spec are also encouraged to add a
   requirement to comply with [RFC6040] as updated by the present
   specification.  In contrast, VXLAN-GPE is being documented under IETF
   change control and it does require compliance with [RFC6040].

   The Network Service Header (NSH) [RFC8300] has been defined as a
   shim-based encapsulation to identify the Service Function Path (SFP)
   in the Service Function Chaining (SFC) architecture [RFC7665].  A
   proposal has been made for the processing of ECN when handling
   transport encapsulation [SFC-NSH-ECN].

   The specification of Geneve already refers to [RFC6040] for ECN
   encapsulation.

   Section 3.1.11 of [RFC8085] already explains that a tunnel that
   encapsulates an IP header within a UDP/IP datagram needs to follow
   [RFC6040] when propagating the ECN field between inner and outer IP
   headers.  Section 3 of the present specification updates [RFC6040] to
   clarify that its scope includes cases with a shim header between the
   IP headers.  So it indirectly updates the scope of [RFC8085] to
   include cases with a shim header as well as a UDP header between the
   IP headers.

   The requirements in Section 4 update [RFC6040], and hence also
   indirectly update the UDP usage guidelines in [RFC8085] to add the
   important but previously unstated requirement that, if the UDP tunnel
   egress does not, or might not, support ECN propagation, a UDP tunnel
   ingress has to clear the outer IP ECN field to 0b00, e.g., by
   configuration.

   Section 9.5 of [RFC9329] already recommends the compatibility mode of
   [RFC6040] in this case because there is not a one-to-one mapping
   between inner and outer packets when TCP encapsulates IKE or IPsec.

6.1.  Specific Updates to Protocols under IETF Change Control

6.1.1.  L2TP (v2 and v3) ECN Extension

   The L2TP terminology used here is defined in [RFC2661] and [RFC3931].

   L2TPv3 [RFC3931] is used as a shim header between any packet-switched
   network (PSN) header (e.g., IPv4, IPv6, and MPLS) and many types of
   L2 headers.  The L2TPv3 shim header encapsulates an L2-specific sub-
   layer, then an L2 header that is likely to contain an inner IP header
   (v4 or v6).  Then this whole stack of headers can be encapsulated
   within an optional outer UDP header and an outer PSN header that is
   typically IP (v4 or v6).

   L2TPv2 is used as a shim header between any PSN header and a PPP
   header, which is in turn likely to encapsulate an IP header.

   Even though these shims are rather fat (particularly in the case of
   L2TPv3), they still fit the definition of a tightly coupled shim
   header over an encapsulating header (Section 3.1) because all the
   headers encapsulating the L2 header are added (or removed) together.
   L2TPv2 and L2TPv3 are therefore within the scope of [RFC6040], as
   updated by Section 3.

   Implementation of the ECN extension to L2TPv2 and L2TPv3 defined in
   Section 6.1.1.2 is RECOMMENDED in order to provide the benefits of
   ECN [RFC8087] whenever a node within an L2TP tunnel becomes the
   bottleneck for an end-to-end traffic flow.

6.1.1.1.  Safe Configuration of a "Non-ECN" Ingress LCCE

   The following text is appended to both Section 5.3 of [RFC2661] and
   Section 4.5 of [RFC3931] as an update to the base L2TPv2 and L2TPv3
   specifications:

   |  The operator of an LCCE that does not support the ECN extension in
   |  Section 6.1.1.2 of RFC 9601 MUST follow the configuration
   |  requirements in Section 4 of RFC 9601 to ensure it clears the
   |  outer IP ECN field to 0b00 when the outer PSN header is IP (v4 or
   |  v6).

   In particular, for an L2TP Control Connection Endpoint (LCCE)
   implementation that does not support the ECN extension, this means
   that configuration of how it propagates the ECN field between inner
   and outer IP headers MUST be independent of any configuration of the
   Diffserv extension of L2TP [RFC3308].

6.1.1.2.  ECN Extension for L2TP (v2 or v3)

   When the outer PSN header and the payload inside the L2 header are
   both IP (v4 or v6), an LCCE will propagate the ECN field at ingress
   and egress by following the rules in Section 4 of [RFC6040].

   Before encapsulating any data packets, [RFC6040] requires an ingress
   LCCE to check that the egress LCCE supports ECN propagation as
   defined in [RFC6040] or one of its compatible predecessors ([RFC4301]
   or the full functionality mode of [RFC3168]).  If the egress supports
   ECN propagation, the ingress LCCE can use the normal mode of
   encapsulation (copying the ECN field from inner to outer).
   Otherwise, the ingress LCCE has to use compatibility mode [RFC6040]
   (clearing the outer IP ECN field to 0b00).

   An LCCE can determine the remote LCCE's support for ECN either
   statically (by configuration) or by dynamic discovery during setup of
   each control connection between the LCCEs using the ECN Capability
   Attribute-Value Pair (AVP) defined in Section 6.1.1.2.1.

   Where the outer PSN header is some protocol other than IP that
   supports ECN, the appropriate ECN propagation specification will need
   to be followed, e.g., [RFC5129] for MPLS.  Where no specification
   exists for ECN propagation by a particular PSN, [RFC9599] gives
   general guidance on how to design ECN propagation into a protocol
   that encapsulates IP.

6.1.1.2.1.  ECN Capability AVP for Negotiation between LCCEs

   The ECN Capability AVP defined here has Attribute Type 103.  The AVP
   has the following format:

    0                   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
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |M|H|0|0|0|0|      Length       |          Vendor ID            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |             103               |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

              Figure 1: ECN Capability AVP for L2TP (v2 or v3)

   This AVP MAY be present in the Start-Control-Connection-Request
   (SCCRQ) and Start-Control-Connection-Reply (SCCRP) message types.
   This AVP MAY be hidden (the H-bit is set to 0 or 1) and is optional
   (the M-bit is not set).  The length (before hiding) of this AVP is 6
   octets.  The Vendor ID is the IETF Vendor ID of 0.

   When an LCCE sends an ECN Capability AVP, it indicates that it
   supports ECN propagation.  When no ECN Capability AVP is present, it
   indicates that the sender does not support ECN propagation.

   If an LCCE initiating a control connection supports ECN propagation,
   it will send an SCCRQ containing an ECN Capability AVP.  If the
   tunnel terminator supports ECN, it will return an SCCRP that also
   includes an ECN Capability AVP.  Then, for any sessions created by
   that control connection, both ends of the tunnel can use the normal
   mode of [RFC6040]; i.e., they can copy the IP ECN field from inner to
   outer when encapsulating data packets.

   On the other hand, if the tunnel terminator does not support ECN, it
   will ignore the ECN Capability AVP and send an SCCRP to the tunnel
   initiator without an ECN Capability AVP.  The tunnel initiator
   interprets the absence of the ECN Capability flag in the SCCRP as an
   indication that the tunnel terminator is incapable of supporting ECN.
   When encapsulating data packets for any sessions created by that
   control connection, the tunnel initiator will then use the
   compatibility mode of [RFC6040] to clear the ECN field of the outer
   IP header to 0b00.

   If the tunnel terminator does not support this ECN extension, the
   network operator is still expected to configure it to comply with the
   safety provisions set out in Section 6.1.1.1 when it acts as an
   ingress LCCE.

   If ECN support by the ingress and egress LCCEs is configured
   statically, as allowed in Section 6.1.1.2, they both ignore the
   presence or absence of any ECN capability AVP.

6.1.2.  GRE

   The GRE terminology used here is defined in [RFC2784].  GRE is often
   used as a tightly coupled shim header between IP headers.  Sometimes,
   the GRE shim header encapsulates an L2 header, which might in turn
   encapsulate an IP header.  Therefore, GRE is within the scope of
   [RFC6040] as updated by Section 3.

   Implementation of support for [RFC6040] as updated by the present
   specification is RECOMMENDED for GRE tunnel endpoints in order to
   provide the benefits of ECN [RFC8087] whenever a node within a GRE
   tunnel becomes the bottleneck for an end-to-end IP traffic flow
   tunnelled over GRE using IP as the delivery protocol (outer header).

   GRE itself does not support dynamic setup and configuration of
   tunnels.  However, control plane protocols, such as Next Hop
   Resolution Protocol (NHRP) [RFC2332], Mobile IPv4 (MIP4) [RFC5944],
   Mobile IPv6 (MIP6) [RFC6275], Proxy Mobile IP (PMIP) [RFC5845], and
   IKEv2 [RFC7296], are sometimes used to set up GRE tunnels
   dynamically.

   When these control protocols set up IP-in-IP or IPsec tunnels, it is
   likely that the resulting tunnels will propagate the ECN field as
   defined in [RFC6040] or one of its compatible predecessors ([RFC4301]
   or the full functionality mode of [RFC3168]).  However, if they use a
   GRE encapsulation, this presumption is less sound.

   Therefore, if the outer delivery protocol is IP (v4 or v6), the
   operator is obliged to follow the safe configuration requirements in
   Section 4.  Section 6.1.2.1 updates the base GRE specification with
   this requirement to emphasize its importance.

   Where the delivery protocol is some protocol other than IP that
   supports ECN, the appropriate ECN propagation specification will need
   to be followed, e.g., [RFC5129] for MPLS.  Where no specification
   exists for ECN propagation by a particular PSN, [RFC9599] gives more
   general guidance on how to propagate ECN to and from protocols that
   encapsulate IP.

6.1.2.1.  Safe Configuration of a "Non-ECN" GRE Ingress

   The following text is appended to Section 3 of [RFC2784] as an update
   to the base GRE specification:

   |  The operator of a GRE tunnel ingress MUST follow the configuration
   |  requirements in Section 4 of RFC 9601 when the outer delivery
   |  protocol is IP (v4 or v6).

6.1.3.  Teredo

   Teredo [RFC4380] provides a way to tunnel IPv6 over an IPv4 network
   with a UDP-based shim header between the two.

   For Teredo tunnel endpoints to provide the benefits of ECN, the
   Teredo specification would have to be updated to include negotiation
   of the ECN capability between Teredo tunnel endpoints.  Otherwise, it
   would be unsafe for a Teredo tunnel ingress to copy the ECN field to
   the IPv6 outer.

   Those implementations known to the authors at the time of writing do
   not support propagation of ECN, but they do safely zero the ECN field
   in the outer IPv6 header.  However, the specification does not
   mention anything about this.

   To make existing Teredo deployments safe, it would be possible to add
   ECN capability negotiation to those that are subject to remote OS
   update.  However, for those implementations not subject to remote OS
   update, it will not be feasible to require them to be configured
   correctly because Teredo tunnel endpoints are generally deployed on
   hosts.

   Therefore, until ECN support is added to the specification of Teredo,
   the only feasible further safety precaution available here is to
   update the specification of Teredo implementations with the following
   text as a new section:

   |  5.1.3.  Safe "Non-ECN" Teredo Encapsulation
   |  
   |  A Teredo tunnel ingress implementation that does not support ECN
   |  propagation as defined in [RFC6040] or one of its compatible
   |  predecessors ([RFC4301] or the full functionality mode of
   |  [RFC3168]) MUST zero the ECN field in the outer IPv6 header.

6.1.4.  AMT

   AMT [RFC7450] is a tightly coupled shim header that encapsulates an
   IP packet and is encapsulated within a UDP/IP datagram.  Therefore,
   AMT is within the scope of [RFC6040] as updated by Section 3.

   Implementation of support for [RFC6040] as updated by the present
   specification is RECOMMENDED for AMT tunnel endpoints in order to
   provide the benefits of ECN [RFC8087] whenever a node within an AMT
   tunnel becomes the bottleneck for an IP traffic flow tunnelled over
   AMT.

   To comply with [RFC6040], an AMT relay and gateway will follow the
   rules for propagation of the ECN field at ingress and egress,
   respectively, as described in Section 4 of [RFC6040].

   Before encapsulating any data packets, [RFC6040] requires an ingress
   AMT relay to check that the egress AMT gateway supports ECN
   propagation as defined in [RFC6040] or one of its compatible
   predecessors ([RFC4301] or the full functionality mode of [RFC3168]).
   If the egress gateway supports ECN, the ingress relay can use the
   normal mode of encapsulation (copying the IP ECN field from inner to
   outer).  Otherwise, the ingress relay has to use compatibility mode,
   which means it has to clear the outer ECN field to zero [RFC6040].

   An AMT tunnel is created dynamically (not manually), so the relay
   will need to determine the remote gateway's support for ECN using the
   ECN capability declaration defined in Section 6.1.4.2.

6.1.4.1.  Safe Configuration of a "Non-ECN" Ingress AMT Relay

   The following text is appended to Section 4.2.2 of [RFC7450] as an
   update to the AMT specification:

   |  The operator of an AMT relay that does not support [RFC6040] or
   |  one of its compatible predecessors ([RFC4301] or the full
   |  functionality mode of [RFC3168]) MUST follow the configuration
   |  requirements in Section 4 of RFC 9601 to ensure it clears the
   |  outer IP ECN field to zero.

6.1.4.2.  ECN Capability Declaration of an AMT Gateway

    0                   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
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |  V=0  |Type=3 |  Reserved |E|P|            Reserved           |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                         Request Nonce                         |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                Figure 2: Updated AMT Request Message Format

   Bit 14 of the AMT Request Message counting from 0 (or bit 7 of the
   Reserved field counting from 1) is defined here as the AMT Gateway
   ECN Capability flag (E) as shown in Figure 2.  The definitions of all
   other fields in the AMT Request Message are unchanged from [RFC7450].

   When the E flag is set to 1, it indicates that the sender of the
   message supports [RFC6040] ECN propagation.  When it is cleared to
   zero, it indicates the sender of the message does not support
   [RFC6040] ECN propagation.  An AMT gateway "that supports [RFC6040]
   ECN propagation" means one that propagates the ECN field to the
   forwarded data packet based on the combination of arriving inner and
   outer ECN fields as defined in Section 4 of [RFC6040].

   The other bits of the Reserved field remain reserved.  They will
   continue to be cleared to zero when sent and ignored when either
   received or forwarded as specified in Section 5.1.3.3 of [RFC7450].

   An AMT gateway that does not support [RFC6040] MUST NOT set the E
   flag of its Request Message to 1.

   An AMT gateway that supports [RFC6040] ECN propagation MUST set the E
   flag of its Relay Discovery Message to 1.

   The action of the corresponding AMT relay that receives a Request
   message with the E flag set to 1 depends on whether the relay itself
   supports [RFC6040] ECN propagation:

   *  If the relay supports [RFC6040] ECN propagation, it will store the
      ECN capability of the gateway along with its address.  Then,
      whenever it tunnels datagrams towards this gateway, it MUST use
      the normal mode of [RFC6040] to propagate the ECN field when
      encapsulating datagrams (i.e., it copies the IP ECN field from
      inner to outer header).

   *  If the discovered AMT relay does not support [RFC6040] ECN
      propagation, it will ignore the E flag in the Reserved field as
      per Section 5.1.3.3 of [RFC7450].

      If the AMT relay does not support [RFC6040] ECN propagation, the
      network operator is still expected to configure it to comply with
      the safety provisions set out in Section 6.1.4.1.

7.  IANA Considerations

   IANA has assigned the following AVP in the L2TP "Control Message
   Attribute Value Pairs" registry:

              +================+================+===========+
              | Attribute Type | Description    | Reference |
              +================+================+===========+
              | 103            | ECN Capability | RFC 9601  |
              +----------------+----------------+-----------+

                                  Table 1

8.  Security Considerations

   The Security Considerations in [RFC6040] and [RFC9599] apply equally
   to the scope defined for the present specification.

9.  References

9.1.  Normative References

   [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>.

   [RFC2474]  Nichols, K., Blake, S., Baker, F., and D. Black,
              "Definition of the Differentiated Services Field (DS
              Field) in the IPv4 and IPv6 Headers", RFC 2474,
              DOI 10.17487/RFC2474, December 1998,
              <https://www.rfc-editor.org/info/rfc2474>.

   [RFC2661]  Townsley, W., Valencia, A., Rubens, A., Pall, G., Zorn,
              G., and B. Palter, "Layer Two Tunneling Protocol "L2TP"",
              RFC 2661, DOI 10.17487/RFC2661, August 1999,
              <https://www.rfc-editor.org/info/rfc2661>.

   [RFC2784]  Farinacci, D., Li, T., Hanks, S., Meyer, D., and P.
              Traina, "Generic Routing Encapsulation (GRE)", RFC 2784,
              DOI 10.17487/RFC2784, March 2000,
              <https://www.rfc-editor.org/info/rfc2784>.

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

   [RFC3931]  Lau, J., Ed., Townsley, M., Ed., and I. Goyret, Ed.,
              "Layer Two Tunneling Protocol - Version 3 (L2TPv3)",
              RFC 3931, DOI 10.17487/RFC3931, March 2005,
              <https://www.rfc-editor.org/info/rfc3931>.

   [RFC4301]  Kent, S. and K. Seo, "Security Architecture for the
              Internet Protocol", RFC 4301, DOI 10.17487/RFC4301,
              December 2005, <https://www.rfc-editor.org/info/rfc4301>.

   [RFC4380]  Huitema, C., "Teredo: Tunneling IPv6 over UDP through
              Network Address Translations (NATs)", RFC 4380,
              DOI 10.17487/RFC4380, February 2006,
              <https://www.rfc-editor.org/info/rfc4380>.

   [RFC5129]  Davie, B., Briscoe, B., and J. Tay, "Explicit Congestion
              Marking in MPLS", RFC 5129, DOI 10.17487/RFC5129, January
              2008, <https://www.rfc-editor.org/info/rfc5129>.

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

   [RFC6660]  Briscoe, B., Moncaster, T., and M. Menth, "Encoding Three
              Pre-Congestion Notification (PCN) States in the IP Header
              Using a Single Diffserv Codepoint (DSCP)", RFC 6660,
              DOI 10.17487/RFC6660, July 2012,
              <https://www.rfc-editor.org/info/rfc6660>.

   [RFC7450]  Bumgardner, G., "Automatic Multicast Tunneling", RFC 7450,
              DOI 10.17487/RFC7450, February 2015,
              <https://www.rfc-editor.org/info/rfc7450>.

   [RFC9599]  Briscoe, B. and J. Kaippallimalil, "Guidelines for Adding
              Congestion Notification to Protocols that Encapsulate IP",
              RFC 9599, DOI 10.17487/RFC9599, August 2024,
              <https://www.rfc-editor.org/info/rfc9599>.

9.2.  Informative References

   [decap-test]
              Briscoe, B., "A Test for IP-ECN Propagation by a Remote
              Tunnel Endpoint", Technical Report, TR-BB-2023-003,
              DOI 10.48550/arXiv.2311.16825, November 2023,
              <https://arxiv.org/abs/2311.16825>.

   [GTPv1]    3GPP, "General Packet Radio Service (GPRS); GPRS
              Tunnelling Protocol (GTP) across the Gn and Gp interface",
              Technical Specification 29.060.

   [GTPv1-U]  3GPP, "General Packet Radio System (GPRS) Tunnelling
              Protocol User Plane (GTPv1-U)", Technical
              Specification 29.281.

   [GTPv2-C]  3GPP, "3GPP Evolved Packet System (EPS); Evolved General
              Packet Radio Service (GPRS) Tunnelling Protocol for
              Control plane (GTPv2-C); Stage 3", Technical
              Specification 29.274.

   [INTAREA-TUNNELS]
              Touch, J. D. and M. Townsley, "IP Tunnels in the Internet
              Architecture", Work in Progress, Internet-Draft, draft-
              ietf-intarea-tunnels-13, 26 March 2023,
              <https://datatracker.ietf.org/doc/html/draft-ietf-intarea-
              tunnels-13>.

   [NVO3-VXLAN-GPE]
              Maino, F., Kreeger, L., and U. Elzur, "Generic Protocol
              Extension for VXLAN (VXLAN-GPE)", Work in Progress,
              Internet-Draft, draft-ietf-nvo3-vxlan-gpe-13, 4 November
              2023, <https://datatracker.ietf.org/doc/html/draft-ietf-
              nvo3-vxlan-gpe-13>.

   [RFC1701]  Hanks, S., Li, T., Farinacci, D., and P. Traina, "Generic
              Routing Encapsulation (GRE)", RFC 1701,
              DOI 10.17487/RFC1701, October 1994,
              <https://www.rfc-editor.org/info/rfc1701>.

   [RFC2332]  Luciani, J., Katz, D., Piscitello, D., Cole, B., and N.
              Doraswamy, "NBMA Next Hop Resolution Protocol (NHRP)",
              RFC 2332, DOI 10.17487/RFC2332, April 1998,
              <https://www.rfc-editor.org/info/rfc2332>.

   [RFC2637]  Hamzeh, K., Pall, G., Verthein, W., Taarud, J., Little,
              W., and G. Zorn, "Point-to-Point Tunneling Protocol
              (PPTP)", RFC 2637, DOI 10.17487/RFC2637, July 1999,
              <https://www.rfc-editor.org/info/rfc2637>.

   [RFC2983]  Black, D., "Differentiated Services and Tunnels",
              RFC 2983, DOI 10.17487/RFC2983, October 2000,
              <https://www.rfc-editor.org/info/rfc2983>.

   [RFC3260]  Grossman, D., "New Terminology and Clarifications for
              Diffserv", RFC 3260, DOI 10.17487/RFC3260, April 2002,
              <https://www.rfc-editor.org/info/rfc3260>.

   [RFC3308]  Calhoun, P., Luo, W., McPherson, D., and K. Peirce, "Layer
              Two Tunneling Protocol (L2TP) Differentiated Services
              Extension", RFC 3308, DOI 10.17487/RFC3308, November 2002,
              <https://www.rfc-editor.org/info/rfc3308>.

   [RFC5415]  Calhoun, P., Ed., Montemurro, M., Ed., and D. Stanley,
              Ed., "Control And Provisioning of Wireless Access Points
              (CAPWAP) Protocol Specification", RFC 5415,
              DOI 10.17487/RFC5415, March 2009,
              <https://www.rfc-editor.org/info/rfc5415>.

   [RFC5845]  Muhanna, A., Khalil, M., Gundavelli, S., and K. Leung,
              "Generic Routing Encapsulation (GRE) Key Option for Proxy
              Mobile IPv6", RFC 5845, DOI 10.17487/RFC5845, June 2010,
              <https://www.rfc-editor.org/info/rfc5845>.

   [RFC5944]  Perkins, C., Ed., "IP Mobility Support for IPv4, Revised",
              RFC 5944, DOI 10.17487/RFC5944, November 2010,
              <https://www.rfc-editor.org/info/rfc5944>.

   [RFC6275]  Perkins, C., Ed., Johnson, D., and J. Arkko, "Mobility
              Support in IPv6", RFC 6275, DOI 10.17487/RFC6275, July
              2011, <https://www.rfc-editor.org/info/rfc6275>.

   [RFC7059]  Steffann, S., van Beijnum, I., and R. van Rein, "A
              Comparison of IPv6-over-IPv4 Tunnel Mechanisms", RFC 7059,
              DOI 10.17487/RFC7059, November 2013,
              <https://www.rfc-editor.org/info/rfc7059>.

   [RFC7296]  Kaufman, C., Hoffman, P., Nir, Y., Eronen, P., and T.
              Kivinen, "Internet Key Exchange Protocol Version 2
              (IKEv2)", STD 79, RFC 7296, DOI 10.17487/RFC7296, October
              2014, <https://www.rfc-editor.org/info/rfc7296>.

   [RFC7348]  Mahalingam, M., Dutt, D., Duda, K., Agarwal, P., Kreeger,
              L., Sridhar, T., Bursell, M., and C. Wright, "Virtual
              eXtensible Local Area Network (VXLAN): A Framework for
              Overlaying Virtualized Layer 2 Networks over Layer 3
              Networks", RFC 7348, DOI 10.17487/RFC7348, August 2014,
              <https://www.rfc-editor.org/info/rfc7348>.

   [RFC7637]  Garg, P., Ed. and Y. Wang, Ed., "NVGRE: Network
              Virtualization Using Generic Routing Encapsulation",
              RFC 7637, DOI 10.17487/RFC7637, September 2015,
              <https://www.rfc-editor.org/info/rfc7637>.

   [RFC7665]  Halpern, J., Ed. and C. Pignataro, Ed., "Service Function
              Chaining (SFC) Architecture", RFC 7665,
              DOI 10.17487/RFC7665, October 2015,
              <https://www.rfc-editor.org/info/rfc7665>.

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

   [RFC8087]  Fairhurst, G. and M. Welzl, "The Benefits of Using
              Explicit Congestion Notification (ECN)", RFC 8087,
              DOI 10.17487/RFC8087, March 2017,
              <https://www.rfc-editor.org/info/rfc8087>.

   [RFC8159]  Konstantynowicz, M., Ed., Heron, G., Ed., Schatzmayr, R.,
              and W. Henderickx, "Keyed IPv6 Tunnel", RFC 8159,
              DOI 10.17487/RFC8159, May 2017,
              <https://www.rfc-editor.org/info/rfc8159>.

   [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>.

   [RFC8300]  Quinn, P., Ed., Elzur, U., Ed., and C. Pignataro, Ed.,
              "Network Service Header (NSH)", RFC 8300,
              DOI 10.17487/RFC8300, January 2018,
              <https://www.rfc-editor.org/info/rfc8300>.

   [RFC8311]  Black, D., "Relaxing Restrictions on Explicit Congestion
              Notification (ECN) Experimentation", RFC 8311,
              DOI 10.17487/RFC8311, January 2018,
              <https://www.rfc-editor.org/info/rfc8311>.

   [RFC8926]  Gross, J., Ed., Ganga, I., Ed., and T. Sridhar, Ed.,
              "Geneve: Generic Network Virtualization Encapsulation",
              RFC 8926, DOI 10.17487/RFC8926, November 2020,
              <https://www.rfc-editor.org/info/rfc8926>.

   [RFC9300]  Farinacci, D., Fuller, V., Meyer, D., Lewis, D., and A.
              Cabellos, Ed., "The Locator/ID Separation Protocol
              (LISP)", RFC 9300, DOI 10.17487/RFC9300, October 2022,
              <https://www.rfc-editor.org/info/rfc9300>.

   [RFC9329]  Pauly, T. and V. Smyslov, "TCP Encapsulation of Internet
              Key Exchange Protocol (IKE) and IPsec Packets", RFC 9329,
              DOI 10.17487/RFC9329, November 2022,
              <https://www.rfc-editor.org/info/rfc9329>.

   [RFC9331]  De Schepper, K. and B. Briscoe, Ed., "The Explicit
              Congestion Notification (ECN) Protocol for Low Latency,
              Low Loss, and Scalable Throughput (L4S)", RFC 9331,
              DOI 10.17487/RFC9331, January 2023,
              <https://www.rfc-editor.org/info/rfc9331>.

   [SFC-NSH-ECN]
              Eastlake 3rd, D., Briscoe, B., Zhuang, S., Malis, A., and
              X. Wei, "Explicit Congestion Notification (ECN) and
              Congestion Feedback Using the Network Service Header (NSH)
              and IPFIX", Work in Progress, Internet-Draft, draft-ietf-
              sfc-nsh-ecn-support-13, 15 April 2024,
              <https://datatracker.ietf.org/doc/html/draft-ietf-sfc-nsh-
              ecn-support-13>.

Acknowledgements

   Thanks to Ing-jyh (Inton) Tsang for initial discussions on the need
   for ECN propagation in L2TP and its applicability.  Thanks also to
   Carlos Pignataro, Tom Herbert, Ignacio Goyret, Alia Atlas, Praveen
   Balasubramanian, Joe Touch, Mohamed Boucadair, David Black, Jake
   Holland, Sri Gundavelli, Gorry Fairhurst, and Martin Duke for helpful
   advice and comments.  [RFC7059] helped to identify a number of
   tunnelling protocols to include within the scope of this document.

   Bob Briscoe was part-funded by the Research Council of Norway through
   the TimeIn project for early drafts, and he was funded by Apple Inc.
   for later draft versions (from -17).  The views expressed here are
   solely those of the authors.

Author's Address