Internet Engineering Task Force (IETF)                         S. Bryant
Request for Comments: 8469                                      A. Malis
Updates: 4448                                                     Huawei
Category: Standards Track                                    I. Bagdonas
ISSN: 2070-1721                                                  Equinix
                                                          November 2018


           Recommendation to Use the Ethernet Control Word

Abstract

  The pseudowire (PW) encapsulation of Ethernet, as defined in
  RFC 4448, specifies that the use of the control word (CW) is
  optional.  In the absence of the CW, an Ethernet PW packet can be
  misidentified as an IP packet by a label switching router (LSR).
  This may lead to the selection of the wrong equal-cost multipath
  (ECMP) path for the packet, leading in turn to the misordering of
  packets.  This problem has become more serious due to the deployment
  of equipment with Ethernet Media Access Control (MAC) addresses that
  start with 0x4 or 0x6.  The use of the Ethernet PW CW addresses this
  problem.  This document RECOMMENDS the use of the Ethernet PW CW in
  all but exceptional circumstances.

  This document updates RFC 4448.

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/rfc8469.












Bryant, et al.               Standards Track                    [Page 1]

RFC 8469               Ethernet CW Recommendation          November 2018


Copyright Notice

  Copyright (c) 2018 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 ....................................................3
  2. Specification of Requirements ...................................3
  3. Background ......................................................4
  4. Recommendation ..................................................5
  5. Equal-Cost Multipath (ECMP) .....................................5
  6. Mitigations .....................................................6
  7. Operational Considerations ......................................6
  8. Security Considerations .........................................7
  9. IANA Considerations .............................................7
  10. References .....................................................7
     10.1. Normative References ......................................7
     10.2. Informative References ....................................8
  Acknowledgments ....................................................9
  Authors' Addresses .................................................9




















Bryant, et al.               Standards Track                    [Page 2]

RFC 8469               Ethernet CW Recommendation          November 2018


1.  Introduction

  The pseudowire (PW) encapsulation of Ethernet, as defined in
  [RFC4448], specifies that the use of the control word (CW) is
  optional.  It is common for label switching routers (LSRs) to search
  past the end of the label stack to determine whether the payload is
  an IP packet and then, if it is, select the next hop based on the
  so-called "five-tuple" (IP source address, IP destination address,
  protocol/next-header, transport-layer source port, and transport-
  layer destination port).  In the absence of a PW CW, an Ethernet PW
  packet can be misidentified as an IP packet by a label switching
  router (LSR) selecting the ECMP path based on the five-tuple.  This
  may lead to the selection of the wrong ECMP path for the packet,
  leading in turn to the misordering of packets.  Further discussion of
  this topic is published in [RFC4928].

  Flow misordering can also happen in a single-path scenario when
  traffic classification and differential forwarding treatment
  mechanisms are in use.  These errors occur when a forwarder
  incorrectly assumes that the packet is IP and applies a forwarding
  policy based on fields in the PW payload.

  IPv4 and IPv6 packets start with the values 0x4 and 0x6,
  respectively.  Misidentification can arise if an Ethernet PW packet
  without a CW is carrying an Ethernet packet with a destination
  address that starts with either of these values.

  This problem has recently become more serious for a number of
  reasons.  First, the IEEE Registration Authority Committee (RAC) has
  assigned Ethernet MAC addresses that start with 0x4 or 0x6, and
  equipment that uses MAC addresses in these series has been deployed
  in networks.  Second, concerns over privacy have led to the use of
  MAC address randomization, which assigns local MAC addresses randomly
  for privacy.  Random assignment results in addresses starting with
  one of these two values approximately one time in eight.

  The use of the Ethernet PW CW addresses this problem.

  This document RECOMMENDS the use of the Ethernet PW CW in all but
  exceptional circumstances.

2.  Specification of Requirements

  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.



Bryant, et al.               Standards Track                    [Page 3]

RFC 8469               Ethernet CW Recommendation          November 2018


3.  Background

  Ethernet PW encapsulation is specified in [RFC4448].  Of particular
  relevance is Section 4.6, part of which is quoted below for the
  convenience of the reader.  Note that RFC 4448 uses the citation
  [PWE3-CW] to refer to [RFC4385] and the citation [VCCV] to refer to
  the document that was eventually published as [RFC5085].

     The control word defined in this section is based on the Generic
     PW MPLS Control Word as defined in [PWE3-CW].  It provides the
     ability to sequence individual frames on the PW, avoidance of
     equal-cost multiple-path load-balancing (ECMP) [RFC2992], and
     Operations and Management (OAM) mechanisms including VCCV [VCCV].

     [PWE3-CW] states, "If a PW is sensitive to packet misordering and
     is being carried over an MPLS PSN that uses the contents of the
     MPLS payload to select the ECMP path, it MUST employ a mechanism
     which prevents packet misordering."  This is necessary because
     ECMP implementations may examine the first nibble after the MPLS
     label stack to determine whether the labeled packet is IP or not.
     Thus, if the source MAC address of an Ethernet frame carried over
     the PW without a control word present begins with 0x4 or 0x6, it
     could be mistaken for an IPv4 or IPv6 packet.  This could,
     depending on the configuration and topology of the MPLS network,
     lead to a situation where all packets for a given PW do not follow
     the same path.  This may increase out-of-order frames on a given
     PW, or cause OAM packets to follow a different path than actual
     traffic (see Section 4.4.3, "Frame Ordering").

     The features that the control word provides may not be needed for
     a given Ethernet PW.  For example, ECMP may not be present or
     active on a given MPLS network, strict frame sequencing may not be
     required, etc.  If this is the case, the control word provides
     little value and is therefore optional.  Early Ethernet PW
     implementations have been deployed that do not include a control
     word or the ability to process one if present.  To aid in
     backwards compatibility, future implementations MUST be able to
     send and receive frames without the control word present.

  When PWs were first deployed, some equipment of commercial
  significance was unable to process the Ethernet CW.  In addition, at
  that time, it was believed that no Ethernet MAC address had been
  issued by the IEEE Registration Authority Committee (RAC) that
  started with 0x4 or 0x6; thus, it was thought to be safe to deploy
  Ethernet PWs without the CW.






Bryant, et al.               Standards Track                    [Page 4]

RFC 8469               Ethernet CW Recommendation          November 2018


  Since that time, the RAC has issued Ethernet MAC addresses that start
  with 0x4 or with 0x6.  Therefore, the assumption that, in practical
  networks, there would be no confusion between an Ethernet PW packet
  without the CW and an IP packet is no longer correct.

  Possibly through the use of unauthorized Ethernet MAC addresses, this
  assumption has been unsafe for a while, leading some equipment
  vendors to implement more complex, proprietary methods to
  discriminate between Ethernet PW packets and IP packets.  Such
  mechanisms rely on the heuristics of examining the transit packets to
  try to find out the exact payload type of the packet and cannot be
  reliable due to the random nature of the payload carried within such
  packets.

  A posting on the NANOG email list highlighted this problem:

  <https://mailman.nanog.org/pipermail/nanog/2016-December/089395.html>

4.  Recommendation

  The ambiguity between an MPLS payload that is an Ethernet PW and one
  that is an IP packet is resolved when the Ethernet PW CW is used.
  This document updates [RFC4448] to state that both the ingress
  provider edge (PE) and the egress PE SHOULD support the Ethernet PW
  CW and that, if supported, the CW MUST be used.

  Where the application of ECMP to Ethernet PW traffic is required and
  where both the ingress and the egress PEs support Entropy Label
  Indicator / Entropy Label (ELI/EL) [RFC6790] and Flow-Aware Transport
  of Pseudowires (FAT PW) [RFC6391], then either method may be used.
  The use of both methods on the same PW is not normally necessary and
  should be avoided unless circumstances require it.  In the case of
  multi-segment PWs, if ELI/EL is used, then it SHOULD be used on every
  segment of the PW.  The method by which usage of ELI/EL on every
  segment is guaranteed is out of the scope of this document.

5.  Equal-Cost Multipath (ECMP)

  Where the volume of traffic on an Ethernet PW is such that ECMP is
  required, then one of two methods may be used:

  o  Flow-Aware Transport of Pseudowires over an MPLS Packet Switched
     Network, specified in [RFC6391], or

  o  Label Switched Path (LSP) entropy labels, specified in [RFC6790].






Bryant, et al.               Standards Track                    [Page 5]

RFC 8469               Ethernet CW Recommendation          November 2018


  RFC 6391 works by increasing the entropy of the bottom-of-stack
  label.  It requires that both the ingress and egress PEs support this
  feature.  It also requires that sufficient LSRs on the LSP between
  the ingress and egress PE be able to select an
  ECMP path on an MPLS packet with the resultant stack depth.

  RFC 6790 works by including an entropy value in the LSP part of the
  label stack.  This requires that the ingress and egress PEs support
  the insertion and removal of the EL and the ELI and that sufficient
  LSRs on the LSP are able to perform ECMP based on the EL.

  In both cases, there are considerations in getting Operations,
  Administration, and Maintenance (OAM) packets to follow the same path
  as a data packet.  This is described in detail in Section 7 of
  [RFC6391] and Section 6 of [RFC6790].  However, in both cases, the
  situation is improved compared to the ECMP behavior in the case where
  the Ethernet PW CW was not used, since there is currently no known
  method of getting a PW OAM packet to follow the same path as a PW
  data packet subjected to ECMP based on the five-tuple of the IP
  payload.

  The PW label is pushed before the LSP label.  As the ELI/EL labels
  are part of the LSP layer rather than part of the PW layer, they are
  pushed after the PW label has been pushed.

6.  Mitigations

  Where it is not possible to use the Ethernet PW CW, the effects of
  ECMP can be disabled by carrying the PW over a traffic-engineered
  path that does not subject the payload to load balancing (for
  example, RSVP-TE [RFC3209]).  However, such paths may be subjected to
  link-bundle load balancing, and, of course, the single LSP has to
  carry the full PW load.

7.  Operational Considerations

  In some cases, the inclusion of a CW in the PW is determined by
  equipment configuration.  Furthermore, it is possible that the
  default configuration in such cases is to disable use of the CW.
  Care needs to be taken to ensure that software that implements this
  recommendation does not depend on existing configuration settings
  that prevent the use of a CW.  It is recommended that platform
  software emit a rate-limited message indicating that the CW can be
  used but is disabled due to existing configuration.

  Instead of including a payload type in the packet, MPLS relies on the
  control plane to signal the payload type that follows the bottom of
  the label stack.  Some LSRs attempt to deduce the packet type by MPLS



Bryant, et al.               Standards Track                    [Page 6]

RFC 8469               Ethernet CW Recommendation          November 2018


  payload inspection, in some cases looking past the PW CW.  If the
  payload appears to be IP or IP carried in an Ethernet header, they
  perform an ECMP calculation based on what they assume to be the
  five-tuple fields.  However, deduction of the payload type in this
  way is not an exact science, and where a packet that is not IP is
  mistaken for an IP packet, the result can be packets delivered out of
  order.  Misordering of this type can be difficult for an operator to
  diagnose.  When enabling capability that allows information gleaned
  from packet inspection past the PW CW to be used in any ECMP
  calculation, operators should be aware that this may cause Ethernet
  frames to be delivered out of order despite the presence of the CW.

8.  Security Considerations

  This document expresses a preference for one existing and widely
  deployed Ethernet PW encapsulation over another.  These methods have
  identical security considerations, which are discussed in [RFC4448].
  This document introduces no additional security issues.

9.  IANA Considerations

  This document has no IANA actions.

10.  References

10.1.  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>.

  [RFC4385]  Bryant, S., Swallow, G., Martini, L., and D. McPherson,
             "Pseudowire Emulation Edge-to-Edge (PWE3) Control Word for
             Use over an MPLS PSN", RFC 4385, DOI 10.17487/RFC4385,
             February 2006, <https://www.rfc-editor.org/info/rfc4385>.

  [RFC4448]  Martini, L., Ed., Rosen, E., El-Aawar, N., and G. Heron,
             "Encapsulation Methods for Transport of Ethernet over MPLS
             Networks", RFC 4448, DOI 10.17487/RFC4448, April 2006,
             <https://www.rfc-editor.org/info/rfc4448>.

  [RFC4928]  Swallow, G., Bryant, S., and L. Andersson, "Avoiding Equal
             Cost Multipath Treatment in MPLS Networks", BCP 128,
             RFC 4928, DOI 10.17487/RFC4928, June 2007,
             <https://www.rfc-editor.org/info/rfc4928>.





Bryant, et al.               Standards Track                    [Page 7]

RFC 8469               Ethernet CW Recommendation          November 2018


  [RFC6391]  Bryant, S., Ed., Filsfils, C., Drafz, U., Kompella, V.,
             Regan, J., and S. Amante, "Flow-Aware Transport of
             Pseudowires over an MPLS Packet Switched Network",
             RFC 6391, DOI 10.17487/RFC6391, November 2011,
             <https://www.rfc-editor.org/info/rfc6391>.

  [RFC6790]  Kompella, K., Drake, J., Amante, S., Henderickx, W., and
             L. Yong, "The Use of Entropy Labels in MPLS Forwarding",
             RFC 6790, DOI 10.17487/RFC6790, November 2012,
             <https://www.rfc-editor.org/info/rfc6790>.

  [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>.

10.2.  Informative References

  [RFC2992]  Hopps, C., "Analysis of an Equal-Cost Multi-Path
             Algorithm", RFC 2992, DOI 10.17487/RFC2992, November 2000,
             <https://www.rfc-editor.org/info/rfc2992>.

  [RFC3209]  Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan, V.,
             and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP
             Tunnels", RFC 3209, DOI 10.17487/RFC3209, December 2001,
             <https://www.rfc-editor.org/info/rfc3209>.

  [RFC5085]  Nadeau, T., Ed. and C. Pignataro, Ed., "Pseudowire Virtual
             Circuit Connectivity Verification (VCCV): A Control
             Channel for Pseudowires", RFC 5085, DOI 10.17487/RFC5085,
             December 2007, <https://www.rfc-editor.org/info/rfc5085>.





















Bryant, et al.               Standards Track                    [Page 8]

RFC 8469               Ethernet CW Recommendation          November 2018


Acknowledgments

  The authors thank Job Snijders for drawing attention to this problem.
  The authors also thank Pat Thaler for clarifying the matter of local
  MAC address assignment.  We thank Sasha Vainshtein for his valuable
  review comments.

Authors' Addresses

  Stewart Bryant
  Huawei

  Email: [email protected]


  Andrew G. Malis
  Huawei

  Email: [email protected]


  Ignas Bagdonas
  Equinix

  Email: [email protected]>


























Bryant, et al.               Standards Track                    [Page 9]