<?xml version='1.0' encoding='utf-8'?>

<!DOCTYPE rfc SYSTEM "rfc2629-xhtml.ent">

<rfc xmlns:xi="http://www.w3.org/2001/XInclude" number="8654" category="std"
     consensus="true" ipr="trust200902" docName="draft-ietf-idr-bgp-extended-messages-36"
     updates="4271" obsoletes="" submissionType="IETF" xml:lang="en"
     sortRefs="true" symRefs="true" tocInclude="true" version="3">

  <!-- xml2rfc v2v3 conversion 2.27.0 -->
  <front>
    <title abbrev="Extended Message Support for BGP">
           Extended Message Support for BGP</title>

    <seriesInfo name="RFC" value="8654"/>

    <author fullname="Randy Bush" initials="R." surname="Bush">
      <organization>Arrcus &amp; IIJ</organization>
      <address>
        <postal>
          <street>5147 Crystal Springs</street>
          <city>Bainbridge Island</city>
          <region>WA</region>
          <code>98110</code>
          <country>United States of America</country>
        </postal>
        <email>randy@psg.com</email>
      </address>
    </author>
    <author fullname="Keyur Patel" initials="K" surname="Patel">
      <organization>Arrcus, Inc.</organization>
      <address>
        <email>keyur@arrcus.com</email>
      </address>
    </author>
    <author fullname="Dave Ward" initials="D." surname="Ward">
      <organization>Cisco Systems</organization>
      <address>
        <postal>
          <street>170 W. Tasman Drive</street>
          <city>San Jose</city>
          <region>CA</region>
          <code>95134</code>
          <country>United States of America</country>
        </postal>
        <email>dward@cisco.com</email>
      </address>
    </author>

    <date month="October" year="2019" />

    <abstract>
      <t>The BGP specification (RFC 4271) mandates a maximum BGP message size of 4,096
    octets.  As BGP is extended to support new Address Family Identifiers
    (AFIs), Subsequent AFIs (SAFIs), and other
    features, there is a need to extend the maximum message size beyond
    4,096 octets.  This document updates the BGP specification by
    extending the maximum message size from 4,096 octets to 65,535 octets
    for all messages except for OPEN and KEEPALIVE messages.</t>
    </abstract>


  </front>

  <middle>
    <section anchor="introduction" numbered="true" toc="default">
      <name>Introduction</name>
      <t>The BGP specification <xref target="RFC4271" format="default"/> mandates a maximum
    BGP message size of 4,096 octets.  As BGP is extended to support
    new AFIs, SAFIs, and other capabilities (e.g., BGPsec <xref
    target="RFC8205" format="default"/> and BGP - Link
    State (BGP-LS) <xref target="RFC7752" format="default"/>), there is a
    need to extend the maximum message size beyond 4,096 octets. This
    document provides an extension to BGP to extend the message size limit
    from 4,096 octets to 65,535 octets for all messages except for OPEN and
    KEEPALIVE messages.</t>
    
    <section anchor="sec-term" numbered="true" toc="default">
      <name>Requirements Language</name>

        <t>
    The key words "<bcp14>MUST</bcp14>", "<bcp14>MUST NOT</bcp14>", "<bcp14>REQUIRED</bcp14>", "<bcp14>SHALL</bcp14>", "<bcp14>SHALL
    NOT</bcp14>", "<bcp14>SHOULD</bcp14>", "<bcp14>SHOULD NOT</bcp14>", "<bcp14>RECOMMENDED</bcp14>", "<bcp14>NOT RECOMMENDED</bcp14>",
    "<bcp14>MAY</bcp14>", and "<bcp14>OPTIONAL</bcp14>" in this document are to be interpreted as
    described in BCP&nbsp;14 <xref target="RFC2119"/> <xref target="RFC8174"/> 
    when, and only when, they appear in all capitals, as shown here.
        </t>

    </section>


    </section>
    <section anchor="extmsg" numbered="true" toc="default">
      <name>BGP Extended Message</name>
      <t>A BGP message over 4,096 octets in length is a BGP Extended
    Message.</t>
      <t>BGP Extended Messages have a maximum message size of 65,535
    octets.  The smallest message that may be sent is a BGP
    KEEPALIVE, which consists of 19 octets.</t>
    </section>
    <section numbered="true" toc="default">
      <name>BGP Extended Message Capability</name>
      <t>The BGP Extended Message Capability is a new BGP capability <xref
      target="RFC5492" format="default"/> defined with Capability Code 6 and
      Capability Length 0.</t>
      <t>To advertise the BGP Extended Message Capability to a peer, a BGP
      speaker uses BGP Capabilities Advertisement <xref target="RFC5492"
      format="default"/>.  By advertising the BGP Extended Message Capability
      to a peer, a BGP speaker conveys that it is able to receive and properly
      handle BGP Extended Messages (see <xref target="opns"
      format="default"/>).</t>
      <t>Peers that wish to use the BGP Extended Message Capability <bcp14>MUST</bcp14>
    support error handling for BGP UPDATE messages per <xref target="RFC7606"
    format="default"/>.</t>

    </section>
    <section anchor="opns" numbered="true" toc="default">
      <name>Operation</name>

      <t>The BGP Extended Message Capability applies to all messages except
    for OPEN and KEEPALIVE messages.  These exceptions 
    reduce the complexity of providing backward compatibility.</t>
      <t>A BGP speaker that is capable of receiving BGP
    Extended Messages <bcp14>SHOULD</bcp14> advertise the BGP Extended Message
    Capability to its peers using BGP Capabilities Advertisement <xref
    target="RFC5492" format="default"/>.  A BGP speaker <bcp14>MAY</bcp14>
    send BGP Extended Messages to a
    peer only if the BGP Extended Message Capability was received from that
    peer.</t>

      <t>An implementation that advertises the BGP Extended Message
    Capability <bcp14>MUST</bcp14> be capable of receiving a message with a length up
    to and including 65,535 octets.</t>

      <t>Applications generating information that might be encapsulated
    within BGP messages <bcp14>MUST</bcp14> limit the size of their payload to take the
    maximum message size into account.</t>


      <t>If a BGP message with a length greater than 4,096 octets is
    received by a BGP listener who has not advertised the BGP Extended
    Message Capability, the listener will generate a NOTIFICATION with
    the Error Subcode set to Bad Message Length (<xref target="RFC4271"
    sectionFormat="comma" section="6.1"/>).</t>

      <t>A BGP UPDATE will (if allowed by policy, best path, etc.) typically
    propagate throughout the BGP-speaking Internet and hence to BGP
    speakers that may not support BGP Extended Messages.  Therefore, an
    announcement in a BGP Extended Message where the size of the attribute
    set plus the NLRI is larger than 4,096 octets may cause lack of
    reachability.</t>
      <t>A BGP speaker that has advertised the BGP Extended Message
    Capability to its peers may receive an UPDATE from one of its peers
    that produces an ongoing announcement that is larger than 4,096
    octets.  When propagating that UPDATE onward to a neighbor that has
    not advertised the BGP Extended Message Capability, the speaker
    <bcp14>SHOULD</bcp14> try to reduce the outgoing message size by removing
    attributes eligible under the "attribute discard" approach of <xref target="RFC7606" format="default"/>.  If the message is still too big, then it must
    not be sent to the neighbor (<xref target="RFC4271" sectionFormat="comma"
    section="9.2"/>).

    Additionally, if the NLRI was previously advertised to that peer, it
    must be withdrawn from service (<xref target="RFC4271"
    sectionFormat="comma" section="9.1.3"/>).
    </t>


      <t>If an Autonomous System (AS) has multiple internal BGP speakers
    and also has multiple external BGP neighbors, care must be taken to ensure a consistent view within the
    AS in order to present a consistent
    external view. In the context of BGP Extended Messages, a
    consistent view can only be guaranteed if all the Internal BGP (iBGP) speakers
    advertise the BGP Extended Message Capability.  If that is not the
    case, then the operator should consider whether or not the BGP Extended
    Message Capability should be advertised to external peers.
    </t>


      <t>During the incremental deployment of BGP Extended Messages and
    use of the "attribute discard" approach of <xref target="RFC7606"
    format="default"/> in an iBGP mesh or with
    External BGP (eBGP) peers, the
    operator should monitor any routes dropped and any discarded
    attributes.</t>
    </section>
    <section anchor="error" numbered="true" toc="default">
      <name>Error Handling</name>

      <t>A BGP speaker that has the ability to use BGP Extended Messages but
    has not advertised the BGP Extended Message Capability, presumably
    due to configuration, <bcp14>MUST NOT</bcp14> accept a BGP Extended Message.  A
    speaker <bcp14>MUST NOT</bcp14> implement a more liberal policy accepting BGP
    Extended Messages.</t>

      <t>A BGP speaker that does not advertise the BGP Extended Message
    Capability might also genuinely not support BGP Extended Messages.  Such
    a speaker will follow the error-handling procedures of <xref
    target="RFC4271" format="default"/> if it receives a BGP Extended Message.  Similarly,
    any speaker that treats an improper BGP Extended Message as a fatal
    error <bcp14>MUST</bcp14> follow the error-handling procedures of <xref
    target="RFC4271" format="default"/>.
      </t>

      <t>Error handling for UPDATE messages, as specified in 
    <xref target="RFC4271" sectionFormat="of" section="6.3"/>, is unchanged.  However, if a
    NOTIFICATION is to be sent to a BGP speaker that has not advertised
    the BGP Extended Message Capability, the size of the message <bcp14>MUST
    NOT</bcp14> exceed 4,096 octets.</t>
      <t>It is <bcp14>RECOMMENDED</bcp14> that BGP protocol developers and implementers
    are conservative in their application and use of BGP Extended Messages.
    Future protocol specifications <bcp14>MUST</bcp14> describe how to handle peers
    that can only accommodate 4,096 octet messages.</t>
    </section>

    <section anchor="rfc4171" numbered="true" toc="default">
      <name>Changes to RFC 4271</name>
      <t><xref target="RFC4271" format="default"/> states "The value of the Length field
    <bcp14>MUST</bcp14> always be at least 19 and no greater than 4096."  This document
    changes the latter number to 65,535 for all messages except for OPEN and
    KEEPALIVE messages.</t>


      <t><xref target="RFC4271" sectionFormat="of" section="6.1"/> specifies
      raising an error if the length of a message is over 4,096 octets.  For
      all messages except for OPEN and KEEPALIVE messages, if the receiver has advertised the
      BGP Extended Message Capability, this document raises that limit to
      65,535.</t>
    </section>
    <section anchor="IANA" numbered="true" toc="default">
      <name>IANA Considerations</name>
      <t>IANA has made the following allocation in the "Capability Codes"
      registry:</t>

<table anchor="ianaregistry" align="left">
<name>Addition to "Capability Codes" Registry</name>
  <thead>
    <tr>
      <th>Value</th>
      <th>Description</th>
      <th>Reference</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>6</td>
      <td>BGP Extended Message</td>
      <td>RFC 8654</td>
    </tr>
  </tbody>
</table>

    </section>
    <section anchor="Security" numbered="true" toc="default">
      <name>Security Considerations</name>
      <t>This extension to BGP does not change BGP's underlying security
    issues <xref target="RFC4272" format="default"/>.</t>
      <t>Due to increased memory requirements for buffering, there may be
    increased exposure to resource exhaustion, intentional or
    unintentional.</t>

      <t>If a remote speaker is able to craft a large BGP Extended Message
    to send on a path where one or more peers do not support BGP
    Extended Messages, peers that support BGP Extended Messages may:
    </t>

<ul spacing="normal">

    <li>act to reduce the outgoing message (see <xref target="opns"
    format="default"/>) and, in doing so, cause an attack by discarding
    attributes one or more of its peers may be expecting.  The attributes eligible under the
    "attribute discard" approach must have no effect on route selection or
    installation <xref target="RFC7606" format="default"/>.</li>


      <li>act to reduce the outgoing message (see <xref target="opns"
      format="default"/>) and, in doing so, allow a downgrade attack.  This
      would only affect the attacker's message, where 'downgrade' has
      questionable meaning.</li>

      <li>incur resource load (processing, message resizing, etc.)
    when reformatting the large messages.</li>
</ul>

    
    </section>
   
  </middle>
  <back>

    <references>
      <name>References</name>

      <references>
        <name>Normative References</name>

<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.2119.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.4271.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5492.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.7606.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8174.xml"/>

      </references>

      <references>
        <name>Informative References</name>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.4272.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.7752.xml"/>
<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8205.xml"/>
      </references>
    </references>
    <section anchor="acknowledgements" numbered="false" toc="default">
     <name>Acknowledgments</name>

     <t>The authors thank Alvaro Retana for an amazing review; Enke Chen,
     Susan Hares, John Scudder, John Levine, and Job Snijders for their
     input; and Oliver Borchert and Kyehwan Lee for their implementations
     and testing.</t>
    </section>

  </back>
</rfc>
