<?xml version='1.0' encoding='UTF-8'?>
<rfc version="3" category="info" docName="draft-jags-6man-icmp-error-srv6-evpn-00" ipr="trust200902" submissionType="IETF" consensus="false" tocInclude="true" tocDepth="4" symRefs="true" sortRefs="true">
  <front>
    <title abbrev="ICMP Errors for SRv6 EVPN-VPWS">ICMP Error Handling for EVPN-VPWS in SRv6 Networks</title>
    <seriesInfo name="Internet-Draft" value="draft-jags-6man-icmp-error-srv6-evpn-00"/>
    <author fullname="Jaganbabu Rajamanickam" initials="J." surname="Rajamanickam">
      <organization>Cisco Systems, Inc.</organization>
      <address>
        <email>jrajaman@cisco.com</email>
      </address>
    </author>
    <author fullname="Madhan Sankaranarayanan" initials="M." surname="Sankaranarayanan">
      <organization>Cisco Systems, Inc.</organization>
      <address>
        <email>madsanka@cisco.com</email>
      </address>
    </author>
    <author fullname="Zafar Ali" initials="Z." surname="Ali">
      <organization>Cisco Systems, Inc.</organization>
      <address>
        <email>zali@cisco.com</email>
      </address>
    </author>
    <date year="2026" month="November" day="16"/>
    <area>Internet</area>
    <workgroup>6man Working Group</workgroup>
    <keyword>EVPN</keyword>
    <keyword>VPWS</keyword>
    <keyword>SRv6</keyword>
    <keyword>Path MTU Discovery</keyword>
    <keyword>ICMP</keyword>
    <abstract>
      <t>This document describes procedures for handling ICMP error messages for Ethernet VPN Virtual Private Wire Service (EVPN-VPWS) in Segment Routing over IPv6 (SRv6) networks. It describes an ingress-PE approach that associates a received ICMPv6 error with the EVPN service, constructs eligible customer ICMP feedback, and uses local End.DX2 disposition to apply the appropriate Ethernet and VLAN treatment. The discussion covers IPv4 and IPv6 customer traffic, the handling of the different ICMP error classes, and the additional procedures applicable to Packet Too Big messages and Path MTU Discovery.</t>
    </abstract>
  </front>
  <middle>
    <section anchor="intro">
      <name>Introduction</name>
      <t>Troubleshooting and error reporting in an Ethernet VPN require the provider network to relate a transport-network event to the customer service that carried the invoking packet. Provider core nodes do not generally have the service information needed to return an error directly to the customer source. This is also the case when an EVPN-VPWS service uses an SRv6 transport network.</t>
      <t>EVPN-VPWS provides point-to-point Ethernet connectivity between customer sites <xref target="RFC8214"/>. With SRv6-based BGP services, the ingress PE encapsulates a customer Ethernet frame for delivery to a remote End.DX2 service SID <xref target="RFC9252"/>. An error generated within the SRv6 domain is normally addressed to the ingress PE, using the source address of the outer IPv6 header.</t>
      <t>The ingress PE needs to identify the affected EVPN service, interpret the received error, and obtain the attachment circuit through which eligible customer feedback can be returned. Because the service carries Ethernet frames, that feedback also needs the appropriate MAC addresses and VLAN treatment.</t>
      <t>This Informational document describes an ingress-PE approach for handling these cases using existing IPv6, ICMP, and SRv6 mechanisms. It discusses the operation, characteristics, and limitations of the approach and illustrates IPv6 and IPv4 customer traffic. It does not define a new ICMP message format, SRv6 endpoint behavior, or BGP extension.</t>
      <t>The related approaches for ICMP error handling in SRv6 VPNs are described in <xref target="I-D.alivar-spring-icmp-error-srv6-vpn"/>. The discussion here addresses the Ethernet framing, local cross-connect disposition, and VLAN treatment of EVPN-VPWS. Packet Too Big processing and Path MTU Discovery are described as particular applications of the general ICMP error-handling procedure.</t>
      <section anchor="scope">
        <name>Scope and Applicability</name>
        <t>The discussion covers Destination Unreachable, Packet Too Big, Time Exceeded, and Parameter Problem messages received for SRv6 packets carrying point-to-point EVPN-VPWS Ethernet traffic using End.DX2. It assumes a managed SRv6 domain in which the operator permits the ingress PE to inspect customer headers and generate eligible customer feedback. Some received errors are handled locally as provider diagnostics without generating a customer packet.<xref target="RFC4443"/>, as received for SRv6 packets carrying point-to-point EVPN-VPWS Ethernet traffic using End.DX2. The procedures assume a managed SRv6 domain and administrative authorization for the PE to inspect customer headers and generate eligible customer errors. Error classification covers provider processing as well as customer notification; some errors are handled locally without a customer packet.</t>
        <t>Customer ICMPv4 and ICMPv6 packets carried inside the VPWS are ordinary service payload and continue to use normal Ethernet forwarding. Informational ICMP messages, including Echo Request and Echo Reply, are not treated as received error reports by this procedure. Locally originated diagnostic probes and their responses are handled in their own context.</t>
        <t>Customer IPv4 TTL and IPv6 Hop Limit are not decremented by transparent VPWS bridging. The outer SRv6 Hop Limit is independent. Consequently, reporting an outer Hop Limit failure does not imply that the customer TTL expired, and ordinary customer traceroute is not assumed to expose the SRv6 core.</t>
        <t>Multipoint EVPN, replication, and other Layer 2 endpoint behaviors require additional service-association and return-path considerations. The procedures described here apply to the point-to-point End.DX2 case.<xref target="I-D.alivar-spring-icmp-error-srv6-vpn"/>. This document addresses the Ethernet framing, local cross-connect disposition, VLAN treatment, and error-specific processing needed for VPWS.</t>
      </section>
    </section>
    <section anchor="terms">
      <name>Terminology</name>
      <section anchor="terms-used">
        <name>Terms Used in This Document</name>
        <t>The terms EVPN, VPWS, attachment circuit, and Ethernet Auto-Discovery (A-D) route follow <xref target="RFC8214"/>. SRv6 service SID and End.DX2 follow <xref target="RFC9252"/> and <xref target="RFC8986"/>.</t>
        <dl spacing="normal">
          <dt>Error disposition</dt>
          <dd>The decision to consume a report for local diagnostics, notify a local probe originator, update permitted error-specific state, or generate a customer error.</dd>
          <dt>AC</dt>
          <dd>Attachment Circuit connecting a PE to the customer Ethernet service.</dd>
          <dt>Remote service SID</dt>
          <dd>The complete End.DX2 SID advertised by the remote PE and installed for the selected VPWS forwarding path.</dd>
          <dt>Local service SID</dt>
          <dd>The local End.DX2 SID whose disposition sends Ethernet frames to the local AC of the same VPWS service.</dd>
          <dt>Service path context</dt>
          <dd>The association among the VPWS service, remote service SID, selected SR policy or ordinary transport path, forwarding generation, and relevant encapsulation and load-balancing parameters.</dd>
          <dt>Invoking packet</dt>
          <dd>The packet that caused an ICMP error. A copy of its initial bytes is carried in the ICMP error payload; that copy may be truncated.</dd>
          <dt>FIB backwalk</dt>
          <dd>A local implementation mechanism that follows forwarding-object associations from a remote SID or path to the VPWS cross-connect, local AC, and local service SID.</dd>
          <dt>PTB validation probe</dt>
          <dd>A locally originated probe used to corroborate a reported size limitation on a selected SRv6 service path. A successful correlation provides path evidence, rather than cryptographic authentication.</dd>
        </dl>
      </section>
      <section>
        <name>Requirements Language</name>
        <t>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 <xref target="RFC2119"/>
          <xref target="RFC8174"/> when, and only when, they appear in all capitals, as shown here.</t>
        <t>The requirement words describe safeguards and conditions of the procedures discussed here, including requirements from the referenced RFCs. Their use does not change the Informational status of this document or establish a new general conformance requirement for EVPN or IPv6 implementations.</t>
      </section>
    </section>
    <section anchor="scenario">
      <name>ICMP Error Handling for EVPN-VPWS in SRv6 Networks</name>
      <section anchor="network">
        <name>Network Scenario and Reference Topology</name>
        <figure anchor="topology">
          <name>EVPN-VPWS over an SRv6 Domain</name>
          <artwork type="ascii-art"><![CDATA[
                  |<--------- SRv6 Domain --------->|
                  |                                 |
     +-----+   +-----+    +----+    +----+       +-----+   +-----+
     | CE1 |---| PE1 |----| P1 |----| P2 |-------| PE2 |---| CE2 |
     +-----+   +-----+    +----+    +----+       +-----+   +-----+
                 AC1                               AC2
                  |<---------- EVPN-VPWS ---------->|

        Customer frame: Ethernet / optional VLAN tags / IPv4 or IPv6
        Core packet:    IPv6 / optional SRH / Ethernet / IP / payload
]]></artwork>
        </figure>
        <t>PE1 encapsulates frames received from CE1 and sends them to PE2 using a remote End.DX2 SID. PE2 decapsulates each packet and delivers the Ethernet frame on AC2. In the reverse direction, PE1 has a local End.DX2 SID associated with AC1.</t>
        <t>When P2 detects a forwarding or IPv6 processing failure, it generates the applicable ICMPv6 error, subject to normal error-generation rules, and addresses it to PE1. An example is outer Hop Limit expiration; another is a packet exceeding the outgoing link MTU. The received error's outer IPv6 destination identifies PE1, but generally does not identify the VPWS service. The ingress PE examines the quoted invoking SRv6 packet to recover the remote service SID and the service path context.</t>
        <t>The common issues are validation of the received error, association with the correct VPWS and path, selection of an error-appropriate disposition, and construction of an eligible customer error frame with the correct access VLAN treatment. MTU validation, size conversion, and PMTU cache changes apply specifically to PTB processing.</t>
        <t>In the examples, PE1::1 and P2::1 denote provider IPv6 addresses; PE1:DX2:: and PE2:DX2:: denote local and remote service SIDs; A6 and B6 denote customer IPv6 addresses; A4 and B4 denote customer IPv4 addresses; and M1 and M2 denote customer MAC addresses. These labels are symbolic placeholders and are not literal address syntax. E6 and E4 denote configured service error-source addresses on PE1.</t>
      </section>
      <section anchor="current-operation">
        <name>Current Operation and Limitations</name>
        <t>Consider an Ethernet frame received from CE1 and encapsulated by PE1 for the remote End.DX2 SID on PE2. If the outer Hop Limit expires at P2, ordinary IPv6 processing generates an ICMPv6 Time Exceeded message addressed to PE1. The message quotes the invoking SRv6 packet, including as much of the carried customer frame as is available within the ICMP size limit <xref target="RFC4443"/>.</t>
        <artwork type="ascii-art"><![CDATA[
Invoking packet received at P2:
  IPv6(SA=PE1::1, DA=PE2:DX2::, HL=1, NH=143)
    Ethernet(SA=M1, DA=M2, tags=service representation)
      IPv6(SA=A6, DA=B6, HL=64, NH=UDP) / customer data

ICMP error returned by P2:
  IPv6(SA=P2::1, DA=PE1::1, NH=58)
    ICMPv6(Type=3, Code=0)
      [copy of the invoking SRv6 packet and customer frame]
]]></artwork>
        <t>The destination of the received error identifies the ingress PE. It does not, by itself, identify a local attachment circuit, a local service SID, or the access VLAN rewrite. These associations are present in the installed EVPN forwarding state. Without service-associated processing, PE1 can record the provider error but does not have a general way to construct and deliver eligible customer feedback.</t>
        <t>The same service-association issue occurs for a received Destination Unreachable or Packet Too Big message and for supported Parameter Problem cases. The interpretation and permitted customer response depend on the specific error. In the example, the customer Hop Limit remains 64; the outer tunnel failure is not customer Hop Limit expiration.</t>
      </section>
    </section>
    <section anchor="procedure">
      <name>Improving ICMP Error Handling for EVPN-VPWS in SRv6 Networks</name>
      <section anchor="local-method">
        <name>Local ICMP Error Processing (Changing Ingress-PE Operation)</name>
        <section anchor="overview">
          <name>Overview</name>
          <t>In the approach described here, the P node performs its existing IPv6 and ICMP processing. The ingress PE relates the received error to the EVPN service using the remote service SID in the quoted invoking packet and locally installed forwarding associations. It then constructs eligible customer feedback and delivers the resulting Ethernet frame through the local End.DX2 disposition associated with the source attachment circuit.</t>
          <t>The node functions are as follows:</t>
          <ol spacing="normal">
            <li>Ingress PE: PE1 encapsulates a frame received from CE1 using the remote End.DX2 SID and the selected SRv6 transport path. The outer source is a local PE1 IPv6 address.</li>
            <li>P node: when an applicable forwarding or IPv6 processing error occurs, P2 sends the corresponding ICMPv6 error to the outer source address according to the existing IPv6 and ICMP procedures.</li>
            <li>Ingress PE: PE1 validates the report, extracts the remote service SID from the quoted invoking packet, and resolves the associated EVPN-VPWS connection, local attachment circuit, and local End.DX2 SID. The particular ICMP error determines the local action and any eligible customer notification.</li>
            <li>Local service disposition: PE1 creates an Ethernet frame containing the eligible ICMPv4 or ICMPv6 response and injects it toward its local End.DX2 SID. The local disposition applies the attachment circuit's Ethernet and VLAN treatment and sends the frame toward CE1.</li>
          </ol>
          <t>The regenerated customer error is delivered locally at PE1. The egress PE does not take part in returning that error. The approach therefore avoids requiring an error packet to traverse the remaining forward service path before it can return to the source.</t>
        </section>
        <section anchor="details">
          <name>Details of ICMP Error Handling</name>
          <t>IPv6 header-chain parsing follows <xref target="RFC8200"/> and the applicable SRv6 endpoint processing. Packet quotations can contain mutable fields updated in transit; correlation distinguishes those permitted changes from inconsistent or fabricated information.</t>
          <section anchor="associate">
            <name>Validate and Associate the Received Error</name>
            <t>The ingress PE performs these steps for every received ICMPv6 error associated with its SRv6 traffic:</t>
            <ol spacing="normal">
              <li>Validate the ICMPv6 checksum, message length, type and code, destination, and quoted-packet bounds. Correlate the quoted outer source and encapsulation with traffic originated by this PE, using recent transmission or forwarding context where available. Apply provider-domain filtering and bounded control-plane processing. A service SID match alone does not authenticate an error. The received PTB Code field is ignored as required by RFC 4443; generated PTBs use Code 0.</li>
              <li>Determine whether the quoted packet belongs to a locally originated probe, provider control traffic, or customer data. Deliver probe-related errors to the corresponding local process; do not generate customer feedback from a validation probe or recursively launch probes for its errors.</li>
              <li>For a customer-data report, parse the quoted invoking SRv6 packet and recover the complete remote service SID. With an ordinary uncompressed SRH, the final segment is Segment List[0] <xref target="RFC8754"/>. Without an SRH, use the quoted IPv6 destination only when it identifies the final service destination. The outer header of the received ICMPv6 message is not the location of the quoted service SID.</li>
              <li>Match the recovered remote End.DX2 SID to active imported service state and resolve the VPWS cross-connect, selected forwarding path, local attachment circuit, and local End.DX2 SID. A forwarding-identifier and L2 forwarding-entry backwalk is one implementation. A locator-only match is insufficient, and the remote SID alone may not distinguish several policies using that SID.</li>
              <li>Classify the error and select its disposition using <xref target="classification"/>. Check any class-specific fields before state changes or customer feedback. For PTB, reject reported outer MTUs below 1280, check packet-size plausibility, and apply <xref target="pmtu"/>. Other error classes MUST NOT change PMTU state.</li>
              <li>For an eligible customer notification, parse sufficient quoted Ethernet, VLAN, and customer IP information and apply the customer error-generation rules in <xref target="eligibility"/>. Construct the return frame and deliver it through local disposition. If the service or return attachment circuit is ambiguous, the PE MUST NOT infer a destination for customer feedback. If sufficient customer bytes are unavailable, it MUST NOT fabricate a quote.</li>
            </ol>
            <t>Compressed SID reconstruction uses the applicable SID structure and retained headend context <xref target="RFC9800"/>. Nested headers and repair encapsulation can require additional context as described in <xref target="complex"/>. Malformed, uncorrelated, unsupported, or incomplete reports receive bounded local handling; an unknown error type is passed to an identifiable local originator as required by <xref target="RFC4443"/>, without inventing a customer mapping.</t>
          </section>
          <section anchor="classification">
            <name>Error Classification and Customer Notification</name>
            <t>The table summarizes whether a received provider error results in a newly generated ICMP notification toward the CE. The provider ICMP packet is processed by the ingress PE; it is not forwarded unchanged to the CE.</t>
            <table anchor="error-matrix">
              <name>CE Notification Summary</name>
              <thead>
                <tr>
                  <th>Provider error or event</th>
                  <th>Notify CE?</th>
                  <th>Response or local action</th>
                </tr>
              </thead>
              <tbody>
                <tr>
                  <td>Packet Too Big</td>
                  <td>Yes*</td>
                  <td>Customer MTU feedback.</td>
                </tr>
                <tr>
                  <td>Destination Unreachable</td>
                  <td>Yes*</td>
                  <td>Customer Destination Unreachable.</td>
                </tr>
                <tr>
                  <td>Outer Hop Limit exceeded</td>
                  <td>Yes*</td>
                  <td>Customer Destination Unreachable.</td>
                </tr>
                <tr>
                  <td>Tunnel encapsulation-limit exhaustion</td>
                  <td>Yes*</td>
                  <td>Customer Destination Unreachable.</td>
                </tr>
                <tr>
                  <td>Other provider Parameter Problem</td>
                  <td>No</td>
                  <td>Local provider diagnostics.</td>
                </tr>
                <tr>
                  <td>Provider reassembly timeout</td>
                  <td>No</td>
                  <td>Local provider diagnostics.</td>
                </tr>
                <tr>
                  <td>Error for a PE-generated probe</td>
                  <td>No</td>
                  <td>Local probe processing.</td>
                </tr>
              </tbody>
            </table>
            <t>Yes* means that the report is valid, the service and customer source are identified, customer notification is enabled, and ICMP generation is permitted. Errors for PE-generated probes remain local even when their type appears in a Yes* row. Mandatory suppression rules in <xref target="eligibility"/> also override the Yes* rows. No means that this approach handles the event at the PE without generating a CE-side error.</t>
            <t>Customer Destination Unreachable uses ICMPv6 Type 1, Code 3, or ICMPv4 Type 3, Code 1, following <xref target="RFC2473"/>, Sections 8.1 through 8.3. MTU feedback uses ICMPv6 PTB or ICMPv4 fragmentation needed for DF-set traffic, subject to <xref target="pmtu"/>. The PE retains the original provider type and code for diagnostics and does not blindly copy them into the customer response.</t>
            <section anchor="time-exceeded">
              <name>Outer Hop Limit and Reassembly Errors</name>
              <t>An outer Hop Limit error reports failure of the SRv6 transport. The customer TTL or Hop Limit is unchanged. The common return-frame procedure therefore carries the generic tunnel-unreachable notification rather than customer Time Exceeded. This distinguishes a failed provider tunnel from expiration of a customer IP forwarding hop.</t>
              <t>Code 1 indicates reassembly timeout for the invoking outer packet. It does not establish that the customer was reassembling fragments. The PE records that provider event locally; it MUST NOT translate it automatically to customer fragment-reassembly timeout.</t>
            </section>
            <section anchor="parameter-problem">
              <name>Parameter Problem and Pointer Handling</name>
              <t>A Parameter Problem pointer is interpreted in the address space of the quoted invoking IPv6 packet. Removing SRv6 and Ethernet headers changes that address space. The PE MUST NOT copy the pointer, or subtract header lengths to create a customer pointer, without a separately defined and applicable customer-header mapping.</t>
              <t>A recognized tunnel encapsulation-limit exhaustion uses the generic tunnel-error notification described above. Other errors in provider IPv6 headers or SRHs remain provider diagnostics in this revision. A Code 4 SR Upper-layer Header Error received for a diagnostic probe is delivered to its local probe context <xref target="RFC8986"/>. Neither this error nor a header-length-related code is a PTB report.</t>
            </section>
          </section>
          <section anchor="eligibility">
            <name>Customer Error Eligibility and Suppression</name>
            <t>The decision to generate a customer error is applied to the original quoted customer IP packet. The received underlay ICMPv6 wrapper does not, by itself, trigger the rule prohibiting errors in response to customer ICMP errors. Conversely, if the quoted customer packet is itself an ICMP error or another packet for which generation is forbidden, the PE MUST suppress customer feedback.</t>
            <t>IPv6 processing follows <xref target="RFC4443"/>, including invalid-source checks, the restrictions for ICMP errors and Redirects, multicast-related exceptions, quote construction, and rate limits. IPv4 processing follows <xref target="RFC792"/> and <xref target="RFC1812"/>, including restrictions for ICMP errors, non-initial fragments, broadcasts, multicast, and invalid sources. Informational customer Echo Requests are not suppressed merely because they use ICMP.</t>
            <t>The examples cover eligible unicast customer traffic. If a group-addressed packet is eligible under an ICMP exception, a valid configured unicast source MAC and applicable source IP are needed; blindly reversing a multicast destination MAC is invalid. The generated IPv6 error, including its customer IPv6 header, is limited to 1280 octets or a smaller applicable return constraint, and quotes as much of the customer packet as allowed. IPv4 errors include at least the customer IPv4 header and the required initial payload bytes; insufficient quotes are not invented.</t>
          </section>
          <section anchor="return-frame">
            <name>Construct the Customer Error Frame</name>
            <t>For a quoted unicast customer IP packet eligible for ICMP feedback, the ingress PE creates a new Ethernet frame as follows:</t>
            <ol spacing="normal">
              <li>Set the return Ethernet destination to the source MAC address of the invoking customer frame. In the simple point-to-point unicast case, set the return source MAC to the original destination MAC. This reverses the frame's MAC addresses. A deployment MAY instead use a configured service error-source MAC when required by its service policy.</li>
              <li>Preserve or reconstruct VLAN tags in the representation expected by the local service disposition. A tag removed at ingress can be absent from the quoted frame; the service state supplies the information needed for the local AC. Do not blindly copy the remote AC VLAN or perform the local disposition rewrite twice.</li>
              <li>Create a customer IPv4 or IPv6 packet addressed to the original customer IP source. Select the IP source as described in <xref target="error-source"/>. Set the Ethernet EtherType for the generated IP version and create the corresponding ICMP error.</li>
              <li>Place the original customer IP packet, to the extent required and available, in the ICMP quote. Provider SRv6 and customer Ethernet headers are not part of this IP quote. Do not reverse addresses or otherwise fabricate bytes inside the quoted customer IP packet.</li>
              <li>Set the ICMP type and code according to the error disposition. Only a PTB-related response carries an adjusted MTU, computed under the address-family and minimum-size conditions in <xref target="pmtu"/>. Initialize unused fields to zero. Recalculate all affected lengths and checksums, initialize a new IP TTL or Hop Limit, and size the error for the return AC. Apply <xref target="eligibility"/> before generation.</li>
            </ol>
            <t>The PE MUST NOT reverse a multicast or broadcast destination MAC into an invalid return source MAC. Group-addressed traffic needs the applicable ICMP eligibility rules and a valid configured source identity; it is outside the unicast examples in this document. Customer traffic that cannot be inspected, such as encrypted Layer 2 payloads, cannot use this construction without additional service knowledge.</t>
          </section>
          <section anchor="error-source">
            <name>Customer ICMP Source Address</name>
            <t>The IP source of a regenerated error identifies its origin to the customer. The base procedure uses a valid configured service error-source IPv4 or IPv6 address owned by PE1 and suitable for the customer context. Source selection follows the applicable ICMP rules <xref target="RFC792"/>
              <xref target="RFC4443"/>. A service with no suitable source address requires an explicitly defined proxy profile before it can enable customer error generation.</t>
            <t>A transparent address-swap profile is a proposed extension for consideration: set the generated IP source to the original customer destination and the generated IP destination to the original customer source. This preserves the address-swap construction for an unnumbered Layer 2 service, but it causes PE1 to use an address it does not own. It is not ordinary ICMP source selection and is not enabled by the base procedure. The ownership, authorization, filtering, and interoperability requirements for that profile remain an open issue in <xref target="issues"/>. In all profiles, the embedded quote retains the original addresses.</t>
          </section>
          <section anchor="reinjection">
            <name>Local End.DX2 Disposition and VLAN Translation</name>
            <t>The ingress PE can deliver the regenerated frame by building a local SRv6 packet whose outer IPv6 source is a local PE1 address and whose destination is the local End.DX2 SID associated with AC1. The upper-layer protocol is Ethernet (Next Header 143), independent of the customer IP version. If an SRH is present, the local End.DX2 is the final segment and Segments Left is zero. The inner sequence is Ethernet, customer IP, and ICMP.</t>
            <t>The PE injects this packet directly into trusted local service disposition. It MUST NOT send it across the SRv6 core, allow it to enter normal customer ingress learning, or recursively apply the error-generation procedure. The local End.DX2 processing selects the AC and applies the access Ethernet and VLAN rewrite. End.DX2 permits Ethernet header rewriting as part of disposition <xref target="RFC8986"/>.</t>
            <t>For asymmetric VLAN connectivity, the local service disposition restores the VLAN expected by CE1. For example, if CE1 uses VLAN 100 and CE2 uses VLAN 200, the returned frame uses the configured AC1 rewrite to VLAN 100. The internal frame and tag representation MUST be consistent with that rewrite. An implementation MAY deliver the frame through an equivalent local disposition interface without constructing an actual outer IPv6 header; the resulting access frame and service isolation MUST be the same.</t>
          </section>
        </section>
        <section anchor="optional-capabilities">
          <name>Optional Capabilities</name>
          <t>Local policy can control which supported error classes produce customer feedback, the rate at which errors are generated, and the source identities used in that feedback. Provider-specific diagnostic information can be retained locally while the customer receives a generic tunnel-failure notification with the semantics described in <xref target="classification"/>.</t>
          <t>For Packet Too Big reports, the ingress PE can use the additional acceptance checks and optional validation probes described in <xref target="pmtu"/>. These capabilities are specific to PTB processing and are not prerequisites for handling every other ICMP error class.</t>
        </section>
        <section anchor="characteristics">
          <name>Characteristics of the Approach</name>
          <t>The local ingress-PE approach has the following characteristics:</t>
          <ul spacing="normal">
            <li>Service-associated processing is performed on the ingress PE. Transit P nodes use their existing IPv6 and ICMP behavior.</li>
            <li>The remote End.DX2 SID and installed service state identify the affected VPWS connection and the local return attachment circuit; a customer VRF lookup is not assumed.</li>
            <li>The egress PE is not involved in forwarding the regenerated error. A failure farther along the forward service path need not prevent local feedback, provided that the original error can reach PE1 and the return attachment circuit remains usable.</li>
            <li>Local End.DX2 disposition reuses the Ethernet and VLAN treatment configured for that attachment circuit, including asymmetric access VLAN configurations.</li>
            <li>The same service association and return-frame construction support the applicable IPv4 and IPv6 customer error responses. The meaning of the received error determines whether feedback is generated.</li>
            <li>Sufficient quoted customer headers, an unambiguous active service association, and a usable source identity are needed. Incomplete reports and non-IP customer frames can limit the available feedback.</li>
          </ul>
        </section>
      </section>
    </section>
    <section anchor="ipv6">
      <name>Illustration of ICMP Error Handling from IPv6 Customer Edge</name>
      <t>The following examples use the common remote-SID association and local End.DX2 return-frame procedure. A6 and B6 are customer addresses, and E6 is PE1's configured service error-source address. The customer Hop Limit remains unchanged by VPWS bridging. Only the PTB example updates size state.</t>
      <section anchor="ipv6-unreachable">
        <name>Destination Unreachable</name>
        <t>P2 cannot deliver an invoking SRv6 customer-data packet and reports a supported Destination Unreachable error. PE1 validates the quoted service SID and recovers the return AC. Under the configured customer notification policy it constructs an ICMPv6 tunnel-unreachable response for A6.</t>
        <artwork type="ascii-art">
P2 to PE1 (example: no route to the SRv6 destination):
  IPv6(SA=P2::1, DA=PE1::1, NH=58)
    ICMPv6(Type=1, Code=0)
      [quoted SRv6 / Ethernet / IPv6(A6 -&gt; B6, HL=64) / data]

PE1 local reinjection:
  IPv6(SA=PE1::1, DA=PE1:DX2::, HL=64, NH=143)
    Ethernet(SA=M2, DA=M1, tags=local disposition format)
      IPv6(SA=E6, DA=A6, HL=64, NH=58)
        ICMPv6(Type=1, Code=3)
          [quoted customer IPv6(A6 -&gt; B6, HL=64) / data]
</artwork>
        <t>Local End.DX2 removes the local outer IPv6 header, applies AC1's VLAN treatment, and delivers the Ethernet error frame toward the customer source. The provider's original type and code remain available to the operator.</t>
      </section>
      <section anchor="ipv6-time">
        <name>Time Exceeded in the Provider Tunnel</name>
        <t>If the invoking outer Hop Limit expires at P2, the received message instead has ICMPv6 Type 3, Code 0 and quotes an outer Hop Limit of one. The customer packet still has Hop Limit 64. PE1 uses the same return-frame construction as <xref target="ipv6-unreachable"/>, with customer Type 1, Code 3. It MUST NOT report customer Hop Limit expiration, and PMTU state is unchanged.</t>
      </section>
      <section anchor="ipv6-parameter">
        <name>Parameter Problem</name>
        <t>An error pointing into the provider SRH is associated with the service for local diagnosis. PE1 does not send A6 a Parameter Problem pointer into a header that the customer never supplied. A recognized tunnel encapsulation-limit exhaustion can use the generic notification in <xref target="ipv6-unreachable"/>; a probe-related SR Upper-layer Header Error is consumed by the local probe process. Neither changes PMTU state.</t>
      </section>
      <section anchor="ipv6-ptb">
        <name>Packet Too Big</name>
        <t>CE1 sends a unicast Ethernet frame carrying an IPv6 packet from A6 to B6. Its customer Hop Limit is 64. PE1 bridges it without changing that field and applies an SRv6 encapsulation with an independently initialized outer Hop Limit. The customer IPv6 packet is 1500 octets long. The example uses no SRH and one retained VLAN tag, so H = 40 and L = 18. The resulting outer IPv6 packet is 1558 octets long and exceeds the 1500-octet core link MTU.</t>
        <artwork type="ascii-art">
CE1 to PE1, after the configured ingress VLAN treatment:
  Ethernet(SA=M1, DA=M2, VLAN=V)
    IPv6(SA=A6, DA=B6, HL=64, NH=UDP) / customer data

PE1 to the core:
  IPv6(SA=PE1::1, DA=PE2:DX2::, HL=64, NH=143)
    Ethernet(SA=M1, DA=M2, VLAN=V)
      IPv6(SA=A6, DA=B6, HL=64, NH=UDP) / customer data

P2 to PE1 when its outgoing IPv6 MTU is 1500:
  IPv6(SA=P2::1, DA=PE1::1, NH=58)
    ICMPv6(Type=2, Code=0, MTU=1500)
      [quoted invoking SRv6 packet, including customer frame]
</artwork>
        <t>PE1 recovers PE2:DX2:: from the quoted packet and matches it to the installed VPWS and path. After validation, it records M = 1500 and derives F = 1460 and C = 1442. It generates an ICMPv6 PTB, Type 2, Code 0, carrying MTU 1442, for A6.</t>
        <artwork type="ascii-art">
PE1 local reinjection (E6 is PE1's service error-source address):
  IPv6(SA=PE1::1, DA=PE1:DX2::, HL=64, NH=143)
    Ethernet(SA=M2, DA=M1, VLAN=Vlocal)
      IPv6(SA=E6, DA=A6, HL=64, NH=58)
        ICMPv6(Type=2, Code=0, MTU=1442)
          [quoted customer IPv6 packet: A6 -&gt; B6, HL=64]

Local End.DX2 disposition to AC1:
  Ethernet(SA=M2, DA=M1, VLAN=AC1 configured VLAN)
    IPv6(SA=E6, DA=A6, HL=64, NH=58)
      ICMPv6(Type=2, Code=0, MTU=1442)
        [quoted customer IPv6 packet]
</artwork>
        <t>Vlocal denotes the tag representation expected at the local disposition input, not necessarily the access VLAN on the original frame. CE1, or a host reached through CE1, receives the feedback in the correct Ethernet service. An oversized customer IPv6 packet is discarded; PE1 MUST NOT fragment the customer's IPv6 packet as a router.</t>
        <t>If C is less than 1280, the PE MUST NOT execute the generic customer PTB generation step with that value. It invokes a separately supported adaptation that preserves the required IPv6 service MTU or reports that the service cannot meet that requirement. A larger transport MTU or lower encapsulation overhead can restore service. Merely advertising a value below 1280 is ineffective because IPv6 PMTU processing discards it <xref target="RFC8201"/>. PE1 MUST NOT fragment the customer's IPv6 packet.</t>
      </section>
    </section>
    <section anchor="ipv4">
      <name>Illustration of ICMP Error Handling from IPv4 Customer Edge</name>
      <t>The outer SRv6 payload remains Ethernet for every customer error type. A4 and B4 are customer IPv4 addresses, and E4 is PE1's configured service error-source address. The outer upper-layer Next Header is 143; the regenerated customer IPv4 Protocol field is ICMP. Customer TTL is unchanged by bridging.</t>
      <section anchor="ipv4-unreachable">
        <name>Destination Unreachable</name>
        <t>After validating an underlay tunnel-delivery failure and finding the matching VPWS and return AC, PE1 constructs a generic customer ICMPv4 Destination Unreachable message, Type 3, Code 1. A provider-specific port, scope, or policy code is not copied into the customer error.</t>
        <artwork type="ascii-art">
P2 to PE1 (example: no route to the SRv6 destination):
  IPv6(SA=P2::1, DA=PE1::1, NH=58)
    ICMPv6(Type=1, Code=0)
      [quoted SRv6 / Ethernet / IPv4(A4 -&gt; B4, TTL=64) / data]

PE1 local reinjection:
  IPv6(SA=PE1::1, DA=PE1:DX2::, HL=64, NH=143)
    Ethernet(SA=M2, DA=M1, tags=local disposition format)
      IPv4(SA=E4, DA=A4, TTL=64, Protocol=ICMP)
        ICMPv4(Type=3, Code=1)
          [quoted customer IPv4(A4 -&gt; B4, TTL=64) / data]
</artwork>
        <t>Local End.DX2 delivers the error frame through AC1 with its configured VLAN treatment. No MTU field is copied from the underlay report and PMTU state is unchanged.</t>
      </section>
      <section anchor="ipv4-time">
        <name>Time Exceeded in the Provider Tunnel</name>
        <t>A received ICMPv6 Type 3, Code 0 reports expiration of the outer SRv6 Hop Limit. PE1 can notify A4 using the generic Type 3, Code 1 construction in <xref target="ipv4-unreachable"/>. It MUST NOT generate customer ICMPv4 Time Exceeded, Type 11, Code 0, for this event, because the customer's TTL did not expire.</t>
      </section>
      <section anchor="ipv4-parameter">
        <name>Parameter Problem</name>
        <t>A provider IPv6 Parameter Problem is handled according to <xref target="parameter-problem"/>. PE1 does not manufacture an IPv4 Parameter Problem pointer from a provider IPv6 or SRH offset. A recognized tunnel encapsulation-limit exhaustion can use the generic Type 3, Code 1 response; other provider-header and probe errors remain local as specified in the classification procedure.</t>
      </section>
      <section anchor="ipv4-ptb">
        <name>Packet Too Big and DF Handling</name>
        <t>The IPv4 case uses the same remote and local End.DX2 association. The outer SRv6 payload is still Ethernet, so its upper-layer Next Header remains 143. The invoking customer IPv4 packet is 1500 octets long. Its untagged Ethernet and SRv6 encapsulation yield an outer IPv6 packet of 1554 octets. For a 1500-octet core link, M = 1500, H = 40, L = 14, F = 1460, and C = 1446.</t>
        <artwork type="ascii-art">
CE1 to PE1:
  Ethernet(SA=M1, DA=M2)
    IPv4(SA=A4, DA=B4, TTL=64, DF=1, Protocol=UDP) / data

PE1 to the core:
  IPv6(SA=PE1::1, DA=PE2:DX2::, HL=64, NH=143)
    Ethernet(SA=M1, DA=M2)
      IPv4(SA=A4, DA=B4, TTL=64, DF=1, Protocol=UDP) / data

P2 to PE1:
  IPv6(SA=P2::1, DA=PE1::1, NH=58)
    ICMPv6(Type=2, Code=0, MTU=1500)
      [quoted invoking SRv6 packet, including customer frame]
</artwork>
        <t>For an oversized IPv4 packet with Don't Fragment (DF) set, PE1 generates ICMPv4 Destination Unreachable, Type 3, Code 4, with Next-Hop MTU C, following <xref target="RFC1191"/>. The advertised value MUST fit the 16-bit Next-Hop MTU field and MUST NOT exceed 65535. A limit too small to support the minimum IPv4 datagram is an unusable service condition, not a valid advertised MTU. It quotes the original customer IPv4 header and the required available payload, and sends the error to A4 in the return Ethernet frame.</t>
        <artwork type="ascii-art">
PE1 local reinjection (E4 is PE1's service error-source address):
  IPv6(SA=PE1::1, DA=PE1:DX2::, HL=64, NH=143)
    Ethernet(SA=M2, DA=M1, tags=local disposition representation)
      IPv4(SA=E4, DA=A4, TTL=64, Protocol=ICMP)
        ICMPv4(Type=3, Code=4, Next-Hop MTU=1446)
          [quoted customer IPv4 packet: A4 -&gt; B4, DF=1]

Local End.DX2 disposition to AC1:
  Ethernet(SA=M2, DA=M1, tags=AC1 configured tags)
    IPv4(SA=E4, DA=A4, TTL=64, Protocol=ICMP)
      ICMPv4(Type=3, Code=4, Next-Hop MTU=1446)
        [quoted customer IPv4 packet]
</artwork>
        <t>For DF-clear IPv4 traffic, a PE MAY support an explicitly enabled service adaptation that fragments the customer IPv4 packet, reconstructs an Ethernet frame for each fragment, and applies SRv6 encapsulation. Every resulting frame MUST fit F and every fragment MUST follow the IPv4 fragmentation rules <xref target="RFC791"/>. The original customer TTL is preserved by the bridging service; fragmentation does not add an IP forwarding hop.</t>
        <t>Customer IP fragmentation is not an inherent function of a transparent Ethernet service. Implementations and operators need to account for its effect on service transparency and performance. If this adaptation is disabled, the PE uses an explicitly supported outer fragmentation mode or drops the oversized DF-clear frame with counters and an operational indication. It MUST NOT generate a DF-set fragmentation-needed error merely because a DF-clear frame cannot be carried.</t>
      </section>
    </section>
    <section anchor="complex">
      <name>Dealing with Complex Scenarios</name>
      <section>
        <name>Non-IP Ethernet Traffic</name>
        <t>The PE can associate a provider error with a non-IP service frame for local diagnostics, but this revision does not define a customer ICMP response for that payload. The same restriction applies to otherwise uninspectable customer payloads. Existing customer Layer 2 control protocols continue to follow their service rules.</t>
        <t>For non-IP Ethernet payloads, including Layer 2 control traffic, PE1 cannot construct a meaningful customer ICMP error. For oversized non-IP frames, it applies the configured frame-size policy and records drops. An implementation MAY support a separate outer IPv6 fragmentation and reassembly service, but only when both tunnel endpoints and applicable SRv6 processing support it. IPv6 transit routers never fragment packets. Such a mode needs its own resource limits, security treatment, and interoperability specification; it is not described here.</t>
      </section>
      <section>
        <name>Compressed SIDs and Nested Encapsulation</name>
        <t>With compressed SID encodings, the PE reconstructs the ultimate service destination using the SID format and installed policy state <xref target="RFC9800"/>. With nested SRv6 encapsulation or a repair tunnel, the PE identifies the header chain and service context that actually carried the frame. The MTU-specific accounting uses the transport headers present at the reporting link, as defined in <xref target="mtu"/>. Other error classes use the header chain to establish ownership and interpret provider errors; they do not derive an MTU. If a truncated quote cannot supply the required information and retained context cannot resolve it safely, the PE declines customer error generation and reports the limitation.</t>
        <t>The service SID recovered after path processing need not be the current outer destination: that address may identify an active transit segment. Matching a final service SID also does not uniquely identify a policy when several policies share that SID. The PE MUST resolve ambiguity before choosing a path-specific disposition or changing path state.</t>
      </section>
      <section>
        <name>Multihoming and Service Changes</name>
        <t>The return AC and local service SID follow the current VPWS binding and eligibility state. Before local injection, PE1 checks that the association remains active and authorized. It MUST NOT return an error through a withdrawn or reassigned cross-connect. Services with multiple eligible return ACs need an explicit local selection rule; a remote SID match alone does not authorize replication to every AC.</t>
      </section>
    </section>
    <section anchor="pmtu">
      <name>Packet Too Big and Path MTU Discovery Considerations</name>
      <t>This section applies specifically to PTB reports and subsequent oversize frames. Destination Unreachable, Time Exceeded, and Parameter Problem messages use the common framework without invoking MTU changes or these optional validation probes.</t>
      <section anchor="mtu">
        <name>Compute the Effective MTU</name>
        <t>The following quantities distinguish the core IP MTU from the Ethernet frame and customer IP limits:</t>
        <dl spacing="normal">
          <dt>M</dt>
          <dd>The validated PMTU of the outer IPv6 packet on the selected transport path, excluding the core link-layer header.</dd>
          <dt>H</dt>
          <dd>The transport overhead present at the link whose MTU constraint is being evaluated, including outer IPv6 headers, SRHs, and any repair encapsulation present at that point.</dd>
          <dt>L</dt>
          <dd>The carried Ethernet header length, including VLAN tags retained inside the tunnel. Preamble, inter-packet gap, and Ethernet FCS are excluded for the H.Encaps.L2 encapsulation used here; FCS is regenerated on Ethernet disposition.</dd>
          <dt>F</dt>
          <dd>The maximum carried Ethernet frame length for a constant-overhead path, F = M - H.</dd>
          <dt>C</dt>
          <dd>The maximum customer IP packet length for that frame, C = F - L.</dd>
        </dl>
        <t>For Ethernet II, L is 14 octets without a tag, 18 octets with one tag, or 22 octets with two tags. H is derived from the actual forwarding context, not a fixed assumption about SRH length. Subtractions MUST be checked for underflow, and invalid or unusable values MUST NOT be installed.</t>
        <t>The per-service frame limit is F. The MTU advertised in a generated customer IP error is C, further limited by any known applicable service or AC IP-size constraint. Subtracting SRv6 overhead alone yields the frame limit, not the customer IP MTU. This applies the tunnel-overhead principle of <xref target="RFC2473"/> while accounting for the additional Ethernet payload.</t>
        <t>A known remote AC limit can further constrain the service, but an underlay probe does not measure the remote AC or the customer's downstream network. The procedure therefore discovers a transport constraint for the selected service path, rather than claiming to discover the complete customer end-to-end PMTU.</t>
        <t>When encapsulation changes along the path, each known link constraint is normalized to the carried Ethernet-frame length: F is no greater than the minimum of M_i minus H_i over the applicable constraints. The PE uses the overhead at the reporting link, and does not blindly subtract an SRH already removed upstream or ignore a repair header added later. If that context cannot be established, it uses a conservative supported bound or treats the report as unresolved.</t>
        <t>A received outer PTB reporting M below 1280 MUST be discarded before MTU cache or customer-response changes <xref target="RFC8201"/>. This outer-packet check is distinct from the customer IPv6 requirement. A computed customer IPv6 C below 1280 invokes the service adaptation or failure handling in <xref target="ipv6-ptb"/>, not a generated customer PTB carrying that smaller value.</t>
      </section>
      <section anchor="validation">
        <name>PTB Validation Probes</name>
        <section>
          <name>Validation Policy and Probe Construction</name>
          <t>Matching a quoted SID to a local service is necessary but does not authenticate the MTU claim. A forged PTB can name a valid SID. A validation policy therefore combines structural checks, trusted-domain controls, recent forwarding history, and, when needed, an active probe. A suspicious report can enter a pending-validation state with a finite configured deadline and bounded retry count. The PE distinguishes that unaccepted candidate from an accepted valid PTB. Once a PTB is accepted as valid, it MUST promptly reduce the applicable PMTU and avoid continued oversized transmission; active probing MUST NOT indefinitely gate that action. At the validation deadline, the PE either accepts the report under its documented checks or rejects it as unverified, ends pending state, and records an operational event. It MUST bound loss during pending validation and MUST NOT represent the previous size as known usable merely because probes timed out.</t>
          <t>PE1 constructs a fresh probe using the selected service's installed transport path and complete SID list, including the remote End.DX2 service destination. It MUST NOT substitute the ordinary route to the remote PE loopback when that differs from the service path. The source is a local PE1 IPv6 address. Relevant traffic class, flow label, and other entropy inputs SHOULD match those of the affected flow as far as feasible.</t>
          <t>The probe length is greater than the candidate M. It can equal the length of the original invoking outer IPv6 packet, recovered from its header and local transmission context. This is not the length of the received ICMP error or the truncated quoted bytes. The PE bounds probe size and rate by policy.</t>
          <t>The probe contains an unpredictable correlation value and an identifier for its outstanding validation context. These MUST be placed where a returned ICMP quote can include them. The PE checks the quoted addresses, SID/path context, length, token, and applicable forwarding generation before accepting a response. A response without sufficient quoted correlation information is inconclusive. No new globally assigned probe wire format is defined in this revision.</t>
        </section>
        <section anchor="probe-containment">
          <name>Preventing Probe Delivery to the Customer</name>
          <t>A validation probe MUST NOT become a customer Ethernet frame. One option uses a non-Ethernet upper-layer protocol that the remote End.DX2 is explicitly configured to reject or consume locally. The ingress PE sets the upper-layer Next Header in the SRH, or in the IPv6 header when no extension header intervenes. It does not set that field to Ethernet (143). An arbitrary currently unassigned protocol number is not an interoperable allocation strategy.</t>
          <t>When a disallowed upper-layer protocol reaches an End.DX2 destination, SRv6 endpoint processing can return ICMPv6 Parameter Problem, Code 4, SR Upper-layer Header Error <xref target="RFC8986"/>. A permitted ICMPv6 Echo request terminated locally at the endpoint is another possible probe profile. The selected profile MUST guarantee local termination or rejection, even if the path becomes shorter.</t>
          <t>An outer Hop Limit chosen to expire near the reported bottleneck MAY supplement containment. A guessed hop count, including a value intended to reach P2 plus one hop, MUST NOT be the sole protection against access delivery. Topology changes and final endpoint processing can defeat that assumption. Time Exceeded is a diagnostic outcome, not a measurement of a size limit.</t>
        </section>
        <section>
          <name>Interpreting Probe Responses</name>
          <dl spacing="normal">
            <dt>Correlated PTB</dt>
            <dd>A fresh PTB that quotes the outstanding probe and passes the applicable checks corroborates a size constraint on the sampled path. If accepted as valid, its MTU promptly updates the applicable service path. The reporting router and value can differ after a route change; the PE re-evaluates the path association.</dd>
            <dt>Correlated terminal response</dt>
            <dd>A reply from the intended endpoint, such as an expected upper-layer error or permitted Echo reply, indicates that this probe reached that endpoint at this size. It does not corroborate the claimed smaller MTU on the sampled path. PE1 retains the previous validated state and can retry according to policy.</dd>
            <dt>Time Exceeded, timeout, or uncorrelated response</dt>
            <dd>The result is inconclusive for PMTU validation. PE1 does not classify the original PTB as forged solely from this outcome. It can retry with backoff within the configured deadline. At that deadline it resolves the pending report using the acceptance and loss-protection policy above. A timeout does not cancel the prompt processing required for an already accepted PTB.</dd>
          </dl>
          <t>Probing is evidence of reachability or a size constraint at a particular time. It cannot prove the original PTB's authenticity or defeat an on-path attacker. Changing payload protocol can also change an ECMP choice. If the implementation cannot reproduce the relevant load-balancing context, it MUST NOT claim that one successful probe validates every member of a service's multipath forwarding set. Conservative per-path limits or additional samples are needed.</t>
          <t>Probe failures MUST NOT recursively trigger additional probes. The PE bounds outstanding validation state, coalesces equivalent requests, applies backoff, and records the result and its confidence. Packetization-layer discovery principles and security limitations are discussed in <xref target="RFC8899"/>; this procedure is a tunnel validation mechanism and does not replace customer transport PMTU discovery.</t>
        </section>
      </section>
      <section anchor="state">
        <name>PMTU State and Subsequent Packet Handling</name>
        <t>The cache is keyed by sufficient service path context to distinguish different forwarding treatments of the same remote service SID. A useful key contains the local service identifier, complete remote SID, SR policy or transport-path identifier, selected segment list or forwarding generation, and the encapsulation profile. Where necessary, it also distinguishes ECMP members or uses a conservative limit valid for the relevant forwarding set.</t>
        <t>The cache records the validated outer PMTU M, derived frame limit F, validation method and timestamp, pending candidate information, and expiry or revalidation state. The PE derives C for each frame's actual carried Layer 2 overhead. It MUST NOT apply one customer's service association or return AC to another service merely because they share a locator.</t>
        <t>Once a lower PTB-derived limit is accepted, PE1 checks subsequent frames before SRv6 encapsulation. Oversized customer IPv6 packets and DF-set IPv4 packets receive the feedback described in <xref target="ipv6-ptb"/> and <xref target="ipv4-ptb"/>, subject to rate limits and error suppression. Frames within the limit continue forwarding. Repeated oversized packets need not cause repeated core loss or a fresh validation exchange for every packet.</t>
        <t>The PE ages state and probes cautiously for increases, consistent with the applicable PMTU discovery procedures. A PTB MUST NOT increase the cached PMTU. Non-PTB errors MUST NOT modify PMTU state. A policy, segment list, service binding, or encapsulation change invalidates or revalidates the affected association. A repair path may have a different overhead and bottleneck; the original limit cannot automatically be transferred to it.</t>
      </section>
    </section>
    <section anchor="operations">
      <name>Operational Considerations</name>
      <t>The feature is enabled per service or service class. Operators configure the permitted error classes, customer notification policy, service error-source addresses, VLAN disposition treatment, and rates for generated errors. PTB-specific configuration additionally covers acceptance and probe policy, finite validation deadlines, permitted adaptation modes, and minimum service MTU. Services requiring transparent preservation of all customer frames need an adequate transport MTU or a separately supported adaptation rather than unrestricted IP inspection.</t>
      <t>Useful counters include received errors by type and code, accepted and rejected reports, unknown or ambiguous SIDs, insufficient quotes, local-only dispositions, suppressed customer errors, regenerated errors by customer type and code, and local injection failures. PTB-specific counters include probes, correlated PTBs, terminal responses, timeouts, PMTU changes, fragmentation activity, and non-IP oversize drops. Operators should be able to inspect the service/path key, H, frame limit, last validation result, and cache age.</t>
      <t>A direct PE-to-PE probe through the default route can be useful for infrastructure diagnostics but does not establish the PMTU of a different traffic-engineered service path. Operators also need to consider ICMP filtering, ECMP sampling, encapsulation changes, and access MTUs when interpreting discovery results.</t>
    </section>
    <section anchor="security">
      <name>Security Considerations</name>
      <t>Forged ICMP errors can create misleading reachability feedback, consume control-plane resources, or, for PTB, reduce a service's usable packet size. Service association, quoted-packet validation, domain filtering, probe correlation, bounded state, and rate limits reduce this exposure. Active probing does not supply cryptographic authentication. An attacker on the selected path can observe or fabricate responses and can also drop the data traffic.</t>
      <t>Type, code, and pointer validation prevents a provider-header error from being misrepresented as a customer fault. Reports referring to locally generated probes MUST be consumed by the local diagnostic process and MUST NOT be injected into a customer service. Rate limits apply across error classes as well as per service.</t>
      <t>Local End.DX2 reinjection is a privileged operation. The selected AC MUST come from the validated active service association, not an arbitrary SID or interface supplied by the received packet. An implementation MUST preserve service isolation and prevent recirculation loops, duplicate access rewrites, and customer injection into unrelated service contexts.</t>
      <t>Returning ICMP errors can disclose provider behavior and customer header information. Generated quotes are limited to the required customer packet bytes; provider SRv6 headers are removed. Configured service error-source addresses can avoid exporting provider infrastructure addresses. The address-swap proxy proposal requires additional authorization and anti-spoofing analysis because it uses a customer's remote address.</t>
      <t>Probe generation can become a denial-of-service amplifier unless its rate, size, and outstanding state are bounded. Every probe MUST be contained before customer Ethernet delivery. Implementations also apply the applicable ICMP error-suppression rules and restrict customer IP fragmentation by policy. The security considerations of the referenced EVPN, SRv6, ICMP, and PMTU specifications also apply.</t>
    </section>
    <section anchor="iana">
      <name>IANA Considerations</name>
      <t>This document makes no IANA requests. It uses existing SRv6 endpoint behaviors and ICMP message types. No new Next Header value, ICMP type or code, BGP attribute, or probe port is allocated by this revision.</t>
    </section>
    <section>
      <name>Acknowledgements</name>
      <t>The authors welcome discussion of these proposed procedures in the IETF.</t>
    </section>
  </middle>
  <back>
    <references>
      <name>References</name>
      <references>
        <name>Normative References</name>
        <reference anchor="RFC791" target="https://www.rfc-editor.org/info/rfc791">
          <front>
            <title>Internet Protocol</title>
            <author initials="J." surname="Postel"/>
            <date year="1981" month="September"/>
          </front>
          <seriesInfo name="STD" value="5"/>
          <seriesInfo name="RFC" value="791"/>
          <seriesInfo name="DOI" value="10.17487/RFC791"/>
        </reference>
        <reference anchor="RFC792" target="https://www.rfc-editor.org/info/rfc792">
          <front>
            <title>Internet Control Message Protocol</title>
            <author initials="J." surname="Postel"/>
            <date year="1981" month="September"/>
          </front>
          <seriesInfo name="STD" value="5"/>
          <seriesInfo name="RFC" value="792"/>
          <seriesInfo name="DOI" value="10.17487/RFC792"/>
        </reference>
        <reference anchor="RFC1191" target="https://www.rfc-editor.org/info/rfc1191">
          <front>
            <title>Path MTU discovery</title>
            <author initials="J." surname="Mogul"/>
            <author initials="S." surname="Deering"/>
            <date year="1990" month="November"/>
          </front>
          <seriesInfo name="RFC" value="1191"/>
          <seriesInfo name="DOI" value="10.17487/RFC1191"/>
        </reference>
        <reference anchor="RFC2119" target="https://www.rfc-editor.org/info/rfc2119">
          <front>
            <title>Key words for use in RFCs to Indicate Requirement Levels</title>
            <author initials="S." surname="Bradner"/>
            <date year="1997" month="March"/>
          </front>
          <seriesInfo name="BCP" value="14"/>
          <seriesInfo name="RFC" value="2119"/>
          <seriesInfo name="DOI" value="10.17487/RFC2119"/>
        </reference>
        <reference anchor="RFC2473" target="https://www.rfc-editor.org/info/rfc2473">
          <front>
            <title>Generic Packet Tunneling in IPv6 Specification</title>
            <author initials="A." surname="Conta"/>
            <author initials="S." surname="Deering"/>
            <date year="1998" month="December"/>
          </front>
          <seriesInfo name="RFC" value="2473"/>
          <seriesInfo name="DOI" value="10.17487/RFC2473"/>
        </reference>
        <reference anchor="RFC4443" target="https://www.rfc-editor.org/info/rfc4443">
          <front>
            <title>Internet Control Message Protocol (ICMPv6) for the Internet Protocol Version 6 (IPv6) Specification</title>
            <author initials="A." surname="Conta"/>
            <author initials="S." surname="Deering"/>
            <author initials="M." surname="Gupta" role="editor"/>
            <date year="2006" month="March"/>
          </front>
          <seriesInfo name="STD" value="89"/>
          <seriesInfo name="RFC" value="4443"/>
          <seriesInfo name="DOI" value="10.17487/RFC4443"/>
        </reference>
        <reference anchor="RFC8174" target="https://www.rfc-editor.org/info/rfc8174">
          <front>
            <title>Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words</title>
            <author initials="B." surname="Leiba"/>
            <date year="2017" month="May"/>
          </front>
          <seriesInfo name="BCP" value="14"/>
          <seriesInfo name="RFC" value="8174"/>
          <seriesInfo name="DOI" value="10.17487/RFC8174"/>
        </reference>
        <reference anchor="RFC8201" target="https://www.rfc-editor.org/info/rfc8201">
          <front>
            <title>Path MTU Discovery for IP version 6</title>
            <author initials="J." surname="McCann"/>
            <author initials="S." surname="Deering"/>
            <author initials="J." surname="Mogul"/>
            <author initials="R." surname="Hinden" role="editor"/>
            <date year="2017" month="July"/>
          </front>
          <seriesInfo name="STD" value="87"/>
          <seriesInfo name="RFC" value="8201"/>
          <seriesInfo name="DOI" value="10.17487/RFC8201"/>
        </reference>
        <reference anchor="RFC8214" target="https://www.rfc-editor.org/info/rfc8214">
          <front>
            <title>Virtual Private Wire Service Support in Ethernet VPN</title>
            <author initials="S." surname="Boutros"/>
            <author initials="A." surname="Sajassi"/>
            <author initials="S." surname="Salam"/>
            <author initials="J." surname="Drake"/>
            <author initials="J." surname="Rabadan"/>
            <date year="2017" month="August"/>
          </front>
          <seriesInfo name="RFC" value="8214"/>
          <seriesInfo name="DOI" value="10.17487/RFC8214"/>
        </reference>
        <reference anchor="RFC8754" target="https://www.rfc-editor.org/info/rfc8754">
          <front>
            <title>IPv6 Segment Routing Header (SRH)</title>
            <author initials="C." surname="Filsfils" role="editor"/>
            <author initials="D." surname="Dukes" role="editor"/>
            <author initials="S." surname="Previdi"/>
            <author initials="J." surname="Leddy"/>
            <author initials="S." surname="Matsushima"/>
            <author initials="D." surname="Voyer"/>
            <date year="2020" month="March"/>
          </front>
          <seriesInfo name="RFC" value="8754"/>
          <seriesInfo name="DOI" value="10.17487/RFC8754"/>
        </reference>
        <reference anchor="RFC8986" target="https://www.rfc-editor.org/info/rfc8986">
          <front>
            <title>Segment Routing over IPv6 (SRv6) Network Programming</title>
            <author initials="C." surname="Filsfils" role="editor"/>
            <author initials="P." surname="Camarillo" role="editor"/>
            <author initials="J." surname="Leddy"/>
            <author initials="D." surname="Voyer"/>
            <author initials="S." surname="Matsushima"/>
            <author initials="Z." surname="Li"/>
            <date year="2021" month="February"/>
          </front>
          <seriesInfo name="RFC" value="8986"/>
          <seriesInfo name="DOI" value="10.17487/RFC8986"/>
        </reference>
        <reference anchor="RFC9252" target="https://www.rfc-editor.org/info/rfc9252">
          <front>
            <title>BGP Overlay Services Based on Segment Routing over IPv6 (SRv6)</title>
            <author initials="G." surname="Dawra" role="editor"/>
            <author initials="K." surname="Talaulikar" role="editor"/>
            <author initials="R." surname="Raszuk"/>
            <author initials="B." surname="Decraene"/>
            <author initials="S." surname="Zhuang"/>
            <author initials="J." surname="Rabadan"/>
            <date year="2022" month="July"/>
          </front>
          <seriesInfo name="RFC" value="9252"/>
          <seriesInfo name="DOI" value="10.17487/RFC9252"/>
        </reference>
        <reference anchor="RFC9800" target="https://www.rfc-editor.org/info/rfc9800">
          <front>
            <title>Compressed SRv6 Segment List Encoding</title>
            <author initials="W." surname="Cheng" role="editor"/>
            <author initials="C." surname="Filsfils"/>
            <author initials="Z." surname="Li"/>
            <author initials="B." surname="Decraene"/>
            <author initials="F." surname="Clad" role="editor"/>
            <date year="2025" month="June"/>
          </front>
          <seriesInfo name="RFC" value="9800"/>
          <seriesInfo name="DOI" value="10.17487/RFC9800"/>
        </reference>
        <reference anchor="RFC1812" target="https://www.rfc-editor.org/info/rfc1812">
          <front>
            <title>Requirements for IP Version 4 Routers</title>
            <author initials="F." surname="Baker" role="editor"/>
            <date year="1995" month="June"/>
          </front>
          <seriesInfo name="RFC" value="1812"/>
          <seriesInfo name="DOI" value="10.17487/RFC1812"/>
        </reference>
        <reference anchor="RFC8200" target="https://www.rfc-editor.org/info/rfc8200">
          <front>
            <title>Internet Protocol, Version 6 (IPv6) Specification</title>
            <author initials="S." surname="Deering"/>
            <author initials="R." surname="Hinden"/>
            <date year="2017" month="July"/>
          </front>
          <seriesInfo name="STD" value="86"/>
          <seriesInfo name="RFC" value="8200"/>
          <seriesInfo name="DOI" value="10.17487/RFC8200"/>
        </reference>
      </references>
      <references>
        <name>Informative References</name>
        <reference anchor="RFC8899" target="https://www.rfc-editor.org/info/rfc8899">
          <front>
            <title>Packetization Layer Path MTU Discovery for Datagram Transports</title>
            <author initials="G." surname="Fairhurst"/>
            <author initials="T." surname="Jones"/>
            <author initials="M." surname="Tuexen"/>
            <author initials="I." surname="Ruengeler"/>
            <author initials="T." surname="Voelker"/>
            <date year="2020" month="September"/>
          </front>
          <seriesInfo name="RFC" value="8899"/>
          <seriesInfo name="DOI" value="10.17487/RFC8899"/>
        </reference>
        <reference anchor="I-D.alivar-spring-icmp-error-srv6-vpn" target="https://datatracker.ietf.org/doc/html/draft-alivar-spring-icmp-error-srv6-vpn-00">
          <front>
            <title>ICMP Error Handling for VPNs in SRv6 Networks</title>
            <author initials="Z." surname="Ali"/>
            <author initials="B." surname="Varga"/>
            <author initials="K." surname="Szarkowicz"/>
            <author initials="M." surname="Manish"/>
            <author initials="S." surname="Joy"/>
            <date year="2026" month="November" day="15"/>
          </front>
          <seriesInfo name="Internet-Draft" value="draft-alivar-spring-icmp-error-srv6-vpn-00"/>
        </reference>
      </references>
    </references>
    <section anchor="issues">
      <name>Further Considerations</name>
      <t>The following points identify further work and applicability questions associated with the approach described in this Informational document.</t>
      <ul spacing="normal">
        <li>Define an interoperable validation probe profile, including exact correlation placement, endpoint termination behavior, and treatment of entropy differences. The current procedure deliberately does not assign an unknown protocol value.</li>
        <li>Determine whether a transparent customer IP address-swap proxy profile should be standardized, and specify its source ownership, authorization, error suppression, and filtering requirements. The examples use configured PE service error-source addresses.</li>
        <li>Specify any supported outer fragmentation and reassembly mode for services whose frame requirement exceeds the available SRv6 path budget, including interaction with extension headers and endpoint processing.</li>
        <li>Specify the minimum quoted-packet coverage and retained forwarding context needed for long segment lists, nested encapsulation, and probe correlation when an ICMP quote is truncated.</li>
        <li>Evaluate extensions to multipoint EVPN and other Layer 2 endpoint behaviors, including replication, source-MAC selection, and return-service identification.</li>
        <li>Consider additional error-code mappings only where their customer semantics are defined. In particular, provider Parameter Problem pointers and provider fragment-reassembly errors require explicit rules before customer translation.</li>
      </ul>
    </section>
  </back>
</rfc>
