Internet Engineering Task Force (IETF)                   J. Rabadan, Ed.
Request for Comments: 9136                                 W. Henderickx
Category: Standards Track                                          Nokia
ISSN: 2070-1721                                                 J. Drake
                                                                  W. Lin
                                                                 Juniper
                                                              A. Sajassi
                                                                   Cisco
                                                            October 2021


             IP Prefix Advertisement in Ethernet VPN (EVPN)

Abstract

   The BGP MPLS-based Ethernet VPN (EVPN) (RFC 7432) mechanism provides
   a flexible control plane that allows intra-subnet connectivity in an
   MPLS and/or Network Virtualization Overlay (NVO) (RFC 7365) network.
   In some networks, there is also a need for dynamic and efficient
   inter-subnet connectivity across Tenant Systems and end devices that
   can be physical or virtual and do not necessarily participate in
   dynamic routing protocols.  This document defines a new EVPN route
   type for the advertisement of IP prefixes and explains some use-case
   examples where this new route type is used.

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/rfc9136.

Copyright Notice

   Copyright (c) 2021 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

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

Table of Contents

   1.  Introduction
     1.1.  Terminology
   2.  Problem Statement
     2.1.  Inter-Subnet Connectivity Requirements in Data Centers
     2.2.  The Need for the EVPN IP Prefix Route
   3.  The BGP EVPN IP Prefix Route
     3.1.  IP Prefix Route Encoding
     3.2.  Overlay Indexes and Recursive Lookup Resolution
   4.  Overlay Index Use Cases
     4.1.  TS IP Address Overlay Index Use Case
     4.2.  Floating IP Overlay Index Use Case
     4.3.  Bump-in-the-Wire Use Case
     4.4.  IP-VRF-to-IP-VRF Model
       4.4.1.  Interface-less IP-VRF-to-IP-VRF Model
       4.4.2.  Interface-ful IP-VRF-to-IP-VRF with SBD IRB
       4.4.3.  Interface-ful IP-VRF-to-IP-VRF with Unnumbered SBD IRB
   5.  Security Considerations
   6.  IANA Considerations
   7.  References
     7.1.  Normative References
     7.2.  Informative References
   Acknowledgments
   Contributors
   Authors' Addresses

1.  Introduction

   [RFC7365] provides a framework for Data Center (DC) Network
   Virtualization over Layer 3 and specifies that the Network
   Virtualization Edge (NVE) devices must provide Layer 2 and Layer 3
   virtualized network services in multi-tenant DCs.  [RFC8365]
   discusses the use of EVPN as the technology of choice to provide
   Layer 2 or intra-subnet services in these DCs.  This document, along
   with [RFC9135], specifies the use of EVPN for Layer 3 or inter-subnet
   connectivity services.

   [RFC9135] defines some fairly common inter-subnet forwarding
   scenarios where Tenant Systems (TSs) can exchange packets with TSs
   located in remote subnets.  In order to achieve this, [RFC9135]
   describes how Media Access Control (MAC) and IPs encoded in TS RT-2
   routes are not only used to populate MAC Virtual Routing and
   Forwarding (MAC-VRF) and overlay Address Resolution Protocol (ARP)
   tables but also IP-VRF tables with the encoded TS host routes (/32 or
   /128).  In some cases, EVPN may advertise IP prefixes and therefore
   provide aggregation in the IP-VRF tables, as opposed to propagating
   individual host routes.  This document complements the scenarios
   described in [RFC9135] and defines how EVPN may be used to advertise
   IP prefixes.  Interoperability between EVPN and Layer 3 Virtual
   Private Network (VPN) [RFC4364] IP Prefix routes is out of the scope
   of this document.

   Section 2.1 describes the inter-subnet connectivity requirements in
   DCs.  Section 2.2 explains why a new EVPN route type is required for
   IP prefix advertisements.  Sections 3, 4, and 5 will describe this
   route type and how it is used in some specific use cases.

1.1.  Terminology

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
   "OPTIONAL" in this document are to be interpreted as described in BCP
   14 [RFC2119] [RFC8174] when, and only when, they appear in all
   capitals, as shown here.

   AC:       Attachment Circuit

   ARP:      Address Resolution Protocol

   BD:       Broadcast Domain.  As per [RFC7432], an EVI consists of a
             single BD or multiple BDs.  In case of VLAN-bundle and
             VLAN-based service models (see [RFC7432]), a BD is
             equivalent to an EVI.  In case of a VLAN-aware bundle
             service model, an EVI contains multiple BDs.  Also, in this
             document, "BD" and "subnet" are equivalent terms.

   BD Route Target:  Refers to the broadcast-domain-assigned Route
             Target [RFC4364].  In case of a VLAN-aware bundle service
             model, all the BD instances in the MAC-VRF share the same
             Route Target.

   BT:       Bridge Table.  The instantiation of a BD in a MAC-VRF, as
             per [RFC7432].

   CE:       Customer Edge

   DA:       Destination Address

   DGW:      Data Center Gateway

   Ethernet A-D Route:  Ethernet Auto-Discovery (A-D) route, as per
             [RFC7432].

   Ethernet NVO Tunnel:  Refers to Network Virtualization Overlay
             tunnels with Ethernet payload.  Examples of this type of
             tunnel are VXLAN or GENEVE.

   EVI:      EVPN Instance spanning the NVE/PE devices that are
             participating on that EVPN, as per [RFC7432].

   EVPN:     Ethernet VPN, as per [RFC7432].

   GENEVE:   Generic Network Virtualization Encapsulation, as per
             [RFC8926].

   GRE:      Generic Routing Encapsulation

   GW IP:    Gateway IP address

   IPL:      IP Prefix Length

   IP NVO Tunnel:  Refers to Network Virtualization Overlay tunnels with
             IP payload (no MAC header in the payload).

   IP-VRF:   A Virtual Routing and Forwarding table for IP routes on an
             NVE/PE.  The IP routes could be populated by EVPN and IP-
             VPN address families.  An IP-VRF is also an instantiation
             of a Layer 3 VPN in an NVE/PE.

   IRB:      Integrated Routing and Bridging interface.  It connects an
             IP-VRF to a BD (or subnet).

   MAC:      Media Access Control

   MAC-VRF:  A Virtual Routing and Forwarding table for MAC addresses on
             an NVE/PE, as per [RFC7432].  A MAC-VRF is also an
             instantiation of an EVI in an NVE/PE.

   ML:       MAC Address Length

   ND:       Neighbor Discovery

   NVE:      Network Virtualization Edge

   NVO:      Network Virtualization Overlay

   PE:       Provider Edge

   RT-2:     EVPN Route Type 2, i.e., MAC/IP Advertisement route, as
             defined in [RFC7432].

   RT-5:     EVPN Route Type 5, i.e., IP Prefix route, as defined in
             Section 3.

   SBD:      Supplementary Broadcast Domain.  A BD that does not have
             any ACs, only IRB interfaces, and is used to provide
             connectivity among all the IP-VRFs of the tenant.  The SBD
             is only required in IP-VRF-to-IP-VRF use cases (see
             Section 4.4).

   SN:       Subnet

   TS:       Tenant System

   VA:       Virtual Appliance

   VM:       Virtual Machine

   VNI:      Virtual Network Identifier.  As in [RFC8365], the term is
             used as a representation of a 24-bit NVO instance
             identifier, with the understanding that "VNI" will refer to
             a VXLAN Network Identifier in VXLAN, or a Virtual Network
             Identifier in GENEVE, etc., unless it is stated otherwise.

   VSID:     Virtual Subnet Identifier

   VTEP:     VXLAN Termination End Point, as per [RFC7348].

   VXLAN:    Virtual eXtensible Local Area Network, as per [RFC7348].

   This document also assumes familiarity with the terminology of
   [RFC7365], [RFC7432], and [RFC8365].

2.  Problem Statement

   This section describes the inter-subnet connectivity requirements in
   DCs and why a specific route type to advertise IP prefixes is needed.

2.1.  Inter-Subnet Connectivity Requirements in Data Centers

   [RFC7432] is used as the control plane for an NVO solution in DCs,
   where NVE devices can be located in hypervisors or Top-of-Rack (ToR)
   switches, as described in [RFC8365].

   The following considerations apply to TSs that are physical or
   virtual systems identified by MAC (and possibly IP addresses) and are
   connected to BDs by Attachment Circuits:

   *  The Tenant Systems may be VMs that generate traffic from their own
      MAC and IP.

   *  The Tenant Systems may be VA entities that forward traffic to/from
      IP addresses of different end devices sitting behind them.

      -  These VAs can be firewalls, load balancers, NAT devices, other
         appliances, or virtual gateways with virtual routing instances.

      -  These VAs do not necessarily participate in dynamic routing
         protocols and hence rely on the EVPN NVEs to advertise the
         routes on their behalf.

      -  In all these cases, the VA will forward traffic to other TSs
         using its own source MAC, but the source IP will be the one
         associated with the end device sitting behind the VA or a
         translated IP address (part of a public NAT pool) if the VA is
         performing NAT.

      -  Note that the same IP address and endpoint could exist behind
         two of these TSs.  One example of this would be certain
         appliance resiliency mechanisms, where a virtual IP or floating
         IP can be owned by one of the two VAs running the resiliency
         protocol (the Master VA).  The Virtual Router Redundancy
         Protocol (VRRP) [RFC5798] is one particular example of this.
         Another example is multihomed subnets, i.e., the same subnet is
         connected to two VAs.

      -  Although these VAs provide IP connectivity to VMs and the
         subnets behind them, they do not always have their own IP
         interface connected to the EVPN NVE; Layer 2 firewalls are
         examples of VAs not supporting IP interfaces.

   Figure 1 illustrates some of the examples described above.

                       NVE1
                    +-----------+
           TS1(VM)--|  (BD-10)  |-----+
             M1/IP1 +-----------+     |               DGW1
                                  +---------+    +-------------+
                                  |         |----|  (BD-10)    |
     SN1---+           NVE2       |         |    |    IRB1\    |
           |        +-----------+ |         |    |     (IP-VRF)|---+
     SN2---TS2(VA)--|  (BD-10)  |-|         |    +-------------+  _|_
           | M2/IP2 +-----------+ |  VXLAN/ |                    (   )
     IP4---+  <-+                 |  GENEVE |         DGW2      ( WAN )
                |                 |         |    +-------------+ (___)
             vIP23 (floating)     |         |----|  (BD-10)    |   |
                |                 +---------+    |    IRB2\    |   |
     SN1---+  <-+      NVE3         |  |  |      |     (IP-VRF)|---+
           | M3/IP3 +-----------+   |  |  |      +-------------+
     SN3---TS3(VA)--|  (BD-10)  |---+  |  |
           |        +-----------+      |  |
     IP5---+                           |  |
                                       |  |
                    NVE4               |  |      NVE5            +--SN5
              +---------------------+  |  | +-----------+        |
     IP6------|  (BD-1)             |  |  +-|  (BD-10)  |--TS4(VA)--SN6
              |       \             |  |    +-----------+        |
              |    (IP-VRF)         |--+                ESI4     +--SN7
              |       /  \IRB3      |
          |---|  (BD-2)  (BD-10)    |
       SN4|   +---------------------+


     Note:
     ESI4 = Ethernet Segment Identifier 4

                    Figure 1: DC Inter-subnet Use Cases

   Where:

   NVE1, NVE2, NVE3, NVE4, NVE5, DGW1, and DGW2 share the same BD for a
   particular tenant.  BD-10 is comprised of the collection of BD
   instances defined in all the NVEs.  All the hosts connected to BD-10
   belong to the same IP subnet.  The hosts connected to BD-10 are
   listed below:

   *  TS1 is a VM that generates/receives traffic to/from IP1, where IP1
      belongs to the BD-10 subnet.

   *  TS2 and TS3 are VAs that send/receive traffic to/from the subnets
      and hosts sitting behind them (SN1, SN2, SN3, IP4, and IP5).
      Their IP addresses (IP2 and IP3) belong to the BD-10 subnet, and
      they can also generate/receive traffic.  When these VAs receive
      packets destined to their own MAC addresses (M2 and M3), they will
      route the packets to the proper subnet or host.  These VAs do not
      support routing protocols to advertise the subnets connected to
      them and can move to a different server and NVE when the cloud
      management system decides to do so.  These VAs may also support
      redundancy mechanisms for some subnets, similar to VRRP, where a
      floating IP is owned by the Master VA and only the Master VA
      forwards traffic to a given subnet.  For example, vIP23 in
      Figure 1 is a floating IP that can be owned by TS2 or TS3
      depending on which system is the Master.  Only the Master will
      forward traffic to SN1.

   *  Integrated Routing and Bridging interfaces IRB1, IRB2, and IRB3
      have their own IP addresses that belong to the BD-10 subnet too.
      These IRB interfaces connect the BD-10 subnet to Virtual Routing
      and Forwarding (IP-VRF) instances that can route the traffic to
      other subnets for the same tenant (within the DC or at the other
      end of the WAN).

   *  TS4 is a Layer 2 VA that provides connectivity to subnets SN5,
      SN6, and SN7 but does not have an IP address itself in the BD-10.
      TS4 is connected to a port on NVE5 that is assigned to Ethernet
      Segment Identifier 4 (ESI4).

   For a BD to which an ingress NVE is attached, "Overlay Index" is
   defined as an identifier that the ingress EVPN NVE requires in order
   to forward packets to a subnet or host in a remote subnet.  As an
   example, vIP23 (Figure 1) is an Overlay Index that any NVE attached
   to BD-10 needs to know in order to forward packets to SN1.  The IRB3
   IP address is an Overlay Index required to get to SN4, and ESI4 is an
   Overlay Index needed to forward traffic to SN5.  In other words, the
   Overlay Index is a next hop in the overlay address space that can be
   an IP address, a MAC address, or an ESI.  When advertised along with
   an IP prefix, the Overlay Index requires a recursive resolution to
   find out the egress NVE to which the EVPN packets need to be sent.

   All the DC use cases in Figure 1 require inter-subnet forwarding;
   therefore, the individual host routes and subnets:

   a)  must be advertised from the NVEs (since VAs and VMs do not
       participate in dynamic routing protocols) and

   b)  may be associated with an Overlay Index that can be a VA IP
       address, a floating IP address, a MAC address, or an ESI.  The
       Overlay Index is further discussed in Section 3.2.

2.2.  The Need for the EVPN IP Prefix Route

   [RFC7432] defines a MAC/IP Advertisement route (also referred to as
   "RT-2") where a MAC address can be advertised together with an IP
   address length and IP address (IP).  While a variable IP address
   length might have been used to indicate the presence of an IP prefix
   in a route type 2, there are several specific use cases in which
   using this route type to deliver IP prefixes is not suitable.

   One example of such use cases is the "floating IP" example described
   in Section 2.1.  In this example, it is necessary to decouple the
   advertisement of the prefixes from the advertisement of a MAC address
   of either M2 or M3; otherwise, the solution gets highly inefficient
   and does not scale.

   For example, if 1,000 prefixes are advertised from M2 (using RT-2)
   and the floating IP owner changes from M2 to M3, 1,000 routes would
   be withdrawn by M2 and readvertised by M3.  However, if a separate
   route type is used, 1,000 routes can be advertised as associated with
   the floating IP address (vIP23), and only one RT-2 can be used for
   advertising the ownership of the floating IP, i.e., vIP23 and M2 in
   the route type 2.  When the floating IP owner changes from M2 to M3,
   a single RT-2 withdrawal/update is required to indicate the change.
   The remote DGW will not change any of the 1,000 prefixes associated
   with vIP23 but will only update the ARP resolution entry for vIP23
   (now pointing at M3).

   An EVPN route (type 5) for the advertisement of IP prefixes is
   described in this document.  This new route type has a differentiated
   role from the RT-2 route and addresses the inter-subnet connectivity
   scenarios for DCs (or NVO-based networks in general) described in
   this document.  Using this new RT-5, an IP prefix may be advertised
   along with an Overlay Index, which can be a GW IP address, a MAC, or
   an ESI.  The IP prefix may also be advertised without an Overlay
   Index, in which case the BGP next hop will point at the egress NVE,
   Area Border Router (ABR), or ASBR, and the MAC in the EVPN Router's
   MAC Extended Community will provide the inner MAC destination address
   to be used.  As discussed throughout the document, the EVPN RT-2 does
   not meet the requirements for all the DC use cases; therefore, this
   EVPN route type 5 is required.

   The EVPN route type 5 decouples the IP prefix advertisements from the
   MAC/IP Advertisement routes in EVPN.  Hence:

   a)  The clean and clear advertisements of IPv4 or IPv6 prefixes in a
       Network Layer Reachability Information (NLRI) message without MAC
       addresses are allowed.

   b)  Since the route type is different from the MAC/IP Advertisement
       route, the current procedures described in [RFC7432] do not need
       to be modified.

   c)  A flexible implementation is allowed where the prefix can be
       linked to different types of Overlay/Underlay Indexes: overlay IP
       addresses, overlay MAC addresses, overlay ESIs, underlay BGP next
       hops, etc.

   d)  An EVPN implementation not requiring IP prefixes can simply
       discard them by looking at the route type value.

   The following sections describe how EVPN is extended with a route
   type for the advertisement of IP prefixes and how this route is used
   to address the inter-subnet connectivity requirements existing in the
   DC.

3.  The BGP EVPN IP Prefix Route

   The BGP EVPN NLRI as defined in [RFC7432] is shown below:

       +-----------------------------------+
       |    Route Type (1 octet)           |
       +-----------------------------------+
       |     Length (1 octet)              |
       +-----------------------------------+
       | Route Type specific (variable)    |
       +-----------------------------------+

                          Figure 2: BGP EVPN NLRI

   This document defines an additional route type (RT-5) in the IANA
   "EVPN Route Types" registry [EVPNRouteTypes] to be used for the
   advertisement of EVPN routes using IP prefixes:

      Value:  5
      Description:  IP Prefix

   According to Section 5.4 of [RFC7606], a node that doesn't recognize
   the route type 5 (RT-5) will ignore it.  Therefore, an NVE following
   this document can still be attached to a BD where an NVE ignoring RT-
   5s is attached.  Regular procedures described in [RFC7432] would
   apply in that case for both NVEs.  In case two or more NVEs are
   attached to different BDs of the same tenant, they MUST support the
   RT-5 for the proper inter-subnet forwarding operation of the tenant.

   The detailed encoding of this route and associated procedures are
   described in the following sections.

3.1.  IP Prefix Route Encoding

   An IP Prefix route type for IPv4 has the Length field set to 34 and
   consists of the following fields:

       +---------------------------------------+
       |      RD (8 octets)                    |
       +---------------------------------------+
       |Ethernet Segment Identifier (10 octets)|
       +---------------------------------------+
       |  Ethernet Tag ID (4 octets)           |
       +---------------------------------------+
       |  IP Prefix Length (1 octet, 0 to 32)  |
       +---------------------------------------+
       |  IP Prefix (4 octets)                 |
       +---------------------------------------+
       |  GW IP Address (4 octets)             |
       +---------------------------------------+
       |  MPLS Label (3 octets)                |
       +---------------------------------------+

                Figure 3: EVPN IP Prefix Route NLRI for IPv4

   An IP Prefix route type for IPv6 has the Length field set to 58 and
   consists of the following fields:

       +---------------------------------------+
       |      RD (8 octets)                    |
       +---------------------------------------+
       |Ethernet Segment Identifier (10 octets)|
       +---------------------------------------+
       |  Ethernet Tag ID (4 octets)           |
       +---------------------------------------+
       |  IP Prefix Length (1 octet, 0 to 128) |
       +---------------------------------------+
       |  IP Prefix (16 octets)                |
       +---------------------------------------+
       |  GW IP Address (16 octets)            |
       +---------------------------------------+
       |  MPLS Label (3 octets)                |
       +---------------------------------------+

                Figure 4: EVPN IP Prefix Route NLRI for IPv6

   Where:

   *  The Length field of the BGP EVPN NLRI for an EVPN IP Prefix route
      MUST be either 34 (if IPv4 addresses are carried) or 58 (if IPv6
      addresses are carried).  The IP prefix and gateway IP address MUST
      be from the same IP address family.

   *  The Route Distinguisher (RD) and Ethernet Tag ID MUST be used as
      defined in [RFC7432] and [RFC8365].  In particular, the RD is
      unique per MAC-VRF (or IP-VRF).  The MPLS Label field is set to
      either an MPLS label or a VNI, as described in [RFC8365] for other
      EVPN route types.

   *  The Ethernet Segment Identifier MUST be a non-zero 10-octet
      identifier if the ESI is used as an Overlay Index (see the
      definition of "Overlay Index" in Section 3.2).  It MUST be all
      bytes zero otherwise.  The ESI format is described in [RFC7432].

   *  The IP prefix length can be set to a value between 0 and 32 (bits)
      for IPv4 and between 0 and 128 for IPv6, and it specifies the
      number of bits in the prefix.  The value MUST NOT be greater than
      128.

   *  The IP prefix is a 4- or 16-octet field (IPv4 or IPv6).

   *  The GW IP Address field is a 4- or 16-octet field (IPv4 or IPv6)
      and will encode a valid IP address as an Overlay Index for the IP
      prefixes.  The GW IP field MUST be all bytes zero if it is not
      used as an Overlay Index.  Refer to Section 3.2 for the definition
      and use of the Overlay Index.

   *  The MPLS Label field is encoded as 3 octets, where the high-order
      20 bits contain the label value, as per [RFC7432].  When sending,
      the label value SHOULD be zero if a recursive resolution based on
      an Overlay Index is used.  If the received MPLS label value is
      zero, the route MUST contain an Overlay Index, and the ingress
      NVE/PE MUST perform a recursive resolution to find the egress NVE/
      PE.  If the received label is zero and the route does not contain
      an Overlay Index, it MUST be "treat as withdraw" [RFC7606].

   The RD, Ethernet Tag ID, IP prefix length, and IP prefix are part of
   the route key used by BGP to compare routes.  The rest of the fields
   are not part of the route key.

   An IP Prefix route MAY be sent along with an EVPN Router's MAC
   Extended Community (defined in [RFC9135]) to carry the MAC address
   that is used as the Overlay Index.  Note that the MAC address may be
   that of a TS.

   As described in Section 3.2, certain data combinations in a received
   route would imply a treat-as-withdraw handling of the route
   [RFC7606].

3.2.  Overlay Indexes and Recursive Lookup Resolution

   RT-5 routes support recursive lookup resolution through the use of
   Overlay Indexes as follows:

   *  An Overlay Index can be an ESI or IP address in the address space
      of the tenant or MAC address, and it is used by an NVE as the next
      hop for a given IP prefix.  An Overlay Index always needs a
      recursive route resolution on the NVE/PE that installs the RT-5
      into one of its IP-VRFs so that the NVE knows to which egress NVE/
      PE it needs to forward the packets.  It is important to note that
      recursive resolution of the Overlay Index applies upon
      installation into an IP-VRF and not upon BGP propagation (for
      instance, on an ASBR).  Also, as a result of the recursive
      resolution, the egress NVE/PE is not necessarily the same NVE that
      originated the RT-5.

   *  The Overlay Index is indicated along with the RT-5 in the ESI
      field, GW IP field, or EVPN Router's MAC Extended Community,
      depending on whether the IP prefix next hop is an ESI, an IP
      address, or a MAC address in the tenant space.  The Overlay Index
      for a given IP prefix is set by local policy at the NVE that
      originates an RT-5 for that IP prefix (typically managed by the
      cloud management system).

   *  In order to enable the recursive lookup resolution at the ingress
      NVE, an NVE that is a possible egress NVE for a given Overlay
      Index must originate a route advertising itself as the BGP next
      hop on the path to the system denoted by the Overlay Index.  For
      instance:

      -  If an NVE receives an RT-5 that specifies an Overlay Index, the
         NVE cannot use the RT-5 in its IP-VRF unless (or until) it can
         recursively resolve the Overlay Index.

      -  If the RT-5 specifies an ESI as the Overlay Index, a recursive
         resolution can only be done if the NVE has received and
         installed an RT-1 (auto-discovery per EVI) route specifying
         that ESI.

      -  If the RT-5 specifies a GW IP address as the Overlay Index, a
         recursive resolution can only be done if the NVE has received
         and installed an RT-2 (MAC/IP Advertisement route) specifying
         that IP address in the IP Address field of its NLRI.

      -  If the RT-5 specifies a MAC address as the Overlay Index, a
         recursive resolution can only be done if the NVE has received
         and installed an RT-2 (MAC/IP Advertisement route) specifying
         that MAC address in the MAC Address field of its NLRI.

      Note that the RT-1 or RT-2 routes needed for the recursive
      resolution may arrive before or after the given RT-5 route.

   *  Irrespective of the recursive resolution, if there is no IGP or
      BGP route to the BGP next hop of an RT-5, BGP MUST NOT install the
      RT-5 even if the Overlay Index can be resolved.

   *  The ESI and GW IP fields may both be zero at the same time.
      However, they MUST NOT both be non-zero at the same time.  A route
      containing a non-zero GW IP and a non-zero ESI (at the same time)
      SHOULD be treat as withdraw [RFC7606].

   *  If either the ESI or the GW IP are non-zero, then the non-zero one
      is the Overlay Index, regardless of whether the EVPN Router's MAC
      Extended Community is present or the value of the label.  In case
      the GW IP is the Overlay Index (hence, ESI is zero), the EVPN
      Router's MAC Extended Community is ignored if present.

   *  A route where ESI, GW IP, MAC, and Label are all zero at the same
      time SHOULD be treat as withdraw.

   The indirection provided by the Overlay Index and its recursive
   lookup resolution is required to achieve fast convergence in case of
   a failure of the object represented by the Overlay Index (see the
   example described in Section 2.2).

   Table 1 shows the different RT-5 field combinations allowed by this
   specification and what Overlay Index must be used by the receiving
   NVE/PE in each case.  Cases where there is no Overlay Index are
   indicated as "None" in Table 1.  If there is no Overlay Index, the
   receiving NVE/PE will not perform any recursive resolution, and the
   actual next hop is given by the RT-5's BGP next hop.

      +==========+==========+==========+============+===============+
      | ESI      | GW IP    | MAC*     | Label      | Overlay Index |
      +==========+==========+==========+============+===============+
      | Non-Zero | Zero     | Zero     | Don't Care | ESI           |
      +----------+----------+----------+------------+---------------+
      | Non-Zero | Zero     | Non-Zero | Don't Care | ESI           |
      +----------+----------+----------+------------+---------------+
      | Zero     | Non-Zero | Zero     | Don't Care | GW IP         |
      +----------+----------+----------+------------+---------------+
      | Zero     | Zero     | Non-Zero | Zero       | MAC           |
      +----------+----------+----------+------------+---------------+
      | Zero     | Zero     | Non-Zero | Non-Zero   | MAC or None** |
      +----------+----------+----------+------------+---------------+
      | Zero     | Zero     | Zero     | Non-Zero   | None***       |
      +----------+----------+----------+------------+---------------+

              Table 1: RT-5 Fields and Indicated Overlay Index

   Table Notes:

   *     MAC with "Zero" value means no EVPN Router's MAC Extended
         Community is present along with the RT-5.  "Non-Zero" indicates
         that the extended community is present and carries a valid MAC
         address.  The encoding of a MAC address MUST be the 6-octet MAC
         address specified by [IEEE-802.1Q].  Examples of invalid MAC
         addresses are broadcast or multicast MAC addresses.  The route
         MUST be treat as withdraw in case of an invalid MAC address.
         The presence of the EVPN Router's MAC Extended Community alone
         is not enough to indicate the use of the MAC address as the
         Overlay Index since the extended community can be used for
         other purposes.

   **    In this case, the Overlay Index may be the RT-5's MAC address
         or "None", depending on the local policy of the receiving NVE/
         PE.  Note that the advertising NVE/PE that sets the Overlay
         Index SHOULD advertise an RT-2 for the MAC Overlay Index if
         there are receiving NVE/PEs configured to use the MAC as the
         Overlay Index.  This case in Table 1 is used in the IP-VRF-to-
         IP-VRF implementations described in Sections 4.4.1 and 4.4.3.
         The support of a MAC Overlay Index in this model is OPTIONAL.

   ***   The Overlay Index is "None".  This is a special case used for
         IP-VRF-to-IP-VRF where the NVE/PEs are connected by IP NVO
         tunnels as opposed to Ethernet NVO tunnels.

   If the combination of ESI, GW IP, MAC, and Label in the receiving
   RT-5 is different than the combinations shown in Table 1, the router
   will process the route as per the rules described at the beginning of
   this section (Section 3.2).

   Table 2 shows the different inter-subnet use cases described in this
   document and the corresponding coding of the Overlay Index in the
   route type 5 (RT-5).

       +=========+=====================+===========================+
       | Section | Use Case            | Overlay Index in the RT-5 |
       +=========+=====================+===========================+
       | 4.1     | TS IP address       | GW IP                     |
       +---------+---------------------+---------------------------+
       | 4.2     | Floating IP address | GW IP                     |
       +---------+---------------------+---------------------------+
       | 4.3     | "Bump-in-the-wire"  | ESI or MAC                |
       +---------+---------------------+---------------------------+
       | 4.4     | IP-VRF-to-IP-VRF    | GW IP, MAC, or None       |
       +---------+---------------------+---------------------------+

            Table 2: Use Cases and Overlay Indexes for Recursive
                                 Resolution

   The above use cases are representative of the different Overlay
   Indexes supported by the RT-5 (GW IP, ESI, MAC, or None).

4.  Overlay Index Use Cases

   This section describes some use cases for the Overlay Index types
   used with the IP Prefix route.  Although the examples use IPv4
   prefixes and subnets, the descriptions of the RT-5 are valid for the
   same cases with IPv6, except that IP Prefixes, IPL, and GW IP are
   replaced by the corresponding IPv6 values.

4.1.  TS IP Address Overlay Index Use Case

   Figure 5 illustrates an example of inter-subnet forwarding for
   subnets sitting behind VAs (on TS2 and TS3).

   IP4---+           NVE2                            DGW1
         |        +-----------+ +---------+    +-------------+
   SN2---TS2(VA)--|  (BD-10)  |-|         |----|  (BD-10)    |
         | M2/IP2 +-----------+ |         |    |    IRB1\    |
    -+---+                      |         |    |     (IP-VRF)|---+
     |                          |         |    +-------------+  _|_
    SN1                         |  VXLAN/ |                    (   )
     |                          |  GENEVE |         DGW2      ( WAN )
    -+---+           NVE3       |         |    +-------------+ (___)
         | M3/IP3 +-----------+ |         |----|  (BD-10)    |   |
   SN3---TS3(VA)--|  (BD-10)  |-|         |    |    IRB2\    |   |
         |        +-----------+ +---------+    |     (IP-VRF)|---+
   IP5---+                                     +-------------+

                      Figure 5: TS IP Address Use Case

   An example of inter-subnet forwarding between subnet SN1, which uses
   a 24-bit IP prefix (written as SN1/24 in the future), and a subnet
   sitting in the WAN is described below.  NVE2, NVE3, DGW1, and DGW2
   are running BGP EVPN.  TS2 and TS3 do not participate in dynamic
   routing protocols, and they only have a static route to forward the
   traffic to the WAN.  SN1/24 is dual-homed to NVE2 and NVE3.

   In this case, a GW IP is used as an Overlay Index.  Although a
   different Overlay Index type could have been used, this use case
   assumes that the operator knows the VA's IP addresses beforehand,
   whereas the VA's MAC address is unknown and the VA's ESI is zero.
   Because of this, the GW IP is the suitable Overlay Index to be used
   with the RT-5s.  The NVEs know the GW IP to be used for a given
   prefix by policy.

   (1)  NVE2 advertises the following BGP routes on behalf of TS2:

        *  Route type 2 (MAC/IP Advertisement route) containing: ML = 48
           (MAC address length), M = M2 (MAC address), IPL = 32 (IP
           prefix length), IP = IP2, and BGP Encapsulation Extended
           Community [RFC9012] with the corresponding tunnel type.  The
           MAC and IP addresses may be learned via ARP snooping.

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = 0, and GW IP address = IP2.  The prefix and GW IP
           are learned by policy.

   (2)  Similarly, NVE3 advertises the following BGP routes on behalf of
        TS3:

        *  Route type 2 (MAC/IP Advertisement route) containing: ML =
           48, M = M3, IPL = 32, IP = IP3 (and BGP Encapsulation
           Extended Community).

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = 0, and GW IP address = IP3.

   (3)  DGW1 and DGW2 import both received routes based on the Route
        Targets:

        *  Based on the BD-10 Route Target in DGW1 and DGW2, the MAC/IP
           Advertisement route is imported, and M2 is added to the BD-10
           along with its corresponding tunnel information.  For
           instance, if VXLAN is used, the VTEP will be derived from the
           MAC/IP Advertisement route BGP next hop and VNI from the MPLS
           Label1 field.  M2/IP2 is added to the ARP table.  Similarly,
           M3 is added to BD-10, and M3/IP3 is added to the ARP table.

        *  Based on the BD-10 Route Target in DGW1 and DGW2, the IP
           Prefix route is also imported, and SN1/24 is added to the IP-
           VRF with Overlay Index IP2 pointing at the local BD-10.  In
           this example, it is assumed that the RT-5 from NVE2 is
           preferred over the RT-5 from NVE3.  If both routes were
           equally preferable and ECMP enabled, SN1/24 would also be
           added to the routing table with Overlay Index IP3.

   (4)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table, and Overlay Index = IP2 is found.  Since IP2 is an
           Overlay Index, a recursive route resolution is required for
           IP2.

        *  IP2 is resolved to M2 in the ARP table, and M2 is resolved to
           the tunnel information given by the BD FIB (e.g., remote VTEP
           and VNI for the VXLAN case).

        *  The IP packet destined to IPx is encapsulated with:

           -  Inner source MAC = IRB1 MAC.

           -  Inner destination MAC = M2.

           -  Tunnel information provided by the BD (VNI, VTEP IPs, and
              MACs for the VXLAN case).

   (5)  When the packet arrives at NVE2:

        *  Based on the tunnel information (VNI for the VXLAN case), the
           BD-10 context is identified for a MAC lookup.

        *  Encapsulation is stripped off and, based on a MAC lookup
           (assuming MAC forwarding on the egress NVE), the packet is
           forwarded to TS2, where it will be properly routed.

   (6)  Should TS2 move from NVE2 to NVE3, MAC Mobility procedures will
        be applied to the MAC route M2/IP2, as defined in [RFC7432].
        Route type 5 prefixes are not subject to MAC Mobility
        procedures; hence, no changes in the DGW IP-VRF table will occur
        for TS2 mobility -- i.e., all the prefixes will still be
        pointing at IP2 as the Overlay Index.  There is an indirection
        for, e.g., SN1/24, which still points at Overlay Index IP2 in
        the routing table, but IP2 will be simply resolved to a
        different tunnel based on the outcome of the MAC Mobility
        procedures for the MAC/IP Advertisement route M2/IP2.

   Note that in the opposite direction, TS2 will send traffic based on
   its static-route next-hop information (IRB1 and/or IRB2), and regular
   EVPN procedures will be applied.

4.2.  Floating IP Overlay Index Use Case

   Sometimes TSs work in active/standby mode where an upstream floating
   IP owned by the active TS is used as the Overlay Index to get to some
   subnets behind the TS.  This redundancy mode, already introduced in
   Sections 2.1 and 2.2, is illustrated in Figure 6.

                    NVE2                           DGW1
                 +-----------+ +---------+    +-------------+
    +---TS2(VA)--|  (BD-10)  |-|         |----|  (BD-10)    |
    |     M2/IP2 +-----------+ |         |    |    IRB1\    |
    |      <-+                 |         |    |     (IP-VRF)|---+
    |        |                 |         |    +-------------+  _|_
   SN1    vIP23 (floating)     |  VXLAN/ |                    (   )
    |        |                 |  GENEVE |         DGW2      ( WAN )
    |      <-+      NVE3       |         |    +-------------+ (___)
    |     M3/IP3 +-----------+ |         |----|  (BD-10)    |   |
    +---TS3(VA)--|  (BD-10)  |-|         |    |    IRB2\    |   |
                 +-----------+ +---------+    |     (IP-VRF)|---+
                                              +-------------+

            Figure 6: Floating IP Overlay Index for Redundant TS

   In this use case, a GW IP is used as an Overlay Index for the same
   reasons as in Section 4.1.  However, this GW IP is a floating IP that
   belongs to the active TS.  Assuming TS2 is the active TS and owns
   vIP23:

   (1)  NVE2 advertises the following BGP routes for TS2:

        *  Route type 2 (MAC/IP Advertisement route) containing: ML =
           48, M = M2, IPL = 32, and IP = vIP23 (as well as BGP
           Encapsulation Extended Community).  The MAC and IP addresses
           may be learned via ARP snooping.

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = 0, and GW IP address = vIP23.  The prefix and GW
           IP are learned by policy.

   (2)  NVE3 advertises the following BGP route for TS3 (it does not
        advertise an RT-2 for M3/vIP23):

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = 0, and GW IP address = vIP23.  The prefix and GW
           IP are learned by policy.

   (3)  DGW1 and DGW2 import both received routes based on the Route
        Target:

        *  M2 is added to the BD-10 FIB along with its corresponding
           tunnel information.  For the VXLAN use case, the VTEP will be
           derived from the MAC/IP Advertisement route BGP next hop and
           VNI from the VNI field.  M2/vIP23 is added to the ARP table.

        *  SN1/24 is added to the IP-VRF in DGW1 and DGW2 with Overlay
           Index vIP23 pointing at M2 in the local BD-10.

   (4)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table, and Overlay Index = vIP23 is found.  Since vIP23 is an
           Overlay Index, a recursive route resolution for vIP23 is
           required.

        *  vIP23 is resolved to M2 in the ARP table, and M2 is resolved
           to the tunnel information given by the BD (remote VTEP and
           VNI for the VXLAN case).

        *  The IP packet destined to IPx is encapsulated with:

           -  Inner source MAC = IRB1 MAC.

           -  Inner destination MAC = M2.

           -  Tunnel information provided by the BD FIB (VNI, VTEP IPs,
              and MACs for the VXLAN case).

   (5)  When the packet arrives at NVE2:

        *  Based on the tunnel information (VNI for the VXLAN case), the
           BD-10 context is identified for a MAC lookup.

        *  Encapsulation is stripped off and, based on a MAC lookup
           (assuming MAC forwarding on the egress NVE), the packet is
           forwarded to TS2, where it will be properly routed.

   (6)  When the redundancy protocol running between TS2 and TS3
        appoints TS3 as the new active TS for SN1, TS3 will now own the
        floating vIP23 and will signal this new ownership using a
        gratuitous ARP REPLY message (explained in [RFC5227]) or
        similar.  Upon receiving the new owner's notification, NVE3 will
        issue a route type 2 for M3/vIP23, and NVE2 will withdraw the
        RT-2 for M2/vIP23.  DGW1 and DGW2 will update their ARP tables
        with the new MAC resolving the floating IP.  No changes are made
        in the IP-VRF table.

4.3.  Bump-in-the-Wire Use Case

   Figure 7 illustrates an example of inter-subnet forwarding for an IP
   Prefix route that carries subnet SN1.  In this use case, TS2 and TS3
   are Layer 2 VA devices without any IP addresses that can be included
   as an Overlay Index in the GW IP field of the IP Prefix route.  Their
   MAC addresses are M2 and M3, respectively, and are connected to BD-
   10.  Note that IRB1 and IRB2 (in DGW1 and DGW2, respectively) have IP
   addresses in a subnet different than SN1.

                      NVE2                           DGW1
               M2 +-----------+ +---------+    +-------------+
     +---TS2(VA)--|  (BD-10)  |-|         |----|  (BD-10)    |
     |      ESI23 +-----------+ |         |    |    IRB1\    |
     |        +                 |         |    |     (IP-VRF)|---+
     |        |                 |         |    +-------------+  _|_
    SN1       |                 |  VXLAN/ |                    (   )
     |        |                 |  GENEVE |         DGW2      ( WAN )
     |        +      NVE3       |         |    +-------------+ (___)
     |      ESI23 +-----------+ |         |----|  (BD-10)    |   |
     +---TS3(VA)--|  (BD-10)  |-|         |    |    IRB2\    |   |
               M3 +-----------+ +---------+    |     (IP-VRF)|---+
                                               +-------------+

                    Figure 7: Bump-in-the-Wire Use Case

   Since TS2 and TS3 cannot participate in any dynamic routing protocol
   and neither has an IP address assigned, there are two potential
   Overlay Index types that can be used when advertising SN1:

   a)  an ESI, i.e., ESI23, that can be provisioned on the attachment
       ports of NVE2 and NVE3, as shown in Figure 7 or

   b)  the VA's MAC address, which can be added to NVE2 and NVE3 by
       policy.

   The advantage of using an ESI as the Overlay Index as opposed to the
   VA's MAC address is that the forwarding to the egress NVE can be done
   purely based on the state of the AC in the Ethernet segment (notified
   by the Ethernet A-D per EVI route), and all the EVPN multihoming
   redundancy mechanisms can be reused.  For instance, the mass
   withdrawal mechanism described in [RFC7432] for fast failure
   detection and propagation can be used.  It is assumed per this
   section that an ESI Overlay Index is used in this use case, but this
   use case does not preclude the use of the VA's MAC address as an
   Overlay Index.  If a MAC is used as the Overlay Index, the control
   plane must follow the procedures described in Section 4.4.3.

   The model supports VA redundancy in a similar way to the one
   described in Section 4.2 for the floating IP Overlay Index use case,
   except that it uses the EVPN Ethernet A-D per EVI route instead of
   the MAC advertisement route to advertise the location of the Overlay
   Index.  The procedure is explained below:

   (1)  Assuming TS2 is the active TS in ESI23, NVE2 advertises the
        following BGP routes:

        *  Route type 1 (Ethernet A-D route for BD-10) containing: ESI =
           ESI23 and the corresponding tunnel information (VNI field),
           as well as the BGP Encapsulation Extended Community as per
           [RFC8365].

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = ESI23, and GW IP address = 0.  The EVPN Router's
           MAC Extended Community defined in [RFC9135] is added and
           carries the MAC address (M2) associated with the TS behind
           which SN1 sits.  M2 may be learned by policy; however, the
           MAC in the Extended Community is preferred if sent with the
           route.

   (2)  NVE3 advertises the following BGP route for TS3 (no AD per EVI
        route is advertised):

        *  Route type 5 (IP Prefix route) containing: IPL = 24, IP =
           SN1, ESI = 23, and GW IP address = 0.  The EVPN Router's MAC
           Extended Community is added and carries the MAC address (M3)
           associated with the TS behind which SN1 sits.  M3 may be
           learned by policy; however, the MAC in the Extended Community
           is preferred if sent with the route.

   (3)  DGW1 and DGW2 import the received routes based on the Route
        Target:

        *  The tunnel information to get to ESI23 is installed in DGW1
           and DGW2.  For the VXLAN use case, the VTEP will be derived
           from the Ethernet A-D route BGP next hop and VNI from the
           VNI/VSID field (see [RFC8365]).

        *  The RT-5 coming from the NVE that advertised the RT-1 is
           selected, and SN1/24 is added to the IP-VRF in DGW1 and DGW2
           with Overlay Index ESI23 and MAC = M2.

   (4)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table, and Overlay Index = ESI23 is found.  Since ESI23 is an
           Overlay Index, a recursive route resolution is required to
           find the egress NVE where ESI23 resides.

        *  The IP packet destined to IPx is encapsulated with:

           -  Inner source MAC = IRB1 MAC.

           -  Inner destination MAC = M2 (this MAC will be obtained from
              the EVPN Router's MAC Extended Community received along
              with the RT-5 for SN1).  Note that the EVPN Router's MAC
              Extended Community is used in this case to carry the TS's
              MAC address, as opposed to the MAC address of the NVE/PE.

           -  Tunnel information for the NVO tunnel is provided by the
              Ethernet A-D route per EVI for ESI23 (VNI and VTEP IP for
              the VXLAN case).

   (5)  When the packet arrives at NVE2:

        *  Based on the tunnel demultiplexer information (VNI for the
           VXLAN case), the BD-10 context is identified for a MAC lookup
           (assuming a MAC-based disposition model [RFC7432]), or the
           VNI may directly identify the egress interface (for an MPLS-
           based disposition model, which in this context is a VNI-based
           disposition model).

        *  Encapsulation is stripped off and, based on a MAC lookup
           (assuming MAC forwarding on the egress NVE) or a VNI lookup
           (in case of VNI forwarding), the packet is forwarded to TS2,
           where it will be forwarded to SN1.

   (6)  If the redundancy protocol running between TS2 and TS3 follows
        an active/standby model and there is a failure, TS3 is appointed
        as the new active TS for SN1.  TS3 will now own the connectivity
        to SN1 and will signal this new ownership.  Upon receiving the
        new owner's notification, NVE3's AC will become active and issue
        a route type 1 for ESI23, whereas NVE2 will withdraw its
        Ethernet A-D route for ESI23.  DGW1 and DGW2 will update their
        tunnel information to resolve ESI23.  The inner destination MAC
        will be changed to M3.

4.4.  IP-VRF-to-IP-VRF Model

   This use case is similar to the scenario described in Section 9.1 of
   [RFC9135]; however, the new requirement here is the advertisement of
   IP prefixes as opposed to only host routes.

   In the examples described in Sections 4.1, 4.2, and 4.3, the BD
   instance can connect IRB interfaces and any other Tenant Systems
   connected to it.  EVPN provides connectivity for:

   1.  Traffic destined to the IRB or TS IP interfaces, as well as

   2.  Traffic destined to IP subnets sitting behind the TS, e.g., SN1
       or SN2.

   In order to provide connectivity for (1), MAC/IP Advertisement routes
   (RT-2) are needed so that IRB or TS MACs and IPs can be distributed.
   Connectivity type (2) is accomplished by the exchange of IP Prefix
   routes (RT-5) for IPs and subnets sitting behind certain Overlay
   Indexes, e.g., GW IP, ESI, or TS MAC.

   In some cases, IP Prefix routes may be advertised for subnets and IPs
   sitting behind an IRB.  This use case is referred to as the "IP-VRF-
   to-IP-VRF" model.

   [RFC9135] defines an asymmetric IRB model and a symmetric IRB model
   based on the required lookups at the ingress and egress NVE.  The
   asymmetric model requires an IP lookup and a MAC lookup at the
   ingress NVE, whereas only a MAC lookup is needed at the egress NVE;
   the symmetric model requires IP and MAC lookups at both the ingress
   and egress NVE.  From that perspective, the IP-VRF-to-IP-VRF use case
   described in this section is a symmetric IRB model.

   Note that in an IP-VRF-to-IP-VRF scenario, out of the many subnets
   that a tenant may have, it may be the case that only a few are
   attached to a given IP-VRF of the NVE/PE.  In order to provide inter-
   subnet connectivity among the set of NVE/PEs where the tenant is
   connected, a new SBD is created on all of them if a recursive
   resolution is needed.  This SBD is instantiated as a regular BD (with
   no ACs) in each NVE/PE and has an IRB interface that connects the SBD
   to the IP-VRF.  The IRB interface's IP or MAC address is used as the
   Overlay Index for a recursive resolution.

   Depending on the existence and characteristics of the SBD and IRB
   interfaces for the IP-VRFs, there are three different IP-VRF-to-IP-
   VRF scenarios identified and described in this document:

   1.  Interface-less model: no SBD and no Overlay Indexes required.

   2.  Interface-ful with an SBD IRB model: requires SBD as well as GW
       IP addresses as Overlay Indexes.

   3.  Interface-ful with an unnumbered SBD IRB model: requires SBD as
       well as MAC addresses as Overlay Indexes.

   Inter-subnet IP multicast is outside the scope of this document.

4.4.1.  Interface-less IP-VRF-to-IP-VRF Model

   Figure 8 depicts the Interface-less IP-VRF-to-IP-VRF model.

                      NVE1(M1)
             +------------+
     IP1+----|  (BD-1)    |                DGW1(M3)
             |      \     |    +---------+ +--------+
             |    (IP-VRF)|----|         |-|(IP-VRF)|----+
             |      /     |    |         | +--------+    |
         +---|  (BD-2)    |    |         |              _+_
         |   +------------+    |         |             (   )
      SN1|                     |  VXLAN/ |            ( WAN )--H1
         |            NVE2(M2) |  GENEVE/|             (___)
         |   +------------+    |  MPLS   |               +
         +---|  (BD-2)    |    |         | DGW2(M4)      |
             |       \    |    |         | +--------+    |
             |    (IP-VRF)|----|         |-|(IP-VRF)|----+
             |       /    |    +---------+ +--------+
     SN2+----|  (BD-3)    |
             +------------+

              Figure 8: Interface-less IP-VRF-to-IP-VRF Model

   In this case:

   a)  The NVEs and DGWs must provide connectivity between hosts in SN1,
       SN2, and IP1 and hosts sitting at the other end of the WAN -- for
       example, H1.  It is assumed that the DGWs import/export IP and/or
       VPN-IP routes to/from the WAN.

   b)  The IP-VRF instances in the NVE/DGWs are directly connected
       through NVO tunnels, and no IRBs and/or BD instances are
       instantiated to connect the IP-VRFs.

   c)  The solution must provide Layer 3 connectivity among the IP-VRFs
       for Ethernet NVO tunnels -- for instance, VXLAN or GENEVE.

   d)  The solution may provide Layer 3 connectivity among the IP-VRFs
       for IP NVO tunnels -- for example, GENEVE (with IP payload).

   In order to meet the above requirements, the EVPN route type 5 will
   be used to advertise the IP prefixes, along with the EVPN Router's
   MAC Extended Community as defined in [RFC9135] if the advertising
   NVE/DGW uses Ethernet NVO tunnels.  Each NVE/DGW will advertise an
   RT-5 for each of its prefixes with the following fields:

   *  RD as per [RFC7432].

   *  Ethernet Tag ID = 0.

   *  IP prefix length and IP address, as explained in the previous
      sections.

   *  GW IP address = 0.

   *  ESI = 0.

   *  MPLS label or VNI corresponding to the IP-VRF.

   Each RT-5 will be sent with a Route Target identifying the tenant
   (IP-VRF) and may be sent with two BGP extended communities:

   *  The first one is the BGP Encapsulation Extended Community, as per
      [RFC9012], identifying the tunnel type.

   *  The second one is the EVPN Router's MAC Extended Community, as per
      [RFC9135], containing the MAC address associated with the NVE
      advertising the route.  This MAC address identifies the NVE/DGW
      and MAY be reused for all the IP-VRFs in the NVE.  The EVPN
      Router's MAC Extended Community must be sent if the route is
      associated with an Ethernet NVO tunnel -- for instance, VXLAN.  If
      the route is associated with an IP NVO tunnel -- for instance,
      GENEVE with an IP payload -- the EVPN Router's MAC Extended
      Community should not be sent.

   The following example illustrates the procedure to advertise and
   forward packets to SN1/24 (IPv4 prefix advertised from NVE1):

   (1)  NVE1 advertises the following BGP route:

        *  Route type 5 (IP Prefix route) containing:

           -  IPL = 24, IP = SN1, Label = 10.

           -  GW IP = set to 0.

           -  BGP Encapsulation Extended Community [RFC9012].

           -  EVPN Router's MAC Extended Community that contains M1.

           -  Route Target identifying the tenant (IP-VRF).

   (2)  DGW1 imports the received routes from NVE1:

        *  DGW1 installs SN1/24 in the IP-VRF identified by the RT-5
           Route Target.

        *  Since GW IP = ESI = 0, the label is a non-zero value, and the
           local policy indicates this interface-less model, DGW1, will
           use the label and next hop of the RT-5, as well as the MAC
           address conveyed in the EVPN Router's MAC Extended Community
           (as the inner destination MAC address) to set up the
           forwarding state and later encapsulate the routed IP packets.

   (3)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table.  The lookup yields SN1/24.

        *  Since the RT-5 for SN1/24 had a GW IP = ESI = 0, a non-zero
           label, and a next hop, and since the model is interface-less,
           DGW1 will not need a recursive lookup to resolve the route.

        *  The IP packet destined to IPx is encapsulated with: inner
           source MAC = DGW1 MAC, inner destination MAC = M1, outer
           source IP (tunnel source IP) = DGW1 IP, and outer destination
           IP (tunnel destination IP) = NVE1 IP.  The source and inner
           destination MAC addresses are not needed if IP NVO tunnels
           are used.

   (4)  When the packet arrives at NVE1:

        *  NVE1 will identify the IP-VRF for an IP lookup based on the
           label (the inner destination MAC is not needed to identify
           the IP-VRF).

        *  An IP lookup is performed in the routing context, where SN1
           turns out to be a local subnet associated with BD-2.  A
           subsequent lookup in the ARP table and the BD FIB will
           provide the forwarding information for the packet in BD-2.

   The model described above is called an "interface-less" model since
   the IP-VRFs are connected directly through tunnels, and they don't
   require those tunnels to be terminated in SBDs instead, as in
   Sections 4.4.2 or 4.4.3.

4.4.2.  Interface-ful IP-VRF-to-IP-VRF with SBD IRB

   Figure 9 depicts the Interface-ful IP-VRF-to-IP-VRF with SBD IRB
   model.

                    NVE1
           +------------+                       DGW1
   IP10+---+(BD-1)      | +---------------+ +------------+
           |  \         | |               | |            |
           |(IP-VRF)-(SBD)|               |(SBD)-(IP-VRF)|-----+
           |  /    IRB(M1/IP1)           IRB(M3/IP3)     |     |
       +---+(BD-2)      | |               | +------------+    _+_
       |   +------------+ |               |                  (   )
    SN1|                  |     VXLAN/    |                 ( WAN )--H1
       |            NVE2  |     GENEVE/   |                  (___)
       |   +------------+ |     MPLS      |     DGW2           +
       +---+(BD-2)      | |               | +------------+     |
           |  \         | |               | |            |     |
           |(IP-VRF)-(SBD)|               |(SBD)-(IP-VRF)|-----+
           |  /    IRB(M2/IP2)           IRB(M4/IP4)     |
   SN2+----+(BD-3)      | +---------------+ +------------+
           +------------+

                 Figure 9: Interface-ful with SBD IRB Model

   In this model:

   a)  As in Section 4.4.1, the NVEs and DGWs must provide connectivity
       between hosts in SN1, SN2, and IP10 and in hosts sitting at the
       other end of the WAN.

   b)  However, the NVE/DGWs are now connected through Ethernet NVO
       tunnels terminated in the SBD instance.  The IP-VRFs use IRB
       interfaces for their connectivity to the SBD.

   c)  Each SBD IRB has an IP and a MAC address, where the IP address
       must be reachable from other NVEs or DGWs.

   d)  The SBD is attached to all the NVE/DGWs in the tenant domain BDs.

   e)  The solution must provide Layer 3 connectivity for Ethernet NVO
       tunnels -- for instance, VXLAN or GENEVE (with Ethernet payload).

   EVPN type 5 routes will be used to advertise the IP prefixes, whereas
   EVPN RT-2 routes will advertise the MAC/IP addresses of each SBD IRB
   interface.  Each NVE/DGW will advertise an RT-5 for each of its
   prefixes with the following fields:

   *  RD as per [RFC7432].

   *  Ethernet Tag ID = 0.

   *  IP prefix length and IP address, as explained in the previous
      sections.

   *  GW IP address = IRB-IP of the SBD (this is the Overlay Index that
      will be used for the recursive route resolution).

   *  ESI = 0.

   *  Label value should be zero since the RT-5 route requires a
      recursive lookup resolution to an RT-2 route.  It is ignored on
      reception, and the MPLS label or VNI from the RT-2's MPLS Label1
      field is used when forwarding packets.

   Each RT-5 will be sent with a Route Target identifying the tenant
   (IP-VRF).  The EVPN Router's MAC Extended Community should not be
   sent in this case.

   The following example illustrates the procedure to advertise and
   forward packets to SN1/24 (IPv4 prefix advertised from NVE1):

   (1)  NVE1 advertises the following BGP routes:

        *  Route type 5 (IP Prefix route) containing:

           -  IPL = 24, IP = SN1, Label = SHOULD be set to 0.

           -  GW IP = IP1 (SBD IRB's IP).

           -  Route Target identifying the tenant (IP-VRF).

        *  Route type 2 (MAC/IP Advertisement route for the SBD IRB)
           containing:

           -  ML = 48, M = M1, IPL = 32, IP = IP1, Label = 10.

           -  A BGP Encapsulation Extended Community [RFC9012].

           -  Route Target identifying the SBD.  This Route Target may
              be the same as the one used with the RT-5.

   (2)  DGW1 imports the received routes from NVE1:

        *  DGW1 installs SN1/24 in the IP-VRF identified by the RT-5
           Route Target.

           -  Since GW IP is different from zero, the GW IP (IP1) will
              be used as the Overlay Index for the recursive route
              resolution to the RT-2 carrying IP1.

   (3)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table.  The lookup yields SN1/24, which is associated with
           the Overlay Index IP1.  The forwarding information is derived
           from the RT-2 received for IP1.

        *  The IP packet destined to IPx is encapsulated with: inner
           source MAC = M3, inner destination MAC = M1, outer source IP
           (source VTEP) = DGW1 IP, and outer destination IP
           (destination VTEP) = NVE1 IP.

   (4)  When the packet arrives at NVE1:

        *  NVE1 will identify the IP-VRF for an IP lookup based on the
           label and the inner MAC DA.

        *  An IP lookup is performed in the routing context, where SN1
           turns out to be a local subnet associated with BD-2.  A
           subsequent lookup in the ARP table and the BD FIB will
           provide the forwarding information for the packet in BD-2.

   The model described above is called an "interface-ful with SBD IRB"
   model because the tunnels connecting the DGWs and NVEs need to be
   terminated into the SBD.  The SBD is connected to the IP-VRFs via SBD
   IRB interfaces, and that allows the recursive resolution of RT-5s to
   GW IP addresses.

4.4.3.  Interface-ful IP-VRF-to-IP-VRF with Unnumbered SBD IRB

   Figure 10 depicts the Interface-ful IP-VRF-to-IP-VRF with unnumbered
   SBD IRB model.  Note that this model is similar to the one described
   in Section 4.4.2, only without IP addresses on the SBD IRB
   interfaces.

                    NVE1
           +------------+                       DGW1
   IP1+----+(BD-1)      | +---------------+ +------------+
           |  \         | |               | |            |
           |(IP-VRF)-(SBD)|               (SBD)-(IP-VRF) |-----+
           |  /    IRB(M1)|               | IRB(M3)      |     |
       +---+(BD-2)      | |               | +------------+    _+_
       |   +------------+ |               |                  (   )
    SN1|                  |     VXLAN/    |                 ( WAN )--H1
       |            NVE2  |     GENEVE/   |                  (___)
       |   +------------+ |     MPLS      |     DGW2           +
       +---+(BD-2)      | |               | +------------+     |
           |  \         | |               | |            |     |
           |(IP-VRF)-(SBD)|               (SBD)-(IP-VRF) |-----+
           |  /    IRB(M2)|               | IRB(M4)      |
   SN2+----+(BD-3)      | +---------------+ +------------+
           +------------+

           Figure 10: Interface-ful with Unnumbered SBD IRB Model

   In this model:

   a)  As in Sections 4.4.1 and 4.4.2, the NVEs and DGWs must provide
       connectivity between hosts in SN1, SN2, and IP1 and in hosts
       sitting at the other end of the WAN.

   b)  As in Section 4.4.2, the NVE/DGWs are connected through Ethernet
       NVO tunnels terminated in the SBD instance.  The IP-VRFs use IRB
       interfaces for their connectivity to the SBD.

   c)  However, each SBD IRB has a MAC address only and no IP address
       (which is why the model refers to an "unnumbered" SBD IRB).  In
       this model, there is no need to have IP reachability to the SBD
       IRB interfaces themselves, and there is a requirement to limit
       the number of IP addresses used.

   d)  As in Section 4.4.2, the SBD is composed of all the NVE/DGW BDs
       of the tenant that need inter-subnet forwarding.

   e)  As in Section 4.4.2, the solution must provide Layer 3
       connectivity for Ethernet NVO tunnels -- for instance, VXLAN or
       GENEVE (with Ethernet payload).

   This model will also make use of the RT-5 recursive resolution.  EVPN
   type 5 routes will advertise the IP prefixes along with the EVPN
   Router's MAC Extended Community used for the recursive lookup,
   whereas EVPN RT-2 routes will advertise the MAC addresses of each SBD
   IRB interface (this time without an IP).

   Each NVE/DGW will advertise an RT-5 for each of its prefixes with the
   same fields as described in Section 4.4.2, except:

   *  GW IP address = set to 0.

   Each RT-5 will be sent with a Route Target identifying the tenant
   (IP-VRF) and the EVPN Router's MAC Extended Community containing the
   MAC address associated with the SBD IRB interface.  This MAC address
   may be reused for all the IP-VRFs in the NVE.

   The example is similar to the one in Section 4.4.2:

   (1)  NVE1 advertises the following BGP routes:

        *  Route type 5 (IP Prefix route) containing the same values as
           in the example in Section 4.4.2, except:

           -  GW IP = SHOULD be set to 0.

           -  EVPN Router's MAC Extended Community containing M1 (this
              will be used for the recursive lookup to an RT-2).

        *  Route type 2 (MAC route for the SBD IRB) with the same values
           as in Section 4.4.2, except:

           -  ML = 48, M = M1, IPL = 0, Label = 10.

   (2)  DGW1 imports the received routes from NVE1:

        *  DGW1 installs SN1/24 in the IP-VRF identified by the RT-5
           Route Target.

           -  The MAC contained in the EVPN Router's MAC Extended
              Community sent along with the RT-5 (M1) will be used as
              the Overlay Index for the recursive route resolution to
              the RT-2 carrying M1.

   (3)  When DGW1 receives a packet from the WAN with destination IPx,
        where IPx belongs to SN1/24:

        *  A destination IP lookup is performed on the DGW1 IP-VRF
           table.  The lookup yields SN1/24, which is associated with
           the Overlay Index M1.  The forwarding information is derived
           from the RT-2 received for M1.

        *  The IP packet destined to IPx is encapsulated with: inner
           source MAC = M3, inner destination MAC = M1, outer source IP
           (source VTEP) = DGW1 IP, and outer destination IP
           (destination VTEP) = NVE1 IP.

   (4)  When the packet arrives at NVE1:

        *  NVE1 will identify the IP-VRF for an IP lookup based on the
           label and the inner MAC DA.

        *  An IP lookup is performed in the routing context, where SN1
           turns out to be a local subnet associated with BD-2.  A
           subsequent lookup in the ARP table and the BD FIB will
           provide the forwarding information for the packet in BD-2.

   The model described above is called an "interface-ful with unnumbered
   SBD IRB" model (as in Section 4.4.2) but without the SBD IRB having
   an IP address.

5.  Security Considerations

   This document provides a set of procedures to achieve inter-subnet
   forwarding across NVEs or PEs attached to a group of BDs that belong
   to the same tenant (or VPN).  The security considerations discussed
   in [RFC7432] apply to the intra-subnet forwarding or communication
   within each of those BDs.  In addition, the security considerations
   in [RFC4364] should also be understood, since this document and
   [RFC4364] may be used in similar applications.

   Contrary to [RFC4364], this document does not describe PE/CE route
   distribution techniques but rather considers the CEs as TSs or VAs
   that do not run dynamic routing protocols.  This can be considered a
   security advantage, since dynamic routing protocols can be blocked on
   the NVE/PE ACs, not allowing the tenant to interact with the
   infrastructure's dynamic routing protocols.

   In this document, the RT-5 may use a regular BGP next hop for its
   resolution or an Overlay Index that requires a recursive resolution
   to a different EVPN route (an RT-2 or an RT-1).  In the latter case,
   it is worth noting that any action that ends up filtering or
   modifying the RT-2 or RT-1 routes used to convey the Overlay Indexes
   will modify the resolution of the RT-5 and therefore the forwarding
   of packets to the remote subnet.

6.  IANA Considerations

   IANA has registered value 5 in the "EVPN Route Types" registry
   [EVPNRouteTypes] defined by [RFC7432] as follows:

                    +=======+=============+===========+
                    | Value | Description | Reference |
                    +=======+=============+===========+
                    | 5     | IP Prefix   | RFC 9136  |
                    +-------+-------------+-----------+

                                  Table 3

7.  References

7.1.  Normative References

   [EVPNRouteTypes]
              IANA, "EVPN Route Types",
              <https://www.iana.org/assignments/evpn>.

   [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>.

   [RFC7432]  Sajassi, A., Ed., Aggarwal, R., Bitar, N., Isaac, A.,
              Uttaro, J., Drake, J., and W. Henderickx, "BGP MPLS-Based
              Ethernet VPN", RFC 7432, DOI 10.17487/RFC7432, February
              2015, <https://www.rfc-editor.org/info/rfc7432>.

   [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>.

   [RFC8365]  Sajassi, A., Ed., Drake, J., Ed., Bitar, N., Shekhar, R.,
              Uttaro, J., and W. Henderickx, "A Network Virtualization
              Overlay Solution Using Ethernet VPN (EVPN)", RFC 8365,
              DOI 10.17487/RFC8365, March 2018,
              <https://www.rfc-editor.org/info/rfc8365>.

   [RFC9012]  Patel, K., Van de Velde, G., Sangli, S., and J. Scudder,
              "The BGP Tunnel Encapsulation Attribute", RFC 9012,
              DOI 10.17487/RFC9012, April 2021,
              <https://www.rfc-editor.org/info/rfc9012>.

   [RFC9135]  Sajassi, A., Salam, S., Thoria, S., Drake, J., and J.
              Rabadan, "Integrated Routing and Bridging in Ethernet VPN
              (EVPN)", RFC 9135, DOI 10.17487/RFC9135, October 2021,
              <https://www.rfc-editor.org/info/rfc9135>.

7.2.  Informative References

   [IEEE-802.1Q]
              IEEE, "IEEE Standard for Local and Metropolitan Area
              Networks -- Bridges and Bridged Networks",
              DOI 10.1109/IEEESTD.2018.8403927, IEEE Std 802.1Q, July
              2018,
              <https://standards.ieee.org/standard/802_1Q-2018.html>.

   [RFC4364]  Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private
              Networks (VPNs)", RFC 4364, DOI 10.17487/RFC4364, February
              2006, <https://www.rfc-editor.org/info/rfc4364>.

   [RFC5227]  Cheshire, S., "IPv4 Address Conflict Detection", RFC 5227,
              DOI 10.17487/RFC5227, July 2008,
              <https://www.rfc-editor.org/info/rfc5227>.

   [RFC5798]  Nadas, S., Ed., "Virtual Router Redundancy Protocol (VRRP)
              Version 3 for IPv4 and IPv6", RFC 5798,
              DOI 10.17487/RFC5798, March 2010,
              <https://www.rfc-editor.org/info/rfc5798>.

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

   [RFC7365]  Lasserre, M., Balus, F., Morin, T., Bitar, N., and Y.
              Rekhter, "Framework for Data Center (DC) Network
              Virtualization", RFC 7365, DOI 10.17487/RFC7365, October
              2014, <https://www.rfc-editor.org/info/rfc7365>.

   [RFC7606]  Chen, E., Ed., Scudder, J., Ed., Mohapatra, P., and K.
              Patel, "Revised Error Handling for BGP UPDATE Messages",
              RFC 7606, DOI 10.17487/RFC7606, August 2015,
              <https://www.rfc-editor.org/info/rfc7606>.

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

Acknowledgments

   The authors would like to thank Mukul Katiyar, Jeffrey Zhang, and
   Alex Nichol for their valuable feedback and contributions.  Tony
   Przygienda and Thomas Morin also helped improve this document with
   their feedback.  Special thanks to Eric Rosen for his detailed
   review, which really helped improve the readability and clarify the
   concepts.  We also thank Alvaro Retana for his thorough review.

Contributors

   In addition to the authors listed on the front page, the following
   coauthors have also contributed to this document:

      Senthil Sathappan
      Florin Balus
      Aldrin Isaac
      Senad Palislamovic
      Samir Thoria

Authors' Addresses

   Jorge Rabadan (editor)
   Nokia
   777 E. Middlefield Road
   Mountain View, CA 94043
   United States of America

   Email: jorge.rabadan@nokia.com


   Wim Henderickx
   Nokia

   Email: wim.henderickx@nokia.com


   John Drake
   Juniper

   Email: jdrake@juniper.net


   Wen Lin
   Juniper

   Email: wlin@juniper.net


   Ali Sajassi
   Cisco

   Email: sajassi@cisco.com