RFC 9135 | IRB EVPN | October 2021 |
Sajassi, et al. | Standards Track | [Page] |
Ethernet VPN (EVPN) provides an extensible and flexible multihoming VPN solution over an MPLS/IP network for intra-subnet connectivity among Tenant Systems and end devices that can be physical or virtual. However, there are scenarios for which there is a need for a dynamic and efficient inter-subnet connectivity among these Tenant Systems and end devices while maintaining the multihoming capabilities of EVPN. This document describes an Integrated Routing and Bridging (IRB) solution based on EVPN to address such requirements.¶
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/rfc9135.¶
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.¶
EVPN [RFC7432] provides an extensible and flexible multihoming VPN solution over an MPLS/IP network for intra-subnet connectivity among Tenant Systems (TSs) and end devices that can be physical or virtual, where an IP subnet is represented by an EVPN instance (EVI) for a VLAN-based service or by an (EVI, VLAN) association for a VLAN-aware bundle service. However, there are scenarios for which there is a need for a dynamic and efficient inter-subnet connectivity among these Tenant Systems and end devices while maintaining the multihoming capabilities of EVPN. This document describes an Integrated Routing and Bridging (IRB) solution based on EVPN to address such requirements.¶
Inter-subnet communication is typically performed by centralized Layer 3 (L3) gateway (GW) devices, which enforce all inter-subnet communication policies and perform all inter-subnet forwarding. When two TSs belonging to two different subnets connected to the same Provider Edge (PE) wanted to communicate with each other, their traffic needed to be backhauled from the PE all the way to the centralized gateway where inter-subnet switching is performed and then sent back to the PE. For today's large multi-tenant Data Center (DC), this scheme is very inefficient and sometimes impractical.¶
In order to overcome the drawback of the centralized L3 GW approach, IRB functionality is needed on the PEs (also referred to as EVPN Network Virtualization Edges (NVEs)) attached to TSs in order to avoid inefficient forwarding of tenant traffic (i.e., avoid backhauling and hair pinning). When a PE with IRB capability receives tenant traffic over an Attachment Circuit (AC), it cannot only locally bridge the tenant intra-subnet traffic but also locally route the tenant inter-subnet traffic on a packet-by-packet basis, thus meeting the requirements for both intra- and inter-subnet forwarding and avoiding non-optimal traffic forwarding associated with a centralized L3 GW approach.¶
Some TSs run non-IP protocols in conjunction with their IP traffic. Therefore, it is important to handle both kinds of traffic optimally -- e.g., to bridge non-IP and intra-subnet traffic and to route inter-subnet IP traffic. Therefore, the solution needs to meet the following requirements:¶
This document also assumes familiarity with the terminology of [RFC7365], [RFC7432], and [RFC8365].¶
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.¶
Since this document discusses IRB operation in relationship to EVPN MAC-VRF, IP-VRF, EVI, BD, bridge table, and IRB interfaces, it is important to understand the relationship between these components. Therefore, the PE model is illustrated below to a) describe these components and b) illustrate the relationship among them.¶
A tenant needing IRB services on a PE requires an IP-VRF table along with one or more MAC-VRF tables. An IP-VRF, as defined in [RFC4364], is the instantiation of an IP-VPN instance in a PE. A MAC-VRF, as defined in [RFC7432], is the instantiation of an EVI in a PE. A MAC-VRF consists of one or more bridge tables, where each bridge table corresponds to a VLAN (broadcast domain). If service interfaces for an EVPN PE are configured in VLAN-based mode (i.e., Section 6.1 of [RFC7432]), then there is only a single bridge table per MAC-VRF (per EVI) -- i.e., there is only one tenant VLAN per EVI. However, if service interfaces for an EVPN PE are configured in VLAN-aware bundle mode (i.e., Section 6.3 of [RFC7432]), then there are several bridge tables per MAC-VRF (per EVI) -- i.e., there are several tenant VLANs per EVI.¶
Each bridge table is connected to an IP-VRF via an L3 interface called an "IRB interface". Since a single tenant subnet is typically (and in this document) represented by a VLAN (and thus supported by a single bridge table), for a given tenant, there are as many bridge tables as there are subnets. Thus, there are also as many IRB interfaces between the tenant IP-VRF and the associated bridge tables as shown in the PE model above.¶
IP-VRF is identified by its corresponding Route Target and Route Distinguisher, and MAC-VRF is also identified by its corresponding Route Target and Route Distinguisher. If operating in EVPN VLAN-based mode, then a receiving PE that receives an EVPN route with a MAC-VRF Route Target can identify the corresponding bridge table; however, if operating in EVPN VLAN-aware bundle mode, then the receiving PE needs both the MAC-VRF Route Target and VLAN ID in order to identify the corresponding bridge table.¶
This document defines and describes two types of IRB solutions -- namely, symmetric and asymmetric IRB. The description of symmetric and asymmetric IRB procedures relating to data path operations and tables in this document is a logical view of data path lookups and related tables. Actual implementations, while following this logical view, may not strictly adhere to it for performance trade-offs. Specifically,¶
In symmetric IRB, as its name implies, the lookup operation is symmetric at both the ingress and egress PEs -- i.e., both ingress and egress PEs perform lookups on both MAC and IP addresses. The ingress PE performs a MAC lookup followed by an IP lookup, and the egress PE performs an IP lookup followed by a MAC lookup, as depicted in the following figure.¶
In symmetric IRB, as shown in Figure 2, the inter-subnet forwarding between two PEs is done between their associated IP-VRFs. Therefore, the tunnel connecting these IP-VRFs can be either an IP-only tunnel (e.g., in the case of MPLS or GPE encapsulation) or an Ethernet NVO tunnel (e.g., in the case of VXLAN encapsulation). If it is an Ethernet NVO tunnel, the TS1's IP packet is encapsulated in an Ethernet header consisting of ingress and egress PE MAC addresses -- i.e., there is no need for the ingress PE to use the destination TS2's MAC address. Therefore, in symmetric IRB, there is no need for the ingress PE to maintain ARP entries for the association of the destination TS2's IP and MAC addresses in its ARP table. Each PE participating in symmetric IRB only maintains ARP entries for locally connected hosts and MAC-VRFs/BTs for only locally configured subnets.¶
In asymmetric IRB, the lookup operation is asymmetric and the ingress PE performs three lookups, whereas the egress PE performs a single lookup -- i.e., the ingress PE performs a MAC lookup, followed by an IP lookup, followed by a MAC lookup again. The egress PE performs just a single MAC lookup as depicted in Figure 3 below.¶
In asymmetric IRB, as shown in Figure 3, the inter-subnet forwarding between two PEs is done between their associated MAC-VRFs/BTs. Therefore, the MPLS or NVO tunnel used for inter-subnet forwarding MUST be of type Ethernet. Since only MAC lookup is performed at the egress PE (e.g., no IP lookup), the TS1's IP packets need to be encapsulated with the destination TS2's MAC address. In order for the ingress PE to perform such encapsulation, it needs to maintain TS2's IP and MAC address association in its ARP table. Furthermore, it needs to maintain destination TS2's MAC address in the corresponding bridge table even though it may not have any TSs of the corresponding subnet locally attached. In other words, each PE participating in asymmetric IRB MUST maintain ARP entries for remote hosts (hosts connected to other PEs) as well as maintain MAC-VRFs/BTs and IRB interfaces for ALL subnets in an IP-VRF, including subnets that may not be locally attached. Therefore, careful consideration of the PE scale aspects for its ARP table size, its IRB interfaces, and the number and size of its bridge tables should be given for the application of asymmetric IRB.¶
It should be noted that whenever a PE performs a host IP lookup for a packet that is routed, the IPv4 Time To Live (TTL) or IPv6 hop limit for that packet is decremented by one, and if it reaches zero, the packet is discarded. In the case of symmetric IRB, the TTL / hop limit is decremented by both ingress and egress PEs (once by each), whereas in the case of asymmetric IRB, the TTL / hop limit is decremented only once by the ingress PE.¶
The following sections define the control and data plane procedures for symmetric and asymmetric IRB on ingress and egress PEs. The following figure is used to describe these procedures, showing a single IP-VRF and a number of BDs on each PE for a given tenant. That is, an IP-VRF connects one or more EVIs, and each EVI contains one MAC-VRF; each MAC VRF consists of one or more bridge tables, one per BD; and a PE has an associated IRB interface for each BD.¶
To support inter-subnet forwarding on a PE, the PE acts as an IP default gateway from the perspective of the attached Tenant Systems where default gateway MAC and IP addresses are configured on each IRB interface associated with its subnet and fall into one of the following two options:¶
It is worth noting that if the applications that are running on the TSs are employing or relying on any form of MAC security, then the first option (i.e., using an anycast MAC address) should be used to ensure that the applications receive traffic from the same IRB interface MAC address to which they are sending. If the second option is used, then the IRB interface MAC address MUST be the one used in the initial ARP reply or ND Neighbor Advertisement (NA) for that TS.¶
Although both of these options are applicable to both symmetric and asymmetric IRB, option 1 is recommended because of the ease of anycast MAC address provisioning on not only the IRB interface associated with a given subnet across all the PEs corresponding to that VLAN but also on all IRB interfaces associated with all the tenant's subnets across all the PEs corresponding to all the VLANs for that tenant. Furthermore, it simplifies the operation as there is no need for Default Gateway extended community advertisement and its associated MAC aliasing procedure. Yet another advantage is that following host mobility, the host does not need to refresh the default GW ARP/ND entry.¶
If option 1 is used, an implementation MAY choose to auto-derive the anycast MAC address. If auto-derivation is used, the anycast MAC MUST be auto-derived out of the following ranges (which are defined in [RFC5798]):¶
Where the last octet is generated based on a configurable Virtual Router ID (VRID) (range 1-255). If not explicitly configured, the default value for the VRID octet is '1'. Auto-derivation of the anycast MAC can only be used if there is certainty that the auto-derived MAC does not collide with any customer MAC address.¶
In addition to IP anycast addresses, IRB interfaces can be configured with non-anycast IP addresses for the purpose of OAM (such as sending a traceroute/ping to these interfaces) for both symmetric and asymmetric IRB. These IP addresses need to be distributed as VPN routes when PEs operate in symmetric IRB mode. However, they don't need to be distributed if the PEs are operating in asymmetric IRB mode as the non-anycast IP addresses are configured along with their individual MACs, and they get distributed via the EVPN route type 2 advertisement.¶
For option 1 -- irrespective of whether only the anycast MAC address or both anycast and non-anycast MAC addresses (where the latter one is used for the purpose of OAM) are used on the same IRB -- when a TS sends an ARP request or ND Neighbor Solicitation (NS) to the PE to which it is attached, the request is sent for the anycast IP address of the IRB interface associated with the TS's subnet. The reply will use an anycast MAC address (in both the source MAC in the Ethernet header and sender hardware address in the payload). For example, in Figure 4, TS1 is configured with the anycast IPx address as its default gateway IP address; thus, when it sends an ARP request for IPx (anycast IP address of the IRB interface for BT1), the PE1 sends an ARP reply with the MACx, which is the anycast MAC address of that IRB interface. Traffic routed from IP-VRF1 to TS1 uses the anycast MAC address as the source MAC address.¶
Symmetric and asymmetric IRB modes may coexist in the same network, and an ingress PE that supports both forwarding modes for a given tenant can interwork with egress PEs that support either IRB mode. The egress PE will indicate the desired forwarding mode for a given host based on the presence of the Label2 field and the IP-VRF Route Target in the EVPN MAC/IP Advertisement route. If the Label2 field of the received MAC/IP Advertisement route for host H1 is non-zero, and one of its Route Targets identifies the IP-VRF, the ingress PE will use symmetric IRB mode when forwarding packets destined to H1. If the Label2 field is zero and the MAC/IP Advertisement route for H1 does not carry any Route Target that identifies the IP-VRF, the ingress PE will use asymmetric mode when forwarding traffic to H1.¶
As an example that illustrates the previous statement, suppose PE1 and PE2 need to forward packets from TS2 to TS4 in Figure 4. Since both PEs are attached to the bridge table of the destination host, symmetric and asymmetric IRB modes are both possible as long as the ingress PE, PE1, supports both modes. The forwarding mode will depend on the mode configured in the egress PE, PE2. That is:¶
Refer to [EVPN] for more information about interoperability between symmetric and asymmetric forwarding modes.¶
The choice between symmetric or asymmetric mode is based on the operator's preference, and it is a trade-off between scale (which is better in the symmetric IRB mode) and control plane simplicity (asymmetric IRB mode simplifies the control plane). In cases where a tenant has hosts for every subnet attached to all (or most of) the PEs, the ARP and MAC entries need to be learned by all PEs anyway; therefore, the asymmetric IRB mode simplifies the forwarding model and saves space in the IP-VRF route table, since host routes are not installed in the route table. However, if the tenant does not need to stretch subnets (broadcast domains) to multiple PEs and inter-subnet forwarding is needed, the symmetric IRB model will save ARP and bridge table space in all the PEs (in comparison with the asymmetric IRB model).¶
When a PE (e.g., PE1 in Figure 4 above) learns the MAC and IP address of a TS (e.g., via an ARP request or Neighbor Solicitation), it adds the MAC address to the corresponding MAC-VRF/BT of that tenant's subnet and adds the IP address to the IP-VRF for that tenant. Furthermore, it adds this TS's MAC and IP address association to its ARP table or Neighbor Discovery Protocol (NDP) cache. It then builds an EVPN MAC/IP Advertisement route (type 2) as follows and advertises it to other PEs participating in that tenant's VPN.¶
Just as in [RFC7432], the RD, Ethernet Tag ID, MAC Address Length, MAC Address, IP Address Length, and IP Address fields are part of the route key used by BGP to compare routes. The rest of the fields are not part of the route key.¶
This route is advertised along with the following two extended communities:¶
This route is advertised with one or more Encapsulation Extended Communities [RFC9012], one for each encapsulation type supported by the advertising PE. If one or more encapsulation types require an Ethernet frame, a single EVPN Router's MAC Extended Community (Section 8.1) is also advertised. This extended community specifies the MAC address to be used as the inner destination MAC address in an Ethernet frame sent to the advertising PE.¶
This route MUST be advertised with two Route Targets, one corresponding to the MAC-VRF of the tenant's subnet and another corresponding to the tenant's IP-VRF.¶
When a PE (e.g., PE2 in Figure 4 above) receives this EVPN MAC/IP Advertisement route, it performs the following:¶
If the MPLS Label2 field is non-zero, it means that this route is to be used for symmetric IRB, and the MPLS label2 value is to be used when sending a packet for this IP address to the advertising PE.¶
If the receiving PE supports asymmetric IRB mode and receives this route with both the MAC-VRF and IP-VRF Route Targets but the MAC/IP Advertisement route does not include the MPLS Label2 field, then the receiving PE installs the MAC address in the corresponding MAC-VRF and the (IP, MAC) association in the ARP table for that tenant (identified by the corresponding IP-VRF Route Target).¶
If the receiving PE receives this route with both the MAC-VRF and IP-VRF Route Targets, and if the receiving PE does not support either asymmetric or symmetric IRB modes but has the corresponding MAC-VRF, then it only imports the MAC address.¶
If the receiving PE receives this route with both the MAC-VRF and IP-VRF Route Targets and the MAC/IP Advertisement route includes the MPLS Label2 field but the receiving PE only supports asymmetric IRB mode, then the receiving PE MUST ignore the MPLS Label2 field and install the MAC address in the corresponding MAC-VRF and (IP, MAC) association in the ARP table for that tenant (identified by the corresponding IP-VRF Route Target).¶
In the case of symmetric IRB, a Layer 3 subnet and IRB interface corresponding to a MAC-VRF/BT are required to be provisioned at a PE only if that PE has locally attached hosts in that subnet. In order to enable inter-subnet routing across PEs in a deployment where not all subnets are provisioned at all PEs participating in an EVPN IRB instance, PEs MUST advertise local subnet routes as EVPN RT-5. These subnet routes are required for bootstrapping host (IP, MAC) learning using gleaning procedures initiated by an inter-subnet data packet.¶
That is, if a given host's (IP, MAC) association is unknown, and an ingress PE needs to send a packet to that host, then that ingress PE needs to know which egress PEs are attached to the subnet in which the host resides in order to send the packet to one of those PEs, causing the PE receiving the packet to probe for that host. For example, consider a subnet A that is locally attached to PE1 and subnet B that is locally attached to PE2 and PE3. Host A in subnet A, which is attached to PE1, initiates a data packet destined to host B in subnet B, which is attached to PE3. If host B's (IP, MAC) has not yet been learned via either a gratuitous ARP OR a prior gleaning procedure, a new gleaning procedure MUST be triggered for host B's (IP, MAC) to be learned and advertised across the EVPN network. Since host B's subnet is not local to PE1, an IP lookup for host B at PE1 will not trigger this gleaning procedure for host B's (IP, MAC). Therefore, PE1 MUST learn subnet B's prefix route via EVPN RT-5 advertised from PE2 and PE3, so it can route the packet to one of the PEs that have subnet B locally attached. Once the packet is received at PE2 OR PE3, and the route lookup yields a glean result, an ARP request is triggered and flooded across the Layer 2 overlay. This ARP request would be received and replied to by host B, resulting in host B (IP, MAC) learning at PE3 and its advertisement across the EVPN network. Packets from host A to host B can now be routed directly from PE1 to PE3. Advertisement of local subnet EVPN RT-5 for an IP-VRF MAY typically be achieved via provisioning-connected route redistribution to BGP.¶
When an Ethernet frame is received by an ingress PE (e.g., PE1 in Figure 4 above), the PE uses the AC ID (e.g., VLAN ID) to identify the associated MAC-VRF/BT, and it performs a lookup on the destination MAC address. If the MAC address corresponds to its IRB interface MAC address, the ingress PE deduces that the packet must be inter-subnet routed. Hence, the ingress PE performs an IP lookup in the associated IP-VRF table. The lookup identifies the BGP next hop of the egress PE along with the tunnel/encapsulation type and the associated MPLS/VNI values. The ingress PE also decrements the TTL / hop limit for that packet by one, and if it reaches zero, the ingress PE discards the packet.¶
If the tunnel type is that of an MPLS or IP-only NVO tunnel, then the TS's IP packet is sent over the tunnel without any Ethernet header. However, if the tunnel type is that of an Ethernet NVO tunnel, then an Ethernet header needs to be added to the TS's IP packet. The source MAC address of this inner Ethernet header is set to the ingress PE's router MAC address, and the destination MAC address of this inner Ethernet header is set to the egress PE's router MAC address learned via the EVPN Router's MAC Extended Community attached to the route. The MPLS VPN label is set to the received label2 in the route. In the case of the Ethernet NVO tunnel type, the VNI may be set one of two ways:¶
PEs may be configured to operate in one of these two modes depending on the administrative domain boundaries across PEs participating in the NVO and the PE's capability to support downstream VNI mode.¶
In the case of NVO tunnel encapsulation, the outer source and destination IP addresses are set to the ingress and egress PE BGP next-hop IP addresses, respectively.¶
When the tenant's MPLS or NVO encapsulated packet is received over an MPLS or NVO tunnel by the egress PE, the egress PE removes the NVO tunnel encapsulation and uses the VPN MPLS label (for MPLS encapsulation) or VNI (for NVO encapsulation) to identify the IP-VRF in which IP lookup needs to be performed. If the VPN MPLS label or VNI identifies a MAC-VRF instead of an IP-VRF, then the procedures in Section 6.4 for asymmetric IRB are executed.¶
The lookup in the IP-VRF identifies a local adjacency to the IRB interface associated with the egress subnet's MAC-VRF/BT. The egress PE also decrements the TTL / hop limit for that packet by one, and if it reaches zero, the egress PE discards the packet.¶
The egress PE gets the destination TS's MAC address for that TS's IP address from its ARP table or NDP cache. It encapsulates the packet with that destination MAC address and a source MAC address corresponding to that IRB interface and sends the packet to its destination subnet MAC-VRF/BT.¶
The destination MAC address lookup in the MAC-VRF/BT results in the local adjacency (e.g., local interface) over which the Ethernet frame is sent.¶
When a PE (e.g., PE1 in Figure 4 above) learns the MAC and IP address of an attached TS (e.g., via an ARP request or ND Neighbor Solicitation), it populates its MAC-VRF/BT, IP-VRF, and ARP table or NDP cache just as in the case for symmetric IRB. It then builds an EVPN MAC/IP Advertisement route (type 2) as follows and advertises it to other PEs participating in that tenant's VPN.¶
Just as in [RFC7432], the RD, Ethernet Tag ID, MAC Address Length, MAC Address, IP Address Length, and IP Address fields are part of the route key used by BGP to compare routes. The rest of the fields are not part of the route key.¶
This route is advertised along with the following extended community:¶
For asymmetric IRB mode, the EVPN Router's MAC Extended Community is not needed because forwarding is performed using destination TS's MAC address, which is carried in this EVPN route type 2 advertisement.¶
This route MUST always be advertised with the MAC-VRF Route Target. It MAY also be advertised with a second Route Target corresponding to the IP-VRF.¶
When a PE (e.g., PE2 in Figure 4 above) receives this EVPN MAC/IP Advertisement route, it performs the following:¶
If the receiving PE receives the MAC/IP Advertisement route with the MPLS Label2 field but the receiving PE only supports asymmetric IRB mode, then the receiving PE MUST ignore the MPLS Label2 field and install the MAC address in the corresponding MAC-VRF and (IP, MAC) association in the ARP table or NDP cache for that tenant (with the IRB interface identified by the MAC-VRF).¶
When an Ethernet frame is received by an ingress PE (e.g., PE1 in Figure 4 above), the PE uses the AC ID (e.g., VLAN ID) to identify the associated MAC-VRF/BT, and it performs a lookup on the destination MAC address. If the MAC address corresponds to its IRB interface MAC address, the ingress PE deduces that the packet must be inter-subnet routed. Hence, the ingress PE performs an IP lookup in the associated IP-VRF table. The lookup identifies a local adjacency to the IRB interface associated with the egress subnet's MAC-VRF/ bridge table. The ingress PE also decrements the TTL / hop limit for that packet by one, and if it reaches zero, the ingress PE discards the packet.¶
The ingress PE gets the destination TS's MAC address for that TS's IP address from its ARP table or NDP cache. It encapsulates the packet with that destination MAC address and a source MAC address corresponding to that IRB interface and sends the packet to its destination subnet MAC-VRF/BT.¶
The destination MAC address lookup in the MAC-VRF/BT results in a BGP next-hop address of the egress PE along with label1 (L2 VPN MPLS label or VNI). The ingress PE encapsulates the packet using the Ethernet NVO tunnel of the choice (e.g., VXLAN or NVGRE) and sends the packet to the egress PE. Because the packet forwarding is between the ingress PE's MAC-VRF/BT and the egress PE's MAC-VRF/ bridge table, the packet encapsulation procedures follow that of [RFC7432] for MPLS and [RFC8365] for VXLAN encapsulations.¶
When a tenant's Ethernet frame is received over an NVO tunnel by the egress PE, the egress PE removes the NVO tunnel encapsulation and uses the VPN MPLS label (for MPLS encapsulation) or VNI (for NVO encapsulation) to identify the MAC-VRF/BT in which the MAC lookup needs to be performed.¶
The MAC lookup results in a local adjacency (e.g., local interface) over which the packet needs to get sent.¶
Note that the forwarding behavior on the egress PE is the same as the EVPN intra-subnet forwarding described in [RFC7432] for MPLS and [RFC8365] for NVO networks. In other words, all the packet processing associated with the inter-subnet forwarding semantics is confined to the ingress PE for asymmetric IRB mode.¶
It should also be noted that [RFC7432] provides a different level of granularity for the EVPN label. Besides identifying the bridge domain table, it can be used to identify the egress interface or a destination MAC address on that interface. If an EVPN label is used for an egress interface or individual MAC address identification, then no MAC lookup is needed in the egress PE for MPLS encapsulation, and the packet can be directly forwarded to the egress interface just based on the EVPN label lookup.¶
When a TS moves from one NVE (aka source NVE) to another NVE (aka target NVE), it is important that the MAC Mobility procedures be properly executed and the corresponding MAC-VRF and IP-VRF tables on all participating NVEs be updated. [RFC7432] describes the MAC Mobility procedures for L2-only services for both single-homed TS and multihomed TS. This section describes the incremental procedures and BGP Extended Communities needed to handle the MAC Mobility for IRB. In order to place the emphasis on the differences between L2-only and IRB use cases, the incremental procedure is described for a single-homed TS with the expectation that the additional steps needed for a multihomed TS can be extended per Section 15 of [RFC7432]. This section describes mobility procedures for both symmetric and asymmetric IRB. Although the language used in this section is for IPv4 ARP, it equally applies to IPv6 ND.¶
When a TS moves from a source NVE to a target NVE, it can behave in one of the following three ways:¶
Depending on the expected TS's behavior, an NVE needs to handle at least the first option and should be able to handle the second and third options. The following subsections describe the procedures for each scenario where it is assumed that the MAC and IP addresses of a TS have a one-to-one relationship (i.e., there is one IP address per MAC address and vice versa). The procedures for host mobility detection in the presence of a many-to-one relationship is outside the scope of this document, and it is covered in [EXTENDED-MOBILITY]. The "many-to-one relationship" refers to many host IP addresses corresponding to a single host MAC address or many host MAC addresses corresponding to a single IP address. It should be noted that in the case of IPv6, a link-local IP address does not count in a many-to-one relationship because that address is confined to a single Ethernet segment, and it is not used for host mobility (i.e., by definition, host mobility is between two different Ethernet segments). Therefore, when an IPv6 host is configured with both a Global Unicast address (or a Unique Local address) and a link-local address, for the purpose of host mobility, it is considered with a single IP address.¶
In this scenario, when a TS moves from a source NVE to a target NVE, the TS initiates a gratuitous ARP upon the move to the target NVE.¶
The target NVE, upon receiving this ARP message, updates its MAC-VRF, IP-VRF, and ARP table with the host MAC, IP, and local adjacency information (e.g., local interface).¶
Since this NVE has previously learned the same MAC and IP addresses from the source NVE, it recognizes that there has been a MAC move, and it initiates MAC Mobility procedures per [RFC7432] by advertising an EVPN MAC/IP Advertisement route with both the MAC and IP addresses filled in (per Sections 5.1 and 6.1) along with the MAC Mobility extended community, with the sequence number incremented by one. The target NVE also exercises the MAC duplication detection procedure in Section 15.1 of [RFC7432].¶
The source NVE, upon receiving this MAC/IP Advertisement route, realizes that the MAC has moved to the target NVE. It updates its MAC-VRF and IP-VRF table accordingly with the adjacency information of the target NVE. In the case of the asymmetric IRB, the source NVE also updates its ARP table with the received adjacency information, and in the case of the symmetric IRB, the source NVE removes the entry associated with the received (IP, MAC) from its local ARP table. It then withdraws its EVPN MAC/IP Advertisement route. Furthermore, it sends an ARP probe locally to ensure that the MAC is gone. If an ARP response is received, the source NVE updates its ARP entry for that (IP, MAC) and re-advertises an EVPN MAC/IP Advertisement route for that (IP, MAC) along with the MAC Mobility extended community, with the sequence number incremented by one. The source NVE also exercises the MAC duplication detection procedure in Section 15.1 of [RFC7432].¶
All other remote NVE devices, upon receiving the MAC/IP Advertisement route with the MAC Mobility extended community, compare the sequence number in this advertisement with the one previously received. If the new sequence number is greater than the old one, then they update the MAC/IP addresses of the TS in their corresponding MAC-VRF and IP-VRF tables to point to the target NVE. Furthermore, upon receiving the MAC/IP withdraw for the TS from the source NVE, these remote PEs perform the cleanups for their BGP tables.¶
In this scenario, when a TS moves from a source NVE to a target NVE, the TS starts sending data traffic without first initiating an ARP request.¶
The target NVE, upon receiving the first data packet, learns the MAC address of the TS in the data plane and updates its MAC-VRF table with the MAC address and the local adjacency information (e.g., local interface) accordingly. The target NVE realizes that there has been a MAC move because the same MAC address has been learned remotely from the source NVE.¶
If EVPN-IRB NVEs are configured to advertise MAC-only routes in addition to MAC-and-IP EVPN routes, then the following steps are taken:¶
If an EVPN-IRB NVE is configured not to advertise MAC-only routes, then upon receiving the first data packet, it learns the MAC address of the TS and updates the MAC entry in the corresponding MAC-VRF table with the local adjacency information (e.g., local interface). It also realizes that there has been a MAC move because the same MAC address has been learned remotely from the source NVE. It uses the local MAC route to find the corresponding local MAC-IP route and sends a unicast ARP request to the host. When receiving an ARP response, it follows the procedure outlined in Section 7.1. In the prior case, where MAC-only routes are also advertised, this procedure of triggering a unicast ARP probe at the target PE MAY also be used in addition to the source PE broadcast ARP probing procedure described earlier for better convergence.¶
In this scenario, when a TS moves from a source NVE to a target NVE, the TS is silent, and it neither initiates an ARP request nor sends any data traffic. Therefore, neither the target nor the source NVEs are aware of the MAC move.¶
On the source NVE, an age-out timer (for the silent host that has moved) is used to trigger an ARP probe. This age-out timer can be either an ARP timer or a MAC age-out timer, and this is an implementation choice. The ARP request gets sent both locally to all the attached TSs on that subnet as well as to all the remote NVEs (including the target NVE) participating in that subnet. The source NVE also withdraws the EVPN MAC/IP Advertisement route with only the MAC address (if it has previously advertised such a route).¶
The target NVE passes the ARP request to its locally attached TSs, and when it receives the ARP response, it updates its MAC-VRF, IP-VRF, and ARP table with the host (IP, MAC) and local adjacency information (e.g., local interface). It also sends an EVPN MAC/IP Advertisement route with both the MAC and IP address fields filled in along with the MAC Mobility extended community, with the sequence number incremented by one.¶
When the source NVE receives the EVPN MAC/IP Advertisement route, it updates its IP-VRF table with the new adjacency information (pointing to the target NVE). In the case of the asymmetric IRB, the source NVE also updates its ARP table with the received adjacency information, and in the case of the symmetric IRB, the source NVE removes the entry associated with the received (IP, MAC) from its local ARP table. Furthermore, it withdraws its previously advertised EVPN MAC/IP route with both the MAC and IP address fields filled in.¶
All other remote NVE devices, upon receiving the MAC/IP Advertisement route with the MAC Mobility extended community, compare the sequence number in this advertisement with the one previously received. If the new sequence number is greater than the old one, then they update the MAC/IP addresses of the TS in their corresponding MAC-VRF, IP-VRF, and ARP (in the case of asymmetric IRB) tables to point to the new NVE. Furthermore, upon receiving the MAC/IP withdraw for the TS from the old NVE, these remote PEs perform the cleanups for their BGP tables.¶
This document defines one new BGP Extended Community for EVPN.¶
A new EVPN BGP Extended Community called "EVPN Router's MAC" is introduced here. This new extended community is a transitive extended community with a Type field of 0x06 (EVPN) and a Sub-Type field of 0x03. It may be advertised along with the Encapsulation Extended Community defined in Section 4.1 of [RFC9012].¶
The EVPN Router's MAC Extended Community is encoded as an 8-octet value as follows:¶
This extended community is used to carry the PE's MAC address for symmetric IRB scenarios, and it is sent with EVPN RT-2. The advertising PE SHALL only attach a single EVPN Router's MAC Extended Community to a route. In case the receiving PE receives more than one EVPN Router's MAC Extended Community with a route, it SHALL process the first one in the list and not store and propagate the others.¶
The following sections describe two main symmetric IRB forwarding scenarios (within a DC -- i.e., intra-DC) along with the corresponding procedures. In the following scenarios, without loss of generality, it is assumed that a given tenant is represented by a single IP-VPN instance. Therefore, on a given PE, a tenant is represented by a single IP-VRF table and one or more MAC-VRF tables.¶
This section covers the symmetric IRB procedures for the scenario where each TS is attached to one or more NVEs, and its host IP and MAC addresses are learned by the attached NVEs and are distributed to all other NVEs that are interested in participating in both intra-subnet and inter-subnet communications with that TS.¶
In this scenario, without loss of generality, it is assumed that NVEs operate in VLAN-based service interface mode with one bridge table(s) per MAC-VRF. Thus, for a given tenant, an NVE has one MAC-VRF for each tenant subnet (e.g., each VLAN) that is configured for extension via VXLAN or NVGRE encapsulation. In the case of VLAN-aware bundling, each MAC-VRF consists of multiple bridge tables (e.g., one bridge table per VLAN). The MAC-VRFs on an NVE for a given tenant are associated with an IP-VRF corresponding to that tenant (or IP-VPN instance) via their IRB interfaces.¶
Since VXLAN and NVGRE encapsulations require an inner Ethernet header (inner MAC SA/DA) and since a TS MAC address cannot be used for inter-subnet traffic, the ingress NVE's MAC address is used as an inner MAC SA. The NVE's MAC address is the device MAC address, and it is common across all MAC-VRFs and IP-VRFs. This MAC address is advertised using the new EVPN Router's MAC Extended Community (Section 8.1).¶
Figure 6 below illustrates this scenario, where a given tenant (e.g., an IP-VPN instance) has three subnets represented by MAC-VRF1, MAC-VRF2, and MAC-VRF3 across two NVEs. There are five TSs that are associated with these three MAC-VRFs -- i.e., TS1, TS4, and TS5 are on the same subnet (e.g., the same MAC-VRF/VLAN). TS1 and TS5 are associated with MAC-VRF1 on NVE1, while TS4 is associated with MAC-VRF1 on NVE2. TS2 is associated with MAC-VRF2 on NVE1, and TS3 is associated with MAC-VRF3 on NVE2. MAC-VRF1 and MAC-VRF2 on NVE1 are, in turn, associated with IP-VRF1 on NVE1, and MAC-VRF1 and MAC-VRF3 on NVE2 are associated with IP-VRF1 on NVE2. When TS1, TS5, and TS4 exchange traffic with each other, only the L2 forwarding (bridging) part of the IRB solution is exercised because all these TSs belong to the same subnet. However, when TS1 wants to exchange traffic with TS2 or TS3, which belong to different subnets, both the bridging and routing parts of the IRB solution are exercised. The following subsections describe the control and data plane operations for this IRB scenario in detail.¶
Each NVE advertises a MAC/IP Advertisement route (i.e., route type 2) for each of its TSs with the following field set:¶
Each NVE advertises an EVPN RT-2 route with two Route Targets (one corresponding to its MAC-VRF and the other corresponding to its IP-VRF). Furthermore, the EVPN RT-2 is advertised with two BGP Extended Communities. The first BGP Extended Community identifies the tunnel type, and it is called "Encapsulation Extended Community" as defined in [RFC9012], and the second BGP Extended Community includes the MAC address of the NVE (e.g., MACx for NVE1 or MACy for NVE2) as defined in Section 8.1. The EVPN Router's MAC Extended Community MUST be added when the Ethernet NVO tunnel is used. If the IP NVO tunnel type is used, then there is no need to send this second Extended Community. It should be noted that the IP NVO tunnel type is only applicable to symmetric IRB procedures.¶
Upon receiving this advertisement, the receiving NVE performs the following:¶
If the receiving NVE receives an EVPN RT-2 with only label1 and only a single Route Target corresponding to IP-VRF; an EVPN RT-2 with only a single Route Target corresponding to MAC-VRF but with both label1 and label2; or an EVPN RT-2 with a MAC address length of zero, then it MUST use the treat-as-withdraw approach [RFC7606] and SHOULD log an error message.¶
The following description of the data plane operation describes just the logical functions, and the actual implementation may differ. Let's consider the data plane operation when TS1 in subnet-1 (MAC-VRF1) on NVE1 wants to send traffic to TS3 in subnet-3 (MAC-VRF3) on NVE2.¶
In this symmetric IRB scenario, inter-subnet traffic between NVEs will always use the IP-VRF VNI/MPLS label. For instance, traffic from TS2 to TS4 will be encapsulated by NVE1 using NVE2's IP-VRF VNI/MPLS label, as long as TS4's host IP is present in NVE1's IP-VRF.¶
This section covers the symmetric IRB procedures for the scenario where some TSs support one or more subnets and these TSs are associated with one or more NVEs. Therefore, besides the advertisement of MAC/IP addresses for each TS, which can be multihomed with All-Active redundancy mode, the associated NVE needs to also advertise the subnets statically configured on each TS.¶
The main difference between this solution and the previous one is the additional advertisement corresponding to each subnet. These subnet advertisements are accomplished using the EVPN IP Prefix route defined in [RFC9136]. These subnet prefixes are advertised with the IP address of their associated TS (which is in an overlay address space) as their next hop. The receiving NVEs perform recursive route resolution to resolve the subnet prefix with its advertising NVE so that they know which NVE to forward the packets to when they are destined for that subnet prefix.¶
The advantage of this recursive route resolution is that when a TS moves from one NVE to another, there is no need to re-advertise any of the subnet prefixes for that TS. All that is needed is to advertise the IP/MAC addresses associated with the TS itself and exercise the MAC Mobility procedures for that TS. The recursive route resolution automatically takes care of the updates for the subnet prefixes of that TS.¶
Figure 7 illustrates this scenario where a given tenant (e.g., an IP-VPN service) has three subnets represented by MAC-VRF1, MAC-VRF2, and MAC-VRF3 across two NVEs. There are four TSs associated with these three MAC-VRFs -- i.e., TS1 is connected to MAC-VRF1 on NVE1, TS2 is connected to MAC-VRF2 on NVE1, TS3 is connected to MAC-VRF3 on NVE2, and TS4 is connected to MAC-VRF1 on NVE2. TS1 has two subnet prefixes (SN1 and SN2), and TS3 has a single subnet prefix (SN3). The MAC-VRFs on each NVE are associated with their corresponding IP-VRF using their IRB interfaces. When TS4 and TS1 exchange intra-subnet traffic, only the L2 forwarding (bridging) part of the IRB solution is used (i.e., the traffic only goes through their MAC-VRFs); however, when TS3 wants to forward traffic to SN1 or SN2 sitting behind TS1 (inter-subnet traffic), then both the bridging and routing parts of the IRB solution are exercised (i.e., the traffic goes through the corresponding MAC-VRFs and IP-VRFs). If TS4, for example, wants to reach SN1, it uses its default route and sends the packet to the MAC address associated with the IRB interface on NVE2; NVE2 then performs an IP lookup in its IP-VRF and finds an entry for SN1. The following subsections describe the control and data plane operations for this IRB scenario in detail.¶
Note that in Figure 7, above, SN1 and SN2 are configured on NVE1, which then advertises each in an IP Prefix route. Similarly, SN3 is configured on NVE2, which then advertises it in an IP Prefix route.¶
Each NVE advertises a route type 5 (EVPN RT-5, IP Prefix route defined in [RFC9136]) for each of its subnet prefixes with the IP address of its TS as the next hop (Gateway Address field) as follows:¶
This EVPN RT-5 is advertised with one or more Route Targets associated with the IP-VRF from which the route is originated.¶
Each NVE also advertises an EVPN RT-2 (MAC/IP Advertisement route) along with its associated Route Targets and Extended Communities for each of its TSs exactly as described in Section 9.1.1.¶
Upon receiving the EVPN RT-5 advertisement, the receiving NVE performs the following:¶
When receiving the EVPN RT-2 advertisement, the receiving NVE imports the MAC/IP addresses of the TS into the corresponding MAC-VRF and IP-VRF per Section 9.1.1. When both routes exist, recursive route resolution is performed to resolve the IP prefix (received in EVPN RT-5) to its corresponding NVE's IP address (e.g., its BGP next hop). The BGP next hop will be used as the underlay tunnel destination address (e.g., VTEP DA for VXLAN encapsulation), and the EVPN Router's MAC will be used as the inner MAC for VXLAN encapsulation.¶
The following description of the data plane operation describes just the logical functions, and the actual implementation may differ. Let's consider the data plane operation when a host in SN1 behind TS1 wants to send traffic to a host in SN3 behind TS3.¶
The security considerations for Layer 2 forwarding in this document follow those of [RFC7432] for MPLS encapsulation and those of [RFC8365] for VXLAN or NVGRE encapsulations. This section describes additional considerations.¶
This document describes a set of procedures for inter-subnet forwarding of tenant traffic across PEs (or NVEs). These procedures include both Layer 2 forwarding and Layer 3 routing on a packet-by-packet basis. The security consideration for Layer 3 routing in this document follows that of [RFC4365], with the exception of the application of routing protocols between CEs and PEs. Contrary to [RFC4364], this document does not describe route distribution techniques between CEs and PEs 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.¶
The VPN scheme described in this document does not provide the quartet of security properties mentioned in [RFC4365] (confidentiality protection, source authentication, integrity protection, and replay protection). If these are desired, they must be provided by mechanisms that are outside the scope of the VPN mechanisms.¶
In this document, the EVPN RT-5 is used for certain scenarios. This route uses an Overlay Index that requires a recursive resolution to a different EVPN route (an EVPN RT-2). Because of this, it is worth noting that any action that ends up filtering or modifying the EVPN RT-2 route used to convey the Overlay Indexes will modify the resolution of the EVPN RT-5 and therefore the forwarding of packets to the remote subnet.¶
IANA has allocated Sub-Type value 0x03 in the "EVPN Extended Community Sub-Types" registry as follows:¶
Sub-Type Value | Name | Reference |
---|---|---|
0x03 | EVPN Router's MAC Extended Community | RFC 9135 |
This document has been listed as an additional reference for the MAC/IP Advertisement route in the "EVPN Route Types" registry.¶
The authors would like to thank Sami Boutros, Jeffrey Zhang, Krzysztof Szarkowicz, Lukas Krattiger and Neeraj Malhotra for their valuable comments. The authors would also like to thank Linda Dunbar, Florin Balus, Yakov Rekhter, Wim Henderickx, Lucy Yong, and Dennis Cai for their feedback and contributions.¶