EVPN for VXLAN tunnels (Layer 2)

EVPN Layer 2 (without multi-homing) includes the implementation of the BGP-EVPN address family and support for the following route types:

• EVPN MAC/IP route (or type 2, RT2)

• EVPN Inclusive Multicast Ethernet Tag route (IMET or type 3, RT3)

The MAC/IP route is used to convey the MAC and IP information of hosts connected to subinterfaces in the MAC-VRF. The IMET route is advertised as soon as bgp-evpn is enabled in the MAC-VRF; it has the following purpose:

• Auto-discovery of the remote VTEPs attached to the same EVI

• Creation of a default flooding list in the MAC-VRF so that BUM frames are replicated

Advertisement of the MAC/IP and IMET routes is configured on a per-MAC-VRF basis. The following example shows the default setting advertise true, which advertises MAC/IP and IMET routes.

Note that changing the setting of the advertise parameter and committing the change internally flaps the BGP instance.

--{ candidate shared default }--[ network-instance blue protocols bgp-evpn bgp-
instance 1 ]--
# info detail
    admin-state enable
    vxlan-interface vxlan1.1
    evi 1
    ecmp 1
    default-admin-tag 0
    routes {
        next-hop use-system-ipv4-address
        mac-ip {
            advertise true
        }
        inclusive-mcast {
            advertise true
        }
    }

The creation of VXLAN destinations of type unicast, unicast ES (Ethernet Segment), and multicast for each vxlan-interface is driven by the reception of EVPN routes.

The created unicast, unicast ES, and multicast VXLAN destinations are visible in state. Each destination is allocated a system-wide unique destination index and is an internal NHG-ID (next-hop group ID). The destination indexes for the VXLAN destinations are shown in the following example for destination 10.22.22.4, vni 1

--{ [FACTORY] + candidate shared default }--[  ]--
# info from state tunnel-interface vxlan1 vxlan-interface 1 bridge-table unicast-
destinations destination * vni *
    tunnel-interface vxlan1 {
        vxlan-interface 1 {
            bridge-table {
                unicast-destinations {
                    destination 10.44.44.4 vni 1 {
                        destination-index 677716962904 // destination index
                        statistics {
                        }
                        mac-table {
                            mac 00:00:00:01:01:04 {
                                type evpn-static
                                last-update "16 hours ago"
                            }
                        }
                    }
                }
            }
        }
    }
--{ [FACTORY] + candidate shared default }--[  ]--
# info from state network-instance blue bridge-table mac-table mac 00:00:00:01:01:04
    network-instance blue {
        bridge-table {
            mac-table {
                mac 00:00:00:01:01:04 {
                    destination-type vxlan
                    destination-index 677716962904 // destination index
                    type evpn-static
                    last-update "16 hours ago"
                    destination "vxlan-interface:vxlan1.1 vtep:10.44.44.4 vni:1"
                }
            }
        }
    }

The following is an example of dynamically created multicast destinations for a vxlan-interface:

--{ [FACTORY] + candidate shared default }--[  ]--
A:dut1# info from state tunnel-interface vxlan1 vxlan-interface 1 bridge-
table multicast-destinations
    tunnel-interface vxlan1 {
        vxlan-interface 1 {
            bridge-table {
                multicast-destinations {
                    destination 40.1.1.2 vni 1 {
                        multicast-forwarding BUM
                        destination-index 46428593833
                    }
                    destination 40.1.1.3 vni 1 {
                        multicast-forwarding BUM
                        destination-index 46428593835
                    }
                    destination 40.1.1.4 vni 1 {
                        multicast-forwarding BUM
                        destination-index 46428593829
                    }
                }
            }
        }
    }

When a MAC is received from multiple sources, the route is selected based on the priority listed in MAC selection. Learned and EVPN-learned routes have equal priority; the latest received route is selected.

When multiple EVPN-learned MAC/IP routes arrive for the same MAC but with a different key (for example, two routes for MAC M1 with different route-distinguishers), a selection is made based on the following priority:

1. EVPN MACs with higher SEQ number

2. EVPN MACs with lower IP next-hop

3. EVPN MACs with lower Ethernet Tag

4. EVPN MACs with lower RD

You can configure the BGP next hop to be used for the EVPN routes advertised for a network-instance. This next hop is by default the IPv4 address configured in interface system 0.0 of the default network-instance. However, the next-hop address can be changed to any IPv4 address.

The system does not check that the configured IP address exists in the default network-instance. Any valid IP address can be used as next hop of the EVPN routes advertised for the network-instance, irrespective of its existence in any subinterface of the system. However, the receiver leaf nodes create their unicast, multicast and ES destinations to this advertised next-hop, so it is important that the configured next-hop is a valid IPv4 address that exists in the default network-instance.

When the system or loopback interface configured for the BGP next-hop is administratively disabled, EVPN still advertises the routes, as long as a valid IP address is available for the next-hop. However, received traffic on that interface is dropped.

The following example configures a BGP next hop to be used for the EVPN routes advertised for a network-instance.

--{ candidate shared default }--[ network-instance 1 protocols bgp-evpn bgp-
instance 1 ]--
# info
    routes {
        next-hop 1.1.1.1
        }
    }

EVPN relies on three different procedures to handle multi-homing: DF election, split-horizon, and aliasing. DF election is relevant to single-active and all-active multi-homing; split-horizon and aliasing are relevant only to all-active multi-homing.

• DF Election – The Designated Forwarder (DF) is the leaf that forwards BUM traffic in the ES. Only one DF can exist per ES at a time, and it is elected based on the exchange of ES routes (type 4) and the subsequent DF Election Algorithm (DF Election Alg).

  In single-active multi-homing, the non-DF leafs bring down the subinterface associated with the ES.

  In all-active multi-homing, the non-DF leafs do not forward BUM traffic received from remote EVPN PEs.

• Split-horizon – This is the mechanism by which BUM traffic received from a peer ES PE is filtered so that it is not looped back to the CE that first transmitted the frame. Local bias is applied in VXLAN services, as described in RFC 8365.

• Aliasing – This is the procedure by which PEs that are not attached to the ES can process non-zero ESI MAC/IP routes and AD routes and create ES destinations to which per-flow ECMP can be applied.

To support multi-homing, EVPN-VXLAN supports two additional route types:

• ES routes (type 4) – Used for ES discovery on all the leafs attached to the ES and DF Election.

  ES routes use an ES-import route target extended community (its value is derived from the ESI), so that its distribution is limited to only the leafs that are attached to the ES.

  The ES route is advertised with the DF Election extended community, which indicates the intent to use a specific DF Election Alg and capabilities.

  Upon reception of the remote ES routes, each PE builds a DF candidate list based on the originator IP of the ES routes. Then, based on the agreed DF Election Alg, each PE elects one of the candidates as DF for each mac-vrf where the ES is defined.

• AD route (type 1) – Advertised to the leafs attached to an ES. There are two versions of AD routes:

  • AD per-ES route – Used to advertise the multi-homing mode (single-active or all-active) and the ESI label, which is not advertised or processed in case of VXLAN. Its withdrawal enables the mass withdrawal procedures in the remote PEs.

  • AD per-EVI route – Used to advertise the availability of an ES in an EVI and its VNI. It is needed by the remote leafs for the aliasing procedures.

  Both versions of AD routes can influence the DF Election. Their withdrawal from a leaf results in removing that leaf from consideration for DF Election for the associated EVI, as long as ac-df exclude is configured. (The AC-DF capability can be set to exclude only if the DF election algorithm type is set to preference.)

Local bias for all-active multi-homing is based on the following behavior at the ingress and egress leafs:

• At the ingress leaf, any BUM traffic received on an all-active multi-homing LAG subinterface (associated with an EVPN-VXLAN mac-vrf) is flooded to all local subinterfaces, irrespective of their DF or NDF status, and VXLAN tunnels.

• At the egress leaf, any BUM traffic received on a VXLAN subinterface (associated with an EVPN-VXLAN mac-vrf) is flooded to single-homed subinterfaces and multi-homed subinterfaces whose ES is not shared with the owner of the source VTEP if the leaf is DF for the ES.

In SR Linux, the local bias filtering entries on the egress leaf are added or removed based on the ES routes, and they are not modified by the removal of AD per EVI/ES routes. This may cause blackholes in the multi-homed CE for BUM traffic if the local subinterfaces are administratively disabled.

EVPN L2 single-homing configuration shows a single-active ES attached to two leaf nodes. In this configuration, the ES in single-active mode can be configured to do the following:

• Associate to an Ethernet interface or a LAG interface (as all-active ESes)

• Coexist with all-active ESes on the same node, as well as in the same MAC-VRF service.

• Signal the non-DF state to the CE by using LACP out-of-synch signaling or power off.

  Optionally, the ES can be configured not to signal to the CE. When the LACP synch flag or power off is used to signal the non-DF state to the CE/server, all of the subinterfaces are active on the same node; that is, load balancing is not per-service, but rather per-port. This mode of operation is known as EVPN multi-homing port-active mode.

• Connect to a CE that uses a single LAG to connect to the ES or separate LAG/ports per leaf in the ES.

All peers in the ES must be configured with the same multi-homing mode; if the nodes are not configured consistently, the oper-multi-homing-mode in state is single-active. From a hardware resource perspective, no local-bias-table entries are populated for ESes in single-active mode.

The following features work in conjunction with single-active mode:

• Preference-based DF election / non-revertive option configures ES peers to elect a DF based on a preference value, as well as an option to prevent traffic from reverting back to a former DF node.

• Attachment Circuit influenced DF Election (AC-DF) allows the DF election candidate list for a network-instance to be modified based on the presence of the AD per-EVI and per-ES routes.

• Standby LACP-based or power-off signaling configures how the node’s non-DF state is signaled to the multi-homed CE.

After the system boots, the reload-delay timer keeps an interface shut down with the laser off for a configured amount of time until connectivity with the rest of network is established. When applied to an access multi-homed interface (typically an Ethernet Segment interface), this delay can prevent black-holing traffic coming from the multi-homed server or CE.

In EVPN multi-homing scenarios, if one leaf in the ES peer group is rebooting, coming up after an upgrade or a failure, it is important for the ES interface not to become active until after the node is ready to forward traffic to the core. If the ES interface comes up too quickly and the node has not programmed its forwarding tables yet, traffic from the server is black-holed. To prevent this from happening, you can configure a reload-delay timer on the ES interface so that the interface does not become active until after network connectivity is established.

When a reload-delay timer is configured, the interface port is shut down and the laser is turned off from the time that the system determines the interface state following a reboot or reload of the XDP process, until the number of seconds specified in the reload-delay timer elapse.

The reload-delay timer is only supported on Ethernet interfaces that are not enabled with breakout mode. For a multi-homed LAG interface, the reload-delay timer should be configured on all the interface members. The reload-delay timer can be from 1-86,400 seconds. There is no default value; if not configured for an interface, there is no reload-delay timer.

Only ES interfaces should be configured with a non-zero reload-delay timer. Single-homed interfaces and network interfaces (used to forward VXLAN traffic) should not have a reload-delay timer configured.

The following example sets the reload-delay timer for an interface to 20 seconds. The timer starts following a system reboot or when the IMM is reconnected, and the system determines the interface state. During the timer period, the interface is deactivated and the port laser is inactive.

--{ * candidate shared default }--[  ]--
# info interface ethernet-1/1
    interface ethernet-1/1 {
        admin-state enable
        ethernet {
            reload-delay 20
        }
    }

When the reload-delay timer is running, the port-oper-down-reason for the port is shown as interface-reload-timer-active. The reload-delay-expires state indicates the amount of time remaining until the port becomes active. For example:

--{ * candidate shared default }--[  ]--
# info from state interface ethernet-1/1
    interface ethernet-1/1 {
        description eth_seg_1
        admin-state enable
        mtu 9232
        loopback-mode false
        ifindex 671742
        oper-state down
        oper-down-reason interface-reload-time-active
        last-change "51 seconds ago"
        vlan-tagging true
        ...
        ethernet {
            auto-negotiate false
            lacp-port-priority 32768
            port-speed 100G
            hw-mac-address 00:01:01:FF:00:15
            reload-delay 20
reload-delay-expires "18 seconds from now"
            flow-control {
                receive false
                transmit false
            }
        }
    }

The following is an example of a single-active multi-homing configuration, including standby signaling power-off, AC-DF capability, and preference-based DF algorithm.

The following configures power-off signaling and the reload delay timer for an interface:

--{ * candidate shared default }--[  ]--
# info interface ethernet-1/21 ethernet
    standby-signaling power-off // needed to signal non-DF state to the CE
    reload-delay 100 // upon reboot, this is required to avoid attracting traffic 
from the multi-homed CE until the node is ready to forward.
// The time accounts for the time it takes all network protocols to be up 
and forwarding entries ready.

The following configures DF election settings for the ES, including preference-based DF election and a preference value for the DF election alg. The ac-df setting is set to exclude, which disables the AC-DF capability. The non-revertive option is enabled, which prevents traffic from reverting back to a former DF node when the node reconnects to the network.

--{ * candidate shared default }--[ system network-instance protocols evpn ethernet-
segments bgp-instance 1 ethernet-segment eth_seg_1 ]--
# info
    admin-state enable
    esi 00:01:00:00:00:00:00:00:00:00
    interface ethernet-1/21
    multi-homing-mode single-active
    df-election {
        interface-standby-signaling-on-non-df { // presence container that enables 
the standby-signaling for the ES
        }
        algorithm {
            type preference // enables the use of preference based DF election
            preference-alg {
                preference-value 100 // changes the default 32767 to a 
specific value
                capabilities {
                    ac-df exclude // turns off the default ac-df capability
                    non-revertive true // enables the non-revertive mode
                }
            }
        }
    }

The following shows the state (and consequently the configuration) of an ES for single-active multi-homing and indicates the default settings for the algorithm/oper-type. All of the peers in the ES should be configured with the same algorithm type. However, if that is not the case, all the peers fall back to the default algorithm.

--{ * candidate shared default }--[ system network-instance protocols evpn ethernet-
segments bgp-instance 1 ethernet-segment eth_seg_1 ]--
# info
    admin-state enable
    oper-state up
    esi 00:01:00:00:00:00:00:00:00:00
    interface ethernet-1/21
    multi-homing-mode single-active
    oper-multi-homing-mode single-active // oper mode may be different if not all 
the ES peers are configured in the same way
    df-election {
        interface-standby-signaling-on-non-df {
        }
        algorithm {
            type preference
            oper-type preference // if at least one peer in the ES is in type 
default, all the peers will fall back to default
            preference-alg {
                preference-value 100
                capabilities {
                    ac-df exclude
                    non-revertive true
                }
            }
        }
    }
    routes {
        next-hop use-system-ipv4-address
        ethernet-segment {
            originating-ip use-system-ipv4-address
        }
    }
    association {
        network-instance blue {
            bgp-instance 1 {
                designated-forwarder-role-last-change "2 seconds ago"
                designated-forwarder-activation-start-time "2 seconds ago"
                designated-forwarder-activation-time 3
                computed-designated-forwarder-candidates {
                    designated-forwarder-candidate 40.1.1.1 {
                        add-time "2 seconds ago"
                        designated-forwarder true
                    }
                    designated-forwarder-candidate 40.1.1.2 {
                        add-time "2 minutes ago"
                        designated-forwarder false
                    }
                }
            }
        }
    }
}

To display information about the ES, use the show system network-instance ethernet-segments command. For example:

--{ [FACTORY] + candidate shared default }--[  ]--
# show system network-instance ethernet-segments eth_seg_1
------------------------------------------------------------------------------------
eth_seg_1 is up, single-active
  ESI      : 00:01:00:00:00:00:00:00:00:00
  Alg      : preference
  Peers    : 40.1.1.2
  Interface: ethernet-1/21
  Network-instances:
     blue
      Candidates : 40.1.1.1 (DF), 40.1.1.2
      Interface : ethernet-1/21.1
------------------------------------------------------------------------------------
Summary
 1 Ethernet Segments Up
 0 Ethernet Segments Down
------------------------------------------------------------------------------------

The detail option displays more information about the ES. For example:

--{ [FACTORY] + candidate shared default }--[  ]--
# show system network-instance ethernet-segments eth_seg_1 detail
====================================================================================
Ethernet Segment
====================================================================================
Name                 : eth_seg_1
40.1.1.1 (DF)
Admin State          : enable              Oper State        : up
ESI                  : 00:01:00:00:00:00:00:00:00:00
Multi-homing         : single-active          Oper Multi-homing : single-active
Interface            : ethernet-1/21
ES Activation Timer  : None
DF Election          : preference          Oper DF Election  : preference
 
Last Change          : 2021-04-06T08:49:44.017Z
====================================================================================
 MAC-VRF    Actv Timer Rem   DF
eth_seg_1   0                Yes
------------------------------------------------------------------------------------
DF Candidates
------------------------------------------------------------------------------------
Network-instance       ES Peers
blue                   40.1.1.1 (DF)
blue                   40.1.1.2
====================================================================================

On the DF node, the info from state command displays the following:

--{ [FACTORY] + candidate shared default }--[  ]--
# info from state interface ethernet-1/21 | grep oper
        oper-state up
            oper-state not-present
            oper-state up

--{ [FACTORY] + candidate shared default }--[  ]--
# info from state network-instance blue interface ethernet-1/21.1
    network-instance blue {
        interface ethernet-1/21.1 {
            oper-state up
            oper-mac-learning up
            index 6
            multicast-forwarding BUM
        }
    }

On the non-DF node, the info from state command displays the following:

--{ [FACTORY] + candidate shared default }--[  ]--
# info from state interface ethernet-1/21 | grep oper
        oper-state down
        oper-down-reason standby-signaling
            oper-state not-present
            oper-state down
            oper-down-reason port-down

--{ [FACTORY] + candidate shared default }--[  ]--
# info from state network-instance blue interface ethernet-1/21.1
    network-instance blue {
        interface ethernet-1/21.1 {
            oper-state down
            oper-down-reason subif-down
            oper-mac-learning up
            index 7
            multicast-forwarding none
        }
    }

When dynamic learning for proxy-ARP/ND is enabled, all frames coming into bridged subinterfaces of the MAC-VRF that have Ethertype 0x0806 (for proxy-ARP) or ICMPv6 Neighbor Discovery (for proxy-ND) are sent to the CPM for learning. This includes ARP-request and ARP-reply (including gratuitous ARP) messages and Neighbor Solicitation and Neighbor Advertisement messages (for ND). Dynamic entries in the proxy-ARP/ND table are created from both message types. For proxy-ND, dynamic entries are created only from Neighbor Advertisement messages. Learning an entry is done irrespective of the MAC Destination Address (DA) of the ARP or ND packet (unicast or broadcast).

The dynamically learned information in the table is based on the ARP/ND payload MAC Source Address (SA) and IP DA, and not the frame outer MAC SA (although they normally match). A valid MAC SA must be present for the entry to be learned.

In addition to learning the dynamic entry from the ARP-request, ARP-reply, or Neighbor Advertisement, the system re-injects and forwards messages as follows:

• For received ARP-request or Neighbor Solicitation messages, the system looks up the requested IP address, and if the lookup is successful, it sends an ARP-reply or Neighbor Advertisement with the MAC→IP information. If the lookup is not successful, the ARP-request / Neighbor Solicitation is re-injected and flooded based on the flood list and the configured flooding option, with source squelching. See Configuring proxy-ARP/ND traffic flooding options.

  The re-injected ARP-request or Neighbor Solicitation keeps all existing non-service-delimiting tags of the original frame. Unicast ARP-requests are replied to if there is an entry in the proxy table. If the lookup is not successful, the frame is forwarded to the MAC DA.

• For ARP-reply / Neighbor Advertisement messages, the MAC DA (unicast) is looked up in the MAC table. In case of a hit, the frame is re-injected and unicasted based on the MAC table information. If there is no hit, the frame is re-injected and flooded based on the flood list information (with source squelching). The re-injected ARP-reply / Neighbor Advertisement keeps all existing non-service-delimiting tags of the original frame.

• For ARP/ND frames that are sent to the CPM, the ARP reply / Neighbor Advertisement is always unicasted to the subinterface on which the ARP-request / Neighbor Solicitation arrived, even if the MAC itself has not yet been learned in the MAC table.

If a MAC ACL is bound to the ingress of a bridged subinterface (see the SR Linux ACL and Policy-Based Routing Guide), and proxy-ARP/ND is enabled in the network-instance, the rules in the MAC ACL are applied before extraction to the CPM. Consequently, if the MAC ACL drops an ARP packet, it never reaches the CPM.

Disabling dynamic learning for proxy-ARP/ND causes the dynamically learned entries to be flushed from the proxy-ARP/ND table. You can also set an age timer (default disabled) for dynamic entries, after which they are flushed from the proxy table.

In addition, you can set a timer to refresh dynamic entries. At a configured interval (default never) the system generates ARP-requests / Neighbor Solicitations with the intent to refresh the proxy entry; if no response is received, another one is attempted.

You can configure static entries in the proxy-ARP/ND table. Static entries have higher priority than snooped dynamic and EVPN entries in the table.

A static proxy-ARP/ND entry requires a static or dynamic MAC entry in the MAC table to become active. The static entries are advertised in MAC/IP routes, with the MAC Mobility extended community following the information associated with the MAC entry (static bit or sequence number).

The system does not validate the MAC addresses configured in static entries.

When proxy-arp or proxy-nd is disabled, the configured static entries remain in the proxy table, unlike dynamically learned and EVPN-learned entries, which are flushed from the table.

When proxy-ARP/ND is configured in an EVPN, the PE devices dynamically learn IP-MAC bindings for their local hosts and advertise them in EVPN MAC/IP routes to remote PE devices. The remote PE devices add these IP-MAC bindings to the proxy-ARP/ND table as EVPN-learned entries.

When a host sends an ARP-request or Neighbor Solicitation for a host on the remote side of the EVPN, if the PE device has the EVPN-learned IP-MAC binding in its proxy-ARP/ND table, it sends back an ARP-reply or Neighbor Advertisement with the remote host's MAC. If the IP-MAC binding does not exist in the PE device's proxy-ARP/ND table, the ARP-request or Neighbor Solicitation is flooded in the BD (ARP-request / Neighbor Solicitation flooding can be optionally disabled). See Layer 2 proxy-ARP example for an illustration of this process.

Proxy-ARP/ND duplicate IP detection is a security mechanism described in RFC 9161 to detect ARP/ND-spoofing attacks. In an ARP/ND-spoofing attack, an attacker sends false ARP/ND messages into a BD, with the goal of associating the attacker's IP address with a target host and directing traffic for that host to the attacker.

The proxy-ARP/ND duplicate IP detection feature monitors changes to active entries in the proxy-ARP/ND table. When an IP move occurs (for example, IP1→MAC1 is replaced by IP1→MAC2 in the table), a monitoring-window timer is started (default 3 minutes). If a specified number of IP moves (default 5) is detected before the monitoring-window timer expires, the IP is considered to be a duplicate.

When the system detects an IP move in the proxy table (for example, IP1→MAC1 changing to IP1→MAC2) it places the IP1→MAC2 proxy table entry in pending-confirmation state for a maximum of 30 seconds. During the pending-confirmation period, the ARP/ND entry is inactive, and a confirm-message is unicast to MAC1. If no reply from MAC1 is received during the pending-confirmation period, the IP1→MAC2 entry is confirmed as legitimate and becomes active. If a reply from MAC1 is received, then MAC2 is sent a confirm-message. If MAC2 replies, an additional confirm-message is sent to MAC1. If both MAC1 and MAC2 keep replying to the confirm-messages, it triggers the duplicate IP detection procedure for IP1, because the number of IP moves exceeds the maximum allowed during the monitoring window.

When an IP is detected as a duplicate, the proxy table cannot be updated with new dynamic or EVPN-learned entries for the same IP (although you can configure a static entry for the IP). The duplicate IP is subject to this restriction until a configured hold-down-time expires (default 9 minutes), after which the entry for the IP is removed from the proxy table, and the monitoring process for the IP is restarted.

When you enable proxy-ARP/ND for a MAC-VRF, this creates a table containing proxy-ARP/ND entries learned dynamically by snooping ARP/ND messages, configured statically, or learned from EVPN MAC/IP routes from remote PE nodes. By default, this table can contain up to 250 entries. You can configure the size of this table to be from 1 to 8,192 entries.

The system generates a log event when the size of the table reaches 90% of the maximum size and when the table reaches 95% of the maximum size. When the table reaches 100% of its maximum size, entries for an IP can be replaced (that is, a different MAC can be learned and added for the IP), but no new IP entries can be added to the table, regardless of the type (dynamic, static, or EVPN-learned).