Import IPv4/6 routes from VRF tables to the Global RIB as EVPN IP Prefix

[lucasolim]

Adding the option to import IPv4/6 routes from VRF tables to the Global table as EVPN Type 5

The current behavior is:

For neighbors attached to VRFs, IPv4/6 routes are imported as VPNv4/6 to the Global RIB. For example:

root@9652c769b373:/go# gobgp vrf blue neighbor
Peer          AS  Up/Down State       |#Received  Accepted
172.17.0.5 65001 00:00:48 Establ      |        1         1
172.17.0.6 65001 00:00:48 Establ      |        1         1

root@9652c769b373:/go# gobgp vrf blue neighbor
Peer          AS  Up/Down State       |#Received  Accepted
172.17.0.5 65001 00:00:48 Establ      |        1         1
172.17.0.6 65001 00:00:48 Establ      |        1         1

root@9652c769b373:/go# gobgp vrf blue rib
   Network              Next Hop             AS_PATH              Age        Attrs
*  10.0.0.0/24          172.17.0.5                                00:01:50   [{Origin: ?} {LocalPref: 100}]

root@9652c769b373:/go# gobgp vrf red rib
   Network              Next Hop             AS_PATH              Age        Attrs
*  10.0.0.0/24          172.17.0.2                                00:01:50   [{Origin: ?} {LocalPref: 100}]

root@9652c769b373:/go# gobgp global rib -a vpnv4
   Network              Labels     Next Hop             AS_PATH              Age        Attrs
*> 10:10:10.0.0.0/24    [0]        172.17.0.2                                00:02:43   [{Origin: ?} {LocalPref: 100} {Extcomms: [10:10]}]
*> 20:20:10.0.0.0/24    [0]        172.17.0.5                                00:02:43   [{Origin: ?} {LocalPref: 100} {Extcomms: [20:20]}]

After this change, we can configure the VRF to convert to EVPN Type 5 (IP Prefix) instead of VPNv4/v6:

config file:

...
...
[[vrfs]]
    [vrfs.config]
        name = "blue"
        rd = "20:20"
        both-rt-list = ["20:20"]
        import-as-evpn-ipprefix = true
        routers-mac = "ca:fe:00:00:fe:ed"
        ethernet-tag = 200
[[vrfs]]
    [vrfs.config]
        name = "red"
        rd = "10:10"
        both-rt-list = ["10:10"]
        import-as-evpn-ipprefix = true
        routers-mac = "ca:fe:00:00:be:ef"
        ethernet-tag = 100

and

root@dc00a25785dd:/go# gobgp vrf
  Name                 RD                   Import RT            Export RT            Router's MAC         ID    Import as EVPN       Ethernet Tag
  blue                 20:20                20:20                20:20                ca:fe:00:00:fe:ed    0     true                 200
  red                  10:10                10:10                10:10                ca:fe:00:00:be:ef    0     true                 100

root@dc00a25785dd:/go# gobgp global rib -a evpn
*> [type:Prefix][rd:20:20][etag:200][prefix:10.0.0.0/24] [0]        172.17.0.5                                00:00:41   [{Origin: ?} {LocalPref: 100} {Extcomms: [20:20], [router's mac: ca:fe:00:00:fe:ed]} [ESI: single-homed] [GW: 0.0.0.0]]
*> [type:Prefix][rd:10:10][etag:100][prefix:10.0.0.0/24] [0]        172.17.0.2                                00:00:28   [{Origin: ?} {LocalPref: 100} {Extcomms: [10:10], [router's mac: ca:fe:00:00:be:ef]} [ESI: single-homed] [GW: 0.0.0.0]]

[lucasolim]

Tests fixed

[fujita]

What use cases is this feature intended for?

[lucasolim]

The scenario where this feature would be useful is the following:

• A server running a hypervisor to deploy VMs
• The network virtualization layer is implemented by the DPU attached to the server
• VMs live in overlay networks (VXLAN), same server can deploys VMs in different VXLANs (VXLAN encap/decap layer is in the DPU)
• GoBGP runs in the DPU, that is a linux box
• Each VXLAN overlay network has its correspondent network namespace in the DPU, so VMs can access link-local services. For example, if a server deploys VM 1 in VXLAN 1, VM2 and VM3 in VXLAN2, the DPU implements 2 netns, one related to VXLAN1 and one related to VXLAN2
• GoBGP also runs in every netns, so VMs can peer with them to announce anycasts
• GoBGP in net namespaces peer with GoBGP in the standard namespace, each one in a separate VRF to allow overlapping anycast ranges
• IP4/6 Routes in the VRFs are imported to the global rib as EVPN Type 5 routes (instead of L3VPN)
• Anycast routes are propagated to other DPUs through Route Reflectors.

[fujita]

Any other BGP implementation to support such feature?