How-To
Services
Internal
Historical
External Tools
Original Article: https://blog.sherpherd.top/2024/02/11/RunYourMPLSNetworkWithBIRD_en.html
Now, most tutorials about running MPLS on Linux are based on FRR. Because in a long time, FRR and its predecessor Quagga are the only choices who provide industry standard MPLS related protocol (LDP, BGP-LU, BGP IPv4/IPv6 MPLS L3VPN, etc.) implementation, by the time, most of other routing software don't even have availiable MPLS support.
The most popular routing software among DN42 users, BIRD, has added availiable MPLS support in its newest version (2.14), now BIRD has MPLS-aware, labeled route producing routing protocol, too. (No LDP still though) [1]
The newest BIRD 2.0 User Guide has added MPLS function related chapter[2], this is an excerpt:
In BIRD, the whole process generally works this way: A MPLS-aware routing protocol (say BGP) receives routing information including remote label. It produces a route with attribute mpls policy (p. 30) specifying desired MPLS label policy (p. 18). Such route then passes the import filter (which could modify the MPLS label policy or perhaps assign a static label) and when it is accepted, a local MPLS label is selected (according to the label policy) and attached to the route, producing labeled route. When a new MPLS label is allocated, the MPLS-aware protocol automatically produces corresponding MPLS route. When all labeled routes that use specific local MPLS label are retracted, the corresponding MPLS route is retracted too.
There are three important concepts for MPLS in BIRD: MPLS domains, MPLS tables and MPLS channels. MPLS domain represents an independent label space, all MPLS-aware protocols are associated with some MPLS domain. It is responsible for label management, handling label allocation requests from MPLS-aware protocols. MPLS table is just a routing table for MPLS routes. Routers usually have one MPLS domain and one MPLS table, with Kernel protocol to export MPLS routes into kernel FIB.
MPLS channels make protocols MPLS-aware, they are responsible for keeping track of active FECs (and corresponding allocated labels), selecting FECs / local labels for labeled routes, and maintaining correspondence between labeled routes and MPLS routes.
As mentioned above, the current BGP implementation of BIRD is MPLS-awared, can be used to assign and distribute MPLS labeled route.
In this article, I will make use of official BIRD document to show readers how to construct a simple MPLS VPN network running with BIRD.
----------------------------------------------------------
-------- | -------- -------- -------- | --------
| | eth0 ens20 | | ens19 ens19| | ens20 ens19| | ens20 eth0 | |
| PC1 | O==========O | R1 |O==========O| R3 |O==========O| R2 | O==========O | PC2 |
| | | | | | | | | | | |
-------- | -------- -------- -------- | --------
| |
| Confederation ASN: 100 |
| R1 ASN: 64512 |
| R2 ASN: 64513 |
| R3 ASN: 64514 |
| RT: 100:500 |
| R1 RT: 203.0.113.1:500 |
| R2 RT: 203.0.113.2:500 |
| |
----------------------------------------------------------
| Address Assignment |
----------------------------------------------------------
| |
| eth0@PC1: 192.168.1.2/24 |
| ens20@R1: 192.168.1.1/24 |
| (vrf blue) |
| lo@R1: 203.0.113.1/32 |
| ens19@R1: 203.0.113.1/32 |
| lo@R3: 203.0.113.3/32 |
| ens19@R3: 203.0.113.3/32 |
| ens20@R3: 203.0.113.3/32 |
| lo@R2: 203.0.113.2/32 |
| ens19@R2: 203.0.113.2/32 |
| ens20@R2: 192.168.2.1/24 |
| (vrf blue) |
| eth0@PC2: 192.168.2.2/24 |
| |
----------------------------------------------------------
PC1, R1, R2 and PC2 all running Debian 12. R1 and R2 both installed newest version BIRD by the time I finished this, the BIRD 2.14.
Notice: Remember to add third port for R1, R2 and R3 to make them able to access Internet for downloading BIRD software package or compiling dependencies
Run these command on R1 and R2 with root permission:
modprobe mpls_router
modprobe mpls_iptunnel
modprobe mpls_gso
Run these command on R1, R2 and R3 with root permission to adjust parameters related to IP routing and MPLS, make them able to work[3]:
cat >/etc/sysctl.d/90-mpls-router.conf <<EOF
net.ipv4.ip_forward=1
net.ipv6.conf.all.forwarding=1
net.ipv4.conf.all.rp_filter=0
net.mpls.platform_labels=1048575
net.ipv4.tcp_l3mdev_accept=1
net.ipv4.udp_l3mdev_accept=1
net.mpls.conf.lo.input=1
EOF
sysctl -p /etc/sysctl.d/90-mpls-router.conf
Every port transits MPLS traffic need to enable MPLS input, run these command on R1 and R2 with root permission to enable MPLS input for their port ens19:
sysctl -w net.mpls.conf.ens19.input=1
So do on R3:
sysctl -w net.mpls.conf.ens19.input=1
sysctl -w net.mpls.conf.ens20.input=1
Notice: Every MPLS traffic transiting port need this configuration
Run these command on R1 and R2 with root permission to create a VRF interface named "blue":
ip link add blue type vrf table 500
ip link set blue up
Run these command on R1 and R2 with root permission to assign ens20 to VRF blue then enable it:
ip link set ens20 master blue up
In practice, increasing MTU of core network link is always harder than decreasing client-faced port MTU, and using MPLS incur additional packet header overhead (4 bytes per label), this made large packet may get fragmented when entering MPLS network. To avoid this, we need to approviately decrease the MTU of client-faced port.
Run these command on PC1 and PC2 with root permission to adjust MTU of eth0 then enable it:
ip link set eth0 mtu 1492 up
Run these command on R1 and R2 with root permission to adjust MTU of ens20:
ip link set ens20 mtu 1492
Run command with root permission on nodes below to done this. R1:
ip addr add 203.0.113.1/32 dev lo
ip addr add 203.0.113.1/32 dev ens19 peer 203.0.113.3/32
ip addr add 192.168.1.1/24 dev ens20
R2:
ip addr add 203.0.113.2/32 dev lo
ip addr add 203.0.113.2/32 dev ens19 peer 203.0.113.3/32
ip addr add 192.168.2.1/24 dev ens20
R3:
ip addr add 203.0.113.3/32 dev lo
ip addr add 203.0.113.3/32 dev ens19 peer 203.0.113.1/32
ip addr add 203.0.113.3/32 dev ens20 peer 203.0.113.2/32
PC1:
ip addr add 192.168.1.2/24 dev eth0
ip route add 192.168.2.0/24 via 192.168.1.1
PC1:
ip addr add 192.168.2.2/24 dev eth0
ip route add 192.168.1.0/24 via 192.168.2.1
The compile installed BIRD is incomplete, it lacks system service file, docs, etc.
If you want complete BIRD, you have to build software package then install from it.
If you don't want build yourself, you can download them here (deb package):
https://drive.google.com/drive/folders/1DUaFJgZGsEXI-RlreNxCG9mnERiAkIVB?usp=drive_link
If hints dependency miss, follow the hint to install missed dependency.
If you want to build it yourself, please read the rest of this chapter.
Prepare a compile node running Debian 12, no need of high spec, mine got 4 cores and 4 gigs of RAM, then do as follow.
apt install -y git linuxdoc-tools autoconf build-essential libssh-dev libreadline-dev libncurses-dev flex bison checkinstall debhelper docbook-xsl libssh-gcrypt-dev quilt xsltproc linuxdoc-tools-latex texlive-latex-extra
pipx install apkg
git clone --branch v2.14 https://gitlab.nic.cz/labs/bird.git
Enter "bird" then run command below:
apkg build
Once finished, apkg gives hint like this:
built 3 packages in: pkg/pkgs/debian-12/bird2_2.14.1707409394.0e1fbaa5-cznic.1
The built software package is located in the location the hint mentioned, make use of it.
Currently BIRD don't have implementation of distributing MPLS labeled route through IGP topo, so we use BGP-LU to do that.
Having a static Router ID is always not a bad thing.
R1:
router id 203.0.113.1;
So do on R2 and R3.
Use R1 as example, so do on other nodes:
mpls domain mpls_dom;
mpls table bgp_mpls_table;
vpn4 table bgp_vpn4;
ipv4 table vrf_blue4; # This one is no need on R3
Use R1 as example:
protocol kernel krt_mpls {
mpls {
table bgp_mpls_table;
export all;
};
}
protocol kernel vrf_blue_4 { # No need for R3, since it doesn't run any VRF instance
vrf "blue";
ipv4 {
table vrf_blue4;
export all;
import all;
};
kernel table 500;
}
protocol static {
ipv4;
route 203.0.113.1/32 reject; # Inject direct route through static, for advertising it in BGP
}
protocol static { # Same, no need for R3
ipv4 { table vrf_blue4; };
route 192.168.1.0/24 reject;
}
Use R1 as example:
protocol bgp r3 {
local 203.0.113.1 as 64512;
neighbor 203.0.113.3 as 64514;
confederation 100;
confederation member;
ipv4 mpls { # Enable IPv4 Labeled Unicast channel, to enable MPLS reachability between MPLS nodes
import all;
export all;
};
vpn4 mpls { # Enable VPNv4 channel, carring IPv4 VPN route
table bgp_vpn4;
import all;
export all;
};
mpls {
label policy aggregate;
};
}
No need for R3, it doesn't run any VRF instance.
Use R1 as example:
protocol l3vpn vpn_blue4 {
vrf "blue";
ipv4 { table vrf_blue4; }; # # Binding VRF Table
vpn4 { table bgp_vpn4; }; # Binding VPNv4 Table
mpls { label policy vrf; };
rd 203.0.113.1:500;
import target [(rt,100,500)]; # Define RT the desired route import from binded VPNv4 to VRF have
export target [(rt,100,500)]; # Define RT the route export from VRF to binded VPNv4 will be attached
}
log syslog all;
router id 203.0.113.1;
mpls domain mpls_dom;
mpls table bgp_mpls_table;
vpn4 table bgp_vpn4;
ipv4 table vrf_blue4;
protocol device {
}
protocol direct {
disabled; # Disable by default
ipv4; # Connect to default IPv4 table
ipv6; # ... and to default IPv6 table
}
protocol kernel {
ipv4 { # Connect protocol to IPv4 table by channel
export all; # Export to protocol. default is export none
import all;
};
}
protocol kernel {
ipv6 { export all; };
}
protocol kernel krt_mpls {
mpls {
table bgp_mpls_table;
export all;
};
}
protocol kernel vrf_blue_4 {
vrf "blue";
ipv4 {
table vrf_blue4;
export all;
import all;
};
kernel table 500;
}
protocol static {
ipv4; # Again, IPv4 channel with default options
route 203.0.113.1/32 reject;
}
protocol static {
ipv4 { table vrf_blue4; };
route 192.168.1.0/24 reject;
}
protocol bgp r3 {
local 203.0.113.1 as 64512;
neighbor 203.0.113.3 as 64514;
confederation 100;
confederation member;
ipv4 mpls {
import all;
export all;
};
vpn4 mpls {
table bgp_vpn4;
import all;
export all;
};
mpls {
label policy aggregate;
};
}
protocol l3vpn vpn_blue4 {
vrf "blue";
ipv4 { table vrf_blue4; };
vpn4 { table bgp_vpn4; };
mpls { label policy vrf; };
rd 203.0.113.1:500;
import target [(rt,100,500)];
export target [(rt,100,500)];
}
router id 203.0.113.2;
log syslog all;
mpls domain mpls_dom;
mpls table bgp_mpls_table;
vpn4 table bgp_vpn4;
ipv4 table vrf_blue4;
protocol device {
}
protocol direct {
disabled; # Disable by default
ipv4; # Connect to default IPv4 table
ipv6; # ... and to default IPv6 table
}
protocol kernel krt_mpls {
mpls {
table bgp_mpls_table;
export all;
};
}
protocol kernel vrf_blue_4 {
vrf "blue";
ipv4 {
table vrf_blue4;
export all;
import all;
};
kernel table 500;
}
protocol kernel {
ipv4 { # Connect protocol to IPv4 table by channel
export all; # Export to protocol. default is export none
};
}
protocol kernel {
ipv6 { export all; };
}
protocol static {
ipv4; # Again, IPv4 channel with default options
route 203.0.113.2/32 reject;
}
protocol static {
ipv4 { table vrf_blue4; };
route 192.168.2.0/24 reject;
}
protocol bgp r3 {
local 203.0.113.2 as 64513;
neighbor 203.0.113.3 as 64514;
confederation 100;
confederation member;
ipv4 mpls {
import all;
export all;
};
vpn4 mpls {
table bgp_vpn4;
import all;
export all;
};
mpls {
label policy aggregate;
};
}
protocol l3vpn vpn_blue4 {
vrf "blue";
ipv4 { table vrf_blue4; };
vpn4 { table bgp_vpn4; };
mpls { label policy vrf; };
rd 203.0.113.2:500;
import target [(rt,100,500)];
export target [(rt,100,500)];
}
log syslog all;
router id 203.0.113.3;
mpls domain mpls_dom;
mpls table bgp_mpls_table;
vpn4 table bgp_vpn4;
protocol device {
}
protocol direct {
disabled; # Disable by default
ipv4; # Connect to default IPv4 table
ipv6; # ... and to default IPv6 table
}
protocol kernel {
ipv4 { # Connect protocol to IPv4 table by channel
export all; # Export to protocol. default is export none
};
}
protocol kernel {
ipv6 { export all; };
}
protocol kernel krt_mpls {
mpls {
table bgp_mpls_table;
export all;
};
};
protocol static {
ipv4; # Again, IPv4 channel with default options
}
protocol bgp r1 {
local 203.0.113.3 as 64514;
neighbor 203.0.113.1 as 64512;
confederation 100;
confederation member;
ipv4 mpls {
next hop self;
import all;
export all;
};
vpn4 mpls {
next hop self;
table bgp_vpn4;
import all;
export all;
};
mpls {
label policy aggregate;
};
}
protocol bgp r2 {
local 203.0.113.3 as 64514;
neighbor 203.0.113.2 as 64513;
confederation 100;
confederation member;
ipv4 mpls {
next hop self;
import all;
export all;
};
vpn4 mpls {
next hop self;
table bgp_vpn4;
import all;
export all;
};
mpls {
label policy aggregate;
};
}
R1:
bird> show route table bgp_vpn4
Table bgp_vpn4:
203.0.113.2:500 192.168.2.0/24 mpls 1001 unicast [r3 23:20:55.236] * (100) [AS64513i]
via 203.0.113.3 on ens19 mpls 1002
203.0.113.1:500 192.168.1.0/24 mpls 1002 unicast [vpn_blue4 22:58:48.918] * (120/0)
dev blue
bird>
R2:
bird> show route table bgp_vpn4
Table bgp_vpn4:
203.0.113.2:500 192.168.2.0/24 mpls 1001 unicast [vpn_blue4 23:20:55.219] * (120/0)
dev blue
203.0.113.1:500 192.168.1.0/24 mpls 1002 unicast [r3 22:58:56.352] * (100) [AS64512i]
via 203.0.113.3 on ens19 mpls 1003
bird>
R1:
bird> show route table master4
Table master4:
203.0.113.2/32 mpls 1000 unicast [r3 22:58:56.355] * (100) [AS64513i]
via 203.0.113.3 on ens19 mpls 1000
203.0.113.1/32 unreachable [static1 22:33:27.446] * (200)
bird>
R2:
bird> show route table master4
Table master4:
203.0.113.2/32 unreachable [static1 22:32:38.874] * (200)
203.0.113.1/32 mpls 1000 unicast [r3 22:58:56.352] * (100) [AS64512i]
via 203.0.113.3 on ens19 mpls 1001
bird>
R1:
bird> show route table vrf_blue4
Table vrf_blue4:
192.168.1.0/24 unreachable [static2 22:58:48.918] * (200)
192.168.2.0/24 unicast [vpn_blue4 23:20:55.236] * (80/0)
via 203.0.113.3 on ens19 mpls 1002
bird>
R2:
bird> show route table vrf_blue4
Table vrf_blue4:
192.168.1.0/24 unicast [vpn_blue4 23:20:55.219] * (80/0)
via 203.0.113.3 on ens19 mpls 1003
192.168.2.0/24 unreachable [static2 22:54:38.777] * (200)
bird>
PC1:
root@pc1:~# ping -c 4 192.168.2.2
PING 192.168.2.2 (192.168.2.2) 56(84) bytes of data.
64 bytes from 192.168.2.2: icmp_seq=1 ttl=61 time=5.53 ms
64 bytes from 192.168.2.2: icmp_seq=2 ttl=61 time=5.03 ms
64 bytes from 192.168.2.2: icmp_seq=3 ttl=61 time=3.73 ms
64 bytes from 192.168.2.2: icmp_seq=4 ttl=61 time=5.97 ms
--- 192.168.2.2 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3005ms
rtt min/avg/max/mdev = 3.729/5.063/5.965/0.838 ms
root@pc1:~# traceroute 192.168.2.2
traceroute to 192.168.2.2 (192.168.2.2), 30 hops max, 60 byte packets
1 192.168.1.1 (192.168.1.1) 5.787 ms 6.165 ms *
2 * * *
3 * * *
4 192.168.2.2 (192.168.2.2) 36.865 ms 37.489 ms 44.775 ms
root@pc1:~#
PC2:
root@pc2:~# ping -c 4 192.168.1.2
PING 192.168.1.2 (192.168.1.2) 56(84) bytes of data.
64 bytes from 192.168.1.2: icmp_seq=1 ttl=61 time=21.7 ms
64 bytes from 192.168.1.2: icmp_seq=2 ttl=61 time=4.35 ms
64 bytes from 192.168.1.2: icmp_seq=3 ttl=61 time=13.6 ms
64 bytes from 192.168.1.2: icmp_seq=4 ttl=61 time=4.67 ms
--- 192.168.1.2 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3007ms
rtt min/avg/max/mdev = 4.352/11.098/21.731/7.181 ms
root@pc2:~# traceroute 192.168.1.2
traceroute to 192.168.1.2 (192.168.1.2), 30 hops max, 60 byte packets
1 192.168.2.1 (192.168.2.1) 17.272 ms 17.125 ms 17.175 ms
2 * * *
3 * * *
4 192.168.1.2 (192.168.1.2) 27.517 ms 27.945 ms 32.354 ms
root@pc2:~#
1. BIRD Team. (2023, October 7). News Archive. bird.network.cz. https://bird.network.cz/?o_news/
2. BIRD Team. (2023, October 7). BIRD 2.0 User’s Guide. MPLS, 9-10. https://bird.network.cz/download/bird-doc-2.14.tar.gz
3. James Swineson. (2020, February 22). Use Linux as an MPLS Router. blog.swineson.me. https://blog.swineson.me/en/use-linux-as-an-mpls-router/
Hosted by: BURBLE-MNT, GRMML-MNT, XUU-MNT, JAN-MNT, LARE-MNT, SARU-MNT, ANDROW-MNT, MARK22K-MNT | Accessible via: dn42, dn42.dev, dn42.eu, wiki.dn42.us, dn42.de (IPv6-only), dn42.cc (wiki-ng), dn42.wiki, dn42.pp.ua, dn42.obl.ong
Last edited by famfo, 2024-02-14 08:58:02