BGP peer

This container image provides a BGP peer.

The current implementation uses gobgpd for bgp protocol and FRR/zebra for FIB manipulation. But the idea is to have the implementation details abstracted away.

FIB manipulation currently requires some capability: NET_ADMIN, SYS_ADMIN, SETPCAP, NET_RAW

Run this container

Example for running this container:

# docker run \
	-e BGP_ROUTER_ID=192.0.2.1 \
	-e BGP_FIB_MANIPULATION=yes \
	openvnf/vnf-bgp

There is also a tests/ subfolder which has some scripts prepared for easy playground setup. Meet Alice and Bob there. They are neighbors and live in docker containers.

Configuration

Configuration is done by environment variables. See below for available options. Alternatively it is (will be) also possible to point $ENVFILE to an environment file which is sourced before startup.

# source specified file before startup.
ENVFILE=/tmp/generated-config.env

# the local AS:
BGP_LOCAL_AS=65000

# the identification (and peer address) of ourself
BGP_ROUTER_ID=192.0.2.1

# The simplest way of specifying neighbors
# (comma separated list of AS/ip pairs):
[email protected],[email protected]

# The more sophisticated way of specifying neighbors
# this allows more neighbor specific options. Counting
# starts at zero.
# (These neighbors will be added in addition to $BGP_NEIGHBORS)
BGP_NEIGHBOR_COUNT=3
BGP_NEIGHBOR_0_PEERAS=65030
BGP_NEIGHBOR_0_ADDRESS=192.0.2.30
BGP_NEIGHBOR_0_AUTHPASSWORD=secret
BGP_NEIGHBOR_0_LOCAL_AS=64999
BGP_NEIGHBOR_0_REMOVE_PRIVATE_AS=all # (none|all|replace)
BGP_NEIGHBOR_0_TIMER_CONNECT_RETRY=5
BGP_NEIGHBOR_0_TIMER_HOLD=9
BGP_NEIGHBOR_0_TIMER_KEEPALIVE=3
BGP_NEIGHBOR_0_MAX_IMPORT_PATH_LENGTH=1 (not yet implemented)
BGP_NEIGHBOR_1_...
BGP_NEIGHBOR_2_...

# statically inject routes into global rib immediately after startup:
BGP_STATIC_ROUTES=192.0.2.0/28,192.0.2.128/25

# do fib manipulation
BGP_FIB_MANIPULATION=yes

# put routes from FIB into RIB (announce routes found in kernel)
# (only applicable with BGP_FIB_MANIPULATION enabled)
BGP_FIB_ANNOUNCE=yes

# maximum acceptable as-path length for imports
# This option is deprecated. Use policy document instead.
# (Option is not available when policy document is
# specified.)
BGP_MAX_PATH=1

# BGP authentication password. Shared over all neighbors
# (Currently only one shared password for all neighbors.)
BGP_AUTHPASSWORD=secret

# Policy document. Path to a file containing the policy
# specification. The format is toml with syntax borrowed
# from gobgp:
# https://github.com/osrg/gobgp/blob/master/docs/sources/policy.md
BGP_POLICY_DOCUMENT=/volume1/bgp.policy

Caveats and Pitfalls

Routes pushed towards FIB maniupulation are considered "inactive" unless the next hop is considered directly connected by the FIB manipulator (currently being Frr/Zebra). Routes in the kernel table are not enough for being "directly connected". When you have your next hop connected via a vti interface without having an IP address in that network you can "connect" your neighbor with the peer specification of ip addr add:

ip addr add YOUR_IP/32 peer YOUR_PEER dev vti42

This will make YOUR_PEER appear as directly connected and allow FIB propagation to the kernel table(s).