ferrix/docs/content/docs/guides/peers.md

+++ title = "Configuring BGP Neighbors" description = "Describes how to configure BGP neighbors, address families, link-local neighbors, ADD-PATH, and session timers, and how Ferrix filters inbound connections." weight = 1

[extra] nav = "BGP Neighbors" +++

A BGP neighbor (peer) is a BGP speaker with which Ferrix establishes a session over TCP. Each neighbor is defined by a peers statement in the configuration.

Configuring a Neighbor

The following example configures an external BGP (EBGP) neighbor:

peers {
  name: "transit-a"
  endpoint { address: "203.0.113.1" }
  remote_asn: 64496
  local_address: "203.0.113.2"
  hold_time_seconds: 90
  role: PROVIDER
  import { max_prefixes: 1500000 }
}

Table 1 describes the statements in the example.

Statement	Description
name	Name of the neighbor in logs, metrics, and the status page.
endpoint	Address and TCP port of the neighbor. The default port is 179.
remote_asn	AS number of the neighbor. If it equals local_asn, the neighbor is an internal BGP (IBGP) neighbor.
local_address	Source address for connections to the neighbor, and the next hop of routes advertised to it.
hold_time_seconds	Hold time proposed in the OPEN message.
role	BGP role of the neighbor (RFC 9234). See BGP Roles.
import	Import policy for routes received from the neighbor. See Import Policy.

Configure local_address when the router has several addresses and the neighbor expects a specific one.

Session Establishment

Ferrix initiates a TCP connection to each configured neighbor and also accepts connections from it. If both connections succeed, Ferrix resolves the collision as specified in RFC 4271 Section 6.8: the connection initiated by the speaker with the higher BGP identifier is retained.

If a connection attempt fails, Ferrix retries after connect_retry_seconds (default: 5 seconds). The interval doubles after each failure, up to a maximum of 2 minutes.

Ferrix rejects a neighbor whose OPEN message contains an AS number other than remote_asn, with the NOTIFICATION message OPEN Message Error / Bad Peer AS.

Passive Neighbors

A passive neighbor is a neighbor to which Ferrix does not initiate connections. Ferrix accepts connections from the neighbor's address and establishes the session with the configured settings when the neighbor connects. Use a passive neighbor to configure a session before the neighbor's side is ready, or when the neighbor initiates the connections:

peers {
  name: "new-peer"
  endpoint { address: "192.0.2.10" }
  remote_asn: 64510
  role: PEER
  passive: true
}

Until the neighbor connects, the status page and /api/peers show the state Passive. After a session ends, they show the reason in the neighbor's last error. Ferrix does not log connection attempts to a passive neighbor, because it makes none.

Inbound Connection Filtering

Ferrix accepts inbound connections only from the addresses of configured neighbors, and limits the number of concurrent connections from each address. Ferrix closes any other connection before reading from it and increments the bgp_connections_refused_total metric.

Ferrix treats each configured neighbor as a single router. A second connection from the same address with a different BGP identifier is rejected. This prevents a neighbor from exceeding its prefix limit by establishing additional sessions.

Caution

The accept_unconfigured_peers statement accepts sessions from any address, with any AS number, and accepts their routes. Use it only for testing on an isolated network.

Listening Addresses

By default, Ferrix listens on 0.0.0.0:179 and [::]:179. To listen on specific addresses, configure listen_endpoints:

listen_endpoints { address: "192.0.2.1" }
listen_endpoints { address: "2001:db8::1" }

Address Families

Ferrix supports the IPv4 unicast and IPv6 unicast address families. Each speaker lists the families that it supports in its OPEN message, and a family is enabled on the session only if both speakers list it (RFC 4760). By default, Ferrix proposes both families. To propose only IPv6 unicast:

address_families: IPV6_UNICAST

If the neighbor supports none of the proposed families, Ferrix rejects the session.

IPv4 Routes with IPv6 Next Hops

Ferrix supports the advertisement of IPv4 routes with IPv6 next hops over a session between IPv6 addresses (RFC 8950). On every IPv6 session that carries IPv4 unicast, Ferrix proposes the Extended Next Hop capability; no configuration is required. On the neighbor, enable the capability as follows:

• FRR: neighbor ... capability extended-nexthop
• BIRD: extended next hop on in the IPv4 channel

When the capability is negotiated, Ferrix advertises IPv4 routes with the session's local IPv6 address as the next hop. It accepts IPv4 routes with an IPv6 next hop, or with a global and a link-local next hop. The kernel installs such routes through the IPv6 neighbor, for example 10.2.0.0/24 via inet6 2001:db8::2 dev eth0. This requires Linux 5.2 or later. The status page displays "IPv4 with IPv6 next hops" for the session.

Link-Local Neighbors

Ferrix can establish a session with a neighbor at an IPv6 link-local address (fe80::/10). Because a link-local address is significant only on one link, you specify the interface with the address:

peers {
  name: "upstream"
  endpoint { address: "fe80::1%eno3" }
  remote_asn: 64496
}

Link-local neighbors have the following behavior:

• Ferrix resolves the interface name to an interface index each time it is needed. If the interface does not exist, the neighbor state displays that the interface is missing; the session is established when the interface is created. An interface that is recreated with a new index, such as a restarted tunnel, is found automatically.
• The same link-local address on two interfaces identifies two neighbors. Ferrix rejects a link-local address without an interface, and an interface specified with a non-link-local address.
• To accept connections from link-local neighbors, Ferrix must listen on ::, which is the default.
• Ferrix advertises routes with its link-local address as the next hop, in both next hop fields of MP_REACH_NLRI. It accepts a link-local next hop in any of the encodings in use: the link-local address alone, the link-local address twice, or the link-local address after ::.
• The kernel installs the routes through the link-local address on the session's interface, for example 10.2.0.0/24 via inet6 fe80::1 dev eno3.
• If local_address is a link-local address, Ferrix uses it on the neighbor's interface.

Tip

On FRR, configure the neighbor by interface (neighbor eno3 interface remote-as ...). FRR discovers the neighbor's address from IPv6 router advertisements, which Ferrix does not send; run radvd on the interface. With the neighbor fe80::1 interface eno3 form, FRR 10.5 establishes the session but marks the received routes as inaccessible. On BIRD, configure neighbor fe80::1%eno3 as ....

ADD-PATH

By default, a neighbor advertises one path for each prefix. With ADD-PATH (RFC 7911), a neighbor can advertise multiple paths for the same prefix. To accept multiple paths from a neighbor:

add_path: ADD_PATH_RECEIVE

ADD-PATH is enabled for each address family and direction only if the neighbor also supports it. Ferrix negotiates ADD_PATH_SEND and ADD_PATH_BOTH, but it advertises only one path for each prefix.

Hold Timer

hold_time_seconds specifies the hold time that Ferrix proposes in its OPEN message. The session uses the lower of the two proposed values. If no message is received from the neighbor within the hold time, Ferrix tears down the session.

Range: 0, or 3 through 65535 seconds. A value of 0 disables keepalive messages and the hold timer. Default: 90 seconds.

Ferrix also implements the send hold timer (RFC 9687). If Ferrix cannot write to the neighbor for 8 minutes or twice the hold time, whichever is longer, it tears down the session. This prevents a neighbor that does not read from the connection from blocking route advertisement indefinitely.

Session Flap Damping

If sessions with a neighbor are established and then fail within 1 minute repeatedly, Ferrix places the neighbor in hold-down. The hold-down begins at the third such failure with a duration of 1 second, and doubles at each further failure, up to a maximum of 5 minutes. During hold-down, Ferrix does not establish a session with the neighbor.

The end of a session requires Ferrix to remove the neighbor's routes from the routing table. Hold-down limits the processing that a flapping neighbor causes.

A neighbor that exceeds its prefix limit is placed in hold-down for hold_down_seconds. See Prefix Limits.

Administrative Shutdown

When the daemon receives SIGINT or SIGTERM, it sends a Cease NOTIFICATION message with the subcode Administrative Shutdown to every neighbor. If the kernel statement is configured, it then removes its routes from the kernel. See Kernel Route Installation.

Verifying BGP Sessions

To verify the state of each neighbor, use one of the following:

• The status page, /statusz, lists each neighbor with its session state, address families, role, and number of routes.
• The HTTP API, /api/peers. See HTTP API Reference.
• The metric bgp_peer_up{peer}, which is 1 while the session is established and 0 otherwise.