Newansage v0.4.0 is out
ansage
Guides / Kernel routes

Kernel routes

Install the best routes learned from neighbors into the Linux routing table.

View as Markdown Edit on GitHub

With a [kernel] section, the best route of every prefix learned from a neighbor with import = "all" goes into the kernel's routing table (protocol bgp, ip route show proto bgp). Linux 5.3 or newer.

[kernel]
table = 254            # main (default)
metric = 20            # FRR's for eBGP (default)
families = ["ipv4", "ipv6"]
resolve_via_default = false

[[neighbor]]
address = "10.0.0.1"
asn = 64501
import = "all"
install = true         # default with [kernel]; false keeps this neighbor's routes out

Next hops

Each BGP next hop is one kernel nexthop object (ids from 3120562176 = 0xBA000000, ip nexthop show), so a next hop that moves is one change, not one per route.

  • A next hop is resolved through the kernel's own route to it: connected, or through that route's gateway (iBGP, multihop; for an ECMP route or a nexthop group, its first leg that is up). Routes learned by BGP and the default route do not count (resolve_via_default = true allows the latter).
  • An IPv6 link-local next hop (RFC 2545) from a neighbor on a shared link goes out of the session's interface: used when the neighbor sends only a link-local next hop, and when its global one does not resolve.
  • ansage watches interfaces, addresses and routes: when a next hop becomes unreachable its routes leave the kernel at once and the next best path takes over; what the kernel refuses is tried again every 30 s.
  • Our own prefixes (announce, also after a runtime withdraw) never go in.

Inspecting

ansage show fib lists the counts and every next hop (reachable through which gateway and interface, or not); ansage show rib ADDRESS shows the longest matching prefix and whether it is in the kernel. Metrics: ansage_fib_routes, ansage_fib_next_hops, ansage_fib_errors_total, ansage_fib_adopted_routes.

Permissions and lifecycle

A stop removes ansage's routes. ansage upgrade and a crash leave them: the next start keeps them forwarding, replaces each as its neighbor sends it again, and removes the rest once every installing neighbor has sent End-of-RIB (at most 6 minutes). A reload moves the routes to a new table or metric without a gap; adding or removing [kernel] needs a restart.

Esc to close