Discrepancy Between Kernel Routing Table and Zebra's Internal Routing Table for "proto kernel" Routes

[omarshaarawi]

Hello,

I am investigating a setup where Zebra manages routing on my system, and I came upon a discrepancy between the kernel routing table and Zebra's internal routing table regarding "proto kernel" routes. Specifically, certain "proto kernel" routes visible in the kernel routing table do not appear in Zebra's internal routing table.

One clarification I'd like to make is that these routes are not associated with interfaces that have addresses on the host. Unlike the routes of addressed interfaces that Zebra installs regardless of their presence in the kernel routing table, these routes seem to be ignored by Zebra.

Here's a snippet from the kernel routing table:

ip route show
default nhid 80 proto bgp metric 20
        nexthop via inet6 fe80::973c:s3fc:redc:6v18 dev enp24s0f1.100 weight 1
        nexthop via inet6 fe80::894c:b4fa:fvs6:7v31 dev enp24s0f0.100 weight 1
10.10.16.147 dev hn-26r2r-host scope link
10.10.22.5 dev veth_26s5h scope link
10.10.22.6 dev hn-26s5i-host scope link
10.10.22.7 dev hn-26s5j-host scope link
10.10.22.8 dev hn-26s5k-host proto kernel scope link
10.10.22.10 dev hn-26s5m-host scope link
10.10.22.11 dev hn-26s5l-host proto kernel scope link
172.17.0.0/16 dev docker0 proto kernel scope link src 172.17.0.1

And here's a snippet from Zebra's routing table:

show ip route
Codes: K - kernel route, C - connected, S - static, R - RIP,
       O - OSPF, I - IS-IS, B - BGP, E - EIGRP, N - NHRP,
       T - Table, v - VNC, V - VNC-Direct, A - Babel, F - PBR,
       f - OpenFabric,
       > - selected route, * - FIB route, q - queued, r - rejected, b - backup
       t - trapped, o - offload failure

B>* 0.0.0.0/0 [20/0] via fe80::973c:s3fc:redc:6v18, enp24s0f1.100, weight 1, 22:08:20
  *                  via fe80::894c:b4fa:fvs6:7v31, enp24s0f0.100, weight 1, 22:08:20
K>* 10.10.16.147/32 [0/0] is directly connected, hn-26r2r-host, 22:08:26
C>* 10.10.18.26/32 is directly connected, lo, 22:08:26
C>* 10.10.18.27/32 is directly connected, br0, 22:08:26
C>* 10.10.22.4/32 is directly connected, cilium_host, 22:08:26
K>* 10.10.22.5/32 [0/0] is directly connected, veth_26s5h, 22:08:26
K>* 10.10.22.6/32 [0/0] is directly connected, hn-26s5i-host, 22:08:26
K>* 10.10.22.7/32 [0/0] is directly connected, hn-26s5j-host, 22:08:26
K>* 10.10.22.10/32 [0/0] is directly connected, hn-26s5m-host, 22:08:26
C>* 172.17.0.0/16 is directly connected, docker0, 22:08:26

The routes for 10.10.22.8 and 10.10.22.11 are missing in Zebra's table

I have reviewed the source code and came across the netlink_route_change_read_unicast_internal function. It seems to have a check for rtm->rtm_protocol == RTPROT_KERNEL that causes the function to return early without further processing of the "proto kernel" routes.

These routes are created by Cilium, and I am curious to understand if this behavior is intended or if there's a configuration within Zebra to control the import of "proto kernel" routes from the kernel routing table.

I'd appreciate any insight or guidance on this issue

[tkgeng]

Same problem cilium/cilium#31243