This is solved. The problem was that the "import all" line in my OSPF protocols was supposed to be calling a filter to restrict the imported routes, but it had been removed for testing. So, nat boxes would import a default route from via OSPF from the other nat box if it was already started, which is not what we want. As to why it would work when using the static protocol to inject the default route, I'm not sure, but with redistribution, the default route is exported to the OSPF protocol twice, and with the static protocol it's only done once. There would likely be a difference in the way announcements are made. As to the distance of 65535 shown by "show ospf state", that was a red herring. For whatever reason, on the DR only, I see "normal" distances that appear to be the sum of the costs, and on every other router than the DR I see 65535 for the distance of all routers, except the local router, which shows 0. It seems that during testing, I thought this difference was due to switching between static and redistributed routes, but it was actually due to a change in the router's DR status. I do not see these 65535 distance values in a production environment on routers running BIRD 1.x, so either it has to do with the BIRD version, or something else in my setup. We'll see... Thanks, Pete On Thu, 2025-07-10 at 10:33 +0200, Pete Heist wrote:
Hi,
I'm seeing a problem with route redistribution that I hope someone can help with...
When redistributing a default route from iBGP to OSPF, other OSPF routers do not seem to respect the value of ospf_metric2 set on exported routes. If I define the default route in a static protocol instead of redistributing it, I do not see the problem.
I *think* I see a clue in the output from "show ospf state" below in the two cases. When using a static protocol, there are different route distances for each default route- 0 and 10. When doing route redistribution, both routers show a distance of 65535. That seems odd, and I'm not sure why it is. Also note that all iBGP peerings are IPv6, so BGP-MP is in use.
The topology is as follows:
^ ^ | | eBGP eBGP | | +------+ +------+ | gw1 |-----------------------------------------| gw2 | +--+---+ <-- 0.0.0.0/0 ---+---+--+ | | \ iBGP / | | | | \ / | | | | \ / | | | | \ / | | | 0.0.0.0/0 \ / 0.0.0.0/0 | | | iBGP \ \ / / iBGP | | | v v \ / v v | | \ / | | \ / | | \ / | | x | | / \ | | / \ | | / \ | | ^ ^ / \ ^ ^ | | | iBGP / / \ \ iBGP | | | A.B.C.D/X / \ A.B.C.D/X | | | / \ | | | | / \ | | | | / \ | | | | / \ | | +--+---+ +---+--+ | nat1 |-----------------------------------------| nat2 | +------+ +------+ | | OSPF OSPF E2 metric 10 E2 metric 100 0.0.0.0/0 0.0.0.0/0 | | v v
* Four routers in an iBGP full mesh * gw1 and gw2 are the ASBRs, with full BGP tables * nat1 and nat2 are core routers doing NAT44 * NOTE: all iBGP peerings are IPv6 only, so BGP-MP is in use * Default route: * exported by gw1 and gw2 using iBGP * imported by nat1 and nat2 using iBGP * redistributed by nat1 and nat2 using OSPF into the LAN * Different metrics from nat1 and nat2 seem not to be respected * Problem occurs on both BIRD 2.0.12 and 2.17.1
--------------------------------------------------------------------- --
STATIC PROTOCOL (what works)
If I define default routes in a static protocol on nat1 and nat2, and set ospf_metric2 on an export filter, it works as expected. Routers on the LAN respect the metric and prefer the route from nat1.
# # Routing table on another OSPF host running BIRD # # Regardless of which order nat1 and nat2 are started in, the # default route chosen is the one with the lower metric, as expected. #
bird> show ospf state ospf4
area 0.0.0.0
router <gw1 IPv4 IP> distance 0 ... external 0.0.0.0/0 metric2 10
router <gw2 IPv4 IP> distance 10 ... external 0.0.0.0/0 metric2 100
bird> show route for 0.0.0.0/0 all Table master4: 0.0.0.0/0 unicast [ospf4 09:26:57.341] * E2 (150/65535/10) [<gw1 IPv4 IP>] via <gw1 IPv4 IP> on eth0 Type: OSPF-E2 univ OSPF.metric1: 65535 OSPF.metric2: 10 OSPF.tag: 0x00000000 OSPF.router_id: <gw1 IPv4 IP>
# # Partial config from gw1/gw2... #
# NO_KERNEL_EXPORT, if defined, flags a route for NOT exporting to the # kernel. attribute int NO_KERNEL_EXPORT;
# IBGP_EXPORT, if defined, flags a route for export to iBGP peers. attribute int IBGP_EXPORT;
ipv4 table Tibgp4;
# default route injected on gw1/gw2 with static protocol into custom table protocol static Sibgp4 { ipv4 { table Tibgp4; import filter { IBGP_EXPORT = 1; NO_KERNEL_EXPORT = 1; accept; }; }; route 0.0.0.0/0 unreachable; }
# pipe to master protocol pipe Pibgp4 { table master4; peer table Tibgp4; import where !defined(NO_KERNEL_EXPORT); export where defined(IBGP_EXPORT); }
# sample peering for nat1, with a bunch of things left out protocol bgp nat1 from IBGP { ipv4 { ... table Tibgp4; export where defined(IBGP_EXPORT); }; ipv6 { ... table Tibgp6; export where defined(IBGP_EXPORT); }; # anonymized addresses below, but shows that IPv6 LLA is used interface "bond0.A"; source address fe80::A:1; neighbor fe80::A:2 internal; ... }
# # Partial config for nat1/nat2... #
# Default route added in a static protocol. This works. protocol static Sospf4 { ipv4 { table Tospf4; import filter { OSPF_EXPORT = 1; accept; }; }; route 0.0.0.0/0 unreachable; }
# ospf_metric2 set on an export filter. protocol ospf v2 ospf4 { ipv4 { table Tospf4; import all; export filter { if !defined(OSPF_EXPORT) then reject; ospf_metric2 = 10; # for nat1, or 100 for nat2 accept; }; }; ... }
--------------------------------------------------------------------- --
ROUTE REDISTRIBUTION (what does not work)
# # Routing table on another OSPF host running BIRD # # !!! Even though nat1 and nat2 are both running, and nat2 has a # lower value for OSPF.metric2, its default route is not chosen. # The default route in the OSPF database is from the router that # started first.
bird> show ospf state ospf4
area 0.0.0.0 ... router <gw1 IPv4 IP> # !!! distance is 65535- with static protocol it's 0 distance 65535 ... # default route not present when started second
router <gw2 IPv4 IP> # !!! distance is 65535- with static protocol it's 10 distance 65535 ... # external default route with higher metric 100 is used external 0.0.0.0/0 metric2 100
# The default route is gw2, which is unexpected when it has a # higher metric2. bird> show route for 0.0.0.0/0 all Table master4: 0.0.0.0/0 unicast [ospf4 08:37:10.342] * E2 (150/65535/100) [<gw2 IPv4 IP>] via <gw2 IPv4 IP> on eth0 Type: OSPF-E2 univ OSPF.metric1: 65535 OSPF.metric2: 100 OSPF.tag: 0x00000000 OSPF.router_id: <gw2 IPv4 IP>
# # Config for gw1/gw2 is same as above, when using static protocol... #
# # Partial config from nat1/nat2... #
# OSPF_EXPORT, if defined, flags a route for export via OSPF. attribute int OSPF_EXPORT;
ipv4 table Tibgp4;
# ibgp_in_primary4 filters primary IPv4 routes imported via iBGP. filter ibgp_in_primary4 { if net = 0.0.0.0/0 then { OSPF_EXPORT = 1; } bgp_local_pref = 100; accept; }
# pipe to master protocol pipe Pibgp4 { table master4; peer table Tibgp4; import all; export where defined(IBGP_EXPORT); }
protocol bgp gw1 from IBGP { description "gw1 - VLAN 3313"; ipv4 { table Tibgp4; import filter ibgp_in_primary4; export where defined(IBGP_EXPORT); ... }; ipv6 { table Tibgp6; import filter ibgp_in_primary6; export where defined(IBGP_EXPORT); ... }; interface "bond0.A"; source address fe80::A:2; neighbor fe80::A:1 internal; ... }
# ospf_metric2 set on an export filter the same as with the static # protocol case, but the metric seems to not be respected when # the route is redistributed from iBGP. protocol ospf v2 ospf4 { ipv4 { table Tospf4; import all; export filter { if !defined(OSPF_EXPORT) then reject; ospf_metric2 = 10; # 10 for nat1, 100 for nat2 accept; }; }; ... }
Thanks in advance for any help...
Pete