Potential bug: Inconsistent next-hop resolution in multi-homed same-subnet setup
David Petera
david.petera at nic.cz
Tue Jan 13 12:23:07 CET 2026
Hi Liu,
are you talking about the channel option `gateway recursive`
(https://bird.nic.cz/doc/bird-2.18.html#bgp-gateway) and that even with
it on it still does not work?
If I understand your issue correctly, this should resolve the issue you
are having (keep in mind that the option in not compatible with sorted
tables).
In BIRD v3 you should be able to see item host-entry in the route info
(via `show route table all all` or similar), unfortunately this
information is not present in v2.
You are right that BIRD should do this better right away. However to fix
this, a much bigger refactor of the code is required.
It is on our backlog, but we can not promise you that the changes will
made in time for the release of v2.19.
Hope this helps a bit,
David
David Petera (he/him) | BIRD Tech Support | CZ.NIC, z.s.p.o.
On 1/11/26 22:40, Pysio wrote:
> Hi Alexander,
>
> Thanks for the tips!
>
> I just tried the multihop (recursive) approach, but unfortunately, it
> didn't change anything. Even with multihop on, BIRD still picks ens23
> for the next hop resolution. It seems like it's still looking at the
> global interface list instead of sticking to the route in the specific
> table I've assigned.
>
> I'll give VRFs a shot later tonight to see if that fixes
>
> While VRFs should work as a workaround, I still feel like BIRD should
> ideally handle this better out of the box. In a multi-table setup,
> you'd expect it to prioritize the lookup within the attached table first.
>
> Anyway, I’ll let you know how the VRF test goes!
>
> Best,
>
> Liu HaoRan
>
> On Mon, Jan 12, 2026 at 3:03 AM Alexander Zubkov <green at qrator.net> wrote:
>
> I probably saw something like that bug for IPv6, so not sure if it
> is the same. And I did not have separate tables for the
> interfaces. But worth a check anyway. It seemed to me that direct
> sessions created routes with the interface and ip through that the
> bgi session was installed. And recursive (multihop) sessions set
> the routes to the first found interface (I remind, I had all them
> in one table).
> So you can try to change protocol to recursive to check (but it
> could have other drawbacks) if it would use the appropriate route
> from the attached table.
> Also, if you wander of the possible solutions, I think you could
> play with vrfs to divide the interfaces routing, bird is aware of
> vrfs.
>
> On Sun, Jan 11, 2026 at 7:09 PM Pysio <mcyouyousever at gmail.com> wrote:
>
> Additionally, I should mention that the issue persists even
> when I explicitly declare the IGP table using the following
> configuration file.
>
> protocol bgp cloudflare_tpix_210440_v4 {
> local as 210440;
> source address 203.163.223.48;
> neighbor 203.163.222.39 as 13335;
> ipv4 {
> table table_tpix_210440_v4;
> import keep filtered on;
> import where import_filter_v4();
> export where export_filter_tpix_210440();
> graceful restart on;
> export limit 99 action block;
> igp table table_tpix_210440_v4;
> };
> }
>
> BIRD 2.17.1 ready.
> Table table_tpix_210440_v4:
> 1.1.1.0/24 <http://1.1.1.0/24> unicast
> [cloudflare_tpix_210440_v4 2026-01-12 02:07:03] * (100) [AS13335i]
> via 203.163.222.39 on ens23
> Type: BGP univ
> BGP.origin: IGP
> BGP.as_path: 13335
> BGP.next_hop: 203.163.222.39
> BGP.local_pref: 100
> BGP.aggregator: 162.158.240.1 AS13335
> BGP.community: (13335,10080)
> BGP.large_community: (213605, 10, 1) (213605, 10, 4)
>
>
> BIRD 2.17.1 ready.
> Table table_tpix_210440_v4:
> 203.163.222.0/23 <http://203.163.222.0/23> unicast
> [direct_tpix_210440 2026-01-12 02:04:33] * (240)
> dev ens4
> Type: device univ
>
>
> On Mon, Jan 12, 2026 at 1:57 AM Pysio
> <mcyouyousever at gmail.com> wrote:
>
> Hi Alexander,
>
> Thank you for your response. To answer your question: yes,
> I am using specific tables (e.g., |table_tpix_210440_v4|)
> for each BGP session, and I have corresponding |protocol
> direct| instances to populate those tables with the
> respective interface routes.
>
> However, despite having a valid direct route in the
> specific table, BIRD still resolves the BGP next hop via
> |ens23| (the first interface in the kernel's subnet list)
> instead of honoring the device route defined in the local
> table.
>
> Below is the relevant part of my configuration for your
> review.
>
> ```config|
> table table_tpix_210440_v4;
> table table_tpix_210440_v6;
>
> protocol direct direct_tpix_210440 {
> ipv4 { table table_tpix_210440_v4; };
> ipv6 { table table_tpix_210440_v6; };
> interface "ens4";
> }
>
>
> protocol bgp cloudflare_tpix_210440_v4 {
> local as 210440;
> source address 203.163.223.48;
> neighbor 203.163.222.39 as 13335;
> ipv4 {
> table table_tpix_210440_v4;
> import keep filtered on;
> import where import_filter_v4();
> export where export_filter_tpix_210440();
> graceful restart on;
> export limit 99 action block;
> };
> }
> |
>
> ```
>
> Even though |table_tpix_210440_v4| contains a direct route
> for |203.163.222.0/23 <http://203.163.222.0/23>| via
> |ens4|, BIRD's |show route| output for the BGP session on
> |ens4| indicates the next hop is resolved via |ens23|.
>
>
> On Mon, Jan 12, 2026 at 1:06 AM Alexander Zubkov
> <green at qrator.net> wrote:
>
> Hi Lui,
>
> Does cloudflare_tpix_210440_v4 bgp protocol
> use table_tpix_210440_v4 as IGP table also? It might
> be better to have an overview of your config file.
>
> Regards,
> Alexander
>
> On Sun, Jan 11, 2026 at 3:14 PM HaoRanLiu
> <mcyouyousever at gmail.com> wrote:
>
> Hello,
>
> I am writing to report a potential issue (or seek
> clarification on
> resolution logic) in BIRD 2.17.1 regarding how
> next hops are resolved
> when multiple interfaces exist in the same subnet.
>
> [ Environment ]
> - BIRD version: 2.17.1
> - OS: Linux
> - Setup: Multiple physical interfaces (ens4,
> ens23, ens2) are configured
> with IP addresses in the same prefix:
> 203.163.222.0/23 <http://203.163.222.0/23>.
>
> [ Observation ]
> Within a specific routing table
> (table_tpix_210440_v4), BIRD's recursive
> resolution for a BGP route does not seem to honor
> the "Direct/Device"
> route present in that same table.
>
> [ Evidence ]
>
> 1. The Direct Route in table
> "table_tpix_210440_v4" points to ens4:
>
> bird> show route for 203.163.222.39/32
> <http://203.163.222.39/32> table
> table_tpix_210440_v4 all
> Table table_tpix_210440_v4:
> 203.163.222.0/23 <http://203.163.222.0/23>
> unicast [direct_tpix_210440 2026-01-11 21:10:29] *
> (240)
> dev ens4
> Type: device univ
>
> 2. However, a BGP route in the SAME table resolves
> via ens23:
>
> bird> show route for 1.1.1.1 table
> table_tpix_210440_v4 all
> Table table_tpix_210440_v4:
> 1.1.1.0/24 <http://1.1.1.0/24> unicast
> [cloudflare_tpix_210440_v4 2026-01-11
> 21:33:43] * (100) [AS13335i]
> via 203.163.222.39 on ens23
> Type: BGP univ
> BGP.origin: IGP
> BGP.as_path: 13335
> BGP.next_hop: 203.163.222.39
> ...
>
> [ System Network State ]
> The OS kernel shows three interfaces in this subnet:
>
> # ip route show 203.163.222.0/23
> <http://203.163.222.0/23>
> 203.163.222.0/23 <http://203.163.222.0/23> dev
> ens23 proto kernel scope link src 203.163.223.49
> 203.163.222.0/23 <http://203.163.222.0/23> dev
> ens2 proto kernel scope link src 203.163.223.50
> 203.163.222.0/23 <http://203.163.222.0/23> dev
> ens4 proto kernel scope link src 203.163.223.48
>
> [ Questions ]
> 1. Is this a bug in the recursive resolution
> logic? It appears BIRD
> might be picking an interface from the global
> interface list (perhaps
> the first one UP) rather than following the best
> route available in the
> specific table being queried.
> 2. Why does the BGP route's "via" field not honor
> the interface (ens4)
> defined by the device route in the same table?
> 3. Is there any configuration to force BIRD to
> resolve next hops
> strictly based on the current table's best path in
> such multi-homed
> scenarios?
>
> Any insights would be appreciated.
>
> Best regards,
> Liu HaoRan
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://trubka.network.cz/pipermail/bird-users/attachments/20260113/f0607b5e/attachment.htm>
More information about the Bird-users
mailing list