From: [email protected] <[email protected]> on behalf of Inder via 
lists.fd.io <[email protected]>
Date: Thursday, 23 April 2026 at 9:06 pm
To: [email protected] <[email protected]>, [email protected] 
<[email protected]>
Subject: [vpp-dev] IPv6 mFIB specific entries missing interfaces — ND broken 
due to Multicast RPF check failure
Hi,

We're seeing an issue where IPv6 Neighbor Discovery is completely broken on 
certain VPP sub-interfaces because the IPv6 mFIB specific entries (ff02::1/128, 
ff02::2/128, ff02::16/128, ff02::1:ff00:0/104) created by the ND subsystem 
under the "Special" source do not include all IPv6-enabled interfaces in the 
VRF.

VPP version: v25.10-release built on SUSE

SETUP
-----
Two servers connected back-to-back via 10G Intel X710 NICs with SR-IOV. VPP 
binds to a VF (iavf driver) on each server for the direct link. Each server 
also has a PF (i40e driver) connected to an upstream router. Both the VF 
sub-interface and PF sub-interface are in the same IPv6 VRF (table 2). OSPFv3 
runs on the VF sub-interface between the servers (PtP link). Multiple servers 
deployed with identical VPP version and configuration.

Interfaces per server in VRF table 2:
- GigE-PF.200 — PF sub-interface (i40e), upstream router, IPv6 2001:db8:1::1/64
- TenG-VF.4091 — VF sub-interface (iavf), server-to-server direct, IPv6 
fd00:ab:cd::1/125
- TenG-VF.4092 — VF sub-interface (iavf), additional link, IPv6 
fd00:ab:cd::19/125
- host-internal.4091 — internal af_packet, IPv6 fd00:ab:cd::9/125

PROBLEM
-------
IPv6 ping between servers on TenG-VF.4091 fails with 100% loss. IPv4 on the 
same interface works fine. OSPFv3 neighbor is Full (uses ff02::5 multicast 
which is unaffected). The issue is that incoming Neighbor Solicitation 
multicast packets on TenG-VF.4091 are dropped by VPP with "Multicast RPF check 
failed" because TenG-VF.4091 is missing from the specific mFIB entries.

The general ff00::/8 entry correctly includes all 4 interfaces. But the 
specific entries only include GigE-PF.200. Since specific entries win via 
longest-prefix-match, the RPF check fails for any interface not listed in them.

This is not consistent across servers — with identical VPP version and config, 
some servers have all 4 interfaces in the specific entries (working), others 
have only 1 or 2 (broken). The issue is deterministic per server and survives 
VPP restart.

WHAT WE TRIED (none fixed the mFIB)
------------------------------------
1. ip mroute add/del via CLI — silently ignored, cannot modify "Special" source 
entries
2. ip mroute add with Accept-all-itf flag — added as CLI source but Special 
source takes precedence
3. Removing and re-adding IPv6 addresses on affected interface — re-joins 
ff00::/8 but not specific entries
4. Removing IPv6 from ALL interfaces and re-adding — specific entries persist 
with empty lists, only one interface gets re-added
5. Interface admin down/up — no effect on mFIB
6. VPP restart — reproduces same broken state
7. Static IPv6 neighbors (bypassing ND) — unicast IPv6 works perfectly, 
confirming data plane is fine

QUESTIONS
---------
1. Is there a known issue with the ND subsystem not adding interfaces to 
existing specific mFIB entries when they join the same multicast group?

Nope. Given it works on some of your servers but not others, if there is a bug 
it depends on the differences between your servers. Perhaps in the order the 
config was entered? Do the interfaces’ addresses appear in a different 
[incorrect] table (i.e. they weren’t correctly moved when the interface was 
added to a VRF).

2. Is there a way to force the Special source to re-evaluate interface 
membership at runtime?

nope

3. Is there a startup config option that controls this behavior?

Nope

/neale

Detailed logs and mFIB dumps from both working and broken servers are in the 
attached file.

Thanks for any guidance.
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#26998): https://lists.fd.io/g/vpp-dev/message/26998
Mute This Topic: https://lists.fd.io/mt/118969596/21656
Group Owner: [email protected]
Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to