All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Pv6 on Proxmox VMs only works after inbound ping (Kimsufi / OVH)
Hi,
I've a problem with ipv6 on a Kimsufi VM (Proxmox)
Setup:
- Host IPv6: 2001:41d0:xxx:2002::1/64
- VMs have IPv6 addresses from the same /64 configured as /128
- VM default gateway: 2001:41d0:xxx:2002::1
- VMs are bridged directly to vmbr0
- ndppd disabled for testing
- proxy_ndp disabled for testing
- net.ipv6.conf.all.forwarding=1
Problem:
After a host reboot (or after several hours of inactivity), a VM loses outbound IPv6 connectivity.
Symptoms:
- VM can ping 2001:41d0:xxx:2002::1 (PVE Node)
- VM has a valid default route
- VM sends IPv6 packets (confirmed with tcpdump)
- No replies arrive
- Host IPv6 works fine
The interesting part:
As soon as I ping the VM’s IPv6 address from an external host, IPv6 immediately starts working again in both directions.
Example:
1. VM cannot ping Google IPv6.
2. External host sends a single ping to VM IPv6.
3. VM replies.
4. Immediately afterwards the VM can reach the IPv6 Internet again.
Tcpdump shows that the external ping triggers Neighbor Solicitation/Advertisement traffic for the VM address.
What confuses me is that the problem also reappears after several hours without rebooting anything, which makes me think this is some kind of NDP/neighbor cache timeout rather than a reboot issue.
Has anyone seen similar behavior on OVH/Kimsufi IPv6 setups with Proxmox bridges?
Thanks!
Comments
It's a QEMU issue with /128 addresses.
I ran into this exact issue on CHR whilst only having a /128 assigned to the VPS, just replicated this on my own residential connection where I get a /56 and split it into /64s(so I directly assigned a /128 as a test) AND I've also replicated this in eve-ng with ULAs.
You can probably ping your gateway(which you've mentioned so) and go no further under this condition which is what I see also.
The fix is not to use a /128 assigned to your VM.
Thank you! That solved the problem.
Using a /64 and setting the OVH gateway as on-link (since it’s outside the subnet) fixed it. No ndppd and no IPv6 forwarding required on the Proxmox host.
Great!