Trunk network profile (untagged / trunk mode) stopped working for VMs

[joaquimrbjr]

Hi,

After upgrading my environment to oVirt engine 4.5.7-1.el10 and upgrading the cluster compatibility level to 4.8, I replaced my old hosts with new ones.

Since then, my trunk network profile (untagged / trunk mode) stopped working for VMs.

Environment

oVirt Engine: 4.5.7-1.el10

Cluster compatibility level: 4.8

New hosts added, old hosts removed

Physical NIC configured as bond0 (trunk) on the host

VLANs are attached to the trunk interface in the host network configuration

Switch port connected to the host is configured as trunk and working

Scenario

I have:

One pfSense VM configured with a trunk interface and multiple VLAN subinterfaces (each running DHCP)

One Rocky Linux 9 VM connected to the same trunk profile

VLAN interfaces configured inside the OS (e.g. eth0.<vlan_id>)

DHCP client running on the VLAN interface

This setup worked correctly before the upgrade.

Current Problem

After the upgrade and host replacement:

VMs connected to the trunk profile do not receive DHCP on VLAN interfaces

VLANs appear correctly configured in oVirt

Host bond interface shows trunk configuration

Switch configuration has not changed

It looks like VLAN-tagged traffic is not properly reaching the VM.

Questions

Has there been any change in trunk network profile behavior in 4.8?

Is there any additional configuration required for trunk profiles after upgrading cluster compatibility?

Could this be related to Open vSwitch / Linux bridge changes in newer versions?

Any guidance would be appreciated.

[dupondje]

Nothing changed in this area...
From which OS to which OS did you update your hosts?

And for the VM that has the issues, please post the libvirt dumpxml output.
Next to that also post the network config (ip addr output, bridge config, etc)

[joaquimrbjr]

Hi,

Thank you for the response.

I have identified the root cause of the issue and it does not appear to be related to oVirt or the cluster compatibility level.

The problem is related to the Supermicro AOC-S25G-b2S (Broadcom-based) NIC firmware and its VLAN offloading behavior.

In another environment using VMware with the same hardware, the issue was resolved by disabling VLAN offloading on the physical NIC, which confirms this is a hardware/firmware offload problem rather than a virtualization configuration issue.

I attempted the same workaround on the oVirt hosts by disabling VLAN offloading on the physical interfaces and the bond:

ethtool -K ens3f1np1 rx-vlan-offload off tx-vlan-offload off
ethtool -K ens4f1np1 rx-vlan-offload off tx-vlan-offload off
ethtool -K bond0 rx-vlan-offload off tx-vlan-offload off

However, in oVirt this did not resolve the issue, and VLAN-tagged return traffic is still not reaching the VMs.

At this point I have opened a support case with the hardware vendor (Supermicro) to obtain an updated NIC firmware (Broadcom fix for VLAN handling).

Since the same behavior occurs outside of oVirt and is hardware-dependent, we can consider this issue not related to oVirt networking.

Thank you for your assistance.