Switching two ports in a bridge, I can't get it to work correctly

[fuomag9]

Hi everyone, I have a very strange issue that I can't figure out how to solve

vmbr0 is the lan bridge, I have another port (2.5Gbit/s) on the router that I wanted to switch along with the main one (10Gbit/s), as you can see from the picture, so that I could connect the ubiquiti AP to that (as I am doing now with the hardware switch) to use the 2.5Gbit (my switch has only 1/10 ports unfortunately)



If I configure it like in the picture and. plug the access point in it, it works halfway, I can actually surf the web but DHCP breaks and some devices won't receive the DHCPOFFER that is supposedly sent by the OPNsense router that is on the machine (I can see it being sent via tcpdump on proxmox on the network interface) and go into a loop of requesting the IP address.



tcpdump on proxmox, the DHCPOFFER is never received by the device plugged in

The issue is very similar to this thread https://forum.proxmox.com/threads/h...and-wire-it-to-vmbr0-to-expand-my-lan.112254/ as I've been trying to do the same exact thing, but openvswitch doesn't seem to be the solution for me (i.e. has the same issues)

I've tried with both openvswitch and linux bridges, and also tried on mikrotik chr and the same behaviour appears. It also happens if I connect a computer instead of the access point, so it's not the ubiquiti hardware


I believe this might even be a software bug in proxmox or a hardware fault, as there should not be any particular reason for this behaviour

 

[milew]

vmbr0(enp2s0f1np1 enp89s0) connect to opensense to two phisical interfaces?

 

[aaron]

Ideally you can set up some bonding, also known as nic teaming, link aggregation.

There are quite some options available. Which one works best depends on the other side that you connect to. The simplest "active-backup" one will always work, as it will use the primary NIC if available, and then switch to the secondary if the primary fails.

LACP is ideal, as it can give you overall better performance (not for a single connection) if both links are available. But the other side needs to be configured accordingly for it to work.

 

[fuomag9]

Ideally you can set up some bonding, also known as nic teaming, link aggregation.

There are quite some options available. Which one works best depends on the other side that you connect to. The simplest "active-backup" one will always work, as it will use the primary NIC if available, and then switch to the secondary if the primary fails.

LACP is ideal, as it can give you overall better performance (not for a single connection) if both links are available. But the other side needs to be configured accordingly for it to work.

I've tried bonding, but it seems not to work (and not the right solution) as I'd need both ports active and working at the same time.

vmbr0(enp2s0f1np1 enp89s0) connect to opensense to two phisical interfaces?

yes, they are two physical interfaces

 

[aaron]

I've tried bonding, but it seems not to work (and not the right solution) as I'd need both ports active and working at the same time.

Do you plan to use both ports for different things or for redunancy/performance? If it is the former, why not create a new vmbrX interface that uses the second NIC as bridge port?

 

[fuomag9]

Do you plan to use both ports for different things or for redunancy/performance? If it is the former, why not create a new vmbrX interface that uses the second NIC as bridge port?

What I want to do is having the two ports behave like they are ports of a switch and use the linux bridge/openvswitch as the "downlink" of opnsense as those ports are connected to the lan devices. (only using enp2s0f1np1 for this scope works perfectly) I cannot create a second bridge because that would count as a different interface inside opnsense and I'd just move the bridging/switching issue inside it (which is probably more inefficient as it's virtualized)

 

[fuomag9]

I believe this might be a hardware (or driver/firmware) issue with the

I226-LM port

as I've tried using the other one (the ms01 has two 2.5Gbit/s ports) that uses a I226-V controller instead with vmbr0 on openvswitch, and I've not seen issues for now.

What I've seen instead is the DHCP loop on the I226-LM port attached to a bridge (as the device that was using that port was configured to have a bridge by itself)

 

[fuomag9]

https://forum.netgate.com/topic/175592/can-not-get-dhcp-leases-on-new-intel-i225-lm-based-machine/24 seems to document that this is really an issue with I226-LM on the minisforum ms01

 

[milew]

It is your configuration. You have loop on layer2
Schema network

Solution is spanning tree or other to detect loop.

 

[fuomag9]

View attachment 72198
It is your configuration. You have loop on layer2
Schema network

Solution is spanning tree or other to detect loop.

It's not a loop, it's a hardware issue as per the post above