CRS326-24G-2S+ & E50UG VLAN Issues

RouterOS
1

Hi guys,

setup:
1x CRS326-24G-2S
1x E50UG
1x proxmox server
I configured on alldevices my bridge and correct vlans - not the first time i do that.

CRS326-24G-2S connected to proxmox vlans working fine - tagged and untagged.

Now the setup i need was:

CRS326-24G-2S - (eth1) E50UG (eth5) - proxmox server
I configured everthing like usaly but i coudnt get tagged traffic into my vms- untagged worked.
Now after 3h of pain I disabled hw-offloading on the trunk port eth5 on the E50UG and its working.
btw the trunk port eth1 is still with hw offloading.

can someone explain me that behavior??

2

This is a usual problem. It occurs when it’s not a single switch chip handling all ports of a given vlan (both tagged and untagged.)

The new hex series is prone to this, because the integrated switch chip only handles eth2-5.

The solution is to assign the bridge itself as a member of every such vlan, usually as a tagged port. (The same can be accomplished by attaching a vlan interface for these vlans to the bridge.)

This should really be handled by the software automatically, and there is a changelog item in the newer betas implying that this is now fixed. I would apply the manual fix anyway, because that’s exactly what the fix in routeros does.

3

thank you. got it.
btw is there a way to pm you…?

4

I’m not really into how Discourse works, but I think PMs are disabled globally.

You can always @ me, or if there’s some totally private matter, some other way may be established.

5

Thanks for the hint!
Indeed, with E50UG when disabling HW-offloading on all ports on the bridge switching between tagged VLANs on one port and untagged on access ports works.
Just limiting ports to ether2-5 does not help. I tested with ROS 7.19.4 and ROS 7.20beta9.
As a "fix" MT could at least disable HW-offloading...

6

Glad to be of help :slight_smile: But really, the proper solution is not to disable hw offloading, but adding the bridge itself as a tagged member port manually for all vlans spanning outside of a switch.

There seems to be a general problem around this is the software, because it was "officially fixed" many times. Maybe it's done by updating some internal table manually for each device? No idea: but these sorts of reports are always appreciated, so if you can reproduce it, file a ticket.