BUG VXLAN and MTU

A

andaga

Member
Proxmox Subscriber
May 24, 2022
137
17
23
Switzerland
Hello,

Why we need setup the MTU of the SDN zone into the settings and into the NIC interface of the VM when we select the "bridge" concerning the VXLAN the MTU by default is 1500?

Normally for VXLAN the MTU need to be 1450 as configured into the capture attached!

It's a bug? Why we need setup it again manually into the NIC interface?
 

Attachments

  • VXLAN MTU.png
    VXLAN MTU.png
    20.7 KB · Views: 15
  • VXLAN MTU2.png
    VXLAN MTU2.png
    38.2 KB · Views: 16
You don't need to setup it again, it is only a visual bug since 1500 was the default MTU for everything before SDN.
 
Ah sorry, I've only looked on the host where tap interfaces get created with the proper MTU, but it seems like in the VM the interface has an MTU of 1500 - I'll look closer into it. Thanks for reporting this!
 
Hi,

We are experiencing the exact same issue. Even though the VXLAN MTU is configured correctly (1450) in the SDN zone settings, the NIC interface inside the VM still defaults to 1500 when using "bridge" mode.

As mentioned, VXLAN typically requires an MTU of 1450, but we still have to manually adjust it in the VM's NIC settings. This behavior suggests it's not just a visual bug, as andaga pointed out previously.

Is there any progress or update on fixing this?

Best regards,
MP
 
can you send the result of "ip addr", just to be sure that the mtu is correctly applied on the bridge itself ?

also, does it work if you stop/start the vm ?

I don't think we change the tap mtu itself currently , but the mtu should be lower inside the vm guest. (it's working with virtio-net only).


The tap mtu could be a problem if it'll be higher than 1500, but not the reverse. (as this is the guest os which define the source mtu, and a lower mtu can travel a interface with a bigger mtu)
 
Last edited:
spirit said:
I don't think we change the tap mtu itself currently , but the mtu should be lower inside the vm guest. (it's working with virtio-net only).
Click to expand...
It's a bug, because host_mtu never gets passed to QEMU, unless MTU is explicitly set in the network device [1]. I've got a fix on my machine but haven't sent it to the mailing list yet, because I wanted to check some other MTU related stuff for the patch series.

As a fix, I think we should just always set host_mtu (if it's virtio) and default to the bridge MTU if it isn't set explicitly in the VM configuration.

(Just rechecked on my machine, bridge is properly created with MTU 1450)

[1] https://git.proxmox.com/?p=qemu-ser...df918aa802619486cc39ab8d775136e;hb=HEAD#l1614
 
You must log in or register to reply here.
Top Bottom