[libvirt-users] macvtap vlan and tcp header overhead (and mtu size)

I have a host setup for libvirt kvm/qemu vms. And I wonder a bit about the overhead of the macvtap and how to configure the mtu's properly. To be able to communicate with the host, I have moved the ip address of the host from the adapter to a macvtap to allow host communication. I have the below setup on hosts. +---------+ | macvtap0| +--| ip | | | mtu1500 | | +---------+ +---------+ | +---------+ |eth0.v100| | | macvtap1| +--| no ip +-+--| ip | +---------+ | | mtu1500 | | mtu1500 | | eth0 | | +---------+ +---------+ | +--+ | mtu9000 | | +---------+ +---------+ | |eth0.v101| +--| ip | | mtu9000 | +---------+ https://pastebin.com/9jJrMCTD I can do a ping -M do -s 9000 between hosts via the vlan interface eth0.v101. That is as expected. The ping -M do -s 1500 macvtap0 and another host or macvtap1 fails. The maximum size that does not fragment is 1472. That is 28 bytes??? Where have they gone? I am only using macvtap, can this be the combination of the parent interface being a vlan and that macvtap is not properly handling this? Anyone experienced something similar? Or can explain where these 28 bytes go?