[libvirt-users] network checksum offloading broken using virtio

Dennis Jacobfeuerborn dennisml at conversis.de
Fri Mar 4 18:17:57 UTC 2016


Hi,
with recent guest installs (both centos 5 and 7) on centos 7 hosts I
seem to have to disable checksum offloading using "ethtool -K eth0 tx
off" in order to allow traffic to flow a specific route.

Basically the guest is installed with IP 192.168.21.10 and a default
gateway of 192.168.21.254. Up until that point I can ssh into the system
normally.
There exists an OpenVPN System with the IP 192.168.21.1 that choses
client IP's for the vpn connections from the pool 192.168.20.0/24.
In order to pass the reponses back to the OpenVPN system I installed the
route "192.168.20.0/24 via 192.168.21.1 dev eth0".

When I now ping the IP 192.168.21.10 through the VPN connection this
works fine but when I try to ssh into that system the connection just
hangs. Looking at a tcpdump I noticed that the checksum for the packets
weren't quite right so I issued a "ethtool -K eth0 tx off" and suddenly
everything worked as expected.

What is strange here is that I'm seeing this with both CentOS 5 and
CentOS 7 guests and only when dealing with th routed traffic and not the
regular one.

Does anyone have an idea what is going on here? Could this be an issue
with the virtio driver?

Regards,
  Dennis




More information about the libvirt-users mailing list