[Xen-users] checksum offloading causing problems?

["James Harper" <james.harper@xxxxxxxxxxxxxxxx>]

Can anyone tell me under what circumstances checksum offloading might
cause problems? I have the following configuration:

1. Remote router (OpenWRT) with PC's on the LAN and a GRE tunnel to the
head office

2. Central office with a server running Xen 3.1.2. Dom0 is doing PPPoE
and is terminating the GRE tunnels

3. Main corporate application running on a Linux DomU, accessed via ssh.
The DomU is bridged to the central office LAN.

Sometimes, completely out of the blue, traffic from the remote LAN will
be unable to connect to the ssh session. Ping works fine, but ssh
sessions (even just a telnet to port 22) don't do anything.

When I do a tcpdump on the DomU, I see the SYN from the remote branch,
the SYN+ACK from the DomU, the ACK from remote branch, and data packets,
which remain unanswered.

When I do a tcpdump on the Dom0 on the LAN bridge, I see exactly the
same as the tcpdump on the DomU.

When I do a tcpdump on the Dom0 on the GRE interface, I see the TCP
handshake but no data packets.

The data packets coming out of the DomU have an incorrect checksum
because tx offloading is in place. I think that is why Dom0 isn't
forwarding them.

The funny thing is, this will work for weeks and then suddenly stop, but
only on one or two GRE tunnels, and different GRE tunnels all the time.

I turned off tx checksum offload on the DomU and a few seconds later
things started working again, so I think that fixed it but I'm not
completely sure.

Any suggestions? Are there any known fixes in newer versions of xen?

I'm running an almost identical setup (same Dom0 kernel) on similar
hardware (HP ML370 vs HP DL385), and the DL385 works fine...

James

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users