Since 2 weeks I'm faced with some problems at a Xen hosting environment based on Debian wheezy.
The issue is, that one specific domU is regularly failing network. The investigations showed, that no network packets from and to this system are seen on the the network interface in the domU and on the bridge interface in the dom0.
The funny part is, that the firewall on the dom0 now is logging packet rejects on the interface of the effected system and one interface of an other specific domU for the destination address. One pack and rejects on both interfaces ... really strange!
Shutting down the domU and recreating them doesn't help. Even if restarting other domUs, those are also effected by now. The only way to fix is to reboot the entire dom0.
After several investigations I stumpled upon the following in the dmesg:
vif-xm116: Frag is bigger than frame.
vif-xm116: fatal error; disabling device
xenbr0: port 4(vif-xm116) entering disabled state
A quick search at the famous search engine revealed, there seems something bigger. Even a Debian bug is open since February. Unfortunately this problem is still unfixed in wheezy and squeeze.
But ... there is hope in sight! Ian Campbell provided packages based on top of the latest kernel packages with a call for testing. Installing them seems to fix the issue in my environment.
So we just have to hope this will get fixed soon, as this regression is really annoying. Thanks to all who make this happen in advance!