Le 20 janv. 09 à 01:24, James Harper a écrit : This is a common problem, and appears to be related to bugs in one or more of: . Xen frontend network driver (unlikely) . Xen backend network driver . Linux bridge code . Linux routing code . Linux physical NIC adapter
I suspect it is the last one, but it's complicated enough that the bug could lie elsewhere but only show up on certain NIC's.
You can be thankful that your problem was consistent. I have a setup where the packets from DomU are routed in Dom0 to one of 6 GRE tunnels, and occasionally for 30 minutes or so TCP packets stopped working. Took me ages to figure out that turning off tx checksum offload in the DomU would resolve the problem. It was made harder because by the time I could connect to the client to investigate, it had started working again so all I could do was to set up tcpdumps of some fairly high volume links and wade through those based on the times that the problem was occurring...
James
I had the same kind of issue under redhat 5.2 and succeed in reproducing it 100% of the time. It seams to be an interraction between tx offload and netfilter nat. I have opened a case :
I have got a reply from Herbet Xu indicating that effectively, currently there is an issue in the netfilter infrastructure.
I don't know if this issue is well identified in the xen or netfilter projects.
Regards, |