> > Could you try this diff? > > I have a kernel running with your diff over the last hours and created > quite some network traffic and the error didn't appear so far. > Previously, I was able to create it quite fast. So definitely an > improvement.
That's strange. The diff would help only if routing domains are involved. But you said that you don't have them. tobhe@ and I could reproduce the problem on our setup. It looks like using enc0 as iked interface may trigger it. But the panic is not clearly reproducable, we don't see the pattern that actually causes the stack overflow. Sometimes it happens when a dynamic PMTU route expires, or when IPsec traffic begins, or when iked is restarted, or it does not panic at all. And the PMTU is not always calculated correctly. So please continue testing and report configurations that work and don't work. Our test setup uses: - laptop with iked as road warrior and IPv4 on enc0 - VPN gateway as IPsec endpoint - host route from gateway to server with -mtu 1400 - server running tcpbench -s When running tcpbench from laptop it crashes. Sometimes ... bluhm
