> -----Messaggio originale----- > Da: Paul Wouters <p...@nohats.ca> > Inviato: domenica 14 ottobre 2018 01:49 > A: libreswa...@iotti.biz > Cc: swan@lists.libreswan.org > Oggetto: Re: R: [Swan] R: Packets dropped strangely > > On Tue, 9 Oct 2018, libreswa...@iotti.biz wrote: > > > ]# cat /proc/net/xfrm_stat > > XfrmInError 0 > > XfrmInBufferError 0 > > XfrmInHdrError 0 > > XfrmInNoStates 2 > > XfrmInStateProtoError 0 > > XfrmInStateModeError 0 > > XfrmInStateSeqError 0 > > XfrmInStateExpired 0 > > XfrmInStateMismatch 0 > > XfrmInStateInvalid 69 > > XfrmInTmplMismatch 119 > > XfrmInNoPols 13 > > XfrmInPolBlock 0 > > XfrmInPolError 0 > > XfrmOutError 0 > > XfrmOutBundleGenError 0 > > XfrmOutBundleCheckError 0 > > XfrmOutNoStates 275 > > anything non-null points to a problem. But these numbers are not reset after > you restart libreswan, only when you restart the kernel. A few of those can > happen at times during race conditions, but if you do a ping to something > that fails these numbers should not increase per ping. > Of that happens, there is a real problem since libreswan thinks there is a > policy or state and gave it to the kernel but the kernel disagrees. > > Paul
After weeks of testing, I think there is such a problem, at least on CentOS7 with the provided kernel and libreswan :) The only thing I am sure of, is that when I do my test in the failing condition, it is the XfrmInTmplMismatch counter that gets increased. Just for clarity, I do my test with "nc -v -s 10.250.14.254 172.16.74.193 5681" since ping is not accepted on the remote site. My box is actually a router/firewall, and the problem was reported by my users. Then for clarity I made the tests with iptables rules flushed, no clients connected, and using nc from the firewall. But I know that if I try to simulate the problem with a client in my local lan, dropwatch indicates a drop in another kernel function than tcp_v4_rcv. I can make the tests and report it back, including the xfrm_stat counter that gets increased when the packet should be routed and not accepted. Luigi _______________________________________________ Swan mailing list Swan@lists.libreswan.org https://lists.libreswan.org/mailman/listinfo/swan