Messages in this thread | | | From | Keith Owens <> | Subject | Re: More broken 2.1.28 things...? | Date | Sat, 08 Mar 1997 13:48:35 +1100 |
| |
On Fri, 7 Mar 1997 14:49:54 +0000 (GMT), Matthew Kirkwood <weejock@ferret.lmh.ox.ac.uk> wrote: >This is an unpatched 2.1.28 kernel, so perhaps I've missed something >important, but.... > >pc29: TCPv4 bad checksum from 1660a2cc:0050 to 818a01a3:0e62, len=20/20/40 >pc29 last message repeated 3 times >pc29 last message repeated 2 times >pc29: TCPv4 bad checksum from 1660a2cc:0050 to 818a01a3:0e62, len=20/20/40 >pc29: TCPv4 bad checksum from 1660a2cc:0050 to 818a01a3:0e62, len=20/20/40 >pc29: TCPv4 bad checksum from 1660a2cc:0050 to 818a01a3:0e62, len=20/20/40
This problem has been raised many times on the IP masquerading list. We see it more because mainline TCP/IP used to silently ignore packets with bad TCP checksums but masq code complained. 2.1.15 added a complaint to mainline TCP/IP as well.
So far every occurrence of this message (that we can explain) has been traced to a faulty serial link. In some cases the UART could not cope with the incoming speed, tuning up the UART to match the modem fixed these. In others the finger is pointing at a problem with VJ compression, possibly a mismatch between Linux and some terminal server's implementation. For these, disabling VJ compression on pppd removes the problem.
| |