On Thu, Jul 25, 2024 at 8:56 AM Ilya Maximets <i.maxim...@ovn.org> wrote: > > On 7/25/24 05:18, Jun Wang via discuss wrote: > > Hi,team. > > I used hping3 to send IPv4 packets and found that the checksum > > of the packet was incorrect when it carried the option field. After > > analyzing the code, I discovered that our IP checksum calculation > > does not consider the length of the options, leading to the > > checksum error. Are we not considering the IP options on purpose, > > or is there another consideration? > > > > Test command : > > hping3 -c 1000 -1 -G 10.252.246.164 > > > > packet: > > > > 0000 58 53 c0 56 07 46 cc d8 1f 1b 7e 5e 81 00 00 5c > > 0010 08 00 45 00 00 76 00 00 40 00 3d 11 5d 33 0a fc > > 0020 e5 46 0a fc e5 05 e1 5c 12 b5 00 62 84 69 08 00 > > 0030 00 00 00 00 64 00 02 7a 3b 9b c2 2a 58 53 c0 56 > > 0040 07 60 08 00 4f 00 00 44 f0 e4 00 00 3e 01 7e 14 > > 0050 0a fc f7 23 0a fc f6 a4 07 27 0c 01 02 03 04 0a > > 0060 fc e9 fc 00 00 00 00 00 00 00 00 00 00 00 00 00 > > 0070 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > 0080 08 00 3e 2f b9 d0 00 00 > > > > > > Calculation method: > > > > ip->ip_csum = csum(ip, sizeof *ip); > > > > The only place where we calculate IP checksum from scratch must > be the checksum offload code, i.e. dp_packet_ip_set_header_csum(). > Are you using checksum offload? > > If so, I'd say it is a bug in the dp_packet_ip_set_header_csum() > function. It should use the length from the ip header.
Yes, it looks like dp_packet_ip_set_header_csum doesn't use the correct length. It seems that the most problematic location where this function is used is in netdev_push_header(), which I believe is what's causing the problem here. I'll send in a patch for this. Thanks, Mike > > In general, OVS doesn't actually support IP options. It doesn't > parse them and doesn't act on them. But it should allow packets > with options to pass through. > > Best regards, Ilya Maximets. > _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
