[ovs-discuss] can not handle fragmented UDP packet

Hui Kang hkang.sunysb at gmail.com
Mon Apr 21 19:38:23 UTC 2014


Hi.
I would like to test the function of IP fragmentation and vlan together of
openvswitch. I found that the OVS switch on the receiver side can not
receive
the fragmented IP correctly. Below is my testbed setup.

On the sender side (a VM), I run iperf with fragmentation and without
fragmentation by

        iperf -c receiver's IP -u -l 2000     ---- has fragmentation
        iperf -c receiver's IP -u             ---- no fragmentation

On the receiver VM's host machine (where the OVS switch runs), I use
tcpdump to
capture the data. I found that I can see all the IP packets with the no
fragmentation Iperf correctly, like

        15:26:34.507647 52:54:00:5d:98:5a > 52:54:00:37:a7:d6, ethertype
802.1Q (0x8100), length 1516: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl
64, id 48308, offset 0, flags [DF], proto UDP (17), length 1498)
                       192.168.10.44.51218 > 192.168.10.61.5001: [udp sum
ok] UDP, length 1470
        15:26:34.758064 52:54:00:5d:98:5a > 52:54:00:37:a7:d6, ethertype
802.1Q (0x8100), length 1516: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl
64, id 48309, offset 0, flags [DF], proto UDP (17), length 1498)
                       192.168.10.44.51218 > 192.168.10.61.5001: [udp sum
ok] UDP, length 1470

However, when I use the iperf command with fragmentation, the VM can not
receive
the UDP packet and the tcpdump on the host shows

        15:32:18.598578 52:54:00:5d:98:5a > 52:54:00:37:a7:d6, ethertype
802.1Q (0x8100), length 70: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl
64, id 41493, offset 1976, flags [none], proto UDP (17), length 52)
                        192.168.10.44 > 192.168.10.61: udp
        15:32:18.629009 52:54:00:5d:98:5a > 52:54:00:37:a7:d6, ethertype
802.1Q (0x8100), length 70: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl
64, id 41495, offset 1976, flags [none], proto UDP (17), length 52)
                        192.168.10.44 > 192.168.10.61: udp
As you can see, all captureed packets have the offset as 1976. So the first
fragment of the IP packets are very likely being dropped. I also use
ovs-ofctl get-frags br to see the fragmentation handling mode is "normal".

Is there anything wrong regarding my setup or any bug in OVS? Thanks.

regards
- Hui Kang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20140421/c2a4dd5a/attachment-0002.html>


More information about the discuss mailing list