A little follow-up:

There is no packets marked as 'dropped by kernel' during entire
tcpdump session and now I`m testing on direct node-node link - packets
can be seen on ethX interface in same bridge as transmitting interface
but there is an empty input on ``receiving'' side. Flushing learning
table or even restart of vswitchd process has no effect at all.

Clearing tag immediately returns working state back as I can see for
now. Once I had managed to get tagged ARP requests asking for router
hwid on the physical neighbor interface when I tried to ping VM on
other host with address belonging to same CIDR with defaultrouter in
it, for now not able to reproduce it in any imaginable way.

What works:

ovs-vsctl set port eth1 tag=X on both sides

What does not:

Clear tag and set trunk to same vid(s).

ovs-vsctl (Open vSwitch) 1.10.0
Compiled May 17 2013 13:45:06
and 7395c05254df87ce52b37b04478e802befd799d9 for kernel module


On Fri, Sep 6, 2013 at 10:22 PM, Jesse Gross <[email protected]> wrote:
> I see one flow that OVS is trying to tag, which is an ARP request
> going from 10.0.3.29 to 10.0.3.2 with VLAN 4. In eth0.dump I also see
> these packets with VLAN 4 on them, so it appears that OVS is operating
> properly. I don't see these packets at all in mirror.dump.
>
> I would run tcpdump on the receive side of the connection and also
> check drop counters. Most vlan related problems that I have seen
> involve missing tags not entirely dropped packets so this is unusual
> unless there is a misconfiguration.
>
> Linux 3.10 introduced support for nested vlan tag offloading so this
> seems a likely culprit.
>
> On Fri, Sep 6, 2013 at 12:40 AM, Andrey Korolyov <[email protected]> wrote:
>> Here please.
>>
>> OpenVSwitch has symmetric topology I showed earlier.
>> - mirror.dump contains traffic from mirrored port on switch maps to
>> eth0 in selected machine
>> - eth0.dump contains traffic from machine holding ``active'' VM doing
>> ICMP requests
>> - dump.flows speaks for itself, from same node
>>
>> I also may ensure you that the broadcast ARP requests from VM with
>> tagged interface will not be shown anywhere more distant than this
>> host. Since stock 3.7 works perfectly in this configuration I assume
>> that we face major bug in kernel datapath implementation.
>>
>>
>>
>> On Fri, Sep 6, 2013 at 4:26 AM, Jesse Gross <[email protected]> wrote:
>>> Can you please provide the information that I originally requested (below)?
>>>
>>> On Thu, Sep 5, 2013 at 1:15 PM, Andrey Korolyov <[email protected]> wrote:
>>>> Hello,
>>>>
>>>> Observing the same issue in quite simple topo:
>>>>
>>>>     Bridge "oswbr0"
>>>>         Port "oswbr0"
>>>>             Interface "oswbr0"
>>>>                 type: internal
>>>>         Port "vnet0"
>>>>             tag: 100
>>>>             Interface "vnet0"
>>>>         Port "vnet1"
>>>>             tag: 5
>>>>             Interface "vnet1"
>>>>         Port "bond0"
>>>>             Interface "bond0"
>>>>     ovs_version: "1.10.0"
>>>>
>>>> bond0 is a regular linux bond configured as a OVS port. There is
>>>> simply no tagged traffic on this port but untagged works just fine.
>>>>
>>>>
>>>> Tested on both 3.10 ``stock'' module and on
>>>> 7395c05254df87ce52b37b04478e802befd799d9. Any suggestions are very
>>>> appreciated since I`m bounded with 3.10 and can not afford downgrade.
>>>>
>>>>
>>>> On Tue, Aug 20, 2013 at 5:10 AM, Jesse Gross <[email protected]> wrote:
>>>>> On Mon, Aug 19, 2013 at 2:07 PM, Ruß, André <[email protected]> wrote:
>>>>>> Hi,
>>>>>>
>>>>>>
>>>>>>
>>>>>> I run two bridges on my openvswitch (ovsbr0, ovsbr1) both have redundant
>>>>>> ports (bond0, bond1 with lacp) since the port groups have different
>>>>>> performance (1gbs and 10gbs). Multiple internal interfaces are connected 
>>>>>> to
>>>>>> the bridges and get network traffic assigned based on the vlan tag.
>>>>>>
>>>>>> Since version 3.10 of the kernel only works with the first bridge 
>>>>>> (ovsbr0)
>>>>>> as expected the second one gets not tagged traffic (untagged works 
>>>>>> fine!).
>>>>>>
>>>>>> It makes no difference if I use the default kernel module or the upstream
>>>>>> datapath module.
>>>>>
>>>>> Can you run ovs-dpctl dump-flows and tcpdump when you have some vlan
>>>>> traffic running through each of the bridges and send that and the
>>>>> output of ovs-dpctl show?
>>>>> _______________________________________________
>>>>> discuss mailing list
>>>>> [email protected]
>>>>> http://openvswitch.org/mailman/listinfo/discuss
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to