[ovs-dev] Datapath kern log msg key attribute error

[Jesse Gross]

The cleanest way seems to be to just reduce the default log level for
these messages so they don't pop up normally. I think we talked about
this before and were concerned about debugging when problems show up
at runtime other than developer testing.

I suppose the other possibility is to pass some kind of flag attribute
with messages indicating that this is a test probe and would silence
logging. Existing kernels would ignore this so they would still log
but the behavior would be otherwise unchanged.

On Thu, Apr 24, 2014 at 2:33 PM, Andy Zhou <azhou at nicira.com> wrote:
> I am sure it can be done and most likely involving some netlink
> changes.  Since netlink message design is your forte, any suggestions?
>
> On Thu, Apr 24, 2014 at 2:22 PM, Jesse Gross <jesse at nicira.com> wrote:
>> That's a good point, I didn't think about that. I wonder if there is a
>> way to avoid the error message in these cases since it's likely to
>> cause confusion given that it is logged by default.
>>
>> On Thu, Apr 24, 2014 at 1:57 PM, Andy Zhou <azhou at nicira.com> wrote:
>>> Whenever we instantiate a new ofproto (i.e. bridge), we probe datapath
>>> to detect if it supports MPLS, and if yes, how many labels it can
>>> support.  The probing produces similar error messages on my system.
>>> It should only appear when you add the bridge, but not afterwards.
>>>
>>> On Thu, Apr 24, 2014 at 12:27 PM, Thomas F Herbert
>>> <thomasfherbert at gmail.com> wrote:
>>>> On 4/24/2014 1:47 PM, Jesse Gross wrote:
>>>>>
>>>>> On Thu, Apr 24, 2014 at 8:38 AM, Thomas F Herbert
>>>>> <thomasfherbert at gmail.com> wrote:
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I think there is a problem with parsing out the OVS ATTRIBUTE nesting in
>>>>>> the
>>>>>> NL message in the datapath and I am having trouble finding the cause. I
>>>>>> think the nl msg is built up wrong in the case of the vlan push action.
>>>>>>
>>>>>> I first saw the problem while testing my patch for 802.1ad. However, I
>>>>>> also
>>>>>> see the same problem WITHOUT the patch on commits in the master branch.
>>>>>>
>>>>>> The message appears in the kern log for commit dcbe78 and subsequent
>>>>>> commits
>>>>>> on the master branch. The message does NOT appear in the 2.1 release
>>>>>> branch.
>>>>>>
>>>>>> Apr 23 07:50:11 lubuntu-12 kernel: [ 1009.493608] openvswitch: netlink:
>>>>>> Key
>>>>>> attribute has unexpected length (type=62, length=4, expected=0).
>>>>>>
>>>>>> Message is from line 300 in datapath/flow_netlink.c
>>>>>>
>>>>>> It is interesting that type 62 is the maximum key attribute value.
>>>>>>
>>>>>> The problem can be also  reproduced with the following command with
>>>>>> commit
>>>>>> f487a:
>>>>>>
>>>>>> ovs-ofctl -vdbg --protocols=OpenFlow11 add-flow br0
>>>>>> in_port=1,actions=push_vlan:0x8100,load:0x64-\>OXM_OF_VLAN_VID[],output:2
>>>>>
>>>>> This attribute is for MPLS, which userspace supports but the kernel
>>>>> doesn't. This shouldn't be generated in any case but is there any MPLS
>>>>> traffic here?
>>>>
>>>> There is no traffic at all when the message appears.
>>>>
>>>>>
>>>>> You could try bisecting since you have points in the code where it is
>>>>> working and not. It might also be worth setting the log level to debug
>>>>> to watch the upcalls and flow installations.
>>>>
>>>> OK, Jesse, I will debug by bisecting the commit history to see where it
>>>> first appeared and get back to you when I find it.
>>>>
>>>> There is no traffic yet and it occurs before any flows are pushed. The
>>>> message appears when the bridge is initially set up with the following
>>>> command.
>>>>
>>>> ovs-vsctl add-br br0
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> dev mailing list
>>>> dev at openvswitch.org
>>>> http://openvswitch.org/mailman/listinfo/dev