[libvirt] [PATCHv2] Configure native vlan modes on Open vSwitch ports

Tue Apr 23 16:13:21 UTC 2013

On 04/23/2013 11:37 AM, james robson wrote:
> On Mon, 2013-04-22 at 17:08 -0400, Laine Stump wrote:
>> (I'm not sure what you did differently when you sent this mail, but
>> somehow your mailer botched the "In-Reply-To:" header, which broke the
>> threaded display in Thunderbird. No big deal, but I thought you might
>> want to know.)
> I expect this was a side affect of getting the digest emails from the
> list.
>
>> On 04/22/2013 12:51 PM, james robson wrote:
>>>> Date: Thu, 18 Apr 2013 14:14:32 -0400
>>>> From: Laine Stump <laine at laine.org>
>>>> To: libvir-list at redhat.com
>>>> Subject: Re: [libvirt] [PATCHv2] Configure native vlan modes on Open
>>>> 	vSwitch ports
>>>> Message-ID: <51703808.2000504 at laine.org>
>>>> Content-Type: text/plain; charset=ISO-8859-1
>>>>
>>>> On 04/18/2013 01:44 PM, james robson wrote:
>>>>> Hello,
>>>>> Has any one been able to review this yet? I realise that the 'Since
>>>>> 1.0.3' in the doc page is now out of date, but is the code itself
>>>>> acceptable?
>>>> I was hoping that someone with more knowledge of Open vSwitch and/or
>>>> vlan tagging/trunking/native mode would repond to the message (Kyle?)
>>>> but there was silence instead...
>>>>
>>>>
>>>>> diff --git a/tests/networkxml2xmlin/openvswitch-net.xml
>>>>> b/tests/networkxml2xmlin/openvswitch-net.xml
>>>>> index a3d82b1..93c49d5 100644
>>>>> --- a/tests/networkxml2xmlin/openvswitch-net.xml
>>>>> +++ b/tests/networkxml2xmlin/openvswitch-net.xml
>>>>> @@ -21,4 +21,13 @@
>>>>>        <parameters profileid='alice-profile'/>
>>>>>      </virtualport>
>>>>>    </portgroup>
>>>>> +  <portgroup name='tagged'>
>>>>> +    <vlan native_mode='tagged' native_tag='123'>
>>>>> +      <tag id='555'/>
>>>>> +      <tag id='444'/>
>>>>> +    </vlan>
>>>>> +    <virtualport>
>>>>> +      <parameters profileid='tagged-profile'/>
>>>>> +    </virtualport>
>>>>> +  </portgroup>
>>>> As brought up again in a separate conversation today, we prefer to use
>>>> camelCase rather than underscored in attribute and element names. So, if
>>>> we were to use the layout you're proposing, the attributes should be
>>>> called "nativeMode" and "nativeTag".
>>>>
>>>> However, I'm wondering if there might be a better way to structure it.
>>>> What about this?
>>>>
>>>>
>>>>    <vlan trunk='yes'>
>>>>      <tag id='123' native='tagged|untagged'/> (or whatever values are
>>>> appropriate)
>>
>> Sounds like this can be "native='yes'", since there is only the
>> possiblity of a tag being the native tag, or *not* being the native tag.
>>
>>
>>>>      <tag id='555'/>
>>>>      <tag id='444'/>
>>>>    </vlan>
>>>>
>>>> Do I understand correctly that native mode is telling what to do with
>>>> packets that come in untagged, and that (using your nomenclature
>>>> "native_mode='yes' native_tag='123'" means "when an untagged packet come
>>>> in from this interface, it should be tagged as 123 before forwarding"?
>>> That is correct, setting the native vlan changes how an untagged packet
>>> is handled when it enters the port. The difference between the 'tagged'
>>> and 'untagged' modes is in how packets on the native vlan are processed
>>> before exiting the port.
>>>
>>>
>>>> And what happens when native_mode='yes' but there is no native_tag?
>>> In that case you configuration is invalid, and will get an error. 
>>
>> Okay, then doing the config the way I suggest would eliminate that
>> possibility, so it has an upside :-)
>>
>>
>>>
>>>> (that's what I was trying to describe with <tag id='123'
>>>> native='untagged'/>, but I don't even know if that makes sense, because
>>>> I don't know exactly what is the native vlan tag and what is done with
>>>> it :-)
>>> That arrangement would make sense, I chose the arrangement I did for two
>>> main reasons. There can only be one native vlan on a port, making it an
>>> attribute of the 'vlan' tag enforces this. Also, I wanted to keep the
>>> validation and processing separate rather than add 'if native' branches
>>> to the loops that operate on the vlan id list.
>>> I can see the advantage of having a single setting to configure the
>>> native vlan, rather than the two attributes I proposed. If the new
>>> suggestion is preferred I can rework my patch to use that format.
>> I think it's simpler, to do it that way, yes.
>>
>>
>>>
>>>> Also, is it valid to have a native_mode/native_tag if trunk='no'? (right
>>>> now trunk is automatically set to 'yes' if there is more than one vlan tag)
>>> It isn't valid to have trunk='no' and the native settings. Therefore
>>> "<vlan trunk='no' native_mode='tagged' native_tag='123'>" will get an
>>> error if you try to enter it. 
>>> If no "trunk" attribute is set explicitly then it will be set to 'yes'.
>>> This means "<vlan native_mode='tagged' native_tag='123'>" is equivalent
>>> to "<vlan trunk='yes' native_mode='tagged' native_tag='123'>".
>> Likewise, if you have more than one <id tag='x'/> element, trunk will
>> automatically be set to yes.
>>
>> So in the end, if you don't foresee any problems with it, I think I do
>> prefer this:
>>
>>    <vlan trunk='yes'>
>>      <tag id='123' native='yes|no'/> (default is 'no', only one allowed to be yes)
>>      <tag id='555'/>
>>      <tag id='444'/>
>>    </vlan>
> The only problem with that proposal is that there is no way to set the
> mode to be 'tagged' or 'untagged'.

Ah. I had inferred from the discussion that (unlike what I'd originally
assumed) there was only an off or an on. What does it mean to have 
"tagged" native mode? I thought the purpose of native was to designate a
tag that, when encountered on a packet, would be stripped from the
packet before forwarding; is "tagged" intended to say "when you get an
untagged packet to send out this port, tag it with this id", and
"untagged" intended to mean "when you get a packet to send out this port
that is tagged with id='x', untag it before forwarding it"?

>  What about the following:
>   <vlan trunk='yes'>
>     <tag id='123' nativeMode='tagged|untagged|none'/> (default is none)
>     ...
>  </vlan>
>
> I think nativeMode would be more appropriate than simply 'native' for
> choosing between tagged or untagged, because it more descriptive. 

Especially if you think there might be other "native-related" attributes
later. Even if not, I don't see a problem with this naming.

>> Note that we'll be going into freeze for 1.0.5 soon, so if you are able
>> to rework the patch within the next couple days, it should go into
>> libvirt-1.0.5 (which *might* end up in Fedora 19?)
> If nothing major goes wrong I should have this done before Friday.

Okay. Sounds like DV is planning to freeze on Friday (which is very late
Thursday night for most of us), so it would be good if there was some
time for review before then.