On Wed, May 1, 2013 at 3:30 PM, Thomas F Herbert
<thomasfherb...@gmail.com> wrote:
> All,
>
> I would like to introduce once again the subject of stacked vlans. We have
> implemented a network edge device using 802.1ad. It is currently implemented
> on Linux with iproute2 and a patched kernel but we would like to
> re-implement it with openvswitch.
>
> I have read earlier discussion in this mailing list that 802.1ad is not
> implemented but it seems to be at least partially in the code.
>
> I am trying to scope how much effort there would be to implement support for
> stacked vlans of some flavor in openflow.However, it looks as if a great
> deal of this work has already been done.
>
> For example, I have been going through the code in the 1.11 branch and I see
> this structure various places:
>
> struct ovs_action_push_vlan {
> __be16 vlan_tpid; /* 802.1Q TPID. */
> __be16 vlan_tci; /* 802.1Q TCI (VLAN ID and priority). */
> };
The protocol between userspace and kernel was designed to be
extensible to allow for this to be cleanly implemented but there's no
code to actually support it.
> I also see this patch in the archives,
> http://openvswitch.org/pipermail/dev/2012-June/017785.html
> that seems to do the job or not? However I also see warnings elsewhere in
> the code about using 88a8 ethertype but why should open vswitch care if it
> is in the kernel?
Both the OVS userspace and kernel module need to know how to parse and
interprete the data from the extra vlan tag in order to be able to do
anything useful with it. The "kernel" (meaning the rest of Linux) will
just deliver packets.
> 1. Is there current support for pseudo stacked vlans if the outer Ethertype
> is 8100 instead of 88a8?
No.
> 2. Can the above push action described above applied recursively?
In the kernel, yes, but userspace will never generate that.
> 3. Is the limitation about 802.1ad because of the presumed lack of support
> in the kernel (in the case of Linux) for 802.1ad. If the limitation is in
> the kernel and not in openvswitch, I can solve that problem easily. As
> stated above I have 802.1ad running on 3.1.10 in conjunction with iproute2.
Both need to support it. There has been some recent work in the
upstream kernel for 3.10.
> 4. We may be interested in contributing a complete implementation of 802.1ad
> and maybe 802.1ah. What is your sense of the LOE?
I suspect the most difficult part is how to handle the concept of
stacking in userspace.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
