On 11/29/2015 02:48 PM, Ben Pfaff wrote:
> Until now, the flow table treated localnet logical ports that have a VLAN
> quite differently from those that don't.  The ones without a VLAN were
> essentially trunk ports: any packets that came in, that weren't picked off
> by a localnet port with a VLAN, were passed to the ones without a VLAN.
> This wasn't the intended behavior.
> 
> This commit changes behavior to the intended behavior.  Now, localnet ports
> without a specific VLAN only receive packets without a VLAN header or those
> with VLAN ID 0 (with that header stripped off).
> 
> Found by inspection.
> 
> Signed-off-by: Ben Pfaff <b...@ovn.org>
> ---
> Repost of a patch from Oct. 15:
> https://patchwork.ozlabs.org/patch/530988/
> 
>  ovn/controller/physical.c | 30 ++++++++++++++++++++----------
>  1 file changed, 20 insertions(+), 10 deletions(-)

First of all, sorry for missing this patch earlier.

I tried testing this patch and I'm having problems with it.  After I
apply this patch, if I exercise this code path using part of
tutorial/OVN-Tutorial.md, ovn-controller gets stuck and eats a CPU.  I
actually accidentally got two instances of ovn-controller running in
this state and almost melted my laptop.  :-)

I'm not sure where the error is, but figured I'd at least report my test
result.

To replicate, just start the sandbox and run env5's setup.

$ make sandbox SANDBOXFLAGS="--ovn"
$ ovn/env5/setup.sh

If I revert this patch, that env works fine again.

Yes, I should turn this into a test case.  I'm happy to do that once we
sort this out.

I do like the idea of this patch.  I looked at this issue when you
mentioned it on IRC one day it seemed less urgent since we do at least
drop any VLAN tagged packets at the beginning of the logical flow table.
 So, it looked like if any unexpectedly tagged packets hit an untagged
localnet port's input flow, it would just get dropped as soon as it got
to the logical flows.  Handling it more explicitly in
physical-to-logical translation does seem like a better thing to do, though.

-- 
Russell Bryant
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev

Reply via email to