[ovs-dev] OVS Push VLAN Tag Hack and Problem

Naga Rohit, IIT Guwahati snagarohit at gmail.com
Wed May 23 08:05:31 UTC 2012


Thanks for the quick response Mr Ben. However, I am bit confused on
which part of the code is responsible for the optimizations..

FILE: ~/openvswitch/lib/dpif.c

Guessing the functionality, I found "dpif_flow_put" to be probably
responsible for it. It is being called by "facet_put__" in
~/openvswitch/ofproto/ofproto.c which is inturn being called by
"facet_install", "facet_revalidate" and "facet_active_timeout" Now,
where does optimizations happen ?

Also, can you please let me know any way of specifying actions in
dpctl so that optimizations doesn't take place ?

Also, another strange thing I have found out is that when I commented
out "modify_vlan_tci" in ~/openvswitch/datapath/actions.c, it din't
throw up any error in make, install time and even at run time using
mininet (with ovsk switch and "dpctl
actions=mod_vlan_vid:xx,output:yy)

Thanks a lot,
Rohit

On Wed, May 23, 2012 at 10:50 AM, Ben Pfaff <blp at nicira.com> wrote:
> On Wed, May 23, 2012 at 10:44:45AM +0530, Naga Rohit, IIT Guwahati wrote:
>> Where am I going wrong ? Does the changes I have made really result in
>> behavior that I intend to see ?
>
> Code in ofproto-dpif.c "optimizes out" actions that it considers
> redundant, including consecutive changes to the VLAN without an
> intervening output action.



More information about the dev mailing list