On Wed, Feb 06, 2013 at 10:53:55PM +0900, Simon Horman wrote:
> Actions may provide information to allow further decoding of
> a flow.
>
> For example:
>
> In the case of MPLS, L3 and L4 information may not initially be decoded
> from the frame as the ethernet type of the frame is an MPLS type and no
> information is known about the type of the inner frame.
>
> However, the type of the inner frame may be provided by an mpls_pop action
> in which case L3 and L4 information may be decoded providing a finer
> grained match than is otherwise possible.
>
> Signed-off-by: Simon Horman <ho...@verge.net.au>
Why does odp_flow_key_from_flow() take a separate encap_dl_type
parameter instead of using flow->encap_dl_type? There may be a good
reason, but I don't understand yet.
The actions_allow_l3_extraction() check is probably not sufficient in
facet_check_consistency(), because it misses ofpacts found via
"resubmit". We will probably need to record this in xlate_actions().
By the way, I think that we may need to adjust how MPLS information is
included in ODP flow key data. Reading datapath/README, I think that we
may need to use a format like:
eth(...), eth_type(0x8847), mpls(...), encap(ip(...), tcp(...))
whereas I believe we are currently omitting the encap and simply writing
eth(...), eth_type(0x8847), mpls(...), ip(...), tcp(...)
I think the latter will make kernel ABI compatibility problematic.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
