[dpdk-dev] [PATCH v2 2/6] ethdev: add GTPC and GTPU items

[Adrien Mazarguil]

Hi Beilei,

On Tue, Sep 12, 2017 at 06:40:50AM +0000, Xing, Beilei wrote:
> Hi Adrien,
> 
> > -----Original Message-----
> > From: Adrien Mazarguil [mailto:adrien.mazarguil at 6wind.com]
> > Sent: Thursday, September 7, 2017 8:20 PM
> > To: Xing, Beilei <beilei.xing at intel.com>
> > Cc: Wu, Jingjing <jingjing.wu at intel.com>; Chilikin, Andrey
> > <andrey.chilikin at intel.com>; dev at dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v2 2/6] ethdev: add GTPC and GTPU items
> > 
> > Hi Beilei,
> > 
> > I assume this patch supersedes [1]?
> > 
> > [1] http://dpdk.org/ml/archives/dev/2017-August/073501.html
> > 
> > Thanks for merging testpmd and the API change as a single patch, I still have
> > a few comments, see below.
> > 
> > (please add "flow API" somewhere in the title by the way)
> 
> Thanks for all your comments.
> Yes, http://dpdk.org/ml/archives/dev/2017-August/073501.html is superseded and did merging testpmd and API change.
> I will update title in next version.

All right, thanks.

> > On Thu, Sep 07, 2017 at 07:20:59PM +0800, Beilei Xing wrote:
> > > This patch adds GTPC and GTPU items to generic rte flow, and also
> > > exposes the following item fields through the flow command:
> > >
> > > - GTPC TEID
> > > - GTPU TEID
> > >
> > > Signed-off-by: Beilei Xing <beilei.xing at intel.com>
> > 
> > Won't there be a need to match nonspecific GTP traffic as well (both GTP-C
> > and GTP-U a once), since they use the same structure?
> > 
> > I'm not familiar with the protocol at all so I wonder if you should maybe leave
> > the GTP item in addition to those two.
> > 
> 
> Agree, I will leave the GTP item in next version.
> 
> GTP-C and GTP-U use the same structure, the difference between them is UDP port, 2123 is for GTP-C, and 2152 is for GTP-U.
> Add GTP-C and GTP -U item since I want to design a user-friendly CLI.
> 
> For example, if user wants to add such flow: assign GTP-U packets with TEID 0x123456 to queue 14. 
> Then use can use following CLI:
> flow create 0 ingress pattern eth / ipv4 / udp / gtpu teid is 0x123456 / end actions queue index 14 / end
> instead of below CLI to distinguish GTP-C and GTP-U with UDP port:
> flow create 0 ingress pattern eth / ipv4 / udp dst spec 2125 dst mask 0 / gtp teid is 0x123456 / end actions queue index 14 / end

I agree with you from a usability standpoint it's much nicer. I have one
question though, could one send GTP-C / GTP-U traffic using nondefault UDP
ports and expect hardware to match it without explicitly specifying a port
in the UDP item, that is, is there some property in GTP-C / GTP-U traffic
outside the UDP port that would theoretically allow a host (even a stateful
one) to tell them apart?

If it really depends on the UDP port only, not specifying one will use
hardware defaults regardless of the item (GTP / GTP-U / GTP-C). However if
like VXLAN, this default value can be modified outside of rte_flow, most
users will have to specify it regardless in order to get consistent results
across various vendors/adapters.

In any case I don't mind three items. GTP encompasses both GTP-U and GTP-C
(possibly two different UDP ports at once), while GTP-U and GTP-C match
exactly one. You only have to describe this properly in the documentation.

Thanks.

-- 
Adrien Mazarguil
6WIND