Hi, The ofprotocol in the reference switch is more than just a relayer process. If I remember correctly it also deals with in-band control and port status handling. For this latter feature it needs to know what ports were assigned to the datapath, and might use a feature request of its own.
Regards, Zoltan. > -----Original Message----- > From: [email protected] > [mailto:[email protected]] On Behalf Of Cayle Spandon > Sent: Friday, January 13, 2012 1:36 PM > To: Murphy McCauley > Cc: [email protected] > Subject: Re: [nox-dev] features-reply without features-request > > Hi Murphy, > > Thanks for the reply. > > No, I am not using NOX (Yes indeed - I should have gone to > the OpenFlow mailing list. I will.). > > This is a new controller which is still under development. I > am sure the controller does not send a features-request > (because that part has not been coded yet). So, yes, the > controller is "out of spec" for now. > > So I am sure that the features request comes from > "ifprotocol" (which is not a controller, as far as I know). > It appears to send it "proactively" as you say, and this is > exactly what seemed surprising (and frankly incorrect) to me. > > Also, I am using the OpenFlow dissector in wireshark, so I > quite sure that the list of packets is accurate in the > summary below (i.e. I did not miss packets because there > multiple OpenFlow packets in a single TCP packet). > > On Wed, Jan 11, 2012 at 3:59 PM, Murphy McCauley > <[email protected]> wrote: > > I can't speak as to the particulars of the reference switch > (you might try one of the OpenFlow mailing lists). > > > > However, the spec does state that part of the handshake > sequence is that the controller sends a features request when > a connection is established. I don't see this in your > sequences of messages below. NOX should do this. Are you > using NOX? If so, is it possible you missed this message > (remember that multiple OpenFlow messages can be in the same > packet, so it is possible it's in packet #4 along with the > HELLO or something -- you'll have to look at the detail view > of the packet, not just the summary line)? > > > > If the message really isn't there, I suppose it's possible > that ofprotocol does this proactively, but in this case I > believe your controller is out of spec. > > > > -- Murphy > > > > On Jan 11, 2012, at 6:00 AM, Cayle Spandon wrote: > > > >> Are OpenFlow controllers expected to process unsolicited > feature-reply messages? > >> > >> I am running the following setup: > >> > >> controller ----- ofprotocol ----- ofdatapath > >> > >> Using wireshark I see the following sequence of messages: > >> > >> #1 ofprotocol -> ofdatapath : hello > >> #2 ofprotocol -> controller : hello > >> #3 ofdatapath -> ofprotocol : hello > >> #4 controller -> ofprotocol : hello > >> #5 ofprotocol -> ofdatapath : features-request > >> #6 ofdatapath -> ofprotocol : features-reply > >> #7 ofprotocol -> controller : features-reply > >> > >> This means that the controller gets a features-reply even > though it > >> never sent a features-request. > >> > >> It appears that ofprotocol originates the features-request to the > >> switch (which is somewhat surprising because I thought it was > >> supposed to be a dumb pass-through) and the forwards the > >> features-reply to the controller. > >> _______________________________________________ > >> nox-dev mailing list > >> [email protected] > >> http://noxrepo.org/mailman/listinfo/nox-dev > > > _______________________________________________ > nox-dev mailing list > [email protected] > http://noxrepo.org/mailman/listinfo/nox-dev > _______________________________________________ nox-dev mailing list [email protected] http://noxrepo.org/mailman/listinfo/nox-dev
