[ovs-discuss] Hello version 0

[Ben Pfaff]

Send a patch that includes a test, and I'll apply it.

On Thu, Jul 24, 2014 at 11:33:56AM -0400, Anup Khadka wrote:
> Ben,
> It was more about sanity tests on OVS than anything else (and hardware
> switches using OVS as its stack). Naturally, if a controller is trying to
> negotiate version 0x00, there is something wrong with the controller, no
> question about it.
> 
> The issue is not that. The issue was more about how OVS should respond to a
> request as such (or few other bad requests that I am testing), and if there
> was anything in the standard that I could have missed that explains the
> right behavior of the switch in such scenarios.
> 
> 0x00 doesn't represent an OpenFlow version, no question, but there are
> error codes for bad versions under bad request. When and why those codes
> should be used isn't completely clear, at least to me, and I was seeking
> insight.
> 
> Just reading through the standard,   the non-experimental versions are:
> 0x01, 0x02, 0x03, 0x04, 0x05, and 0x06. The experimental versions are 0x81
> - 0x99.
> 
> The OVS as currently constructed, I believe, sends a Hello failed with
> incompatible error for every version 0x01 - 0xFF.
> 
> Why does the switch silently ignore a Hello message with version 0, but
> responds with a hello failed for version 0x01 - 0xFF even though its not
> explicitly stated as a valid openflow version. In other words, why is a
> version, say 0xFE, a valid version?
> 
> If you think, the switch's current behavior is expected, then its alright.
> I am trying to get a better insight, and make sure its consistent.
> 
> Thanks,
> Anup
> 
> 
> 
> 
> 
> 
> On Thu, Jul 24, 2014 at 11:07 AM, Ben Pfaff <blp at nicira.com> wrote:
> 
> > On Wed, Jul 23, 2014 at 07:59:10PM -0400, Anup Khadka wrote:
> > > If OVS receives a Hello packet with version set to 0, it logs an error
> > > "received message while expecting hello" and change's vconn's state
> > without
> > > sending any message back to the controller.
> > >
> > > Is this a desired behavior? The standard isn't really clear on how to
> > > process a Hello packet with version == 0 (or if its even a valid OpenFlow
> > > packet at that point), but I think its reasonable for a controller to
> > > expect either one of the following two messages:
> > >
> > > 1. An OFPT_ERROR message with OFPET_HELLO_FAILED error type, and
> > > OFPHFC_INCOMPATIBLE error code (this was the behavior of OVS 1.4.0 stack,
> > > but the code base, and checks were significantly different then. The old
> > > code didn't really check for version number range, simply transitioned to
> > > VCS_SEND_ERROR state, and sent an error back to controller)
> > >
> > > 2. An OFPT_ERROR message with OFPET_BAD_REQUEST error type, and
> > > OFPBRC_BAD_VERSION error code.
> >
> > It's not great behavior.
> >
> > I don't know how OVS would send OFPT_ERROR in this case, though, because
> > 0x00 doesn't represent an OpenFlow version and so there aren't any error
> > codes to use.
> >
> > If your controller is trying to negotiate version 0x00, it's buggy.  Fix
> > it.
> >