You are right.

The user space code needs to distinguish between two cases:
1) The entire mask attribute is missing, then the entire key is exact
match.
2) The mask attribute is there, then the missing key attributes should be
treated as wildcarded fields.

It seems we don't need this patch, but make fixes to the user space program
instead. What do you think?


On Fri, Jun 21, 2013 at 9:55 AM, Jesse Gross <je...@nicira.com> wrote:

> It doesn't really make sense that userspace would interpret a missing
> attribute as exact match because when the flow is installed a missing
> attribute means that it is wildcarded. Therefore, if you echoed back
> the exact same flow it would have a different meaning.
>
> On Fri, Jun 21, 2013 at 9:43 AM, Andy Zhou <az...@nicira.com> wrote:
> > Correction.
> >
> > The ICMP mask will have to be fully masked in this example. But they
> still
> > need to be present in the netlink. A missing ICMP mask attribute will be
> > interpreted by the user space programs as exact match.
> >
> >
> >
> >
> >
> > On Fri, Jun 21, 2013 at 9:38 AM, Andy Zhou <az...@nicira.com> wrote:
> >>
> >> Yes,  you are right about the root cause.
> >>
> >> A fully wildcarded ip proto in the key would prevent the mask of ICMP
> (may
> >> not be fully wildcarded) to be generated.
> >> This patch fixes it.
> >>
> >> Andy
> >>
> >>
> >> On Fri, Jun 21, 2013 at 9:27 AM, Jesse Gross <je...@nicira.com> wrote:
> >>>
> >>> On Thu, Jun 20, 2013 at 11:38 PM, Andy Zhou <az...@nicira.com> wrote:
> >>> > Fix a bug where some mask attributes are missing in the netlink
> packets
> >>> > from kernel to the user space.
> >>> >
> >>> > Reported-by: Justin Pettit <jpet...@nicira.com>
> >>> > Signed-off-by: Andy Zhou <az...@nicira.com>
> >>>
> >>> Can you describe this some more? Presumably, this is because some of
> >>> the prerequisite values are either fully or partially masked out and
> >>> this is preventing latter masks from being emitted. However, since we
> >>> force prerequisites to be exact match, doesn't that mean the omitted
> >>> masks are fully wildcarded?
> >>
> >>
> >
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev

Reply via email to