[ofw] CM ref counting issues...

[Sean Hefty]

>Yes, removing the status assignment should be fine.  But I think entering
>timewait as soon as the CEP is destroyed might be cleaner, as it allows all
>canceled MADs to be treated the same way.

I'm talking only about removing the assignment.  I've removed it from my
next-patch-to-test, along with converting the p_mad->status == success check to
an assert opposite.

I should have a patch by the end of today or tomorrow for wider review.  What I
have now does fix my issues and allows ndconn to complete.  (The ndconn
performance sucks, but that's another matter... :)