2.6 oops eliminated

I got rid of an annoying oops on rmmod in the 2.6 code.  There was a
sync call after the device had been destroyed in the clean_module.  I've
never chased it down before because one could avoid it via a "don't do
that then" and the 2.6 kernel was still pretty fluid as far as
interfaces go.

I am running the enbd code fine on 2.6.12.3, but if anyone has trouble
doing so please tell me.  There has been one person who gets a crash in
the init_module routinr, in what looks entirely innocent and humdrum
initialization code that hasn't changed in ages.  It may be a compiler
issue.  I use gcc 2.95.4, the reporter uses gcc 3.3.4 (IIRC).

Peter