HEADS UP: Lots of minor fifo tweaks

Yeah, the topic says "minor", but the key is "lots".  I believe I have 
fixed a moderate number of minor bugs in the POSIX fifo code (fifofs). 
This was a result of writing a set of regression tests before trying to 
fix a serious race condition in fifofs, as I was afraid I would break 
fifofs subtly in the process.  The result was finding lots of bugs 
instead, which are now believed to be fixed, and the regression test suite 
runs to completion without errors.  However, this doesn't mean I haven't 
broken something, as the semantics of fifos are (at best) tricky.  If you 
use applications that use fifos, please test them on a recent kernel from 
the CVS HEAD.  I plan to merge many if not all of these changes to 
RELENG_6 in the immediate future, and RELENG_5 a bit after that.  I'd much 
rather hear about bugs sooner rather than later.

Thanks,

Robert N M Watson