In some email I received from itojun@iijlab.net, sie wrote:
> 
> >>It would appear that the behaviour of m_copy()/m_copym() is changed,
> >>depending on whether or not M_EXT is set.  If M_EXT is set, then the
> >	I believe the above is the normal behavior, and this has been there
> >	for a very long time.  See TCP/IP Illustrated volume 2, section 2.9.
> >	To handle, for example, IPsec ESP header (which requires total
> >	modification of the packet), we reimplement mbuf chain copying,
> >	without cluster sharing, in few places.
> 
> 	In fact, there are only very few places we would need a deep copy.
> 	IPsec ESP handler processing is the only one item I can imagine.
> 	(do any of drivers need a deep copy?  I believe not)

Ok.

So far I've received two replies from people who've added this behaviour
in (yourself and one other) and I also require it.  Maybe there will be
others in the future, maybe not, but surely if there's two or more uses
for this, it has to be worth adding to the kernel ?

Darren