>> This is also the reason why m_pullup() always copies the header out of a
>> cluster even if the header was originally contiguous in the cluster.
>Hm. This should be documented.
>And we should have a M_REALPULLUP macro that does no copy unless necessary.
>But I guess this should wait until Itojun specifies what he needs to make
>IPv6 work with mbuf chains.

	For IPv6 chained headers, m_pullup is not really useful.  We should
	write a code that handles non-contiguous headers (for example,
	do not use struct tcpiphdr).  I've been trying to figure out
	what is the best route, I'll try to test something locally.

itojun