>I'm still trying to debug my tunnelling problems, and I've found
>something very strange. The setup looks like this:
>
>    A -- B == C -- D
>
>The link between B and C is a GRE tunnel.
>
>When using no IPSec between B and C, A can ping D.
>
>When using only AH between B and C, A can ping D.
>
>When using only ESP, or AH + ESP between B and C, B can still ping C (on
>the non-tunnel addresses), but A can no longer ping D, nor can B ping C
>on the addresses of the tunnel interfaces themselves.

	sys/netinet/in_gre.c:gre_mobile_input() assumes that GRE header
	and IP header are continuous on a single mbuf, but the assumption
	does not hold for ESP case i guess.  i guess we need to fix
	gre_mobile_input() to perform m_pullup().

itojun