>	it seems that PRU_PURGEIF logic still has a possibility for panics.
>
>	the current logic is like below.  therefore, by the time we call the
>	2nd in_pcbpurgeif0(), the interface address can be gone already.
>
>	i guess what should be done is like below.  correct me if i'm mistaken.

	oops sorry, i guess i was mistaken. IFAREF() should protect the
	situation from happening.  but anyway, it is a bit confusing...
	i guess i prefer the following order, not sure if it makes real
	difference or not.
	- remove all reference from pcbs, to in_ifaddr or ifp
	- remove all in_ifaddr
	- refresh route caches for all pcbs

itojun


--- should be
if_detach()
{
	for (all usrreq)
		xx_usrreq(PRU_PURGEIF0);

	for (all protos)
		xx_purgeif();

	for (all usrreq)
		xx_usrreq(PRU_PURGEIF);
}

xx_usrreq()
{
	if (cmd == PRU_PURGEIF0)
		in_pcbpurgeif0();
	else if (cmd == PRU_PURGEIF)
		in_pcbpurgeif();
}