In some email I received from itojun@iijlab.net, sie wrote:
> >On Wed, Sep 25, 2002 at 10:58:02PM +1000, Darren Reed wrote:
> >
> > > Ok, explain to me why we're not fixing the problem of the mbuf being
> > > badly initialised here and instead passing a 'bad mbuf' to bpf_mtap().
> > > 
> > > If M_PKTHDR is set, then the contents of m_pkthdr are *meant* to be
> > > valid or at least that's what everyone _tells me_ (hi Jason!).
> 
> 	the change you should have made with sys/net/bpf.c change was
> 	to update all bpf_mtap() calls from DLT_NULL interfaces (uses "struct
> 	mbuf" without initialization like those done in MGET).  i had no time
> 	for that, and because of the severety of the bug, i backed the change
> 	to sys/net/bpf.c out.

Well, I've fixed up ppp (I hope).

Are there any interfaces, besides lo0, that are DLT_NULL ?

I'll look at loopback next.

Darren