* write via libpcap (since i thought there would be some fancy
api to deal with endianness. after doing 80% of the work i
noticed there wasn't, but went ahead anyway)
only the in-memory one. Since the in-memory one uses struct timeval
and the on-disk one uses 32bit time_t, things didn't work so well
on -current. Generate a suitable local packetheader to fix things.
(don't think it's worth writing via libpcap for something as simple
as this)
noticed both by bouyer and the shmif_dumpbus test
mapped. (I had a very interesting time last night trying to figure
out what was going when things apparently got compiled with a 64bit
time_t on nb5. this caused only 4k of the bus file to be mapped...)
