> I'm going to take a look into this ASAP. However, I don't think implemen-
> ting BOOTP into the kernel (as with 1.3.75) is such a good idea. It blows
> up the kernel binary for being used just once. This should instead go into
> the bootrom, which then sends the necessary parameters from the BOOTP
> handling to the kernel with the appropriate kernel command line parameters.
> This also reduces the number of BOOTP requests from one client. Other-
> wise, the bootrom sends a request, then the kernel, and finally a BOOTP
> user process. So, why should we implement BOOTP into the kernel? IMHO, we
> should discuss this further.
(1) It's the same thing with BOOTP as it was with RARP.
(2) My patch was meant not only for true diskless machines, but also for
machines booting the kernel from a floppy or by LOADLIN -- lots of such
machines exist on our faculty.
(3) I'll make the whole thing optional.
(4) In case the boot loader passes the necessary parameters to the kernel,
no BOOTP/RARP packets are sent, thus the network load in this case is not
affected.
(5) Probably the only reason for the BOOTP user proccess started after my
BOOTP autoconfiguration is for discovery of DNS server addresses (I hope
everything else is set up properly by the kernel), which is done only
ocassionaly as it's usually specified in the config files on the mounted root
disk.
Martin Mares