Mark Atwood wrote:
>
> "David L. Parsley" <pa...@li...> writes:
> >
> > Sure, just don't make a 'linuxrc' file, instead:
> > root=/dev/ram init=/whatever/you/like (/sbin/init even).
> > This is the right way to do things now
> > I use a shell script init which later exec's the real init - still pid=1.
>
> Where does you specify those kernel cmdline parameters. As a bootp
> resource?
I supply them with mknbi-linux:
mknbi-linux --param="root=/dev/ram (etc...)"
> And isn't the kernel still in "initrd mode", wanting to do a
> pivot/init instead of halting the system as soon as that "initial
> process" dies?
No, by supplying a root device of /dev/ram, the kernel assumes that the
initrd ramdisk is the true root filesystem, and either uses the
command-line supplied init or looks for /sbin/init. Then, using
pivot_root, you can change the root filesystem whenever you like - but
usually just once during boot.
> > > Or am I stuck with having to construct an initrd that tftp's a
> > > rootdisk image down and mounts and pivots to that?
> >
> > I use userspace tftp quite a bit for my work; what tftp are you using?
> > I've patched RedHat's tftp-0.17 for command-line operation, and like it
> > pretty well. The tftp included with busybox wasn't very robust and
> > would frequently time out and fail.
>
> Thanks for the advanced warning on that.
Let me know if you want my tftp patches. It makes about a 10k
dynamically-linked binary, but that's not too bad.
regards,
David
--
David L. Parsley
Network Administrator
Roanoke College
|