On Tue, Mar 30, 2010 at 1:28 AM, Michael C. Robinson <[email protected]> wrote: > filesystem? > > I have a hunch that I want to do an initial root file system that mounts > the machine specific root file system and pivots to it. Yes. A gz compressed cpio file with the contents of you script, shell and any necessary libraries and modules if needed. Yes you can have modules for nfs root when you have one of these.
> > How do I go about doing this? I have a Linux from scratch system > that I am trying to get the necessary pieces from. The init scripts > are designed for local disk based booting, I need to create new ones. > A gentoo howto on diskless machines shows ip=dhcp with a share > assigned for the root filesystem that is obviously not a dynamic > reference. If I am doing 192.168.5.x for each work station > respectively, I should determine somehow at boot which share is needed > based on which machine is booting. The host machine would just export all workstation paths from the same location (or not if you like complicated). >From the client, use the hostname or ip address as part of the path. > > Any pointers on information about how to write rc scripts for system V > style init? What is the first script that gets executed during boot? Nope. In the initrd you have a single script not a system V set of scripts. If you have a Fedora or Centos 5 machine then you have some examples of init scripts for your initrd. > > On a hunch, I figure that I need a customized etc, a copy of bin, sbin, > and lib off of the LFS system, a couple of directories for the kernel > to create an initial root file system along with a device directory. > I'm wondering if my initial root filesystem can use udev and how to > set that up? Directories such as home and usr should be, I figure, > separate NFS shares that are shared by all workstations. Yes a full copy of your server file system for each workstation. Hardlinks and read-only on things that you don't really need to change (or want changed). Home directories -> automounter with NIS. Udev is just a couple of processes associated kernel options that mknod based on kernel devices found on a tmpfs mounted directory. > > The following shows the base directory for what I plan on exporting via > NFS: > > [ad...@goose ~]$ cd /nfsroot/lfsp3/ > [ad...@goose lfsp3]$ ls > boot eagle home usr vmeagle > > The following is what I have so far for the initial root system: > > [ad...@goose lfsp3]$ cd boot > [ad...@goose boot]$ ls > bin dev etc lib proc sbin srv sys > > The following shows the local disk installation of Linux From Scratch on > the server side: > > [ad...@goose boot]$ ls /lfsp3/ > bin dev home lost+found mnt proc sbin sys usr > boot etc lib media opt root srv tmp var > > Note that there are some directories that I am not copying at all and so > far etc in the boot directory is being created from scratch. > > So far what I have come up with does not allow for an initial console. > > I suppose an initrd is an option, but I don't think I have to do that in > the traditional way. I'm thinking mount the machine specific NFS share > on the mounted root file system and do a chroot. > > There are certain things I want to synchronize such as the password > files under etc, but I don't know how to go about that. It would be > nice if for workstation A there could be configuration information > specific to that machine. The purpose of this NFS root system is > to make doing backups to a network server easy. I am thinking an > nfs root system that can mount and dump local drives. I am also > intending to have network troubleshooting tools installed. > You might want to reference the LTSP and derivatives. http://k12ltsp.org/mediawiki/index.php/Main_Page http://en.wikipedia.org/wiki/Diskless_Remote_Boot_in_Linux _______________________________________________ PLUG mailing list [email protected] http://lists.pdxlinux.org/mailman/listinfo/plug
