Diskless Boot Steps

Diskless Booting Howto
pauln@psc.edu

Parts of this document refers to steps taken from:

Overview
This document describes the steps needed to create a basic environment for suporting diskless linux nodes. This howto mentions three different node types: the prototype (golden client), server, and client. The prototype node (described below) is a node with a disk that is used for the initial installation of the operating system. For the purpose of this document, the server node also runs linux. The server node is responsible for holding and serving files and filesystems that the client will need. The client is the machine which will be booted using the diskless methods described here. The prototype node may be used as the client.

Client Configuration

Install Linux on a prototype node. The prototype's filesystems will serve as the basis for our diskless nodes. This installation should remain so that it may be updated later. I was using kickstart for node imaging so now I can use it to create the prototype. Here's the ks file I was using for the initial install.

Cleanup unwanted cruft, this script can help get rid of unwanted rpms.

Build a kernel that uses few or no modules, here's an example CONFIG. After installing the kernel (make install) and you've ensured it's correct, run "make clean" - this save about 700MB of disk space. Be sure to set the ramdisk size to 64MB or above.

Copy the kernel to your server's /tftpboot/ directory. # scp prototype:/boot/vmlinuz /tftpboot/vmlinuz

Server Configuration

Services
- DHCPd v3.0 or greater w/ PXE extensions
- TFTPd server
- Enable Syslog from other hosts's on the network
Setup
- Example dhcpd.conf
- PXE Linux
  - mkdir /tftpboot/pxelinux.cfg/
  - cp pxelinux.0 /tftpboot/
    The "pxelinux.0" file may be found on your system if you've installed the syslinux rpm, otherwise download pxelinux and use the prebuilt "pxelinux.0" file.
  - Create bootfile for client(s) in /tftpboot/pxelinux.cfg/, the name of the bootfile is the client's ip address in hex. /tftpboot/pxelinux.cfg/0A000069 would be for the client 10.0.0.105. Bootfiles may also be named after the clients MAC address (i.e. pxelinux.cfg/01-00-04-23-a8-db-52).

Preparing RootFS Image
Now that the "golden client" (borrowed from systemimager) is created you want to appropriate the root and usr filesystems to be used respectively as a ramdisk and nfs mounted /usr directory.

** Take caution in this phase not to accidentally delete vital directories on your server. When making modifications to your ramdisk make sure to use "chroot".

mkdir /cluster/rootfs; cd /cluster/rootfs # Create a directory to hold the root dump

ssh prototype "tar cl /" | tar xf - # Grab the entire root directory

mkdir /cluster/userfs; cd /cluster/userfs # Create a directory to hold the usr dump
If the prototype doesn't have a separate /usr/ partion then move the contents of the /cluster/rootfs/usr directory to /cluster/userfs
...else...

Add /cluster/userfs to /etc/exports for nfs mounting.

          (pauln@panda:pxelinux.cfg)$ cat /etc/exports
	  /cluster/userfs localhost(rw,async,no_root_squash)
	  /cluster/userfs client(ro,async,no_root_squash)

chroot /cluster/rootfs; # change root to our netboot /
mount localhost:/cluster/userfs /usr # mount /usr in the same manner as the real clients

Setup of "/" Partition
- rm -Rf /lib/modules/"any kernel you're not using"
  # FC2 "cd /lib/modules; rm -Rf 2.6.5-1.358 2.6.5-1.358smp"
- If you have files in /var/ or /boot/ then remove them all.
- cd var; mkdir -p ./log ./lock ./lock/subsys ./lib ./lib/ntp ./lib/nfs ./lib/dhcp ./lib/rpm ./preserve ./run ./spool ./tmp ./db ./cache ./empty ./empty/sshd ; chmod 111 empty/sshd
- Replace/edit /etc/fstab, /etc/rc.sysinit and /etc/syslog.conf

Type "ctrl-d" to escape ramdisk
umount `pwd`/usr # to unmount the usr directory

Creating the Ramdisk
Next step is take our wonderfully tailored rootfs and turn it into a compressed file to be used as a initrd ramdisk by our clients.

cd /tftpboot; dd if=/dev/zero of=rootfs.img count=1024 bs=65536 # Make a 64MB, zero-filled file to serve as the ramdisk image
/sbin/mke2fs -F -N 25000 -L ROOT rootfs.img < /dev/null # Create a filesystem in the image file
/sbin/losetup /dev/loop2 rootfs.img # Bind the image file to a block device
/sbin/tune2fs -r 0 -C 0 -c -1 /dev/loop2; /sbin/fsck.ext2 /dev/loop2 # Tune up and fsck the filesystem on the image
mkdir /loop; mount /dev/loop2 /loop # Verify the image is mountable.
cp -a /cluster/rootfs/* /loop # copy the rootfs files into the ramdisk.
umount /loop; losetup -d /dev/loop2
gzip -c rootfs.img | dd of=rootfs.img.gz # zip the ramdisk file.

Now you have an initrd file that your kernel can use as a root filesystem!

Back to PXE

Verify that the bootfile is correct. The first line is the name of the kernel that the client will request - you should have a kernel by that name in the /tftpboot/ directory. The append line is where the initrd root filesystem is specified. In this example we used the "rootfs.img.gz"

Set your client's bios to boot from the network card. This can be done via the boot menu in your system bios. Something labeled "IBA" or "INT18" is a probably network device. The network devices should be placed before the hard drive.

Reboot your client while watching (tail -f) /var/log/messages on your dhcp/tftp server. Your client machine should boot up with the diskless image.