3. Preparing the server

Now it's time to prepare the server to serve diskless clients.

3.1 Building a kernel

The first thing todo is build a kernel with the nescesarry stuff in to support root over nfs. Take the following steps to build your kernel:

1. Since we'll be using redhat-5.2 with kernel-2.2 you should asure yourself that your redhat-5.2 is kernel-2.2 ready. RedHat has got an excellent howto on this.
2. I use the same kernel for both server and ws, to avoid module conflicts since they share the same /lib/modules. If this is not possible in your situation, fake different kernel versions by editing the version number in the kernel's top makefile. These different versionsnumbers will avoid any conflicts.
3. Besides the usual stuff the kernel should have the following:
   • ext2 compiled in (if used on server, or for both)
   • nfs and root-over-nfs compiled in (if used on client or both), to get the nfs over root option in 2.2 enable ip-autoconfig in the network options. We'll use bootp as configuration method.
   • ws networkcard support compiled in (if used on client or both)
   • compile devfs in (required for client, also nice for server)
   • anything else you normally use, modules for all other devices used on either the server or all / some ws etc.
4. The kernel-src needs to be edited to make the default root-over-nfs mount: /tftpboot/<ip>/root instead of just /tftpboot/<ip>. This is to get a clean tree in /tftpboot with one dir per ws containing both the root for it (a link to the actual server root) and any ws specific dirs.
   • For 2.0 This is a define in: "include/linux/nfs_fs.h" called "NFS_ROOT"
   • For 2.2 This is a define in: "fs/nfs/nfsroot.c"
5. Now just compile the kernel as usual, see the kernel-howto.
6. If you don't have /dev/nfsroot yet, create it by typing:

   mknod /dev/nfsroot b 0 255.

7. After compiling the kernel set the root to nfsroot by typing:

   rdev <path-to-zImage>/zImage /dev/nfsroot

8. Before booting with devfs you need to make a few changes to /etc/conf.modules, append the contents of the conf.modules in the devfs documentation to it.
9. Since this new kernel is compiled for autoconfig of ip's it will try to autoconf the ip of the server during bootup. Which ofcourse will fail since it gives out the ip's. To avoid a long timeout add: append="ip=off" To the linux section of /etc/lilo.conf.
10. Run lilo and boot the new kernel.
11. Due to devfs you'll have lost all symlinks on the server. With redhat this is usually /dev/mouse and /dev/cdrom. Recreate these. If you also used to use special ownerships, chown to appropiate files in /dev. Now save the /dev settings (in /etc/sysconfig, since they might be ws specific):
    • Copy rc.devfs from the devfs documentation in the kernel source to /etc/rc.d/rc.devfs and make it executable
    • Save the settings by typing:

      /etc/rc.d/rc.devfs save /etc/sysconfig

3.2 Creating and populating /tftpboot, making symlinks for /tmp etc.

The next step is to create and populate /tftpboot

The automagic part

This is all handled by a big script since putting a long list of commands into this howto seemed pretty useless to me. If you want todo this manual just read the script and type it in as you go ;)

This setup script thus some nasty things like nuke /tmp, temporary kill syslog, umount /proc. So make sure that noone is using the machine during this, and that X isn't running. Just making sure your the only one logged in on a text-console is enough, no need to change runlevels.

DISCLAIMER: this script has been tested but nevertheless if it messes up your server your on your own. I can take no responsibility what so ever. Lett me repeat this howto is only for experienced linux sysadmins. Also this is script is designed to be run once and I really mean once. Running it twice will nuke: /etc/fstab, /etc/X11/XF86Config, /etc/X11/X and /etc/conf.modules.

Now with that said, just cut and paste the script make it executable, execute it and pray to the holy penguin that it works ;)

#!/bin/sh

SERVER_NAME=`hostname -s`

###
echo creating /etc/rc.d/rc.ws
#this basicly just echos the entire script ;)
echo "#root on nfs stuff

SERVER=$SERVER_NAME

#we need proc for mtab, route etc
mount -t proc /proc /proc

IP=\`ifconfig eth0|grep inet|cut --field 2 -d ':'|cut --field 1 -d ' '\`

#if the first mount fails we're probably the server, or atleast something is
#pretty wrong, so only do the other stuff if the first mount succeeds
mount \$SERVER:/tftpboot/\$IP/sysconfig /etc/sysconfig -o nolock &&
{
   #other mounts
   mount \$SERVER:/home /home -o nolock
   mount \$SERVER:/ /\$SERVER -o ro,nolock

   #/var
   echo Creating /var ...
   mke2fs -q -i 1024 /dev/ram1 1024
   mount /dev/ram1 /var -o defaults,rw
   cp -a /tftpboot/var /

   #network stuff
   . /etc/sysconfig/network
   HOSTNAME=\`cat /etc/hosts|grep \$IP|cut --field 2\`
   route add default gw \$GATEWAY
   ifup lo
}

#restore devfs settings
/etc/rc.d/rc.devfs restore /etc/sysconfig

umount /proc" > /etc/rc.d/rc.ws

###
echo splitting runlevel 3 for the client and server
mv /etc/rc.d/rc3.d /etc/rc.d/rc3.server
cp -a /etc/rc.d/rc3.server /etc/rc.d/rc3.ws
rm /etc/rc.d/rc3.ws/*network
rm /etc/rc.d/rc3.ws/*nfs
rm /etc/rc.d/rc3.ws/*nfsfs
rm /etc/rc.d/rc3.ws/S99local
ln -s /etc/sysconfig/rc.local /etc/rc.d/rc3.ws/S99local
ln -s /etc/rc.d/rc3.server /etc/sysconfig/rc3.d
ln -s /etc/sysconfig/rc3.d /etc/rc.d/rc3.d

###
echo making tmp a link to /var/tmp
rm -fR /tmp
ln -s var/tmp /tmp

###
echo moving various files around and create symlinks for them
echo mtab
/etc/rc.d/init.d/syslog stop
umount /proc
touch /proc/mounts
mount /proc
/etc/rc.d/init.d/syslog start
rm /etc/mtab
ln -s /proc/mounts /etc/mtab
echo fstab
mv /etc/fstab /etc/sysconfig
ln -s sysconfig/fstab /etc/fstab
echo X-config files
mkdir /etc/sysconfig/X11
mv /etc/X11/X /etc/sysconfig/X11
ln -s ../sysconfig/X11/X /etc/X11/X
mv /etc/X11/XF86Config /etc/sysconfig/X11
ln -s ../sysconfig/X11/XF86Config /etc/X11/XF86Config
echo conf.modules
mv /etc/conf.modules /etc/sysconfig
ln -s sysconfig/conf.modules /etc/conf.modules
echo isapnp.conf
mv /etc/isapnp.conf /etc/sysconfig
ln -s sysconfig/isapnp.conf /etc/isapnp.conf

###
echo creating a template dir for the ws directories
echo /tftpboot/template
mkdir /home/tftpboot
ln -s home/tftpboot /tftpboot
mkdir /tftpboot/template
mkdir /$SERVER_NAME
echo root
ln -s / /tftpboot/template/root
echo sysconfig
cp -a /etc/sysconfig /tftpboot/template/sysconfig
rm -fR /tftpboot/template/sysconfig/network-scripts
ln -s /$SERVER_NAME/etc/sysconfig/network-scripts \
 /tftpboot/template/sysconfig/network-scripts
echo NETWORKING=yes > /tftpboot/template/sysconfig/network
echo `grep "GATEWAY=" /etc/sysconfig/network` >> /tftpboot/template/sysconfig/network
echo "/dev/nfsroot / nfs defaults 1 1" > /tftpboot/template/sysconfig/fstab
echo "none /proc proc defaults 0 0" >> /tftpboot/template/sysconfig/fstab
echo "#!/bin/sh" > /tftpboot/template/sysconfig/rc.local
chmod 755 /tftpboot/template/sysconfig/rc.local
rm /tftpboot/template/sysconfig/rc3.d
ln -s /etc/rc.d/rc3.ws /tftpboot/template/sysconfig/rc3.d
rm /tftpboot/template/sysconfig/isapnp.conf
echo var
cp -a /var /tftpboot/var
rm -fR /tftpboot/var/lib
ln -s /$SERVER_NAME/var/lib /tftpboot/var/lib
rm -fR /tftpboot/var/catman
ln -s /$SERVER_NAME/var/catman /tftpboot/var/catman
rm -fR /tftpboot/var/log/httpd
rm -f /tftpboot/var/log/samba/*
for i in `find /tftpboot/var/log -type f`; do cat /dev/null > $i; done
rm `find /tftpboot/var/lock -type f`
rm `find /tftpboot/var/run -type f`
echo /sbin/fsck.nfs
echo "#!/bin/sh
exit 0" > /sbin/fsck.nfs
chmod 755 /sbin/fsck.nfs

echo all done
 

Manual adjustments to some files

Now we need to make a few manual adjustments to the server:

1. The ws setup script has to be sourced at the very beginning of rc.sysinit, so add the following lines directly after setting the PATH:

   ---

   #for root over nfs workstations.
   /etc/rc.d/rc.ws
     
   

   ---

2. Strip /etc/rc.d/rc3.ws to a bare minimum. It might be useful to create something like rc.local.ws but I'll leave that up to you. Network and nfsfs are already setup.The following have been already removed / updated by the automagic script:
   • network
   • nfsfs
   • nfs
   • rc.local

3.3 Exporting the appropriate file systems and setting up bootp

The server must ofcourse export the appropriate filesystems and asign the ip addresses to the clients.

Exporting the appropriate file systems

We need to export some dir's for the workstations so for the situation here at the university I would add the following to /etc/exports:

/ *.st.hhs.nl(ro,no_root_squash)
/home *.st.hhs.nl(rw,no_root_squash)
 

Ofcourse use the apropriate domain ;) and restart nfs by typing:

/etc/rc.d/init.d/nfs restart

Note for knfsd users: knfsd doesn't allow you to have multiple exports on one partition with different permissions. Also knfsd doesn't allow clients to go past partition boundaries for example if a client mounts / and /usr is a different partition it won't have access to /usr. Thus if you use knfsd, at least /home should be on a different partition, the server prepare script already puts /tftpboot in /home so that doesn't need a seperate partition. If you've got any other partitions your clients should have access to export them seperatly and add mount commands for them to /etc/rc.d/rc.ws.

Setting up bootp

1. If bootp isn't installed yet install it. It comes with RedHat.
2. Edit /etc/inetd.conf and uncomment the line beginning with bootps, if you want to use a bootprom uncomment tftp while your at it.
3. Restart inetd by typing:

   /etc/rc.d/init.d/inetd restart