How to set up diskless boot using Fedora Core 1

Under construction

This documents my experiences setting up a bunch of workstations to = boot Fedora diskless. These = are actually=20 nodes of a beowulf cluster, and that usage colours this description.=20

Right now these are just quick notes made as I went along. I'll write = this up=20 properly later.=20

Server setup

Standard server install for Fedora.=20

In the beowulf context set up two ethernet cards:=20

DHCP setup

Install dhcp-3.0pl2-6.16.rpm from distribution (or from http://isc.org/sw/dhcp/). This is = the ISC=20 DHCP server.=20

Configure /etc/dhcpd.conf. IP addresses I use are LAN addresses for = the=20 beowulf: 

option domain-name "my.domain.name";
option subnet-mask 255.255.255.0;
option broadcast-address 192.168.1.255;
ddns-update-style none;
option routers 192.168.1.254;

subnet 192.168.1.0 netmask 255.255.255.0 {
  range 192.168.1.101 192.168.1.150;
  next-server 192.168.1.249;
  filename "/pxelinux.0";
}

Test: # /usr/sbin/dhcpd -d -f.=20

Start: 

# chkconfig dhcpd on
# service dhcpd start

tftp setup

Install from fedora=20 core: # yum install tftp-server. This is standard tftp server HPA (by H. = Peter=20 Anvin).=20

Test: # /usr/sbin/in.tftpd -vvv -l -s /tftpboot=20

Configure in /etc/xinetd.d/tftp: 

server_args     =
        =3D -s /tftpboot

Start: # chkconfig tftp on (xinetd will start it when needed).=20

PXE setup

Latest syslinux from http://www.kernel.org/pub/linux/utils/boot/syslinux/. = Includes=20 pxelinux.0. Copy to /tftpboot.=20

Create /tftpboot/pxelinux.cfg/default: 

display =
pxeboot.msg
prompt 1

label linux
    kernel vmlinuz
    append ip=3Ddhcp root=3D/dev/nfs \
       nfsroot=3D192.168.1.249:/home/cns/local/beowulf/client-root =
vga=3D0x0f01

label local
    localboot 0

Client installation

We need to set up a network boot kernel, and a = Fedora = installation hosted=20 on the server.=20

Kernel compile and install

Block devices:=20

Networking options:=20

Copy bzImage to /tftpboot/linux.=20

Set root to nfsroot: 

# mknod /dev/nfsroot b 0 =
255
# rdev path/to/bzImage /dev/nfsroot

Install client linux distribution

Install to a spare partition. Boot another = partition and=20 use tar (to preserve links, char and block devices, etc) to copy it = across: 

spare# tar cvpf client_root.tar =
/mnt/spare_partition/
server# tar oxvpf client_root.tar /home/beowulf/.
NOTE: 'p' option is very important to preserve file permissions = (e.g.=20 setuids, etc).=20

Make mount points for virtual filesystems: 

# cd =
/home/beowulf/client_root
# mkdir proc
# rm -rf dev; mkdir dev

Need to write: /dev, /var, /tmp. We will use devfs, and use a ramdisk = for=20 /var. /tmp will be a symlink to /var/tmp.=20

Set up shadow passwords: 

# chroot =
/home/beowulf/client_root /usr/sbin/pwconv

In order to login make sure you install passwd and set root password. =

set hostname

We set this by lookup from /etc/hosts in rc.diskless. Normally = hostname=20 is set in /etc/rc.d/rc.sysinit to value in /etc/sysconfig/network, so = make sure=20 this file doesn't have a HOSTNAME entry, or it will overwrite our value. =

sshd startup

Make keys by hand ... they can be same for all client, since they're = not too=20 secret anyway (not on local filesystem). 

# cd =
/home/beowulf/client_root/etc/ssh
# ssh-keygen -t rsa1 -f ./ssh_host_key -C '' -N ''
# ssh-keygen -t rsa -f ./ssh_host_rsa_key -C '' -N ''
# ssh-keygen -t dsa -f ./ssh_host_dsa_key -C '' -N ''
# chmod 600 *_key
# chmod 644 *_key.pub

syslog setup

We want nodes to log to the master. In node etc/syslog.conf we just = need one=20 line: 

*.*                                           =
          @master
And on the master syslog must be started with the -r option, so = edit=20 :/etc/sysconfig/syslog to have: 
SYSLOGD_OPTIONS=3D"-r -m 0"

can't su?

Need to make sure setuid bit set on su: 

# chroot =
/home/beowulf/client_root  =20
# chmod u+s /bin/su
Ditto unix_chkpasswd.=20

why can't ssh?

May be that /etc/pam.d/sshd isn't there. It should be installed from=20 openssh-server RPM.=20

ypbind

In client /etc/sysconfig/network, add: 

NISDOMAIN=3Dypcns
and in client /etc/yp.conf: 
domain ypcns =
server coleman
and in client /etc/nsswitch.conf set all the right things (see = standard=20 CNS setup notes). In client /etc/{passed,shadow,group} add the usual = things.=20

rc.diskless = will also=20 need 

mkdir -p /var/yp/binding
then 
# chroot /home/client/root
# chkconfig --add ypbind
# chkconfig ypbind on

Installing and upgrading client packages

rpm will not work on clients, since /var is a ramdisk and doesn't = contain the=20 real RPM database.=20

Instead, simply chroot and do it on the master: 

# cp foo.rpm /home/client/root/.
# chroot /home/client/root
# rpm -ivh /foo.rpm
# exit