LDP/LDP/retired/KickStart-HOWTO.sgml

1185 lines
40 KiB
Plaintext

<!doctype linuxdoc system>
<article>
<title>RedHat Linux KickStart HOWTO
<author>Martin Hamilton <tt/&lt;martinh@gnu.org>/
<date>v0.2, 11 January 1999
<abstract>
Archived: 2004-05-17 at the request of the author by Emma Jane Hogbin
(Technical Review Coodinator). Additional information
can be obtained from: <url name="Red Hat Linux Customization Guide"
url="http://www.redhat.com/docs/manuals/linux/RHL-9-Manual/custom-guide/part-install-info.html">
<p>
This HOWTO briefly describes how to use the RedHat Linux
<em>KickStart</em> system to rapidly install large numbers of
identical Linux boxes. For advanced users, we describe how to modify
the KickStart installation procedure to do your own thing, and give a
quick guide to building RPM packages of your own.
</abstract>
<toc>
<sect>Copyright<p>
Copyright (c) 1998 Martin Hamilton, All rights reserved. This is free
documentware; you can redistribute it and/or modify it under the terms of
version 2 or later of the <url name="GNU General Public License"
url="http://www.gnu.org/copyleft/gpl.html">.<p>
<sect>Homepage<p>
If you got this document from a Linux HOWTO mirror site or a CD-ROM,
you might want to check back to the <url name="KickStart HOWTO home
page" url="http://wwwcache.ja.net/dev/kickstart/"> to see if there's a
newer version around.<p>
<sect>Introduction<p>
RedHat Linux version 5 comes with a a little-known (and until now, not
hugely documented) feature called <em>KickStart</em>. This lets you
automate most/all of a RedHat Linux installation, including:<p>
<itemize>
<item> Language selection
<item> Network configuration and distribution source selection
<item> Keyboard selection
<item> Boot loader installation (e.g. lilo)
<item> Disk partitioning and filesystem creation
<item> Mouse selection
<item> X Window system server configuration
<item> Timezone selection
<item> Selection of an (initial) root password
<item> Which packages to install
</itemize><p>
Eagle eyed RedHat users will probably have realised by now that these
are essentially the main steps involved in the manual installation of
a RedHat Linux system. KickStart makes it possible for you to script
the regular installation process, by putting the information you would
normally type at the keyboard into a configuration file.<p>
<em>But wait - there's more :-)</em><p>
Having finished the normal installation process, KickStart also lets
you specify a list of shell level commands which you would like to be
executed. This means that you can automatically install extra local
software not distributed as part of RedHat Linux (yes, there are even
more free software programs than the ones you get with the RedHat
distribution. Some can't be distributed by RedHat on legal grounds,
e.g. the <tt>ssh</tt> and <tt>PGP</tt> encryption systems) and carry
out any tidying up you may need to do in order to get a fully
operational system.<p>
<sect>Prerequisites<p>
There are two approaches to a KickStart install - one is to simply
copy your KickStart configuration file to a RedHat boot floppy. The
other is to use a regular boot floppy and get your KickStart config
file off the network.<p>
In both cases, you'll need:
<enum>
<item> Intel (i386) class machines - KickStart appears to only work
on these at the time of writing.
<item> KickStart config file - we'll talk about this in the next
section.
<item> RedHat boot disk - preferably from the <em>updates</em>
directory, to take advantage of any fixes/driver updates.
<item> DNS entries for the IP addresses you'll be using - optional,
but will stop the installation from prompting you for your
machine's domain name.
</enum><p>
If you want to fetch your config file over the network, you'll need to
export it via NFS - this is the only access method supported at the
moment. The config file lets you specify a different NFS server to
fetch the RedHat distribution itself from.<p>
You can configure a static IP address for your machine - e.g. a special
one reserved for KickStart installations. Alternatively, if you don't
want to hard code an IP address in the config file you can tell KickStart
to use a BOOTP/DHCP server to fetch this. Some servers will allocate new
addresses in a given range automatically, e.g. <url name="the CMU BOOTP
server with dynamic addressing extensions"
url="ftp://ftp.ntplx.net/pub/networking/bootp">.<p>
More information on NFS and BOOTP/DHCP is in Appendix A.<p>
<sect>Setting up a boot floppy<p>
Essentially, all you have to do is copy your KickStart config file
to <em>/ks.cfg</em> on the RedHat boot floppy, e.g.<p>
<tscreen><verb>
mcopy ks.cfg a:
</verb></tscreen><p>
However - the RedHat boot floppy is pretty full, and you may find that
you have to delete some of the other files to make room for the
KickStart config file. I was able to make enough room for mine by
deleting the various message files normally displayed by the
<tt>SYSLINUX</tt> boot loader, e.g.<p>
<tscreen><verb>
mdel a:\*.msg
</verb></tscreen><p>
Another approach would be to throw away the drivers for some of the
hardware you don't have - see the section on modifying the boot
floppy below.<p>
You may also want to edit <em>syslinux.cfg</em>, the <tt>SYSLINUX</tt>
config file. This also lives in the top level directory of the RedHat
boot floppy. For instance, the following <em>syslinux.cfg</em> will
cause KickStart mode to be entered into automatically as the machine
boots up, without the normal delay:<p>
<tscreen><verb>
default ks
prompt 0
label ks
kernel vmlinuz
append ks=floppy initrd=initrd.img
</verb></tscreen><p>
Note that you almost probably want to base your boot and supplementary
floppies on the most recent disk images available in the
<em>updates/i386</em> on your local RedHat mirror site. Older images
may be buggy or have driver support for less hardware.<p>
<sect>The KickStart config file<p>
There are three main sections to the config file:<p>
<enum>
<item> System info, e.g. disk partitioning and network config
<item> RedHat packages to install
<item> Post-installation shell commands to run
</enum<p>
There are some other possibilities which we won't talk about here, but
<bf>might</bf> work. For more information check out the sample
KickStart config in <em>misc/src/install/ks.samp</em> and
<em>doc/README.ks</em> under the top level <em>i386</em> RedHat
distribution directory on your CD-ROM or local RedHat mirror site.<p>
<sect1>System info<p>
The available directives which I've been using are:<p>
<descrip>
<tag/lang/Language configuration, e.g. for English
<tscreen><verb>
lang en
</verb></tscreen>
<tag/network/Network configuration, e.g. to use BOOTP/DHCP
<tscreen><verb>
network --bootp
</verb></tscreen>
<tag/nfs/NFS server and directory to install from, e.g.
<tscreen><verb>
nfs --server chicken.swedish-chef.org /mnt/cdrom
</verb></tscreen>
to use the NFS server <em>chicken.swedish-chef.org</em> and try to
mount the RedHat distribution from the directory <em>/mnt/cdrom</em>.
<tag/keyboard/Select keyboard type, e.g. for UK keyboards
<tscreen><verb>
keyboard uk
</verb></tscreen>
<tag/zerombr/Clear the Master Boot Record - removes any existing
operating system boot loader from your disk
<tag/clearpart/Clear existing partitions - e.g. to remove all existing
disk partitions prior to installation
<tscreen><verb>
clearpart --all
</verb></tscreen>
<tag/part/Partition the disk, e.g. to make a root filesystem of
500MB
<tscreen><verb>
part / --size 500
</verb></tscreen>
<tag/install/Make a fresh installation of RedHat Linux.
<tag/mouse/Set the mouse being used, e.g. for a PS/2 or compatible
"bus mouse"
<tscreen><verb>
mouse ps/2
</verb></tscreen>
<tag/timezone/Set the timezone, e.g. for local time in the UK
<tscreen><verb>
timezone --utc Europe/London
</verb></tscreen>
<tag/rootpw/Set the initial root password, based on a previously
derived encrypted password
<tscreen><verb>
rootpw --iscrypted XaacoeGPmf/A.
</verb></tscreen>
<tag/lilo/Install the LILO boot loader, e.g. in the Master Boot Record
<tscreen><verb>
lilo --location mbr
</verb></tscreen>
<tag/%packages/Packages to install - see below.
<tag/%post/Post-installation shell commands - see below.
</descrip><p>
Note that the directory where KickStart is looking for the RedHat
distribution should have a subdirectory <em>RedHat</em>, which
contains the RedHat distribution tree for the platform in question.
In the above example, we should see something like the following files
and directories:<p>
<tscreen><verb>
/mnt/cdrom/RedHat
/mnt/cdrom/RedHat/base
/mnt/cdrom/RedHat/contents
/mnt/cdrom/RedHat/i386
/mnt/cdrom/RedHat/instimage
/mnt/cdrom/RedHat/RPMS
/mnt/cdrom/RPM-PGP-KEY
</verb></tscreen><p>
If you're installing off a CD-ROM rather than off the network, the
contents should look something like this:<p>
<tscreen><verb>
RedHat
RedHat/base
RedHat/contents
RedHat/i386
RedHat/instimage
RedHat/RPMS
RPM-PGP-KEY
</verb></tscreen><p>
If you have the RedHat distribution for multiple architectures (e.g.
on an NFS server - they're too big to fit more than one architecture's
version onto a single CD-ROM), you'll notice that each distribution
has the same files and directories under a subdirectory, e.g.<p>
<tscreen><verb>
alpha/RPM-PGP-KEY
i386/RPM-PGP-KEY
sparc/RPM-PGP-KEY
</verb></tscreen><p>
There should be a file <tt>architecture</tt>/Redhat/<tt>architecture</tt>,
e.g. <em>i386/Redhat/i386</em>.<p>
If you want to create your own encrypted passwords, it's very easy
using Perl, e.g.<p>
<tscreen><verb>
% perl -e 'print crypt("schmurrdegurr", "Xa") . "\n";'p
</verb></tscreen><p>
Other options (or mooted options), which I've not tried:<p>
<descrip>
<tag/cdrom/Install off CD-ROM rather than network.
<tag/device/Explicitly declare device details, e.g.
<tscreen><verb>
device ethernet 3c509 --opts "io=0x330, irq=7"
</verb></tscreen>
Alternative values of <tt>device</tt> include <tt>scsi</tt> for SCSI
controllers and <tt>cdrom</tt> for proprietary CD-ROM drives.
<tag/upgrade/Upgrade an existing installation rather than make
a fresh installation.
<tag/xconfig/Configure X Window server, graphics card and monitor.
e.g.
<tscreen><verb>
xconfig --server "Mach64" --monitor "tatung cm14uhe"
</verb></tscreen>
</descrip><p>
I've not delved too deeply into this last one, because I'm not ever
planning to run X on the console of any of my KickStarted machines.
I'm told that running <tt>xconfig</tt> within KickStart itself is a
bit flaky, but the same functionality is also available from the
command line via <tt>Xconfigurator</tt> - so you might be best off
leaving this to the post-installation script.<p>
Here's how this first part of a KickStart config file looks when we
put all the bits together:<p>
<tscreen><verb>
lang en
network --static --ip 198.168.254.253 --netmask 255.255.255.0
--gateway 198.168.254.1 --nameserver 198.168.254.2
nfs --server chicken.swedish-chef.org /mnt/cdrom
keyboard uk
zerombr yes
clearpart --all
part / --size 500
part swap --size 120
install
mouse ps/2
timezone --utc Europe/London
rootpw --iscrypted XaacoeGPmf/A.
lilo --location mbr
</verb></tscreen><p>
Note that some of the RedHat documentation refers to an invocation
of the <tt>network</tt> directive which doesn't actually work in
practice: <tt>network --option</tt>. The correct invocation is to
put <tt>network</tt> followed by <tt>--static</tt>, <tt>--bootp</tt>
or <tt>--dhcp</tt>. Be aware that the BOOTP and DHCP options are
different - to the extent that they even use different code.<p>
You can add the <tt>--grow</tt> parameter to a <tt>part</tt>
directive to indicate that it's OK to grow the partition beyond
the size you specify. It probably only makes sense to have one
partition tagged with <tt>--grow</tt>.<p>
<sect1>Packages to install<p>
The start of the packages section of the KickStart config file is
indicated by the presence of a <tt>%packages</tt> directive on a line
of its own. This is followed by one or both of two types of package
specifier - individual packages may be installed by giving the name of
their RPM (excluding the version and platform information), and groups
of packages may be installed by giving their group name.<p>
Here's a sample <tt>packages</tt> section for a KickStart config file:
<tscreen><verb>
%packages
@ Base
netkit-base
bind-utils
ncftp
rdate
tcp_wrappers
traceroute
cmu-snmp
</verb></tscreen><p>
So, what are these groups ? Well, there are a number of groups
defined by default in a file called <em>base/comps</em> under the
RedHat distribution's top level directory. Here are the ones which
were current at the time of writing:<p>
<itemize>
<item> Base
<item> Printer Support
<item> X Window System
<item> Mail/WWW/News Tools
<item> DOS/Windows Connectivity
<item> File Managers
<item> Graphics Manipulation
<item> X Games
<item> Console Games
<item> X multimedia support
<item> Console Multimedia
<item> Print Server
<item> Networked Workstation
<item> Dialup Workstation
<item> News Server
<item> NFS Server
<item> SMB (Samba) Connectivity
<item> IPX/Netware(tm) Connectivity
<item> Anonymous FTP/Gopher Server
<item> Web Server
<item> DNS Name Server
<item> Postgres (SQL) Server
<item> Network Management Workstation
<item> TeX Document Formatting
<item> Emacs
<item> Emacs with X windows
<item> C Development
<item> Development Libraries
<item> C++ Development
<item> X Development
<item> Extra Documentation
</itemize><p>
You'll notice that they correspond to the various configurations which
you're prompted for during a manual installation. Note that some of
the packages in a given package group are duplicated in other groups,
and that you can install multiple groups of packages without this
causing problems. Each group's entry in the <em>comps</em> listing
looks similar to this:<p>
<tscreen><verb>
0 Extra Documentation
sag
lpg
howto
faq
man-pages
end
</verb></tscreen><p>
It seems that groups with a <em>1</em> next to their name (the first
line above) are selected for installation by default. You can customise
the Linux installation process even further by creating your own groups
or redefine existing ones by editing this file.<p>
<sect1>Post-installation shell commands<p>
This is probably the best feature of all, and something which there is
no direct equivalent to in the manual installation process. What we
can do here is specify a sequence of shell level commands which should
be executed after the main installation (disk partitioning, package
installation, and so on) is complete.<p>
The beginning of this section is signified by the <tt>%post</tt>
directive in the KickStart config file. In what follows you can take
advantage of all of the utilities which have been installed on your
newly built Linux system, e.g.<p>
<tscreen><verb>
%post
ln -s /etc/rc.d/init.d /etc/init.d
ln -s /etc/rc.d/rc.local /etc/rc.local
ln -s /usr/bin/md5sum /usr/bin/md5
ln -s /usr/bin/perl /usr/local/bin/perl
chmod ug-s /bin/linuxconf
mkdir /var/tmp/tmp
perl -spi -e 's!image=/boot/vmlinuz-.*!image=/boot/vmlinuz!' /etc/lilo.conf
rm /etc/rc.d/rc*.d/*sendmail
</verb></tscreen><p>
You can also use I/O redirection and here documents:<p>
<tscreen><verb>
cat <<EOF >>/etc/passwd
squid:*:102:3500:Squid Proxy:/usr/squid:/bin/bash
EOF
cat <<EOF >>/etc/group
cache:x:3500:
EOF
</verb></tscreen><p>
Modify the run-time startup scripts:<p>
<tscreen><verb>
cat <<EOF >>/etc/rc.local
echo 8192 > /proc/sys/kernel/file-max
echo 32768 > /proc/sys/kernel/inode-max
[ -x /usr/sbin/sshd ] && /usr/sbin/sshd
[ -x /usr/sbin/cfd ] && /usr/sbin/cfd
EOF
</verb></tscreen><p>
Set up <em>crontab</em> entries:<p>
<tscreen><verb>
cat <<EOF >/tmp/crontab.root
# Keep the time up to date
0,15,30,45 * * * * /usr/sbin/ntpdate -s eggtimer 2>&1 >/dev/null
# Recycle Exim log files
1 0 * * * /usr/exim/bin/exicyclog
# Flush the Exim queue
0,15,30,45 * * * * /usr/exim/bin/exim -q
EOF
crontab /tmp/crontab.root
rm /tmp/crontab.root
</verb></tscreen><p>
And even install other RPMs which you made yourself:<p>
<tscreen><verb>
rpm -i ftp://chicken.swedish-chef.org/rpms/squid.rpm
rpm -i ftp://chicken.swedish-chef.org/rpms/ssh.rpm
rpm -i ftp://chicken.swedish-chef.org/rpms/exim.rpm
rpm -i ftp://chicken.swedish-chef.org/rpms/cfengine.rpm
rpm -i ftp://chicken.swedish-chef.org/rpms/linux.rpm
ssh-keygen -b 1024 -f /etc/ssh_host_key -N ""
depmod -a
</verb></tscreen><p>
Note that you can achieve the same effect by making your own RPMs
containing the commands you want executed - see below for more
information. Give them a carefully chosen name and you can force them
to be installed first (e.g. name starts with 'aaa') or last (e.g.
name starts with 'zzz').<p>
Be aware that a less painful way of doing root crontab entries is to
create them as files in one or more of the directories
<em>/etc/cron.hourly</em>, <em>/etc/cron.daily</em>,
<em>/etc/cron.weekly</em> and <em>/etc/cron.monthly</em>.<p>
More information about making your own RPMs is available in Appendix B.<p>
<sect>Installation itself<p>
Boot the to-be-installed machine off your RedHat boot floppy as usual,
but instead of pressing <tt>RETURN</tt> at the <tt>SYSLINUX</tt>
prompt, type <tt>linux ks</tt>.<p>
If you're lucky, this will be all you have to type!<p>
If you customised your RedHat boot floppy as outlined above, you won't
even need to do this bit :-)<p>
Since we're really just automating the normal steps involved in a
RedHat installation, the normal dialogs may appear if/when KickStart
gets confused about what to do next. The most likely case is that
your network interface won't be detected automatically, and you'll be
prompted for its IRQ and I/O address space. KickStart tends to need
help for ISA bus cards, but detects PCI bus cards automatically.<p>
You can keep an eye on what KickStart is doing by by switching virtual
consoles as usual:<p>
<itemize>
<item> Alt-F1 - installation dialog
<item> Alt-F2 - shell prompt
<item> Alt-F3 - install log (messages from install program)
<item> Alt-F4 - system log (messages from kernel, etc.)
<item> Alt-F5 - other messages
</itemize><p>
<sect>Mounting the boot/supp disks<p>
The RedHat boot disk <em>boot.img</em> is in MS-DOS format, using the
<tt>SYSLINUX</tt> program to boot up. The supplementary disk
<em>supp.img</em> is a Linux <tt>ext2</tt> filesystem. If you have
support for the loopback filesystem in your Linux kernel, you can mount
both of these files in your filesystem and hack at them:<p>
<tscreen><verb>
# mkdir -p /mnt/boot /mnt/supp
# mount -o loop -t msdos boot.img /mnt/boot
# mount -o loop supp.img /mnt/supp
</verb></tscreen><p>
Now you should be able to see and manipulate the files on the boot and
supplementary disk under <em>/mnt/boot</em> and <em>/mnt/supp</em>
respectively. Phew! Note that older versions of <tt>mount</tt> may
not be able to handle the <tt>-o loop</tt> option. In these cases
you'll need to explicitly use <tt>losetup</tt> to configure the
loopback device for each file, e.g.<p>
<tscreen><verb>
# losetup /dev/loop0 boot.img
# mount -t msdos /dev/loop0 /mnt/boot
</verb></tscreen><p>
You might also need to explicitly use the <tt>-t ext2</tt> option when
mounting an <tt>ext2</tt> filesystem like the one on the supplementary disk.
But, it looks like people with modern Linux distributions shouldn't have to
worry about this.<p>
Of course, if you don't want to mess around too much, you can cut a
corner and manipulate actual floppy disks rather than these floppy
disk images. If time is important, you'll probably prefer to use the
loopback devices, since you can hack around with the disk images
without incurring the latency associated with a genuine floppy disk
read/write.<p>
<sect>Modifying the RedHat installer<p>
If you want to mess around with the installation procedure itself, the
source code can be found on the RedHat CD-ROM or your local RedHat
mirror site. It's in <em>misc/src/install</em> under the
<em>i386</em> distribution top level directory.<p>
If you examine the RedHat boot disk you'll see that, in addition to
the Linux kernel <em>vmlinuz</em>, there's a large file
<em>initrd.img</em>:<p>
<tscreen><verb>
- -rwxr-xr-x 1 root root 559 May 11 15:48 boot.msg
- -rwxr-xr-x 1 root root 668 May 11 15:48 expert.msg
- -rwxr-xr-x 1 root root 986 May 11 15:48 general.msg
- -rwxr-xr-x 1 root root 968842 May 11 15:48 initrd.img
- -rwxr-xr-x 1 root root 1120 May 11 15:48 kickit.msg
- -r-xr-xr-x 1 root root 5352 May 11 15:48 ldlinux.sys
- -rwxr-xr-x 1 root root 875 May 11 15:48 param.msg
- -rwxr-xr-x 1 root root 1239 May 11 15:48 rescue.msg
- -rwxr-xr-x 1 root root 402 May 11 15:48 syslinux.cfg
- -rwxr-xr-x 1 root root 444602 May 11 15:48 vmlinuz
</verb></tscreen><p>
You guessed it, this is another <tt>ext2</tt> filesystem saved as a file
- - but with a twist. It's actually compressed as well. You can uncompress
it and then mount the result, e.g.<p>
<tscreen><verb>
# gzip -dc /mnt/boot/initrd.img >/tmp/initrd.ext2
# mkdir /mnt/initrd
# mount -o loop /tmp/initrd.ext2 /mnt/initrd
</verb></tscreen><p>
Probably the most important part of this filesystem is the collection
of loadable kernel modules which are included with the boot disk. If
you need to merge in a new version of a driver, you'll need to either
replace <em>vmlinuz</em> with a new kernel which has this statically
linked, or replace it in the modules collection. What's more, you may
need to throw other modules away to make room.<p>
The modules collection is the file <em>modules/modules.cgz</em>.
Wondering what that might be ? It's actually a compressed
<tt>cpio</tt> archive, believe it or not! And you thought nobody used
<tt>cpio</tt> any more... Actually RPM itself uses <tt>cpio</tt>
internally, too. Here's how to hack around with it:<p>
<tscreen><verb>
# gzip -dc /mnt/initrd/modules/modules.cgz >/tmp/modules.cpio
# cpio -itv <modules.cpio >modules.listing
# mkdir modules
# cpio -idumv <../modules.cpio
</verb></tscreen><p>
I don't believe that there is currently a way under Linux (at least in
mainstream distributions) to transparently access compressed
filesystems. Let me know if you know better!<p>
If you change anything, remember to:<p>
<enum>
<item> Use <tt>cpio</tt> to recreate the archive. How to do this is
left as an exercise for the reader...
<item> Use <tt>gzip</tt> to compress the resulting archive.
<item> Copy it to <em>/mnt/initrd</em>, or wherever you put the
uncompressed <em>initrd.img</em> archive.
<item> Unmount <em>/mnt/initrd</em> (or whatever you called it).
<item> Compress the new <em>initrd.img</em> using <tt>gzip</tt>
again.
<item> Copy the resulting archive onto the boot disk image -
<em>/mnt/boot/initrd.img</em> in our example.
<item> Unmount the boot disk image, e.g. <em>/mnt/boot</em>.
</enum><p>
Finally, you can now create new boot floppies using this modified boot
disk setup, e.g.<p>
<tscreen><verb>
# cat boot.img >/dev/fd0
</verb></tscreen><p>
<sect>FAQs/Wish list<p>
<bf>Q:</bf> After KickStart installation, my machine won't boot up.
The BIOS complains with a message like <tt>Missing operating system</tt>.<p>
<bf>A:</bf> Sounds like the partition with the root filesystem on isn't
bootable. Use <tt>fdisk</tt> to toggle its bootable status.<p>
<bf>Q:</bf> After the floppy boots, I get the message:
<tt>Error opening files for kickstart copy: File exists</tt>.<p>
<bf>A:</bf> Use a more recent version of <em>boot.img</em> and
<em>supp.img</em> - look in the <em>updates</em> directory of your
local RedHat mirror site. There was a bug in some older versions
of these for RedHat 5.1.<p>
<bf>Q:</bf> Can you have all outstanding patches (update RPMs) applied
automatically too ? How ?<p>
<bf>A1:</bf> Copy the RPMs you want installing to the RPMS directory
from which the installation is going to take place, get rid of the
older RPMs, and update the file <em>RedHat/base/hdlist</em> with the
new RPM details. See Appendix C for a script from Eric Doutreleau to do
this for you. If you do this yourself, remember to run <em>genhdlist</em>
afterwards!<p>
<bf>A2:</bf> Try this Perl script: <url name="patchup"
url="http://wwwcache.ja.net/dev/patchup/">. This compares the RPMs
your system has installed with those in a nominated directory and
reports on the ones it thinks need updating. It can even install them
for you if you trust it to.<p>
<bf>A3:</bf> <url name="rpm2hml" url="http://rufus.w3.org/linux/rpm2html/">
has a much more powerful (12MB of C code vs. a page of Perl!) version of A2.<p>
<bf>Q:</bf> A single config file on the install server for all of the clients,
perhaps as a fallback after trying <em>IPADDR-kickstart</em> ?<p>
<bf>A1:</bf> Use the BOOTP/DHCP 'boot file' parameter <em>bf</em> to set
the filename.<p>
<bf>A2:</bf> Add a a record <tt>bf=/kickstart/ks.cfg</tt> to the relevant
entry in <em>/etc/bootptab</em>.
<bf>Q:</bf> More flexibility when things go wrong - e.g. prompt for alternate
locations if distribution not found on CD-ROM.<p>
<bf>A:</bf> ?<p>
<bf>Q:</bf> Explicit exclusion of packages - e.g. everything apart from
<em>sendmail</em>.<p>
<bf>A:</bf> Rebuild the <bf>BASE</bf> package without sendmail.<p>
<bf>Q:</bf> Choose which services are started automatically on boot-up by the
run-level scripts under <em>/etc/rc.d/</em>.<p>
<bf>A:</bf> The <em>chkconfig</em> utility lets you configure which
services are run automatically on boot-up. You can run this in your
post-installation script section, e.g. to run <em>ypbind</em> in run
levels 3, 4 and 5:<p>
<tscreen><verb>
chkconfig --level 345 ypbind on
</verb></tscreen><p>
and it will start the ypbind level on the 345 level.
<bf>Q:</bf> When executing the shell commands in the <tt>%post</tt> section,
bring any output up in another virtual console rather than overwriting the
main screen. <em>Could be done in the shell commands section using
<tt>open</tt>?</em>.<p>
<bf>A:</bf> No problem - do something like this:<p>
<tscreen><verb>
exec >/dev/tty5
</verb></tscreen><p>
<bf>Q:</bf> Does the filesystem creation code check for bad blocks ?<p>
<bf>A:</bf> If you switch to the virtual console where the filesystem
creation output is being displayed, you won't see any mention of the
'read-only' test being performed. Looks like the answer is no.<p>
<bf>Q:</bf> Can I arrange things so some of my machines are configured
differently from others ?<p>
<bf>A:</bf> You could move the host dependent stuff into the
scripted section of the KickStart config - e.g. only install a given
RPM if on a given machine. It would be useful to have a conditional
installation feature in the packages section of the config file,
e.g. switching on architecture, or hostname/domain name/IP address.<p>
<bf>Q:</bf> Are there any changes between RedHat 5.1 and 5.2 ?<p>
<bf>A1:</bf> Lots of changes in the installer, but mostly bug fixes
or cosmetic improvements. No impact on KickStart as far as I can
tell - from a <em>diff -rcs</em> of the two <em>misc/src/install</em>
directories.<p>
<bf>A2:</bf> RH5.2 now apparently includes the automatic IP allocation/DHCP
patches to <tt>bootpd</tt>, but they have left out the documentation
which tells you how to use it.
<bf>Q:</bf> (How) can you clear a specific partition or partitions ?
e.g. to leave <em>/home</em> but zap <em>/</em>.<p>
<bf>A:</bf> You can't - yet!<p>
<bf>Q:</bf> Can you arrange to have your partitions created across
multiple drives ? e.g. <em>/</em> on <tt>sda</tt> and <em>/home</em>
on <tt>sdb</tt>.<p>
<bf>A:</bf> Don't think so - looks like you only get access to the
first drive from the partitioning tool.<p>
<bf>Q:</bf> Is it possible to specify existing partitions to be
included in the mount table, or is it only possible to specify the
creation of new partitions that will then be included?<p>
<bf>A:</bf> ?<p>
<bf>Q:</bf> After running <tt>mkkickstart</tt>, where is the file
it creates?<p>
<bf>A:</bf> It doesn't create a file - it dumps the KickStart
config to <tt>stdout</tt>.<p>
<bf>Q:</bf> In virtual console 4 (Alt-F4) I get <tt>Unable to load NLS
charset cp437(nls_cp437)</tt>. What does this mean ? Should I be
worried ?<p>
<bf>A:</bf> Sounds like you're trying to mount a CD-ROM burned with
the Joliet (Unicode extensions to ISO 9660. In theory the filenames
on the CD-ROM might get munched and not make it through to Linux
correctly. In practice it doesn't seem to cause any problems -
could be a spurious dependency ?<p>
<bf>Q:</bf> Why am i getting the X Window System installed ? I didn't
put it in my list of packages.<p>
<bf>A:</bf> The <tt>XFree86-VGA16</tt> RPM is a 'base' component, and
as such always gets installed - unless you change the definition of
the base class.<p>
<bf>Q:</bf> In my post-installation script, can I use the packages
which have been installed by now to do funky things not possible with
the limited tools on the floppies ?<p>
<bf>A:</bf> Yep - e.g. if you chose to install Perl when you put your
KickStart config together, almost anything is possible in about five
lines :-)<p>
<sect>Credits<p>
Thanks to Eric Doutreleau for the info about <em>chkconfig</em>, the
<tt>SYSLINUX</tt> config file hack, and the Perl script for updating
your distribution server's RPMs. Thanks to Robert Kaminsky for
extensive investigations. Thanks to Piete Brooks, Flavia Regina
Munhoz, Tom Toffoli, Bob Robbins, Charlie Brady and Ragen Herrington,
for their comments and questions.<p>
<sect>Appendix A - Configuring BOOTP/DHCP and NFS<p>
If you're wondering what on earth this BOOTP and DHCP stuff is, more
information is available at <url name="the DHCP WWW site"
url="http://www.dhcp.org/">. NFS is documented separately in detail
in the NFS HOWTO, and there's a DHCP mini-HOWTO too. I've tried to
provide enough details here to help you get started, whilst not
treating the topics in depth - let me know if you think this is
overkill.<p>
In the BOOTP/DHCP + NFS configuration we're discussing, the KickStart
config file should be NFS mountable by the machine being installed
from <em>/kickstart/IPADDR-kickstart</em> on the BOOTP/DHCP server,
where <em>IPADDR</em> is the IP address of the new machine, e.g.
<em>/kickstart/198.168.254.254-kickstart</em> for the machine
<em>198.168.254.254</em>.<p>
You should be able to override this location by returning
the <tt>bf</tt> parameter (boot file) in your BOOTP/DHCP response. It
may even be possible to have this NFS mounted off another machine
entirely.<p>
To NFS export some directories from an existing Linux box, create the
file <em>/etc/exports</em> with contents something like:<p>
<tscreen><verb>
/kickstart *.swedish-chef.org(ro,no_root_squash)
/mnt/cdrom *.swedish-chef.org(ro,no_root_squash)
</verb></tscreen><p>
Note that if you didn't register the IP addresses you're going to be
using in the DNS, you may be told to get lost by the NFS server and/or
the RPC portmapper. In this you can probably get away with putting
IP address/netmask pairs in the config files, e.g.<p>
<tscreen><verb>
/kickstart 198.168.254.0/255.255.255.0(ro,no_root_squash)
</verb></tscreen><p>
and in <em>/etc/hosts.allow</em>:<p>
<tscreen><verb>
ALL: 194.82.103.0/255.255.255.0: ALLOW
</verb></tscreen><p>
This is because most Linux distributions use TCP wrappers to do
access control for some or all of the NFS related daemons. Be aware
that the <em>/etc/exports</em> syntax tends to be different on other
Unix variants - the NFS servers bundled with Linux distributions
tend to offer a much wider range of options than the ones shipped
with other versions of Unix.
Be aware that if you include a root password in your KickStart config
file, or NFS export directories containing sensitive information, you
should take care to expose this information to as few people as
possible. This can be done by making the NFS export permissions as
fine grained as you can, e.g. by specifying a particular host or
subnet to export to rather than a whole domain. If you keep a special
IP address free for KickStart installations, everything's nice and
simple, but you'll have to change it later - or reconfigure the
machine to get its IP address via BOOTP/DHCP.<p>
Most NFS servers require you to tell <tt>mountd</tt> and <tt>nfsd</tt>
(on some versions of Unix they're prefixed with a <tt>rpc.</tt>) that
the <em>/etc/exports</em> file has changed - usually by sending a
<tt>SIGHUP</tt>. There's often a program or script called
<tt>exportfs</tt>, which will do this for you, e.g.<p>
<tscreen><verb>
# exportfs -a
</verb></tscreen><p>
If you didn't have NFS up and running when this machine was booted,
the directories may not be exported automatically. Try rebooting, or
running the following programs as root:<p>
<tscreen><verb>
# portmap
# rpc.nfsd
# rpc.mountd
</verb></tscreen><p>
As noted, on some systems the <tt>rpc.</tt> prefix isn't used. In
most modern Unix distributions, these programs can be found in the
<em>/usr/sbin</em> or <em>/usr/libexec</em> directories. These might
not be in your path already, e.g. if you used <tt>su</tt> to become
<em>root</em>. The <tt>portmap</tt> program is also sometimes called
<tt>rpcbind</tt>, e.g. on Solaris, some versions of <tt>nfsd</tt>
require a command line argument specifying the number of instances of
the server to run, and you may find you also need to run another
daemon called <tt>biod</tt>. The above should suffice on most (all?)
Linux systems.<p>
If you're using the CMU BOOTP server with DHCP and dynamic addressing
extensions referred to earlier, a sample <em>/etc/bootptab</em> entry
(<em>/etc/bootptab</em> is the normal location of the BOOTP/DHCP
configuration file) would look something like this:<p>
<tscreen><verb>
.dynamic-1:ip=198.168.254.128:T254=0x30:T250="ds=198.168.254.2:
dn=swedish-chef.org:sm=255.255.255.0:gw=198.168.254.1:
dl=0xFFFFFFFF":
</verb></tscreen><p>
(wrapped for clarity)<p>
This says to allocate IP addresses dynamically on encountering new
machines, starting at <em>198.168.254.128</em> and continuing for the
next 48 (the hexadecimal value <em>30</em>) addresses. Each client
will be passed back the value of <em>T250</em>. In this case that
sets:<p>
<itemize>
<item> the DNS server <tt>ds</tt> to <em>198.168.254.2</em>
<item> the domain name <tt>dn</tt> to <em>swedish-chef.org</em>
<item> the subnet mask <tt>sm</tt> to <em>255.255.255.0</em>
<item> the default gateway <tt>gw</tt> to <em>198.168.254.1</em>
<item> the lease length <tt>dl</tt> (how long the address is
valid for) to "forever"
</itemize><p>
There seem to be a number of other versions of this server kicking
around which do not support dynamic addressing. For these, you would
have to list the hardware (typically Ethernet MAC) address of each
to-be-installed machine in <em>/etc/bootptab</em>, and the entries
would look something like this:<p>
<tscreen><verb>
bork.swedish-chef.org:ip=198.168.254.128:ha=0000E8188E56:
ds=198.168.254.2:dn=swedish-chef.org:sm=255.255.255.0:
gw=198.168.254.1:dl=0xFFFFFFFF":
</verb></tscreen><p>
(wrapped for clarity)<p>
Note that the parameter <tt>ha</tt> corresponds to the hardware
address of the machine being installed.<p>
<sect>Appendix B - Making your own RPMs<p>
The RPM package format is already very well documented, particularly
in the book <em>Maximum RPM</em> by Ed Bailey, which you can download
from the <url url="http://www.rpm.org/" name="RPM WWW site"> - also
available from all good book stores! This is just a couple of quick
hints for people in a hurry.<p>
RPM packages are built from a <em>spec</em> file. This consists (in a
similar fashion to the KickStart config file) of a recipe of steps
that need to be taken in order to build the package - it's expected
that you'll have to build it from source, potentially for multiple
platforms, and may need to apply patches before compiling. Once built
and installed, a binary RPM will be created from the files and
directories you specify as being associated with the package. It's
important to note that RPM has no idea of which files and directories
are related to a given package - you have to tell it.
Here's a sample specification for a custom RPM of the <url name="Squid
WWW cache server" url="http://squid.nlanr.net/">:<p>
<tscreen><verb>
Summary: Squid Web Cache server
Name: squid
Version: 1.NOVM.22
Release: 1
Copyright: GPL/Harvest
Group: Networking/Daemons
Source: squid-1.NOVM.22-src.tar.gz
Patch: retry-1.NOVM.20.patch
%description
This is just a first attempt to package up the Squid Web Cache for easy
installation on our RedHat Linux servers
%prep
%setup
%build
configure --prefix=/usr/squid
perl -spi -e 's!#( -DALLOW_HOSTNAME_UNDERSCORES)!$1!' src/Makefile
make
%install
make install
%files
/usr/squid
</verb></tscreen><p>
Here's how to build this RPM:<p>
<tscreen><verb>
% mkdir -p SOURCES BUILD SRPMS RPMS/i386
% cp ~/squid-1.NOVM.22-src.tar.gz SOURCES
% cp ~/retry-1.NOVM.20.patch SOURCES
% rpm -ba squid-1.NOVM.22+retry-1.spec
</verb></tscreen><p>
This will automatically create a subdirectory under the <em>BUILD</em>
directory, into which it'll unpack the source code and then apply the
patch (there are a number of options available for patching - check
the book for details). Now, RPM will automatically build the package
by running <tt>configure</tt> and then <tt>make</tt>, install it using
<tt>make install</tt>, and take a snapshot of the files under
<em>/usr/squid</em>. It's the latter which will form the binary RPM
of the Squid software.<p>
Note that we can insert arbitrary shell commands into the unpacking,
building and installing processes, e.g. the call to <tt>perl</tt>
which tweaks one of Squid's compile-time parameters.<p>
The final binary RPM will be left under the <em>RPMS</em> directory in
the platform specific subdirectory <em>i386</em>. In this case it
will be called <em>squid-1.NOVM.22-1.i386.rpm</em>. Note that the
filename is created by concatenating the values of the following
parameters from the spec file: <tt>Name</tt>, <tt>Version</tt> and
<tt>Release</tt> - plus the hardware platform in question,
<em>i386</em> in this case. Try to bear this in mind when creating
your own RPMs, to avoid giving them overly long or painful names!<p>
It's also worth bearing in mind that you can build RPMs without having
to rebuild the whole software package, e.g.<p>
<tscreen><verb>
Summary: Linux 2.0.36 kernel + filehandle patch + serial console patch
Name: linux
Version: 2.0.36+filehandle+serial_console
Release: 1
Copyright: GPL
Group: Base/Kernel
Source: linux-2.0.36+filehandle+serial_console.tar.gz
%description
This is just a first attempt to package up the Linux kernel with patches
for installation on our RedHat Linux servers
%prep
echo
%setup
echo
%build
echo
%install
echo
%post
/sbin/lilo
%files
/lib/modules/2.0.36
/boot/vmlinuz
</verb></tscreen><p>
In this case we simply create an RPM based on the
<em>/boot/vmlinuz</em> file and the contents of the directory
<em>/lib/modules/2.0.36</em>, and execute <em>/sbin/lilo</em> after
the package has been installed on a target machine. Let me know if
you know much neater way of writing the spec file than this.<p>
<sect>Appendix C - Munging your own RPMs into the distribution<p>
Here is Eric's script for munging updated RPMs into the RedHat
distribution area:
<tscreen><verb>
#!/usr/bin/perl
#
$redhatdir="/cdrom/i386";
$rpmdir="/cdrom/i386/RedHat/RPMS/";
$updatedir="/cdrom/updates/";
@OTHERDIR=($updatedir);
foreach $dir (@OTHERDIR)
{
print "update for $dir\n";
system(" find $dir -name \"*.rpm\" -exec cp {} $rpmdir \\; ");
}
chdir($contribdir) || die "peux pas aller dans $contribdir $!\n";
system("chmod -R 755 $redhatdir");
chdir($rpmdir) || die "problem to go in $rpmdir $!\n";
#
# remove the old file
#
opendir(DIR,'.');
@package=grep(/\.rpm$/,readdir(DIR));
foreach $file (@package)
{
$file =~ /(.*)\-([\d+|\.]+\w*)\-(\d+)\.[i386|noarch].*/;
$nom=$1;
$version=$2;
$buildvers=$3;
if ($NOM{$nom})
{
$version2=$VERSION{$nom};
$buildver2=$BUILDVERS{$nom};
$file2=$FILE{$nom};
$nom2=$NOM{$nom};
if ( $version2 gt $version )
{
print "$file2 is newer than $file\n";
unlink($file);
}
else
{
if ( $version2 lt $version )
{
print "$file is newer than $file2\n";
unlink($file2);
$VERSION{$nom}=$version;
$BUILDVERS{$nom}=$buildvers;
$FILE{$nom}=$file;
$NOM{$nom}=$nom;
}
else
{
if ( $buildver2 > $buildvers )
{
print "$file2 : $buildver2 est mieux que $file : $buildvers\n";
unlink($file);
}
else
{
print "$file2 : $buildver2 is older than $file : $buildvers\n";
unlink($file2);
$VERSION{$nom}=$version;
$BUILDVERS{$nom}=$buildvers;
$FILE{$nom}=$file;
$NOM{$nom}=$nom;
}
}
}
}
else
{
$VERSION{$nom}=$version;
$BUILDVERS{$nom}=$buildvers;
$FILE{$nom}=$file;
$NOM{$nom}=$nom;
}
}
# we do the hard thing here
#
system("$redhatdir/misc/src/install/genhdlist $redhatdir");
</verb></tscreen><p>
</article>