http://kbase.redhat.com/faq/docs/DOC-7273
Thanks to David Cantrell for writing this up.
WHAT IS THE STAGE 1 IMAGE?
The stage 1 image in anaconda confuses a lot of people. We've done a good job making stage 1 and stage 2 appear as a single entity, but they are really separate. The whole job of stage 1 is to load the necessary drivers to find anaconda (which we call the stage 2 image). Stage 2 could be located on a network source (NFS, HTTP, FTP), a local hard drive, or a CD-ROM drive. Stage 1 also takes care of configuring your network interface for installation if you indicate the stage 2 location is a network source.
When a bug is found in stage 1, patching it and testing it requires building a new ramdisk image to boot in to the installation environment. We cannot provide a mechanism like updates.img for several reasons. First, the stage 1 portion of anaconda is responsible for loading updates.img after it loads. Second, the stage 1 portion is basically one static program called /sbin/loader, so the only way to test it is to build a new one.
This document explains one of several ways to build a new stage 1 image for testing purposes. The example platform is Red Hat Enterprise Linux 5, but the techniques apply to Fedora Core 6 and higher.
SETTING UP YOUR DEVELOPMENT SYSTEM
I will assume you are building a new ramdisk image for the same architecture as your workstation. If you are not, you will need to explore other means of building the anaconda SRPM. You must build on the target architecture.
First, set up RPM to build locally:
mkdir -p ~/rpmbuild/RPMS
mkdir -p ~/rpmbuild/SRPMS
mkdir -p ~/rpmbuild/SOURCES
mkdir -p ~/rpmbuild/BUILD
mkdir -p ~/rpmbuild/SPECS
echo '%_topdir %(echo $HOME)/rpmbuild' > ~/.rpmmacros
Next, obtain the anaconda source RPM from the RHEL-5 media and install it. You should not be doing this as root:
rpm -Uvh anaconda-11.1.2.36-1.src.rpm
Now you are ready to patch anaconda to your liking.
PATCHING ANACONDA'S STAGE 1 COMPONENTS
Extract the source:
RPM may complain about missing dependencies. Install them if you lack them. Anaconda needs a lot of stuff in order to compile.
cd ~/rpmbuild/SPECS
rpmbuild -bp anaconda.spec
Patch the source:
cd ~/rpmbuild/BUILD/anaconda-11.1.2.36
The stage 1 sources are located in the loader2 subdirectory. The stage 1 components also link against libisys which is in the isys subdirectory. Either way, you will want to make patches against this source tree and put the patches in the ~/rpmbuild/SOURCES directory. For this example, we are patching loader2/loader.c:
Now go and edit ~/rpmbuild/SPECS/anaconda.spec and add anaconda.patch as a patch and make sure it's applied in the %prep section after the %setup macro is run.
cd loader2
cp -a loader.c loader.c.orig
vim loader.c
# la la la...making my changes.... :wq
cd ~/rpmbuild/BUILD
gendiff anaconda-11.1.2.36/ .orig > ~/rpmbuild/SOURCES/anaconda.patch
BUILDING A PATCHED ANACONDA
Simple, use rpmbuild:
The resulting packages will be written to ~/rpmbuild/RPMS
cd ~/rpmbuild/SPECS
rpmbuild -ba anaconda.spec
BUILDING A NEW RAMDISK
First, we need to get the two stage 1 components that matter: loader and init.
cd ~/rpmbuild/RPMS/
rpmdev-extract anaconda-runtime-11.1.2.36-1.i386.rpm
cd anaconda-runtime-11.1.2.36-1
cp -a usr/lib/anaconda-runtime/loader/init ~/init
cp -a usr/lib/anaconda-runtime/loader/loader ~/loader
Second, get the ramdisk image that you want to update. For this example, I am updating the pxeboot/initrd.img from the RHEL-5 tree. I have copied the initrd.img file to my home directory:
Now the initrd.img file in my home directory contains the new loader and init binaries. Copying this to the boot server and you are ready to go with a new initrd.img for RHEL-5.
cd ~
mkdir tmp-initrd
cd tmp-initrd
gzip -dc ~/initrd.img | cpio -id
cat ~/init > sbin/init
cat ~/loader > sbin/loader
(find . | cpio -c -o | gzip -9) > ~/initrd.img
cd ~
CREATING AN INSTALLATION BOOT CD-ROM WITH YOUR NEW initrd.img
isolinux (not available for Itanium systems, you'll need to use a loop back mount of the ia64 boot.iso) is used for booting the Red Hat Enterprise Linux installation CD. To create your own CD-ROM to boot the installation program, use the following instructions:
Copy the isolinux/ directory from the Red Hat Enterprise Linux CD #1 into a temporary directory (referred to here as path-to-workspace) using the following command:
Change directories to the path-to-workspace directory you have created:
cp -r path-to-cd/isolinux/ path-to-workspace
Copy the new initrd.img to path-to-workspace
cd path-to-work-space
Make sure the file(s) you have copied have appropriate permissions:
cp ~/initd.img path-to-workspace/isolinux/
Finally, issue the following command to create the ISO image file:
chmod u+w isolinux/*
mkisofs -o file.iso -b isolinux.bin -c boot.cat -no-emul-boot \\
-boot-load-size 4 -boot-info-table -R -J -v -T isolinux/
POTENTIAL STUMBLING BLOCKS
Your development workstation may be a different architecture than your target system. If you are using a Thinkpad to prepare a new ramdisk for the Itanium platform, you will be building binaries for the wrong platform. Unfortunately, there is no easy way around this. You must build the binaries on the target platform. If you don't have that sitting in front of you, find one somehow.
At Red Hat, engineers can submit packages to the build system as a scratch build. The package gets built on the desired architecture, but thrown off to the side, which works great for testing purposes.