Take a moment to skim through the Bootdisk-HOWTO or the From-PowerUp-to-BASH-Prompt-HOWTO. These HOWTO documents can be found online at http://www.tldp.org/docs.html#howto. Both documents offer an excellent view of what it takes to get a GNU/Linux system up and running. There is also a lot of information to digest. Remember that one of our goals is, "keep it simple to avoid stressing out," so we want to ignore everything but the absolutely critical pieces of a boot / root diskset.

Basically it boils down to the following required items:

We don't even need an init daemon. The kernel can be told to run the shell directly by passing it an option through the boot loader.

For easy construction we will build a two-disk boot / root set rather than trying to get everything onto a single diskette. The boot loader and kernel will go on the boot disk and the shell will reside on the root disk.

For the boot disk we simply need to install the GRUB bootloader and a Linux kernel. We will need to use a kernel that does not require modules for the hardware we need to access. Mainly, it should have compiled-in support for the floppy drive, ram disk, second extended filesystem, proc filesystem, ELF binaries, and a text-based console. If such a kernel is not available, it will need to be built from source code. Kwan Lowe's Kernel Rebuild Guide is a good reference for this task, however we can ignore the sections that deal with modules and the initial ramdisk.

For the root disk we will need a floppy that has been prepared with a filesystem. We will also need a BASH shell that is statically-linked so we can avoid the additional complexities of shared libraries. The configure program in the BASH source code recognizes the --enable-static-link option for this feature. We will also be using the --enable-minimal-config option to keep the BASH binary down to a manageable size. Additional requirements for the root disk are a /dev directory and a device file for the console. The console device is required for BASH to be able to communicate with the keyboard and video display.

There is one other, less obvious requirement to keep in mind and that is CPU compatibility. Each generation of CPU features a more complex architecture than its predecessor. Late generation chips have additional registers and instructions when compared to an older 486 or 386. So a kernel optimized for a new, fast 6x86 machine will not run on an older box. (See the README file in the Linux kernel source code for details.) A BASH shell built for a 6x86 will probably not run on an older processor either. To avoid this problem, we can choose the 386 as a lowest common denominator CPU and build all the code for that architecture.

Insert a blank diskette labeled "boot disk".

bash# mke2fs -m0 /dev/fd0
bash# mount /dev/fd0 /mnt

Get the GRUB source code from ftp://alpha.gnu.org/gnu/grub/ and unpack it into the /usr/src directory.

Configure and build the GRUB source code for an i386 processor by using the following commands:

bash# cd /usr/src/grub-0.95
bash# export CC="gcc -mcpu=i386"
bash# ./configure --host=i386-pc-linux-gnu --without-curses
bash# make

Normally, after compiling source code, one would use the command make install to copy the finished files to their proper destinations in the filesystem. However, using make install does not work well with small media like the floppy disks we are using. The problem is that there are many files in a package besides the actual binaries that get the job done. For example, there are often man or info pages that provide documentation. These extra files can take up more space than we can spare on the diskette. We can work around this limitation by copying essential files manually rather than using make install.

For GRUB to boot we will need to copy the stage1 and stage2 bootloader files to the /boot/grub directory on the boot floppy.

bash# mkdir -p /mnt/boot/grub
bash# cp /usr/src/grub-0.95/stage1/stage1 /mnt/boot/grub
bash# cp /usr/src/grub-0.95/stage2/stage2 /mnt/boot/grub

Once the bootloader's files are copied to the boot disk we can enter the grub shell to finish the installation.

bash# /usr/src/grub-0.95/grub/grub
grub> root (fd0)
grub> setup (fd0)
grub> quit

The steps for building the kernel were tested using Linux kernel version 2.4.26 and should work any 2.4.x or 2.6.x kernel. The latest version of the kernel source code may be downloaded from http://www.kernel.org/ or one of its mirrors.

bash# cd /usr/src/linux
bash# make menuconfig

Be sure to configure support for the following:

bash# make dep
bash# make clean
bash# make bzImage

bash# cp /usr/src/linux/arch/i386/boot/bzImage /mnt/boot/vmlinuz

bash# cd /
bash# umount /mnt

Insert a blank diskette labeled "root disk".

bash# mke2fs -m0 /dev/fd0
bash# mount /dev/fd0 /mnt

Get the bash-3.0 source code package from ftp://ftp.gnu.org/gnu/bash/ and untar it into the /usr/src directory.

Build BASH for an i386 CPU with the following commands:

bash# cd /usr/src/bash-3.0
bash# export CC="gcc -mcpu=i386"
bash# ./configure --enable-static-link \
  --enable-minimal-config --host=i386-pc-linux-gnu
bash# make
bash# strip bash

bash# mkdir /mnt/bin
bash# cp bash /mnt/bin/bash
bash# ln -s bash /mnt/bin/sh

bash# mkdir /mnt/dev
bash# mknod /mnt/dev/console c 5 1

bash# cd /
bash# umount /mnt

Follow these steps to boot the system:

If all goes well the screen should look something like the example shown below.

GNU GRUB version 0.95

grub> kernel (fd0)/boot/vmlinuz init=/bin/sh root=/dev/fd0 load_ramdisk=1 prompt_ramdisk=1
   [Linux-bzImage, setup=0xc00, size=0xce29b]

grub> boot

Linux version 2.4.26
..
.. [various kernel messages]
..
VFS: Insert root floppy disk to be loaded into RAM disk and press ENTER
RAMDISK: ext2 filesystem found at block 0
RAMDISK: Loading 1440 blocks [1 disk] into ram disk... done.
VFS: Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 178k freed
# _

Try out a few of BASH's built-in commands to see if things are working properly.

bash# echo "Hello World"
bash# cd /
bash# pwd
bash# echo *

Try out a few other familiar commands.

bash# ls /var
bash# mkdir /var/tmp

Notice that only commands internal to BASH actually work and that external commands like ls and mkdir do not work at all. This shortcoming is something that can be addressed in a future phase of the project. For now we should just enjoy the fact that our prototype boot / root diskset works and that it was not all that hard to build.

Remove the diskette from fd0 and restart the system using CTRL-ALT-DELETE.

First, in order to use shared libraries we will need to rebuild the BASH shell. This time we will configure it without using the --enable-static-link option. Once BASH is rebuilt we need to figure out which libraries it is linked with and be sure to include them on the root disk. The ldd command makes this job easy. By typing ldd bash on the command-line we can see a list of all the shared libraries that BASH uses. As long as all these libraries are copied to the root disk, the new BASH build should work fine.

Next, we should strip any binaries that get copied to the root disk. The manpage for strip does not give much description of what it does other than to say, "strip discards all symbols from the object files." It seems like removing pieces of a binary would render it useless, but this is not the case. The reason it works is because a large number of these discarded symbols are used for debugging. While debugging symbols are very helpful to programmers working to improve the code, they do not do much for the average end-user other than take up more disk space. And since space is at a premium, we should definitely remove as many symbols as possible from BASH and any other binaries before we copy over them to the ramdisk.

The process of stripping files to save space also works with shared library files. But when stripping libraries it is important to use the --strip-unneeded option so as not to break them. Using --strip-unneeded shrinks the file size, but leaves the symbols needed for relocation intact which is something that shared libraries need to function properly.

Finally, we can tackle the problem of how to build a compressed root filesystem. The Bootdisk-HOWTO suggests three ways of constructing a compressed root filesystem using either a ramdisk, a spare hard drive partition or a loopback device. This project will concentrate on using the ramdisk approach. It seems logical that if the root filesystem is going to be run from a ramdisk, it may as well be built on a ramdisk. All we have to do is create a second extended filesystem on a ramdisk device, mount it and copy files to it. Once the filesystem is populated with all the files that the root disk needs, we simply unmount it, compress it and write it out to floppy.

bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt

bash# cd /usr/src/bash-3.0
bash# make distclean
bash# export CC="gcc -mcpu=i386"
bash# ./configure --enable-minimal-config --host=i386-pc-linux-gnu
bash# make
bash# strip bash

bash# ldd bash

View the output from the ldd command. It should look similar to the example below.

bash# ldd bash
  libdl.so.2 => /lib/libdl.so.2 (0x4001d000)
  libc.so.6 => /lib/libc.so.6 (0x40020000)
  /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

bash# mkdir /mnt/bin
bash# cp bash /mnt/bin
bash# ln -s bash /mnt/bin/sh
bash# mkdir /mnt/lib
bash# strip --strip-unneeded -o /mnt/lib/libdl.so.2 /lib/libdl.so.2
bash# strip --strip-unneeded -o /mnt/lib/libc.so.6 /lib/libc.so.6
bash# strip --strip-unneeded -o /mnt/lib/ld-linux.so.2 /lib/ld-linux.so.2
bash# chmod +x /mnt/lib/ld-linux.so.2

bash# mkdir /mnt/dev
bash# mknod /mnt/dev/console c 5 1

bash# cd /
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase2-image bs=1k count=4096
bash# gzip -9 ~/phase2-image

Insert the floppy labeled "root disk" into drive fd0.

bash# dd if=~/phase2-image.gz of=/dev/fd0 bs=1k

Follow these steps to boot:

The screen output should be similar to the following example:

GNU GRUB version 0.95

grub> kernel (fd0)/boot/vmlinuz init=/bin/sh root=/dev/fd0 load_ramdisk=1 prompt_ramdisk=1
   [Linux-bzImage, setup=0xc00, size=0xce29b]

grub> boot

Linux version 2.4.26
..
.. [various kernel messages]
..
VFS: Insert root floppy disk to be loaded into RAM disk and press ENTER
RAMDISK: Compressed image found at block 0
VFS: Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 178k freed
# _

If the implementation was successful, this new root disk should behave exactly like the root disk from the previous chapter. The key difference is that this compressed root disk has much more room to grow and we will put this extra space to good use in the next phase of the project.

Remove the diskette from fd0 and restart the system using CTRL-ALT-DELETE.

The first question that might come to mind is, "How do we know which commands are needed?" It is possible to just start with cat and ls then install other commands as we discover a need for them. But this is terribly inefficient. We need a plan or a blueprint to work from. For this we can turn to the Filesystem Hierarchy Standard (FHS) available from http://www.pathname.com/fhs/. The FHS dictates which commands should be present on a Linux system and where they should be placed in the directory structure.

The next logical question is, "Now that we know what we need, where do we get the source code?" One way to find the answer to this question is to check the manpages. We can either search the manpages included with one of the popular GNU/Linux distributions or use one of the manpage search engines listed at http://www.tldp.org/docs.html#man. One thing that should tip us off as to where to find the source code for a particular command is the email address listed for reporting bugs. For example the cat manpage lists bug-textutils@gnu.org. From this email address we can deduce that cat is part of the textutils package from GNU.

So let's look at the FHS requirements for the /bin directory. The first few commands in the list are cat, chgrp, chmod, chown and cp. We already know that cat is part of GNU's textutils. Using the next few commands as keywords in a manpage search we discover that we need GNU's fileutils package for chmod, chgrp, chown and cp. In fact quite a few of the commands in /bin come from GNU's fileutils. The date command also comes from a GNU package called sh-utils. So a good way to tackle the problem of finding source code might be to group the commands together by package as shown below.

These four packages do not contain all of the commands in the /bin directory, but they do represent of over 70% of them. That should be enough to accomplish our goal of adding some of the commonly used external commands. We can worry about the other commands in later phases of the project.

To fetch the source code we simply need to connect to GNU's FTP site and navigate to the appropriate package directory.

When we get to the directory for textutils there are several versions available. There is also a note informing us that the package has been renamed to coreutils. The same message about coreutils appears in the fileutils and sh-utils directories as well. So instead of downloading three separate packages we can get everything in one convenient bundle in the coreutils directory.

bash# mkdir ~/staging
bash# cd ~/staging
bash# mkdir bin boot dev etc home lib mnt opt proc root sbin tmp usr var
bash# mkdir var/log var/run

bash# dd if=~/phase2-image.gz | gunzip -c > /dev/ram7
bash# mount /dev/ram7 /mnt
bash# cp -dpR /mnt/* ~/staging
bash# umount /dev/ram7
bash# rmdir ~/staging/lost+found

Download a recent version of coreutils from ftp://ftp.gnu.org/gnu/coreutils/

bash# cd /usr/src/coreutils-5.2.1
bash# export CC="gcc -mcpu=i386"
bash# ./configure --host=i386-pc-linux-gnu
bash# make
bash# cd src
bash# cp cat chgrp chmod chown cp date dd df ~/staging/bin
bash# cp hostname ln ls mkdir mkfifo mknod ~/staging/bin
bash# cp mv rm rmdir stty su sync uname ~/staging/bin

Check library requirements by using ldd on some of the new binaries.

bash# ldd ~/staging/bin/cat
bash# ldd ~/staging/bin/ls
bash# ldd ~/staging/bin/su
bash# ls ~/staging/lib

Note the differences in the required libraries, as shown by the ldd command, and the libraries present in the staging area, as shown by the ls command, then copy any missing libraries to the staging area.

bash# cp /lib/librt.so.1 ~/staging/lib
bash# cp /lib/libpthread.so.0 ~/staging/lib
bash# cp /lib/libcrypt.so.1 ~/staging/lib

bash# strip ~/staging/bin/*
bash# strip --strip-unneeded ~/staging/lib/*

bash# cd /
bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt
bash# cp -dpR ~/staging/* /mnt
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase3-image bs=1k count=4096
bash# gzip -9 ~/phase3-image

Insert the diskette labeled "root disk" into drive fd0.

bash# dd if=~/phase3-image.gz of=/dev/fd0 bs=1k

Follow these steps to get the system running.

The terminal display should look similar to the example below.

GNU GRUB version 0.95

grub> kernel (fd0)/boot/vmlinuz rw init=/bin/sh root=/dev/fd0 load_ramdisk=1 prompt_ramdisk=1
   [Linux-bzImage, setup=0xc00, size=0xce29b]

grub> boot

Linux version 2.4.26
..
.. [various kernel messages]
..
VFS: Insert root floppy disk to be loaded into RAM disk and press ENTER
RAMDISK: Compressed image found at block 0
VFS: Mounted root (ext2 filesystem) read-write.
Freeing unused kernel memory: 178k freed
# _

Now that the system is up and running, try using some of the new commands.

bash# uname -a
bash# ls /etc
bash# echo "PocketLinux" > /etc/hostname
bash# hostname $(cat /etc/hostname)
bash# uname -n
bash# mkdir /home/stuff
bash# cd /home/stuff

If everything goes well the commands like cat, ls and hostname should work now. Even mkdir should work since the root filesystem is mounted read-write. Of course since we are using a ramdisk, any changes will be lost once the PC is reset.

Remove the diskette from fd0 and restart the system using CTRL-ALT-DELETE.

We can use the Filesystem Hierarchy Standard (FHS) document to help find the names of utilities we need and where they reside in the directory structure. The FHS /sbin directory lists fsck and something called fsck.* for checking filesystems. Since we are using a Second Extended (ext2) filesystem the fsck.* becomes fsck.ext2 for our purposes. Mounting filesystems is done using the commands mount and umount in the /bin directory. However, the name of a script to automatically mount local filesystems cannot be found. On most systems this type of script is in the /etc directory, but while FHS does list requirements for /etc, it does not currently make recommendations for startup scripts. Several GNU/Linux distributions use /etc/init.d as the place to hold startup scripts so we will put our filesystem mounting script there.

In the previous chapter we used manpages to help us find source code. In this chapter we will use a tool called the Linux Software Map (LSM). LSM is a database of GNU/Linux software that tracks such things as package name, author, names of binaries that make up the package and download sites. Using an LSM search engine we can locate packages using command names as keywords.

If we search Ibiblio's Linux Software Map (LSM) at http://www.ibiblio.org/pub/Linux/ for the keyword "fsck" we get a large number of matches. Since we are using a Second Extended filesystem, called ext2 for short, we can refine the search using "ext2" as a keyword. Supplying both keywords to the LSM search engine comes up with a package called e2fsprogs. Looking at the LSM entry for e2fsprogs we find out that this package contains the utilities e2fsck, mke2fs, dumpe2fs, fsck and more. We also find out that the LSM entry for e2fsprogs has not been updated for a while. There is almost certainly a newer version out there somewhere. Another good Internet resource for source code is SourceForge at http://sourceforge.net/. Using the keyword "e2fsprogs" in the SourceForge search engine results in a much newer version of e2fsprogs.

Finding fsck was quite an adventure, but now we can move on to finding mount and umount. A search on LSM comes up with a number of matches, but most of them point to various versions of a package called util-linux. All we have to do is scroll through and pick the most recent release. The LSM entry for util-linux lists a lot of utilities besides just mount and umount. We should definitely scan through the list to see if any of the other util-linux commands show up in the FHS requirements for /bin and /sbin.

Below is a list of packages we have gathered so far and the utilities that match up with FHS.

Now that we have fsck and mount commands we need to come up with a shell script to automate checking and mounting the local filesystems. An easy way to do this would be to write a short, two line script that calls fsck and then mount. But, what if the filesystems are not clean? The system should definitely not try to mount a corrupted filesystem. Therefore we need to devise a way of determining the status of the filesystems before mounting them. The manpage for fsck gives some insight into how this can be accomplished using return codes. Basically, if fsck returns a code of zero or one it means the filesystem is okay and a return code of two or greater means some kind of manual intervention is needed. A simple if-then statement could evaluate the fsck return code to determine whether or not the filesystem should be mounted. For help on writing shell scripts we can turn to the BASH(1) manpage and the Advanced-BASH-Scripting-Guide. Both references are freely available from the Linux Documentation Project web site at http://www.tldp.org/.

The last thing to do is to figure out if any other files besides the binaries are needed. We learned about using ldd to check for library dependencies in the last phase of the project and we will use it to check the utilities in this phase too. There are also some other files that fsck and mount will need and the fsck(8) and mount(8) manpages give some insight into what those files are. There is /etc/fstab that lists devices and their mount points, /etc/mtab that keeps track of what is mounted, and a number of /dev files that represent the various disks. We will need to include all of these to have everything work right.

Download the e2fsprogs source code package from http://sourceforge.net/projects/e2fsprogs/

bash# cd /usr/src/e2fsprogs-1.35
bash# export CC="gcc -mcpu=i386"
bash# ./configure --host=i386-pc-linux-gnu
bash# make
bash# cd e2fsck
bash# cp e2fsck.shared ~/staging/sbin/e2fsck
bash# ln -s e2fsck ~/staging/sbin/fsck.ext2
bash# cd ../misc
bash# cp fsck mke2fs ~/staging/sbin
bash# ln -s mke2fs ~/staging/sbin/mkfs.ext2

Get the latest util-linux source from ftp://ftp.win.tue.nl/pub/linux-local/utils/util-linux/

bash# cd /usr/src/util-linux-2.12h

Use a text editor to make the following changes to MCONFIG:

bash# ./configure
bash# make
bash# cp disk-utils/mkfs ~/staging/sbin
bash# cp fdisk/fdisk ~/staging/sbin
bash# cp login-utils/agetty ~/staging/sbin
bash# ln -s agetty ~/staging/sbin/getty
bash# cp login-utils/login ~/staging/bin
bash# cp misc-utils/kill ~/staging/bin
bash# cp mount/mount ~/staging/bin
bash# cp mount/umount ~/staging/bin
bash# cp mount/swapon ~/staging/sbin
bash# cp sys-utils/dmesg ~/staging/bin

bash# ldd ~/staging/bin/* | more
bash# ldd ~/staging/sbin/* | more
bash# ls ~/staging/lib

All of the dependencies revealed by the ldd command are for libraries already present in the staging area so there is no need to copy anything new.

bash# strip ~/staging/bin/*
bash# strip ~/staging/sbin/*

bash# mknod ~/staging/dev/ram0 b 1 0
bash# mknod ~/staging/dev/fd0 b 2 0
bash# mknod ~/staging/dev/null c 1 3

bash# cd ~/staging/etc

Use an editor like vi, emacs or pico to create the following file and save it as ~/staging/etc/fstab.

proc        /proc   proc   defaults   0   0
/dev/ram0   /       ext2   defaults   1   1

Create an empty mtab file.

bash# echo -n >mtab

Use an editor to create the following shell script and save it as ~/staging/etc/init.d/local_fs:

#!/bin/sh
#
# local_fs - check and mount local filesystems
#
PATH=/sbin:/bin ; export PATH

fsck -ATCp
if [ $? -gt 1 ]; then
  echo "Filesystem errors still exist!  Manual intervention required."
  /bin/sh
else
  echo "Remounting / as read-write."
  mount -n -o remount,rw /
  echo -n >/etc/mtab
  mount -f -o remount,rw /
  echo "Mounting local filesystems."
  mount -a -t nonfs,nosmbfs
fi
#
# end of local_fs

Set execute permissions on the script.

bash# chmod +x local_fs

bash# cd /
bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt
bash# cp -dpR ~/staging/* /mnt
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase4-image bs=1k count=4096
bash# gzip -9 ~/phase4-image

Insert the diskette labeled "root disk" into drive fd0.

bash# dd if=~/phase4-image.gz of=/dev/fd0 bs=1k

Start the system using the following procedure:

The output should resemble the example below:

GNU GRUB version 0.95

grub> kernel (fd0)/boot/vmlinuz init=/bin/sh root=/dev/fd0 load_ramdisk=1 prompt_ramdisk=1
   [Linux-bzImage, setup=0xc00, size=0xce29b]

grub> boot

Linux version 2.4.26
..
.. [various kernel messages]
..
VFS: Insert root floppy disk to be loaded into RAM disk and press ENTER
RAMDISK: Compressed image found at block 0
VFS: Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 178k freed
# _

Run the script by typing the following commands at the shell prompt:

bash# PATH=/sbin:/bin:/etc/init.d ; export PATH
bash# cat /etc/mtab
bash# local_fs
bash# cat /etc/mtab
bash# df

If everything is working properly, then the screen output should look something like the example below.

bash# PATH=/sbin:/bin:/etc/init.d ; export PATH
bash# cat /etc/mtab
bash# local_fs
/dev/ram0: clean 74/1024 files 3178/4096 blocks
Remounting / as read-write.
Mounting local filesystems.
bash# cat /etc/mtab
/dev/ram0 / ext2 rw 0 0
proc /proc proc rw 0 0
bash# df
Filesystem      1k-blocks       Used Available Use% Mounted on
/dev/ram0       3963            3045 918        77% /

Procure a blank floppy disk and label it as "home". Remove the root disk floppy and insert the "home" diskette. Type the following commands:

bash# mkfs -t ext2 /dev/fd0
bash# fsck /dev/fd0
bash# mount /dev/fd0 /home
bash# mkdir /home/floyd
bash# cd /home/floyd
bash# echo "Goodbye cruel world." > goodbye.txt
bash# cat goodbye.txt

bash# cd /
bash# umount /home

Remove the diskette from fd0 and restart the system using CTRL-ALT-DELETE.

Loading the kernel without manually typing parameters is easy to do if we read the grub info page. According to the section entitled "configuration" all of the commands used for booting can be put in a file called menu.lst and placed in the /boot/grub directory.

To automate system start-up we will need an init daemon. We know this because the Bootdisk-HOWTO and From-Powerup-To-BASH-Prompt-HOWTO both make mention of init as the first program to start after the kernel loads. The latter HOWTO also goes into some detail about the /etc/inittab file and the organization of startup scripts. This could be helpful since FHS, the blueprint we have used so far, makes no recommendation for init scripts.

We will also need to find the shutdown command to fulfill the second goal of graceful shutdown capability.

Searching the Linux Software Map on Ibiblio for the keyword "init" gives a large number of results. From reading the From-Powerup-To-BASH-Prompt-HOWTO however, we know that most Linux systems use a System V style init daemon. Narrowing the search with the additional key phrase of "System V" gives much better results. The sysvinit package contains init, shutdown, halt and reboot which is everything we need. The version listed in the LSM entry looks to be pretty old, but there is a primary-site URL that will probably lead to the latest version.

The manpage for init mentions a FIFO called /dev/initctl that is required for init to communicate with other programs in the sysvinit package. We will have to create this file for init to function properly.

Using a GRUB configuration file is slightly more complex than specifying the bootloader commands manually. There are directives for features like menus, default selections and timeouts that need to be specified in the configuration file as well as the familiar kernel loading command. The info page for GRUB gives much of the necessary information. We may also be able to use the GRUB configuration file on the development system as a template. However, there is some inconsistency between vendors as to the name and location of the file. Regardless of what the path is on the development system it should be /boot/grub/menu.lst on the Pocket Linux System.

Many of the popular GNU/Linux distributions use System V style init scripts. Since we are using a "sysvinit" daemon it makes sense to use System V style scripts as well. The following documents all touch upon the System V style init scripts in some way and will serve as references when building the scripts for this project:

After glancing at one or two of the above references we should have a pretty good idea of how the System V style system initialization process works. We should also know what it takes to create System V style init scripts for the Pocket Linux project. Below is a brief list of what needs to be done:

As always, the BASH(1) manpage and the Advanced BASH Scripting Guide are very helpful for writing and understanding shell scripts.

Insert and mount the floppy labeled "boot disk".

bash# mount /dev/fd0 /mnt
bash# cd /mnt/boot/grub

Use your favorite text editor to create the following file and save it as /mnt/boot/grub/menu.lst:

default 0
timeout 3
title Pocket Linux Boot Disk
kernel (fd0)/boot/vmlinuz root=/dev/fd0 load_ramdisk=1 prompt_ramdisk=1

Download the latest sysvinit source from ftp://ftp.cistron.nl/pub/people/miquels/software/

bash# cd /usr/src/sysvinit-2.85/src
bash# make CC="gcc -mcpu=i386"
bash# cp halt init shutdown ~/staging/sbin
bash# ln -s halt ~/staging/sbin/reboot
bash# ln -s init ~/staging/sbin/telinit
bash# mknod ~/staging/dev/initctl p

Use a text editor to create the following file and save it as ~/staging/etc/inittab

# /etc/inittab - init daemon configuration file
#
# Default runlevel
id:1:initdefault:
#
# System initialization
si:S:sysinit:/etc/init.d/rc S
#
# Runlevel scripts
r0:0:wait:/etc/init.d/rc 0
r1:1:respawn:/bin/sh
r2:2:wait:/etc/init.d/rc 2
r3:3:wait:/etc/init.d/rc 3
r4:4:wait:/etc/init.d/rc 4
r5:5:wait:/etc/init.d/rc 5
r6:6:wait:/etc/init.d/rc 6
#
# end of /etc/inittab

Use a text editor to create the following file and save it as ~/staging/etc/init.d/rc

#!/bin/sh
#
# /etc/init.d/rc - runlevel change script
#
PATH=/sbin:/bin
SCRIPT_DIR="/etc/rc$1.d"
#
# Check that the rcN.d directory really exists.
if [ -d $SCRIPT_DIR ]; then
#
# Execute the kill scripts first.
  for SCRIPT in $SCRIPT_DIR/K*; do
    if [ -x $SCRIPT ]; then
      $SCRIPT stop;
    fi;
  done;
#
# Do the Start scripts last.
  for SCRIPT in $SCRIPT_DIR/S*; do
    if [ -x $SCRIPT ]; then
      $SCRIPT start;
    fi;
  done;
fi
#
# end of /etc/init.d/rc

Make the file executable.

bash# chmod +x ~/staging/etc/init.d/rc

A case statement is added to allow the script to either mount or unmount local filesystems depending on the command-line argument given. The original script is contained inside the "start" portion of the case statement. The "stop" portion is new.

#!/bin/sh
#
# local_fs - check and mount local filesystems
#
PATH=/sbin:/bin ; export PATH

case $1 in

start)
  echo "Checking local filesystem integrity."
  fsck -ATCp
  if [ $? -gt 1 ]; then
    echo "Filesystem errors still exist!  Manual intervention required."
    /bin/sh
  else
    echo "Remounting / as read-write."
    mount -n -o remount,rw /
    echo -n > /etc/mtab
    mount -f -o remount,rw /
    echo "Mounting local filesystems."
    mount -a -t nonfs,smbfs
  fi
;;

stop)
  echo "Unmounting local filesystems."
  umount -a -r
;;

*)
  echo "usage: $0 start|stop";
;;

esac
#
# end of local_fs

Use a text editor to create the following script and save it as ~/staging/etc/init.d/hostname

#!/bin/sh
#
# hostname - set the system name to the name stored in /etc/hostname
#
PATH=/sbin:/bin ; export PATH

echo "Setting hostname."
if [ -f /etc/hostname ]; then
  hostname $(cat /etc/hostname)
else
  hostname gnu-linux
fi
#
# end of hostname

Use a text editor to create ~/staging/etc/init.d/halt as shown below.

#!/bin/sh
#
# halt - halt the system
#
PATH=/sbin:/bin ; export PATH

echo "Initiating system halt."
halt
#
# end of /etc/init.d/halt

Create the following script and save it as ~/staging/etc/init.d/reboot

#!/bin/sh
#
# reboot - reboot the system
#
PATH=/sbin:/bin ; export PATH

echo "Initiating system reboot."
reboot
#
# end of /etc/init.d/reboot

Flag all script files as executable.

bash# chmod +x ~/staging/etc/init.d/*

bash# cd ~/staging/etc
bash# mkdir rc0.d rc1.d rc2.d rc3.d rc4.d rc5.d rc6.d rcS.d
bash# cd ~/staging/etc/rcS.d
bash# ln -s ../init.d/local_fs S20local_fs
bash# ln -s ../init.d/hostname S30hostname
bash# cd ~/staging/etc/rc0.d
bash# ln -s ../init.d/local_fs K10local_fs
bash# ln -s ../init.d/halt K90halt
bash# cd ~/staging/etc/rc6.d
bash# ln -s ../init.d/local_fs K10local_fs
bash# ln -s ../init.d/reboot K90reboot

bash# cd /
bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt
bash# cp -dpR ~/staging/* /mnt
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase5-image bs=1k
bash# gzip -9 ~/phase5-image

Insert the diskette labeled "root disk" into drive fd0.

bash# dd if=~/phase5-image.gz of=/dev/fd0 bs=1k

Boot the PC using the floppy labeled "boot disk". Place the recently created root disk in fd0 when prompted. The output should resemble the example below:

GNU GRUB version 0.95

Uncompressing Linux... Ok, booting kernel.
..
.. [various kernel messages]
..
VFS: Insert root floppy to be loaded into RAM disk and press ENTER
RAMDISK: Compressed image found at block 0
VFS: Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 178k freed
Checking local filesystem integrity.
/dev/ram0: clean 105/1024 files 2842/4096 blocks
Remounting / as read-write.
Mounting local filesystems.
Setting the hostname.
INIT: Entering runlevel: 1
# _

Use the mount command to check that local filesystems are mounted as read-write. The output should look like the example below.

bash# mount
/dev/root on / type ext2 (rw)
proc on /proc type proc (rw)

Check the hostname.

bash# uname -n
gnu-linux

Bring the system down gracefully with the shutdown command.

bash# shutdown -h now

We should see the following output from init and the shutdown scripts:

INIT: Switching to runlevel: 0
INIT: Sending processes the TERM signal
Terminated
INIT: Sending processes the KILL signal
Unmounting local filesystems.
Initiating system halt.
System halted.

The From-Powerup-To-BASH-Prompt-HOWTO does a good job of outlining the steps in the login process. Basically it works like this.

The getty and login programs were already installed as part of the util-linux package so there is no need to download any new source code.

Running ldd on the login program from util-linux will reveal that it is linked to the libraries libcrypt.so.1, libc.so.6 and ld-linux.so.2. In addition to these libraries there is another, unseen dependency on libnss_files.so.2 and the configuration file /etc/nsswitch.conf.

The name service switch library libnss_files.so.2 and nsswitch.conf are required for libc.so.6, and consequently the login program, to access the /etc/passwd file. Without libnss and its configuration file, all logins will mysteriously fail. More information about glibc's use of the name service switch libraries can be found at http://www.gnu.org/software/libc/manual/html_node/Name-Service-Switch.html.

Previously, with the single user system, there was no need to worry about permissions when installing directories, files and device nodes. The shell was effectively operating as root, so everything was accessible. Things become more complex with the addition of multiple user capability. Now we need to make sure that every user has access to what they need and at the same time gets blocked from what they do not need.

A good guideline for assigning ownership and permissions would be to give the minimum level of access required. Take the /bin directory as an example. The Filesystem Hierarchy (FHS) document says, "/bin contains commands that may be used by both the system administrator and by users". From that statement we can infer that /bin should have read and execute permission for everyone. On the other hand, the /boot directory contains files for the boot loader. Chances are good that regular users will not need to access anything in the /boot directory. So the minimum level of access would be read permission for the root user and other administrators who are members of the root group. Normal users would have no permissions assigned on the /boot directory.

Most of the time we can assign similar permissions to all the commands in a directory, but there are some programs that prove to be exceptions to the rule. The su command is a good example. Other commands in the /bin directory have a minimum requirement of read and execute, but the su command needs to be setuid root in order to run correctly. Since it is a setuid binary, it might not be a good idea to allow just anyone to run it. Ownership of 0:0 (root user, root group) and permissions of rwsr-x--- (octal 4750) would be a good fit for su.

The same logic can be applied to other directories and files in the root filesystem using the following steps:

bash# ls ~/staging/sbin/getty
bash# ls ~/staging/bin/login

Modify ~/staging/etc/inittab by changing the default runlevel and adding getty entries as shown below.

# /etc/inittab - init daemon configuration file
#
# Default runlevel
id:2:initdefault:
#
# System initialization
si:S:sysinit:/etc/init.d/rc S
#
# Runlevel scripts
r0:0:wait:/etc/init.d/rc 0
r1:1:respawn:/bin/sh
r2:2:wait:/etc/init.d/rc 2
r3:3:wait:/etc/init.d/rc 3
r4:4:wait:/etc/init.d/rc 4
r5:5:wait:/etc/init.d/rc 5
r6:6:wait:/etc/init.d/rc 6
#
# Spawn virtual terminals
1:235:respawn:/sbin/getty 38400 tty1 linux
2:235:respawn:/sbin/getty 38400 tty2 linux
3:235:respawn:/sbin/getty 38400 tty3 linux
4:235:respawn:/sbin/getty 38400 tty4 linux
5:235:respawn:/sbin/getty 38400 tty5 linux
6:2345:respawn:/sbin/getty 38400 tty6 linux
#
# end of /etc/inittab

bash# cd ~/staging/dev
bash# mknod ~/staging/dev/tty0 c 4 0
bash# mknod ~/staging/dev/tty1 c 4 1
bash# mknod ~/staging/dev/tty2 c 4 2
bash# mknod ~/staging/dev/tty3 c 4 3
bash# mknod ~/staging/dev/tty4 c 4 4
bash# mknod ~/staging/dev/tty5 c 4 5
bash# mknod ~/staging/dev/tty6 c 4 6
bash# mknod ~/staging/dev/tty c 5 0

bash# cp /lib/libnss_files.so.2 ~/staging/lib
bash# strip --strip-unneeded ~/staging/lib/*

Set minimal privileges on all files and directories under ~/staging. Everything is owned by the root user and the root group. Permissions are read-write for the owner and read-only for the group. Exceptions to the blanket permissions are handled case by case.

bash# cd ~/staging
bash# chown -R 0:0 ~/staging/*
bash# chmod -R 640 ~/staging/*

Set execute permission on all directories. (Note the capital "X")

bash# chmod -R +X ~/staging/*

Files in /bin are read and execute for all, but su is an exception.

bash# chmod 755 ~/staging/bin/*
bash# chmod 4750 ~/staging/bin/su

Files in /dev have various permissions. Disk devices should be accessible to administrators only. Other files like /dev/null should have full privileges granted to everyone.

bash# chmod 660 ~/staging/dev/fd0 dev/ram0
bash# chmod 666 ~/staging/dev/null
bash# chmod 622 ~/staging/dev/console
bash# chmod 600 ~/staging/dev/initctl
bash# chmod 622 ~/staging/dev/tty
bash# chmod 622 ~/staging/dev/tty?

The passwd and group files must be world readable.

bash# chmod 644 ~/staging/etc/passwd
bash# chmod 644 ~/staging/etc/group

The scripts in /etc/init.d are read and execute for administrators.

bash# chmod 750 ~/staging/etc/init.d/*

Libraries need read and execute permissions for everyone.

bash# chmod 755 ~/staging/lib/*

Only root should have access to the /root directory.

bash# chmod 700 ~/staging/root

Make files in /sbin read and execute for administrators.

bash# chmod 750 ~/staging/sbin/*

Temp should be read-write for all with the sticky bit set.

bash# chmod 1777 ~/staging/tmp

bash# cd /
bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt
bash# cp -dpR ~/staging/* /mnt
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase6-image bs=1k count=4096
bash# gzip -9 ~/phase6-image

Insert the diskette labeled "root disk" into drive fd0.

bash# dd if=~/phase6-image.gz of=/dev/fd0 bs=1k

If everything goes well, the virtual console display should look similar to the following example:

Connected to tty1 at 38400 bps.
gnu-linux login:

Log in as root.

Create a new, unprivileged user and new group by appending a line to the /etc/passwd and /etc/group files, respectively. Be sure to use a double greater-than (>>) to avoid accidentally overwriting the files.

bash# echo "floyd::501:500:User:/home/floyd:/bin/sh" >>/etc/passwd
bash# echo "users::500:" >>/etc/group
bash# mkdir /home/floyd
bash# chown floyd.users /home/floyd
bash# chmod 700 /home/floyd

Switch to virtual terminal tty2 by pressing ALT+F2.

Log in as floyd.

Try the following commands and verify that they work.

bash$ pwd
bash$ ls -l /
bash$ cat /etc/passwd

Try the following commands and verify that they do not work.

bash$ ls /root
bash$ /sbin/shutdown -h now
bash$ su -

Switch back to tty1 where root is logged in.

bash# shutdown -h now

There is a more command that comes with util-linux, but it will not work for this project. The reason is because of library dependencies and space constraints. The util-linux supplied more needs either the libncurses or libtermcap to work and there just is not enough space on the root disk floppy to fit everything in. So, in order to have a more command we will have to get creative.

The more command is used to display a file page by page. It's a little like having a cat command that pauses every twenty-five lines. The basic logic is outlined below.

Of course there are some details left out like what to do if the screen dimensions are not what we anticipated, but overall it is a fair representation of what more does. Given this simple program logic, it should not be hard to put together a short shell script that emulates the basic functionality of more. The BASH(1) manpage and Adv-BASH-Scripting-Guide will serve as references.

The more script will need access to device files that are not on the root disk yet. Specifically more needs to have stdin, stdout and stderr, but while we are at it we should check for any other missing /dev files. The Linux Standard Base requires null, zero and tty to be present in the /dev directory. Files for null and tty already exist from previous phases of the project, but we still need /dev/zero. We can refer to devices.txt in the Linux source code Documentation directory for major and minor numbers.

These three packages can be found by using the Internet resources we have used before plus one new site. The "sed" and "ed" packages can be found at the same place we found BASH, on the GNU FTP server. The procps package shows up in an Ibiblio LSM search, but it is an old version. In order to find the latest version we can go to the Freshmeat website at http://freshmeat.net and search for "procps" in projects.

Both "sed" and "ed" packages feature GNU's familiar configure script and are therefore very easy to build. There is no configure script for "procps" but this does not make things too difficult. We can just read the package's README file to find out about how to set various configuration options. We can use one of these options to avoid the complexity of using and installing libproc. Setting SHARED=0 makes libproc an integrated part of ps rather than a separate, shared library.

Create the following script with a text editor and save it as ~/staging/bin/more.sh

#!/bin/sh
#
# more.sh - emulates the basic functions of the "more" binary without
#           requiring ncurses or termcap libraries.
#
# Assume input is coming from STDIN unless a valid file is given as
# a command-line argument. 
if [ -f $1 ]; then
  INPUT="$1"
else
  INPUT="/dev/stdin"
fi
#
# Set IFS to newline only. See BASH(1) manpage for details on IFS.
IFS=$'\n'
#
# If terminal dimensions are not already set as shell variables, take
# a guess of 80x25.
if [ "$COLUMNS" = "" ]; then
  let COLUMNS=80;
fi
if [ "$LINES" = "" ]; then
  let LINES=25;
fi
#
# Initialize line counter variable
let LINE_COUNTER=$LINES
#
# Read the input file one line at a time and display on STDOUT until
# the page fills up. Display "Press <Enter>" message on STDERR and wait
# for keypress from STDERR.  Continue until the end of the input file.
# Any input line greater than $COLUMNS characters in length is wrapped
# and counts as multiple lines.
#
while read -n $COLUMNS LINE_BUFFER; do
  echo "$LINE_BUFFER"
  let LINE_COUNTER=$LINE_COUNTER-1
  if [ $LINE_COUNTER -le 1 ]; then
    echo "Press <ENTER> for next page or <CTRL>+C to quit.">/dev/stderr
    read</dev/stderr
    let LINE_COUNTER=$LINES
  fi
done<$INPUT
#
# end of more.sh

Create a symbolic link for more

bash# ln -s more.sh ~/staging/bin/more

bash# ln -s /proc/self/fd ~/staging/dev/fd
bash# ln -s fd/0 ~/staging/dev/stdin
bash# ln -s fd/1 ~/staging/dev/stdout
bash# ln -s fd/2 ~/staging/dev/stderr
bash# mknod -m644 ~/staging/dev/zero c 1 5

Get the latest procps source package from http://procps.sourceforge.net/

bash# cd /usr/src/procps-3.2.3
bash# make SHARED=0 CC="gcc -mcpu=i386"
bash# cd ps
bash# cp ps ~/staging/bin

Download GNU's sed from ftp://ftp.gnu.org/gnu/sed/

bash# cd /usr/src/sed-4.1.2
bash# export CC="gcc -mcpu=i386"
bash# ./configure --host=i386-pc-linux-gnu
bash# make
bash# cd sed
bash# cp sed ~/staging/bin

The ed package also comes from GNU at ftp://ftp.gnu.org/gnu/ed/

bash# cd /usr/src/ed-0.2
bash# ./configure --host=i386-pc-linux-gnu
bash# make
bash# cp ed ~/staging/bin

bash# strip ~/staging/bin/*

bash# chown 0:0 ~/staging/bin/*
bash# chmod -R 755 ~/staging/bin
bash# chmod 4750 ~/staging/bin/su

bash# cd /
bash# dd if=/dev/zero of=/dev/ram7 bs=1k count=4096
bash# mke2fs -m0 /dev/ram7 4096
bash# mount /dev/ram7 /mnt
bash# cp -dpR ~/staging/* /mnt
bash# umount /dev/ram7
bash# dd if=/dev/ram7 of=~/phase7-image bs=1k
bash# gzip -9 ~/phase7-image

Insert the diskette labeled "root disk" into drive fd0.

bash# dd if=~/phase7-image.gz of=/dev/fd0 bs=1k

Boot from the diskset in the usual way and log in as root.

Display kernel messages by piping the output of dmesg to more.

bash# dmesg | more

Examine the local_fs script by using more with a command-line argument.

bash# more /etc/init.d/local_fs

Display processes for the user currently logged in.

bash# ps

Display all available information about all running processes.

bash# ps -ef

Use sed to display an alternate version of /etc/passwd.

bash# sed -e "s/Legacy/Old School/" /etc/passwd

Verify that sed did not make the changes permanent.

bash# cat /etc/passwd

Use ed to change properties on the "daemon" user.

bash# ed -p*
ed* r /etc/passwd
ed* %p
ed* /daemon/s/Legacy/Old School/
ed* %p
ed* w
ed* q

Verify that the changes are permanent (at least until the system is restarted.)

bash# cat /etc/passwd

Bring the system down gracefully with the shutdown command.