1Using the initial RAM disk (initrd)
2===================================
3
4Written 1996,2000 by Werner Almesberger <werner.almesberger@epfl.ch> and
5                     Hans Lermen <lermen@fgan.de>
6
7
8initrd provides the capability to load a RAM disk by the boot loader.
9This RAM disk can then be mounted as the root file system and programs
10can be run from it. Afterwards, a new root file system can be mounted
11from a different device. The previous root (from initrd) is then moved
12to a directory and can be subsequently unmounted.
13
14initrd is mainly designed to allow system startup to occur in two phases,
15where the kernel comes up with a minimum set of compiled-in drivers, and
16where additional modules are loaded from initrd.
17
18This document gives a brief overview of the use of initrd. A more detailed
19discussion of the boot process can be found in [1].
20
21
22Operation
23---------
24
25When using initrd, the system typically boots as follows:
26
27  1) the boot loader loads the kernel and the initial RAM disk
28  2) the kernel converts initrd into a "normal" RAM disk and
29     frees the memory used by initrd
30  3) if the root device is not /dev/ram0, the old (deprecated)
31     change_root procedure is followed. see the "Obsolete root change
32     mechanism" section below.
33  4) root device is mounted. if it is /dev/ram0, the initrd image is
34     then mounted as root
35  5) /sbin/init is executed (this can be any valid executable, including
36     shell scripts; it is run with uid 0 and can do basically everything
37     init can do).
38  6) init mounts the "real" root file system
39  7) init places the root file system at the root directory using the
40     pivot_root system call
41  8) init execs the /sbin/init on the new root filesystem, performing
42     the usual boot sequence
43  9) the initrd file system is removed
44
45Note that changing the root directory does not involve unmounting it.
46It is therefore possible to leave processes running on initrd during that
47procedure. Also note that file systems mounted under initrd continue to
48be accessible.
49
50
51Boot command-line options
52-------------------------
53
54initrd adds the following new options:
55
56  initrd=<path>    (e.g. LOADLIN)
57
58    Loads the specified file as the initial RAM disk. When using LILO, you
59    have to specify the RAM disk image file in /etc/lilo.conf, using the
60    INITRD configuration variable.
61
62  noinitrd
63
64    initrd data is preserved but it is not converted to a RAM disk and
65    the "normal" root file system is mounted. initrd data can be read
66    from /dev/initrd. Note that the data in initrd can have any structure
67    in this case and doesn't necessarily have to be a file system image.
68    This option is used mainly for debugging.
69
70    Note: /dev/initrd is read-only and it can only be used once. As soon
71    as the last process has closed it, all data is freed and /dev/initrd
72    can't be opened anymore.
73
74  root=/dev/ram0
75
76    initrd is mounted as root, and the normal boot procedure is followed,
77    with the RAM disk mounted as root.
78
79Compressed cpio images
80----------------------
81
82Recent kernels have support for populating a ramdisk from a compressed cpio
83archive. On such systems, the creation of a ramdisk image doesn't need to
84involve special block devices or loopbacks; you merely create a directory on
85disk with the desired initrd content, cd to that directory, and run (as an
86example):
87
88find . | cpio --quiet -H newc -o | gzip -9 -n > /boot/imagefile.img
89
90Examining the contents of an existing image file is just as simple:
91
92mkdir /tmp/imagefile
93cd /tmp/imagefile
94gzip -cd /boot/imagefile.img | cpio -imd --quiet
95
96Installation
97------------
98
99First, a directory for the initrd file system has to be created on the
100"normal" root file system, e.g.
101
102# mkdir /initrd
103
104The name is not relevant. More details can be found on the pivot_root(2)
105man page.
106
107If the root file system is created during the boot procedure (i.e. if
108you're building an install floppy), the root file system creation
109procedure should create the /initrd directory.
110
111If initrd will not be mounted in some cases, its content is still
112accessible if the following device has been created:
113
114# mknod /dev/initrd b 1 250 
115# chmod 400 /dev/initrd
116
117Second, the kernel has to be compiled with RAM disk support and with
118support for the initial RAM disk enabled. Also, at least all components
119needed to execute programs from initrd (e.g. executable format and file
120system) must be compiled into the kernel.
121
122Third, you have to create the RAM disk image. This is done by creating a
123file system on a block device, copying files to it as needed, and then
124copying the content of the block device to the initrd file. With recent
125kernels, at least three types of devices are suitable for that:
126
127 - a floppy disk (works everywhere but it's painfully slow)
128 - a RAM disk (fast, but allocates physical memory)
129 - a loopback device (the most elegant solution)
130
131We'll describe the loopback device method:
132
133 1) make sure loopback block devices are configured into the kernel
134 2) create an empty file system of the appropriate size, e.g.
135    # dd if=/dev/zero of=initrd bs=300k count=1
136    # mke2fs -F -m0 initrd
137    (if space is critical, you may want to use the Minix FS instead of Ext2)
138 3) mount the file system, e.g.
139    # mount -t ext2 -o loop initrd /mnt
140 4) create the console device:
141    # mkdir /mnt/dev
142    # mknod /mnt/dev/console c 5 1
143 5) copy all the files that are needed to properly use the initrd
144    environment. Don't forget the most important file, /sbin/init
145    Note that /sbin/init's permissions must include "x" (execute).
146 6) correct operation the initrd environment can frequently be tested
147    even without rebooting with the command
148    # chroot /mnt /sbin/init
149    This is of course limited to initrds that do not interfere with the
150    general system state (e.g. by reconfiguring network interfaces,
151    overwriting mounted devices, trying to start already running demons,
152    etc. Note however that it is usually possible to use pivot_root in
153    such a chroot'ed initrd environment.)
154 7) unmount the file system
155    # umount /mnt
156 8) the initrd is now in the file "initrd". Optionally, it can now be
157    compressed
158    # gzip -9 initrd
159
160For experimenting with initrd, you may want to take a rescue floppy and
161only add a symbolic link from /sbin/init to /bin/sh. Alternatively, you
162can try the experimental newlib environment [2] to create a small
163initrd.
164
165Finally, you have to boot the kernel and load initrd. Almost all Linux
166boot loaders support initrd. Since the boot process is still compatible
167with an older mechanism, the following boot command line parameters
168have to be given:
169
170  root=/dev/ram0 rw
171
172(rw is only necessary if writing to the initrd file system.)
173
174With LOADLIN, you simply execute
175
176     LOADLIN <kernel> initrd=<disk_image>
177e.g. LOADLIN C:\LINUX\BZIMAGE initrd=C:\LINUX\INITRD.GZ root=/dev/ram0 rw
178
179With LILO, you add the option INITRD=<path> to either the global section
180or to the section of the respective kernel in /etc/lilo.conf, and pass
181the options using APPEND, e.g.
182
183  image = /bzImage
184    initrd = /boot/initrd.gz
185    append = "root=/dev/ram0 rw"
186
187and run /sbin/lilo
188
189For other boot loaders, please refer to the respective documentation.
190
191Now you can boot and enjoy using initrd.
192
193
194Changing the root device
195------------------------
196
197When finished with its duties, init typically changes the root device
198and proceeds with starting the Linux system on the "real" root device.
199
200The procedure involves the following steps:
201 - mounting the new root file system
202 - turning it into the root file system
203 - removing all accesses to the old (initrd) root file system
204 - unmounting the initrd file system and de-allocating the RAM disk
205
206Mounting the new root file system is easy: it just needs to be mounted on
207a directory under the current root. Example:
208
209# mkdir /new-root
210# mount -o ro /dev/hda1 /new-root
211
212The root change is accomplished with the pivot_root system call, which
213is also available via the pivot_root utility (see pivot_root(8) man
214page; pivot_root is distributed with util-linux version 2.10h or higher
215[3]). pivot_root moves the current root to a directory under the new
216root, and puts the new root at its place. The directory for the old root
217must exist before calling pivot_root. Example:
218
219# cd /new-root
220# mkdir initrd
221# pivot_root . initrd
222
223Now, the init process may still access the old root via its
224executable, shared libraries, standard input/output/error, and its
225current root directory. All these references are dropped by the
226following command:
227
228# exec chroot . what-follows <dev/console >dev/console 2>&1
229
230Where what-follows is a program under the new root, e.g. /sbin/init
231If the new root file system will be used with udev and has no valid
232/dev directory, udev must be initialized before invoking chroot in order
233to provide /dev/console.
234
235Note: implementation details of pivot_root may change with time. In order
236to ensure compatibility, the following points should be observed:
237
238 - before calling pivot_root, the current directory of the invoking
239   process should point to the new root directory
240 - use . as the first argument, and the _relative_ path of the directory
241   for the old root as the second argument
242 - a chroot program must be available under the old and the new root
243 - chroot to the new root afterwards
244 - use relative paths for dev/console in the exec command
245
246Now, the initrd can be unmounted and the memory allocated by the RAM
247disk can be freed:
248
249# umount /initrd
250# blockdev --flushbufs /dev/ram0
251
252It is also possible to use initrd with an NFS-mounted root, see the
253pivot_root(8) man page for details.
254
255
256Usage scenarios
257---------------
258
259The main motivation for implementing initrd was to allow for modular
260kernel configuration at system installation. The procedure would work
261as follows:
262
263  1) system boots from floppy or other media with a minimal kernel
264     (e.g. support for RAM disks, initrd, a.out, and the Ext2 FS) and
265     loads initrd
266  2) /sbin/init determines what is needed to (1) mount the "real" root FS
267     (i.e. device type, device drivers, file system) and (2) the
268     distribution media (e.g. CD-ROM, network, tape, ...). This can be
269     done by asking the user, by auto-probing, or by using a hybrid
270     approach.
271  3) /sbin/init loads the necessary kernel modules
272  4) /sbin/init creates and populates the root file system (this doesn't
273     have to be a very usable system yet)
274  5) /sbin/init invokes pivot_root to change the root file system and
275     execs - via chroot - a program that continues the installation
276  6) the boot loader is installed
277  7) the boot loader is configured to load an initrd with the set of
278     modules that was used to bring up the system (e.g. /initrd can be
279     modified, then unmounted, and finally, the image is written from
280     /dev/ram0 or /dev/rd/0 to a file)
281  8) now the system is bootable and additional installation tasks can be
282     performed
283
284The key role of initrd here is to re-use the configuration data during
285normal system operation without requiring the use of a bloated "generic"
286kernel or re-compiling or re-linking the kernel.
287
288A second scenario is for installations where Linux runs on systems with
289different hardware configurations in a single administrative domain. In
290such cases, it is desirable to generate only a small set of kernels
291(ideally only one) and to keep the system-specific part of configuration
292information as small as possible. In this case, a common initrd could be
293generated with all the necessary modules. Then, only /sbin/init or a file
294read by it would have to be different.
295
296A third scenario is more convenient recovery disks, because information
297like the location of the root FS partition doesn't have to be provided at
298boot time, but the system loaded from initrd can invoke a user-friendly
299dialog and it can also perform some sanity checks (or even some form of
300auto-detection).
301
302Last not least, CD-ROM distributors may use it for better installation
303from CD, e.g. by using a boot floppy and bootstrapping a bigger RAM disk
304via initrd from CD; or by booting via a loader like LOADLIN or directly
305from the CD-ROM, and loading the RAM disk from CD without need of
306floppies. 
307
308
309Obsolete root change mechanism
310------------------------------
311
312The following mechanism was used before the introduction of pivot_root.
313Current kernels still support it, but you should _not_ rely on its
314continued availability.
315
316It works by mounting the "real" root device (i.e. the one set with rdev
317in the kernel image or with root=... at the boot command line) as the
318root file system when linuxrc exits. The initrd file system is then
319unmounted, or, if it is still busy, moved to a directory /initrd, if
320such a directory exists on the new root file system.
321
322In order to use this mechanism, you do not have to specify the boot
323command options root, init, or rw. (If specified, they will affect
324the real root file system, not the initrd environment.)
325  
326If /proc is mounted, the "real" root device can be changed from within
327linuxrc by writing the number of the new root FS device to the special
328file /proc/sys/kernel/real-root-dev, e.g.
329
330  # echo 0x301 >/proc/sys/kernel/real-root-dev
331
332Note that the mechanism is incompatible with NFS and similar file
333systems.
334
335This old, deprecated mechanism is commonly called "change_root", while
336the new, supported mechanism is called "pivot_root".
337
338
339Mixed change_root and pivot_root mechanism
340------------------------------------------
341
342In case you did not want to use root=/dev/ram0 to trigger the pivot_root
343mechanism, you may create both /linuxrc and /sbin/init in your initrd image.
344
345/linuxrc would contain only the following:
346
347#! /bin/sh
348mount -n -t proc proc /proc
349echo 0x0100 >/proc/sys/kernel/real-root-dev
350umount -n /proc
351
352Once linuxrc exited, the kernel would mount again your initrd as root,
353this time executing /sbin/init. Again, it would be the duty of this init
354to build the right environment (maybe using the root= device passed on
355the cmdline) before the final execution of the real /sbin/init.
356
357
358Resources
359---------
360
361[1] Almesberger, Werner; "Booting Linux: The History and the Future"
362    http://www.almesberger.net/cv/papers/ols2k-9.ps.gz
363[2] newlib package (experimental), with initrd example
364    http://sources.redhat.com/newlib/
365[3] util-linux: Miscellaneous utilities for Linux
366    http://www.kernel.org/pub/linux/utils/util-linux/
367