1 .. SPDX-License-Identifier: GPL-2.0
2
3 ===========================
4 Ramfs, rootfs and initramfs
5 ===========================
6
7 October 17, 2005
8
9 :Author: Rob Landley <rob@landley.net>
10
11 What is ramfs?
12 --------------
13
14 Ramfs is a very simple filesystem that exports
15 mechanisms (the page cache and dentry cache) a
16 RAM-based filesystem.
17
18 Normally all files are cached in memory by Lin
19 backing store (usually the block device the fi
20 around in case it's needed again, but marked a
21 Virtual Memory system needs the memory for som
22 written to files is marked clean as soon as it
23 store, but kept around for caching purposes un
24 memory. A similar mechanism (the dentry cache
25 directories.
26
27 With ramfs, there is no backing store. Files
28 dentries and page cache as usual, but there's
29 This means the pages are never marked clean, s
30 VM when it's looking to recycle memory.
31
32 The amount of code required to implement ramfs
33 work is done by the existing Linux caching inf
34 you're mounting the disk cache as a filesystem
35 an optional component removable via menuconfig
36 space savings.
37
38 ramfs and ramdisk:
39 ------------------
40
41 The older "ram disk" mechanism created a synth
42 an area of RAM and used it as backing store fo
43 device was of fixed size, so the filesystem mo
44 size. Using a ram disk also required unnecess
45 fake block device into the page cache (and cop
46 as creating and destroying dentries. Plus it
47 (such as ext2) to format and interpret this da
48
49 Compared to ramfs, this wastes memory (and mem
50 unnecessary work for the CPU, and pollutes the
51 to avoid this copying by playing with the page
52 complicated and turn out to be about as expens
53 More to the point, all the work ramfs is doing
54 since all file access goes through the page an
55 disk is simply unnecessary; ramfs is internall
56
57 Another reason ramdisks are semi-obsolete is t
58 loopback devices offered a more flexible and c
59 synthetic block devices, now from files instea
60 See losetup (8) for details.
61
62 ramfs and tmpfs:
63 ----------------
64
65 One downside of ramfs is you can keep writing
66 up all memory, and the VM can't free it becaus
67 should get written to backing store (rather th
68 got any backing store. Because of this, only
69 be allowed write access to a ramfs mount.
70
71 A ramfs derivative called tmpfs was created to
72 to write the data to swap space. Normal users
73 tmpfs mounts. See Documentation/filesystems/t
74
75 What is rootfs?
76 ---------------
77
78 Rootfs is a special instance of ramfs (or tmpf
79 always present in 2.6 systems. You can't unmo
80 same reason you can't kill the init process; r
81 to check for and handle an empty list, it's sm
82 to just make sure certain lists can't become e
83
84 Most systems just mount another filesystem ove
85 amount of space an empty instance of ramfs tak
86
87 If CONFIG_TMPFS is enabled, rootfs will use tm
88 default. To force ramfs, add "rootfstype=ramf
89 line.
90
91 What is initramfs?
92 ------------------
93
94 All 2.6 Linux kernels contain a gzipped "cpio"
95 extracted into rootfs when the kernel boots up
96 checks to see if rootfs contains a file "init"
97 1. If found, this init process is responsible
98 rest of the way up, including locating and mou
99 any). If rootfs does not contain an init prog
100 archive is extracted into it, the kernel will
101 to locate and mount a root partition, then exe
102 out of that.
103
104 All this differs from the old initrd in severa
105
106 - The old initrd was always a separate file,
107 linked into the linux kernel image. (The
108 devoted to generating this archive during
109
110 - The old initrd file was a gzipped filesyst
111 such as ext2, that needed a driver built i
112 initramfs archive is a gzipped cpio archiv
113 see cpio(1) and Documentation/driver-api/e
114 The kernel's cpio extraction code is not o
115 __init text and data that can be discarded
116
117 - The program run by the old initrd (which w
118 some setup and then returned to the kernel
119 initramfs is not expected to return to the
120 off control it can overmount / with a new
121 program. See the switch_root utility, bel
122
123 - When switching another root device, initrd
124 umount the ramdisk. But initramfs is root
125 rootfs, nor unmount it. Instead delete ev
126 free up the space (find -xdev / -exec rm '
127 with the new root (cd /newmount; mount --m
128 stdin/stdout/stderr to the new /dev/consol
129
130 Since this is a remarkably persnickety pro
131 commands before you can run them), the kli
132 program (utils/run_init.c) to do all this
133 (such as busybox) have named this command
134
135 Populating initramfs:
136 ---------------------
137
138 The 2.6 kernel build process always creates a
139 archive and links it into the resulting kernel
140 archive is empty (consuming 134 bytes on x86).
141
142 The config option CONFIG_INITRAMFS_SOURCE (in
143 and living in usr/Kconfig) can be used to spec
144 initramfs archive, which will automatically be
145 resulting binary. This option can point to an
146 archive, a directory containing files to be ar
147 specification such as the following example::
148
149 dir /dev 755 0 0
150 nod /dev/console 644 0 0 c 5 1
151 nod /dev/loop0 644 0 0 b 7 0
152 dir /bin 755 1000 1000
153 slink /bin/sh busybox 777 0 0
154 file /bin/busybox initramfs/busybox 755 0 0
155 dir /proc 755 0 0
156 dir /sys 755 0 0
157 dir /mnt 755 0 0
158 file /init initramfs/init.sh 755 0 0
159
160 Run "usr/gen_init_cpio" (after the kernel buil
161 documenting the above file format.
162
163 One advantage of the configuration file is tha
164 set permissions or create device nodes in the
165 two example "file" entries expect to find file
166 a directory called "initramfs", under the linu
167 Documentation/driver-api/early-userspace/early
168
169 The kernel does not depend on external cpio to
170 directory instead of a configuration file, the
171 creates a configuration file from that directo
172 usr/gen_initramfs.sh), and proceeds to package
173 using the config file (by feeding it to usr/ge
174 from usr/gen_init_cpio.c). The kernel's build
175 entirely self-contained, and the kernel's boot
176 (obviously) self-contained.
177
178 The one thing you might need external cpio uti
179 or extracting your own preprepared cpio files
180 (instead of a config file or directory).
181
182 The following command line can extract a cpio
183 or by the kernel build) back into its componen
184
185 cpio -i -d -H newc -F initramfs_data.cpio --
186
187 The following shell script can create a prebui
188 use in place of the above config file::
189
190 #!/bin/sh
191
192 # Copyright 2006 Rob Landley <rob@landley.net
193 # Licensed under GPL version 2
194
195 if [ $# -ne 2 ]
196 then
197 echo "usage: mkinitramfs directory imagena
198 exit 1
199 fi
200
201 if [ -d "$1" ]
202 then
203 echo "creating $2 from $1"
204 (cd "$1"; find . | cpio -o -H newc | gzip)
205 else
206 echo "First argument must be a directory"
207 exit 1
208 fi
209
210 .. Note::
211
212 The cpio man page contains some bad advice
213 archive if you follow it. It says "A typic
214 of filenames is with the find command; you
215 option to minimize problems with permission
216 unwritable or not searchable." Don't do th
217 initramfs.cpio.gz images, it won't work. T
218 won't create files in a directory that does
219 entries must go before the files that go in
220 The above script gets them in the right ord
221
222 External initramfs images:
223 --------------------------
224
225 If the kernel has initrd support enabled, an e
226 be passed into a 2.6 kernel in place of an ini
227 will autodetect the type (initramfs, not initr
228 archive into rootfs before trying to run /init
229
230 This has the memory efficiency advantages of i
231 device) but the separate packaging of initrd (
232 non-GPL code you'd like to run from initramfs,
233 the GPL licensed Linux kernel binary).
234
235 It can also be used to supplement the kernel's
236 files in the external archive will overwrite a
237 the built-in initramfs archive. Some distribu
238 a single kernel image with task-specific initr
239
240 Contents of initramfs:
241 ----------------------
242
243 An initramfs archive is a complete self-contai
244 If you don't already understand what shared li
245 you need to get a minimal root filesystem up a
246 references:
247
248 - https://www.tldp.org/HOWTO/Bootdisk-HOWTO/
249 - https://www.tldp.org/HOWTO/From-PowerUp-To-B
250 - http://www.linuxfromscratch.org/lfs/view/sta
251
252 The "klibc" package (https://www.kernel.org/pu
253 designed to be a tiny C library to statically
254 code against, along with some related utilitie
255
256 I use uClibc (https://www.uclibc.org) and busy
257 myself. These are LGPL and GPL, respectively.
258 package is planned for the busybox 1.3 release
259
260 In theory you could use glibc, but that's not
261 uses like this. (A "hello world" program stat
262 over 400k. With uClibc it's 7k. Also note th
263 name lookups, even when otherwise statically l
264
265 A good first step is to get initramfs to run a
266 program as init, and test it under an emulator
267 User Mode Linux, like so::
268
269 cat > hello.c << EOF
270 #include <stdio.h>
271 #include <unistd.h>
272
273 int main(int argc, char *argv[])
274 {
275 printf("Hello world!\n");
276 sleep(999999999);
277 }
278 EOF
279 gcc -static hello.c -o init
280 echo init | cpio -o -H newc | gzip > test.cp
281 # Testing external initramfs using the initr
282 qemu -kernel /boot/vmlinuz -initrd test.cpio
283
284 When debugging a normal root filesystem, it's
285 "init=/bin/sh". The initramfs equivalent is "
286 just as useful.
287
288 Why cpio rather than tar?
289 -------------------------
290
291 This decision was made back in December, 2001.
292
293 http://www.uwsg.iu.edu/hypermail/linux/kerne
294
295 And spawned a second thread (specifically on t
296
297 http://www.uwsg.iu.edu/hypermail/linux/kerne
298
299 The quick and dirty summary version (which is
300 the above threads) is:
301
302 1) cpio is a standard. It's decades old (from
303 widely used on Linux (inside RPM, Red Hat's
304 a Linux Journal article about it from 1996:
305
306 http://www.linuxjournal.com/article/1213
307
308 It's not as popular as tar because the trad
309 require _truly_hideous_ command line argume
310 either way about the archive format, and th
311 such as:
312
313 http://freecode.com/projects/afio
314
315 2) The cpio archive format chosen by the kerne
316 thus easier to create and parse) than any o
317 various tar archive formats. The complete
318 explained in buffer-format.txt, created in
319 extracted in init/initramfs.c. All three t
320 total of human-readable text.
321
322 3) The GNU project standardizing on tar is app
323 Windows standardizing on zip. Linux is not
324 to make its own technical decisions.
325
326 4) Since this is a kernel internal format, it
327 something brand new. The kernel provides i
328 extract this format anyway. Using an exist
329 but not essential.
330
331 5) Al Viro made the decision (quote: "tar is u
332 supported on the kernel side"):
333
334 http://www.uwsg.iu.edu/hypermail/linux/k
335
336 explained his reasoning:
337
338 - http://www.uwsg.iu.edu/hypermail/linux/
339 - http://www.uwsg.iu.edu/hypermail/linux/
340
341 and, most importantly, designed and impleme
342
343 Future directions:
344 ------------------
345
346 Today (2.6.16), initramfs is always compiled i
347 kernel falls back to legacy boot code that is
348 not contain an /init program. The fallback is
349 smooth transition and allowing early boot func
350 "early userspace" (I.E. initramfs).
351
352 The move to early userspace is necessary becau
353 root device is complex. Root partitions can s
354 separate journal). They can be out on the net
355 specific MAC address, logging into a server, e
356 media, with dynamically allocated major/minor
357 issues requiring a full udev implementation to
358 compressed, encrypted, copy-on-write, loopback
359 and so on.
360
361 This kind of complexity (which inevitably incl
362 in userspace. Both klibc and busybox/uClibc a
363 packages to drop into a kernel build.
364
365 The klibc package has now been accepted into A
366 The kernel's current early boot code (partitio
367 be migrated into a default initramfs, automati
368 kernel build.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.