ramdisk和initramfs比较

Linux内核下的关于ramdisk核ramfs的文档Linux Kernel Documentation  filesystems  ramfs-rootfs-initramfs.txt

Linux Kernel Documentation :: filesystems : ramfs-rootfs-initramfs.txtAbout
Kernel Documentation Linux Kernel Contact Linux Resources Linux
BlogDocumentation / filesystems / ramfs-rootfs-initramfs.txt

 

    
 Custom Search
Based on kernel version 2.6.34. Page generated on 2010-05-31 16:02 EST.
1 ramfs, rootfs and initramfs
2 October 17, 2005
3 Rob Landley
4 =============================
5 
6 What is ramfs?
7 --------------
8 
9 Ramfs is a very simple filesystem that exports Linux's disk caching
10 mechanisms (the page cache and dentry cache) as a dynamically resizable
11 RAM-based filesystem.
12 
13 Normally all files are cached in memory by Linux.  Pages of data read from
14 backing store (usually the block device the filesystem is mounted on) are kept
15 around in case it's needed again, but marked as clean (freeable) in case the
16 Virtual Memory system needs the memory for something else.  Similarly, data
17 written to files is marked clean as soon as it has been written to backing
18 store, but kept around for caching purposes until the VM reallocates the
19 memory.  A similar mechanism (the dentry cache) greatly speeds up access to
20 directories.
21 
22 With ramfs, there is no backing store.  Files written into ramfs allocate
23 dentries and page cache as usual, but there's nowhere to write them to.
24 This means the pages are never marked clean, so they can't be freed by the
25 VM when it's looking to recycle memory.
26 
27 The amount of code required to implement ramfs is tiny, because all the
28 work is done by the existing Linux caching infrastructure.  Basically,
29 you're mounting the disk cache as a filesystem.  Because of this, ramfs is not
30 an optional component removable via menuconfig, since there would be negligible
31 space savings.
32 
33 ramfs and ramdisk:
34 ------------------
35 
36 The older "ram disk" mechanism created a synthetic block device out of
37 an area of RAM and used it as backing store for a filesystem.  This block
38 device was of fixed size, so the filesystem mounted on it was of fixed
39 size.  Using a ram disk also required unnecessarily copying memory from the
40 fake block device into the page cache (and copying changes back out), as well
41 as creating and destroying dentries.  Plus it needed a filesystem driver
42 (such as ext2) to format and interpret this data.
43 
44 Compared to ramfs, this wastes memory (and memory bus bandwidth), creates
45 unnecessary work for the CPU, and pollutes the CPU caches.  (There are tricks
46 to avoid this copying by playing with the page tables, but they're unpleasantly
47 complicated and turn out to be about as expensive as the copying anyway.)
48 More to the point, all the work ramfs is doing has to happen _anyway_,
49 since all file access goes through the page and dentry caches.  The RAM
50 disk is simply unnecessary; ramfs is internally much simpler.
51 
52 Another reason ramdisks are semi-obsolete is that the introduction of
53 loopback devices offered a more flexible and convenient way to create
54 synthetic block devices, now from files instead of from chunks of memory.
55 See losetup (8) for details.
56 
57 ramfs and tmpfs:
58 ----------------
59 
60 One downside of ramfs is you can keep writing data into it until you fill
61 up all memory, and the VM can't free it because the VM thinks that files
62 should get written to backing store (rather than swap space), but ramfs hasn't
63 got any backing store.  Because of this, only root (or a trusted user) should
64 be allowed write access to a ramfs mount.
65 
66 A ramfs derivative called tmpfs was created to add size limits, and the ability
67 to write the data to swap space.  Normal users can be allowed write access to
68 tmpfs mounts.  See Documentation/filesystems/tmpfs.txt for more information.
69 
70 What is rootfs?
71 ---------------
72 
73 Rootfs is a special instance of ramfs (or tmpfs, if that's enabled), which is
74 always present in 2.6 systems.  You can't unmount rootfs for approximately the
75 same reason you can't kill the init process; rather than having special code
76 to check for and handle an empty list, it's smaller and simpler for the kernel
77 to just make sure certain lists can't become empty.
78 
79 Most systems just mount another filesystem over rootfs and ignore it.  The
80 amount of space an empty instance of ramfs takes up is tiny.
81 
82 What is initramfs?
83 ------------------
84 
85 All 2.6 Linux kernels contain a gzipped "cpio" format archive, which is
86 extracted into rootfs when the kernel boots up.  After extracting, the kernel
87 checks to see if rootfs contains a file "init", and if so it executes it as PID
88 1.  If found, this init process is responsible for bringing the system the
89 rest of the way up, including locating and mounting the real root device (if
90 any).  If rootfs does not contain an init program after the embedded cpio
91 archive is extracted into it, the kernel will fall through to the older code
92 to locate and mount a root partition, then exec some variant of /sbin/init
93 out of that.
94 
95 All this differs from the old initrd in several ways:
96 
97   - The old initrd was always a separate file, while the initramfs archive is
98     linked into the linux kernel image.  (The directory linux-*/usr is devoted
99     to generating this archive during the build.)
100 
101   - The old initrd file was a gzipped filesystem image (in some file format,
102     such as ext2, that needed a driver built into the kernel), while the new
103     initramfs archive is a gzipped cpio archive (like tar only simpler,
104     see cpio(1) and Documentation/early-userspace/buffer-format.txt).  The
105     kernel's cpio extraction code is not only extremely small, it's also
106     __init text and data that can be discarded during the boot process.
107 
108   - The program run by the old initrd (which was called /initrd, not /init) did
109     some setup and then returned to the kernel, while the init program from
110     initramfs is not expected to return to the kernel.  (If /init needs to hand
111     off control it can overmount / with a new root device and exec another init
112     program.  See the switch_root utility, below.)
113 
114   - When switching another root device, initrd would pivot_root and then
115     umount the ramdisk.  But initramfs is rootfs: you can neither pivot_root
116     rootfs, nor unmount it.  Instead delete everything out of rootfs to
117     free up the space (find -xdev / -exec rm '{}' ';'), overmount rootfs
118     with the new root (cd /newmount; mount --move . /; chroot .), attach
119     stdin/stdout/stderr to the new /dev/console, and exec the new init.
120 
121     Since this is a remarkably persnickety process (and involves deleting
122     commands before you can run them), the klibc package introduced a helper
123     program (utils/run_init.c) to do all this for you.  Most other packages
124     (such as busybox) have named this command "switch_root".
125 
126 Populating initramfs:
127 ---------------------
128 
129 The 2.6 kernel build process always creates a gzipped cpio format initramfs
130 archive and links it into the resulting kernel binary.  By default, this
131 archive is empty (consuming 134 bytes on x86).
132 
133 The config option CONFIG_INITRAMFS_SOURCE (in General Setup in menuconfig,
134 and living in usr/Kconfig) can be used to specify a source for the
135 initramfs archive, which will automatically be incorporated into the
136 resulting binary.  This option can point to an existing gzipped cpio
137 archive, a directory containing files to be archived, or a text file
138 specification such as the following example:
139 
140   dir /dev 755 0 0
141   nod /dev/console 644 0 0 c 5 1
142   nod /dev/loop0 644 0 0 b 7 0
143   dir /bin 755 1000 1000
144   slink /bin/sh busybox 777 0 0
145   file /bin/busybox initramfs/busybox 755 0 0
146   dir /proc 755 0 0
147   dir /sys 755 0 0
148   dir /mnt 755 0 0
149   file /init initramfs/init.sh 755 0 0
150 
151 Run "usr/gen_init_cpio" (after the kernel build) to get a usage message
152 documenting the above file format.
153 
154 One advantage of the configuration file is that root access is not required to
155 set permissions or create device nodes in the new archive.  (Note that those
156 two example "file" entries expect to find files named "init.sh" and "busybox" in
157 a directory called "initramfs", under the linux-2.6.* directory.  See
158 Documentation/early-userspace/README for more details.)
159 
160 The kernel does not depend on external cpio tools.  If you specify a
161 directory instead of a configuration file, the kernel's build infrastructure
162 creates a configuration file from that directory (usr/Makefile calls
163 scripts/gen_initramfs_list.sh), and proceeds to package up that directory
164 using the config file (by feeding it to usr/gen_init_cpio, which is created
165 from usr/gen_init_cpio.c).  The kernel's build-time cpio creation code is
166 entirely self-contained, and the kernel's boot-time extractor is also
167 (obviously) self-contained.
168 
169 The one thing you might need external cpio utilities installed for is creating
170 or extracting your own preprepared cpio files to feed to the kernel build
171 (instead of a config file or directory).
172 
173 The following command line can extract a cpio image (either by the above script
174 or by the kernel build) back into its component files:
175 
176   cpio -i -d -H newc -F initramfs_data.cpio --no-absolute-filenames
177 
178 The following shell script can create a prebuilt cpio archive you can
179 use in place of the above config file:
180 
181   #!/bin/sh
182 
183   # Copyright 2006 Rob Landley and TimeSys Corporation[DOT]
184   # Licensed under GPL version 2
185 
186   if [ $# -ne 2 ]
187   then
188     echo "usage: mkinitramfs directory imagename.cpio.gz"
189     exit 1
190   fi
191 
192   if [ -d "$1" ]
193   then
194     echo "creating $2 from $1"
195     (cd "$1"; find . | cpio -o -H newc | gzip) > "$2"
196   else
197     echo "First argument must be a directory"
198     exit 1
199   fi
200 
201 Note: The cpio man page contains some bad advice that will break your initramfs
202 archive if you follow it.  It says "A typical way to generate the list
203 of filenames is with the find command; you should give find the -depth option
204 to minimize problems with permissions on directories that are unwritable or not
205 searchable."  Don't do this when creating initramfs.cpio.gz images, it won't
206 work.  The Linux kernel cpio extractor won't create files in a directory that
207 doesn't exist, so the directory entries must go before the files that go in
208 those directories.  The above script gets them in the right order.
209 
210 External initramfs images:
211 --------------------------
212 
213 If the kernel has initrd support enabled, an external cpio.gz archive can also
214 be passed into a 2.6 kernel in place of an initrd.  In this case, the kernel
215 will autodetect the type (initramfs, not initrd) and extract the external cpio
216 archive into rootfs before trying to run /init.
217 
218 This has the memory efficiency advantages of initramfs (no ramdisk block
219 device) but the separate packaging of initrd (which is nice if you have
220 non-GPL code you'd like to run from initramfs, without conflating it with
221 the GPL licensed Linux kernel binary).
222 
223 It can also be used to supplement the kernel's built-in initramfs image.  The
224 files in the external archive will overwrite any conflicting files in
225 the built-in initramfs archive.  Some distributors also prefer to customize
226 a single kernel image with task-specific initramfs images, without recompiling.
227 
228 Contents of initramfs:
229 ----------------------
230 
231 An initramfs archive is a complete self-contained root filesystem for Linux.
232 If you don't already understand what shared libraries, devices, and paths
233 you need to get a minimal root filesystem up and running, here are some
234 references:
235 
236 
237 
238 
239 The "klibc" package () is
240 designed to be a tiny C library to statically link early userspace
241 code against, along with some related utilities.  It is BSD licensed.
242 
243 I use uClibc () and busybox ()
244 myself.  These are LGPL and GPL, respectively.  (A self-contained initramfs
245 package is planned for the busybox 1.3 release.)
246 
247 In theory you could use glibc, but that's not well suited for small embedded
248 uses like this.  (A "hello world" program statically linked against glibc is
249 over 400k.  With uClibc it's 7k.  Also note that glibc dlopens libnss to do
250 name lookups, even when otherwise statically linked.)
251 
252 A good first step is to get initramfs to run a statically linked "hello world"
253 program as init, and test it under an emulator like qemu () or
254 User Mode Linux, like so:
255 
256   cat > hello.c << EOF
257   #include
258   #include
259 
260   int main(int argc, char *argv[])
261   {
262     printf("Hello world!\n");
263     sleep(999999999);
264   }
265   EOF
266   gcc -static hello.c -o init
267   echo init | cpio -o -H newc | gzip > test.cpio.gz
268   # Testing external initramfs using the initrd loading mechanism.
269   qemu -kernel /boot/vmlinuz -initrd test.cpio.gz /dev/zero
270 
271 When debugging a normal root filesystem, it's nice to be able to boot with
272 "init=/bin/sh".  The initramfs equivalent is "rdinit=/bin/sh", and it's
273 just as useful.
274 
275 Why cpio rather than tar?
276 -------------------------
277 
278 This decision was made back in December, 2001.  The discussion started here:
279 
280  
281 
282 And spawned a second thread (specifically on tar vs cpio), starting here:
283 
284  
285 
286 The quick and dirty summary version (which is no substitute for reading
287 the above threads) is:
288 
289 1) cpio is a standard.  It's decades old (from the AT&T days), and already
290    widely used on Linux (inside RPM, Red Hat's device driver disks).  Here's
291    a Linux Journal article about it from 1996:
292 
293      
294 
295    It's not as popular as tar because the traditional cpio command line tools
296    require _truly_hideous_ command line arguments.  But that says nothing
297    either way about the archive format, and there are alternative tools,
298    such as:
299 
300     
301 
302 2) The cpio archive format chosen by the kernel is simpler and cleaner (and
303    thus easier to create and parse) than any of the (literally dozens of)
304    various tar archive formats.  The complete initramfs archive format is
305    explained in buffer-format.txt, created in usr/gen_init_cpio.c, and
306    extracted in init/initramfs.c.  All three together come to less than 26k
307    total of human-readable text.
308 
309 3) The GNU project standardizing on tar is approximately as relevant as
310    Windows standardizing on zip.  Linux is not part of either, and is free
311    to make its own technical decisions.
312 
313 4) Since this is a kernel internal format, it could easily have been
314    something brand new.  The kernel provides its own tools to create and
315    extract this format anyway.  Using an existing standard was preferable,
316    but not essential.
317 
318 5) Al Viro made the decision (quote: "tar is ugly as hell and not going to be
319    supported on the kernel side"):
320 
321      
322 
323    explained his reasoning:
324 
325      
326      
327 
328    and, most importantly, designed and implemented the initramfs code.
329 
330 Future directions:
331 ------------------
332 
333 Today (2.6.16), initramfs is always compiled in, but not always used.  The
334 kernel falls back to legacy boot code that is reached only if initramfs does
335 not contain an /init program.  The fallback is legacy code, there to ensure a
336 smooth transition and allowing early boot functionality to gradually move to
337 "early userspace" (I.E. initramfs).
338 
339 The move to early userspace is necessary because finding and mounting the real
340 root device is complex.  Root partitions can span multiple devices (raid or
341 separate journal).  They can be out on the network (requiring dhcp, setting a
342 specific MAC address, logging into a server, etc).  They can live on removable
343 media, with dynamically allocated major/minor numbers and persistent naming
344 issues requiring a full udev implementation to sort out.  They can be
345 compressed, encrypted, copy-on-write, loopback mounted, strangely partitioned,
346 and so on.
347 
348 This kind of complexity (which inevitably includes policy) is rightly handled
349 in userspace.  Both klibc and busybox/uClibc are working on simple initramfs
350 packages to drop into a kernel build.
351 
352 The klibc package has now been accepted into Andrew Morton's 2.6.17-mm tree.
353 The kernel's current early boot code (partition detection, etc) will probably
354 be migrated into a default initramfs, automatically created and used by the
355 kernel build.
Hide Line Numbers[ filesystems ]
   00-INDEX
   9p.txt
   adfs.txt
   affs.txt
   afs.txt
   autofs4-mount-control.txt
   automount-support.txt
   befs.txt
   bfs.txt
   btrfs.txt
   [ caching ]
   ceph.txt
   cifs.txt
   coda.txt
   [ configfs ]
   cramfs.txt
   debugfs.txt
   dentry-locking.txt
   devpts.txt
   directory-locking
   dlmfs.txt
   dnotify.txt
   dnotify_test.c
   ecryptfs.txt
   exofs.txt
   ext2.txt
   ext3.txt
   ext4.txt
   fiemap.txt
   files.txt
   fuse.txt
   gfs2-glocks.txt
   gfs2-uevents.txt
   gfs2.txt
   hfs.txt
   hfsplus.txt
   hpfs.txt
   inotify.txt
   isofs.txt
   jfs.txt
   Locking
   locks.txt
   logfs.txt
   Makefile
   mandatory-locking.txt
   ncpfs.txt
   [ nfs ]
   nilfs2.txt
   ntfs.txt
   ocfs2.txt
   omfs.txt
   [ pohmelfs ]
   porting
   proc.txt
   quota.txt
   ramfs-rootfs-initramfs.txt
   relay.txt
   romfs.txt
   seq_file.txt
   sharedsubtree.txt
   smbfs.txt
   spufs.txt
   squashfs.txt
   sysfs-pci.txt
   sysfs.txt
   sysv-fs.txt
   tmpfs.txt
   ubifs.txt
   udf.txt
   ufs.txt
   vfat.txt
   vfs.txt
   xfs.txt
   xip.txt
     About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Information is copyright its respective author. All material is available from
the Linux Kernel Source distributed under a GPL License. This page is provided
as a free service by mjmwired.net.