369 lines · plain
1.. SPDX-License-Identifier: GPL-2.02 3===========================4Ramfs, rootfs and initramfs5===========================6 7October 17, 20058 9:Author: Rob Landley <rob@landley.net>10 11What is ramfs?12--------------13 14Ramfs is a very simple filesystem that exports Linux's disk caching15mechanisms (the page cache and dentry cache) as a dynamically resizable16RAM-based filesystem.17 18Normally all files are cached in memory by Linux. Pages of data read from19backing store (usually the block device the filesystem is mounted on) are kept20around in case it's needed again, but marked as clean (freeable) in case the21Virtual Memory system needs the memory for something else. Similarly, data22written to files is marked clean as soon as it has been written to backing23store, but kept around for caching purposes until the VM reallocates the24memory. A similar mechanism (the dentry cache) greatly speeds up access to25directories.26 27With ramfs, there is no backing store. Files written into ramfs allocate28dentries and page cache as usual, but there's nowhere to write them to.29This means the pages are never marked clean, so they can't be freed by the30VM when it's looking to recycle memory.31 32The amount of code required to implement ramfs is tiny, because all the33work is done by the existing Linux caching infrastructure. Basically,34you're mounting the disk cache as a filesystem. Because of this, ramfs is not35an optional component removable via menuconfig, since there would be negligible36space savings.37 38ramfs and ramdisk:39------------------40 41The older "ram disk" mechanism created a synthetic block device out of42an area of RAM and used it as backing store for a filesystem. This block43device was of fixed size, so the filesystem mounted on it was of fixed44size. Using a ram disk also required unnecessarily copying memory from the45fake block device into the page cache (and copying changes back out), as well46as creating and destroying dentries. Plus it needed a filesystem driver47(such as ext2) to format and interpret this data.48 49Compared to ramfs, this wastes memory (and memory bus bandwidth), creates50unnecessary work for the CPU, and pollutes the CPU caches. (There are tricks51to avoid this copying by playing with the page tables, but they're unpleasantly52complicated and turn out to be about as expensive as the copying anyway.)53More to the point, all the work ramfs is doing has to happen _anyway_,54since all file access goes through the page and dentry caches. The RAM55disk is simply unnecessary; ramfs is internally much simpler.56 57Another reason ramdisks are semi-obsolete is that the introduction of58loopback devices offered a more flexible and convenient way to create59synthetic block devices, now from files instead of from chunks of memory.60See losetup (8) for details.61 62ramfs and tmpfs:63----------------64 65One downside of ramfs is you can keep writing data into it until you fill66up all memory, and the VM can't free it because the VM thinks that files67should get written to backing store (rather than swap space), but ramfs hasn't68got any backing store. Because of this, only root (or a trusted user) should69be allowed write access to a ramfs mount.70 71A ramfs derivative called tmpfs was created to add size limits, and the ability72to write the data to swap space. Normal users can be allowed write access to73tmpfs mounts. See Documentation/filesystems/tmpfs.rst for more information.74 75What is rootfs?76---------------77 78Rootfs is a special instance of ramfs (or tmpfs, if that's enabled), which is79always present in 2.6 systems. You can't unmount rootfs for approximately the80same reason you can't kill the init process; rather than having special code81to check for and handle an empty list, it's smaller and simpler for the kernel82to just make sure certain lists can't become empty.83 84Most systems just mount another filesystem over rootfs and ignore it. The85amount of space an empty instance of ramfs takes up is tiny.86 87If CONFIG_TMPFS is enabled, rootfs will use tmpfs instead of ramfs by88default. To force ramfs, add "rootfstype=ramfs" to the kernel command89line.90 91What is initramfs?92------------------93 94All 2.6 Linux kernels contain a gzipped "cpio" format archive, which is95extracted into rootfs when the kernel boots up. After extracting, the kernel96checks to see if rootfs contains a file "init", and if so it executes it as PID971. If found, this init process is responsible for bringing the system the98rest of the way up, including locating and mounting the real root device (if99any). If rootfs does not contain an init program after the embedded cpio100archive is extracted into it, the kernel will fall through to the older code101to locate and mount a root partition, then exec some variant of /sbin/init102out of that.103 104All this differs from the old initrd in several ways:105 106 - The old initrd was always a separate file, while the initramfs archive is107 linked into the linux kernel image. (The directory ``linux-*/usr`` is108 devoted to generating this archive during the build.)109 110 - The old initrd file was a gzipped filesystem image (in some file format,111 such as ext2, that needed a driver built into the kernel), while the new112 initramfs archive is a gzipped cpio archive (like tar only simpler,113 see cpio(1) and Documentation/driver-api/early-userspace/buffer-format.rst).114 The kernel's cpio extraction code is not only extremely small, it's also115 __init text and data that can be discarded during the boot process.116 117 - The program run by the old initrd (which was called /initrd, not /init) did118 some setup and then returned to the kernel, while the init program from119 initramfs is not expected to return to the kernel. (If /init needs to hand120 off control it can overmount / with a new root device and exec another init121 program. See the switch_root utility, below.)122 123 - When switching another root device, initrd would pivot_root and then124 umount the ramdisk. But initramfs is rootfs: you can neither pivot_root125 rootfs, nor unmount it. Instead delete everything out of rootfs to126 free up the space (find -xdev / -exec rm '{}' ';'), overmount rootfs127 with the new root (cd /newmount; mount --move . /; chroot .), attach128 stdin/stdout/stderr to the new /dev/console, and exec the new init.129 130 Since this is a remarkably persnickety process (and involves deleting131 commands before you can run them), the klibc package introduced a helper132 program (utils/run_init.c) to do all this for you. Most other packages133 (such as busybox) have named this command "switch_root".134 135Populating initramfs:136---------------------137 138The 2.6 kernel build process always creates a gzipped cpio format initramfs139archive and links it into the resulting kernel binary. By default, this140archive is empty (consuming 134 bytes on x86).141 142The config option CONFIG_INITRAMFS_SOURCE (in General Setup in menuconfig,143and living in usr/Kconfig) can be used to specify a source for the144initramfs archive, which will automatically be incorporated into the145resulting binary. This option can point to an existing gzipped cpio146archive, a directory containing files to be archived, or a text file147specification such as the following example::148 149 dir /dev 755 0 0150 nod /dev/console 644 0 0 c 5 1151 nod /dev/loop0 644 0 0 b 7 0152 dir /bin 755 1000 1000153 slink /bin/sh busybox 777 0 0154 file /bin/busybox initramfs/busybox 755 0 0155 dir /proc 755 0 0156 dir /sys 755 0 0157 dir /mnt 755 0 0158 file /init initramfs/init.sh 755 0 0159 160Run "usr/gen_init_cpio" (after the kernel build) to get a usage message161documenting the above file format.162 163One advantage of the configuration file is that root access is not required to164set permissions or create device nodes in the new archive. (Note that those165two example "file" entries expect to find files named "init.sh" and "busybox" in166a directory called "initramfs", under the linux-2.6.* directory. See167Documentation/driver-api/early-userspace/early_userspace_support.rst for more details.)168 169The kernel does not depend on external cpio tools. If you specify a170directory instead of a configuration file, the kernel's build infrastructure171creates a configuration file from that directory (usr/Makefile calls172usr/gen_initramfs.sh), and proceeds to package up that directory173using the config file (by feeding it to usr/gen_init_cpio, which is created174from usr/gen_init_cpio.c). The kernel's build-time cpio creation code is175entirely self-contained, and the kernel's boot-time extractor is also176(obviously) self-contained.177 178The one thing you might need external cpio utilities installed for is creating179or extracting your own preprepared cpio files to feed to the kernel build180(instead of a config file or directory).181 182The following command line can extract a cpio image (either by the above script183or by the kernel build) back into its component files::184 185 cpio -i -d -H newc -F initramfs_data.cpio --no-absolute-filenames186 187The following shell script can create a prebuilt cpio archive you can188use in place of the above config file::189 190 #!/bin/sh191 192 # Copyright 2006 Rob Landley <rob@landley.net> and TimeSys Corporation.193 # Licensed under GPL version 2194 195 if [ $# -ne 2 ]196 then197 echo "usage: mkinitramfs directory imagename.cpio.gz"198 exit 1199 fi200 201 if [ -d "$1" ]202 then203 echo "creating $2 from $1"204 (cd "$1"; find . | cpio -o -H newc | gzip) > "$2"205 else206 echo "First argument must be a directory"207 exit 1208 fi209 210.. Note::211 212 The cpio man page contains some bad advice that will break your initramfs213 archive if you follow it. It says "A typical way to generate the list214 of filenames is with the find command; you should give find the -depth215 option to minimize problems with permissions on directories that are216 unwritable or not searchable." Don't do this when creating217 initramfs.cpio.gz images, it won't work. The Linux kernel cpio extractor218 won't create files in a directory that doesn't exist, so the directory219 entries must go before the files that go in those directories.220 The above script gets them in the right order.221 222External initramfs images:223--------------------------224 225If the kernel has initrd support enabled, an external cpio.gz archive can also226be passed into a 2.6 kernel in place of an initrd. In this case, the kernel227will autodetect the type (initramfs, not initrd) and extract the external cpio228archive into rootfs before trying to run /init.229 230This has the memory efficiency advantages of initramfs (no ramdisk block231device) but the separate packaging of initrd (which is nice if you have232non-GPL code you'd like to run from initramfs, without conflating it with233the GPL licensed Linux kernel binary).234 235It can also be used to supplement the kernel's built-in initramfs image. The236files in the external archive will overwrite any conflicting files in237the built-in initramfs archive. Some distributors also prefer to customize238a single kernel image with task-specific initramfs images, without recompiling.239 240Contents of initramfs:241----------------------242 243An initramfs archive is a complete self-contained root filesystem for Linux.244If you don't already understand what shared libraries, devices, and paths245you need to get a minimal root filesystem up and running, here are some246references:247 248- https://www.tldp.org/HOWTO/Bootdisk-HOWTO/249- https://www.tldp.org/HOWTO/From-PowerUp-To-Bash-Prompt-HOWTO.html250- http://www.linuxfromscratch.org/lfs/view/stable/251 252The "klibc" package (https://www.kernel.org/pub/linux/libs/klibc) is253designed to be a tiny C library to statically link early userspace254code against, along with some related utilities. It is BSD licensed.255 256I use uClibc (https://www.uclibc.org) and busybox (https://www.busybox.net)257myself. These are LGPL and GPL, respectively. (A self-contained initramfs258package is planned for the busybox 1.3 release.)259 260In theory you could use glibc, but that's not well suited for small embedded261uses like this. (A "hello world" program statically linked against glibc is262over 400k. With uClibc it's 7k. Also note that glibc dlopens libnss to do263name lookups, even when otherwise statically linked.)264 265A good first step is to get initramfs to run a statically linked "hello world"266program as init, and test it under an emulator like qemu (www.qemu.org) or267User Mode Linux, like so::268 269 cat > hello.c << EOF270 #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 EOF279 gcc -static hello.c -o init280 echo init | cpio -o -H newc | gzip > test.cpio.gz281 # Testing external initramfs using the initrd loading mechanism.282 qemu -kernel /boot/vmlinuz -initrd test.cpio.gz /dev/zero283 284When debugging a normal root filesystem, it's nice to be able to boot with285"init=/bin/sh". The initramfs equivalent is "rdinit=/bin/sh", and it's286just as useful.287 288Why cpio rather than tar?289-------------------------290 291This decision was made back in December, 2001. The discussion started here:292 293 http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.2/1538.html294 295And spawned a second thread (specifically on tar vs cpio), starting here:296 297 http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.2/1587.html298 299The quick and dirty summary version (which is no substitute for reading300the above threads) is:301 3021) cpio is a standard. It's decades old (from the AT&T days), and already303 widely used on Linux (inside RPM, Red Hat's device driver disks). Here's304 a Linux Journal article about it from 1996:305 306 http://www.linuxjournal.com/article/1213307 308 It's not as popular as tar because the traditional cpio command line tools309 require _truly_hideous_ command line arguments. But that says nothing310 either way about the archive format, and there are alternative tools,311 such as:312 313 http://freecode.com/projects/afio314 3152) The cpio archive format chosen by the kernel is simpler and cleaner (and316 thus easier to create and parse) than any of the (literally dozens of)317 various tar archive formats. The complete initramfs archive format is318 explained in buffer-format.txt, created in usr/gen_init_cpio.c, and319 extracted in init/initramfs.c. All three together come to less than 26k320 total of human-readable text.321 3223) The GNU project standardizing on tar is approximately as relevant as323 Windows standardizing on zip. Linux is not part of either, and is free324 to make its own technical decisions.325 3264) Since this is a kernel internal format, it could easily have been327 something brand new. The kernel provides its own tools to create and328 extract this format anyway. Using an existing standard was preferable,329 but not essential.330 3315) Al Viro made the decision (quote: "tar is ugly as hell and not going to be332 supported on the kernel side"):333 334 http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.2/1540.html335 336 explained his reasoning:337 338 - http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.2/1550.html339 - http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.2/1638.html340 341 and, most importantly, designed and implemented the initramfs code.342 343Future directions:344------------------345 346Today (2.6.16), initramfs is always compiled in, but not always used. The347kernel falls back to legacy boot code that is reached only if initramfs does348not contain an /init program. The fallback is legacy code, there to ensure a349smooth transition and allowing early boot functionality to gradually move to350"early userspace" (I.E. initramfs).351 352The move to early userspace is necessary because finding and mounting the real353root device is complex. Root partitions can span multiple devices (raid or354separate journal). They can be out on the network (requiring dhcp, setting a355specific MAC address, logging into a server, etc). They can live on removable356media, with dynamically allocated major/minor numbers and persistent naming357issues requiring a full udev implementation to sort out. They can be358compressed, encrypted, copy-on-write, loopback mounted, strangely partitioned,359and so on.360 361This kind of complexity (which inevitably includes policy) is rightly handled362in userspace. Both klibc and busybox/uClibc are working on simple initramfs363packages to drop into a kernel build.364 365The klibc package has now been accepted into Andrew Morton's 2.6.17-mm tree.366The kernel's current early boot code (partition detection, etc) will probably367be migrated into a default initramfs, automatically created and used by the368kernel build.369