/dev/initrd is a read-only block device assigned major number 1 and minor number 250. Typically /dev/initrd is owned by root:disk with mode 0400 (read access by root only). If the Linux system does not have /dev/initrd already created, it can be created with the following commands:
mknod -m 400 /dev/initrd b 1 250 chown root:disk /dev/initrd
Also, support for both "RAM disk" and "Initial RAM disk" (e.g., CONFIG_BLK_DEV_RAM=y and CONFIG_BLK_DEV_INITRD=y) must be compiled directly into the Linux kernel to use /dev/initrd. When using /dev/initrd, the RAM disk driver cannot be loaded as a module.
The special file /dev/initrd is a read-only block device. This device is a RAM disk that is initialized (e.g., loaded) by the boot loader before the kernel is started. The kernel then can use /dev/initrd's contents for a two-phase system boot-up.
In the first boot-up phase, the kernel starts up and mounts an initial root filesystem from the contents of /dev/initrd (e.g., RAM disk initialized by the boot loader). In the second phase, additional drivers or other modules are loaded from the initial root device's contents. After loading the additional modules, a new root filesystem (i.e., the normal root filesystem) is mounted from a different device.
When booting up with initrd, the system boots as follows:
The boot loader loads the kernel program and /dev/initrd's contents into memory.
On kernel startup, the kernel uncompresses and copies the contents of the device /dev/initrd onto device /dev/ram0 and then frees the memory used by /dev/initrd.
The kernel then read-write mounts the device /dev/ram0 as the initial root filesystem.
If the indicated normal root filesystem is also the initial root filesystem (e.g., /dev/ram0) then the kernel skips to the last step for the usual boot sequence.
If the executable file /linuxrc is present in the initial root filesystem, /linuxrc is executed with UID 0. (The file /linuxrc must have executable permission. The file /linuxrc can be any valid executable, including a shell script.)
If /linuxrc is not executed or when /linuxrc terminates, the normal root filesystem is mounted. (If /linuxrc exits with any filesystems mounted on the initial root filesystem, then the behavior of the kernel is UNSPECIFIED. See the Notes section for the current kernel behavior.)
If the normal root filesystem has a directory /initrd, the device /dev/ram0 is moved from / to /initrd. Otherwise, if the directory /initrd does not exist, the device /dev/ram0 is unmounted. (When moved from / to /initrd, /dev/ram0 is not unmounted and therefore processes can remain running from /dev/ram0. If directory /initrd does not exist on the normal root filesystem and any processes remain running from /dev/ram0 when /linuxrc exits, the behavior of the kernel is UNSPECIFIED. See the Notes section for the current kernel behavior.)
The usual boot sequence (e.g., invocation of /sbin/init) is performed on the normal root filesystem.
The following boot loader options, when used with initrd, affect the kernel's boot-up operation:
Specifies the file to load as the contents of /dev/initrd. For LOADLIN this is a command-line option. For LILO you have to use this command in the LILO configuration file /etc/lilo.config. The filename specified with this option will typically be a gzipped filesystem image.
This boot option disables the two-phase boot-up operation. The kernel performs the usual boot sequence as if /dev/initrd was not initialized. With this option, any contents of /dev/initrd loaded into memory by the boot loader contents are preserved. This option permits the contents of /dev/initrd to be any data and need not be limited to a filesystem image. However, device /dev/initrd is read-only and can be read only one time after system startup.
Specifies the device to be used as the normal root filesystem. For LOADLIN this is a command-line option. For LILO this is a boot time option or can be used as an option line in the LILO configuration file /etc/lilo.config. The device specified by this option must be a mountable device having a suitable root filesystem.
Changing the normal root filesystem
By default, the kernel's settings (e.g., set in the kernel file with rdev(8) or compiled into the kernel file), or the boot loader option setting is used for the normal root filesystems. For an NFS-mounted normal root filesystem, one has to use the nfs_root_name and nfs_root_addrs boot options to give the NFS settings. For more information on NFS-mounted root see the kernel documentation file Documentation/filesystems/nfs/nfsroot.txt (or Documentation/filesystems/nfsroot.txt before Linux 2.6.33). For more information on setting the root filesystem see also the LILO and LOADLIN documentation.
It is also possible for the /linuxrc executable to change the normal root device. For /linuxrc to change the normal root device, /proc must be mounted. After mounting /proc, /linuxrc changes the normal root device by writing into the proc files /proc/sys/kernel/real-root-dev, /proc/sys/kernel/nfs-root-name, and /proc/sys/kernel/nfs-root-addrs. For a physical root device, the root device is changed by having /linuxrc write the new root filesystem device number into /proc/sys/kernel/real-root-dev. For an NFS root filesystem, the root device is changed by having /linuxrc write the NFS setting into files /proc/sys/kernel/nfs-root-name and /proc/sys/kernel/nfs-root-addrs and then writing 0xff (e.g., the pseudo-NFS-device number) into file /proc/sys/kernel/real-root-dev. For example, the following shell command line would change the normal root device to /dev/hdb1:
echo 0x365 >/proc/sys/kernel/real-root-dev
For an NFS example, the following shell command lines would change the normal root device to the NFS directory /var/nfsroot on a local networked NFS server with IP number 188.8.131.52 for a system with IP number 184.108.40.206 and named "idefix":
echo /var/nfsroot >/proc/sys/kernel/nfs-root-name echo 220.127.116.11:18.104.22.168::255.255.255.0:idefix \ >/proc/sys/kernel/nfs-root-addrs echo 255 >/proc/sys/kernel/real-root-dev
Note: The use of /proc/sys/kernel/real-root-dev to change the root filesystem is obsolete. See the Linux kernel source file Documentation/admin-guide/initrd.rst (or Documentation/initrd.txt before Linux 4.10) as well as pivot_root(2) and pivot_root(8) for information on the modern method of changing the root filesystem.
The main motivation for implementing initrd was to allow for modular kernel configuration at system installation.
A possible system installation scenario is as follows:
The loader program boots from floppy or other media with a minimal kernel (e.g., support for /dev/ram, /dev/initrd, and the ext2 filesystem) and loads /dev/initrd with a gzipped version of the initial filesystem.
The executable /linuxrc determines what is needed to (1) mount the normal root filesystem (i.e., device type, device drivers, filesystem) and (2) the distribution media (e.g., CD-ROM, network, tape, ...). This can be done by asking the user, by auto-probing, or by using a hybrid approach.
The executable /linuxrc loads the necessary modules from the initial root filesystem.
The executable /linuxrc creates and populates the root filesystem. (At this stage the normal root filesystem does not have to be a completed system yet.)
The executable /linuxrc sets /proc/sys/kernel/real-root-dev, unmounts /proc, the normal root filesystem and any other filesystems it has mounted, and then terminates.
The kernel then mounts the normal root filesystem.
Now that the filesystem is accessible and intact, the boot loader can be installed.
The boot loader is configured to load into /dev/initrd a filesystem with the set of modules that was used to bring up the system. (e.g., device /dev/ram0 can be modified, then unmounted, and finally, the image is written from /dev/ram0 to a file.)
The system is now bootable and additional installation tasks can be performed.
The key role of /dev/initrd in the above is to reuse the configuration data during normal system operation without requiring initial kernel selection, a large generic kernel or, recompiling the kernel.
A second scenario is for installations where Linux runs on systems with different hardware configurations in a single administrative network. In such cases, it may be desirable to use only a small set of kernels (ideally only one) and to keep the system-specific part of configuration information as small as possible. In this case, create a common file with all needed modules. Then, only the /linuxrc file or a file executed by /linuxrc would be different.
A third scenario is more convenient recovery disks. Because information like the location of the root filesystem partition is not needed at boot time, the system loaded from /dev/initrd can use a dialog and/or auto-detection followed by a possible sanity check.
Last but not least, Linux distributions on CD-ROM may use initrd for easy installation from the CD-ROM. The distribution can use LOADLIN to directly load /dev/initrd from CD-ROM without the need of any floppies. The distribution could also use a LILO boot floppy and then bootstrap a bigger RAM disk via /dev/initrd from the CD-ROM.
- With the current kernel, any filesystems that remain mounted when /dev/ram0 is moved from / to /initrd continue to be accessible. However, the /proc/mounts entries are not updated.
- With the current kernel, if directory /initrd does not exist, then /dev/ram0 will not be fully unmounted if /dev/ram0 is used by any process or has any filesystem mounted on it. If /dev/ram0 is not fully unmounted, then /dev/ram0 will remain in memory.
- Users of /dev/initrd should not depend on the behavior given in the above notes. The behavior may change in future versions of the Linux kernel.
chown(1), mknod(1), ram(4), freeramdisk(8), rdev(8)
Documentation/admin-guide/initrd.rst (or Documentation/initrd.txt before Linux 4.10) in the Linux kernel source tree, the LILO documentation, the LOADLIN documentation, the SYSLINUX documentation