The Boot Loader Specification

This document defines a set of file formats and naming conventions that allow the boot loader configuration to be shared between multiple operating systems and boot loaders installed on one device.

Operating systems cooperatively manage a boot loader configuration directory that contains drop-in files, making multi-boot scenarios easy to support. Boot menu items are defined via a simple format that can be understood by different boot loader implementations, operating systems, and userspace programs. The same scheme can be used to prepare OS media for cases where the firmware includes a boot loader.

Target audience

The target audience for this specification is:

The boot partition

Everything described below is located on one or two partitions. The boot loader or user-space programs reading the boot loader configuration should locate them in the following manner:

In the text below, $BOOT will be used to refer to (the root of) the first of the two partitions (the boot partition on MBR disks and the ESP on GPT disks), and $XBOOTLDR will be used to refer to (the root of) the optional second partition.

An installer for the operating system should use this logic when selecting or creating partitions:

Those file systems shall be determined during installation time, and an fstab entry may be created. If only one partition is used, it should be mounted on /boot/. If both XBOOTLDR partition and the ESP are used, they should be mounted on /boot and /efi, or on /boot and /boot/efi.

Note: Those file systems are shared among all OS installations on the system. Instead of maintaining one boot partition per installed OS (as /boot/ was traditionally handled), all installed OSes use the same place for boot-time configuration.

For systems where the firmware is able to read file systems directly, the ESP must — and the XBOOTLDR partition should — be a file system readable by the firmware. For most systems this means VFAT (16 or 32 bit). Applications accessing both partitions should hence not assume that fancier file system features such as symlinks, hardlinks, access control or case sensitivity are supported.

Boot loader entries

This specification defines two types of boot loader entries. The first type is text based, very simple, and suitable for a variety of firmware, architecture and image types (“Type #1”). The second type is specific to EFI, but allows single-file images that embed all metadata in the kernel binary itself, which is useful to cryptographically sign them as one file for the purpose of SecureBoot (“Type #2”).

Not all boot loader entries will apply to all systems. For example, Type #1 entries that use the efi key and all Type #2 entries only apply to EFI systems. Entries using the architecture key might specify an architecture that doesn’t match the local one. Boot loaders should ignore all entries that don’t match the local platform and what the boot loader can support, and hide them from the user. Only entries matching the feature set of boot loader and system shall be considered and displayed. This allows image builders to put together images that transparently support multiple different architectures.

Note that the boot partitions are not supposed to be the exclusive territory of this specification. This specification only defines semantics of the /loader/ directory inside the file system (see below), but it doesn’t intend to define ownership of the whole file system. Boot loaders, firmware, and other software implementing this specification may choose to place other files and directories in the same file system. For example, boot loaders that implement this specification might install their own boot code on the same partition; this is particularly common in the case of the ESP. Implementations of this specification must be able to operate correctly if files or directories other than /loader/ are found in the top level directory. Implementations that add their own files or directories to the file systems should use well-named directories, to make name collisions between multiple users of the file system unlikely.

Type #1 Boot Loader Specification Entries

$ESP/loader/ is the main directory containing the configuration for the boot loader.

Note: In all cases the /loader/ directory should be located directly in the root of the file system. Specifically, the /loader/ directory should not be located under the /EFI/ subdirectory on the ESP.

$BOOT/loader/entries/ and $XBOOTLDR/loader/entries/ are the directories containing the drop-in snippets defining boot entries, one .conf file for each boot menu item. Each OS may provide one or more such entries. The boot loader should enumerate both directories and provide a merged list.

The file name is used for identification of the boot item but shall never be presented to the user in the UI. The file name may be chosen freely but should be unique enough to avoid clashes between OS installations. More specifically, it is suggested to include the entry-token (see kernel-install) or machine ID (see /etc/machine-id), and the kernel version (as returned by uname -r, including the OS identifier), so that the whole filename is $BOOT/loader/entries/<entry-token-or-machine-id>-<version>.conf.

Example: $BOOT/loader/entries/6a9857a393724b7a981ebb5b8495b9ea-3.8.0-2.fc19.x86_64.conf.

In order to maximize compatibility with file system implementations and restricted boot loader environments, and to minimize conflicting character use with other programs, file names shall be chosen from a restricted character set: ASCII upper and lower case characters, digits, “+”, “-“, “_” and “.”. Also, the file names should have a length of at least one and at most 255 characters (including the file name suffix).

These configuration snippets shall be UNIX-style text files (i.e. lines separated by a single newline character), in the UTF-8 encoding. The configuration snippets are loosely inspired by Grub1’s configuration syntax. Lines beginning with “#” are used for comments and shall be ignored. The first word of a line is used as key and is separated by one or more spaces from the value.

Type #1 Boot Loader Entry Keys

The following keys are recognized:

Each configuration drop-in snippet must include at least a linux or an efi key. Here is an example for a complete drop-in file:

# /boot/loader/entries/6a9857a393724b7a981ebb5b8495b9ea-3.8.0-2.fc19.x86_64.conf
title        Fedora 19 (Rawhide)
sort-key     fedora
machine-id   6a9857a393724b7a981ebb5b8495b9ea
version      3.8.0-2.fc19.x86_64
options      root=UUID=6d3376e4-fc93-4509-95ec-a21d68011da2 quiet
architecture x64
linux        /6a9857a393724b7a981ebb5b8495b9ea/3.8.0-2.fc19.x86_64/linux
initrd       /6a9857a393724b7a981ebb5b8495b9ea/3.8.0-2.fc19.x86_64/initrd

On EFI systems all Linux kernel images should be EFI images. In order to increase compatibility with EFI systems it is highly recommended only to install EFI kernel images, even on non-EFI systems, if that’s applicable and supported on the specific architecture.

Conversely, in order to increase compatibility it is recommended to install generic kernel images that make few assumptions about the firmware they run on, i.e. it is a good idea that both images shipped as UEFI PE images and those which are not don’t make unnecessary assumption on the underlying firmware, i.e. don’t hard depend on legacy BIOS calls or UEFI boot services.

When Type #1 configuration snippets refer to other files (for linux, initrd, efi, devicetree, and devicetree-overlay), those files must be located on the same partition, and the paths must be absolute paths relative to the root of that file system. The naming of those files can be chosen by the installer. A recommended scheme is described in the next section.

It is recommended to place the kernel and other other files comprising a single boot loader entry in a separate directory: /<entry-token-or-machine-id>/<version>/. This naming scheme uses the same elements as the boot loader configuration snippet, providing the same level of uniqueness.

Example: $BOOT/6a9857a393724b7a981ebb5b8495b9ea/3.8.0-2.fc19.x86_64/linux $BOOT/6a9857a393724b7a981ebb5b8495b9ea/3.8.0-2.fc19.x86_64/initrd

Other naming schemes are possible. In particular, traditionally a flat naming scheme with files in the root directory was used. This is not recommended because it is hard to avoid conflicts in a multi-boot installation.

Standard-conformance Marker File

Unfortunately, there are implementations of boot loading infrastructure that are also using the /loader/entries/ directory, but installing files that do not follow this specification. In order to minimize confusion, a boot loader implementation may place the file /loader/entries.srel next to the /loader/entries/ directory containing the ASCII string type1 (followed by a UNIX newline). Tools that need to determine whether an existing directory implements the semantics described here may check for this file and contents: if it exists and contains the mentioned string, it shall assume a standards-compliant implementation is in place. If it exists but contains a different string it shall assume other semantics are implemented. If the file does not exist, no assumptions should be made.

Type #2 EFI Unified Kernel Images

A unified kernel image is a single EFI PE executable combining an EFI stub loader, a kernel image, an initramfs image, and the kernel command line. See the description of the --uefi option in dracut(8). Such unified images are installed in the$BOOT/EFI/Linux/ and $XBOOTLDR/EFI/Linux/ directories and must have the extension .efi. Support for images of this type is of course specific to systems with EFI firmware. Ignore this section if you work on systems not supporting EFI.

Type #2 file names should be chosen from the same restricted character set as Type #1 described above (but with the file name suffix of .efi instead of .conf).

Images of this type have the advantage that all metadata and payload that makes up the boot entry is contained in a single PE file that can be signed cryptographically as one for the purpose of EFI SecureBoot.

A valid unified kernel image must contain two PE sections:

The PRETTY_NAME= and VERSION_ID= fields in the embedded os-release file are used the same as title and version in the Type #1 entries. The .cmdline section is used instead of the options field. linux and initrd fields are not necessary, and there is no counterpart for the machine-id field.

On EFI, any such images shall be added to the list of valid boot entries.

Additional notes

Note that these configurations snippets do not need to be the only configuration source for a boot loader. It may extend this list of entries with additional items from other configuration files (for example its own native configuration files) or automatically detected other entries without explicit configuration.

To make this explicitly clear: this specification is designed with “free” operating systems in mind, starting Windows or macOS is out of focus with these configuration snippets, use boot-loader specific solutions for that. In the text above, if we say “OS” we hence imply “free”, i.e. primarily Linux (though this could be easily be extended to the BSDs and whatnot).

Note that all paths used in the configuration snippets use a Unix-style “/” as path separator. This needs to be converted to an EFI-style “\” separator in EFI boot loaders.

Locating boot entries

A boot loader locates $BOOT and $XBOOTLDR, then simply reads all the files $BOOT/loader/entries/*.conf and $XBOOTLDR/loader/entries/*.conf, and populates its boot menu. On EFI, it then extends this with any unified kernel images found in $BOOT/EFI/Linux/*.efi and $XBOOTLDR/EFI/Linux/*.efi. It may also add additional entries, for example a “Reboot into firmware” option. Optionally it may sort the menu based on the sort-key, machine-id and version fields, and possibly others. It uses the file name to identify specific items, for example in case it supports storing away default entry information somewhere. A boot loader should generally not modify these files.

For “Boot Loader Specification Entries” (Type #1), the kernel package installer installs the kernel and initrd images to $XBOOTLDR (if used) or $BOOT. It is recommended to place these files in a vendor and OS and installation specific directory. It then generates a configuration snippet, placing it in $BOOT/loader/entries/xyz.conf, with “xyz” as concatenation of machine id and version information (see above). The files created by a kernel package are tied to the kernel package and should be removed along with it.

For “EFI Unified Kernel Images” (Type #2), the vendor or kernel package installer should create the combined image and drop it into $BOOT/EFI/Linux/. This file is also tied to the kernel package and should be removed along with it.

A UI application intended to show available boot options shall operate similarly to a boot loader, but might apply additional filters, for example by filtering the booted OS via the machine ID, or by suppressing all but the newest kernel versions.

An OS installer picks the right place for $BOOT as defined above (possibly creating a partition and file system for it) and creates the /loader/entries/ directory in it. It then installs an appropriate boot loader that can read these snippets. Finally, it installs one or more kernel packages.

Boot counting

The main idea is that when boot entries are initially installed, they are marked as “indeterminate” and assigned a number of boot attempts. Each time the boot loader tries to boot an entry, it decreases this count by one. If the operating system considers the boot as successful, it removes the counter altogether and the entry becomes “good”. Otherwise, once the assigned number of boots is exhausted, the entry is marked as “bad”.

Which boots are “successful” is determined by the operating system. systemd provides a generic mechanism that can be extended with arbitrary checks and actions, see Automatic Boot Assessment, but the boot counting mechanism described in this specifaction can also be used with other implementations.

The boot counting data is stored in the name of the boot loader entry. A boot loader entry file name may contain a plus (+) followed by a number. This may optionally be followed by a minus (-) followed by a second number. The dot (.) and file name suffix (conf of efi) must immediately follow. Boot counting is enabled for entries which match this pattern.

The first number is the “tries left” counter signifying how many attempts to boot this entry shall still be made. The second number is the “tries done” counter, showing how many failed attempts to boot it have already been made. Each time a boot loader entry marked this way is booted, the first counter is decremented, and the second one incremented. (If the second counter is missing, then it is assumed to be equivalent to zero.) If the “tries left” counter is above zero the entry is still considered “indeterminate”. A boot entry with the “tries left” counter at zero is considered “bad”.

If the boot attempt completed successfully the entry’s counters are removed from the name (entry state becomes “good”), thus turning off boot counting for this entry.

Sorting

The boot loader menu should generally show entries in some order meaningful to the user. The title key is free-form and not suitable to be used as the primary sorting key. Instead, the boot loader should use the following rules:

  1. Entries which are subject to boot counting and are marked as “bad”, should be sorted later than all other entries. Entries which are marked as “indeterminate” or “good” (or were not subject to boot counting at all), are thus sorted earlier.

  2. If sort-key is set on both entries, use in order of priority, the sort-key (A-Z, increasing alphanumerical order), machine-id (A-Z, increasing alphanumerical order), and version keys (decreasing version order).

  3. If sort-key is set on one entry, it sorts earlier.

  4. At the end, if necessary, when sort-key is not set or those fields are not set or are all equal, the boot loader should sort using the file name of the entry (decreasing version sort), with the suffix removed.

Note: This description assumes that the boot loader shows entries in a traditional menu, with newest and “best” entries at the top, thus entries with a higher version number are sorter *earlier*. The boot loader is free to use a different direction (or none at all) during display.

Note: The boot loader should allow booting “bad” entries, e.g. in case no other entries are left or they are unusable for other reasons. It may deemphasize or hide such entries by default.

Note: “Bad” boot entries have a suffix of “+0-n”, where n is the number of failed boot attempts. Removal of the suffix is not necessary for comparisons described by the last point above. In the unlikely scenario that we have multiple such boot entries that differ only by the boot counting data, we would sort them by n.

Alphanumerical order

Free-form strings and machine IDs should be compared using a method equivalent to strcmp(3) on their UTF-8 representations. If just one of the strings is unspecified or empty, it compares lower. If both strings are unspecified or empty, they compare equal.

Version order

The following method should be used to compare version strings. The algorithm is based on rpm’s rpmvercmp(), but not identical.

ASCII letters (a-z, A-Z) and digits (0-9) form alphanumerical components of the version. Minus (-) separates the version and release parts. Dot (.) separates parts of version or release. Tilde (~) is a prefix that always compares lower. Caret (^) is a prefix that always compares higher.

Both strings are compared from the beginning until the end, or until the strings are found to compare as different. In a loop:

  1. Any characters which are outside of the set of listed above (a-z, A-Z, 0-9, -, ., ~, ^) are skipped in both strings. In particular, this means that non-ASCII characters that are Unicode digits or letters are skipped too.
  2. If one of the strings has ended: if the other string hasn’t, the string that has remaining characters compares higher. Otherwise, the strings compare equal.
  3. If the remaining part of one of strings starts with ~: if other remaining part does not start with ~, the string with ~ compares lower. Otherwise, both tilde characters are skipped.
  4. The check from point 2. is repeated here.
  5. If the remaining part of one of strings starts with -: if the other remaining part does not start with -, the string with - compares lower. Otherwise, both minus characters are skipped.
  6. If the remaining part of one of strings starts with ^: if the other remaining part does not start with ^, the string with ^ compares higher. Otherwise, both caret characters are skipped.
  7. If the remaining part of one of strings starts with .: if the other remaining part does not start with ., the string with . compares lower. Otherwise, both dot characters are skipped.
  8. If either of the remaining parts starts with a digit, numerical prefixes are compared numerically. Any leading zeroes are skipped. The numerical prefixes (until the first non-digit character) are evaluated as numbers. If one of the prefixes is empty, it evaluates as 0. If the numbers are different, the string with the bigger number compares higher. Otherwise, the comparison continues at the following characters at point 1.
  9. Leading alphabetical prefixes are compared alphabetically. The substrings are compared letter-by-letter. If both letters are the same, the comparison continues with the next letter. Capital letters compare lower than lower-case letters (A < a). When the end of one substring has been reached (a non-letter character or the end of the whole string), if the other substring has remaining letters, it compares higher. Otherwise, the comparison continues at the following characters at point 1.

Examples (with ‘’ meaning the empty string):

Note: systemd-analyze implements this version comparison algorithm as

systemd-analyze compare-versions <version-a> <version-b>

Additional discussion

Why is there a need for this specification?

This specification brings the following advantages:

Why not simply rely on the EFI boot menu logic?

EFI is not ubiquitous, especially not in embedded systems. But even on systems with EFI, which provides a boot options logic that can offer similar functionality, this specification is still needed for the following reasons:

Why is the version comparison logic so complicated?

The sort-key allows us to group entries by “operating system”, e.g. all versions of Fedora together, no matter if they identify themselves as “Fedora Workstation” or “Fedora Rawhide (prerelease)”. The sort-key was introduced only recently, so we need to provide a meaningful order for entries both with and without it. Since it is a new concept, it is assumed that entries with sort-key are newer.

In a traditional menu with entries displayed vertically, we want names to be sorter alpabetically (CentOS, Debian, Fedora, OpenSUSE, …), it would be strange to have them in reverse order. But when multiple kernels are available for the same installation, we want to display the latest kernel with highest priority, i.e. earlier in the list.

Why do you use file renames to store the counter? Why not a regular file?

Mainly two reasons: it’s relatively likely that renames can be implemented atomically even in simpler file systems, as renaming generally avoids allocating or releasing data blocks. Writing to file contents has a much bigger chance to be result in incomplete or corrupt data. Moreover renaming has the benefit that the boot count metadata is directly attached to the boot loader entry file, and thus the lifecycle of the metadata and the entry itself are bound together. This means no additional clean-up needs to take place to drop the boot loader counting information for an entry when it is removed.

Why not use EFI variables for storing the boot counter?

The memory chips used to back the persistent EFI variables are generally not of the highest quality, hence shouldn’t be written to more than necessary. This means we can’t really use it for changes made regularly during boot, but should use it only for seldom-made configuration changes.

Out of Focus

There are a couple of items that are out of focus for this specification:

GUID Partition Table
Boot Loader Interface
Discoverable Partitions Specification
systemd-boot(7)
bootctl(1)
systemd-gpt-auto-generator(8)