The Discoverable Partitions Specification (DPS) #

TL;DR: Let’s automatically discover, mount and enable the root partition, /home/, /srv/, /var/ and /var/tmp/ and the swap partitions based on GUID Partition Tables (GPT)!

This specification describes the use of GUID Partition Table (GPT) UUIDs to enable automatic discovery of partitions and their intended mountpoints. Traditionally Linux has made little use of partition types, mostly just defining one UUID for file system/data partitions and another one for swap partitions. With this specification, we introduce additional partition types for specific uses. This has many benefits:

• OS installers can automatically discover and make sense of partitions of existing Linux installations.
• The OS can discover and mount the necessary file systems with a non-existent or incomplete /etc/fstab file and without the root= kernel command line option.
• Container managers (such as nspawn and libvirt-lxc) can introspect and set up file systems contained in GPT disk images automatically and mount them to the right places, thus allowing booting the same, identical images on bare metal and in Linux containers. This enables true, natural portability of disk images between physical machines and Linux containers.
• As a help to administrators and users partition manager tools can show more descriptive information about partitions tables.

Note that the OS side of this specification is currently implemented in systemd 211 and newer in the systemd-gpt-auto-generator(8) generator tool. Note that automatic discovery of the root only works if the boot loader communicates this information to the OS, by implementing the Boot Loader Interface.

Defined Partition Type UUIDs #

Other GPT type IDs might be used on Linux, for example to mark software RAID or LVM partitions. The definitions of those GPT types is outside of the scope of this specification.

systemd-id128(1)’s show command may be used to list those GPT partition type UUIDs.

Partition Names #

For partitions of the types listed above it is recommended to use human-friendly, descriptive partition names in the GPT partition table, for example “Home”, “Server Data”, “Fedora Root” and similar, possibly localized.

For the Root/Verity/Verity signature partitions it might make sense to use a versioned naming scheme reflecting the OS name and its version, e.g. “fooOS_2021.4” or similar. For details about the version format see the Version Format Specification. The underscore character (_) must be used to separate the version from the name of the image.

Partition Attribute Flags #

This specification defines three GPT partition attribute flags that may be set for the partition types defined above:

1. For the root, /usr/, Verity, Verity signature, home, server data, variable data, temporary data, swap, and extended boot loader partitions, the partition flag bit 63 ("no-auto", SD_GPT_FLAG_NO_AUTO) may be used to turn off auto-discovery for the specific partition. If set, the partition will not be automatically mounted or enabled.

2. For the root, /usr/, Verity, Verity signature home, server data, variable data, temporary data and extended boot loader partitions, the partition flag bit 60 ("read-only", SD_GPT_FLAG_READ_ONLY) may be used to mark a partition for read-only mounts only. If set, the partition will be mounted read-only instead of read-write. Note that the variable data partition and the temporary data partition will generally not be able to serve their purpose if marked read-only, since by their very definition they are supposed to be mutable. (The home and server data partitions are generally assumed to be mutable as well, but the requirement for them is not equally strong.) Because of that, while the read-only flag is defined and supported, it’s almost never a good idea to actually use it for these partitions. Also note that Verity and signature partitions are by their semantics always read-only. The flag is hence of little effect for them, and it is recommended to set it unconditionally for the Verity and signature partition types.

3. For the root, /usr/, home, server data, variable data, temporary data and extended boot loader partitions, the partition flag bit 59 ("grow-file-system", SD_GPT_FLAG_GROWFS) may be used to mark a partition for automatic growing of the contained file system to the size of the partition when mounted. Tools that automatically mount disk image with a GPT partition table are suggested to implicitly grow the contained file system to the partition size they are contained in, if they are found to be smaller. This flag is without effect on partitions marked “read-only”.

Note that the first two flag definitions happen to correspond nicely to the same ones used by Microsoft Basic Data Partitions.

All three of these flags generally affect only auto-discovery and automatic mounting of disk images. If partitions marked with these flags are mounted using low-level commands like mount(8) or directly with mount(2), they typically have no effect.

Verity #

The Root//usr/ partition types and their matching Verity and Verity signature partitions enable relatively automatic handling of dm-verity protected setups. These types are defined with two modes of operation in mind:

1. A trusted Verity root hash is passed in externally, for example is specified on the kernel command line that is signed along with the kernel image using SecureBoot PE signing (which in turn is tested against a set of firmware-provided set of signing keys). If so, discovery and setup of a Verity volume may be fully automatic: if the root partition’s UUID is chosen to match the first 128 bit of the root hash, and the matching Verity partition UUIDs is chosen to match the last 128bit of the root hash, then automatic discovery and match-up of the two partitions is possible, as the root hash is enough to both find the partitions and then combine them in a Verity volume. In this mode a Verity signature partition is not used and unnecessary.

2. A Verity signature partition is included on the disk, with a signature to be tested against a system-provided set of signing keys. The signature partition primarily contains two fields: the root hash to use, and a PKCS#7 signature of it, using a signature key trusted by the OS. If so, discovery and setup of a Verity volume may be fully automatic. First, the specified root hash is validated with the signature and the OS-provided trusted keys. If the signature checks out the root hash is then used in the same way as in the first mode of operation described above.

Both modes of operation may be combined in a single image. This is particularly useful for images that shall be usable in two different contexts: for example an image that shall be able to boot directly on UEFI systems (in which case it makes sense to include the root hash on the kernel command line that is included in the signed kernel image to boot, as per mode of operation #1 above), but also be able to used as image for a container engine (such as systemd-nspawn), which can use the signature partition to validate the image, without making use of the signed kernel image (and thus following mode of operation #2).

The Verity signature partition’s contents should be a serialized JSON object in text form, padded with NUL bytes to the next multiple of 4096 bytes in size. Currently three fields are defined for the JSON object:

1. The (mandatory) rootHash field should be a string containing the Verity root hash, formatted as series of (lowercase) hex characters.

2. The (mandatory) signature field should be a string containing the PKCS#7 signature of the root hash, in Base64-encoded DER format. This should be the same format used by the Linux kernel’s dm-verity signature logic, i.e. the signed data should be the exact string representation of the hash, as stored in rootHash above.

3. The (optional) certificateFingerprint field should be a string containing a SHA256 fingerprint of the X.509 certificate in DER format for the key that signed the root hash, formatted as series of (lowercase) hex characters (no : separators or such).

More fields might be added in later revisions of this specification.

Suggested Mode of Operation #

An installer that repartitions the hard disk should use the above UUID partition types for appropriate partitions it creates.

An installer which supports a “manual partitioning” interface may choose to pre-populate the interface with swap, /home/, /srv/, /var/tmp/ partitions of pre-existing Linux installations, identified with the GPT type UUIDs above. The installer should not pre-populate such an interface with any identified root, /usr or /var/ partition unless the intention is to overwrite an existing operating system that might be installed.

An installer may omit creating entries in /etc/fstab for root, /home/, /srv/, /var/, /var/tmp and for the swap partitions if they use these UUID partition types, and are the first partitions on the disk of each type. If the ESP shall be mounted to /efi/ (or /boot/), it may additionally omit creating the entry for it in /etc/fstab. If the EFI partition shall not be mounted to /efi/ or /boot/, it must create /etc/fstab entries for them. If other partitions are used (for example for /usr/local/ or /var/lib/mysql/), the installer must register these in /etc/fstab. The root= parameter passed to the kernel by the boot loader may be omitted if the root partition is the first one on the disk of its type. If the root partition is not the first one on the disk, the root= parameter must be passed to the kernel by the boot loader. An installer that mounts a root, /usr/, /home/, /srv/, /var/, or /var/tmp/ file system with the partition types defined as above which contains a LUKS header must call the device mapper device “root”, “usr”, “home”, “srv”, “var” or “tmp”, respectively. This is necessary to ensure that the automatic discovery will never result in different device mapper names than any static configuration by the installer, thus eliminating possible naming conflicts and ambiguities.

An operating system should automatically discover and mount the first root partition that does not have the no-auto flag set (as described above) by scanning the disk containing the currently used EFI ESP. It should automatically discover and mount the first /usr/, /home/, /srv/, /var/, /var/tmp/ and swap partitions that do not have the no-auto flag set by scanning the disk containing the discovered root partition. It should automatically discover and mount the partition containing the currently used EFI ESP to /efi/ (or /boot/ as fallback). It should automatically discover and mount the partition containing the currently used Extended Boot Loader Partition to /boot/. It should not discover or automatically mount partitions with other UUID partition types, or partitions located on other disks, or partitions with the no-auto flag set. User configuration shall always override automatic discovery and mounting. If a root, /usr/, /home/, /srv/, /boot/, /var/, /var/tmp/, /efi/, /boot/ or swap partition is listed in /etc/fstab or with root= on the kernel command line, it must take precedence over automatically discovered partitions. If a /home/, /usr/, /srv/, /boot/, /var/, /var/tmp/, /efi/ or /boot/ directory is found to be populated already in the root partition, the automatic discovery must not mount any discovered file system over it. Optionally, in case of the root, /usr/ and their Verity partitions instead of strictly mounting the first suitable partition an OS might choose to mount the partition whose label compares the highest according to strverscmp() or similar logic, in order to implement a simple partition-based A/B versioning scheme. The precise rules are left for the implementation to decide, but when in doubt earlier partitions (by their index) should always win over later partitions if the label comparison is inconclusive.

A container manager should automatically discover and mount the root, /usr/, /home/, /srv/, /var/, /var/tmp/ partitions inside a container disk image. It may choose to mount any discovered ESP and/or XBOOTLDR partition to /efi/ or /boot/. It should ignore any swap should they be included in a container disk image.

If a btrfs file system is automatically discovered and mounted by the operating system/container manager it will be mounted with its default subvolume. The installer should make sure to set the default subvolume correctly using “btrfs subvolume set-default”.

Sharing of File Systems between Installations #

If two Linux-based operating systems are installed on the same disk, the scheme above suggests that they may share the swap, /home/, /srv/, /var/tmp/, ESP, XBOOTLDR. However, they should each have their own root, /usr/ and /var/ partition.

Frequently Asked Questions #

Why are you taking my /etc/fstab away? #

We are not. /etc/fstab always overrides automatic discovery and is indeed mentioned in the specifications. We are simply trying to make the boot and installation processes of Linux a bit more robust and self-descriptive.

Why did you only define the root partition for these listed architectures? #

Please submit a patch that adds appropriate partition type UUIDs for the architecture of your choice should they be missing so far. The only reason they aren’t defined yet is that nobody submitted them yet.

Why define distinct root partition UUIDs for the various architectures? #

This allows disk images that may be booted on multiple architectures to use discovery of the appropriate root partition on each architecture.

Doesn’t this break multi-boot scenarios? #

No, it doesn’t. The specification says that installers may not stop creating /etc/fstab or stop including root= on the kernel command line, unless the used partitions are the first ones of their type on the disk. Additionally, /etc/fstab and root= both override automatic discovery. Multi-boot is hence well supported, since it doesn’t change anything for anything but the first installation.

That all said, it’s not expected that generic installers generally stop setting root= and creating /etc/fstab anyway. The option to drop these configuration bits is primarily something for appliance-like devices. However, generic installers should still set the right GPT partition types for the partitions they create so that container managers, partition tools and administrators can benefit. Phrased differently, this specification introduces A) the recommendation to use the newly defined partition types to tag things properly and B) the option to then drop root= and /etc/fstab. While we advertise A) to all installers, we only propose B) for simpler, appliance-like installations.

What partitioning tools will create a DPS-compliant partition table? #

As of util-linux 2.25.2, the fdisk tool provides type codes to create the root, home, and swap partitions that the DPS expects. By default, fdisk will create an old-style MBR, not a GPT, so typing l to list partition types will not show the choices to let you set the correct UUID. Make sure to first create an empty GPT, then type l in order for the DPS-compliant type codes to be available.

The gdisk tool (from version 1.0.5 onward) and its variants (sgdisk, cgdisk) also support creation of partitions with a matching type code.

Links #

Boot Loader Specification
Boot Loader Interface
Safely Building Images
systemd-boot(7)
bootctl(1)
systemd-gpt-auto-generator(8)