sysupdate.d
systemd
sysupdate.d
5
sysupdate.d
Transfer Definition Files for Automatic Updates
/etc/sysupdate.d/*.conf
/run/sysupdate.d/*.conf
/usr/lib/sysupdate.d/*.conf
Description
sysupdate.d/*.conf files describe how specific resources on the local system
shall be updated from a remote source. Each such file defines one such transfer: typically a remote
HTTP/HTTPS resource as source; and a local file, directory or partition as target. This may be used as a
simple, automatic, atomic update mechanism for the OS itself, for containers, portable services or system
extension images — but in fact may be used to update any kind of file from a remote source.
The
systemd-sysupdate8
command reads these files and uses them to determine which local resources should be updated, and then
executes the update.
Both the remote HTTP/HTTPS source and the local target typically exist in multiple, concurrent
versions, in order to implement flexible update schemes, e.g. A/B updating (or a superset thereof,
e.g. A/B/C, A/B/C/D, …).
Each *.conf file defines one transfer, i.e. describes one resource to
update. Typically, multiple of these files (i.e. multiple of such transfers) are defined together, and
are bound together by a common version identifier in order to update multiple resources at once on each
update operation, for example to update a kernel, a root file system and a Verity partition in a single,
combined, synchronized operation, so that only a combined update of all three together constitutes a
complete update.
Each *.conf file contains three sections: [Transfer], [Source] and [Target].
Basic Mode of Operation
Disk-image based OS updates typically consist of multiple different resources that need to be
updated together, for example a secure OS update might consist of a root file system image to drop into a
partition, a matching Verity integrity data partition image, and a kernel image prepared to boot into the
combination of the two partitions. The first two resources are files that are downloaded and placed in a
disk partition, the latter is a file that is downloaded and placed in a regular file in the boot file
system (e.g. EFI system partition). Hence, during an update of a hypothetical operating system "foobarOS"
to a hypothetical version 47 the following operations should take place:
A file https://download.example.com/foobarOS_47.root.xz should be
downloaded, decompressed and written to a previously unused partition with GPT partition type UUID
4f68bce3-e8cd-4db1-96e7-fbcaf984b709 for x86-64, as per Discoverable Partitions
Specification.
Similarly, a file https://download.example.com/foobarOS_47.verity.xz
should be downloaded, decompressed and written to a previously empty partition with GPT partition type
UUID of 2c7357ed-ebd2-46d9-aec1-23d437ec2bf5 (i.e. the partition type for Verity integrity information
for x86-64 root file systems).
Finally, a file https://download.example.com/foobarOS_47.efi.xz (a
unified kernel, as per Boot Loader
Specification Type #2) should be downloaded, decompressed and written to the $BOOT file system,
i.e. to EFI/Linux/foobarOS_47.efi in the ESP or XBOOTLDR partition.
The version-independent generalization of this would be (using the special marker
@v as wildcard for the version identifier):
A transfer of a file https://download.example.com/foobarOS_@v.root.xz
→ a local, previously empty GPT partition of type 4f68bce3-e8cd-4db1-96e7-fbcaf984b709, with the label to
be set to foobarOS_@v.
A transfer of a file https://download.example.com/foobarOS_@v.verity.xz
→ a local, previously empty GPT partition of type 2c7357ed-ebd2-46d9-aec1-23d437ec2bf5, with the label to be
set to foobarOS_@v_verity.
A transfer of a file https://download.example.com/foobarOS_@v.efi.xz
→ a local file $BOOT/EFI/Linux/foobarOS_@v.efi.
An update can only complete if the relevant URLs provide their resources for the same version,
i.e. for the same value of @v.
The above may be translated into three *.conf files in
sysupdate.d/, one for each resource to transfer. The *.conf
files configure the type of download, and what place to write the download to (i.e. whether to a
partition or a file in the file system). Most importantly these files contain the URL, partition name and
filename patterns shown above that describe how these resources are called on the source and how they
shall be called on the target.
In order to enumerate available versions and figuring out candidates to update to, a mechanism is
necessary to list suitable files:
For partitions: the surrounding GPT partition table contains a list of defined
partitions, including a partition type UUID and a partition label (in this scheme the partition label
plays a role for the partition similar to the filename for a regular file).
For regular files: the directory listing of the directory the files are contained in
provides a list of existing files in a straightforward way.
For HTTP/HTTPS sources a simple scheme is used: a manifest file
SHA256SUMS, following the format defined by sha256sum1,
lists file names and their SHA256 hashes.
Transfers are done in the alphabetical order of the .conf file names they are
defined in. First, the resource data is downloaded directly into a target file/directory/partition. Once
this is completed for all defined transfers, in a second step the files/directories/partitions are
renamed to their final names as defined by the target MatchPattern=, again in the
order the .conf transfer file names dictate. This step is not atomic, however it is
guaranteed to be executed strictly in order with suitable disk synchronization in place. Typically, when
updating an OS one of the transfers defines the entry point when booting. Thus it is generally a good idea
to order the resources via the transfer configuration file names so that the entry point is written
last, ensuring that any abnormal termination does not leave an entry point around whose backing is not
established yet. In the example above it would hence make sense to establish the EFI kernel image last
and thus give its transfer configuration file the alphabetically last name.
See below for an extended, more specific example based on the above.
Resource Types
Each transfer file defines one source resource to transfer to one target resource. The following
resource types are supported:
Resources of type url-file encapsulate a file on a web server,
referenced via a HTTP or HTTPS URL. When an update takes place, the file is downloaded and decompressed
and then written to the target file or partition. This resource type is only available for sources, not
for targets. The list of available versions of resources of this type is encoded in
SHA256SUMS manifest files, accompanied by
SHA256SUMS.gpg detached signatures.
The url-tar resource type is similar, but the file must be a
.tar archive. When an update takes place, the file is decompressed and unpacked
into a directory or btrfs subvolume. This resource type is only available for sources, not for
targets. Just like url-file, url-tar version enumeration makes
use of SHA256SUMS files, authenticated via
SHA256SUMS.gpg.
The regular-file resource type encapsulates a local regular file on
disk. During updates the file is uncompressed and written to the target file or partition. This
resource type is available both as source and as target. When updating no integrity or authentication
verification is done for resources of this type.
The partition resource type is similar to
regular-file, and encapsulates a GPT partition on disk. When updating, the partition
must exist already, and have the correct GPT partition type. A partition whose GPT partition label is
set to _empty is considered empty, and a candidate to place a newly downloaded
resource in. The GPT partition label is used to store version information, once a partition is
updated. This resource type is only available for target resources.
The tar resource type encapsulates local .tar
archive files. When an update takes place, the files are uncompressed and unpacked into a target
directory or btrfs subvolume. Behaviour of tar and url-tar is
generally similar, but the latter downloads from remote sources, and does integrity and authentication
checks while the former does not. The tar resource type is only available for source
resources.
The directory resource type encapsulates local directory trees. This
type is available both for source and target resources. If an update takes place on a source resource
of this type, a recursive copy of the directory is done.
The subvolume resource type is identical to
directory, except when used as the target, in which case the file tree is placed in
a btrfs subvolume instead of a plain directory, if the backing file system supports it (i.e. is
btrfs).
As already indicated, only a subset of source and target resource type combinations are
supported:
Resource Types
Identifier
Description
Usable as Source
When Used as Source: Compatible Targets
When Used as Source: Integrity + Authentication
When Used as Source: Decompression
Usable as Target
When Used as Target: Compatible Sources
url-file
HTTP/HTTPS files
yes
regular-file, partition
yes
yes
no
-
url-tar
HTTP/HTTPS .tar archives
yes
directory, subvolume
yes
yes
no
-
regular-file
Local files
yes
regular-file, partition
no
yes
yes
url-file, regular-file
partition
Local GPT partitions
no
-
-
-
yes
url-file, regular-file
tar
Local .tar archives
yes
directory, subvolume
no
yes
no
-
directory
Local directories
yes
directory, subvolume
no
no
yes
url-tar, tar, directory, subvolume
subvolume
Local btrfs subvolumes
yes
directory, subvolume
no
no
yes
url-tar, tar, directory, subvolume
Match Patterns
Both the source and target resources typically exist in multiple versions concurrently. An update
operation is done whenever the newest of the source versions is newer than the newest of the target
versions. To determine the newest version of the resources a directory listing, partition listing or
manifest listing is used, a subset of qualifying entries selected from that, and the version identifier
extracted from the file names or partition labels of these selected entries. Subset selection and
extraction of the version identifier (plus potentially other metadata) is done via match patterns,
configured in MatchPattern= in the [Source] and [Target] sections. These patterns are
strings that describe how files or partitions are named, with named wildcards for specific fields such as
the version identifier. The following wildcards are defined:
Match Pattern Wildcards
Wildcard
Description
Format
Notes
@v
Version identifier
Valid version string
Mandatory
@u
GPT partition UUID
Valid 128-Bit UUID string
Only relevant if target resource type chosen as partition
@f
GPT partition flags
Formatted hexadecimal integer
Only relevant if target resource type chosen as partition
@a
GPT partition flag NoAuto
Either 0 or 1
Controls NoAuto bit of the GPT partition flags, as per Discoverable Partitions Specification; only relevant if target resource type chosen as partition
@g
GPT partition flag GrowFileSystem
Either 0 or 1
Controls GrowFileSystem bit of the GPT partition flags, as per Discoverable Partitions Specification; only relevant if target resource type chosen as partition
@r
Read-only flag
Either 0 or 1
Controls ReadOnly bit of the GPT partition flags, as per Discoverable Partitions Specification and other output read-only flags, see ReadOnly= below
@t
File modification time
Formatted decimal integer, μs since UNIX epoch Jan 1st 1970
Only relevant if target resource type chosen as regular-file
@m
File access mode
Formatted octal integer, in UNIX fashion
Only relevant if target resource type chosen as regular-file
@s
File size after decompression
Formatted decimal integer
Useful for measuring progress and to improve partition allocation logic
@d
Tries done
Formatted decimal integer
Useful when operating with kernel image files, as per Automatic Boot Assessment
@l
Tries left
Formatted decimal integer
Useful when operating with kernel image files, as per Automatic Boot Assessment
@h
SHA256 hash of compressed file
64 hexadecimal characters
The SHA256 hash of the compressed file; not useful for url-file or url-tar where the SHA256 hash is already included in the manifest file anyway
Of these wildcards only @v must be present in a valid pattern, all other
wildcards are optional. Each wildcard may be used at most once in each pattern. A typical wildcard
matching a file system source image could be MatchPattern=foobar_@v.raw.xz, i.e. any file
whose name begins with foobar_, followed by a version ID and suffixed by
.raw.xz.
Do not confuse the @ pattern matching wildcard prefix with the
% specifier expansion prefix. The former encapsulate a variable part of a match
pattern string, the latter are simple shortcuts that are expanded while the drop-in files are
parsed. For details about specifiers, see below.
[Transfer] Section Options
This section defines general properties of this transfer.
MinVersion=
Specifies the minimum version to require for this transfer to take place. If the
source or target patterns in this transfer definition match files older than this version they will
be considered obsolete, and never be considered for the update operation.
ProtectVersion=
Takes one or more version strings to mark as "protected". Protected versions are
never removed while making room for new, updated versions. This is useful to ensure that the
currently booted OS version (or auxiliary resources associated with it) is not replaced/overwritten
during updates, in order to avoid runtime file system corruptions.
Like many of the settings in these configuration files this setting supports specifier
expansion. It's particularly useful to set this setting to one of the %A,
%B or %w specifiers to automatically refer to the current OS
version of the running system. See below for details on supported specifiers.
Verify=
Takes a boolean, defaults to yes. Controls whether to cryptographically verify
downloaded resources (specifically: validate the GPG signatures for downloaded
SHA256SUMS manifest files, via their detached signature files
SHA256SUMS.gpg in combination with the system keyring
/usr/lib/systemd/import-pubring.gpg or
/etc/systemd/import-pubring.gpg).
This option is essential to provide integrity guarantees for downloaded resources and thus
should be left enabled, outside of test environments.
Note that the downloaded payload files are unconditionally checked against the SHA256 hashes
listed in the manifest. This option only controls whether the signatures of these manifests are
verified.
This option only has an effect if the source resource type is selected as
url-file or url-tar, as integrity and authentication
checking is only available for transfers from remote sources.
[Source] Section Options
This section defines properties of the transfer source.
Type=
Specifies the resource type of the source for the transfer. Takes one of
url-file, url-tar, tar,
regular-file, directory or
subvolume. For details about the resource types, see above. This option is
mandatory.
Note that only certain combinations of source and target resource types are supported, see
above.
Path=
Specifies where to find source versions of this resource.
If the source type is selected as url-file or
url-tar this must be a HTTP/HTTPS URL. The URL is suffixed with
/SHA256SUMS to acquire the manifest file, with
/SHA256SUMS.gpg to acquire the detached signature file for it, and with the file
names listed in the manifest file in case an update is executed and a resource shall be
downloaded.
For all other source resource types this must be a local path in the file system, referring to
a local directory to find the versions of this resource in.
MatchPattern=
Specifies one or more file name match patterns that select the subset of files that
are update candidates as source for this transfer. See above for details on match patterns.
This option is mandatory. Any pattern listed must contain at least the @v
wildcard, so that a version identifier may be extracted from the filename. All other wildcards are
optional.
If the source type is regular-file or directory, the
pattern may contain slash characters. In this case it will match the file or directory in
corresponding subdirectory. For example MatchPattern=foo_@v/bar.efi will match
bar.efi in directory foo_1.
[Target] Section Options
This section defines properties of the transfer target.
Type=
Specifies the resource type of the target for the transfer. Takes one of
partition, regular-file, directory or
subvolume. For details about the resource types, see above. This option is
mandatory.
Note that only certain combinations of source and target resource types are supported, see
above.
Path=
Specifies a file system path where to look for already installed versions or place
newly downloaded versions of this configured resource. If Type= is set to
partition, expects a path to a (whole) block device node, or the special string
auto in which case the block device which contains the root file system of the
currently booted system is automatically determined and used. If Type= is set to
regular-file, directory or subvolume,
must refer to a path in the local file system referencing the directory to find or place the version
files or directories under.
Note that this mechanism cannot be used to create or remove partitions, in case
Type= is set to partition. Partitions must exist already, and
a special partition label _empty is used to indicate empty partitions. To
automatically generate suitable partitions on first boot, use a tool such as
systemd-repart8.
PathRelativeTo=
Specifies what partition Path= should be relative to. Takes one of
root, esp, xbootldr, or boot.
If unspecified, defaults to root.
If set to boot, the specified Path= will be resolved
relative to the mount point of the $BOOT partition (i.e. the ESP or XBOOTLDR), as defined by the
Boot Loader
Specification.
The values esp, xbootldr, and
boot are only supported when Type= is set to
regular-file or directory.
MatchPattern=
Specifies one or more file name or partition label match patterns that select the
subset of files or partitions that are update candidates as targets for this transfer. See above for
details on match patterns.
This option is mandatory. Any pattern listed must contain at least the @v
wildcard, so that a version identifier may be extracted from the filename. All other wildcards are
optional.
This pattern is both used for matching existing installed versions and for determining the name
of new versions to install. If multiple patterns are specified, the first specified is used for
naming newly installed versions.
If the target type is regular-file or directory, the
pattern may contain slash characters. In this case it will match the file or directory in
corresponding subdirectory. For example MatchPattern=foo_@v/bar.efi will match
bar.efi in directory foo_1. Directories in the path will be
created when file is installed. Empty directories will be removed when file is removed.
MatchPartitionType=
When the target Type= is chosen as partition,
specifies the GPT partition type to look for. Only partitions of this type are considered, all other
partitions are ignored. If not specified, the GPT partition type linux-generic
is used. Accepts either a literal type UUID or a symbolic type identifier. For a list of supported
type identifiers, see the Type= setting in
repart.d5.
PartitionUUID=
PartitionFlags=
PartitionNoAuto=
PartitionGrowFileSystem=
When the target Type= is picked as partition,
selects the GPT partition UUID and partition flags to use for the updated partition. Expects a valid
UUID string, a hexadecimal integer, or booleans, respectively. If not set, but the source match
pattern includes wildcards for these fields (i.e. @u, @f,
@a, or @g), the values from the patterns are used. If neither
configured with wildcards or these explicit settings, the values are left untouched. If both the
overall PartitionFlags= flags setting and the individual flag settings
PartitionNoAuto= and PartitionGrowFileSystem= are used (or the
wildcards for them), then the latter override the former, i.e. the individual flag bit overrides the
overall flags value. See Discoverable
Partitions Specification for details about these flags.
Note that these settings are not used for matching, they only have effect on newly written
partitions in case a transfer takes place.
ReadOnly=
Controls whether to mark the resulting file, subvolume or partition read-only. If the
target type is partition this controls the ReadOnly partition flag, as per
Discoverable Partitions
Specification, similar to the PartitionNoAuto= and
PartitionGrowFileSystem= flags described above. If the target type is
regular-file, the writable bit is removed from the access mode. If the
target type is subvolume, the subvolume will be marked read-only as a
whole. Finally, if the target Type= is selected as directory,
the "immutable" file attribute is set, see chattr1 for
details.
Mode=
The UNIX file access mode to use for newly created files in case the target resource
type is picked as regular-file. Expects an octal integer, in typical UNIX
fashion. If not set, but the source match pattern includes a wildcard for this field
(i.e. @t), the value from the pattern is used.
Note that this setting is not used for matching, it only has an effect on newly written
files when a transfer takes place.
TriesDone=
TriesLeft=
These options take positive, decimal integers, and control the number of attempts
done and left for this file. These settings are useful for managing kernel images, following the
scheme defined in Automatic Boot
Assessment, and only have an effect if the target pattern includes the @d
or @l wildcards.
InstancesMax=
Takes a decimal integer equal to or greater than 2. This configures how many concurrent
versions of the resource to keep. Whenever a new update is initiated it is made sure that no more
than the number of versions specified here minus one exist in the target. Any excess versions are
deleted (in case the target Type= of regular-file,
directory, subvolume is used) or emptied (in case the
target Type= of partition is used; emptying in this case
simply means to set the partition label to the special string _empty; note that no
partitions are actually removed). After an update is completed the number of concurrent versions of
the target resources is equal to or below the number specified here.
Note that this setting may be set differently for each transfer. However, it generally is
advisable to keep this setting the same for all transfers, since otherwise incomplete combinations of
files or partitions will be left installed.
If the target Type= is selected as partition, the number
of concurrent versions to keep is additionally restricted by the number of partition slots of the
right type in the partition table. I.e. if there are only 2 partition slots for the selected
partition type, setting this value larger than 2 is without effect, since no more than 2 concurrent
versions could be stored in the image anyway.
RemoveTemporary=
Takes a boolean argument. If this option is enabled (which is the default) before
initiating an update, all left-over, incomplete updates from a previous attempt are removed from the
target directory. This only has an effect if the target resource Type= is selected
as regular-file, directory or
subvolume.
CurrentSymlink=
Takes a symlink name as argument. If this option is used, as the last step of the
update a symlink under the specified name is created/updated pointing to the completed update. This
is useful in to provide a stable name always pointing to the newest version of the resource. This is
only supported if the target resource Type= is selected as
regular-file, directory or
subvolume.
Specifiers
Specifiers may be used in the MinVersion=, ProtectVersion=,
Path=, MatchPattern= and CurrentSymlink=
settings. The following expansions are understood:
Specifiers available
Specifier
Meaning
Details
Do not confuse the % specifier expansion prefix with the @
pattern matching wildcard prefix. The former are simple shortcuts that are expanded while the drop-in
files are parsed, the latter encapsulate a variable part of a match pattern string. For details about
pattern matching wildcards, see above.
Examples
Updates for a Verity Enabled Secure OS
With the following three files we define a root file system partition, a matching Verity
partition, and a unified kernel image to update as one. This example is an extension of the example
discussed earlier in this man page.
# /usr/lib/sysupdate.d/50-verity.conf
[Transfer]
ProtectVersion=%A
[Source]
Type=url-file
Path=https://download.example.com/
MatchPattern=foobarOS_@v_@u.verity.xz
[Target]
Type=partition
Path=auto
MatchPattern=foobarOS_@v_verity
MatchPartitionType=root-verity
PartitionFlags=0
ReadOnly=1
The above defines the update mechanism for the Verity partition of the root file system. Verity
partition images are downloaded from
https://download.example.com/foobarOS_@v_@u.verity.xz and written to a suitable
local partition, which is marked read-only. Under the assumption this update is run from the image
itself the current image version (i.e. the %A specifier) is marked as protected, to
ensure it is not corrupted while booted. Note that the partition UUID for the target partition is
encoded in the source file name. Fixating the partition UUID can be useful to ensure that
roothash= on the kernel command line is sufficient to pinpoint both the Verity and
root file system partition, and also encode the Verity root level hash (under the assumption the UUID
in the file names match their top-level hash, the way
systemd-gpt-auto-generator8
suggests).
# /usr/lib/sysupdate.d/60-root.conf
[Transfer]
ProtectVersion=%A
[Source]
Type=url-file
Path=https://download.example.com/
MatchPattern=foobarOS_@v_@u.root.xz
[Target]
Type=partition
Path=auto
MatchPattern=foobarOS_@v
MatchPartitionType=root
PartitionFlags=0
ReadOnly=1
The above defines a matching transfer definition for the root file system.
# /usr/lib/sysupdate.d/70-kernel.conf
[Transfer]
ProtectVersion=%A
[Source]
Type=url-file
Path=https://download.example.com/
MatchPattern=foobarOS_@v.efi.xz
[Target]
Type=regular-file
Path=/EFI/Linux
PathRelativeTo=boot
MatchPattern=foobarOS_@v+@l-@d.efi \
foobarOS_@v+@l.efi \
foobarOS_@v.efi
Mode=0444
TriesLeft=3
TriesDone=0
InstancesMax=2
The above installs a unified kernel image into the $BOOT partition, as per
Boot Loader
Specification Type #2. This defines three possible patterns for the names of the kernel
images, as per Automatic Boot Assessment,
and ensures when installing new kernels, they are set up with 3 tries left. No more than two parallel
kernels are kept.
With this setup the web server would serve the following files, for a hypothetical version 7 of
the OS:
SHA256SUMS – The manifest file, containing available files and their SHA256 hashes
SHA256SUMS.gpg – The detached cryptographic signature for the manifest file
foobarOS_7_8b8186b1-2b4e-4eb6-ad39-8d4d18d2a8fb.verity.xz – The Verity image for version 7
foobarOS_7_f4d1234f-3ebf-47c4-b31d-4052982f9a2f.root.xz – The root file system image for version 7
foobarOS_7_efi.xz – The unified kernel image for version 7
For each new OS release a new set of the latter three files would be added, each time with an
updated version. The SHA256SUMS manifest should then be updated accordingly,
listing all files for all versions that shall be offered for download.
Updates for Plain Directory Container Image
[Source]
Type=url-tar
Path=https://download.example.com/
MatchPattern=myContainer_@v.tar.gz
[Target]
Type=subvolume
Path=/var/lib/machines
MatchPattern=myContainer_@v
CurrentSymlink=myContainer
On updates this downloads https://download.example.com/myContainer_@v.tar.gz
and decompresses/unpacks it to /var/lib/machines/myContainer_@v. After each update
a symlink /var/lib/machines/myContainer is created/updated always pointing to the
most recent update.
See Also
systemd1,
systemd-sysupdate8,
systemd-repart8