HALF ECHO LEARN / REFERENCE COLLECTION

v262.2

NEWS

Git blob: b0d42b491c5b5bbc5168dc8d19d77308851d87df

Lines 1–1000 of 23,379

systemd System and Service Manager
 
CHANGES WITH 262:
 
        Announcements of Future Feature Removals and Incompatible Changes:
 
        * systemd-logind's integration with the UAPI.1 Boot Loader
          Specification (which allows the systemctl reboot --boot-loader-entry=
          switch to work) so far has supported a special directory
          /run/boot-loader-entries/ which allowed defining boot loader entries
          outside of the ESP/XBOOTLDR partition for compatibility with legacy
          systems that do not natively implement UAPI.1. However, it appears
          that (to our knowledge) it is not actually being used by any project
          (quite unlike UAPI.1 itself, which found adoption far beyond
          systemd), and its implementation is incomplete. With the future v263
          release we intend to remove support for /run/boot-loader-entries/ and
          related interfaces, in order to simplify our codebase. Support for
          UAPI.1 is – of course – kept in place.
 
        * The experimental "systemd-sysupdated" D-Bus API is going to be
          removed in the next release (v263). The plan is that in its place
          clients should directly talk to systemd-sysupdate (i.e. the backend
          of "systemd-sysupdated") via Varlink IPC. The "updatectl" tool will
          be reworked along these lines.
 
        Feature Removals and Incompatible Changes:
 
        * DNS-SD services registered via systemd-resolved's D-Bus method
          org.freedesktop.resolve1.Manager.RegisterService() are now
          unregistered automatically when the client that registered them
          disconnects from the bus, as originally intended and matching the
          behavior of the equivalent Avahi API. Due to a bug the client
          tracking never worked, and such services used to stay registered
          until systemd-resolved exited (but did not survive a restart of the
          daemon either). Clients must now keep their bus connection open for
          as long as the service shall remain registered; one-shot
          registrations, e.g. via busctl or gdbus, will be withdrawn again as
          soon as the tool exits. To register a DNS-SD service independently
          of a client's lifetime, use *.dnssd files, see systemd.dnssd(5).
 
        * Meson options '-Dlibiptc=' (deprecated in v259), '-Dlibidn=',
          '-Drc-local=', '-Dsysvinit-path=', and '-Dsysvrcnd-path=' (deprecated
          and without effect since v260) have been removed.
 
        * Services of Type=notify-reload are now required to catch or block
          ReloadSignal= when they send READY=1. If they do neither at that
          point, the service will fail to start with a protocol error.
          Previously, such services were allowed to start even though a later
          reload could invoke the signal's default action and terminate the
          service.
 
        * systemd-repart's MakeSymlinks= option now expands "%" specifiers in
          the symlink target. Previously, these were only expanded in the
          source.
 
        * ukify inspect --json= changed its machine-readable output for UKIs
          with repeated PE sections and profiles. Repeated sections are now
          emitted as arrays, and multi-profile UKIs describe per-profile
          sections below a new "_profiles" array. Consumers that assumed each
          section name occurred at most once should be updated.
 
        * systemd-repart no longer honours the legacy "FactoryReset" EFI
          variable in the initrd. The newer "FactoryResetRequest" EFI variable
          remains supported and should be used instead.
 
        * As part of changes to systemd-sysupdate, the existing
          "systemd-sysupdate.service" and "systemd-sysupdate.timer" units –
          which periodically ran "systemd-sysupdate" to update the host system –
          have been renamed to "systemd-sysupdate-update.service" and
          "systemd-sysupdate-update.timer" respectively. Compatibility symlinks
          have been provided. This clears the way for a new
          "systemd-sysupdate@.service" unit for varlink activation of sysupdate.
 
        * The default OpenPGP keyring lookup now combines the vendor keyring in
          /usr/ with the local keyring in /etc/. Deployments that relied on a
          single /etc/ keyring to exclude vendor trust should set
          $SYSTEMD_OPENPGP_KEYRING explicitly.
 
        * TPM-sealed credentials are now pinned to the TPM's SRK. This prevents
          MITM interposer attacks from stealing the decrypted credentials by
          ensuring that communication with the TPM is protected by a private key
          only known to the same TPM the credential was sealed to. Additionally,
          this allows usage of TPM-sealed credentials when the TPM's owner
          hierarchy is protected with a PIN.
 
          TPM-bound credentials minted after this change are not recognized by
          older systemd versions, however new systemd continues to accept
          credentials that were created before this change.
 
        * The "tpm2-measure-bank=" crypttab option of systemd-cryptsetup is
          now deprecated and has no effect. Volume key measurements are
          performed via varlink call to systemd-pcrextend, which automatically
          selects all suitable TPM2 PCR banks; the banks can no longer be
          restricted per volume.
 
        * The legacy socket for controlling systemd-udevd has been removed, and
          udevadm now unconditionally uses Varlink IPC. The legacy UNIX socket
          /run/udev/control, which is sometimes used to check whether udevd is
          running, has been replaced with a symlink to the Varlink socket. This
          change does not mean the legacy protocol is supported on this path;
          instead, the symlink is maintained to ensure the socket is bound
          under the same conditions as before.
 
        * UnsetEnvironment= is now applied after environment variable expansion
          of ExecStart= and related command lines. Variables listed in
          UnsetEnvironment= can therefore still be used in Exec*= command
          lines, while being absent from the environment passed to the executed
          process. This fixes an unintended coupling between command expansion
          and final environment cleanup. Units that relied on whole-word $VAR
          references being dropped should be updated.
 
        * Meson option compatibility changed for static/multicall builds:
          -Dbuild-executor-shared=single has been replaced by
          -Dsystemd-multicall-binary=true, and -Dstandalone-binaries= now
          takes a comma-separated list of patterns naming binaries whose
          .standalone variant should be installed. The old true/false values
          are translated to the previous defaults but are deprecated. Existing
          Meson build directories that used the old options need to be
          recreated.
 
        * sd-event rate-limit timers, and thus manager timers such as
          StartLimitIntervalSec=, now use CLOCK_BOOTTIME instead of
          CLOCK_MONOTONIC. Time spent suspended therefore counts toward
          rate-limit expiry. After upgrade/reexecution, already-serialized
          rate-limit state from older managers may expire once earlier than it
          would have previously, because the stored timestamp is compared
          against the new clock base.
 
        * journalctl now rejects -F/--field together with filters such as
          unit, boot, time, cursor or grep filters. Previously field listing
          used sd_journal_query_unique(), which cannot represent those filters,
          so the filters were silently ignored. Scripts relying on filtered
          field lists should be adjusted to query entries explicitly.
 
        * The systemd-sysupdate binary has been moved back from bindir to
          libexecdir, as more breaking changes are forthcoming and it is thus
          again marked as experimental. It will be stabilized in a future
          release. The libexecdir path is the most compatible way to call it.
 
        * The "extensions" field of "systemd-sysext status --json=" and
          "systemd-confext status --json=" is now always an array. A hierarchy
          that is not merged reports an empty array, instead of the string
          "none" returned previously. In the human-readable table output a
          hierarchy that is not merged is now shown as "-" instead of "none".
 
        * The JSON payload returned by systemd-sysupdated's DescribeFeature()
          method and by the "systemd-sysupdate features" command changed the
          "name" key to "id". The "documentation" key is now an array of
          strings. Consumers of this interface must update their JSON parsing
          accordingly.
 
        * The PrivateUsersOwnership= setting in .nspawn files now defaults to
          "auto" when PrivateUsers= is "pick" or "managed", matching the
          --private-users-ownership= command line switch. Previously "pick"
          defaulted to "chown", and "managed" to "off".
 
        Changes in the system and service manager:
 
        * The manager now embeds a basic set of unit files (basic.target,
          sysinit.target, multi-user.target, reboot.target, shutdown.target,
          systemd-poweroff.service, etc.) and will use them if files cannot be
          loaded from disk. Unit files and masks found on disk take precedence
          over these built-in fallbacks. This makes the binary slightly more
          robust and allows a container to be started with systemd as PID1
          without installing any unit files.
 
        * systemd can now be built as a single statically linked PID 1/executor
          binary suitable for very small containers by configuring Meson with
          --default-library=static --prefer-static -Dbuild-static=true
          -Dsystemd-multicall-binary=true. Such builds do not use dlopen() to
          load optional libraries at runtime and use simplified passwd/group
          file lookup instead of NSS.
 
        * Service units gained RestartRandomizedDelaySec=, which adds a
          uniformly distributed extra delay to automatic restarts on top of
          RestartSec= and optional exponential backoff. This can spread out
          synchronized failure/restart storms.
 
        * Slice units gained ActivatingConcurrencyMax= to limit how many units
          in a slice hierarchy may be in the activating state at the same time.
          Further activations are queued and started automatically as slots are
          freed.
 
        * The manager gained the D-Bus method EnqueueUnitJobMany() to queue
          start/stop/restart/reload jobs for multiple units in one transaction,
          so ordering dependencies between named units are honored independent
          of command-line order. systemctl and portablectl use it when
          available and fall back to per-unit calls with older managers or
          unsupported cases.
 
        * The io.systemd.Unit Varlink StartTransient() method now accepts a
          larger set of transient-unit properties, including more Exec context
          settings for users, groups, credentials, root images/directories,
          namespace paths, common sandboxing booleans, UMask=, OOMScoreAdjust=,
          CollectMode=, and standard input/output/error file descriptors passed
          with the Varlink call.
 
        * SecureBits= now recognizes "no-cap-ambient-raise",
          "exec-restrict-file", "exec-deny-interactive", and their "-locked"
          variants.
 
        * NUMAPolicy= now accepts "preferred-many" and "weighted-interleave".
          The former requires Linux 5.15 or newer, the latter requires Linux
          6.9 or newer and uses weights configured in the kernel's
          weighted_interleave sysfs files.
 
        * ConditionCPUFeature= now supports arm64 features reported via ELF
          hwcaps, and CPU feature names may be prefixed as ARCH.FEATURE (for
          example: arm64.bti) to disambiguate mixed-architecture deployments.
 
        * Socket units' MessageQueueMessageSize= now accepts byte values
          suffixed with the usual IEC prefixes (Ki, Mi, Gi, etc.).
 
        * Service units gained LUOSession= to have systemd create Live Update
          Orchestrator sessions and pass them through the file descriptor
          store interface.
 
        * PID 1 now exposes a new D-Bus property KExecCount that indicates the
          number of kexec reboots the system has undergone since the last real
          boot. It also exposes shutdown timestamps of the previous kexec
          cycle. (Only if LUO is available). 'systemd-analyze time' uses them
          to report kexec/live-update timing.
 
        * RestrictFileSystemAccess= now permits execution from overlayfs
          mounts whose file data resides on a signed and verified
          dm-verity-protected filesystem, while still denying files in a
          writable upper layer. This requires kernel v7.2 or newer; on older
          kernels execution from overlayfs mounts remains denied entirely.
 
        * AF_UNIX sockets managed by .socket units can now be marked with
          extended attributes. Socket units gained XAttrEntryPoint=,
          XAttrListen= and XAttrAccept= helpers for tagging entrypoint,
          listening and accepted sockets. This requires socket inode xattr
          support available since Linux kernel 7.0.
 
        * The common io.systemd.service Varlink interface gained GetLogLevel(),
          matching the existing SetLogLevel() method. It is exposed by
          systemd's service daemons, and 'resolvectl log-level' now uses it to
          report the current maximum log level.
 
        * The hostname setup logic now supports a new type of wildcard
          character when configuring hostnames (for example via /etc/hostname).
          In addition to the existing '?' wildcards (which resolve to a hex
          character derived from the machine ID by hashing), there is now '$'
          which resolves to a word picked from a word list file also via
          hashing. This may be used to automatically assign memorable hostnames
          to systems in a random but deterministic and stable fashion. Three
          such word lists are shipped by default.
 
        * The new systemd.credentials_boot_policy= kernel command line option
          controls when boot credentials encrypted with the "null" key are
          accepted. It accepts "strict" (never), "tofu" (first boot or no
          TPM2), "relaxed" (SecureBoot disabled or no TPM2, the default and
          previous behavior), and "off" (always). It does not affect
          credentials encrypted with host or TPM2 keys.
 
        * systemd now imports credentials from /.extra/system_credentials/ in
          the initrd as trusted system credentials, available via
          $CREDENTIALS_DIRECTORY and LoadCredential=. The existing
          /.extra/credentials/ and /.extra/global_credentials/ locations remain
          treated as untrusted boot-loader credentials, routed to
          $ENCRYPTED_CREDENTIALS_DIRECTORY for LoadCredentialEncrypted=.
 
        Changes in sd-varlink and varlinkctl:
 
        * Varlink socket inodes are tagged with an
          user.varlink= xattr where supported, indicating whether they are
          entrypoint, listener or accepted sockets.
 
        * varlinkctl gained a command list-sockets to discover all tagged local
          Varlink entrypoints.
 
        Changes in systemd-boot, systemd-stub & bootctl:
 
        * Random seed handling now respects the "read-only" ESP/vFAT file flag.
          If the backing volume or /loader/random-seed file is marked
          read-only, systemd-boot skips updating the seed and does not credit
          it to the kernel entropy pool, since a seed that cannot be rotated
          would be reused on later boots. This may be used to turn off
          random-seed handling in replicated images. Similarly, the boot
          counting logic is disabled if boot entry files or the file system
          backing them are marked as read-only.
 
        * systemd-stub now sets up all detected consoles instead of only the
          best one. If no console= option was supplied, graphical console
          detection is also preferred over the virtio console heuristic, making
          VM boots with graphical output and unrelated virtio devices more
          reliable while still enabling other detected consoles.
 
        * bootctl gained a new "link-auto" verb, which operates like
          "bootctl link" but automatically discovers the UKI and resources to
          link in /usr/lib/systemd/uki/ and similar directories in /run/ and
          /var/. This logic can be executed automatically when
          systemd-sysupdate completes an update by enabling the
          systemd-sysupdate-notify-bootctl.socket unit.
 
        Changes in systemd-hostnamed:
 
        * The Describe() method (for both D-Bus and Varlink) now exposes the
          parsed machine tag list as a new MachineTags field, so that Varlink
          clients no longer need to parse the raw TAGS= line out of
          MachineInformationData themselves.
 
        * systemd-machine-tag@.service and the SYSTEMD_MACHINE_TAG= udev
          property allow devices to add machine tags to the host early in
          boot. This is useful for automatically tagging a machine based on
          available hardware, for example by setting the udev property via a
          hwdb entry, matching a device (for example the kernel's DMI/SMBIOS
          device).
 
        * systemd-hostnamed's Varlink interface gained setter methods for the
          transient, static and pretty hostnames, icon name, chassis,
          deployment, and location fields.
 
        * systemd-hostnamed's Varlink interface gained a new call SetTags() for
          setting/unsetting machine tags. Its D-Bus API gained
          AddAndRemoveTags() as a call for atomically setting some tags and
          unsetting others. 'hostnamectl tags' accepts +TAG and -TAG arguments
          to add or remove tags atomically without replacing the whole list.
 
        * Machine tags may now be written as key=value pairs. The same key may
          not be assigned multiple distinct values, and
          ConditionMachineTag=key=* matches any value assigned to key.
 
        Changes in systemd-networkd:
 
        * The [Match] sections of .link, .network and .netdev files gained
          MachineTag=, matching the same tag syntax as ConditionMachineTag= in
          unit files. Or in other words: network configuration may now be
          conditioned on the presence or absence of specific machine tags.
 
        * The IPv6ProxyNDPAddress= setting in the [Network] section of .network
          files has been renamed to ProxyNeighbor=, and remains accepted as a
          deprecated alias. ProxyNeighbor= can add both IPv4 and IPv6 addresses
          to the kernel's neighbor proxy tables.
 
        * "networkctl reload" gained --no-reconfigure to reload .network and
          .netdev files from disk without reconfiguring existing links. It uses
          the new io.systemd.Network.Reload() Varlink method and its optional
          reconfigureLinks= parameter. Plain reload falls back to the older
          service Reload() method if needed, while --no-reconfigure requires a
          new enough networkd.
 
        * The IPoIB=, IPVLAN=, IPVTAP=, MACsec=, MACVLAN=, MACVTAP=,
          Tunnel=, VLAN=, VXLAN= and Xfrm= stacked-netdev settings in
          .network files now accept whitespace-separated lists of netdev names.
          Assigning an empty string clears all previously assigned stacked
          netdevs.
 
        Changes in systemd-journald and journalctl:
 
        * Journal readers now attempt to salvage intact entries from active
          journal files whose header/tail state was truncated by an unclean
          shutdown during high-rate writes. Writable opens remain strict and
          still rotate away files whose state is unsafe for appending.
 
        * Journal sealing/FSS support was moved from libgcrypt to OpenSSL.
          libsystemd no longer links against libgcrypt, and sealed journal
          files remain readable when sealing support is unavailable (seal tag
          verification is skipped in that case).
 
        * The io.systemd.JournalAccess.GetEntries Varlink method gained a
          follow mode. When called with "follow=true" it returns the requested
          backlog and then keeps the method call open, streaming new matching
          entries like 'journalctl -f' until the client disconnects.
 
        Changes in systemd-coredump and coredumpctl:
 
        * systemd-coredump now supports the kernel coredump socket protocol
          available since Linux 6.17. On supported kernels,
          systemd-coredumpd.service listens on /run/systemd/coredumpd/kernel,
          and systemd-coredump-register.service registers the socket through
          kernel.core_pattern. Older kernels continue to use the per-coredump
          systemd-coredump@.service path. Changes to coredump.conf require
          reloading systemd-coredumpd.service, and deployments that set a
          custom kernel.core_pattern should stop and disable that service.
 
        * The coredumpctl dump and debug commands now honor --json= and include
          the selected coredump's metadata in the requested JSON format.
 
        Changes in systemd-resolved and resolvectl:
 
        * 'resolvectl openpgp' and 'resolvectl tlsa' now support --json= when
          querying their implied record types. Mismatched explicit --type=
          values are rejected, unauthenticated cryptographic records are not
          emitted as JSON, and DNS lookup failures in JSON mode are reported as
          structured JSON objects with the resolver error fields.
 
        * resolvconf compatibility mode now rejects malformed stdin input
          before sending updates to systemd-resolved, leaving the existing
          per-link DNS state unchanged on parse failure.
 
        * The io.systemd.Resolve.Monitor Varlink interface gained FlushCaches()
          and ResetServerFeatures() methods, and resolvectl uses them where
          available for its flush-caches and reset-server-features verbs.
 
        Changes in systemd-udevd:
 
        * systemd-udevd now runs workers in a sibling "workers" cgroup when
          possible. This keeps them separate from the manager process and lets
          udev clean up remaining worker processes atomically with cgroup.kill
          on kernels that support it.
 
        * sd-device and systemd-udevd now ignore invalid ordinary kernel
          properties in uevents and uevent files instead of rejecting the whole
          otherwise-valid event. Required and typed properties remain strict.
 
        Changes in systemd-homed and homectl:
 
        * New fscrypt-backed home directories now default to fscrypt v2
          policies, so the master key is installed in the filesystem keyring
          and is visible across mount namespaces and containers. Existing v1
          homes remain supported for unlock, rekey and deactivation, but there
          is no in-place v1-to-v2 policy upgrade.
 
        * LUKS-backed home password rotation now adds all replacement keyslots
          before removing the old ones. If adding a new slot fails, the old
          slots are left intact, avoiding lockout on enrollment failure.
 
        * homectl gained a new --recovery-key-file= option which may be used to
          write the recovery key into a file in an atomic and reasonably safe
          fashion, when enrolling one.
 
        Changes in systemd-tmpfiles:
 
        * systemd-tmpfiles now rejects non-empty argument fields for
          tmpfiles.d/ line types that do not use the argument field. This is a
          compatibility change: fields that were previously ignored after a
          warning will now cause the tmpfiles.d configuration to be rejected,
          so downstreams should audit shipped tmpfiles.d snippets before
          upgrading.
 
        * The r and R tmpfiles.d/ line types now honor the age field when
          systemd-tmpfiles is invoked with --clean.
 
        * tmpfiles.d/ lines now understand the %D specifier. In system mode it
          resolves to /usr/share, in user mode it resolves to $XDG_DATA_HOME.
 
        Changes in the TPM Subsystem:
 
        * The way that NvPCRs are anchored has been improved. They are now
          created with a write policy which only permits the initial extend to
          be performed from the initrd environment, and no longer rely on a
          secret that is sealed to the TPM (the "anchor secret"). For NvPCRs to
          work, all NvPCR definitions (/usr/lib/nvpcr/*.nvpcr) must now be
          shipped in the UKI. The UKI must also embed a signed PCR policy that
          is bound to the initrd boot phase and with the "initrd" policy
          reference. This can be created by supplying the --sign-initrd-pcrs
          option to ukify. NvPCRs that are not defined or initialized in the
          initrd will fail to initialize lazily during runtime if an attempt is
          made to extend them with systemd-pcrextend. Existing NvPCRs that were
          created with previous systemd versions will be automatically
          upgraded, and the existing "anchor secret" stored in /var/lib and the
          ESP/XBOOTLDR will be removed.
 
        * systemd-tpm2-setup now creates and persists a TPM Endorsement Key
          (EK) when possible. If an EK certificate is available, the generated
          EK is verified against it before being persisted. Supported
          high-range EK templates are preferred, allowing the endorsement
          hierarchy to remain protected while the persistent EK is used.
 
        * TPM PCR-bank selection now considers SHA384 and SHA512 PCR banks in
          addition to SHA256 and SHA1. SHA256 remains preferred for
          compatibility, followed by SHA384, SHA512 and finally SHA1.
          SHA384/SHA512 are selected only when SHA256 is unavailable.
 
        * NvPCR definitions may now individually select whether to set the
          TPMA_NV_ORDERLY flag (which remains enabled by default). The
          "hardware" NvPCR will now disable by default, as this reduces the RAM
          use on the device, and is generally preferable for NvPCRs that are
          only written to a small number of times each boot, instead of
          repeatedly.
 
        Changes in systemd-pcrlock:
 
        * systemd-pcrlock predict and make-policy gained --strict=. When true,
          the command fails if any requested PCR cannot be included in the
          prediction or policy instead of silently dropping it.
 
        * systemd-pcrlock's Varlink API gained methods to list known pcrlock
          components and to generate or remove pcrlock data for firmware code,
          firmware configuration, Secure Boot policy, and Secure Boot authority
          measurements. (This is useful to relax policy before an
          fwupd-prepared firmware update.)
 
        * Support has been added to redo the pcrlock policy whenever a
          systemd-sysupdate update operation completed. This may be enabled via
          the systemd-sysupdate-notify-pcrlock.socket unit.
 
        * systemd-pcrlock will now display hashes abbreviated to 7 characters
          (similar to git) by default. Use --full to get the full output back.
 
        Changes in systemd-cryptsetup/systemd-cryptenroll:
 
        * TPM2 PIN enrollment can now harden PINs with Argon2id. The normal
          --tpm2-with-pin=yes mode derives key material with Argon2id, so a
          compromised TPM alone is not sufficient to recover volume key
          material. --tpm2-with-pin=direct selects the legacy direct/PBKDF2
          compatible mode. Argon2 parameters can be tuned with
          --tpm2-argon2id-memory=, --tpm2-argon2id-iterations=,
          --tpm2-argon2id-parallelism= and --tpm2-argon2id-iter-time=.
 
        * systemd-cryptenroll now supports an optional "first-boot" wizard for
          enrolling additional unlock mechanisms. This is intended to be used
          in TPM-enabled scenarios which default to unattended TPM-based disk
          encryption, but where enrollment of additional mechanisms to decrypt
          the disks for recovery purposes shall be suggested to the user. This
          wizard is provided by the new systemd-cryptenroll-firstboot.service
          unit.
 
        * systemd-cryptenroll now implements a Varlink API which allows
          programmatic enrollment of additional keys via IPC.
 
        * systemd-cryptenroll gained a new --unlock-empty switch for attempting
          to unlock the disk via an empty password, instead of prompting the
          user for one. There's also a new --unlock-headless switch, which is
          equivalent to --unlock-tpm2 if a TPM is available, and otherwise
          equivalent to --unlock-empty. This is supposed to be useful in
          scenarios where the system defaults to unattended encryption with a
          TPM if one is available, and an empty password otherwise.
 
        * Signed TPM PCR policies can now be scoped by opaque policy
          references. systemd-measure gained --policyref=, and matching
          --tpm2-public-key-policyref= options in systemd-cryptenroll and
          systemd-repart select the reference required when a TPM object is
          unlocked. This allows, for example, an initrd-only signed policy,
          which ukify can generate via --sign-initrd-pcrs without reusing the
          same signature for later boot phases.
 
        Changes in systemd-repart and systemd-dissect:
 
        * systemd-repart and systemd-dissect now support DDIs where a data
          partition combines Encrypt= with Verity=data. In this mode repart
          writes LUKS2 ciphertext first and then builds dm-verity metadata over
          that ciphertext, so verity forms the outer authenticity envelope and
          LUKS2 the inner encryption layer. (Or in other words, this provides
          authenticated confidentiality).
 
        * The io.systemd.Repart ListCandidateDevices() Varlink method gained a
          subscription mode. With the "subscribe" parameter set, clients
          receive the initial device list, a ready marker, and then live
          add/remove events as block devices become suitable or disappear.
          Replies and errors also include additional sizing details such as
          currentSizeBytes, minimalSizeBytes and needFreeBytes where
          appropriate.
 
        * systemd-repart now honors $SOURCE_DATE_EPOCH while populating images.
          Timestamps copied from source trees are clamped to the epoch, and a
          validated value is forwarded to mkfs and mtools, improving
          reproducibility of generated file systems. Byte-for-byte
          reproducibility may still require --offline=yes and suitable
          file-system tooling.
 
        * systemd-repart gained a new --cow= switch that controls whether
          COW/NOCOW shall be enabled for generated image files, on file systems
          that support such a file flag (i.e. on copy-on-write file systems
          such as btrfs).
 
        Changes in systemd-nspawn & systemd-vmspawn:
 
        * When systemd-nspawn creates both user and network namespaces, it now
          initializes net.ipv4.ping_group_range from the container's mapped GID
          range. This permits unprivileged container users to create ICMP echo
          sockets when their GIDs are mapped. Externally supplied network
          namespaces are left unchanged.
 
        * systemd-vmspawn --coco= now supports Intel TDX in addition to
          SEV-SNP. Confidential VM firmware is selected via QEMU firmware
          descriptors, TDVF/OVMF feature requirements are verified, and
          --secure-boot=yes can opt into firmware with pre-enrolled keys, which
          is required because confidential VM firmware has no writable NVRAM for
          runtime key enrollment.
 
        * With --coco=sev-snp, systemd-vmspawn now delivers credentials by
          appending a cpio archive to the initrd under
          /.extra/system_credentials/, so they are covered by the SEV-SNP
          launch measurement. The channel provides integrity and attestability
          but not confidentiality from the host or VMM, and requires guest
          systemd support for importing credentials from
          /.extra/system_credentials/.
 
        Changes in systemd-report:
 
        * The "generate" and "upload" commands of systemd-report can now
          cryptographically sign the reports they produce. Signatures are
          acquired from backend Varlink signers linked into
          /run/systemd/report.sign/. The new "--sign=" option selects the
          signing policy: "no" (the default) emits an unsigned report;
          "best-effort" attaches whatever signatures can be acquired but never
          fails; "require-one" requires at least one signature; and
          "require-all" requires every available signer to succeed. A signed
          report is emitted as a JSON-SEQ stream: the report object followed by
          one signature object per acquired signature.
 
        * Three report signing backends are provided: systemd-report-sign-plain
          maintains a local software Ed25519 key, systemd-report-sign-tsm
          obtains Confidential Computing attestation reports through configfs
          TSM, and systemd-report-sign-tpm2 emits TPM2 PCR quotes, NvPCR
          attestations and audit-session attestations together with measurement
          logs and public key material.
 
        * systemd-report can now run as a socket-activated io.systemd.Report
          Varlink service. The service exposes Generate() for unsigned reports
          and GenerateSigned() for Base64-encoded signed JSON-SEQ report data.
 
        * Report metrics gained additional inputs from systemd-report-basic, as
          well as the networkd, PID 1, IMDS and the journalctl providers.
          Reports now include load averages, swap, aggregate CPU and wait time,
          memory use, system PSI, disk I/O, kernel and systemd version, boot
          timestamps, instance metadata, network interfaces addresses, the last
          10 high-priority journal entries.
 
        * A new report provider systemd-report-files has been added that will
          report any file placed in (or symlinked into)
          {/etc,/run,/var/lib}/systemd/report.files/ as a metric in the report.
 
        Changes in libsystemd:
 
        * libsystemd gained new sd-varlink upgrade helper APIs:
          sd_varlink_call_and_upgradeb() and
          sd_varlink_call_and_upgradebo() build call parameters with the JSON
          builder, and sd_varlink_respond_and_upgrade(),
          sd_varlink_respond_and_upgradeb(), sd_varlink_respond_and_upgradebo()
          plus sd_varlink_bind_upgrade() implement asynchronous server-side
          protocol upgrades without blocking while the final reply is flushed.
 
        * sd-dlopen.h gained SD_ELF_NOTE_DLOPEN_ANCHORED(), which associates
          dlopen metadata notes with a generated anchor symbol so unused notes
          can be garbage-collected with unreachable code when linking with
          --gc-sections. Full builds now require binutils >= 2.35 and
          LLVM/clang >= 18 for the assembler features used by these notes.
          Clang older than 18 falls back to the non-anchored form.
 
        * sd_id128_from_string() now accepts RFC 4122 UUID URNs with the
          "urn:uuid:" prefix in addition to plain 32-character IDs and dashed
          UUID strings.
 
        systemd-userdbd and userdbctl:
 
        * userdb group records gained support for alias names, similar to the
          existing support for alias names for user records.
 
        * userdb services may now advertise via xattrs on their Varlink AF_UNIX
          entrypoint socket inodes for which subset of UID/GID ranges or
          user/group names (via glob expression) they shall be queried. This is
          an efficient way to suppress unnecessary userdb backend IPC requests.
          All of systemd's own userdb providers now set the appropriate xattrs.
 
        Changes in systemd-sysupdate:
 
        * systemd-sysupdate will now maintain a persistent database of all
          files it installed. Files that have previously been installed by
          systemd-sysupdate but to which no current match pattern applies, are
          now automatically removed when the new "cleanup" verb is called. This
          means match patterns now become a stronger form of ownership
          declaration.
 
        * systemd-sysupdate will now synchronously notify all services that
          have a Varlink entrypoint socket bound into the
          /run/systemd/sysupdate/notify/ directory about completed updates.
 
        * systemd-sysupdate gained a basic Varlink API. Currently you may use
          it to list features, and components/targets and check for newer
          versions.
 
        * systemd-sysupdate's support for "components" has been extended: a
          per-component metadata file (*.component) is now loaded if it exists.
          It can be used to enable/disable the component, and may carry a human
          readable description as well as documentation links. The
          'systemd-sysupdate components' command will show the component
          description strings.
 
        * systemd-sysupdate gained support for new
          enable-feature/disable-feature commands that may be used to enable
          disable features. (Previously these operations where available in
          updatectl, i.e. in the frontend, but now they are available in the
          service itself.)  Similar new commands
          'enable-component'/'disable-component' have been added.
 
        * Components and features acquired a "suggestion" concept. This is
          inspired by unit file's ConditionXYZ= settings, and allows marking
          components and features as suggested for installation under various
          conditions (such as architecture, used virtualization, host identity,
          kernel command line, version, firmware properties and machine tags).
          All components currently suggested can be enabled via
          'systemd-sysupdate enable-component --component-suggested' (and
          similarly for features). Note that by providing hwdb files that match
          vendor/product identity and set machine tags, it is now possible to
          automatically determine which system extensions containing drivers
          shall be installed. If the new systemd-sysupdate-auto-enable.service
          unit is enabled, it will automatically enable all suggested
          components and features before each unattended systemd-sysupdate run.
 
        * systemd-sysupdate's "update" command acquired a new --component-all
          switch for updating all installed components in one go.
 
        Changes in systemd-sysctl:
 
        * systemd-sysctl gained --verify, which reads each sysctl value back
          after a successful write and treats read-back failures or mismatches
          as write failures. This mode is off by default because kernels may
          normalize values. Call it with --strict and targeted configuration
          when exact verification is desired.
 
        * systemd-sysctl gained --save=FILENAME and --revert=FILENAME for
          managing runtime configuration snippets below /run/sysctl.d/.
          --save= writes the loaded settings before applying them, and
          --revert= removes the named runtime snippet before loading the
          remaining configuration.
 
        Changes in systemd-sysext:
 
        * systemd-sysext will now honour new fields EXTENSION_RESTART_UNITS=
          and EXTENSION_RELOAD_OR_RESTART_UNITS= in the extension-release file.
          It may list units to restart or reload when a sysext is applied.
 
        * systemd-sysext can now automatically refresh if a systemd-sysupdate
          run completed. To activate this feature, enable the
          systemd-sysupdate-notify-sysext.socket unit.
 
        Other changes:
 
        * Support for OpenSSL 4 has been added.
 
        * systemd-timesyncd now resolves NTP server names through
          systemd-resolved's Varlink API when available. It suppresses DNSSEC
          validation only until the local clock has synchronized, falls back to
          getaddrinfo() if resolved is unavailable, and avoids AAAA lookups
          when IPv6 is disabled.
 
        * systemd-machined: privileged callers that authenticate through polkit
          as administrators may register machines for other users. The Register
          Varlink method now derives the supervisor process from the caller and
          ignores the deprecated supervisor/supervisorProcessId input fields.
          systemd-machined's OpenShell accepts numeric UIDs again.
 
        * systemctl kexec gained --kernel-cmdline-reuse, which appends the
          currently running kernel command line to the kexec kernel command
          line. --kernel-cmdline= may now be specified more than once. The
          final command line is assembled from the boot loader entry, optional
          reused current command line, and explicit additions. After that
          byte-identical duplicate arguments are removed, keeping the last
          occurrence. These options are rejected if a kexec kernel is already
          loaded.
 
        * The "systemd.firstboot=" kernel command line option now accepts the
          special value "headless" in addition to a boolean. Like "no", it
          suppresses all interactive prompts, but unlike "no" it still performs
          non-interactive auto-configuration that requires no user input (such
          as selecting the sole installed locale, or applying settings provided
          via credentials). This is useful for unattended installations that
          should be provisioned as far as possible without blocking on a
          prompt.
 
        * Tools that create or modify files in offline roots/images now use the
          SELinux policy from the target tree, and strip the alternate-root
          prefix for label lookups. This fixes labels written by
          systemd-firstboot, systemd-sysusers, systemd-tmpfiles, and related
          helpers when operating with --root= or --image=. If the target
          /etc/selinux/config names an SELINUXTYPE=, that policy must be
          present in the target.
 
        * run0 gained sudo-compatible -k/--reset-timestamp,
          -K/--remove-timestamp, and -v/--validate options to revoke or
          refresh temporary polkit authorizations. It also accepts
          -n/--non-interactive as an alias for --no-ask-password.
 
        * bootctl, systemd-oomd, systemd-repart, systemd-sysusers, and
          systemd-tmpfiles now accept -n as a short option for --dry-run.
 
        * systemd-escape gained --stdin. It reads strings from standard input,
          one per line, and writes one escaped/unescaped result per line.
          Command line strings and --stdin are mutually exclusive.
 
        * vconsole configuration gained FONT_SCALE=, with matching
          vconsole.font_scale= kernel command line and vconsole.font_scale
          credential support. Values may be written as factors or percentages.
          Currently only 100% and 200% are supported, with other values clamped
          to the nearest supported scale.
 
        * kernel-install gained entry_name_format= in install.conf and the
          matching $KERNEL_INSTALL_ENTRY_NAME_FORMAT environment variable. The
          resolved name is exported to plugins as $KERNEL_INSTALL_ENTRY_NAME
          and is used as the base name for generated UKI files and loader entry
          snippets.
 
        * All command line tools that implement --help now expose
          machine-readable command metadata via --introspect-cli as JSON,
          generated from declarative verb/option descriptions. Scripts that
          need command/option discovery may consume this interface instead of
          parsing human --help output. This feature is experimental and the
          JSON schema will most likely change in the future release until a
          future stabilization announcement.
 
        * OpenPGP verification of downloaded resources now accepts signing
          subkeys embedded in signatures when they are certified by a trusted
          primary key, allowing signing-key rotation without distributing each
          subkey separately. $SYSTEMD_OPENPGP_KEYRING may select a custom
          keyring for unprivileged pulls or custom sysupdate definitions. When
          unset, the vendor keyring in /usr/ and the local keyring in /etc/ are
          combined.
 
        * New dm-clone boot integration was added. /etc/clonetab entries are
          translated by systemd-clonesetup-generator into
          systemd-clonesetup@.service units pulled in by clonesetup.target, and
          systemd-clonesetup creates and removes the device-mapper clone
          devices. clonetab entries may set region-size= to control hydration
          granularity.
 
        * systemd-sysinstall gained a socket-activated io.systemd.SysInstall
          Varlink API. It can run OS installations with progress notifications,
          pass installation credentials and locale/keymap/timezone choices, and
          list candidate block devices with fit and sizing information for
          installer frontends.
 
        * pam_systemd.so gained inhibit= and inhibit-why= options, plus the
          $XDG_SESSION_INHIBIT and $XDG_SESSION_INHIBIT_WHY environment
          variables. When set, pam_systemd takes block inhibitor locks for the
          lifetime of the PAM session and releases them when the session is
          closed.
 
        * FIDO2 support has been improved to show the PIN left attempts counter
          when prompting for a PIN.
 
        Contributions from: Adam Dinwoodie, Akarithos, Aleksa Sarai,
        Alexander Egorov, Alexandre Saou, Alexey Shabalin,
        Alperen Kılıç, Américo Monteiro, Ananth, Ananth Bhaskararaman,
        Anders Jonsson, Andre Eikmeyer, Andrei Stepanov, Andres Beltran,
        Andres Diaz, Andrew Dunn, Ani Sinha, Anton Tiurin,
        Antonio Alvarez Feijoo, Arian van Putten, Arif Budiman,
        Aritra Basu, Armaan Sandhu, Arseniy Kostevich, Aswin Murugan, Axel,
        Baurzhan Muftakhidinov, Beniamino Galvani, Bone NI, Cathal Mullan,
        Chris Ayoub, Chris Coulson, Chris Down, Chris Hasiński,
        Christian Brauner, Christian Glombek, Christian Hesse,
        Clayton Craft, Confusedphoton, CrtlTom, Daan De Meyer,
        Daniel Wagenknecht, Daniel28972897, Dara Adib,
        David 'equinox' Lamparter, David Tardon, Davide Cavalca,
        Davis-Rippon, Deniz Gözen, DongShengyuan, EJ Campbell,
        Edvard Jaegtnes, Elliot Berman, Emanuele Petriglia, Emanuele Rocca,
        Emilio Sepulveda, Eric Boucher, Eric Curtin, Ettore Atalan,
        Fabian Vogt, Fco. Javier F. Serrador, Fede2782, Franck Bui,
        Frantisek Sumsal, Gabriel, Guillaume Kehren, Hans de Goede,
        Hashim Khan, Hijae Song, Hilal Rahmatullah, Honza Hejzl,
        Hüseyin Kutsi Balcı, Ivan Kruglov, Jakub Stasiak, Jan Kuparinen,
        Jan Macku, Jason Yang, Jelle van der Waa, Jesse Guo,
        Jonas Dreßler, Josh Hoffer, Jouke Witteveen, Julian Bouzas,
        Julian Sparber, Justin Cabral, Jörg Behrmann, Kai Lüke,
        Kai Wohlfahrt, Kakueeen, Kevin Charpentier, KingPointer,
        Konstantin Shkolnyy, Lars Sjöstrom, Lefteris T,
        Lennart Poettering, Liu Zhangjian, Liu Zheng, Lokesh Patel,
        Luca Boccassi, Lucas Adriano Salles, LucasTavaresA, Ludwig Nussel,
        Luna Jernberg, Léane GRASSER, Lương Việt Hoàng,
        Marek Adamski, Markus Boehme, Martin Pitt, Martin Srebotnjak,
        Matthew Schwartz, Matthias P. Walther, Michael Renner,
        Michael Vogt, Michal Koutný, Michal Sekletár, Michal Čihař,
        Mike Yuan, Morteza Pourkazemi, Muhammed Yakup Özlük, Méven Car,
        Nandakumar Raghavan, Nicholas Lim, Nick Rosbrook, Oğuz Ersen,
        Parag Gupta, Paul Menzel, Paul Meyer, Pavel Borecki, Petru Rebeja,
        Philip Withnall, Popax21, Pranay Pawar, Pétur, Rafael Fontenelle,
        Rafael Passos, Raito Bezarius, Ramses, Rasmus Villemoes,
        Renan Pagani, Robin Richard Rajan, Rocker Zhang, Roman Vinogradov,
        Ronan Pigott, Ross Burton, Semyon Klintsov, Sergey A., Shihao Ren,
        ShiroKSH, Shivank Sharma, Simon de Vlieger, Simran Singh, Sinity,
        Siteshwar Vashisht, Skye Soss, Srijan Keshri, Stefan Agner,
        Stephanie Wilde-Hobbs, Sven Joachim, Syed Mohammed Nayyar,
        Temuri Doghonadze, Thomas Kopp, Tim Culverhouse, Tim Schumacher,
        Timon de Groot, Tobias Heider, Totoro, Umut Tezduyar Lindskog,
        Val Packett, Victor Verdan, Vsevolod Kozlov, Wang Yu, Yu Watanabe,
        Zbigniew Jędrzejewski-Szmek, acandoo, benkigeek, cidkidnix,
        countgitmick, cverorg, dirhamtriyadi, dongshengyuan, felix h, fuldeka,
        gkehren, gpenz, guolin, hanjinpeng, jmestwa-coder, joo es,
        logical-misha, lucia-huang7, lzwind, nl6720, quanyeyang, r-vdp,
        rashidmya, rodin-ia, totoro, tunaichao, wangzhaohui, wyu, xiahualiu,
        zhaoshuang, zouge666, Марко Костић (Marko Kostić),
        김인수
 
        — Lavanttal Tal, 2026/09/22
 
CHANGES WITH 261:
 
        Announcements of Future Feature Removals and Incompatible Changes:
 
        * systemd-logind's integration with the UAPI.1 Boot Loader
          Specification (which allows the systemctl reboot --boot-loader-entry=
          switch to work) so far has supported a special directory
          /run/boot-loader-entries/ which allowed defining boot loader entries
          outside of the ESP/XBOOTLDR partition for compatibility with legacy
          systems that do not natively implement UAPI.1. However, it appears
          that (to our knowledge) it is not actually being used by any project
          (quite unlike UAPI.1 itself, which found adoption far beyond
          systemd), and its implementation is incomplete. With the future 262
          release we intend to remove support for /run/boot-loader-entries/ and
          related interfaces, in order to simplify our codebase. Support for
          UAPI.1 is – of course – kept in place.
 
        * The experimental "systemd-sysupdated" D-Bus API is going to be
          removed in the next release. The plan is that in its place
          clients should directly talk to systemd-sysupdate (i.e. the backend
          of "systemd-sysupdated") via Varlink IPC. The "updatectl" tool will
          be reworked along these lines.
 
        Feature Removals and Incompatible Changes:
 
        * systemd-nspawn's --user= option has been renamed to --uid=. The -u
          short option continues to work. The old --user NAME and --user=NAME
          forms (with and without "=") are still accepted but deprecated; a
          warning is emitted suggesting --uid=NAME. The --user option (without
          an argument) has been repurposed as a standalone switch to select
          the user service manager scope, matching --system.
 
        * Several configuration fields in the io.systemd.Unit varlink interface
          that were previously exposed as plain strings have been converted to
          proper enum types. This adds type safety and IDL-level validation.
          The output wire format now uses underscores instead of dashes and
          plus signs in enum values (e.g. "tty-force" becomes "tty_force",
          "kmsg+console" becomes "kmsg_console"). The previous use of plain
          strings for these well-defined enumerations is considered a bug.
          Affected enum types: ExecInputType, ExecOutputType, ProtectHome,
          CGroupController, CollectMode, EmergencyAction, JobMode.
 
        * It was discovered that some of the events systemd-stub measures to
          the TPM were not also measured to the hardware CC registers (e.g.
          Intel TDX RTMRs) via EFI_CC_MEASUREMENT_PROTOCOL. In particular,
          devicetree, initrd, ucode addons and the UKI profile were only
          measured to the TPM. The missing measurements for CC have now been
          added; however, this changes the expected register values. This
          may need to be reflected in the attestation environments which use
          hardware CC registers (in place of TPM quotes).
 
        * systemd-nspawn gained a new --restrict-address-families= option (and
          corresponding RestrictAddressFamilies= setting in .nspawn files) to
          restrict which socket address families may be used in the container.
          This is currently opt-in. In a future version, the default will be
          changed to restrict socket address families to AF_INET, AF_INET6 and
          AF_UNIX.
 
        * A new service unit "systemd-pcrosseparator.service" will now measure
          a new separator measurement during early userspace into PCRs 0-7, 9,
          12-14, in order to isolate firmware/pre-boot measurements from host
          measurements. This is a safety concept to protect firmware
          measurements on systems where the regular firmware separator
          measurement is missing. It's also useful in environments where a
          software TPM is used, i.e. where TPM functionality is only available
          starting with the OS, but not before. Note that this new measurement
          has an effect on all indicated PCRs, hence might affect relevant TPM
          policies.
 
        * Support for udev's old database version 0 has been removed. This
          effectively means live upgrades from versions older than v247 are not
          supported anymore.
 
        * systemd-networkd gained a new sd-dhcp-relay backend for DHCP relay
          agent support. As part of this change, the following [DHCPServer]
          settings are deprecated:
          - BindToInterface=
          - RelayTarget=
          - RelayAgentCircuitId=
          - RelayAgentRemoteId=
          They are replaced by DHCPRelay= in [Network], along with new
          [DHCPRelay] section settings in .network files:
          - AgentAddress=
          - GatewayAddress=
          - CircuitId=
          - VirtualSubnetSelection=
          - ExtraOption=
          - InterfacePriority=
          and in networkd.conf:
          - ServerAddress=
          - OverrideServerIdentifier=
          - RemoteId=
          - ExtraOption=
 
        * Required version of musl (when built with -Dlibc=musl) has been raised
          from 1.2.5 to 1.2.6.
 
        * libsystemd is no longer guaranteed to be linked against libm. Whether
          the dependency is recorded depends on whether the compiler chooses to
          emit builtins for all calls to libm symbols. Consumers that rely on
          libsystemd transitively pulling in libm should link against it
          themselves. There is at least one known case that is still unsolved:
          rsyslog crashes on launch due to libfastjson using libm without linking
          to it, which was previously masked because libsystemd linked to it. If
          forcing a link against libm is required as a workaround,
          '-Wl,--push-state,--no-as-needed,-lm,--pop-state' can be added to the
          link flags, or passed to systemd's meson build options via
          '-Dc_link_args=-Wl,--push-state,--no-as-needed,-lm,--pop-state'.
 
        Changes in the system and service manager:
 
        * PID1 now supports the kernel's Live Update Orchestration (LUO) /
          Kexec Handover (KHO) systems when present and enabled. System units'
          FD Stores are now preserved through kexec, and units will get back
          stashed (named) file descriptors after kexec, if the kernel supports
          the FD type (at the time of writing only memfds are supported).
          Units can also create their own LUO Sessions by talking to the kernel
          directly, and store them in their FD Stores, and those will also be
          preserved and passed down to the unit after kexec. Units must set
          'FileDescriptorStorePreserve=yes' in order to enable this feature.
 
        * User session managers now support persisting user units' FD Stores