v1.0.0 - 1a67de1

Clean-room reimplementation against the design in designs/docs/next/. The
whole LD_PRELOAD subsystem, the four committed .so blobs and the old
systemd/, procwait/, journalctl/, systemctl/ and common/ packages are
deleted.

Process tracking

• Replace LD_PRELOAD fork interposition with per-unit subreaper
  supervisors. Each unit runs under a dedicated process that sets
  PR_SET_CHILD_SUBREAPER=1, so orphaned descendants reparent to it rather
  than to PID 1 and unit membership becomes exact. Measured, the old shim saw
  2 of 6 child-creation paths and missed daemon(3) — the case it existed to
  solve — entirely.
• systemctl stop can now stop a Type=forking service. The termination
  ladder signals every member of the unit's live tree individually,
  re-enumerating each round, so it converges even against a process that forks
  on every signal. Process-group kills are used only as a cheap supplement:
  measured, kill(-pgid) does not stop a daemonised service.
• pidfd_open/pidfd_send_signal are used where available, eliminating
  PID-reuse mis-targeting.
• cgroup v2 is auto-detected and used as an accelerator when a writable
  hierarchy is available; it is never required.
• Orphan recovery: if a supervisor dies, its processes are found again by an
  invocation-id environment marker and handed to a replacement.
• --no-pidtrack is accepted and ignored with a deprecation warning.

Correctness

• Unit commands no longer run through /bin/bash -c. systemd's own command
  lexer is implemented: prefixes -, @, :, +, !, !!; quoting;
  C-style escapes; $VAR/${VAR} expansion with systemd's word-splitting
  rule; no globs, pipes, redirection or command substitution.
• Full boolean grammar: RemainAfterExit=yes now parses as true.
• Type=notify is implemented with a real $NOTIFY_SOCKET and
  SCM_CREDENTIALS, rather than being silently treated as forking.
• Before=/After= are honoured: transactions are topologically ordered,
  units in a stratum start in parallel, and an ordering cycle is broken with a
  warning rather than hanging the boot.
• Requisite= no longer starts its dependency; OnFailure=/OnSuccess= no
  longer create a bogus inverse edge; StopWhenUnneeded= works.
• Type=oneshot runs its ExecStart= lines sequentially and no longer forces
  RemainAfterExit. Multiple ExecStart= lines for other types are a parse
  error rather than being started in parallel.
• PIDFile= is parsed strictly and the pid is validated as a member of the
  unit's tree; a stale or hostile pid file fails the unit instead of making
  init adopt — and later kill — an unrelated process.
• TimeoutStartSec= is implemented and TimeoutStopSec= defaults to 90 s;
  a hanging ExecStartPre= fails its own unit instead of the whole boot.
• StartLimitIntervalSec=/StartLimitBurst= stop a crash-looping unit from
  restarting for ever; RestartSec= honours values below one second.
• EnvironmentFile= is parsed properly (comments, blanks, export , quotes,
  continuations, optional - prefix); Environment= accepts several
  space-separated assignments per line.
• Full specifier table (%n %N %p %P %i %I %f %j %t %S %C %L %E %h %u %U %g %G %H %m %b %v %a %%), expanded in every directive systemd expands them in.
• User=/Group= are applied whenever either is set, supplementary groups are
  set, and $HOME/$USER/$LOGNAME/$SHELL are exported. An unresolvable
  User= fails the unit with 217/USER rather than falling back to root.
• Limit*= directives now work, clamped to the container's inherited hard
  limits with a diagnostic naming the docker run --ulimit that would fix it.
• New: UMask=, Nice=, OOMScoreAdjust=, NoNewPrivileges=,
  WorkingDirectory=, RuntimeDirectory= and its State/Cache/Logs/
  Configuration siblings, StandardInput=/StandardOutput=/
  StandardError=, SyslogIdentifier=, the Condition*=/Assert*= families,
  KillMode=/KillSignal=/FinalKillSignal=/SendSIGHUP=/SendSIGKILL=.
• .target units are first-class, with well-known targets synthesised on
  demand so After=network.target resolves.
• Templates are resolved from their template file, never materialised;
  delete-instance removes only the enable symlink.
• set-environment and unset-environment are no longer swapped.
• mask creates the right symlink instead of a file called target in init's
  working directory; mask state is derived from the filesystem on every load.
• enable writes real symlinks into the [Install]-named target's .wants
  or .requires directory, honouring Alias= and Also=.
• Unit discovery de-duplicates directories by realpath, scans
  /run/systemd/system, and applies drop-ins from all four search-path
  directories plus the type-wide and template directories.

Reliability

• PID 1 is a thin dispatcher: it never execs a unit command and never blocks on
  one. Per-unit state is owned by one supervisor, dependency edges are stored
  as names and resolved at use time, and every unit actor has a top-level
  recover — the nine container-fatal crash and hang defects are removed by
  construction.
• Shutdown is reverse-topological and parallel within a stratum, under a global
  budget, with a three-signal operator escape hatch and a final sweep and reap.
• daemon-reload no longer rebuilds the definitions of running units: a
  running job keeps the configuration it was started with.

Protocol, CLI and logging

• New length-prefixed, versioned, typed control protocol: no single-read()
  assumption, no 0x00 sentinel, no 64 KiB cap, and a HELLO handshake that
  diagnoses a stale client instead of misparsing.
• systemctl gains try-restart, reload-or-restart, is-failed,
  is-enabled, list-unit-files, list-dependencies, cat, preset,
  preset-all, link, revert, add-wants, add-requires, kill,
  isolate, reset-failed, daemon-reexec, show -p X --value, and the
  systemd exit-code table. An unknown verb exits 2 rather than 0.
• journalctl is native: -u is optional, -n tails the file (it used to
  tail init's stdin), -f uses inotify, -S/-U/-b work independently of
  each other, plus -p, -o, -g, -r, --disk-usage and --vacuum-*.
• Log records carry an RFC3339-microsecond UTC timestamp, a syslog priority and
  the unit and pid, so time filters work at all; legacy un-timestamped lines
  are still readable. Logs rotate by size with a total cap.
• New systemd-notify and systemd-detect-virt helpers, and
  /run/systemd/system is created so distro maintainer scripts stop skipping
  unit registration.

Security

• Runtime state moved to /run/docker-systemd/ (0700 root:root) with the
  control socket at 0600 and SO_PEERCRED authentication: only uid 0 may
  control the manager. The old /tmp socket let any uid in the container start
  a root unit or power the container off.
• The world-writable pidtrack socket, /var/log/pidtrack.log and
  /etc/ld.so.preload manipulation are gone with the shim.
• Unit log files are 0640 root:adm.

Build and test

• First automated tests: unit tests with the race detector, four fuzz targets,
  a process-level suite that runs real daemonising services, and a conformance
  matrix across ten base images including alpine:3.
• The toolchain is pinned in go.mod; no UPX; no committed binary artefacts;
  every release artefact is reproducible from a clean checkout by CI.

v0.5.0 - 550d55b

CHANGELOG

v0.5.0

• add timeout handling and wpid==0 handling to procwait in FinalReap
• add isShuttingDown method to daemon to check if the system is shutting down to prevent starting new services during shutdown

v0.4.5

• add dockerbuild.sh script to build LD_PRELOAD libraries for different architectures
• build against older supported versions of glibc

v0.4.2

• add error handling to LD_PRELOAD libraries
• fix arm64 versions of LD_PRELOAD libraries

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.5 - 70bf517

CHANGELOG

v0.4.5

• add dockerbuild.sh script to build LD_PRELOAD libraries for different architectures
• build against older supported versions of glibc

v0.4.2

• add error handling to LD_PRELOAD libraries
• fix arm64 versions of LD_PRELOAD libraries

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.4 - b472df8

CHANGELOG

v0.4.2

• add error handling to LD_PRELOAD libraries
• fix arm64 versions of LD_PRELOAD libraries

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.3 - b52d8bf

CHANGELOG

v0.4.2

• add error handling to LD_PRELOAD libraries
• fix arm64 versions of LD_PRELOAD libraries

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.2 - 204e16e

CHANGELOG

v0.4.2

• add error handling to LD_PRELOAD libraries
• fix arm64 versions of LD_PRELOAD libraries

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.1 - 71d0581

CHANGELOG

v0.4.1

• daemons should also inherit os.Environ() of systemd process
• support set-environment and unset-environment features of systemd/systemctl
• disable journalctl pager; feature will be ignored; to page, simply pipe to more or less

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.4.0 - 6c09dbe

CHANGELOG

v0.4.0

• redo the systemd daemon handler, using signals and states instead of long-lasting mutex locks to allow for better state flow and querying capability
• add version systemctl command to print version

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.3.3 - 614cea5

CHANGELOG

v0.3.3

• bump dependencies
• automated docker build system

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes

v0.3.2 - 33b1a64

CHANGELOG

v0.3.2

• improve syscall.Wait4 call error handling in procwait
• improve communication protocol between systemd.command and systemctl so that multiple messages can be sent and printed to user during the command execution process

v0.3.1

• add --now option to systemctl enable

v0.3.0

• big change - not using double-process-init method for controlling and reaping processes; instead using a single init with a single dispatching syscall.Wait4(-1,...)
  • this allows for a more streamlined approach, single-pid init process, as well as improved tracking of other PIDs
  • more importantly, the forking-process wait system now can pause and wait on a mutex instead of timed polling, making this much more efficient

v0.2.1

• support tracking processes that fork-detach themselves late (corner-case)
• running enable/start on a multi-instance will create that instance automatically

v0.2.0

• first runnable release which properly tracks pids and reaps zombie processes
• added ld_preload feature for fork process tracking and multiple bugfixes