minit

Minimal PID1 replacement for embedded systems. A middle ground between BusyBox init and systemd.

Features

• Simple conf format — INI-style [service name] sections (JSON support is planned; examples/minit.json.example shows the intended format)
• Service supervision — track and respawn processes
• Restart policies — always, on-failure, never
• Auto-restart with backoff — exponential backoff to avoid restart storms
• Signal handling — SIGTERM, SIGINT → graceful shutdown; as PID 1, reboot/halt/poweroff (see below)
• PID 1 boot setup — mounts /proc, /sys, /dev, /dev/pts, /dev/shm, /run, /tmp, console, hostname, random seed
• Emergency shell — /bin/sh on the console on fatal boot errors or with single
• Boot time measurement — write boot_ms to /var/run/minit.boot (or /run/minit.boot if /var/run is missing)
• Dependency ordering — topological start order via depends / provides; dependents wait until a service is ready
• Oneshot services and readiness — type = oneshot, ready_fd (s6-style notification) or ready_file
• Service output — to the console, /dev/null, a rotated log file, or the kernel log
• Crash logging — log exits to /var/log/minit/crashes.log
• Minimal footprint — about 100KB as a static musl binary (budget: 500KB)

Build

make                       # dynamic ./minit
make static CC=musl-gcc    # stripped static ./minit-static for the target rootfs
make size CC=musl-gcc      # report the static binary size

Usage

./minit [config]
# Default config: /etc/minit.conf or MINIT_CONFIG env

Configuration

[service network]
type = oneshot
exec = /etc/minit/network.sh

[service httpd]
exec = /usr/sbin/lighttpd -D -f /etc/lighttpd.conf
depends = network
log = file

Service names may use A-Z a-z 0-9 . _ -. Keys per [service name] section:

Key	Default	Description
exec	(required)	Command line, run directly without a shell. '...', "..." and \ quote like a shell, without expansion. Looked up in PATH.
restart	on-failure	always, on-failure or never. Being killed by a signal counts as a failure.
type	simple	oneshot: ready once it exits 0, for setup scripts (restart = always is not allowed).
ready_fd	—	3–63: the service gets a pipe on this fd and is ready once it writes a newline to it (s6 notification-fd).
ready_file	—	Absolute path: ready once it exists. Removed before each start so a stale file does not count.
provides	—	Comma-separated names (up to 4) others can depend on. The service name is always provided too.
depends	—	Comma-separated names; this service starts once all their providers are ready.
respawn_delay	1	Base restart delay in seconds (1–60), doubled per quick restart, capped at 60s.
respawn_max	0	Give up after this many quick restarts in a row (0 = never give up).
respawn_window	60	A run longer than this many seconds resets the restart count.
log	console	console, null, file (<log dir>/<name>.log) or kmsg (one kernel log record per line).
log_size	256	log = file: rotate to <name>.log.1 at this many KB.

Without type = oneshot, ready_fd or ready_file, a service is ready as soon as it is started. Use at most one of the three. The boot report is written once every service is ready (or can never start because a dependency stopped or failed for good). The log directory is /var/log/minit, or /tmp/minit if that is not writable. The kernel rate-limits /dev/kmsg writes unless booted with printk.devkmsg=on.

Running as PID 1

Boot with init=/sbin/minit (or install it as /sbin/init). As PID 1 minit:

1. mounts /proc, /sys, /dev (devtmpfs), /dev/pts, /dev/shm, /run and /tmp, skipping anything that is already a mountpoint (e.g. set up by an initramfs);
2. puts stdin/stdout/stderr on /dev/console;
3. sets the hostname from /etc/hostname and mixes /var/lib/minit/random-seed into /dev/urandom;
4. reads the kernel command line:
   • minit.config=PATH — config file (takes precedence over MINIT_CONFIG, which the kernel also passes);
   • single or minit.single — start an emergency shell before any service; exit it to continue booting.

If the config cannot be loaded, minit starts /bin/sh on the console. Fix the problem and exit the shell to retry the boot. Without a shell it stays alive instead of exiting, because PID 1 exiting panics the kernel.

Shutdown follows the BusyBox init convention, so BusyBox reboot, halt and poweroff work:

Signal	Action
SIGTERM	reboot
SIGINT (Ctrl-Alt-Del)	reboot
SIGUSR1	halt
SIGUSR2	power off

Services are stopped in reverse dependency order, every remaining process gets SIGTERM and then SIGKILL, a fresh random seed is saved, filesystems are unmounted (or remounted read-only if busy), and the system reboots, halts or powers off.