NAME
finit.conf — Finit configuration file format
SYNOPSIS
/etc/finit.conf
/etc/finit.d/*.conf
/etc/finit.d/available/*.conf
/etc/finit.d/enabled/*.conf
DESCRIPTION
Finit based systems can be set up with a single file: /etc/finit.conf. This is the traditional way of doing it which can be ideal for some setups since it gives a great overview of the system configuration.
More comprehensive setups, however, require more careful planning. The recommendation is per-package *.conf files in /etc/finit.d/available/. This allows end users to enable and disable parts of the system configuration at runtime. Finit tracks which tasks and services belong to a given file, so that when the user calls initctl reload to activate the changes they have made, only the affected tasks and services are stopped, started, or restarted. Hence, this approach is useful for modern package-based Linux distributions.
The following sections describe the more of this in detail, starting with files, the file format, and available directives. Remember to also visit the other manual pages (references at the bottom).
FILES
/etc/finit.conf
Main configuration file, optional
/etc/finit.d/*.conf
Static (system) service definitions
/etc/finit.d/available/*.conf
Available (installed) services
/etc/finit.d/enabled/*.conf
Enabled services (symlink back)
Static services, .conf files in /etc/finit.d/, cannot be enabled or disabled using initctl [enable | disable] service[.conf]. An enabled service is a symlink back to the corresponding ../available/ service .conf file. Please use the initctl tool to manage these symlinks to ensure proper operation.
NOTE: Previous versions of Finit created symlinks in /etc/finit.d/ if /etc/finit.d/enabled/ was missing, this is as of v4.4 not supported.
FILE FORMAT
The file format is line based, empty lines and comments, lines starting with ’#’, are ignored. A configuration directive starts with a keyword followed by a space and the rest of the line is treated as the value.
As of Finit v4.4, configuration directives can be broken up in multiple lines using the continuation character ’\’, and trailing comments are also allowed. Example:
# Escape \#
chars if you want them literal in, e.g., descriptions
service name:sysklogd [S123456789] \
env:-/etc/default/sysklogd \
syslogd -F $SYSLOGD_ARGS \
-- System log daemon \# 1 # Comments allowed now
DIRECTIVES
This section lists all supported configuration directives. There also exist deprecated directives, see the Markdown documentation for details on these.
rlimit [
hard|soft] RESOURCE ⟨ LIMIT |
unlimited⟩
Set the hard or soft limit for a resource, or both if that argument is omitted. RESOURCE is the lower-case RLIMIT_ string constants from setrlimit(2), without prefix. E.g. to set RLIMIT_CPU, use cpu.
LIMIT is an integer that depends on the resource being modified, see the man page, or the kernel /proc/PID/limits file, for details. Finit versions before v3.1 used infinity for unlimited, which is still supported, albeit deprecated.
# No process is
allowed more than 8MB of address space
rlimit hard as 8388608
# Core dumps
may be arbitrarily large
rlimit soft core infinity
# CPU limit for
all services, soft & hard = 10 sec
rlimit cpu 10
rlimit can be set globally, in /etc/finit.conf, or locally per each /etc/finit.d/*.conf read. I.e., a set of task/run/service stanzas can share the same rlimits if they are in the same .conf.
runlevel ⟨ N⟩
The system runlevel to go to after bootstrap (S) has completed. N is the runlevel number 0-9, where 6 is reserved for reboot and 0 for halt. All other can be used by operating system administrators. Default: 2
It s recommended to keep runlevel 1 as single-user mode, because finit.conf disables networking in this mode.
Note: only read and executed in runlevel S (bootstrap).
run [
LVLS] ⟨
COND⟩ /path/to/cmd ARGS [-- Optional
description]
One-shot command to run in sequence when entering a runlevel, with optional arguments and description.
run commands are guaranteed to be completed before running the next command. Highly useful if true serialization is needed. Usually only used in the bootstrap (S) runlevel.
⟨ COND⟩ conditions are described in finit(8), see also the Examples section below.
task [
LVLS] ⟨
COND⟩ /path/to/cmd ARGS [-- Optional
description]
One-shot like run, but starts in parallel with the next command.
Both run and task commands are run in a shell, so pipes and redirects can be freely used:
task [s] echo "foo" | cat >/tmp/bar
sysv [
LVLS] ⟨
COND⟩ /path/to/script ARGS [-- Optional
description]
Similar to task is the sysv stanza, which can be used to call SysV style start/stop scripts. The primary intention for this command is to be able to re-use much of existing setup and init scripts in Linux distributions.
When entering an allowed runlevel, Finit calls init-script start, when entering a disallowed runlevel, Finit calls init-script stop, and if the Finit .conf, where the sysv stanza is declared, is modified, Finit calls init-script restart on initctl reload. Similar to how service stanzas work.
Forking services started with sysv scripts can be monitored by Finit by declaring the PID file to look for:
sysv pid:!/path/to/pidfile.pid /path/to/script ...
The leading ’!’ is to prevent Finit from managing the PID file, which is the default behavior for the pid: command modifier.
service [
LVLS] ⟨
COND⟩ /path/to/daemon ARGS [-- Optional
description]
Service, or daemon, to be monitored and automatically restarted if it exits prematurely. Finit tries to restart services that die, by default 10 times, before giving up and marking them as crashed. After which they have to be restarted manually using initctl restart NAME. The number of restarts, the delay between each restart, and more is configurable, see the options below.
Tip: to allow endless restarts, see below option respawn
For daemons that support it, we recommend appending --foreground, --no-background, -n, -F, or similar command line argument to prevent them from forking off a sub-process in the background. This is the most reliable way to monitor a service.
However, not all daemons support running in the foreground, or they may start logging to the foreground as well, these are called forking services and are supported using the same syntax as forking sysv services, using the pid:!/path/to/pidfile.pid command modifier syntax. There is an alternative syntax that may be more intuitive, where Finit can also guess the PID file based on the daemon’s command name:
service type:forking ntpd -- NTP daemon
Here we let BusyBox ntpd daemonize itself. Finit uses the basename of the binary to guess the PID file to watch for the PID: /var/run/ntpd.pid. If Finit guesses wrong, you have to submit the full pid:!/path/to/file.pid option to your service stanza.
Example: in the case of ospfd (below), we omit the -d flag (daemonize) to prevent it from forking to the background:
service [2345] <pid/zebra> /sbin/ospfd -- OSPF daemon
[2345] denote the runlevels ospfd is allowed to run in, they are optional and default to runlevel 2-5 if omitted.
<pid/zebra> is the condition for starting ospfd. In this example Finit waits for another service, zebra, to have created its PID file in /var/run/quagga/zebra.pid before starting ospfd. Finit watches *all* files in /var/run, for each file named *.pid, */pid, Finit opens it and find the matching NAME:ID using the PID.
Some services do not maintain a PID file and rather than patching each application Finit provides a workaround. A pid modifier keyword can be set to have Finit automatically create (when starting) and later remove (when stopping) the PID file. The file is created in the /var/run directory using the basename(3) of the service. The full syntax of the pid modifier is:
pid[:[!][/path/to/]filename[.pid]]
For example, by adding pid:/run/foo.pid to the service /sbin/bar, that PID file will, not only be created and removed automatically, but also be used by the Finit condition subsystem. So a service/run/task can depend on the <pid/bar> condition.
As an alternative "readiness" notification, Finit supports both systemd and s6 style notification. This can be enabled by using the ’notify’ option:
notify:systemd
tells Finit the service uses the sd_notify() API to signal PID 1 when it has completed its startup and is ready to service events. This API expects the environment variable NOTIFY_SOCKET to be set to the socket where the application can send READY=10 when it is starting up or has processed a SIGHUP. For details, see:
https://www.freedesktop.org/software/systemd/man/sd_notify.html
notify:s6
puts Finit in s6 compatibility mode. Compared to the systemd notification, s6 expect compliant daemons to send 0 and then close their socket. For details, see:
https://skarnet.org/software/s6/notifywhenup.html
Finit takes care of "hard-wiring" the READY state as long as the application is running, events across any ’SIGHUP’. Since s6 can give its applications the descriptor number (must be >3) on then command line, Finit provides the following syntax ( %n is replaced by Finit with then descriptor number):
service notify:s6 mdevd -C -O 4 -D %n
When a service is ready, either by Finit detecting its PID file, or their respective readiness mechanism has been triggered, Finit creates then service’s ready condition which other services can depend on:
$ initctl -v
cond get service/mdevd/ready
on
If a service should not be automatically started, it can be configured as manual with the manual:yes command modifier. The service can then be started at any time by running initctl start NAME
The name of a service, shown by the initctl tool, defaults to the basename of the service executable. It can be changed with the name:foo command modifier.
As mentioned previously, services are automatically restarted should they crash, this is configurable with the following options:
restart:NUM
number of times Finit tries to restart a crashing service, default: 10. When this limit is reached the service is marked crashed and must be restarted manually with initctl(8).
restart_sec:SEC
number of seconds before Finit tries to restart a crashing service, default: 2 seconds for the first five retries, then back-off to 5 seconds. The maximum of this configured value and the above (2 and 5) will be used
restart:always
no upper limit on the number of times Finit tries to restart a crashing service. Same as restart:-1
norestart
dont restart on failures, same as restart:0
respawn
bypasses the restart mechanism completely, allows endless restarts. Useful in many use-cases, but not what service was originally designed for so not the default behavior.
oncrash:reboot
when all retries have failed, and the service has crashed, if this option is set the system is rebooted.
oncrash:script
Similar to oncrash:reboot, but instead of rebooting this action calls the post:script (see below) to let the operator decide the best course of action. If the post:script option is not set, this is a no-op.
The post:script is called with the same environment variables except for the EXIT_CODE variable which is set to crashed.
When stopping a service (run/task/sysv/service), either manually or when moving to another runlevel, Finit starts by sending SIGTERM, to allow the process to shut down gracefully. If the process has not been collected within 3 seconds, Finit sends SIGKILL. To halt the process using a different signal, use the command modifier halt:SIGNAL, e.g., halt:SIGPWR. To change the delay between your halt signal and KILL, use the command modifier kill:SEC, e.g., kill:10 to wait 10 seconds before sending SIGKILL.
Services support the pre:script and post:script command actions as well. These run as the same @USER:GROUP as the service itself, with any env:file sourced. The scripts must use an absolute path, but are executed from the $HOME of the given user. The scripts are not called with any argument (currently), but both get the SERVICE_IDENT=foo environment variable set. Here foo denotes the identity of the service, which if there are multiple services named foo, may be foo:1, or any unique identifier specified in the .conf file. The post:script is called with an additional set of environment variables:
EXIT_CODE=[exited,signal,crashed]
set to one of exited, signal, or crashed (see above).
EXIT_STATUS=[num,SIGNAME]
set to one of exit status code from the program, if it exited normally, or the signal name (HUP, TERM, etc.) if it exited due to signal
The scripts have a default execution time of 3 seconds before they are SIGKILLed, this can be adjusted using the kill:SEC modifier syntax.
runparts ⟨ DIR⟩
Call run-parts(8) on DIR to run start scripts. All executable files, or scripts, in the directory are called, in alphabetic order. The scripts in this directory are executed at the very end of bootstrap, runlevel S.
It can be beneficial to use S01name, S02othername, etc. if there is a dependency order between the scripts. Symlinks to existing daemons can talso be used, but make sure they daemonize by default.
Similar to the /etc/rc.local shell script, make sure that all your services and programs either terminate or start in the background or you will block Finit. Note: only read and executed in runlevel S (bootstrap).
include ⟨ CONF⟩
Include another configuration file. Absolute path required.
log size:BYTES count:NUM
Log rotation for run/task/services using the log command modifier with redirection to a log file. Global setting, applies to all services.
The size can be given as bytes, without a specifier, or in ’k’, ’M’, or ’G’, e.g. size:10M, or size:3G. A value of size:0 disables log rotation. The default is size:200k.
The count value is recommended to be between 1-5, with a default 5. Setting count to 0 means the logfile will be truncated when the MAX size limit is reached.
tty [
LVLS] ⟨
COND⟩ DEV [
BAUD] [
noclear] [
nowait] [
nologin] [
TERM]
This form of the tty stanza uses the built-in getty on the given TTY device DEV, in the given runlevels. DEV may be the special keyword @console, or ’console’, which is expanded from ’/sys/class/tty/console/active’, useful on embedded systems.
The default baud rate is 0, i.e., keep kernel default.
The ’tty’ stanza inherits runlevel, condition (and other feature) parsing from the ’service’ stanza. So TTYs can run in one or many runlevels and depend on any condition supported by Finit. This is useful e.g. to depend on ’<pid/elogind>’ before starting a TTY.
tty [12345] /dev/ttyAMA0 115200 noclear vt220
tty [
LVLS] ⟨
COND⟩ CMD DEV [
noclear] [
nowait]
This form of the tty stanza is for using an external getty, like agetty or the BusyBox getty.
By default, these first two syntax variants clear the TTY and wait for the user to press enter before starting getty.
tty [12345]
/sbin/getty -L 115200 /dev/ttyAMA0 vt100
tty [12345] /sbin/agetty -L ttyAMA0 115200 vt100 nowait
The noclear option disables clearing the TTY after each session. Clearing the TTY when a user logs out is usually preferable.
The nowait option disables the Please press Enter to activate console message before actually starting the getty program. On small and embedded systems running multiple unused getty wastes both memory and CPU cycles, so ’wait’ is the preferred default.
The nologin option disables getty and /bin/login, and gives the user a root (login) shell on the given TTY DEV immediately. Needless to say, this is a rather insecure option, but can be very useful for developer builds, during board bringup, or similar.
Notice the ordering, the TERM option to the built-in getty must be the last argument.
Embedded systems may want to enable automatic ’DEV’ by supplying the special @console device. This works regardless weather the system uses ttyS0, ttyAMA0, ttyMXC0, or anything else. Finit figures it out by querying sysfs: /sys/class/tty/console/active. The speed can be omitted to keep the kernel default.
Most systems get by fine by just using ’console’, which will evaluate to /dev/console. If you have to use @console to get any output, you may have some issue with your kernel config.
tty [12345] @console noclear vt220
On really bare bones systems, or for board bringup, Finit can give you a shell prompt as soon as bootstrap is done, without opening any device node:
tty [12345789] notty
This should of course not be enabled on production systems. Because it may give a user root access without having to log in. However, for board bringup and system debugging it can come in handy.
One can also use the service stanza to start a stand-alone shell:
service [12345] /bin/sh -l
tty [
LVLS] ⟨
COND⟩ [
notty] [
rescue]
The third tty form is for board bringup and the rescue boot mode. No device node is required in this variant, the same output that the kernel uses is reused for stdio. If the rescue option is omitted, a shell is started. The flags nologin, noclear, and nowait are implied. If the rescue option is set the bundled /libexec/finit/sulogin is started to present a bare-bones root login prompt. If the root (uid:0, gid:0) user does not have a password set, no rescue is possible.
COMMAND MODIFIERS
The run/task/tty/service/sysv stanzas take modifiers, or options, to control their behavior. This section lists them with their limitations. All modifiers must be placed between the stanza and its command.
@user:group
Every run, task, or service can also list the privileges the /path/to/cmd should be executed with. Prefix the command with @USR[:GRP], group is optional, like this:
run [2345] @joe:users logger "Hello world"
For multiple instances of the same command, e.g. a DHCP client or multiple web servers, add :ID somewhere between the run, task, service keyword and the command, like this:
service :80
[2345] httpd -f -h /http -p 80 -- Web server
service :8080[2345] httpd -f -h /http -p 8080 -- Old web
server
Without the :ID to the service the latter will overwrite the former and only the old web server would be started and supervised.
log:/path/to/file
Redirect stdout/stderr of a command to the given log file. See the global log directive, above, for details on log rotation.
log:console
Redirect stdout/stderr of a command to /dev/console, only use this for debugging or bringup.
log:null
Redirect stdout/stderr of a command to /dev/null.
log:prio:facility.level,tag:ident
Redirect stdout/stderr of a command to syslog using the given priority and tag identity.
service log:prio:user.warn,tag:ntpd /sbin/ntpd pool.ntp.org -- NTP daemon
log
Default prio is daemon.info and the default tag identity is the basename of the service or run/task command.
RESCUE MODE
Finit supports a rescue mode which is activated by the rescue option on the kernel command line. The rescue mode comes in two flavors: traditional and fallback.
Traditional
This is what most users expect. A very early maintenance
login prompt, served by the bundled
/libexec/finit/sulogin program, or the standard
sulogin from util-linux or BusyBox is searched for in
the UNIX default $PATH. If a successful login is
made, or the user exits (Ctrl-D), the rescue mode is ended
and the system boots up normally.
Note: if the user (UID 0 and GID 0) does not have a password, or the account is locked, the user is presented with a password-less prompt: Press enter to enter maintenance mode., which opens up a root shell.
Fallback
If no sulogin program is found, Finit tries to bring
up as much of its own functionality as possible, yet
limiting many aspects, meaning; no network,
no’fsck’ of file systems in /etc/fstab,
no /etc/rc.local, no runparts, and most
plugins are skipped (except those that provide functionality
for the condition subsystem).
Instead of reading /etc/finit.conf et al, system configuration is read from /lib/finit/rescue.conf, which can be freely modified by the system administrator.
The bundled default ’rescue.conf’ contains nothing more than:
runlevel 1
tty [12345] rescue
The tty has the rescue option set, which works similar to the board bring-up tty option notty. The major difference being that ’sulogin’ is started to query for root/admin password. If sulogin is not found, rescue behaves like notty and gives a plain root shell prompt.
If Finit cannot find /lib/finit/rescue.conf it defaults to:
tty [12345] rescue
There is no way to exit the fallback rescue mode.
SERVICE ENVIRONMENT
Finit supports sourcing environment variables from /etc/default/*, or similar. This is a common pattern from SysV init scripts, where the start/stop script is a generic script for the given service, foo, and the options for the service are sourced from the file /etc/default/foo. Like this:
/etc/default/foo:
FOO_OPTIONS=--extra-arg="bar" -s -x
/etc/finit.conf:
service [2345] env:-/etc/default/foo foo -n $FOO_OPTIONS -- Example foo daemon
Here the service foo is started with [--n], to make sure it runs in the foreground, and the with the options found in the environment file. With the ps command we can see that the process is started with:
foo -n --extra-arg=bar -s -x
Note: the leading ’-’ determines if Finit should treat a missing environment file as blocking the start of the service or not. When ’-’ is used, a missing environment file does not block the start.
SERVICE WRAPPER SCRIPTS
If your service requires to run additional commands, executed before the service is actually started, like the systemd ’ExecStartPre’, you can use a wrapper shell script to start your service.
The Finit service .conf file can be put into /etc/finit.d/available, so you can control the service using initctl. Then use the path to the wrapper script in the Finit .conf service stanza. The following example employs a wrapper script in /etc/start.d.
/etc/finit.d/available/program.conf:
service [235] <!> /etc/start.d/program -- Example Program
/etc/start.d/program:
#!/bin/sh
# Prepare the command line options
OPTIONS="-u $(cat /etc/username)"
# Execute the
program
exec /usr/bin/program $OPTIONS
Note: the example sets <!> to denote that it doesn’t support SIGHUP. That way Finit will stop/start the service instead of sending SIGHUP at restart/reload events.
TEMPLATING
Finit comes with rudimentary support for templating, similar to that of systemd. Best illustrated with an example:
$ initctl show
avahi-autoipd@
service :%i avahi-autoipd --syslog %i -- ZeroConf for %i
To enable ZeroConf for, e.g., eth0, use
$ initctl enable avahi-autoipd [AT] eth0.conf
The enabled symlink will be set up to avahi-autoipd@.conf and every instance of %i will in the instantiated directive be replaced with eth0. Inspect the result with:
$ initctl status avahi-autoipd:eth0
CGROUPS
There are three major cgroup configuration directives:
1.
Global top-level group: init, system, user, or a custom group
2.
Selecting a top-level group for a set of run/task/services
3.
Per run/task/service limits
Top-level group configuration.
# Top-level
cgroups and their default settings. All groups mandatory
# but more can be added, max 8 groups in total currently.
The cgroup
# ’root’ is also available, reserved for RT
processes. Settings are
# as-is, only one shorthand ’mem.’ exists, other
than that it’s the
# cgroup v2 controller default names.
cgroup init cpu.weight:100
cgroup user cpu.weight:100
cgroup system cpu.weight:9800
Adding an extra cgroup maint/ will require you to adjust the weight of the above three. We leave init/ and user/ as-is reducing weight of system/ to 9700.
cgroup system cpu.weight:9700
# Example extra
cgroup ’maint’
cgroup maint cpu.weight:100
By default, the system/ cgroup is selected for almost everything. The init/ cgroup is reserved for PID 1 itself and its closest relatives. The user/ cgroup is for local TTY logins spawned by getty.
To select a different top-level cgroup, e.g. maint/, one can either define it for a group of run/task/service directives in a .conf or per each stanza:
cgroup.maint
service [...] <...> /path/to/foo args -- description
service [...] <...> /path/to/bar args --
description
or
service [...] <...> cgroup.maint /path/to/foo args -- description
The latter form also allows per-stanza limits on the form:
service [...] <...> cgroup.maint:cpu.max:10000,mem.max:655360 /path/to/foo args -- description
Notice the comma separation and the mem. exception to the rule: every cgroup setting maps directly to cgroup v2 syntax. I.e., cpu.max maps to the file ±There is no filtering, except for expanding the shorthand mem. to memory., if the file is not available, either the cgroup controller is not available in your Linux kernel, or the name is misspelled.
Linux cgroups and details surrounding values are not explained in the Finit documentation. The Linux admin-guide cover this well: https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html
LIMITATIONS
As of Finit v4 there are no limitations to where .conf settings can be placed. Except for the system/global rlimit and cgroup top-level group declarations, which can only be set from /etc/finit.conf, since it is the first .conf file Finit reads.
Originally,
was the only way to set up a Finit system. Today it is mainly used for bootstrap settings like system hostname, early module loading for watchdogd, network bringup and system shutdown. These can now also be set in any .conf file in /etc/finit.d.
There is, however, nothing preventing you from having all configuration settings in /etc/finit.conf.
SEE ALSO
AUTHORS
Finit was conceived and reverse engineered by Claudio Matsuoka. Since v1.0, maintained by Joachim Wiberg, with contributions by many others.
Linux July 11, 2023 Linux