wt/finit

0 Fork 0

mirror of https://github.com/troglobit/finit.git synced 2026-10-18 10:59:34 +00:00

Files

T

Joachim Wiberg 61b0e0f3e6 Fix #420 : run services inside a PAM session

Apply a PAM session to run/task/sysv/services Finit starts, pam_limits
above all, so a service running as a given user picks up that user's
limits the way a login does.

Add a new `pam` setting for the new block format (only), like the
per-service directories, naming a file in /etc/pam.d:

    service weston {
        user    = "weston"
        pam     = "weston-autologin"
        command = "/usr/bin/weston --continue-without-input"
    }

pam_close_session() has to be called by a process still holding the
handle, and the handle does not survive exec().  Hence the keeper: it
holds the handle, drops to the service's credentials, and waits for a
parent-death signal before closing the session.  Same shape as
systemd's (sd-pam), for the same reason, and one per fork, so the
script hooks open and close their own.

The keeper closes the descriptors it inherited from Finit and only
those.  Closing everything would also take out what pam_open_session()
opened for itself, a keyring fd or a lock file, and leave the modules
to close a session with those pulled out from under them.  Closing
nothing, as (sd-pam) does, would leave it holding the write end of the
notify pipe for the service's whole lifetime and starve notify = "s6"
services of their ready signal.  So the fds open before pam_start()
are snapshotted and exactly those are closed, while the ones PAM opens
after are marked close-on-exec so the daemon does not inherit them
either.

A refused value, a denied account stack, an uninstalled pam.d file,
and a build without PAM support all keep the service from starting
rather than running it with the stacks skipped: one that quietly loses
pam_limits and its private /tmp, with nothing said.  Capabilities a
module like pam_cap.so granted are merged into the IAB Finit applies
instead of being replaced by it, which only helps a service that also
sets capabilities, the other arm being a plain setuid() with nothing
left to restore once permitted is empty.

The test sysroot gains pam_permit.so, pam_deny.so and pam_limits.so,
which ldd cannot see, libpam dlopen()s them, and the test skips when
the host has none to stage.  The negative cases pin the exit status
rather than only asserting crashed, which serv reports for any early
exit, so a bad command or an unwritable pidfile cannot pass for a
rejected session.

Signed-off-by: Joachim Wiberg <troglobit@gmail.com>

2026-09-23 16:30:14 +02:00

13 KiB

Raw Permalink Blame History

Service Options

The run, task, tty, service and sysv blocks take settings as keys. This section lists them with their limitations.

The name of a service, shown by the initctl tool, is the block title. It defaults to the basename of the executable if the title is omitted:

service sshd {
    command = "/usr/sbin/sshd -D"
}

For multiple instances of a service, with the same name, add an identifier after a colon to prevent Finit from replacing previous instances:

service ssdp:eth1 { command = "ssdpd eth1" }
service ssdp:eth2 { command = "ssdpd eth2" }

The initctl tool will list these two services as:

ssdp:eth1
ssdp:eth2

Conflicting services that must be prevented from starting, use the conflicts key:

service udevd { command = "udevd" }
run mdev {
    runlevel  = "S"
    conflicts = { "udevd" }
    command   = "mdev -s"
}

The list may name several, and an instance is named the same way it is declared:

service abc:1 { command = "abc 1" }
service abc:2 { command = "abc 2" }
service cde   { conflicts = { "abc:1", "abc:2" }  command = "cde" }

If a service should not be automatically started, set manual-start. It can then be started at any time with initctl start <service>:

service lldpd {
    manual-start = true
    command      = "/usr/sbin/lldpd -d"
}

Other run/task/service settings are:

capabilities -- see the Linux Capabilities section
cgroup NAME {} -- see the Cgroups section
envfile -- see the Service Environment section
log {} -- see Redirecting Output
tty -- see Controlling TTY
notify -- see Service Synchronization
pam -- see the PAM Sessions section
if -- see Conditional Execution
type = "forking" -- see description of the service block
a leading - on command -- see Conditional Loading

Command Candidates

The command setting may list several candidates for the same service, in which case Finit starts the first one it finds:

service udevd {
    command = { "/lib/systemd/systemd-udevd", "-udevd" }
}

A candidate that is not installed is skipped without a warning, that is the point of listing more than one. The leading - is read from the candidate that wins, or from the last one when none are found, so a block with a single command behaves as it always has.

The candidates are alternative spellings of one service, not a fallback for a service that fails to start. Finit chooses when the .conf file is read and does not revisit the choice.

A tty block takes the same list, where the candidates are the getty to run:

tty ttyAMA0 {
    command = { "/sbin/agetty -L ttyAMA0 115200 vt100",
                "/sbin/getty -L 115200 /dev/ttyAMA0 vt100" }
}

Provided Conditions

A service always asserts pid/<ident> for itself, named after the block title. provides names conditions it asserts in addition to that one:

service syslogd:udev {
    if       = "udevd"
    provides = "pid/syslogd"
    command  = "syslogd -F"
}

This is what lets variants of one service, picked apart by if, share the barrier that everything downstream waits for. Each variant has its own title, so each is its own service to initctl, while the condition they publish stays the same.

Any namespace works, pid/, usr/, service/, since the point is publishing a name existing configuration already waits on. A value without a namespace separator is not a condition and is rejected. The list holds at most four.

A provided condition is asserted and cleared with the service, exactly like its own, so stopping the provider clears the barrier.

Two services cannot supply the same condition. The second claim is refused, with the owner named:

finit.conf: second: provides condition <usr/barrier> already registered by service first, ignoring

The service itself still registers and runs, only the claim is dropped. Which one is "first" is .conf load order, so it follows the file names, see Files & Layout.

A real identity outranks a claim on it. provides = "pid/sshd" in a system that also runs a service titled sshd loses to that service, because pid/<ident> is how Finit tracks the service itself.

initctl cond dump names the owner of every condition, which for a provided one is the service that claimed it:

$ initctl cond dump
PID    IDENT         STATUS  CONDITION
1234   syslogd:udev  on      <pid/syslogd>
1234   syslogd:udev  on      <pid/syslogd:udev>

Duplicate Titles

The title is the service identity, so two blocks sharing a title in one file are two declarations of one service. The parser underneath merges them without a word, leaving a single service holding a mix of both, so Finit rejects the file and names the title:

/etc/finit.d/hotplug.conf:12: found duplicate title 'udevd'

Nothing from a rejected file is loaded and the rest of the configuration is unaffected.

The same title in another file is fine. That is how a system .conf is overridden, the later file wins, see Files & Layout.

Two shapes need a single identity and used to be written as repeated declarations. One is a service with several candidate binaries, which is now a command list. The other is a service gated differently per platform, e.g. a syslog daemon waiting for whichever hotplug daemon the system happens to have. For that one, give each variant its own file, or keep them as one-liners in the line-based format, which has no titles. The migration guide works through it.

Restarting

Services are automatically started, restarted, and stopped, depending on the configuration and conditions. Within the confines of that the following settings are available:

restart-max = 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 restart NAME
restart-sec = SEC -- number of seconds before Finit tries to restart a crashing service. The default is 2 seconds for the first half of restart-max attempts, then a back-off to 5 seconds -- with the default restart-max that is the first five retries. The greater of this configured value and the back-off is used
restart = "always" -- no upper limit on the number of times Finit tries to restart a crashing service
restart = "never" -- do not restart on failures. false is accepted as a synonym, and true selects the default policy
respawn = true -- bypasses the restart mechanism completely, allowing endless restarts. Exiting is treated as normal work rather than failure, so the service is restarted at once instead of being counted and delayed. This is how a tty behaves; it is not what service was designed for, so it is not the default
remain-after-exit = true -- for run and task only. Prevents the task from re-running on runlevel re-entry and ensures the exec-stop-post script runs when the task is explicitly stopped or leaves its valid runlevels. This is systemd's RemainAfterExit=. See Task and Run for more details
oncrash = "reboot" -- when all retries have failed, and the service has crashed, the system is rebooted
oncrash = "script" -- similarly, but instead of rebooting, call the exec-stop-post script with exit code crashed, see below

Service directories

Five settings ask Finit to create a directory for the service before it starts, owned by its user and group, mode 0755. The value is a directory name, resolved under a fixed base -- absolute paths and .. are refused:

Setting	Base	Environment variable
`runtime-dir`	`/run`	`RUNTIME_DIRECTORY`
`state-dir`	`/var/lib`	`STATE_DIRECTORY`
`cache-dir`	`/var/cache`	`CACHE_DIRECTORY`
`logs-dir`	`/var/log`	`LOGS_DIRECTORY`
`config-dir`	`/etc`	`CONFIGURATION_DIRECTORY`

Each resolved path is exported to the process environment under the listed name, the same names systemd uses for RuntimeDirectory= and friends. As in systemd, config-dir is the odd one out: it is created but never chowned.

Each takes a matching -mode, e.g. runtime-dir-mode = 0700, default 0755. Modes are octal, with the leading zero.

The mode of the named directory is locked down again on every start. Its contents are left alone as long as the owner is right; if the owner has drifted, everything under it is chowned back.

The runtime directory is removed again when the service stops, after any exec-stop-post script has run; /run is a tmpfs so it would not survive a reboot anyway. The other four persist. A completed run or task counts as stopped, unless remain-after-exit keeps it alive until stopped for real. runtime-dir-preserve adjusts this, same values as systemd's RuntimeDirectoryPreserve=:

"no" -- removed when the service stops, the default
"restart" -- kept across restarts, removed on a real stop
"yes" -- never removed

This is what lets a service drop privileges and still create, and later touch, its own PID file:

service ntpd {
    user        = "ntp"
    group       = "ntp"
    runtime-dir = "ntpd"
    pidfile     = "/run/ntpd/ntpd.pid"
    command     = "/usr/sbin/ntpd -n -p /run/ntpd/ntpd.pid"
}

These settings exist only in the block format.

Stopping and reloading

When stopping a service, either manually or when moving to another runlevel, Finit starts by sending SIGTERM to let the process shut down gracefully. If it has not been collected within 3 seconds, SIGKILL follows. Both ends of that are configurable:

stop-signal = "SIGPWR" -- send this instead of SIGTERM
stop-timeout = 10 -- seconds to wait before SIGKILL, 1-300

Some services need more than a signal:

exec-stop = "script [args]" -- run instead of sending the stop signal, and instead of stop for sysv
exec-reload = "script [args]" -- run instead of sending SIGHUP. Like systemd, Finit sets $MAINPID as a convenience to scripts, which in effect also allows exec-reload = "kill -HUP $MAINPID"

If a daemon cannot be reloaded with a signal at all, say so and Finit restarts it instead:

service dropbear {
    reload-signal = "none"
    command       = "/usr/sbin/dropbear -R -F"
}

SIGHUP is the default and the only other accepted value, with or without the SIG prefix and in any case.

Scripts

Services, including the sysv variant, support six lifecycle scripts:

exec-start-pre -- called before the sysv/service is started
exec-start-ready -- called when the sysv/service is ready
exec-stop -- called instead of the stop signal
exec-stop-post -- called after the sysv/service has stopped
exec-reload -- called instead of SIGHUP
exec-cleanup -- called when run/task/sysv/service is removed

Each takes a matching -timeout, in seconds, 0-3600, after which Finit kills the script. It defaults to the stop-timeout value and can be disabled by setting it to zero:

service foo {
    exec-start-pre         = "/etc/foo/pre.sh"
    exec-start-pre-timeout = 10
    command                = "/usr/sbin/foo"
}

These scripts run as the same user and group as the service itself, with any envfile sourced. They are executed from the $HOME of the given user. The scripts are not called with any argument, but get a set of environment variables:

SERVICE_IDENT=foo:1
SERVICE_NAME=foo
SERVICE_ID=1

The exec-stop-post script is called with an additional set of environment variables. Yes, the text is correct, the naming was an accident:

EXIT_CODE=[exited,signal,crashed]: normal exit, signaled, or crashed
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

When a run/task/sysv/service is removed (disable + reload) it is first stopped and then removed from the runlevel. The exec-stop-post script always runs when the process has stopped, and exec-cleanup runs when the block has been removed from the runlevel.

Important

These script actions are intended for setup, cleanup, and readiness notification. A script that outlives its timeout is killed, so pick one that suits the work, or set it to zero to opt out and take responsibility for the script terminating.

13 KiB Raw Permalink Blame History