]> git.ipfire.org Git - thirdparty/systemd.git/blame_incremental - man/systemd.service.xml
tmpfiles: fix compilation without acl support
[thirdparty/systemd.git] / man / systemd.service.xml
... / ...
CommitLineData
1<?xml version='1.0'?> <!--*-nxml-*-->
2<?xml-stylesheet type="text/xsl" href="http://docbook.sourceforge.net/release/xsl/current/xhtml/docbook.xsl"?>
3<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
4 "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
5
6<!--
7 This file is part of systemd.
8
9 Copyright 2010 Lennart Poettering
10
11 systemd is free software; you can redistribute it and/or modify it
12 under the terms of the GNU Lesser General Public License as published by
13 the Free Software Foundation; either version 2.1 of the License, or
14 (at your option) any later version.
15
16 systemd is distributed in the hope that it will be useful, but
17 WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 Lesser General Public License for more details.
20
21 You should have received a copy of the GNU Lesser General Public License
22 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23-->
24
25<refentry id="systemd.service">
26 <refentryinfo>
27 <title>systemd.service</title>
28 <productname>systemd</productname>
29
30 <authorgroup>
31 <author>
32 <contrib>Developer</contrib>
33 <firstname>Lennart</firstname>
34 <surname>Poettering</surname>
35 <email>lennart@poettering.net</email>
36 </author>
37 </authorgroup>
38 </refentryinfo>
39
40 <refmeta>
41 <refentrytitle>systemd.service</refentrytitle>
42 <manvolnum>5</manvolnum>
43 </refmeta>
44
45 <refnamediv>
46 <refname>systemd.service</refname>
47 <refpurpose>Service unit configuration</refpurpose>
48 </refnamediv>
49
50 <refsynopsisdiv>
51 <para><filename><replaceable>service</replaceable>.service</filename></para>
52 </refsynopsisdiv>
53
54 <refsect1>
55 <title>Description</title>
56
57 <para>A unit configuration file whose name ends in
58 <filename>.service</filename> encodes information
59 about a process controlled and supervised by
60 systemd.</para>
61
62 <para>This man page lists the configuration options
63 specific to this unit type. See
64 <citerefentry><refentrytitle>systemd.unit</refentrytitle><manvolnum>5</manvolnum></citerefentry>
65 for the common options of all unit configuration
66 files. The common configuration items are configured
67 in the generic <literal>[Unit]</literal> and
68 <literal>[Install]</literal> sections. The service
69 specific configuration options are configured in the
70 <literal>[Service]</literal> section.</para>
71
72 <para>Additional options are listed in
73 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
74 which define the execution environment the commands
75 are executed in, and in
76 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
77 which define the way the processes of the service are
78 terminated, and in
79 <citerefentry><refentrytitle>systemd.resource-control</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
80 which configure resource control settings for the
81 processes of the service.</para>
82
83 <para>Unless <varname>DefaultDependencies=</varname>
84 is set to <option>false</option>, service units will
85 implicitly have dependencies of type
86 <varname>Requires=</varname> and
87 <varname>After=</varname> on
88 <filename>basic.target</filename> as well as
89 dependencies of type <varname>Conflicts=</varname> and
90 <varname>Before=</varname> on
91 <filename>shutdown.target</filename>. These ensure
92 that normal service units pull in basic system
93 initialization, and are terminated cleanly prior to
94 system shutdown. Only services involved with early
95 boot or late system shutdown should disable this
96 option.</para>
97
98 <para>If a service is requested under a certain name
99 but no unit configuration file is found, systemd looks
100 for a SysV init script by the same name (with the
101 <filename>.service</filename> suffix removed) and
102 dynamically creates a service unit from that
103 script. This is useful for compatibility with
104 SysV. Note that this compatibility is quite
105 comprehensive but not 100%. For details about the
106 incompatibilities, see the <ulink
107 url="http://www.freedesktop.org/wiki/Software/systemd/Incompatibilities">Incompatibilities
108 with SysV</ulink> document.
109 </para>
110 </refsect1>
111
112 <refsect1>
113 <title>Options</title>
114
115 <para>Service files must include a
116 <literal>[Service]</literal> section, which carries
117 information about the service and the process it
118 supervises. A number of options that may be used in
119 this section are shared with other unit types. These
120 options are documented in
121 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>
122 and
123 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>. The
124 options specific to the <literal>[Service]</literal>
125 section of service units are the following:</para>
126
127 <variablelist class='unit-directives'>
128 <varlistentry>
129 <term><varname>Type=</varname></term>
130
131 <listitem><para>Configures the process
132 start-up type for this service
133 unit. One of <option>simple</option>,
134 <option>forking</option>,
135 <option>oneshot</option>,
136 <option>dbus</option>,
137 <option>notify</option> or
138 <option>idle</option>.</para>
139
140 <para>If set to
141 <option>simple</option> (the default
142 if neither
143 <varname>Type=</varname> nor
144 <varname>BusName=</varname>, but
145 <varname>ExecStart=</varname> are
146 specified), it is expected that the
147 process configured with
148 <varname>ExecStart=</varname> is the
149 main process of the service. In this
150 mode, if the process offers
151 functionality to other processes on
152 the system, its communication channels
153 should be installed before the daemon
154 is started up (e.g. sockets set up by
155 systemd, via socket activation), as
156 systemd will immediately proceed
157 starting follow-up units.</para>
158
159 <para>If set to
160 <option>forking</option>, it is
161 expected that the process configured
162 with <varname>ExecStart=</varname>
163 will call <function>fork()</function>
164 as part of its start-up. The parent process is
165 expected to exit when start-up is
166 complete and all communication
167 channels are set up. The child continues
168 to run as the main daemon
169 process. This is the behavior of
170 traditional UNIX daemons. If this
171 setting is used, it is recommended to
172 also use the
173 <varname>PIDFile=</varname> option, so
174 that systemd can identify the main
175 process of the daemon. systemd will
176 proceed with starting follow-up units
177 as soon as the parent process
178 exits.</para>
179
180 <para>Behavior of
181 <option>oneshot</option> is similar to
182 <option>simple</option>; however, it
183 is expected that the process has to
184 exit before systemd starts follow-up
185 units. <varname>RemainAfterExit=</varname>
186 is particularly useful for this type
187 of service. This is the implied
188 default if neither
189 <varname>Type=</varname> or
190 <varname>ExecStart=</varname> are
191 specified.</para>
192
193 <para>Behavior of
194 <option>dbus</option> is similar to
195 <option>simple</option>; however, it is
196 expected that the daemon acquires a
197 name on the D-Bus bus, as configured
198 by
199 <varname>BusName=</varname>. systemd
200 will proceed with starting follow-up
201 units after the D-Bus bus name has been
202 acquired. Service units with this
203 option configured implicitly gain
204 dependencies on the
205 <filename>dbus.socket</filename>
206 unit. This type is the default if
207 <varname>BusName=</varname> is
208 specified.</para>
209
210 <para>Behavior of
211 <option>notify</option> is similar to
212 <option>simple</option>; however, it is
213 expected that the daemon sends a
214 notification message via
215 <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>
216 or an equivalent call when it has finished
217 starting up. systemd will proceed with
218 starting follow-up units after this
219 notification message has been sent. If
220 this option is used,
221 <varname>NotifyAccess=</varname> (see
222 below) should be set to open access to
223 the notification socket provided by
224 systemd. If
225 <varname>NotifyAccess=</varname> is
226 not set, it will be implicitly set to
227 <option>main</option>. Note that
228 currently
229 <varname>Type=</varname><option>notify</option>
230 will not work if used in combination with
231 <varname>PrivateNetwork=</varname><option>yes</option>.</para>
232
233 <para>Behavior of
234 <option>idle</option> is very similar
235 to <option>simple</option>; however,
236 actual execution of the service
237 binary is delayed until all jobs are
238 dispatched. This may be used to avoid
239 interleaving of output of shell
240 services with the status output on the
241 console.</para>
242 </listitem>
243 </varlistentry>
244
245 <varlistentry>
246 <term><varname>RemainAfterExit=</varname></term>
247
248 <listitem><para>Takes a boolean value
249 that specifies whether the service
250 shall be considered active even when
251 all its processes exited. Defaults to
252 <option>no</option>.</para>
253 </listitem>
254 </varlistentry>
255
256 <varlistentry>
257 <term><varname>GuessMainPID=</varname></term>
258
259 <listitem><para>Takes a boolean value
260 that specifies whether systemd should
261 try to guess the main PID of a service
262 if it cannot be determined
263 reliably. This option is ignored
264 unless <option>Type=forking</option>
265 is set and <option>PIDFile=</option>
266 is unset because for the other types
267 or with an explicitly configured PID
268 file, the main PID is always known. The
269 guessing algorithm might come to
270 incorrect conclusions if a daemon
271 consists of more than one process. If
272 the main PID cannot be determined,
273 failure detection and automatic
274 restarting of a service will not work
275 reliably. Defaults to
276 <option>yes</option>.</para>
277 </listitem>
278 </varlistentry>
279
280 <varlistentry>
281 <term><varname>PIDFile=</varname></term>
282
283 <listitem><para>Takes an absolute file
284 name pointing to the PID file of this
285 daemon. Use of this option is
286 recommended for services where
287 <varname>Type=</varname> is set to
288 <option>forking</option>. systemd will
289 read the PID of the main process of
290 the daemon after start-up of the
291 service. systemd will not write to the
292 file configured here.</para>
293 </listitem>
294 </varlistentry>
295
296 <varlistentry>
297 <term><varname>BusName=</varname></term>
298
299 <listitem><para>Takes a D-Bus bus
300 name that this service is reachable
301 as. This option is mandatory for
302 services where
303 <varname>Type=</varname> is set to
304 <option>dbus</option>.</para>
305 </listitem>
306 </varlistentry>
307
308 <varlistentry>
309 <term><varname>BusPolicy=</varname></term>
310
311 <listitem><para>If specified, a custom
312 <ulink url="https://code.google.com/p/d-bus/">kdbus</ulink>
313 endpoint will be created and installed as the
314 default bus node for the service. Such a custom
315 endpoint can hold an own set of policy rules
316 that are enforced on top of the bus-wide ones.
317 The custom endpoint is named after the service
318 it was created for, and its node will be
319 bind-mounted over the default bus node
320 location, so the service can only access the
321 bus through its own endpoint. Note that custom
322 bus endpoints default to a 'deny all' policy.
323 Hence, if at least one
324 <varname>BusPolicy=</varname> directive is
325 given, you have to make sure to add explicit
326 rules for everything the service should be able
327 to do.</para>
328 <para>The value of this directive is comprised
329 of two parts; the bus name, and a verb to
330 specify to granted access, which is one of
331 <option>see</option>,
332 <option>talk</option>, or
333 <option>own</option>.
334 <option>talk</option> implies
335 <option>see</option>, and <option>own</option>
336 implies both <option>talk</option> and
337 <option>see</option>.
338 If multiple access levels are specified for the
339 same bus name, the most powerful one takes
340 effect.
341 </para>
342 <para>Examples:</para>
343 <programlisting>BusPolicy=org.freedesktop.systemd1 talk</programlisting>
344 <programlisting>BusPolicy=org.foo.bar see</programlisting>
345 <para>This option is only available on kdbus enabled systems.</para>
346 </listitem>
347 </varlistentry>
348
349 <varlistentry>
350 <term><varname>ExecStart=</varname></term>
351 <listitem><para>Commands with their
352 arguments that are executed when this
353 service is started. The value is split
354 into zero or more command lines is
355 according to the rules described below
356 (see section "Command Lines" below).
357 </para>
358
359 <para>When <varname>Type</varname> is
360 not <option>oneshot</option>, only one
361 command may and must be given. When
362 <varname>Type=oneshot</varname> is
363 used, zero or more commands may be
364 specified. This can be specified by
365 providing multiple command lines in
366 the same directive, or alternatively,
367 this directive may be specified more
368 than once with the same effect. If the
369 empty string is assigned to this
370 option, the list of commands to start
371 is reset, prior assignments of this
372 option will have no effect. If no
373 <varname>ExecStart=</varname> is
374 specified, then the service must have
375 <varname>RemainAfterExit=yes</varname>
376 set.</para>
377
378 <para>For each of the specified
379 commands, the first argument must be
380 an absolute path to an executable.
381 Optionally, if this file name is
382 prefixed with <literal>@</literal>,
383 the second token will be passed as
384 <literal>argv[0]</literal> to the
385 executed process, followed by the
386 further arguments specified. If the
387 absolute filename is prefixed with
388 <literal>-</literal>, an exit code of
389 the command normally considered a
390 failure (i.e. non-zero exit status or
391 abnormal exit due to signal) is
392 ignored and considered success. If
393 both <literal>-</literal> and
394 <literal>@</literal> are used, they
395 can appear in either order.</para>
396
397 <para>If more than one command is
398 specified, the commands are invoked
399 sequentially in the order they appear
400 in the unit file. If one of the
401 commands fails (and is not prefixed
402 with <literal>-</literal>), other
403 lines are not executed, and the unit
404 is considered failed.</para>
405
406 <para>Unless
407 <varname>Type=forking</varname> is
408 set, the process started via this
409 command line will be considered the
410 main process of the daemon.</para>
411
412 </listitem>
413 </varlistentry>
414
415 <varlistentry>
416 <term><varname>ExecStartPre=</varname></term>
417 <term><varname>ExecStartPost=</varname></term>
418 <listitem><para>Additional commands
419 that are executed before or after
420 the command in
421 <varname>ExecStart=</varname>, respectively.
422 Syntax is the same as for
423 <varname>ExecStart=</varname>, except
424 that multiple command lines are allowed
425 and the commands are executed one
426 after the other, serially.</para>
427
428 <para>If any of those commands (not
429 prefixed with <literal>-</literal>)
430 fail, the rest are not executed and
431 the unit is considered failed.</para>
432 </listitem>
433 </varlistentry>
434
435 <varlistentry>
436 <term><varname>ExecReload=</varname></term>
437 <listitem><para>Commands to execute to
438 trigger a configuration reload in the
439 service. This argument takes multiple
440 command lines, following the same
441 scheme as described for
442 <varname>ExecStart=</varname>
443 above. Use of this setting is
444 optional. Specifier and environment
445 variable substitution is supported
446 here following the same scheme as for
447 <varname>ExecStart=</varname>.</para>
448
449 <para>One additional, special
450 environment variable is set: if known,
451 <varname>$MAINPID</varname> is set to
452 the main process of the daemon, and
453 may be used for command lines like the
454 following:</para>
455
456 <programlisting>/bin/kill -HUP $MAINPID</programlisting>
457
458 <para>Note however that reloading a
459 daemon by sending a signal (as with
460 the example line above) is usually not
461 a good choice, because this is an
462 asynchronous operation and hence not
463 suitable to order reloads of multiple
464 services against each other. It is
465 strongly recommended to set
466 <varname>ExecReload=</varname> to a
467 command that not only triggers a
468 configuration reload of the daemon,
469 but also synchronously waits for it to
470 complete.</para>
471 </listitem>
472 </varlistentry>
473
474 <varlistentry>
475 <term><varname>ExecStop=</varname></term>
476 <listitem><para>Commands to execute to
477 stop the service started via
478 <varname>ExecStart=</varname>. This
479 argument takes multiple command lines,
480 following the same scheme as described
481 for <varname>ExecStart=</varname>
482 above. Use of this setting is
483 optional. After the commands configured
484 in this option are run, all processes
485 remaining for a service are
486 terminated according to the
487 <varname>KillMode=</varname> setting
488 (see
489 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>). If
490 this option is not specified, the
491 process is terminated immediately when
492 service stop is requested. Specifier
493 and environment variable substitution
494 is supported (including
495 <varname>$MAINPID</varname>, see
496 above).</para></listitem>
497 </varlistentry>
498
499 <varlistentry>
500 <term><varname>ExecStopPost=</varname></term>
501 <listitem><para>Additional commands
502 that are executed after the service
503 was stopped. This includes cases where
504 the commands configured in
505 <varname>ExecStop=</varname> were used,
506 where the service does not have any
507 <varname>ExecStop=</varname> defined, or
508 where the service exited unexpectedly. This
509 argument takes multiple command lines,
510 following the same scheme as described
511 for <varname>ExecStart</varname>. Use
512 of these settings is
513 optional. Specifier and environment
514 variable substitution is
515 supported.</para></listitem>
516 </varlistentry>
517
518 <varlistentry>
519 <term><varname>RestartSec=</varname></term>
520 <listitem><para>Configures the time to
521 sleep before restarting a service (as
522 configured with
523 <varname>Restart=</varname>). Takes a
524 unit-less value in seconds, or a time
525 span value such as "5min
526 20s". Defaults to
527 100ms.</para></listitem>
528 </varlistentry>
529
530 <varlistentry>
531 <term><varname>TimeoutStartSec=</varname></term>
532 <listitem><para>Configures the time to
533 wait for start-up. If a
534 daemon service does not signal
535 start-up completion within the
536 configured time, the service will be
537 considered failed and will be shut
538 down again.
539 Takes a unit-less value in seconds, or a
540 time span value such as "5min
541 20s". Pass <literal>0</literal> to
542 disable the timeout logic. Defaults to
543 <varname>DefaultTimeoutStartSec=</varname> from
544 the manager configuration file, except
545 when <varname>Type=oneshot</varname> is
546 used, in which case the timeout
547 is disabled by default
548 (see <citerefentry><refentrytitle>systemd-system.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>).
549 </para></listitem>
550 </varlistentry>
551
552 <varlistentry>
553 <term><varname>TimeoutStopSec=</varname></term>
554 <listitem><para>Configures the time to
555 wait for stop. If a service is asked
556 to stop, but does not terminate in the
557 specified time, it will be terminated
558 forcibly via <constant>SIGTERM</constant>,
559 and after another timeout of equal duration
560 with <constant>SIGKILL</constant> (see
561 <varname>KillMode=</varname>
562 in <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>).
563 Takes a unit-less value in seconds, or a
564 time span value such as "5min
565 20s". Pass <literal>0</literal> to disable
566 the timeout logic. Defaults to
567 <varname>DefaultTimeoutStopSec=</varname> from the
568 manager configuration file
569 (see <citerefentry><refentrytitle>systemd-system.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>).
570 </para></listitem>
571 </varlistentry>
572
573 <varlistentry>
574 <term><varname>TimeoutSec=</varname></term>
575 <listitem><para>A shorthand for configuring
576 both <varname>TimeoutStartSec=</varname>
577 and <varname>TimeoutStopSec=</varname>
578 to the specified value.
579 </para></listitem>
580 </varlistentry>
581
582 <varlistentry>
583 <term><varname>WatchdogSec=</varname></term>
584 <listitem><para>Configures the
585 watchdog timeout for a service. The
586 watchdog is activated when the start-up is
587 completed. The service must call
588 <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>
589 regularly with <literal>WATCHDOG=1</literal>
590 (i.e. the "keep-alive ping"). If the time
591 between two such calls is larger than
592 the configured time, then the service
593 is placed in a failed state and it will
594 be terminated with <varname>SIGABRT</varname>.
595 By setting <varname>Restart=</varname> to
596 <option>on-failure</option> or
597 <option>always</option>, the service
598 will be automatically restarted. The
599 time configured here will be passed to
600 the executed service process in the
601 <varname>WATCHDOG_USEC=</varname>
602 environment variable. This allows
603 daemons to automatically enable the
604 keep-alive pinging logic if watchdog
605 support is enabled for the service. If
606 this option is used,
607 <varname>NotifyAccess=</varname> (see
608 below) should be set to open access to
609 the notification socket provided by
610 systemd. If
611 <varname>NotifyAccess=</varname> is
612 not set, it will be implicitly set to
613 <option>main</option>. Defaults to 0,
614 which disables this
615 feature.</para></listitem>
616 </varlistentry>
617
618 <varlistentry>
619 <term><varname>Restart=</varname></term>
620 <listitem><para>Configures whether the
621 service shall be restarted when the
622 service process exits, is killed,
623 or a timeout is reached. The service
624 process may be the main service
625 process, but it may also be one of the
626 processes specified with
627 <varname>ExecStartPre=</varname>,
628 <varname>ExecStartPost=</varname>,
629 <varname>ExecStop=</varname>,
630 <varname>ExecStopPost=</varname>, or
631 <varname>ExecReload=</varname>.
632 When the death of the process is a
633 result of systemd operation (e.g. service
634 stop or restart), the service will not be
635 restarted. Timeouts include missing
636 the watchdog "keep-alive ping"
637 deadline and a service start, reload,
638 and stop operation timeouts.</para>
639
640 <para>Takes one of
641 <option>no</option>,
642 <option>on-success</option>,
643 <option>on-failure</option>,
644 <option>on-abnormal</option>,
645 <option>on-watchdog</option>,
646 <option>on-abort</option>, or
647 <option>always</option>. If set to
648 <option>no</option> (the default), the
649 service will not be restarted. If set
650 to <option>on-success</option>, it
651 will be restarted only when the
652 service process exits cleanly. In
653 this context, a clean exit means an
654 exit code of 0, or one of the signals
655 <constant>SIGHUP</constant>,
656 <constant>SIGINT</constant>,
657 <constant>SIGTERM</constant> or
658 <constant>SIGPIPE</constant>, and
659 additionally, exit statuses and
660 signals specified in
661 <varname>SuccessExitStatus=</varname>.
662 If set to <option>on-failure</option>,
663 the service will be restarted when the
664 process exits with a non-zero exit
665 code, is terminated by a signal
666 (including on core dump, but excluding
667 the aforementiond four signals), when
668 an operation (such as service reload)
669 times out, and when the configured
670 watchdog timeout is triggered. If set
671 to <option>on-abnormal</option>, the
672 service will be restarted when the
673 process is terminated by a signal
674 (including on core dump, excluding the
675 aforementioned four signals), when an
676 operation times out, or when the
677 watchdog timeout is triggered. If set
678 to <option>on-abort</option>, the
679 service will be restarted only if the
680 service process exits due to an
681 uncaught signal not specified as a
682 clean exit status. If set to
683 <option>on-watchdog</option>, the
684 service will be restarted only if the
685 watchdog timeout for the service
686 expires. If set to
687 <option>always</option>, the service
688 will be restarted regardless of
689 whether it exited cleanly or not, got
690 terminated abnormally by a signal, or
691 hit a timeout.</para>
692
693 <table>
694 <title>Exit causes and the effect of the <varname>Restart=</varname> settings on them</title>
695
696 <tgroup cols='2'>
697 <colspec colname='path' />
698 <colspec colname='expl' />
699 <thead>
700 <row>
701 <entry>Restart settings/Exit causes</entry>
702 <entry><option>no</option></entry>
703 <entry><option>always</option></entry>
704 <entry><option>on-success</option></entry>
705 <entry><option>on-failure</option></entry>
706 <entry><option>on-abnormal</option></entry>
707 <entry><option>on-abort</option></entry>
708 <entry><option>on-watchdog</option></entry>
709 </row>
710 </thead>
711 <tbody>
712 <row>
713 <entry>Clean exit code or signal</entry>
714 <entry/>
715 <entry>X</entry>
716 <entry>X</entry>
717 <entry/>
718 <entry/>
719 <entry/>
720 <entry/>
721 </row>
722 <row>
723 <entry>Unclean exit code</entry>
724 <entry/>
725 <entry>X</entry>
726 <entry/>
727 <entry>X</entry>
728 <entry/>
729 <entry/>
730 <entry/>
731 </row>
732 <row>
733 <entry>Unclean signal</entry>
734 <entry/>
735 <entry>X</entry>
736 <entry/>
737 <entry>X</entry>
738 <entry>X</entry>
739 <entry>X</entry>
740 <entry/>
741 </row>
742 <row>
743 <entry>Timeout</entry>
744 <entry/>
745 <entry>X</entry>
746 <entry/>
747 <entry>X</entry>
748 <entry>X</entry>
749 <entry/>
750 <entry/>
751 </row>
752 <row>
753 <entry>Watchdog</entry>
754 <entry/>
755 <entry>X</entry>
756 <entry/>
757 <entry>X</entry>
758 <entry>X</entry>
759 <entry/>
760 <entry>X</entry>
761 </row>
762 </tbody>
763 </tgroup>
764 </table>
765
766 <para>As exceptions to the setting
767 above the service will not be
768 restarted if the exit code or signal
769 is specified in
770 <varname>RestartPreventExitStatus=</varname>
771 (see below). Also, the services will
772 always be restarted if the exit code
773 or signal is specified in
774 <varname>RestartForceExitStatus=</varname>
775 (see below).</para>
776
777 <para>Setting this to
778 <option>on-failure</option> is the
779 recommended choice for long-running
780 services, in order to increase
781 reliability by attempting automatic
782 recovery from errors. For services
783 that shall be able to terminate on
784 their own choice (and avoid
785 immediate restarting),
786 <option>on-abnormal</option> is an
787 alternative choice.</para>
788 </listitem>
789 </varlistentry>
790
791 <varlistentry>
792 <term><varname>SuccessExitStatus=</varname></term>
793 <listitem><para>Takes a list of exit
794 status definitions that when returned
795 by the main service process will be
796 considered successful termination, in
797 addition to the normal successful exit
798 code 0 and the signals <constant>SIGHUP</constant>, <constant>SIGINT</constant>,
799 <constant>SIGTERM</constant>, and <constant>SIGPIPE</constant>. Exit status
800 definitions can either be numeric exit
801 codes or termination signal names,
802 separated by spaces. For example:
803 <programlisting>SuccessExitStatus=1 2 8 SIGKILL</programlisting>
804 ensures that exit codes 1, 2, 8 and
805 the termination signal
806 <constant>SIGKILL</constant> are
807 considered clean service terminations.
808 </para>
809
810 <para>Note that if a process has a
811 signal handler installed and exits by
812 calling
813 <citerefentry><refentrytitle>_exit</refentrytitle><manvolnum>2</manvolnum></citerefentry>
814 in response to a signal, the
815 information about the signal is lost.
816 Programs should instead perform cleanup and kill themselves with the same signal instead. See
817 <ulink url="http://www.cons.org/cracauer/sigint.html">Proper handling of SIGINT/SIGQUIT — How to be a proper program</ulink>.</para>
818
819 <para>This option may appear more than once,
820 in which case the list of successful
821 exit statuses is merged. If the empty
822 string is assigned to this option, the
823 list is reset, all prior assignments
824 of this option will have no
825 effect.</para></listitem>
826 </varlistentry>
827
828 <varlistentry>
829 <term><varname>RestartPreventExitStatus=</varname></term>
830 <listitem><para>Takes a list of exit
831 status definitions that when returned
832 by the main service process will
833 prevent automatic service restarts,
834 regardless of the restart setting
835 configured with
836 <varname>Restart=</varname>. Exit
837 status definitions can either be
838 numeric exit codes or termination
839 signal names, and are separated by
840 spaces. Defaults to the empty list, so
841 that, by default, no exit status is
842 excluded from the configured restart
843 logic. For example:
844 <programlisting>RestartPreventExitStatus=1 6 SIGABRT</programlisting> ensures that exit
845 codes 1 and 6 and the termination
846 signal <constant>SIGABRT</constant> will
847 not result in automatic service
848 restarting. This
849 option may appear more than once, in
850 which case the list of restart-preventing
851 statuses is merged. If the empty
852 string is assigned to this option, the
853 list is reset and all prior assignments
854 of this option will have no
855 effect.</para></listitem>
856 </varlistentry>
857
858 <varlistentry>
859 <term><varname>RestartForceExitStatus=</varname></term>
860 <listitem><para>Takes a list of exit
861 status definitions that when returned
862 by the main service process will force
863 automatic service restarts, regardless
864 of the restart setting configured with
865 <varname>Restart=</varname>. The
866 argument format is similar to
867 <varname>RestartPreventExitStatus=</varname>.</para></listitem>
868 </varlistentry>
869
870 <varlistentry>
871 <term><varname>PermissionsStartOnly=</varname></term>
872 <listitem><para>Takes a boolean
873 argument. If true, the permission-related
874 execution options, as
875 configured with
876 <varname>User=</varname> and similar
877 options (see
878 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>
879 for more information), are only applied
880 to the process started with
881 <varname>ExecStart=</varname>, and not
882 to the various other
883 <varname>ExecStartPre=</varname>,
884 <varname>ExecStartPost=</varname>,
885 <varname>ExecReload=</varname>,
886 <varname>ExecStop=</varname>, and
887 <varname>ExecStopPost=</varname>
888 commands. If false, the setting is
889 applied to all configured commands the
890 same way. Defaults to
891 false.</para></listitem>
892 </varlistentry>
893
894 <varlistentry>
895 <term><varname>RootDirectoryStartOnly=</varname></term>
896 <listitem><para>Takes a boolean
897 argument. If true, the root directory,
898 as configured with the
899 <varname>RootDirectory=</varname>
900 option (see
901 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>
902 for more information), is only applied
903 to the process started with
904 <varname>ExecStart=</varname>, and not
905 to the various other
906 <varname>ExecStartPre=</varname>,
907 <varname>ExecStartPost=</varname>,
908 <varname>ExecReload=</varname>,
909 <varname>ExecStop=</varname>, and
910 <varname>ExecStopPost=</varname>
911 commands. If false, the setting is
912 applied to all configured commands the
913 same way. Defaults to
914 false.</para></listitem>
915 </varlistentry>
916
917 <varlistentry>
918 <term><varname>NonBlocking=</varname></term>
919 <listitem><para>Set the
920 <constant>O_NONBLOCK</constant> flag
921 for all file descriptors passed via
922 socket-based activation. If true, all
923 file descriptors >= 3 (i.e. all except
924 stdin, stdout, and stderr) will have
925 the <constant>O_NONBLOCK</constant> flag
926 set and hence are in
927 non-blocking mode. This option is only
928 useful in conjunction with a socket
929 unit, as described in
930 <citerefentry><refentrytitle>systemd.socket</refentrytitle><manvolnum>5</manvolnum></citerefentry>. Defaults
931 to false.</para></listitem>
932 </varlistentry>
933
934 <varlistentry>
935 <term><varname>NotifyAccess=</varname></term>
936 <listitem><para>Controls access to the
937 service status notification socket, as
938 accessible via the
939 <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>
940 call. Takes one of
941 <option>none</option> (the default),
942 <option>main</option> or
943 <option>all</option>. If
944 <option>none</option>, no daemon status
945 updates are accepted from the service
946 processes, all status update messages
947 are ignored. If <option>main</option>,
948 only service updates sent from the
949 main process of the service are
950 accepted. If <option>all</option>, all
951 services updates from all members of
952 the service's control group are
953 accepted. This option should be set to
954 open access to the notification socket
955 when using
956 <varname>Type=notify</varname> or
957 <varname>WatchdogSec=</varname> (see
958 above). If those options are used but
959 <varname>NotifyAccess=</varname> is not
960 configured, it will be implicitly set
961 to
962 <option>main</option>.</para></listitem>
963 </varlistentry>
964
965 <varlistentry>
966 <term><varname>Sockets=</varname></term>
967 <listitem><para>Specifies the name of
968 the socket units this service shall
969 inherit socket file descriptors
970 from when the service is
971 started. Normally it should not be
972 necessary to use this setting as all
973 socket file descriptors whose unit
974 shares the same name as the service
975 (subject to the different unit name
976 suffix of course) are passed to the
977 spawned process.</para>
978
979 <para>Note that the same socket file
980 descriptors may be passed to multiple
981 processes simultaneously. Also note
982 that a different service may be
983 activated on incoming socket traffic
984 than the one which is ultimately
985 configured to inherit the socket file
986 descriptors. Or in other words: the
987 <varname>Service=</varname> setting of
988 <filename>.socket</filename> units
989 does not have to match the inverse of
990 the <varname>Sockets=</varname>
991 setting of the
992 <filename>.service</filename> it
993 refers to.</para>
994
995 <para>This option may appear more than
996 once, in which case the list of socket
997 units is merged. If the empty string
998 is assigned to this option, the list of
999 sockets is reset, and all prior uses of
1000 this setting will have no
1001 effect.</para></listitem>
1002 </varlistentry>
1003
1004 <varlistentry>
1005 <term><varname>StartLimitInterval=</varname></term>
1006 <term><varname>StartLimitBurst=</varname></term>
1007
1008 <listitem><para>Configure service
1009 start rate limiting. By default,
1010 services which are started more
1011 than 5 times within 10 seconds are not
1012 permitted to start any more times
1013 until the 10 second interval ends. With
1014 these two options, this rate limiting
1015 may be modified. Use
1016 <varname>StartLimitInterval=</varname>
1017 to configure the checking interval (defaults to
1018 <varname>DefaultStartLimitInterval=</varname> in
1019 manager configuration file, set to 0 to disable
1020 any kind of rate limiting). Use
1021 <varname>StartLimitBurst=</varname> to
1022 configure how many starts per interval
1023 are allowed (defaults to
1024 <varname>DefaultStartLimitBurst=</varname> in
1025 manager configuration file). These
1026 configuration options are particularly
1027 useful in conjunction with
1028 <varname>Restart=</varname>; however,
1029 they apply to all kinds of starts
1030 (including manual), not just those
1031 triggered by the
1032 <varname>Restart=</varname> logic.
1033 Note that units which are configured
1034 for <varname>Restart=</varname> and
1035 which reach the start limit are not
1036 attempted to be restarted anymore;
1037 however, they may still be restarted
1038 manually at a later point, from which
1039 point on, the restart logic is again
1040 activated. Note that
1041 <command>systemctl
1042 reset-failed</command> will cause the
1043 restart rate counter for a service to
1044 be flushed, which is useful if the
1045 administrator wants to manually start
1046 a service and the start limit
1047 interferes with
1048 that.</para></listitem>
1049 </varlistentry>
1050
1051 <varlistentry>
1052 <term><varname>StartLimitAction=</varname></term>
1053
1054 <listitem><para>Configure the action
1055 to take if the rate limit configured
1056 with
1057 <varname>StartLimitInterval=</varname>
1058 and
1059 <varname>StartLimitBurst=</varname> is
1060 hit. Takes one of
1061 <option>none</option>,
1062 <option>reboot</option>,
1063 <option>reboot-force</option>,
1064 <option>reboot-immediate</option>,
1065 <option>poweroff</option>,
1066 <option>poweroff-force</option> or
1067 <option>poweroff-immediate</option>. If
1068 <option>none</option> is set, hitting
1069 the rate limit will trigger no action
1070 besides that the start will not be
1071 permitted. <option>reboot</option>
1072 causes a reboot following the normal
1073 shutdown procedure (i.e. equivalent to
1074 <command>systemctl reboot</command>).
1075 <option>reboot-force</option> causes a
1076 forced reboot which will terminate all
1077 processes forcibly but should cause no
1078 dirty file systems on reboot
1079 (i.e. equivalent to <command>systemctl
1080 reboot -f</command>) and
1081 <option>reboot-immediate</option>
1082 causes immediate execution of the
1083 <citerefentry><refentrytitle>reboot</refentrytitle><manvolnum>2</manvolnum></citerefentry>
1084 system call, which might result in
1085 data loss. Similar,
1086 <option>poweroff</option>,
1087 <option>poweroff-force</option>,
1088 <option>poweroff-immediate</option>
1089 have the effect of powering down the
1090 system with similar
1091 semantics. Defaults to
1092 <option>none</option>.</para></listitem>
1093 </varlistentry>
1094
1095 <varlistentry>
1096 <term><varname>FailureAction=</varname></term>
1097 <listitem><para>Configure the action
1098 to take when the service enters a failed
1099 state. Takes the same values as
1100 <varname>StartLimitAction=</varname>
1101 and executes the same actions.
1102 Defaults to <option>none</option>.
1103 </para></listitem>
1104 </varlistentry>
1105
1106 <varlistentry>
1107 <term><varname>RebootArgument=</varname></term>
1108 <listitem><para>Configure the optional
1109 argument for the
1110 <citerefentry><refentrytitle>reboot</refentrytitle><manvolnum>2</manvolnum></citerefentry>
1111 system call if
1112 <varname>StartLimitAction=</varname>
1113 or <varname>FailureAction=</varname>
1114 is a reboot action. This works just
1115 like the optional argument to
1116 <command>systemctl reboot</command>
1117 command.</para></listitem>
1118 </varlistentry>
1119
1120 <varlistentry>
1121 <term><varname>FileDescriptorStoreMax=</varname></term>
1122 <listitem><para>Configure how many
1123 file descriptors may be stored in the
1124 service manager for the service using
1125 <citerefentry><refentrytitle>sd_pid_notify_with_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>'s
1126 <literal>FDSTORE=1</literal>
1127 messages. This is useful for
1128 implementing service restart schemes
1129 where the state is serialized to
1130 <filename>/run</filename> and the file
1131 descriptors passed to the service
1132 manager, to allow restarts without
1133 losing state. Defaults to 0, i.e. no
1134 file descriptors may be stored in the
1135 service manager by default. All file
1136 descriptors passed to the service
1137 manager from a specific service are
1138 passed back to the service's main
1139 process on the next service
1140 restart. Any file descriptors passed
1141 to the service manager are
1142 automatically closed when POLLHUP or
1143 POLLERR is seen on them, or when the
1144 service is fully stopped and no job
1145 queued or being executed for
1146 it.</para></listitem>
1147 </varlistentry>
1148
1149 </variablelist>
1150
1151 <para>Check
1152 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1153 and
1154 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1155 for more settings.</para>
1156
1157 </refsect1>
1158
1159 <refsect1>
1160 <title>Command lines</title>
1161
1162 <para>This section describes command line parsing and
1163 variable and specifier substitions for
1164 <varname>ExecStart=</varname>,
1165 <varname>ExecStartPre=</varname>,
1166 <varname>ExecStartPost=</varname>,
1167 <varname>ExecReload=</varname>,
1168 <varname>ExecStop=</varname>, and
1169 <varname>ExecStopPost=</varname> options.</para>
1170
1171 <para>Multiple command lines may be concatenated in a
1172 single directive by separating them with semicolons
1173 (these semicolons must be passed as separate words).
1174 Lone semicolons may be escaped as
1175 <literal>\;</literal>.</para>
1176
1177 <para>Each command line is split on whitespace, with
1178 the first item being the command to execute, and the
1179 subsequent items being the arguments. Double quotes
1180 ("...") and single quotes ('...') may be used, in
1181 which case everything until the next matching quote
1182 becomes part of the same argument. C-style escapes are
1183 also supported, see table below. Quotes themselves are
1184 removed after parsing and escape sequences
1185 substituted. In addition, a trailing backslash
1186 (<literal>\</literal>) may be used to merge lines.
1187 </para>
1188
1189 <para>This syntax is intended to be very similar to
1190 shell syntax, but only the meta-characters and
1191 expansions described in the following paragraphs are
1192 understood. Specifically, redirection using
1193 <literal>&lt;</literal>, <literal>&lt;&lt;</literal>,
1194 <literal>&gt;</literal>, and
1195 <literal>&gt;&gt;</literal>, pipes using
1196 <literal>|</literal>, running programs in the
1197 background using <literal>&amp;</literal>, and
1198 <emphasis>other elements of shell syntax are not
1199 supported</emphasis>.</para>
1200
1201 <para>The command to execute must an absolute path
1202 name. It may contain spaces, but control characters
1203 are not allowed.</para>
1204
1205 <para>The command line accepts <literal>%</literal>
1206 specifiers as described in
1207 <citerefentry><refentrytitle>systemd.unit</refentrytitle><manvolnum>5</manvolnum></citerefentry>.
1208 Note that the first argument of the command line
1209 (i.e. the program to execute) may not include
1210 specifiers.</para>
1211
1212 <para>Basic environment variable substitution is
1213 supported. Use <literal>${FOO}</literal> as part of a
1214 word, or as a word of its own, on the command line, in
1215 which case it will be replaced by the value of the
1216 environment variable including all whitespace it
1217 contains, resulting in a single argument. Use
1218 <literal>$FOO</literal> as a separate word on the
1219 command line, in which case it will be replaced by the
1220 value of the environment variable split at whitespace
1221 resulting in zero or more arguments. For this type of
1222 expansion, quotes and respected when splitting into
1223 words, and afterwards removed.</para>
1224
1225 <para>Example:</para>
1226
1227 <programlisting>Environment="ONE=one" 'TWO=two two'
1228ExecStart=/bin/echo $ONE $TWO ${TWO}</programlisting>
1229
1230 <para>This will execute <command>/bin/echo</command>
1231 with four arguments: <literal>one</literal>,
1232 <literal>two</literal>, <literal>two</literal>, and
1233 <literal>two two</literal>.</para>
1234
1235 <para>Example:</para>
1236 <programlisting>Environment=ONE='one' "TWO='two two' too" THREE=
1237ExecStart=/bin/echo ${ONE} ${TWO} ${THREE}
1238ExecStart=/bin/echo $ONE $TWO $THREE</programlisting>
1239 <para>This results in <filename>echo</filename> being
1240 called twice, the first time with arguments
1241 <literal>'one'</literal>,
1242 <literal>'two two' too</literal>, <literal></literal>,
1243 and the second time with arguments
1244 <literal>one</literal>, <literal>two two</literal>,
1245 <literal>too</literal>.
1246 </para>
1247
1248 <para>To pass a literal dollar sign, use
1249 <literal>$$</literal>. Variables whose value is not
1250 known at expansion time are treated as empty
1251 strings. Note that the first argument (i.e. the
1252 program to execute) may not be a variable.</para>
1253
1254 <para>Variables to be used in this fashion may be
1255 defined through <varname>Environment=</varname> and
1256 <varname>EnvironmentFile=</varname>. In addition,
1257 variables listed in the section "Environment variables
1258 in spawned processes" in
1259 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
1260 which are considered "static configuration", may be
1261 used (this includes e.g. <varname>$USER</varname>, but
1262 not <varname>$TERM</varname>).</para>
1263
1264 <para>Note that shell command lines are not directly
1265 supported. If shell command lines are to be used, they
1266 need to be passed explicitly to a shell implementation
1267 of some kind. Example:</para>
1268 <programlisting>ExecStart=/bin/sh -c 'dmesg | tac'</programlisting>
1269
1270 <para>Example:</para>
1271
1272 <programlisting>ExecStart=/bin/echo one ; /bin/echo "two two"</programlisting>
1273
1274 <para>This will execute <command>/bin/echo</command>
1275 two times, each time with one argument:
1276 <literal>one</literal> and <literal>two two</literal>,
1277 respectively. Because two commands are specified,
1278 <varname>Type=oneshot</varname> must be used.</para>
1279
1280 <para>Example:</para>
1281
1282 <programlisting>ExecStart=/bin/echo / &gt;/dev/null &amp; \; \
1283/bin/ls</programlisting>
1284
1285 <para>This will execute <command>/bin/echo</command>
1286 with five arguments: <literal>/</literal>,
1287 <literal>&gt;/dev/null</literal>,
1288 <literal>&amp;</literal>, <literal>;</literal>, and
1289 <literal>/bin/ls</literal>.</para>
1290
1291 <table>
1292 <title>C escapes supported in command lines and environment variables</title>
1293 <tgroup cols='2'>
1294 <colspec colname='escape' />
1295 <colspec colname='meaning' />
1296 <thead>
1297 <row>
1298 <entry>Literal</entry>
1299 <entry>Actual value</entry>
1300 </row>
1301 </thead>
1302 <tbody>
1303 <row>
1304 <entry><literal>\a</literal></entry>
1305 <entry>bell</entry>
1306 </row>
1307 <row>
1308 <entry><literal>\b</literal></entry>
1309 <entry>backspace</entry>
1310 </row>
1311 <row>
1312 <entry><literal>\f</literal></entry>
1313 <entry>form feed</entry>
1314 </row>
1315 <row>
1316 <entry><literal>\n</literal></entry>
1317 <entry>newline</entry>
1318 </row>
1319 <row>
1320 <entry><literal>\r</literal></entry>
1321 <entry>carriage return</entry>
1322 </row>
1323 <row>
1324 <entry><literal>\t</literal></entry>
1325 <entry>tab</entry>
1326 </row>
1327 <row>
1328 <entry><literal>\v</literal></entry>
1329 <entry>vertical tab</entry>
1330 </row>
1331 <row>
1332 <entry><literal>\\</literal></entry>
1333 <entry>backslash</entry>
1334 </row>
1335 <row>
1336 <entry><literal>\"</literal></entry>
1337 <entry>double quotation mark</entry>
1338 </row>
1339 <row>
1340 <entry><literal>\'</literal></entry>
1341 <entry>single quotation mark</entry>
1342 </row>
1343 <row>
1344 <entry><literal>\s</literal></entry>
1345 <entry>space</entry>
1346 </row>
1347 <row>
1348 <entry><literal>\x<replaceable>xx</replaceable></literal></entry>
1349 <entry>character number <replaceable>xx</replaceable> in hexadecimal encoding</entry>
1350 </row>
1351 <row>
1352 <entry><literal>\<replaceable>nnn</replaceable></literal></entry>
1353 <entry>character number <replaceable>nnn</replaceable> in octal encoding</entry>
1354 </row>
1355 </tbody>
1356 </tgroup>
1357 </table>
1358 </refsect1>
1359
1360 <refsect1>
1361 <title>Examples</title>
1362
1363 <example>
1364 <title>Simple service</title>
1365
1366 <para>The following unit file creates a service
1367 that will execute
1368 <filename>/usr/sbin/foo-daemon</filename>.
1369 Since no <varname>Type=</varname> is specified,
1370 the default
1371 <varname>Type=</varname><option>simple</option>
1372 will be assumed. systemd will assume the unit
1373 to be started immediately after the program has
1374 begun executing.</para>
1375
1376 <programlisting>[Unit]
1377Description=Foo
1378
1379[Service]
1380ExecStart=/usr/sbin/foo-daemon
1381
1382[Install]
1383WantedBy=multi-user.target</programlisting>
1384
1385 <para>Note that systemd assumes here that the
1386 process started by systemd will continue
1387 running until the service terminates. If the
1388 program daemonizes itself (i.e. forks), please
1389 use
1390 <varname>Type=</varname><option>forking</option>
1391 instead.</para>
1392
1393 <para>Since no <varname>ExecStop=</varname> was
1394 specified, systemd will send SIGTERM to all
1395 processes started from this service, and after
1396 a timeout also SIGKILL. This behavior can be
1397 modified, see
1398 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1399 for details.</para>
1400
1401 <para>Note that this unit type does not include
1402 any type of notification when a service has
1403 completed initialization. For this, you should
1404 use other unit types, such as
1405 <varname>Type=</varname><option>notify</option>
1406 if the service understands systemd's
1407 notification protocol,
1408 <varname>Type=</varname><option>forking</option>
1409 if the service can background itself or
1410 <varname>Type=</varname><option>dbus</option>
1411 if the unit acquires a DBus name once
1412 initialization is complete. See below.</para>
1413 </example>
1414
1415 <example>
1416 <title>Oneshot service</title>
1417
1418 <para>Sometimes units should just execute an
1419 action without keeping active processes, such
1420 as a filesystem check or a cleanup action on
1421 boot. For this,
1422 <varname>Type=</varname><option>oneshot</option>
1423 exists. Units of this type will wait until the
1424 process specified terminates and then fall back
1425 to being inactive. The following unit will
1426 perform a clenaup action:</para>
1427
1428 <programlisting>[Unit]
1429Description=Cleanup old Foo data
1430
1431[Service]
1432Type=oneshot
1433ExecStart=/usr/sbin/foo-cleanup
1434
1435[Install]
1436WantedBy=multi-user.target</programlisting>
1437
1438 <para>Note that systemd will consider the unit
1439 to be in the state 'starting' until the program
1440 has terminated, so ordered dependencies will
1441 wait for the program to finish before starting
1442 themselves. The unit will revert to the
1443 'inactive' state after the execution is
1444 done, never reaching the 'active' state. That
1445 means another request to start the unit will
1446 perform the action again.</para>
1447
1448 <para><varname>Type=</varname><option>oneshot</option>
1449 are the only service units that may have more
1450 than one <varname>ExecStart=</varname>
1451 specified. They will be executed in order until
1452 either they are all successful or one of them
1453 fails.</para>
1454 </example>
1455
1456 <example>
1457 <title>Stoppable oneshot service</title>
1458
1459 <para>Similarly to the oneshot services, there
1460 are sometimes units that need to execute a
1461 program to set up something and then execute
1462 another to shut it down, but no process remains
1463 active while they are considered
1464 'started'. Network configuration can sometimes
1465 fall into this category. Another use case is if
1466 a oneshot service shall not be executed a
1467 each time when they are pulled in as a
1468 dependency, but only the first time.</para>
1469
1470 <para>For this, systemd knows the setting
1471 <varname>RemainAfterExit=</varname><option>yes</option>,
1472 which causes systemd to consider the unit to be
1473 active if the start action exited successfully.
1474 This directive can be used with all types, but
1475 is most useful with
1476 <varname>Type=</varname><option>oneshot</option>
1477 and
1478 <varname>Type=</varname><option>simple</option>.
1479 With
1480 <varname>Type=</varname><option>oneshot</option>
1481 systemd waits until the start action has
1482 completed before it considers the unit to be
1483 active, so dependencies start only after the
1484 start action has succeeded. With
1485 <varname>Type=</varname><option>simple</option>
1486 dependencies will start immediately after the
1487 start action has been dispatched. The following
1488 unit provides an example for a simple static
1489 firewall.</para>
1490
1491 <programlisting>[Unit]
1492Description=Simple firewall
1493
1494[Service]
1495Type=oneshot
1496RemainAfterExit=yes
1497ExecStart=/usr/local/sbin/simple-firewall-start
1498ExecStop=/usr/local/sbin/simple-firewall-stop
1499
1500[Install]
1501WantedBy=multi-user.target</programlisting>
1502
1503 <para>Since the unit is considered to be
1504 running after the start action has exited,
1505 invoking <command>systemctl start</command> on
1506 that unit again will cause no action to be
1507 taken.</para>
1508 </example>
1509
1510 <example>
1511 <title>Traditional forking services</title>
1512
1513 <para>Many traditional daemons/services
1514 background (i.e. fork, daemonize) themselves
1515 when starting. Set
1516 <varname>Type=</varname><option>forking</option>
1517 in the service's unit file to support this mode
1518 of operation. systemd will consider the service
1519 to be in the process of initialization while
1520 the original program is still running. Once
1521 it exits successfully and at least a process
1522 remains (and
1523 <varname>RemainAfterExit=</varname><option>no</option>),
1524 the service is considered started.</para>
1525
1526 <para>Often a traditional daemon only consists
1527 of one process. Therefore, if only one process
1528 is left after the original process terminates,
1529 systemd will consider that process the main
1530 process of the service. In that case, the
1531 <varname>$MAINPID</varname> variable will be
1532 available in <varname>ExecReload=</varname>,
1533 <varname>ExecStop=</varname>, etc.</para>
1534
1535 <para>In case more than one process remains,
1536 systemd will be unable to determine the main
1537 process, so it will not assume there is one.
1538 In that case, <varname>$MAINPID</varname> will
1539 not expand to anything. However, if the process
1540 decides to write a traditional PID file,
1541 systemd will be able to read the main PID from
1542 there. Please set <varname>PIDFile=</varname>
1543 accordingly. Note that the daemon should write
1544 that file before finishing with its
1545 initialization, otherwise systemd might try to
1546 read the file before it exists.</para>
1547
1548 <para>The following example shows a simple
1549 daemon that forks and just starts one process
1550 in the background:</para>
1551
1552 <programlisting>[Unit]
1553Description=Some simple daemon
1554
1555[Service]
1556Type=forking
1557ExecStart=/usr/sbin/my-simple-daemon -d
1558
1559[Install]
1560WantedBy=multi-user.target</programlisting>
1561
1562 <para>Please see
1563 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1564 for details on how you can influence the way
1565 systemd terminates the service.</para>
1566 </example>
1567
1568 <example>
1569 <title>DBus services</title>
1570
1571 <para>For services that acquire a name on the
1572 DBus system bus, use
1573 <varname>Type=</varname><option>dbus</option>
1574 and set <varname>BusName=</varname>
1575 accordingly. The service should not fork
1576 (daemonize). systemd will consider the service
1577 to be initialized once the name has been
1578 acquired on the system bus. The following
1579 example shows a typical DBus service:</para>
1580
1581 <programlisting>[Unit]
1582Description=Simple DBus service
1583
1584[Service]
1585Type=dbus
1586BusName=org.example.simple-dbus-service
1587ExecStart=/usr/sbin/simple-dbus-service
1588
1589[Install]
1590WantedBy=multi-user.target</programlisting>
1591
1592 <para>For <emphasis>bus-activatable</emphasis>
1593 services, don't include a
1594 <literal>[Install]</literal> section in the
1595 systemd service file, but use the
1596 <varname>SystemdService=</varname> option in
1597 the corresponding DBus service file, for
1598 example
1599 (<filename>/usr/share/dbus-1/system-services/org.example.simple-dbus-service.service</filename>):</para>
1600
1601 <programlisting>[D-BUS Service]
1602Name=org.example.simple-dbus-service
1603Exec=/usr/sbin/simple-dbus-service
1604User=root
1605SystemdService=simple-dbus-service.service</programlisting>
1606
1607 <para>Please see
1608 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1609 for details on how you can influence the way
1610 systemd terminates the service.</para>
1611 </example>
1612
1613 <example>
1614 <title>Services that notify systemd about their initialization</title>
1615
1616 <para><varname>Type=</varname><option>simple</option>
1617 services are really easy to write, but have the
1618 major disadvantage of systemd not being able to
1619 tell when initialization of the given service
1620 is complete. For this reason, systemd supports
1621 a simple notification protocol that allows
1622 daemons to make systemd aware that they are
1623 done initializing. Use
1624 <varname>Type=</varname><option>notify</option>
1625 for this. A typical service file for such a
1626 daemon would look like this:</para>
1627
1628 <programlisting>[Unit]
1629Description=Simple notifying service
1630
1631[Service]
1632Type=notify
1633ExecStart=/usr/sbin/simple-notifying-service
1634
1635[Install]
1636WantedBy=multi-user.target</programlisting>
1637
1638 <para>Note that the daemon has to support
1639 systemd's notification protocol, else systemd
1640 will think the service hasn't started yet and
1641 kill it after a timeout. For an example of how
1642 to update daemons to support this protocol
1643 transparently, take a look at
1644 <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>.
1645 systemd will consider the unit to be in the
1646 'starting' state until a readiness notification
1647 has arrived.</para>
1648
1649 <para>Please see
1650 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>
1651 for details on how you can influence the way
1652 systemd terminates the service.</para>
1653 </example>
1654 </refsect1>
1655
1656 <refsect1>
1657 <title>See Also</title>
1658 <para>
1659 <citerefentry><refentrytitle>systemd</refentrytitle><manvolnum>1</manvolnum></citerefentry>,
1660 <citerefentry><refentrytitle>systemctl</refentrytitle><manvolnum>1</manvolnum></citerefentry>,
1661 <citerefentry><refentrytitle>systemd.unit</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
1662 <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
1663 <citerefentry><refentrytitle>systemd.resource-control</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
1664 <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>,
1665 <citerefentry><refentrytitle>systemd.directives</refentrytitle><manvolnum>7</manvolnum></citerefentry>
1666 </para>
1667 </refsect1>
1668
1669</refentry>