chiark / gitweb /
man: update journald rate limit defaults
[elogind.git] / man / daemon.xml
index 650e0fa8b1801d069eca9c8e4cfbec0069557766..b1f43f559475ebeccf24d556154ec16b3e446861 100644 (file)
@@ -8,20 +8,20 @@
   Copyright 2010 Lennart Poettering
 
   systemd is free software; you can redistribute it and/or modify it
   Copyright 2010 Lennart Poettering
 
   systemd is free software; you can redistribute it and/or modify it
-  under the terms of the GNU General Public License as published by
-  the Free Software Foundation; either version 2 of the License, or
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
   (at your option) any later version.
 
   systemd is distributed in the hope that it will be useful, but
   WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
   (at your option) any later version.
 
   systemd is distributed in the hope that it will be useful, but
   WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
-  General Public License for more details.
+  Lesser General Public License for more details.
 
 
-  You should have received a copy of the GNU General Public License
+  You should have received a copy of the GNU Lesser General Public License
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 -->
 
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 -->
 
-<refentry id="systemd.special">
+<refentry id="daemon">
 
         <refentryinfo>
                 <title>daemon</title>
 
         <refentryinfo>
                 <title>daemon</title>
@@ -44,7 +44,7 @@
 
         <refnamediv>
                 <refname>daemon</refname>
 
         <refnamediv>
                 <refname>daemon</refname>
-                <refpurpose>Writing and Packaging System Daemons</refpurpose>
+                <refpurpose>Writing and packaging system daemons</refpurpose>
         </refnamediv>
 
         <refsect1>
         </refnamediv>
 
         <refsect1>
                 functionality to other processes. Traditionally,
                 daemons are implemented following a scheme originating
                 in SysV Unix. Modern daemons should follow a simpler
                 functionality to other processes. Traditionally,
                 daemons are implemented following a scheme originating
                 in SysV Unix. Modern daemons should follow a simpler
-                yet more powerful scheme here called "new-style"
-                daemons, as implemented by systemd. </para>
+                yet more powerful scheme (here called "new-style"
+                daemons), as implemented by
+                <citerefentry><refentrytitle>systemd</refentrytitle><manvolnum>1</manvolnum></citerefentry>. This
+                manual page covers both schemes, and in
+                particular includes recommendations for daemons that
+                shall be included in the systemd init system.</para>
 
                 <refsect2>
                         <title>SysV Daemons</title>
 
                 <refsect2>
                         <title>SysV Daemons</title>
                         <para>When a traditional SysV daemon
                         starts, it should execute the following steps
                         as part of the initialization. Note that these
                         <para>When a traditional SysV daemon
                         starts, it should execute the following steps
                         as part of the initialization. Note that these
-                        steps are unnecessary for new-style daemons,
-                        and should only be implemented if compatbility
+                        steps are unnecessary for new-style daemons (see below),
+                        and should only be implemented if compatibility
                         with SysV is essential.</para>
 
                         <orderedlist>
                                 <listitem><para>Close all open file
                         with SysV is essential.</para>
 
                         <orderedlist>
                                 <listitem><para>Close all open file
-                                descriptors except STDIN, STDOUT,
-                                STDERR (i.e. the first three file
+                                descriptors except standard input, output,
+                                and error (i.e. the first three file
                                 descriptors 0, 1, 2). This ensures
                                 that no accidentally passed file
                                 descriptor stays around in the daemon
                                 descriptors 0, 1, 2). This ensures
                                 that no accidentally passed file
                                 descriptor stays around in the daemon
-                                process. On Linux this is best
+                                process. On Linux, this is best
                                 implemented by iterating through
                                 <filename>/proc/self/fd</filename>,
                                 with a fallback of iterating from file
                                 descriptor 3 to the value returned by
                                 implemented by iterating through
                                 <filename>/proc/self/fd</filename>,
                                 with a fallback of iterating from file
                                 descriptor 3 to the value returned by
-                                getrlimit() for
+                                <function>getrlimit()</function> for
                                 RLIMIT_NOFILE.</para></listitem>
 
                                 <listitem><para>Reset all signal
                                 RLIMIT_NOFILE.</para></listitem>
 
                                 <listitem><para>Reset all signal
                                 best done by iterating through the
                                 available signals up to the limit of
                                 _NSIG and resetting them to
                                 best done by iterating through the
                                 available signals up to the limit of
                                 _NSIG and resetting them to
-                                SIG_DFL.</para></listitem>
+                                <constant>SIG_DFL</constant>.</para></listitem>
 
                                 <listitem><para>Reset the signal mask
 
                                 <listitem><para>Reset the signal mask
-                                using sigprocmask().</para></listitem>
+                                using
+                                <function>sigprocmask()</function>.</para></listitem>
 
 
-                                <listitem><para>Call fork(),
+                                <listitem><para>Sanitize the
+                                environment block, removing or
+                                resetting environment variables that
+                                might negatively impact daemon
+                                runtime.</para></listitem>
+
+                                <listitem><para>Call <function>fork()</function>,
                                 to create a background
                                 process.</para></listitem>
 
                                 <listitem><para>In the child, call
                                 to create a background
                                 process.</para></listitem>
 
                                 <listitem><para>In the child, call
-                                setsid() to detach from any terminal
-                                and create an independent
-                                session.</para></listitem>
+                                <function>setsid()</function> to
+                                detach from any terminal and create an
+                                independent session.</para></listitem>
 
                                 <listitem><para>In the child, call
 
                                 <listitem><para>In the child, call
-                                fork() again, to ensure the daemon can
-                                never re-aquire a terminal
-                                again.</para></listitem>
+                                <function>fork()</function> again, to
+                                ensure that the daemon can never re-acquire
+                                a terminal again.</para></listitem>
 
 
-                                <listitem><para>Call exit() in the
+                                <listitem><para>Call <function>exit()</function> in the
                                 first child, so that only the second
                                 child (the actual daemon process)
                                 stays around. This ensures that the
                                 first child, so that only the second
                                 child (the actual daemon process)
                                 stays around. This ensures that the
-                                daemon process is reparented to
+                                daemon process is re-parented to
                                 init/PID 1, as all daemons should
                                 be.</para></listitem>
 
                                 <listitem><para>In the daemon process,
                                 connect <filename>/dev/null</filename>
                                 init/PID 1, as all daemons should
                                 be.</para></listitem>
 
                                 <listitem><para>In the daemon process,
                                 connect <filename>/dev/null</filename>
-                                to STDIN, STDOUT,
-                                STDERR.</para></listitem>
+                                to standard input, output, and error.
+                                </para></listitem>
 
                                 <listitem><para>In the daemon process,
                                 reset the umask to 0, so that the file
 
                                 <listitem><para>In the daemon process,
                                 reset the umask to 0, so that the file
-                                modes passed to open(), mkdir() and
+                                modes passed to <function>open()</function>, <function>mkdir()</function> and
                                 suchlike directly control the access
                                 mode of the created files and
                                 directories.</para></listitem>
                                 suchlike directly control the access
                                 mode of the created files and
                                 directories.</para></listitem>
                                 blocks mount points from being
                                 unmounted.</para></listitem>
 
                                 blocks mount points from being
                                 unmounted.</para></listitem>
 
+                                <listitem><para>In the daemon process,
+                                write the daemon PID (as returned by
+                                <function>getpid()</function>) to a
+                                PID file, for example
+                                <filename>/run/foobar.pid</filename>
+                                (for a hypothetical daemon "foobar")
+                                to ensure that the daemon cannot be
+                                started more than once. This must be
+                                implemented in race-free fashion so
+                                that the PID file is only updated when
+                                it is verified at the same time that
+                                the PID previously stored in the PID
+                                file no longer exists or belongs to a
+                                foreign process. Commonly, some kind of
+                                file locking is employed to implement
+                                this logic.</para></listitem>
+
                                 <listitem><para>In the daemon process,
                                 drop privileges, if possible and
                                 applicable.</para></listitem>
 
                                 <listitem><para>From the daemon
                                 <listitem><para>In the daemon process,
                                 drop privileges, if possible and
                                 applicable.</para></listitem>
 
                                 <listitem><para>From the daemon
-                                process notify the original process
+                                process, notify the original process
                                 started that initialization is
                                 complete. This can be implemented via
                                 an unnamed pipe or similar
                                 communication channel that is created
                                 started that initialization is
                                 complete. This can be implemented via
                                 an unnamed pipe or similar
                                 communication channel that is created
-                                before the first fork() and available
-                                in both processes.</para></listitem>
+                                before the first
+                                <function>fork()</function> and hence
+                                available in both the original and the
+                                daemon process.</para></listitem>
 
 
-                                <listitem><para>Call exit() in the
+                                <listitem><para>Call
+                                <function>exit()</function> in the
                                 original process. The process that
                                 invoked the daemon must be able to
                                 original process. The process that
                                 invoked the daemon must be able to
-                                rely that this exit() happens after
-                                initialization is complete and all
-                                external communication channels
-                                established and
+                                rely on that this
+                                <function>exit()</function> happens
+                                after initialization is complete and
+                                all external communication channels
+                                are established and
                                 accessible.</para></listitem>
                         </orderedlist>
 
                                 accessible.</para></listitem>
                         </orderedlist>
 
-                        <para>The BSD daemon() function should not be
-                        used, as it does only a subset of these steps.</para>
+                        <para>The BSD <function>daemon()</function> function should not be
+                        used, as it implements only a subset of these steps.</para>
 
                         <para>A daemon that needs to provide
                         compatibility with SysV systems should
                         implement the scheme pointed out
                         above. However, it is recommended to make this
 
                         <para>A daemon that needs to provide
                         compatibility with SysV systems should
                         implement the scheme pointed out
                         above. However, it is recommended to make this
-                        behaviour optional and configurable via a
-                        command line argument, to ease debugging as
+                        behavior optional and configurable via a
+                        command line argument to ease debugging as
                         well as to simplify integration into systems
                         using systemd.</para>
                 </refsect2>
                         well as to simplify integration into systems
                         using systemd.</para>
                 </refsect2>
                         runtime and simplifies their
                         implementation.</para>
 
                         runtime and simplifies their
                         implementation.</para>
 
-                        <para>For developing a new-style daemon none
+                        <para>For developing a new-style daemon, none
                         of the initialization steps recommended for
                         SysV daemons need to be implemented. New-style
                         init systems such as systemd make all of them
                         redundant. Moreover, since some of these steps
                         interfere with process monitoring, file
                         descriptor passing and other functionality of
                         of the initialization steps recommended for
                         SysV daemons need to be implemented. New-style
                         init systems such as systemd make all of them
                         redundant. Moreover, since some of these steps
                         interfere with process monitoring, file
                         descriptor passing and other functionality of
-                        the init system it is recommended not to
+                        the init system, it is recommended not to
                         execute them when run as new-style
                         service.</para>
 
                         execute them when run as new-style
                         service.</para>
 
+                        <para>Note that new-style init systems
+                        guarantee execution of daemon processes in a
+                        clean process context: it is guaranteed that
+                        the environment block is sanitized, that the
+                        signal handlers and mask is reset and that no
+                        left-over file descriptors are passed. Daemons
+                        will be executed in their own session, with
+                        standard input/output/error connected to
+                        <filename>/dev/null</filename> unless
+                        otherwise configured. The umask is reset.
+                        </para>
+
                         <para>It is recommended for new-style daemons
                         to implement the following:</para>
 
                         <orderedlist>
                         <para>It is recommended for new-style daemons
                         to implement the following:</para>
 
                         <orderedlist>
-                                <listitem><para>If SIGTERM is
+                                <listitem><para>If <constant>SIGTERM</constant> is
                                 received, shut down the daemon and
                                 exit cleanly.</para></listitem>
 
                                 received, shut down the daemon and
                                 exit cleanly.</para></listitem>
 
-                                <listitem><para>If SIGHUP is received,
+                                <listitem><para>If <constant>SIGHUP</constant> is received,
                                 reload the configuration files, if
                                 this applies.</para></listitem>
 
                                 reload the configuration files, if
                                 this applies.</para></listitem>
 
                                 this is used by the init system to
                                 detect service errors and problems. It
                                 is recommended to follow the exit code
                                 this is used by the init system to
                                 detect service errors and problems. It
                                 is recommended to follow the exit code
-                                scheme as defined in LSB
-                                recommendations for SysV init scripts
-                                (http://refspecs.freestandards.org/LSB_3.1.1/LSB-Core-generic/LSB-Core-generic/iniscrptact.html).</para></listitem>
-
-                                <listitem><para>As much as possible,
-                                rely on systemd's functionality to
-                                limit the accces of the daemon to
-                                files, services and other
-                                resources. i.e. rely on systemd's
-                                resource limit control instead of
-                                implementing your own, rely on
-                                systemd's privilege dropping code
-                                instead of implementing it in the
-                                daemon, and similar.</para></listitem>
+                                scheme as defined in the <ulink
+                                url="http://refspecs.linuxbase.org/LSB_3.1.1/LSB-Core-generic/LSB-Core-generic/iniscrptact.html">LSB
+                                recommendations for SysV init
+                                scripts</ulink>.</para></listitem>
 
                                 <listitem><para>If possible and
 
                                 <listitem><para>If possible and
-                                applicable expose the daemon's control
+                                applicable, expose the daemon's control
                                 interface via the D-Bus IPC system and
                                 grab a bus name as last step of
                                 initialization.</para></listitem>
 
                                 interface via the D-Bus IPC system and
                                 grab a bus name as last step of
                                 initialization.</para></listitem>
 
+                                <listitem><para>For integration in
+                                systemd, provide a
+                                <filename>.service</filename> unit
+                                file that carries information about
+                                starting, stopping and otherwise
+                                maintaining the daemon. See
+                                <citerefentry><refentrytitle>systemd.service</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                                for details.</para></listitem>
+
+                                <listitem><para>As much as possible,
+                                rely on the init system's
+                                functionality to limit the access of
+                                the daemon to files, services and
+                                other resources, i.e. in the case of
+                                systemd, rely on systemd's resource
+                                limit control instead of implementing
+                                your own, rely on systemd's privilege
+                                dropping code instead of implementing
+                                it in the daemon, and similar. See
+                                <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                                for the available
+                                controls.</para></listitem>
+
                                 <listitem><para>If D-Bus is used, make
                                 <listitem><para>If D-Bus is used, make
-                                your daemon bus-activatable, via
+                                your daemon bus-activatable by
                                 supplying a D-Bus service activation
                                 configuration file. This has multiple
                                 advantages: your daemon may be started
                                 supplying a D-Bus service activation
                                 configuration file. This has multiple
                                 advantages: your daemon may be started
                                 parallel to other daemons requiring it
                                 -- which maximizes parallelization and
                                 boot-up speed; your daemon can be
                                 parallel to other daemons requiring it
                                 -- which maximizes parallelization and
                                 boot-up speed; your daemon can be
-                                restarted on failure, without losing
+                                restarted on failure without losing
                                 any bus requests, as the bus queues
                                 any bus requests, as the bus queues
-                                requests for activatable
-                                services.</para></listitem>
+                                requests for activatable services. See
+                                below for details.</para></listitem>
 
                                 <listitem><para>If your daemon
                                 provides services to other local
 
                                 <listitem><para>If your daemon
                                 provides services to other local
                                 socket, it should be made
                                 socket-activatable following the
                                 scheme pointed out below. Like D-Bus
                                 socket, it should be made
                                 socket-activatable following the
                                 scheme pointed out below. Like D-Bus
-                                activation this enables on-demand
+                                activation, this enables on-demand
                                 starting of services as well as it
                                 starting of services as well as it
-                                allows improved parallization of
+                                allows improved parallelization of
                                 service start-up. Also, for state-less
                                 service start-up. Also, for state-less
-                                protocols (such as syslog, DNS) a
+                                protocols (such as syslog, DNS), a
                                 daemon implementing socket-based
                                 activation can be restarted without
                                 daemon implementing socket-based
                                 activation can be restarted without
-                                losing a single
-                                request.</para></listitem>
+                                losing a single request. See below for
+                                details.</para></listitem>
 
 
-                                <listitem><para>If applicable a daemon
+                                <listitem><para>If applicable, a daemon
                                 should notify the init system about
                                 should notify the init system about
-                                startup completion or status
-                                updates via the sd_notify()
+                                startup completion or status updates
+                                via the
+                                <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>
                                 interface.</para></listitem>
 
                                 <listitem><para>Instead of using the
                                 interface.</para></listitem>
 
                                 <listitem><para>Instead of using the
-                                syslog() call to log directly to the
-                                system logger, a new-style daemon may
-                                choose to simply log to STDERR via
-                                fprintf(), which is then forwarded to
+                                <function>syslog()</function> call to log directly to the
+                                system syslog service, a new-style daemon may
+                                choose to simply log to standard error via
+                                <function>fprintf()</function>, which is then forwarded to
                                 syslog by the init system. If log
                                 syslog by the init system. If log
-                                priorities are necessary these can be
+                                priorities are necessary, these can be
                                 encoded by prefixing individual log
                                 lines with strings like "&lt;4&gt;"
                                 (for log priority 4 "WARNING" in the
                                 syslog priority scheme), following a
                                 similar style as the Linux kernel's
                                 encoded by prefixing individual log
                                 lines with strings like "&lt;4&gt;"
                                 (for log priority 4 "WARNING" in the
                                 syslog priority scheme), following a
                                 similar style as the Linux kernel's
-                                printk() priority system. In fact, using
-                                this style of logging also enables the
-                                init system to optionally direct all
-                                application logging to the kernel log
-                                buffer (kmsg), as accessible via
-                                dmesg.</para></listitem>
+                                <function>printk()</function> priority system. In fact,
+                                using this style of logging also
+                                enables the init system to optionally
+                                direct all application logging to the
+                                kernel log buffer (kmsg), as
+                                accessible via
+                                <citerefentry><refentrytitle>dmesg</refentrytitle><manvolnum>1</manvolnum></citerefentry>. This
+                                kind of logging may be enabled by
+                                setting
+                                <varname>StandardError=syslog</varname>
+                                in the service unit file. For details,
+                                see
+                                <citerefentry><refentrytitle>sd-daemon</refentrytitle><manvolnum>3</manvolnum></citerefentry>
+                                and
+                                <citerefentry><refentrytitle>systemd.exec</refentrytitle><manvolnum>5</manvolnum></citerefentry>.</para></listitem>
 
                         </orderedlist>
 
                         </orderedlist>
+
+                        <para>These recommendations are similar but
+                        not identical to the <ulink
+                        url="http://developer.apple.com/mac/library/documentation/MacOSX/Conceptual/BPSystemStartup/Articles/LaunchOnDemandDaemons.html#//apple_ref/doc/uid/TP40001762-104738">Apple
+                        MacOS X Daemon Requirements</ulink>.</para>
+                </refsect2>
+
+        </refsect1>
+        <refsect1>
+                <title>Activation</title>
+
+                <para>New-style init systems provide multiple
+                additional mechanisms to activate services, as
+                detailed below. It is common that services are
+                configured to be activated via more than one mechanism
+                at the same time. An example for systemd:
+                <filename>bluetoothd.service</filename> might get
+                activated either when Bluetooth hardware is plugged
+                in, or when an application accesses its programming
+                interfaces via D-Bus. Or, a print server daemon might
+                get activated when traffic arrives at an IPP port, or
+                when a printer is plugged in, or when a file is queued
+                in the printer spool directory. Even for services that
+                are intended to be started on system bootup
+                unconditionally, it is a good idea to implement some of
+                the various activation schemes outlined below, in
+                order to maximize parallelization. If a daemon
+                implements a D-Bus service or listening socket,
+                implementing the full bus and socket activation scheme
+                allows starting of the daemon with its clients in
+                parallel (which speeds up boot-up), since all its
+                communication channels are established already, and no
+                request is lost because client requests will be queued
+                by the bus system (in case of D-Bus) or the kernel (in
+                case of sockets) until the activation is
+                completed.</para>
+
+                <refsect2>
+                        <title>Activation on Boot</title>
+
+                        <para>Old-style daemons are usually activated
+                        exclusively on boot (and manually by the
+                        administrator) via SysV init scripts, as
+                        detailed in the <ulink
+                        url="http://refspecs.linuxbase.org/LSB_3.1.1/LSB-Core-generic/LSB-Core-generic/iniscrptact.html">LSB
+                        Linux Standard Base Core
+                        Specification</ulink>. This method of
+                        activation is supported ubiquitously on Linux
+                        init systems, both old-style and new-style
+                        systems. Among other issues, SysV init scripts
+                        have the disadvantage of involving shell
+                        scripts in the boot process. New-style init
+                        systems generally employ updated versions of
+                        activation, both during boot-up and during
+                        runtime and using more minimal service
+                        description files.</para>
+
+                        <para>In systemd, if the developer or
+                        administrator wants to make sure that a service or
+                        other unit is activated automatically on boot,
+                        it is recommended to place a symlink to the
+                        unit file in the <filename>.wants/</filename>
+                        directory of either
+                        <filename>multi-user.target</filename> or
+                        <filename>graphical.target</filename>, which
+                        are normally used as boot targets at system
+                        startup. See
+                        <citerefentry><refentrytitle>systemd.unit</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                        for details about the
+                        <filename>.wants/</filename> directories, and
+                        <citerefentry><refentrytitle>systemd.special</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                        for details about the two boot targets.</para>
+
                 </refsect2>
 
                 <refsect2>
                 </refsect2>
 
                 <refsect2>
-                        <title>Bus Activation</title>
+                        <title>Socket-Based Activation</title>
+
+                        <para>In order to maximize the possible
+                        parallelization and robustness and simplify
+                        configuration and development, it is
+                        recommended for all new-style daemons that
+                        communicate via listening sockets to employ
+                        socket-based activation. In a socket-based
+                        activation scheme, the creation and binding of
+                        the listening socket as primary communication
+                        channel of daemons to local (and sometimes
+                        remote) clients is moved out of the daemon
+                        code and into the init system. Based on
+                        per-daemon configuration, the init system
+                        installs the sockets and then hands them off
+                        to the spawned process as soon as the
+                        respective daemon is to be started.
+                        Optionally, activation of the service can be
+                        delayed until the first inbound traffic
+                        arrives at the socket to implement on-demand
+                        activation of daemons. However, the primary
+                        advantage of this scheme is that all providers
+                        and all consumers of the sockets can be
+                        started in parallel as soon as all sockets
+                        are established. In addition to that, daemons
+                        can be restarted with losing only a minimal
+                        number of client transactions, or even any
+                        client request at all (the latter is
+                        particularly true for state-less protocols,
+                        such as DNS or syslog), because the socket
+                        stays bound and accessible during the restart,
+                        and all requests are queued while the daemon
+                        cannot process them.</para>
+
+                        <para>New-style daemons which support socket
+                        activation must be able to receive their
+                        sockets from the init system instead of
+                        creating and binding them themselves. For
+                        details about the programming interfaces for
+                        this scheme provided by systemd, see
+                        <citerefentry><refentrytitle>sd_listen_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>
+                        and
+                        <citerefentry><refentrytitle>sd-daemon</refentrytitle><manvolnum>3</manvolnum></citerefentry>. For
+                        details about porting existing daemons to
+                        socket-based activation, see below. With
+                        minimal effort, it is possible to implement
+                        socket-based activation in addition to
+                        traditional internal socket creation in the
+                        same codebase in order to support both
+                        new-style and old-style init systems from the
+                        same daemon binary.</para>
+
+                        <para>systemd implements socket-based
+                        activation via <filename>.socket</filename>
+                        units, which are described in
+                        <citerefentry><refentrytitle>systemd.socket</refentrytitle><manvolnum>5</manvolnum></citerefentry>. When
+                        configuring socket units for socket-based
+                        activation, it is essential that all listening
+                        sockets are pulled in by the special target
+                        unit <filename>sockets.target</filename>. It
+                        is recommended to place a
+                        <varname>WantedBy=sockets.target</varname>
+                        directive in the <literal>[Install]</literal>
+                        section to automatically add such a
+                        dependency on installation of a socket
+                        unit. Unless
+                        <varname>DefaultDependencies=no</varname> is
+                        set, the necessary ordering dependencies are
+                        implicitly created for all socket units. For
+                        more information about
+                        <filename>sockets.target</filename>, see
+                        <citerefentry><refentrytitle>systemd.special</refentrytitle><manvolnum>7</manvolnum></citerefentry>. It
+                        is not necessary or recommended to place any
+                        additional dependencies on socket units (for
+                        example from
+                        <filename>multi-user.target</filename> or
+                        suchlike) when one is installed in
+                        <filename>sockets.target</filename>.</para>
                 </refsect2>
 
                 <refsect2>
                 </refsect2>
 
                 <refsect2>
-                        <title>Socket Activation</title>
+                        <title>Bus-Based Activation</title>
+
+                        <para>When the D-Bus IPC system is used for
+                        communication with clients, new-style daemons
+                        should employ bus activation so that they are
+                        automatically activated when a client
+                        application accesses their IPC
+                        interfaces. This is configured in D-Bus
+                        service files (not to be confused with systemd
+                        service unit files!). To ensure that D-Bus
+                        uses systemd to start-up and maintain the
+                        daemon, use the
+                        <varname>SystemdService=</varname> directive
+                        in these service files to configure the
+                        matching systemd service for a D-Bus
+                        service. e.g.: For a D-Bus service whose D-Bus
+                        activation file is named
+                        <filename>org.freedesktop.RealtimeKit.service</filename>,
+                        make sure to set
+                        <varname>SystemdService=rtkit-daemon.service</varname>
+                        in that file to bind it to the systemd
+                        service
+                        <filename>rtkit-daemon.service</filename>. This
+                        is needed to make sure that the daemon is
+                        started in a race-free fashion when activated
+                        via multiple mechanisms simultaneously.</para>
                 </refsect2>
 
                 <refsect2>
                 </refsect2>
 
                 <refsect2>
-                        <title>Writing Service Files</title>
+                        <title>Device-Based Activation</title>
+
+                        <para>Often, daemons that manage a particular
+                        type of hardware should be activated only when
+                        the hardware of the respective kind is plugged
+                        in or otherwise becomes available. In a
+                        new-style init system, it is possible to bind
+                        activation to hardware plug/unplug events. In
+                        systemd, kernel devices appearing in the
+                        sysfs/udev device tree can be exposed as units
+                        if they are tagged with the string
+                        <literal>systemd</literal>. Like any other
+                        kind of unit, they may then pull in other units
+                        when activated (i.e. plugged in) and thus
+                        implement device-based activation. systemd
+                        dependencies may be encoded in the udev
+                        database via the
+                        <varname>SYSTEMD_WANTS=</varname>
+                        property. See
+                        <citerefentry><refentrytitle>systemd.device</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                        for details. Often, it is nicer to pull in
+                        services from devices only indirectly via
+                        dedicated targets. Example: Instead of pulling
+                        in <filename>bluetoothd.service</filename>
+                        from all the various bluetooth dongles and
+                        other hardware available, pull in
+                        bluetooth.target from them and
+                        <filename>bluetoothd.service</filename> from
+                        that target. This provides for nicer
+                        abstraction and gives administrators the
+                        option to enable
+                        <filename>bluetoothd.service</filename> via
+                        controlling a
+                        <filename>bluetooth.target.wants/</filename>
+                        symlink uniformly with a command like
+                        <command>enable</command> of
+                        <citerefentry><refentrytitle>systemctl</refentrytitle><manvolnum>1</manvolnum></citerefentry>
+                        instead of manipulating the udev
+                        ruleset.</para>
                 </refsect2>
 
                 <refsect2>
                 </refsect2>
 
                 <refsect2>
-                        <title>Installing Service Files</title>
+                        <title>Path-Based Activation</title>
+
+                        <para>Often, runtime of daemons processing
+                        spool files or directories (such as a printing
+                        system) can be delayed until these file system
+                        objects change state, or become
+                        non-empty. New-style init systems provide a
+                        way to bind service activation to file system
+                        changes. systemd implements this scheme via
+                        path-based activation configured in
+                        <filename>.path</filename> units, as outlined
+                        in
+                        <citerefentry><refentrytitle>systemd.path</refentrytitle><manvolnum>5</manvolnum></citerefentry>.</para>
+                </refsect2>
+
+                <refsect2>
+                        <title>Timer-Based Activation</title>
+
+                        <para>Some daemons that implement clean-up
+                        jobs that are intended to be executed in
+                        regular intervals benefit from timer-based
+                        activation. In systemd, this is implemented
+                        via <filename>.timer</filename> units, as
+                        described in
+                        <citerefentry><refentrytitle>systemd.timer</refentrytitle><manvolnum>5</manvolnum></citerefentry>.</para>
+                </refsect2>
+
+                <refsect2>
+                        <title>Other Forms of Activation</title>
+
+                        <para>Other forms of activation have been
+                        suggested and implemented in some
+                        systems. However, there are often simpler or
+                        better alternatives, or they can be put
+                        together of combinations of the schemes
+                        above. Example: Sometimes, it appears useful to
+                        start daemons or <filename>.socket</filename>
+                        units when a specific IP address is configured
+                        on a network interface, because network
+                        sockets shall be bound to the
+                        address. However, an alternative to implement
+                        this is by utilizing the Linux IP_FREEBIND
+                        socket option, as accessible via
+                        <varname>FreeBind=yes</varname> in systemd
+                        socket files (see
+                        <citerefentry><refentrytitle>systemd.socket</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                        for details). This option, when enabled,
+                        allows sockets to be bound to a non-local, not
+                        configured IP address, and hence allows
+                        bindings to a particular IP address before it
+                        actually becomes available, making such an
+                        explicit dependency to the configured address
+                        redundant. Another often suggested trigger for
+                        service activation is low system
+                        load. However, here too, a more convincing
+                        approach might be to make proper use of
+                        features of the operating system, in
+                        particular, the CPU or IO scheduler of
+                        Linux. Instead of scheduling jobs from
+                        userspace based on monitoring the OS
+                        scheduler, it is advisable to leave the
+                        scheduling of processes to the OS scheduler
+                        itself. systemd provides fine-grained access
+                        to the CPU and IO schedulers. If a process
+                        executed by the init system shall not
+                        negatively impact the amount of CPU or IO
+                        bandwidth available to other processes, it
+                        should be configured with
+                        <varname>CPUSchedulingPolicy=idle</varname>
+                        and/or
+                        <varname>IOSchedulingClass=idle</varname>. Optionally,
+                        this may be combined with timer-based
+                        activation to schedule background jobs during
+                        runtime and with minimal impact on the system,
+                        and remove it from the boot phase
+                        itself.</para>
                 </refsect2>
 
         </refsect1>
                 </refsect2>
 
         </refsect1>
+        <refsect1>
+                <title>Integration with Systemd</title>
 
 
+                <refsect2>
+                        <title>Writing Systemd Unit Files</title>
+
+                        <para>When writing systemd unit files, it is
+                        recommended to consider the following
+                        suggestions:</para>
+
+                        <orderedlist>
+                                <listitem><para>If possible, do not use
+                                the <varname>Type=forking</varname>
+                                setting in service files. But if you
+                                do, make sure to set the PID file path
+                                using <varname>PIDFile=</varname>. See
+                                <citerefentry><refentrytitle>systemd.service</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                                for details.</para></listitem>
+
+                                <listitem><para>If your daemon
+                                registers a D-Bus name on the bus,
+                                make sure to use
+                                <varname>Type=dbus</varname> in the
+                                service file if
+                                possible.</para></listitem>
+
+                                <listitem><para>Make sure to set a
+                                good human-readable description string
+                                with
+                                <varname>Description=</varname>.</para></listitem>
+
+                                <listitem><para>Do not disable
+                                <varname>DefaultDependencies=</varname>,
+                                unless you really know what you do and
+                                your unit is involved in early boot or
+                                late system shutdown.</para></listitem>
+
+                                <listitem><para>Normally, little if
+                                any dependencies should need to
+                                be defined explicitly. However, if you
+                                do configure explicit dependencies, only refer to
+                                unit names listed on
+                                <citerefentry><refentrytitle>systemd.special</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                                or names introduced by your own
+                                package to keep the unit file
+                                operating
+                                system-independent.</para></listitem>
+
+                                <listitem><para>Make sure to include
+                                an <literal>[Install]</literal>
+                                section including installation
+                                information for the unit file. See
+                                <citerefentry><refentrytitle>systemd.unit</refentrytitle><manvolnum>5</manvolnum></citerefentry>
+                                for details. To activate your service
+                                on boot, make sure to add a
+                                <varname>WantedBy=multi-user.target</varname>
+                                or
+                                <varname>WantedBy=graphical.target</varname>
+                                directive. To activate your socket on
+                                boot, make sure to add
+                                <varname>WantedBy=sockets.target</varname>. Usually,
+                                you also want to make sure that when
+                                your service is installed, your socket
+                                is installed too, hence add
+                                <varname>Also=foo.socket</varname> in
+                                your service file
+                                <filename>foo.service</filename>, for
+                                a hypothetical program
+                                <filename>foo</filename>.</para></listitem>
+
+                        </orderedlist>
+                </refsect2>
+
+                <refsect2>
+                        <title>Installing Systemd Service Files</title>
+
+                        <para>At the build installation time
+                        (e.g. <command>make install</command> during
+                        package build), packages are recommended to
+                        install their systemd unit files in the
+                        directory returned by <command>pkg-config
+                        systemd
+                        --variable=systemdsystemunitdir</command> (for
+                        system services) or <command>pkg-config
+                        systemd
+                        --variable=systemduserunitdir</command>
+                        (for user services). This will make the
+                        services available in the system on explicit
+                        request but not activate them automatically
+                        during boot. Optionally, during package
+                        installation (e.g. <command>rpm -i</command>
+                        by the administrator), symlinks should be
+                        created in the systemd configuration
+                        directories via the <command>enable</command>
+                        command of the
+                        <citerefentry><refentrytitle>systemctl</refentrytitle><manvolnum>1</manvolnum></citerefentry>
+                        tool to activate them automatically on
+                        boot.</para>
+
+                        <para>Packages using
+                        <citerefentry><refentrytitle>autoconf</refentrytitle><manvolnum>1</manvolnum></citerefentry>
+                        are recommended to use a configure script
+                        excerpt like the following to determine the
+                        unit installation path during source
+                        configuration:</para>
+
+                        <programlisting>PKG_PROG_PKG_CONFIG
+AC_ARG_WITH([systemdsystemunitdir],
+     [AS_HELP_STRING([--with-systemdsystemunitdir=DIR], [Directory for systemd service files])],,
+     [with_systemdsystemunitdir=auto])
+AS_IF([test "x$with_systemdsystemunitdir" = "xyes" -o "x$with_systemdsystemunitdir" = "xauto"], [
+     def_systemdsystemunitdir=$($PKG_CONFIG --variable=systemdsystemunitdir systemd)
+
+     AS_IF([test "x$def_systemdsystemunitdir" = "x"],
+         [AS_IF([test "x$with_systemdsystemunitdir" = "xyes"],
+                [AC_MSG_ERROR([systemd support requested but pkg-config unable to query systemd package])])
+          with_systemdsystemunitdir=no],
+         [with_systemdsystemunitdir="$def_systemdsystemunitdir"])])
+AS_IF([test "x$with_systemdsystemunitdir" != "xno"],
+      [AC_SUBST([systemdsystemunitdir], [$with_systemdsystemunitdir])])
+AM_CONDITIONAL([HAVE_SYSTEMD], [test "x$with_systemdsystemunitdir" != "xno"])</programlisting>
+
+                        <para>This snippet allows automatic
+                        installation of the unit files on systemd
+                        machines, and optionally allows their
+                        installation even on machines lacking
+                        systemd. (Modification of this snippet for the
+                        user unit directory is left as an exercise for the
+                        reader.)</para>
+
+                        <para>Additionally, to ensure that
+                        <command>make distcheck</command> continues to
+                        work, it is recommended to add the following
+                        to the top-level <filename>Makefile.am</filename>
+                        file in
+                        <citerefentry><refentrytitle>automake</refentrytitle><manvolnum>1</manvolnum></citerefentry>-based
+                        projects:</para>
+
+                        <programlisting>DISTCHECK_CONFIGURE_FLAGS = \
+        --with-systemdsystemunitdir=$$dc_install_base/$(systemdsystemunitdir)</programlisting>
+
+                        <para>Finally, unit files should be installed in the system with an automake excerpt like the following:</para>
+
+                        <programlisting>if HAVE_SYSTEMD
+systemdsystemunit_DATA = \
+        foobar.socket \
+        foobar.service
+endif</programlisting>
+
+                        <para>In the
+                        <citerefentry><refentrytitle>rpm</refentrytitle><manvolnum>8</manvolnum></citerefentry>
+                        <filename>.spec</filename> file, use snippets
+                        like the following to enable/disable the
+                        service during
+                        installation/deinstallation. This makes use of
+                        the RPM macros shipped along systemd. Consult
+                        the packaging guidelines of your distribution
+                        for details and the equivalent for other
+                        package managers.</para>
+
+                        <para>At the top of the file:</para>
+
+                        <programlisting>BuildRequires: systemd
+%{?systemd_requires}</programlisting>
+
+                        <para>And as scriptlets, further down:</para>
+
+                        <programlisting>%post
+%systemd_post foobar.service foobar.socket
+
+%preun
+%systemd_preun foobar.service foobar.socket
+
+%postun
+%systemd_postun</programlisting>
+
+                        <para>If the service shall be restarted during
+                        upgrades, replace the
+                        <literal>%postun</literal> scriptlet above
+                        with the following:</para>
+
+                        <programlisting>%postun
+%systemd_postun_with_restart foobar.service</programlisting>
+
+                        <para>Note that
+                        <literal>%systemd_post</literal> and
+                        <literal>%systemd_preun</literal> expect the
+                        names of all units that are installed/removed
+                        as arguments, separated by
+                        spaces. <literal>%systemd_postun</literal>
+                        expects no
+                        arguments. <literal>%systemd_postun_with_restart</literal>
+                        expects the units to restart as
+                        arguments.</para>
+
+                        <para>To facilitate upgrades from a package
+                        version that shipped only SysV init scripts to
+                        a package version that ships both a SysV init
+                        script and a native systemd service file, use
+                        a fragment like the following:</para>
+
+                        <programlisting>%triggerun -- foobar &lt; 0.47.11-1
+if /sbin/chkconfig --level 5 foobar ; then
+        /bin/systemctl --no-reload enable foobar.service foobar.socket >/dev/null 2>&amp;1 || :
+fi</programlisting>
+
+                        <para>Where 0.47.11-1 is the first package
+                        version that includes the native unit
+                        file. This fragment will ensure that the first
+                        time the unit file is installed, it will be
+                        enabled if and only if the SysV init script is
+                        enabled, thus making sure that the enable
+                        status is not changed. Note that
+                        <command>chkconfig</command> is a command
+                        specific to Fedora which can be used to check
+                        whether a SysV init script is enabled. Other
+                        operating systems will have to use different
+                        commands here.</para>
+                </refsect2>
+        </refsect1>
+
+        <refsect1>
+                <title>Porting Existing Daemons</title>
+
+                <para>Since new-style init systems such as systemd are
+                compatible with traditional SysV init systems, it is
+                not strictly necessary to port existing daemons to the
+                new style. However, doing so offers additional
+                functionality to the daemons as well as simplifying
+                integration into new-style init systems.</para>
+
+                <para>To port an existing SysV compatible daemon, the
+                following steps are recommended:</para>
+
+                <orderedlist>
+                        <listitem><para>If not already implemented,
+                        add an optional command line switch to the
+                        daemon to disable daemonization. This is
+                        useful not only for using the daemon in
+                        new-style init systems, but also to ease
+                        debugging.</para></listitem>
+
+                        <listitem><para>If the daemon offers
+                        interfaces to other software running on the
+                        local system via local <constant>AF_UNIX</constant> sockets,
+                        consider implementing socket-based activation
+                        (see above). Usually, a minimal patch is
+                        sufficient to implement this: Extend the
+                        socket creation in the daemon code so that
+                        <citerefentry><refentrytitle>sd_listen_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>
+                        is checked for already passed sockets
+                        first. If sockets are passed (i.e. when
+                        <function>sd_listen_fds()</function> returns a
+                        positive value), skip the socket creation step
+                        and use the passed sockets. Secondly, ensure
+                        that the file system socket nodes for local
+                        <constant>AF_UNIX</constant> sockets used in the socket-based
+                        activation are not removed when the daemon
+                        shuts down, if sockets have been
+                        passed. Third, if the daemon normally closes
+                        all remaining open file descriptors as part of
+                        its initialization, the sockets passed from
+                        the init system must be spared. Since
+                        new-style init systems guarantee that no
+                        left-over file descriptors are passed to
+                        executed processes, it might be a good choice
+                        to simply skip the closing of all remaining
+                        open file descriptors if sockets are
+                        passed.</para></listitem>
+
+                        <listitem><para>Write and install a systemd
+                        unit file for the service (and the sockets if
+                        socket-based activation is used, as well as a
+                        path unit file, if the daemon processes a
+                        spool directory), see above for
+                        details.</para></listitem>
+
+                        <listitem><para>If the daemon exposes
+                        interfaces via D-Bus, write and install a
+                        D-Bus activation file for the service, see
+                        above for details.</para></listitem>
+                </orderedlist>
+        </refsect1>
 
         <refsect1>
                 <title>See Also</title>
                 <para>
                         <citerefentry><refentrytitle>systemd</refentrytitle><manvolnum>1</manvolnum></citerefentry>,
 
         <refsect1>
                 <title>See Also</title>
                 <para>
                         <citerefentry><refentrytitle>systemd</refentrytitle><manvolnum>1</manvolnum></citerefentry>,
+                        <citerefentry><refentrytitle>sd-daemon</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+                        <citerefentry><refentrytitle>sd_listen_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+                        <citerefentry><refentrytitle>sd_notify</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
                         <citerefentry><refentrytitle>daemon</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
                         <citerefentry><refentrytitle>daemon</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
-                        <citerefentry><refentrytitle>sd_listen_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>
+                        <citerefentry><refentrytitle>systemd.service</refentrytitle><manvolnum>5</manvolnum></citerefentry>
                 </para>
         </refsect1>
 
                 </para>
         </refsect1>