blob: db95d2f75b52e1170cd69d4e204dc283723b4b4d [file] [log] [blame]
<?xml version='1.0'?> <!--*-nxml-*-->
<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.5//EN"
<!-- SPDX-License-Identifier: LGPL-2.1-or-later -->
<refentry id="daemon">
<refpurpose>Writing and packaging system daemons</refpurpose>
<para>A daemon is a service process that runs in the background
and supervises the system or provides 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
This manual page covers both schemes, and in particular includes
recommendations for daemons that shall be included in the systemd
init system.</para>
<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 steps are unnecessary for new-style daemons (see below),
and should only be implemented if compatibility with SysV is
<listitem><para>Close all open 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 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
<function>getrlimit()</function> for
<constant>RLIMIT_NOFILE</constant>. </para></listitem>
<listitem><para>Reset all signal handlers to their default.
This is best done by iterating through the available signals
up to the limit of <constant>_NSIG</constant> and resetting
them to <constant>SIG_DFL</constant>.</para></listitem>
<listitem><para>Reset the signal mask
<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
<function>setsid()</function> to detach from any terminal and
create an independent session.</para></listitem>
<listitem><para>In the child, call <function>fork()</function> again, to ensure that the daemon can
never re-acquire a terminal again. (This relevant if the program — and all its dependencies — does
not carefully specify `O_NOCTTY` on each and every single `open()` call that might potentially open a
TTY device node.)</para></listitem>
<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 daemon process is
re-parented to init/PID 1, as all daemons should
<listitem><para>In the daemon process, connect
<filename>/dev/null</filename> to standard input, output, and
<listitem><para>In the daemon process, reset the umask to 0,
so that the file modes passed to <function>open()</function>,
<function>mkdir()</function> and suchlike directly control the
access mode of the created files and
<listitem><para>In the daemon process, change the current
directory to the root directory (/), in order to avoid that
the daemon involuntarily blocks mount points from being
<listitem><para>In the daemon process, write the daemon PID
(as returned by <function>getpid()</function>) to a PID file,
for example <filename index='false'>/run/</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
<listitem><para>In the daemon process, drop privileges, if
possible and applicable.</para></listitem>
<listitem><para>From the daemon 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 before the first
<function>fork()</function> and hence available in both the
original and the daemon process.</para></listitem>
<listitem><para>Call <function>exit()</function> in the
original process. The process that invoked the daemon must be
able to rely on that this <function>exit()</function> happens
after initialization is complete and all external
communication channels are established and
<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 behavior optional and
configurable via a command line argument to ease debugging as
well as to simplify integration into systems using
<title>New-Style Daemons</title>
<para>Modern services for Linux should be implemented as
new-style daemons. This makes it easier to supervise and control
them at runtime and simplifies their implementation.</para>
<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 the init system, it is recommended not to
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
connected to <filename>/dev/null</filename> and standard output/error connected to the
logging service, unless otherwise configured. The umask is reset.
<para>It is recommended for new-style daemons to implement the
<listitem><para>If <constant>SIGTERM</constant> is received,
shut down the daemon and exit cleanly.</para></listitem>
<listitem><para>If <constant>SIGHUP</constant> is received,
reload the configuration files, if this
<listitem><para>Provide a correct exit code from the main
daemon process, as 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 the <ulink
recommendations for SysV init
<listitem><para>If possible and 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>
<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
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.
for the available controls.</para></listitem>
<listitem><para>If D-Bus is used, make your daemon
bus-activatable by supplying a D-Bus service activation
configuration file. This has multiple advantages: your daemon
may be started lazily on-demand; it may be started in parallel
to other daemons requiring it — which maximizes
parallelization and boot-up speed; your daemon can be
restarted on failure without losing any bus requests, as the
bus queues requests for activatable services. See below for
<listitem><para>If your daemon provides services to other
local processes or remote clients via a socket, it should be
made socket-activatable following the scheme pointed out
below. Like D-Bus activation, this enables on-demand starting
of services as well as it allows improved parallelization of
service start-up. Also, for state-less protocols (such as
syslog, DNS), a daemon implementing socket-based activation
can be restarted without losing a single request. See below
for details.</para></listitem>
<listitem><para>If applicable, a daemon should notify the init
system about startup completion or status updates via the
<listitem><para>Instead of using the
<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
levels are necessary, these can be encoded by prefixing
individual log lines with strings like
<literal>&lt;4&gt;</literal> (for log level 4 "WARNING" in the
syslog priority scheme), following a similar style as the
Linux kernel's <function>printk()</function> level system. For
details, see
<listitem><para>As new-style daemons are invoked without a controlling TTY (but as their own session
leaders) care should be taken to always specify `O_NOCTTY` on `open()` calls that possibly reference
a TTY device node, so that no controlling TTY is accidentally acquired.</para></listitem>
<para>These recommendations are similar but not identical to the
MacOS X Daemon Requirements</ulink>.</para>
<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>
<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
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></filename> or
<filename></filename>, which are normally used
as boot targets at system startup. See
for details about the <filename>.wants/</filename> directories,
for details about the two boot targets.</para>
<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
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
When configuring socket units for socket-based activation, it is
essential that all listening sockets are pulled in by the
special target unit <filename></filename>. It is
recommended to place a
<varname></varname> directive in the
[Install] 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></filename>, see
It is not necessary or recommended to place any additional
dependencies on socket units (for example from
<filename></filename> or suchlike) when one is
installed in <filename></filename>.</para>
<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>
<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
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 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></filename>
symlink uniformly with a command like <command>enable</command>
instead of manipulating the udev ruleset.</para>
<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
<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
<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
<constant>IP_FREEBIND</constant>/<constant>IPV6_FREEBIND</constant> 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 I/O 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 I/O schedulers. If a process executed by the init system shall not negatively impact the
amount of CPU or I/O 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>
<title>Integration with systemd</title>
<title>Writing systemd Unit Files</title>
<para>When writing systemd unit files, it is recommended to
consider the following suggestions:</para>
<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
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
<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
or names introduced by your own package to keep the unit file
operating system-independent.</para></listitem>
<listitem><para>Make sure to include an
[Install] section including installation
information for the unit file. See
for details. To activate your service on boot, make sure to
add a <varname></varname> or
<varname></varname> directive. To
activate your socket on boot, make sure to add
<varname></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
<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
tool to activate them automatically on boot.</para>
<para>Packages using
<citerefentry project='die-net'><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
[AS_HELP_STRING([--with-systemdsystemunitdir=DIR], [Directory for systemd service files])],,
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])])
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
<para>Additionally, to ensure that
<command>make distcheck</command> continues to
work, it is recommended to add the following
to the top-level <filename></filename>
file in
<citerefentry project='die-net'><refentrytitle>automake</refentrytitle><manvolnum>1</manvolnum></citerefentry>-based
<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 \
<para>In the
<citerefentry project='die-net'><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
<para>At the top of the file:</para>
<programlisting>BuildRequires: systemd
<para>And as scriptlets, further down:</para>
%systemd_post foobar.service foobar.socket
%systemd_preun foobar.service foobar.socket
<para>If the service shall be restarted during upgrades, replace
the <literal>%postun</literal> scriptlet above with the
%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 || :
<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>
<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>
<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
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>
<title>Placing Daemon Data</title>
<para>It is recommended to follow the general guidelines for
placing package files, as discussed in
<title>See Also</title>