chiark / gitweb /
socket: suffix newly added TCP sockopt time properties with "Sec"
[elogind.git] / man / systemd.socket.xml
index f65704d67b68be3dd6d269cbe649ec47d5e43108..238029a2516dc651556cf7396815f5b9e9562109 100644 (file)
                                 replaced with a
                                 <constant>NUL</constant> character
                                 before binding. For details, see
-                                <citerefentry><refentrytitle>unix</refentrytitle><manvolnum>7</manvolnum></citerefentry>.</para>
+                                <citerefentry project='man-pages'><refentrytitle>unix</refentrytitle><manvolnum>7</manvolnum></citerefentry>.</para>
 
                                 <para>If the address string is a
                                 single number, it is read as port
                         </varlistentry>
 
                         <varlistentry>
-                                <term><varname>DirectoryMode=</varname></term>
-                                <listitem><para>If listening on a file
-                                system socket or FIFO, the parent
-                                directories are automatically created
-                                if needed. This option specifies the
-                                file system access mode used when
-                                creating these directories. Takes an
-                                access mode in octal
-                                notation. Defaults to
-                                0755.</para></listitem>
+                                <term><varname>SocketUser=</varname></term>
+                                <term><varname>SocketGroup=</varname></term>
+
+                                <listitem><para>Takes a UNIX
+                                user/group name. When specified,
+                                all AF_UNIX sockets and FIFO nodes in
+                                the file system are owned by the
+                                specified user and group. If unset
+                                (the default), the nodes are owned by
+                                the root user/group (if run in system
+                                context) or the invoking user/group
+                                (if run in user context). If only a
+                                user is specified but no group, then
+                                the group is derived from the user's
+                                default group.</para></listitem>
                         </varlistentry>
 
                         <varlistentry>
                                 0666.</para></listitem>
                         </varlistentry>
 
+                        <varlistentry>
+                                <term><varname>DirectoryMode=</varname></term>
+                                <listitem><para>If listening on a file
+                                system socket or FIFO, the parent
+                                directories are automatically created
+                                if needed. This option specifies the
+                                file system access mode used when
+                                creating these directories. Takes an
+                                access mode in octal
+                                notation. Defaults to
+                                0755.</para></listitem>
+                        </varlistentry>
+
                         <varlistentry>
                                 <term><varname>Accept=</varname></term>
                                 <listitem><para>Takes a boolean
                                 <option>false</option>.</para></listitem>
                         </varlistentry>
 
+                        <varlistentry>
+                                <term><varname>KeepAliveTimeSec=</varname></term>
+                                <listitem><para>Takes time (in seconds) as argument . The connection needs to remain
+                                idle before TCP starts sending keepalive probes. This controls the TCP_KEEPIDLE
+                                socket option (see
+                                <citerefentry><refentrytitle>socket</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                                and the <ulink
+                                url="http://www.tldp.org/HOWTO/html_single/TCP-Keepalive-HOWTO/">TCP
+                                Keepalive HOWTO</ulink> for details.)
+                                Defaults value is 7200 seconds (2 hours).</para></listitem>
+                        </varlistentry>
+
+                        <varlistentry>
+                                <term><varname>KeepAliveIntervalSec=</varname></term>
+                                <listitem><para>Takes time (in seconds) as argument between individual keepalive probes,
+                                if the socket option SO_KEEPALIVE has  been set on this socket seconds as argument.
+                                This controls the TCP_KEEPINTVL socket option (see
+                                <citerefentry><refentrytitle>socket</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                                and the <ulink
+                                url="http://www.tldp.org/HOWTO/html_single/TCP-Keepalive-HOWTO/">TCP
+                                Keepalive HOWTO</ulink> for details.)
+                                Defaults value is 75 seconds.</para></listitem>
+                        </varlistentry>
+
+                        <varlistentry>
+                                <term><varname>KeepAliveProbes=</varname></term>
+                                <listitem><para>Takes interger as argument. It's the number of unacknowledged probes to
+                                send before considering the connection dead and notifying the application layer.
+                                This controls the TCP_KEEPCNT socket option (see
+                                <citerefentry><refentrytitle>socket</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                                and the <ulink
+                                url="http://www.tldp.org/HOWTO/html_single/TCP-Keepalive-HOWTO/">TCP
+                                Keepalive HOWTO</ulink> for details.)
+                                Defaults value is 9.</para></listitem>
+                        </varlistentry>
+
+                        <varlistentry>
+                                <term><varname>NoDelay=</varname></term>
+                                <listitem><para>Takes a boolean
+                                argument. TCP Nagle's algorithm works by combining a number of
+                                small outgoing messages, and sending them all at once.
+                                This controls the TCP_NODELAY socket option (see
+                                <citerefentry><refentrytitle>tcp</refentrytitle><manvolnum>7</manvolnum></citerefentry>
+                                Defaults to
+                                <option>false</option>.</para></listitem>
+                        </varlistentry>
+
                         <varlistentry>
                                 <term><varname>Priority=</varname></term>
                                 <listitem><para>Takes an integer
                                 for details.).</para></listitem>
                         </varlistentry>
 
+                        <varlistentry>
+                                <term><varname>DeferAcceptSec=</varname></term>
+
+                                <listitem><para>Takes time (in
+                                seconds) as argument. If set, the
+                                listening process will be awakened
+                                only when data arrives on the socket,
+                                and not immediately when connection is
+                                established. When this option is set,
+                                the
+                                <constant>TCP_DEFER_ACCEPT</constant>
+                                socket option will be used (see
+                                <citerefentry><refentrytitle>tcp</refentrytitle><manvolnum>7</manvolnum></citerefentry>),
+                                and the kernel will ignore initial ACK
+                                packets without any data. The argument
+                                specifies the approximate amount of
+                                time the kernel should wait for
+                                incoming data before falling back to
+                                the normal behaviour of honouring
+                                empty ACK packets. This option is
+                                beneficial for protocols where the
+                                client sends the data first (e.g.
+                                HTTP, in contrast to SMTP), because
+                                the server process will not be woken
+                                up unnecessarily before it can take
+                                any action.
+                                </para>
+
+                                <para>If the client also uses the
+                                <constant>TCP_DEFER_ACCEPT</constant>
+                                option, the latency of the initial
+                                connection may be reduced, because the
+                                kernel will send data in the final
+                                packet establishing the connection
+                                (the third packet in the "three-way
+                                handshake").</para>
+
+                                <para>Disabled by default.</para>
+                                </listitem>
+                        </varlistentry>
+
                         <varlistentry>
                                 <term><varname>ReceiveBuffer=</varname></term>
                                 <term><varname>SendBuffer=</varname></term>
                                 finish. If a command does not exit
                                 within the configured time, the socket
                                 will be considered failed and be shut
-                                down again. All commands still running,
+                                down again. All commands still running
                                 will be terminated forcibly via
                                 <constant>SIGTERM</constant>, and after another delay of
                                 this time with <constant>SIGKILL</constant>. (See
                                 <option>KillMode=</option> in <citerefentry><refentrytitle>systemd.kill</refentrytitle><manvolnum>5</manvolnum></citerefentry>.)
                                 Takes a unit-less value in seconds, or
                                 a time span value such as "5min
-                                20s". Pass 0 to disable the timeout
-                                logic. Defaults to <varname>TimeoutStartSec=</varname> from the
-                                manager configuration file.</para></listitem>
+                                20s". Pass <literal>0</literal> to disable the timeout
+                                logic. Defaults to <varname>DefaultTimeoutStartSec=</varname> from the
+                                manager configuration file
+                                (see <citerefentry><refentrytitle>systemd-systemd.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>).
+                                </para></listitem>
                         </varlistentry>
 
                         <varlistentry>
                         <varlistentry>
                                 <term><varname>RemoveOnStop=</varname></term>
                                 <listitem><para>Takes a boolean
-                                argument. If enabled any file nodes
+                                argument. If enabled, any file nodes
                                 created by this socket unit are
                                 removed when it is stopped. This
                                 applies to AF_UNIX sockets in the file
-                                system, POSIX message queues as well
-                                as FIFOs, as well as any symlinks to
+                                system, POSIX message queues, FIFOs,
+                                as well as any symlinks to
                                 them configured with
-                                <varname>Symlinks=</varname>. Normally
+                                <varname>Symlinks=</varname>. Normally,
                                 it should not be necessary to use this
                                 option, and is not recommended as
                                 services might continue to run after
                                 system paths. The specified paths will
                                 be created as symlinks to the AF_UNIX
                                 socket path or FIFO path of this
-                                socket unit.  If this setting is used
+                                socket unit.  If this setting is used,
                                 only one AF_UNIX socket in the file
                                 system or one FIFO may be configured
                                 for the socket unit. Use this option