chrony/doc/chrony.conf.adoc

// This file is part of chrony
//
// Copyright (C) Richard P. Curnow  1997-2003
// Copyright (C) Miroslav Lichvar  2009-2016
//
// This program is free software; you can redistribute it and/or modify
// it under the terms of version 2 of the GNU General Public License as
// published by the Free Software Foundation.
//
// This program 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.
//
// You should have received a copy of the GNU General Public License along
// with this program; if not, write to the Free Software Foundation, Inc.,
// 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.

= chrony.conf(5)
:doctype: manpage
:man manual: Configuration Files
:man source: chrony @CHRONY_VERSION@

== NAME
chrony.conf - chronyd configuration file

== SYNOPSIS
*chrony.conf*

== DESCRIPTION

This file configures the *chronyd* daemon. The compiled-in location is
_@SYSCONFDIR@/chrony.conf_, but other locations can be specified on the
*chronyd* command line with the *-f* option.

Each directive in the configuration file is placed on a separate line. The
following sections describe each of the directives in turn. The directives can
occur in any order in the file and they are not case-sensitive.

The configuration directives can also be specified directly on the *chronyd*
command line. In this case each argument is parsed as a new line and the
configuration file is ignored.

While the number of supported directives is large, only a few of them are
typically needed. See the <<examples,*EXAMPLES*>> section for configuration in
typical operating scenarios.

The configuration file might contain comment lines. A comment line is any line
that starts with zero or more spaces followed by any one of the following
characters: *!*, *;*, *#*, *%*. Any line with this format will be ignored.

== DIRECTIVES

=== Time sources

[[server]]*server* _hostname_ [_option_]...::
The *server* directive specifies an NTP server which can be used as a time
source. The client-server relationship is strictly hierarchical: a client might
synchronise its system time to that of the server, but the server's system time
will never be influenced by that of a client.
+
The *server* directive is immediately followed by either the name of the
server, or its IP address. The *server* directive supports the following
options:
+
*minpoll* _poll_:::
Although *chronyd* will trim the rate at which it samples the server during
normal operation, the user might want to constrain the minimum polling interval.
This is always defined as a power of 2, so *minpoll 5* would mean that the
polling interval cannot drop below 32 seconds. The default is 6 (64 seconds).
*maxpoll* _poll_:::
In a similar way, the user might want to constrain the maximum polling interval.
Again this is specified as a power of 2, *maxpoll 9* indicates that the polling
interval must stay at or below 512 seconds. The default is 10 (1024 seconds).
*iburst*:::
If this option is set, the interval between the first four polls will be 2
seconds instead of _minpoll_. This is useful to quickly get the first update of
the clock after *chronyd* is started.
*key* _id_:::
The NTP protocol supports the inclusion of checksums in the packets, to prevent
computers having their system time upset by rogue packets being sent to them.
The checksums are generated as a function of a password, using the
cryptographic hash function set in the key file, which is specified by the
<<keyfile,*keyfile*>> directive.
+
If the key option is present, *chronyd* will attempt to use authenticated
packets when communicating with this server. The key number used will be the
single argument to the key option (an unsigned integer in the range 1 through
2^32-1). The server must have the same password for this key number configured,
otherwise no relationship between the computers will be possible.
*maxdelay* _delay_:::
*chronyd* uses the network round-trip delay to the server to determine how
accurate a particular measurement is likely to be. Long round-trip delays
indicate that the request, or the response, or both were delayed. If only one
of the messages was delayed the measurement error is likely to be substantial.
+
For small variations in the round-trip delay, *chronyd* uses a weighting scheme
when processing the measurements. However, beyond a certain level of delay the
measurements are likely to be so corrupted as to be useless. (This is
particularly so on dial-up or other slow links, where a long delay probably
indicates a highly asymmetric delay caused by the response waiting behind a lot
of packets related to a download of some sort).
+
If the user knows that round trip delays above a certain level should cause the
measurement to be ignored, this level can be defined with the *maxdelay*
option. For example, *maxdelay 0.3* would indicate that measurements with a
round-trip delay of 0.3 seconds or more should be ignored. The default value is
3 seconds.
*maxdelayratio* _ratio_:::
This option is similar to the maxdelay option above. *chronyd* keeps a record
of the minimum round-trip delay amongst the previous measurements that it has
buffered. If a measurement has a round trip delay that is greater than the
maxdelayratio times the minimum delay, it will be rejected.
*maxdelaydevratio* _ratio_:::
If a measurement has a ratio of the increase in the round-trip delay from the
minimum delay amongst the previous measurements to the standard deviation of
the previous measurements that is greater than the specified ratio, it will be
rejected. The default is 10.0.
*minsamples* _samples_:::
Set the minimum number of samples kept for this source. This overrides the
<<minsamples,*minsamples*>> directive.
*maxsamples* _samples_:::
Set the maximum number of samples kept for this source. This overrides the
<<maxsamples,*maxsamples*>> directive.
*offline*:::
If the server will not be reachable when *chronyd* is started, the *offline*
option can be specified. *chronyd* will not try to poll the server until it is
enabled to do so (by using the <<chronyc.adoc#online,*online*>> command in
*chronyc*).
*auto_offline*:::
If this option is set, the server will be assumed to have gone offline when 2
requests have been sent to it without receiving a response. This option avoids
the need to run the <<chronyc.adoc#offline,*offline*>> command from *chronyc*
when disconnecting the network link. (It will still be necessary to use the
<<chronyc.adoc#online,*online*>> command when the link has been established, to
enable measurements to start.)
*prefer*:::
Prefer this source over sources without prefer option.
*noselect*:::
Never select this source. This is particularly useful for monitoring.
*trust*:::
Assume time from this source is always true. It can be rejected as a
falseticker in the source selection only if another source with this option
does not agree with it.
*require*:::
Require that at least one of the sources specified with this option is
selectable (i.e. recently reachable and not a falseticker) before updating the
clock. Together with the *trust* option this might be useful to allow a trusted
authenticated source to be safely combined with unauthenticated sources in
order to improve the accuracy of the clock. They can be selected and used for
synchronisation only if they agree with the trusted and required source.
*polltarget* _target_:::
Target number of measurements to use for the regression algorithm which
*chronyd* will try to maintain by adjusting the polling interval between
*minpoll* and *maxpoll*. A higher target makes *chronyd* prefer shorter polling
intervals. The default is 6 and a useful range is from 6 to 60.
*port* _port_:::
This option allows the UDP port on which the server understands NTP requests to
be specified. For normal servers this option should not be required (the
default is 123, the standard NTP port).
*presend* _poll_:::
If the timing measurements being made by *chronyd* are the only network data
passing between two computers, you might find that some measurements are badly
skewed due to either the client or the server having to do an ARP lookup on the
other party prior to transmitting a packet. This is more of a problem with long
sampling intervals, which might be similar in duration to the lifetime of entries
in the ARP caches of the machines.
+
In order to avoid this problem, the *presend* option can be used. It takes a
single integer argument, which is the smallest polling interval for which an
extra pair of NTP packets will be exchanged between the client and the server
prior to the actual measurement. For example, with the following option
included in a *server* directive:
+
----
presend 9
----
+
when the polling interval is 512 seconds or more, an extra NTP client packet
will be sent to the server a short time (4 seconds) before making the actual
measurement.
*minstratum* _stratum_:::
When the synchronisation source is selected from available sources, sources
with lower stratum are normally slightly preferred. This option can be used to
increase stratum of the source to the specified minimum, so *chronyd* will
avoid selecting that source. This is useful with low stratum sources that are
known to be unreliable or inaccurate and which should be used only when other
sources are unreachable.
*version* _version_:::
This option sets the NTP version number used in packets sent to the server.
This can be useful when the server runs an old NTP implementation that does not
respond to newer versions. The default version number is 4.

[[pool]]*pool* _name_ [_option_]...::
The syntax of this directive is similar to that for the <<server,*server*>>
directive, except that it is used to specify a pool of NTP servers rather than
a single NTP server. The pool name is expected to resolve to multiple addresses
which might change over time.
+
All options valid in the <<server,*server*>> directive can be used in this
directive too. There is one option specific to the *pool* directive:
*maxsources* sets the maximum number of sources that can be used from the pool,
the default value is 4.
+
On start, when the pool name is resolved, *chronyd* will add up to 16 sources,
one for each resolved address. When the number of sources from which at least
one valid reply was received reaches the number specified by the *maxsources*
option, the other sources will be removed. When a pool source is unreachable,
marked as a falseticker, or has a distance larger than the limit set by the
<<maxdistance,*maxdistance*>> directive, *chronyd* will try to replace the
source with a newly resolved address from the pool.
+
An example of the *pool* directive is
+
----
pool pool.ntp.org iburst maxsources 3
----

[[peer]]*peer* _hostname_ [_option_]...::
The syntax of this directive is identical to that for the <<server,*server*>>
directive, except that it is used to specify an NTP peer rather than an NTP
server.
+
When a key is specified by the *key* option to enable authentication, both
peers must be configured to use the same key and the same key number.
+
Please note that NTP peers that are not configured with a key to enable
authentication are vulnerable to a denial-of-service attack. An attacker
knowing that NTP hosts A and B are peering with each other can send a packet
with random timestamps to host A with source address of B which will set the
NTP state variables on A to the values sent by the attacker. Host A will then
send on its next poll to B a packet with an origin timestamp that does not match
the transmit timestamp of B and the packet will be dropped. If the attacker
does this periodically for both hosts, they will not be able to synchronise to
each other.
+
This attack can be prevented by enabling authentication with the *key* option,
or by using the <<server,*server*>> directive on both sides to specify the other
host as a server instead of a peer. The disadvantage of the *server* directive
is that it will double the network traffic between the two hosts.

[[initstepslew]]*initstepslew* _step-threshold_ [_hostname_]...::
In normal operation, *chronyd* slews the time when it needs to adjust the
system clock. For example, to correct a system clock which is 1 second slow,
*chronyd* slightly increases the amount by which the system clock is advanced
on each clock interrupt, until the error is removed. Note that at no time does
time run backwards with this method.
+
On most Unix systems it is not desirable to step the system clock, because many
programs rely on time advancing monotonically forwards.
+
When the *chronyd* daemon is initially started, it is possible that the system
clock is considerably in error. Attempting to correct such an error by slewing
might not be sensible, since it might take several hours to correct the error by
this means.
+
The purpose of the *initstepslew* directive is to allow *chronyd* to make a
rapid measurement of the system clock error at boot time, and to correct the
system clock by stepping before normal operation begins. Since this would
normally be performed only at an appropriate point in the system boot sequence,
no other software should be adversely affected by the step.
+
If the correction required is less than a specified threshold, a slew is used
instead. This makes it safer to restart *chronyd* whilst the system is in
normal operation.
+
The *initstepslew* directive takes a threshold and a list of NTP servers as
arguments. Each of the servers is rapidly polled several times, and a majority
voting mechanism used to find the most likely range of system clock error that
is present. A step or slew is applied to the system clock to correct this
error. *chronyd* then enters its normal operating mode.
+
An example of the use of the directive is:
+
----
initstepslew 30 foo.example.net bar.example.net
----
+
where 2 NTP servers are used to make the measurement. The _30_ indicates that
if the system's error is found to be 30 seconds or less, a slew will be used to
correct it; if the error is above 30 seconds, a step will be used.
+
The *initstepslew* directive can also be used in an isolated LAN environment,
where the clocks are set manually. The most stable computer is chosen as the
master, and the other computers are slaved to it. If each of the slaves is
configured with the <<local,*local*>> directive, the master can be set up with
an *initstepslew* directive which references some or all of the slaves. Then,
if the master machine has to be rebooted, the slaves can be relied on to act
analogously to a flywheel and preserve the time for a short period while the
master completes its reboot.
+
The *initstepslew* directive is functionally similar to a combination of the
<<makestep,*makestep*>> and <<server,*server*>> directives with the *iburst*
option. The main difference is that the *initstepslew* servers are used only
before normal operation begins and that the foreground *chronyd* process waits
for *initstepslew* to finish before exiting. This is useful to prevent programs
started in the boot sequence after *chronyd* from reading the clock before it
has been stepped.

[[refclock]]*refclock* _driver_ _parameter_ [_option_]...::
The *refclock* directive specifies a hardware reference clock to be used as a
time source. It has two mandatory parameters, a driver name and a
driver-specific parameter.
+
There are currently four drivers included:
+
*PPS*:::
Driver for the kernel PPS (pulse per second) API. The parameter is the path to
the PPS device (typically _/dev/pps?_). The assert events are used for
synchronisation by default. String *:clear* can be appended to the path to use
the clear events instead.
+
As PPS refclocks don't supply full time, *chronyd* needs to be configured with
another time source (NTP or non-PPS refclock) in order to complete samples from
the PPS refclock. An alternative is to enable the <<local,*local*>> directive
to allow synchronisation with some unknown but constant offset.
For example:
+
----
refclock PPS /dev/pps0 lock NMEA
refclock SHM 0 offset 0.5 delay 0.2 refid NMEA noselect
----
+
*SHM*:::
NTP shared memory driver. This driver uses a shared memory segment to receive
samples from another process. The parameter is the number of the shared memory
segment, typically 0, 1, 2 or 3. For example:
+
----
refclock SHM 1 poll 3 refid GPS1
----
+
A driver option in form of *:perm=NNN* can be appended to the segment number to
create the segment with permissions other than the default 0600.
+
Examples of applications that can be used as SHM refclocks are
http://www.catb.org/gpsd/[*gpsd*],
http://www.buzzard.me.uk/jonathan/radioclock.html[*radioclk*], and
https://www.vanheusden.com/time/omnisync/[*omnisync*].
+
*SOCK*:::
Unix domain socket driver. It is similar to the SHM driver, but samples are
received from a Unix domain socket instead of shared memory and the messages
have a different format. The parameter is the path to the socket, which
*chronyd* creates on start. An advantage over the SHM driver is that SOCK does
not require polling and it can receive PPS samples with incomplete time. The
format of the messages is described in the _refclock_sock.c_ file in the chrony
source code.
+
An application which supports the SOCK protocol is the *gpsd* daemon. The path
where *gpsd* expects the socket to be created is described in the *gpsd(8)* man
page. For example:
+
----
refclock SOCK /var/run/chrony.ttyS0.sock
----
+
*PHC*:::
PTP hardware clock (PHC) driver. The parameter is the path to the device of
the PTP clock, which for example can be synchronised by *ptp4l* from
http://linuxptp.sourceforge.net[*linuxptp*]. PTP clocks are typically kept in
TAI instead of UTC, so the *offset* option should be used to compensate for the
current UTC-TAI offset. For example:
+
----
refclock PHC /dev/ptp0 poll 3 dpoll -2 offset -36
----
+
::
The *refclock* directive also supports a number of options:
+
*poll* _poll_:::
Timestamps produced by refclock drivers are not used immediately, but they are
stored and processed by a median filter in the polling interval specified by
this option. This is defined as a power of 2 and can be negative to specify a
sub-second interval. The default is 4 (16 seconds). A shorter interval allows
*chronyd* to react faster to changes in the frequency of the system clock, but
it might have a negative effect on its accuracy if the samples have a lot of
jitter.
*dpoll* _dpoll_:::
Some drivers do not listen for external events and try to produce samples in
their own polling interval. This is defined as a power of 2 and can be negative
to specify a sub-second interval. The default is 0 (1 second).
*refid* _refid_:::
This option is used to specify the reference ID of the refclock, as up to four
ASCII characters. The default reference ID is composed from the first three
characters of the driver name and the number of the refclock. Each refclock
must have a unique reference ID.
*lock* _refid_:::
This option can be used to lock a PPS refclock to another refclock, which is
specified by its reference ID. In this mode received PPS samples are paired
directly with raw samples from the specified refclock.
*rate* _rate_:::
This option sets the rate of the pulses in the PPS signal (in Hz). This option
controls how the pulses will be completed with real time. To actually receive
more than one pulse per second, a negative *dpoll* has to be specified (-3 for
a 5Hz signal). The default is 1.
*offset* _offset_:::
This option can be used to compensate for a constant error. The specified
offset (in seconds) is applied to all samples produced by the reference clock.
The default is 0.0.
*delay* _delay_:::
This option sets the NTP delay of the source (in seconds). Half of this value
is included in the maximum assumed error which is used in the source selection
algorithm. Increasing the delay is useful to avoid having no majority in the
source selection or to make it prefer other sources. The default is 1e-9 (1
nanosecond).
*precision* _precision_:::
This option sets the refclock precision (in seconds). The default is 1e-6 (1
microsecond) for SHM refclock, and 1e-9 (1 nanosecond) for SOCK, PPS and PHC
refclocks.
*maxdispersion* _dispersion_:::
Maximum allowed dispersion for filtered samples (in seconds). Samples with
larger estimated dispersion are ignored. By default, this limit is disabled.
*filter* _samples_:::
This option sets the length of the median filter which is used to reduce the
noise in the measurements. With each poll about 40 percent of the stored
samples are discarded and one final sample is calculated as an average of the
remaining samples. If the length is 4 or more, at least 4 samples have to be
collected between polls. For lengths below 4, the filter has to be full. The
default is 64.
*prefer*:::
Prefer this source over sources without the prefer option.
*noselect*:::
Never select this source. This is useful for monitoring or with sources which
are not very accurate, but are locked with a PPS refclock.
*trust*:::
Assume time from this source is always true. It can be rejected as a
falseticker in the source selection only if another source with this option
does not agree with it.
*require*:::
Require that at least one of the sources specified with this option is
selectable (i.e. recently reachable and not a falseticker) before updating the
clock. Together with the *trust* option this can be useful to allow a trusted,
but not very precise, reference clock to be safely combined with
unauthenticated NTP sources in order to improve the accuracy of the clock. They
can be selected and used for synchronisation only if they agree with the
trusted and required source.
*minsamples* _samples_:::
Set the minimum number of samples kept for this source. This overrides the
<<minsamples,*minsamples*>> directive.
*maxsamples* _samples_:::
Set the maximum number of samples kept for this source. This overrides the
<<maxsamples,*maxsamples*>> directive.

[[manual]]*manual*::
The *manual* directive enables support at run-time for the
<<chronyc.adoc#settime,*settime*>> command in *chronyc*. If no *manual*
directive is included, any attempt to use the *settime* command in *chronyc*
will be met with an error message.
+
Note that the *settime* command can be enabled at run-time using
the <<chronyc.adoc#manual,*manual*>> command in *chronyc*. (The idea of the two
commands is that the *manual* command controls the manual clock driver's
behaviour, whereas the *settime* command allows samples of manually entered
time to be provided.)

[[acquisitionport]]*acquisitionport* _port_::
By default, *chronyd* uses a separate client socket for each configured server
and their source port is chosen arbitrarily by the operating system. However,
you can use the *acquisitionport* directive to explicitly specify a port and
use only one socket (per IPv4 or IPv6 address family) for all configured servers.
This can be useful for getting through some firewalls. If set to 0, the source
port of the socket will be chosen arbitrarily.
+
It can be set to the same port as is used by the NTP server (which can be
configured with the <<port,*port*>> directive) to use only one socket for all
NTP packets.
+
An example of the *acquisitionport* directive is:
+
----
acquisitionport 1123
----
+
This would change the source port used for client requests to UDP port 1123.
You could then persuade the firewall administrator to open that port.

[[bindacqaddress]]*bindacqaddress* _address_::
The *bindacqaddress* directive sets the network interface to which
*chronyd* will bind its NTP client sockets. The syntax is similar to the
<<bindaddress,*bindaddress*>> and <<bindcmdaddress,*bindcmdaddress*>>
directives.
+
For each of the IPv4 and IPv6 protocols, only one *bindacqaddress* directive
can be specified.

[[dumpdir]]*dumpdir* _directory_::
To compute the rate of gain or loss of time, *chronyd* has to store a
measurement history for each of the time sources it uses.
+
Certain systems (Linux, FreeBSD, NetBSD, Solaris) have operating system support
for setting the rate of gain or loss to compensate for known errors. (On Mac OS
X, *chronyd* must simulate such a capability by periodically slewing the system
clock forwards or backwards by a suitable amount to compensate for the error
built up since the previous slew.)
+
For such systems, it is possible to save the measurement history across
restarts of *chronyd* (assuming no changes are made to the system clock
behaviour whilst it is not running). If this capability is to be used (via the
*dumponexit* directive in the configuration file, or the
<<chronyc.adoc#dump,*dump*>> command in *chronyc*), the *dumpdir* directive
should be used to define the directory where the measurement histories are
saved.
+
An example of the directive is:
+
----
dumpdir @CHRONYVARDIR@
----
+
A source whose reference ID (the IP address for IPv4 sources) is _1.2.3.4_
would have its measurement history saved in the file
_/var/lib/chrony/1.2.3.4.dat_.

[[dumponexit]]*dumponexit*::
If this directive is present, it indicates that *chronyd* should save the
measurement history for each of its time sources recorded whenever the program
exits. (See the <<dumpdir,*dumpdir*>> directive above.)

[[maxsamples]]*maxsamples* _samples_::
The *maxsamples* directive sets the default maximum number of samples that
*chronyd* should keep for each source. This setting can be overridden for
individual sources in the <<server,*server*>> and <<refclock,*refclock*>>
directives. The default value is 0, which disables the configurable limit. The
useful range is 4 to 64.

[[minsamples]]*minsamples* _samples_::
The *minsamples* directive sets the default minimum number of samples that
*chronyd* should keep for each source. This setting can be overridden for
individual sources in the <<server,*server*>> and <<refclock,*refclock*>>
directives. The default value is 0. The useful range is 4 to 64.

=== Source selection

[[combinelimit]]*combinelimit* _limit_::
When *chronyd* has multiple sources available for synchronisation, it has to
select one source as the synchronisation source. The measured offsets and
frequencies of the system clock relative to the other sources, however, can be
combined with the selected source to improve the accuracy of the system clock.
+
The *combinelimit* directive limits which sources are included in the combining
algorithm. Their synchronisation distance has to be shorter than the distance
of the selected source multiplied by the value of the limit. Also, their
measured frequencies have to be close to the frequency of the selected source.
+
By default, the limit is 3. Setting the limit to 0 effectively disables the
source combining algorithm and only the selected source will be used to control
the system clock.

[[maxdistance]]*maxdistance* _distance_::
The *maxdistance* directive sets the maximum allowed root distance of the
sources to not be rejected by the source selection algorithm. The distance
includes the accumulated dispersion, which might be large when the source is no
longer synchronised, and half of the total round-trip delay to the primary
source.
+
By default, the maximum root distance is 3 seconds.
+
Setting *maxdistance* to a larger value can be useful to allow synchronisation
with a server that only has a very infrequent connection to its sources and can
accumulate a large dispersion between updates of its clock.

[[minsources]]*minsources* _sources_::
The *minsources* directive sets the minimum number of sources that need to be
considered as selectable in the source selection algorithm before the local
clock is updated. The default value is 1.
+
Setting this option to a larger number can be used to improve the reliability.
More sources will have to agree with each other and the clock will not be
updated when only one source (which could be serving incorrect time) is
reachable.

[[reselectdist]]*reselectdist* _distance_::
When *chronyd* selects a synchronisation source from available sources, it
will prefer the one with the shortest synchronisation distance. However, to
avoid frequent reselecting when there are sources with similar distance, a
fixed distance is added to the distance for sources that are currently not
selected. This can be set with the *reselectdist* directive. By default, the
distance is 100 microseconds.

[[stratumweight]]*stratumweight* _distance_::
The *stratumweight* directive sets how much distance should be added per
stratum to the synchronisation distance when *chronyd* selects the
synchronisation source from available sources.
+
By default, the weight is 0.001 seconds. This means that the stratum of the sources
in the selection process matters only when the differences between the
distances are in milliseconds.

=== System clock

[[corrtimeratio]]*corrtimeratio* _ratio_::
When *chronyd* is slewing the system clock to correct an offset, the rate at
which it is slewing adds to the frequency error of the clock. On Linux,
FreeBSD, NetBSD and Solaris this rate can be controlled.
+
The *corrtimeratio* directive sets the ratio between the duration in which the
clock is slewed for an average correction according to the source history and
the interval in which the corrections are done (usually the NTP polling
interval). Corrections larger than the average take less time and smaller
corrections take more time, the amount of the correction and the correction
time are inversely proportional.
+
Increasing *corrtimeratio* improves the overall frequency error of the system
clock, but increases the overall time error as the corrections take longer.
+
By default, the ratio is set to 3, the time accuracy of the clock is preferred
over its frequency accuracy.
+
The maximum allowed slew rate can be set by the <<maxslewrate,*maxslewrate*>>
directive. The current remaining correction is shown in the
<<chronyc.adoc#tracking,*tracking*>> report as the *System time* value.

[[driftfile]]*driftfile* _file_::
One of the main activities of the *chronyd* program is to work out the rate at
which the system clock gains or loses time relative to real time.
+
Whenever *chronyd* computes a new value of the gain or loss rate, it is desirable
to record it somewhere. This allows *chronyd* to begin compensating the system
clock at that rate whenever it is restarted, even before it has had a chance to
obtain an equally good estimate of the rate during the new run. (This process
can take many minutes, at least.)
+
The *driftfile* directive allows a file to be specified into which *chronyd*
can store the rate information. Two parameters are recorded in the file. The
first is the rate at which the system clock gains or loses time, expressed in
parts per million, with gains positive. Therefore, a value of 100.0 indicates
that when the system clock has advanced by a second, it has gained 100
microseconds in reality (so the true time has only advanced by 999900
microseconds). The second is an estimate of the error bound around the first
value in which the true rate actually lies.
+
An example of the driftfile directive is:
+
----
driftfile @CHRONYVARDIR@/drift
----

[[fallbackdrift]]*fallbackdrift* _min-interval_ _max-interval_::
Fallback drifts are long-term averages of the system clock drift calculated
over exponentially increasing intervals. They are used when the clock is no
longer synchronised to avoid quickly drifting away from true time if there was
a short-term deviation in the drift before the synchronisation was lost.
+
The directive specifies the minimum and maximum interval since the last clock
update to switch between fallback drifts. They are defined as a power of 2 (in
seconds). The syntax is as follows:
+
----
fallbackdrift 16 19
----
+
In this example, the minimum interval is 16 (18 hours) and the maximum interval is
19 (6 days). The system clock frequency will be set to the first fallback 18
hours after last clock update, to the second after 36 hours, etc. This might be
a good setting to cover daily and weekly temperature fluctuations.
+
By default (or if the specified maximum or minimum is 0), no fallbacks are used
and the clock frequency changes only with new measurements from NTP sources,
reference clocks, or manual input.

[[leapsecmode]]*leapsecmode* _mode_::
A leap second is an adjustment that is occasionally applied to UTC to keep it
close to the mean solar time. When a leap second is inserted, the last day of
June or December has an extra second 23:59:60.
+
For computer clocks that is a problem. The Unix time is defined as number of
seconds since 00:00:00 UTC on 1 January 1970 without leap seconds. The system
clock cannot have time 23:59:60, every minute has 60 seconds and every day has
86400 seconds by definition. The inserted leap second is skipped and the clock
is suddenly ahead of UTC by one second. The *leapsecmode* directive selects how
that error is corrected. There are four options:
+
*system*:::
When inserting a leap second, the kernel steps the system clock backwards by
one second when the clock gets to 00:00:00 UTC. When deleting a leap second, it
steps forward by one second when the clock gets to 23:59:59 UTC. This is the
default mode when the system driver supports leap seconds (i.e. on Linux,
FreeBSD, NetBSD and Solaris).
*step*:::
This is similar to the *system* mode, except the clock is stepped by
*chronyd* instead of the kernel. It can be useful to avoid bugs in the kernel
code that would be executed in the *system* mode. This is the default mode
when the system driver does not support leap seconds.
*slew*:::
The clock is corrected by slewing started at 00:00:00 UTC when a leap second
is inserted or 23:59:59 UTC when a leap second is deleted. This might be
preferred over the *system* and *step* modes when applications running on the
system are sensitive to jumps in the system time and it is acceptable that the
clock will be off for a longer time. On Linux with the default
<<maxslewrate,*maxslewrate*>> value the correction takes 12 seconds.
*ignore*:::
No correction is applied to the clock for the leap second. The clock will be
corrected later in normal operation when new measurements are made and the
estimated offset includes the one second error.
::
+
When serving time to NTP clients that cannot be configured to correct their
clocks for a leap second by slewing, or to clients that would correct at
slightly different rates when it is necessary to keep them close together, the
*slew* mode can be combined with the <<smoothtime,*smoothtime*>> directive to
enable a server leap smear.
+
When smearing a leap second, the leap status is suppressed on the server and
the served time is corrected slowly be slewing instead of stepping. The clients
do not need any special configuration as they do not know there is any leap
second and they follow the server time which eventually brings them back to
UTC. Care must be taken to ensure they use only NTP servers which smear the
leap second in exactly the same way for synchronisation.
+
This feature must be used carefully, because the server is intentionally not
serving its best estimate of the true time.
+
A recommended configuration to enable a server leap smear is:
+
----
leapsecmode slew
maxslewrate 1000
smoothtime 400 0.001 leaponly
----
+
The first directive is necessary to disable the clock step which would reset
the smoothing process. The second directive limits the slewing rate of the
local clock to 1000 ppm, which improves the stability of the smoothing process
when the local correction starts and ends. The third directive enables the
server time smoothing process. It will start when the clock gets to 00:00:00
UTC and it will take 17 hours 34 minutes to finish. The frequency offset will
be changing by 0.001 ppm per second and will reach a maximum of 31.623 ppm. The
*leaponly* option makes the duration of the leap smear constant and allows the
clients to safely synchronise with multiple identically configured leap
smearing servers.

[[leapsectz]]*leapsectz* _timezone_::
This directive is used to set the name of the timezone in the system tz
database which *chronyd* can use to find out when will the next leap second
occur. It will periodically check if the times 23:59:59 and 23:59:60 are valid
on Jun 30 and Dec 31 in the timezone. This typically works with the *right/UTC*
timezone.
+
This directive is mainly useful with reference clocks which do not provide
leap second information. It is not necessary to restart *chronyd* if the tz
database is updated with a new leap second at least 12 hours before the event.
+
An example of the directive is:
+
----
leapsectz right/UTC
----
+
The following shell command verifies that the timezone contains leap seconds
and can be used with this directive:
+
----
$ TZ=right/UTC date -d 'Dec 31 2008 23:59:60'
Wed Dec 31 23:59:60 UTC 2008
----

[[makestep]]*makestep* _threshold_ _limit_::
Normally *chronyd* will cause the system to gradually correct any time offset,
by slowing down or speeding up the clock as required. In certain situations,
the system clock might be so far adrift that this slewing process would take a
very long time to correct the system clock.
+
This directive forces *chronyd* to step the system clock if the adjustment is
larger than a threshold value, but only if there were no more clock updates
since *chronyd* was started than a specified limit (a negative value can be
used to disable the limit).
+
This is particularly useful when using reference clocks, because the
<<initstepslew,*initstepslew*>> directive works only with NTP sources.
+
An example of the use of this directive is:
+
----
makestep 0.1 3
----
+
This would step the system clock if the adjustment is larger than 0.1 seconds, but
only in the first three clock updates.

[[maxchange]]*maxchange* _offset_ _start_ _ignore_::
This directive sets the maximum allowed offset corrected on a clock update. The
check is performed only after the specified number of updates to allow a large
initial adjustment of the system clock. When an offset larger than the
specified maximum occurs, it will be ignored for the specified number of times
and then *chronyd* will give up and exit (a negative value can be used to never
exit). In both cases a message is sent to syslog.
+
An example of the use of this directive is:
+
----
maxchange 1000 1 2
----
+
After the first clock update, *chronyd* will check the offset on every clock
update, it will ignore two adjustments larger than 1000 seconds and exit on
another one.

[[maxclockerror]]*maxclockerror* _error-in-ppm_::
The *maxclockerror* directive sets the maximum assumed frequency error that the
system clock can gain on its own between clock updates. It describes the
stability of the clock.
+
By default, the maximum error is 1 ppm.
+
Typical values for _error-in-ppm_ might be 10 for a low quality clock and 0.1
for a high quality clock using a temperature compensated crystal oscillator.

[[maxdrift]]*maxdrift* _drift-in-ppm_::
This directive specifies the maximum assumed drift (frequency error) of the
system clock. It limits the frequency adjustment that *chronyd* is allowed to
use to correct the measured drift. It is an additional limit to the maximum
adjustment that can be set by the system driver (100000 ppm on Linux, 500 ppm
on FreeBSD and NetBSD, 32500 ppm on Solaris).
+
By default, the maximum assumed drift is 500000 ppm, i.e. the adjustment is
limited by the system driver rather than this directive.

[[maxupdateskew]]*maxupdateskew* _skew-in-ppm_::
One of *chronyd*'s tasks is to work out how fast or slow the computer's clock
runs relative to its reference sources. In addition, it computes an estimate of
the error bounds around the estimated value.
+
If the range of error is too large, it probably indicates that the measurements
have not settled down yet, and that the estimated gain or loss rate is not very
reliable.
+
The *maxupdateskew* directive sets the threshold for determining whether an
estimate might be so unreliable that it should not be used. By default, the
threshold is 1000 ppm.
+
Typical values for _skew-in-ppm_ might be 100 for a dial-up connection to
servers over a phone line, and 5 or 10 for a computer on a LAN.
+
It should be noted that this is not the only means of protection against using
unreliable estimates. At all times, *chronyd* keeps track of both the estimated
gain or loss rate, and the error bound on the estimate. When a new estimate is
generated following another measurement from one of the sources, a weighted
combination algorithm is used to update the master estimate. So if *chronyd*
has an existing highly-reliable master estimate and a new estimate is generated
which has large error bounds, the existing master estimate will dominate in the
new master estimate.

[[maxslewrate]]*maxslewrate* _rate-in-ppm_::
The *maxslewrate* directive sets the maximum rate at which *chronyd* is allowed
to slew the time. It limits the slew rate controlled by the correction time
ratio (which can be set by the <<corrtimeratio,*corrtimeratio*>> directive) and
is effective only on systems where *chronyd* is able to control the rate (i.e.
Linux, FreeBSD, NetBSD, Solaris).
+
For each system there is a maximum frequency offset of the clock that
can be set by the driver. On Linux it is 100000 ppm, on FreeBSD and NetBSD
it is 5000 ppm, and on Solaris it is 32500 ppm. Also, due to a kernel
limitation, setting *maxslewrate* on FreeBSD and NetBSD to a value between 500
ppm and 5000 ppm will effectively set it to 500 ppm.
+
By default, the maximum slew rate is set to 83333.333 ppm (one twelfth).

[[tempcomp]]
*tempcomp* _file_ _interval_ _T0_ _k0_ _k1_ _k2_::
*tempcomp* _file_ _interval_ _points-file_::
Normally, changes in the rate of drift of the system clock are caused mainly by
changes in the temperature of the crystal oscillator on the motherboard.
+
If there are temperature measurements available from a sensor close to the
oscillator, the *tempcomp* directive can be used to compensate for the changes
in the temperature and improve the stability and accuracy of the clock.
+
The result depends on many factors, including the resolution of the sensor, the
amount of noise in the measurements, the polling interval of the time source,
the compensation update interval, how well the compensation is specified, and
how close the sensor is to the oscillator. When it is working well, the
frequency reported in the _tracking.log_ file is more stable and the maximum
reached offset is smaller.
+
There are two forms of the directive. The first one has six parameters: a path
to the file containing the current temperature from the sensor (in text
format), the compensation update interval (in seconds), and temperature
coefficients _T0_, _k0_, _k1_, _k2_.
+
The frequency compensation is calculated (in ppm) as
+
----
k0 + (T - T0) * k1 + (T - T0)^2 * k2
----
+
The result has to be between -10 ppm and 10 ppm, otherwise the measurement is
considered invalid and will be ignored. The _k0_ coefficient can be adjusted to
keep the compensation in that range.
+
An example of the use is:
+
----
tempcomp /sys/class/hwmon/hwmon0/temp2_input 30 26000 0.0 0.000183 0.0
----
+
The measured temperature will be read from the file in the Linux sysfs
filesystem every 30 seconds. When the temperature is 26000 (26 degrees
Celsius), the frequency correction will be zero. When it is 27000 (27 degrees
Celsius), the clock will be set to run faster by 0.183 ppm, etc.
+
The second form has three parameters: the path to the sensor file, the update
interval, and a path to a file containing a list of (temperature, compensation)
points, from which the compensation is linearly interpolated or extrapolated.
+
An example is:
+
----
tempcomp /sys/class/hwmon/hwmon0/temp2_input 30 /etc/chrony.tempcomp
----
+
where the _/etc/chrony.tempcomp_ file could have
+
----
20000 1.0
21000 0.64
22000 0.36
23000 0.16
24000 0.04
25000 0.0
26000 0.04
27000 0.16
28000 0.36
29000 0.64
30000 1.0
----
+
Valid measurements with corresponding compensations are logged to the
_tempcomp.log_ file if enabled by the <<log,*log tempcomp*>> directive.

=== NTP server

[[allow]]*allow* [*all*] [_subnet_]::
The *allow* directive is used to designate a particular subnet from which NTP
clients are allowed to access the computer as an NTP server.
+
The default is that no clients are allowed access, i.e. *chronyd* operates
purely as an NTP client. If the *allow* directive is used, *chronyd* will be
both a client of its servers, and a server to other clients.
+
Examples of the use of the directive are as follows:
+
----
allow foo.example.net
allow 1.2
allow 3.4.5
allow 6.7.8/22
allow 6.7.8.9/22
allow 2001:db8::/32
allow 0/0
allow ::/0
allow
----
+
The first directive allows the named node to be an NTP client of this computer.
The second directive allows any node with an IPv4 address of the form _1.2.x.y_
(with _x_ and _y_ arbitrary) to be an NTP client of this computer. Likewise,
the third directive allows any node with an IPv4 address of the form _3.4.5.x_
to have client NTP access. The fourth and fifth forms allow access from any
node with an IPv4 address of the form _6.7.8.x_, _6.7.9.x_, _6.7.10.x_ or
_6.7.11.x_ (with _x_ arbitrary), i.e. the value 22 is the number of bits
defining the specified subnet. In the fifth form, the final byte is ignored.
The sixth form is used for IPv6 addresses. The seventh and eighth forms allow
access by any IPv4 and IPv6 node respectively. The ninth forms allows access by
any node (IPv4 or IPv6).
+
A second form of the directive, *allow all*, has a greater effect, depending on
the ordering of directives in the configuration file. To illustrate the effect,
consider the two examples:
+
----
allow 1.2.3.4
deny 1.2.3
allow 1.2
----
+
and
+
----
allow 1.2.3.4
deny 1.2.3
allow all 1.2
----
+
In the first example, the effect is the same regardless of what order the three
directives are given in. So the _1.2.x.y_ subnet is allowed access, except for
the _1.2.3.x_ subnet, which is denied access, however the host _1.2.3.4_ is
allowed access.
+
In the second example, the *allow all 1.2* directives overrides the effect of
_any_ previous directive relating to a subnet within the specified subnet.
Within a configuration file this capability is probably rather moot; however,
it is of greater use for reconfiguration at run-time via *chronyc* with the
<<chronyc.adoc#allow,*allow all*>> command.
+
Note, if the <<initstepslew,*initstepslew*>> directive is used in the
configuration file, each of the computers listed in that directive must allow
client access by this computer for it to work.

[[deny]]*deny* [*all*] [_subnet_]::
This is similar to the <<allow,*allow*>> directive, except that it denies NTP
client access to a particular subnet or host, rather than allowing it.
+
The syntax is identical.
+
There is also a *deny all* directive with similar behaviour to the *allow all*
directive.

[[bindaddress]]*bindaddress* _address_::
The *bindaddress* directive allows you to restrict the network interface to
which *chronyd* will listen for NTP requests. This provides an additional level
of access restriction above that available through the <<deny,*deny*>>
mechanism.
+
Suppose you have a local network with addresses in the _192.168.1.0_
subnet together with an Internet connection. The network interface's IP
address is _192.168.1.1_. Suppose you want to block all access through the
Internet connection. You could add the line:
+
----
bindaddress 192.168.1.1
----
+
to the configuration file.
+
For each of the IPv4 and IPv6 protocols, only one *bindaddress* directive can be
specified. Therefore, it is not useful on computers which should serve NTP on
multiple network interfaces.

[[broadcast]]*broadcast* _interval_ _address_ [_port_]::
The *broadcast* directive is used to declare a broadcast address to which
chronyd should send packets in the NTP broadcast mode (i.e. make *chronyd* act
as a broadcast server). Broadcast clients on that subnet will be able to
synchronise.
+
The syntax is as follows:
+
----
broadcast 30 192.168.1.255
broadcast 60 192.168.2.255 12123
broadcast 60 ff02::101
----
+
In the first example, the destination port defaults to UDP port 123 (the normal NTP
port). In the second example, the destination port is specified as 12123. The
first parameter in each case (30 or 60 respectively) is the interval in seconds
between broadcast packets being sent. The second parameter in each case is the
broadcast address to send the packet to. This should correspond to the
broadcast address of one of the network interfaces on the computer where
*chronyd* is running.
+
You can have more than 1 *broadcast* directive if you have more than 1 network
interface onto which you want to send NTP broadcast packets.
+
*chronyd* itself cannot act as a broadcast client; it must always be configured
as a point-to-point client by defining specific NTP servers and peers. This
broadcast server feature is intended for providing a time source to other NTP
implementations.
+
If *ntpd* is used as the broadcast client, it will try to measure the
round-trip delay between the server and client with normal client mode packets.
Thus, the broadcast subnet should also be the subject of an <<allow,*allow*>>
directive.

[[clientloglimit]]*clientloglimit* _limit_::
This directive specifies the maximum amount of memory that *chronyd* is allowed
to allocate for logging of client accesses. The default limit is 524288 bytes,
which allows monitoring of several thousands of addresses at the same time.
+
In older *chrony* versions if the limit was set to 0, the memory allocation was
unlimited.
+
An example of the use of this directive is:
+
----
clientloglimit 1048576
----

[[noclientlog]]*noclientlog*::
This directive, which takes no arguments, specifies that client accesses are
not to be logged. Normally they are logged, allowing statistics to be reported
using the <<chronyc.adoc#clients,*clients*>> command in *chronyc*.

[[local]]*local* [_option_]...::
The *local* directive enables a local reference mode, which allows *chronyd*
operating as an NTP server to appear synchronised to real time (from the
viewpoint of clients polling it), even when it was never synchronised or
the last update of the clock happened a long time ago.
+
This directive is normally used in an isolated network, where computers are
required to be synchronised to one another, but not necessarily to real time.
The server can be kept vaguely in line with real time by manual input.
+
The *local* directive has the following options:
+
*stratum* _stratum_:::
This option sets the stratum of the server which will be reported to clients
when the local reference is active. The specified value is in the range 1
through 15, and the default value is 10. It should be larger than the maximum
expected stratum in the network when external NTP servers are accessible.
+
Stratum 1 indicates a computer that has a true real-time reference directly
connected to it (e.g. GPS, atomic clock, etc.), such computers are expected to
be very close to real time. Stratum 2 computers are those which have a stratum
1 server; stratum 3 computers have a stratum 2 server and so on. A value
of 10 indicates that the clock is so many hops away from a reference clock that
its time is fairly unreliable.
*distance* _distance_:::
This option sets the threshold for the root distance which will activate the local
reference. If *chronyd* was synchronised to some source, the local reference
will not be activated until its root distance reaches the specified value (the
rate at which the distance is increasing depends on how well the clock was
tracking the source). The default value is 1 second.
+
The current root distance can be calculated from root delay and root dispersion
(reported by the <<chronyc.adoc#tracking,*tracking*>> command in *chronyc*) as:
+
----
distance = delay / 2 + dispersion
----
*orphan*:::
This option enables a special '`orphan`' mode, where sources with stratum equal
to the local _stratum_ are assumed to not serve real time. They are ignored
unless no other source is selectable and their reference IDs are smaller than
the local reference ID.
+
This allows multiple servers in the network to use the same *local*
configuration and to be synchronised to one another, without confusing clients
that poll more than one server. Each server needs to be configured to poll all
other servers with the *local* directive. This ensures only the server with the
smallest reference ID has the local reference active and others are
synchronised to it. When that server fails, another will take over.
+
The *orphan* mode is compatible with the *ntpd*'s orphan mode (enabled by the
*tos orphan* command).
::
+
An example of the directive is:
+
----
local stratum 10 orphan
----

[[port]]*port* _port_::
This option allows you to configure the port on which *chronyd* will listen for
NTP requests. The port will be open only when an address is allowed by the
<<allow,*allow*>> directive or the <<chronyc.adoc#allow,*allow*>> command in
*chronyc*, an NTP peer is configured, or the broadcast server mode is enabled.
+
The default value is 123, the standard NTP port. If set to 0, *chronyd* will
never open the server port and will operate strictly in a client-only mode. The
source port used in NTP client requests can be set by the
<<acquisitionport,*acquisitionport*>> directive.

[[ratelimit]]*ratelimit* [_option_]...::
This directive enables response rate limiting for NTP packets. Its purpose is
to reduce network traffic with misconfigured or broken NTP clients that are
polling the server too frequently. The limits are applied to individual IP
addresses. If multiple clients share one IP address (e.g. multiple hosts behind
NAT), the sum of their traffic will be limited. If a client that increases its
polling rate when it does not receive a reply is detected, its rate limiting
will be temporarily suspended to avoid increasing the overall amount of
traffic. The maximum number of IP addresses which can be monitored at the same
time depends on the memory limit set by the <<clientloglimit,*clientloglimit*>>
directive.
+
The *ratelimit* directive supports a number of options (which can be defined
in any order):
+
*interval*:::
This option sets the minimum interval between responses. It is defined as a
power of 2 in seconds. The default value is 3 (8 seconds). The minimum value
is -4 and the maximum value is 12.
*burst*:::
This option sets the maximum number of responses that can be sent in a burst,
temporarily exceeding the limit specified by the *interval* option. This is
useful for clients that make rapid measurements on start (e.g. *chronyd* with
the *iburst* option). The default value is 8. The minimum value is 1 and the
maximum value is 255.
*leak*:::
This option sets the rate at which responses are randomly allowed even if the
limits specified by the *interval* and *burst* options are exceeded. This is
necessary to prevent an attacker who is sending requests with a spoofed
source address from completely blocking responses to that address. The leak
rate is defined as a power of 1/2 and it is 3 by default, i.e. on average at
least every eighth request has a response. The minimum value is 1 and the
maximum value is 4.
::
+
An example use of the directive is:
+
----
ratelimit interval 4 burst 4
----
+
This would reduce the response rate for IP addresses that send packets on
average more frequently than once per 16 seconds or send packets in bursts
of more than 4 packets.

[[smoothtime]]*smoothtime* _max-freq_ _max-wander_ [*leaponly*]::
The *smoothtime* directive can be used to enable smoothing of the time that
*chronyd* serves to its clients to make it easier for them to track it and keep
their clocks close together even when large offset or frequency corrections are
applied to the server's clock, for example after being offline for a longer
time.
+
BE WARNED: The server is intentionally not serving its best estimate of the
true time. If a large offset has been accumulated, it can take a very long time
to smooth it out. This directive should be used only when the clients are not
configured to also poll another NTP server, because they could reject this
server as a falseticker or fail to select a source completely.
+
The smoothing process is implemented with a quadratic spline function with two
or three pieces. It is independent from any slewing applied to the local system
clock, but the accumulated offset and frequency will be reset when the clock is
corrected by stepping, e.g. by the <<makestep,*makestep*>> directive or the
<<chronyc.adoc#makestep,*makestep*>> command in *chronyc*. The process can be
reset without stepping the clock by the <<chronyc.adoc#smoothtime,*smoothtime
reset*>> command.
+
The first two arguments of the directive are the maximum frequency offset of
the smoothed time to the tracked NTP time (in ppm) and the maximum rate at
which the frequency offset is allowed to change (in ppm per second). *leaponly*
is an optional third argument which enables a mode where only leap seconds are
smoothed out and normal offset and frequency changes are ignored. The *leaponly*
option is useful in a combination with the <<leapsecmode,*leapsecmode slew*>>
directive to allow the clients to use multiple time smoothing servers safely.
+
The smoothing process is activated automatically when 1/10000 of the estimated
skew of the local clock falls below the maximum rate of frequency change. It
can be also activated manually by the <<chronyc.adoc#smoothtime,*smoothtime
activate*>> command, which is particularly useful when the clock is
synchronised only with manual input and the skew is always larger than the
threshold. The <<chronyc.adoc#smoothing,*smoothing*>> command can be used to
monitor the process.
+
An example suitable for clients using *ntpd* and 1024 second polling interval
could be:
+
----
smoothtime 400 0.001
----
+
An example suitable for clients using *chronyd* on Linux could be:
+
----
smoothtime 50000 0.01
----

=== Command and monitoring access

[[bindcmdaddress]]*bindcmdaddress* _address_::
The *bindcmdaddress* directive allows you to specify the network interface on
which *chronyd* will listen for monitoring command packets (issued by
*chronyc*). This provides an additional level of access restriction above that
available through the <<cmddeny,*cmddeny*>> mechanism.
+
This directive can also change the path of the Unix domain command socket,
which is used by *chronyc* to send configuration commands. The socket must be
in a directory that is accessible only by the root or _chrony_ user. The
directory will be created on start if it does not exist. The compiled-in default
path of the socket is _@CHRONYSOCKDIR@/chronyd.sock_.
+
By default, *chronyd* binds to the loopback interface (with addresses
_127.0.0.1_ and _::1_). This blocks all access except from localhost. To listen
for command packets on all interfaces, you can add the lines:
+
----
bindcmdaddress 0.0.0.0
bindcmdaddress ::
----
+
to the configuration file.
+
For each of the IPv4 and IPv6 protocols, only one *bindcmdaddress* directive can be
specified.
+
An example that sets the path of the Unix domain command socket is:
+
----
bindcmdaddress /var/run/chrony/chronyd.sock
----

[[cmdallow]]*cmdallow* [*all*] [_subnet_]::
This is similar to the <<allow,*allow*>> directive, except that it allows
monitoring access (rather than NTP client access) to a particular subnet or
host. (By '`monitoring access`' is meant that *chronyc* can be run on those
hosts and retrieve monitoring data from *chronyd* on this computer.)
+
The syntax is identical to the *allow* directive.
+
There is also a *cmdallow all* directive with similar behaviour to the *allow
all* directive (but applying to monitoring access in this case, of course).
+
Note that *chronyd* has to be configured with the
<<bindcmdaddress,*bindcmdaddress*>> directive to not listen only on the
loopback interface to actually allow remote access.

[[cmddeny]]*cmddeny* [*all*] [_subnet_]::
This is similar to the <<cmdallow,*cmdallow*>> directive, except that it denies
monitoring access to a particular subnet or host, rather than allowing it.
+
The syntax is identical.
+
There is also a *cmddeny all* directive with similar behaviour to the *cmdallow
all* directive.

[[cmdport]]*cmdport* _port_::
The *cmdport* directive allows the port that is used for run-time monitoring
(via the *chronyc* program) to be altered from its default (323). If set to 0,
*chronyd* will not open the port, this is useful to disable *chronyc*
access from the Internet. (It does not disable the Unix domain command socket.)
+
An example shows the syntax:
+
----
cmdport 257
----
+
This would make *chronyd* use UDP 257 as its command port. (*chronyc* would
need to be run with the *-p 257* switch to inter-operate correctly.)

[[cmdratelimit]]*cmdratelimit* [_option_]...::
This directive enables response rate limiting for command packets. It is
similar to the <<ratelimit,*ratelimit*>> directive, except responses to
localhost are never limited and the default interval is 1 (2 seconds), the default
burst is 16, and the default leak rate is 2.
+
An example of the use of the directive is:
+
----
cmdratelimit interval 2
----

=== Real-time clock (RTC)

[[hwclockfile]]*hwclockfile* _file_::
The *hwclockfile* directive sets the location of the adjtime file which is
used by the *hwclock* program on Linux. *chronyd* parses the file to find out
if the RTC keeps local time or UTC. It overrides the <<rtconutc,*rtconutc*>>
directive.
+
The compiled-in default value is '_@DEFAULT_HWCLOCK_FILE@_'.
+
An example of the directive is:
+
----
hwclockfile /etc/adjtime
----

[[rtcautotrim]]*rtcautotrim* _threshold_::
The *rtcautotrim* directive is used to keep the RTC close to the system clock
automatically. When the system clock is synchronised and the estimated error
between the two clocks is larger than the specified threshold, *chronyd* will
trim the RTC as if the <<chronyc.adoc#trimrtc,*trimrtc*>> command in *chronyc*
was issued.
+
This directive is effective only with the <<rtcfile,*rtcfile*>> directive.
+
An example of the use of this directive is:
+
----
rtcautotrim 30
----
+
This would set the threshold error to 30 seconds.

[[rtcdevice]]*rtcdevice* _device_::
The *rtcdevice* directive sets the path to the device file for accessing the
RTC. The default path is _/dev/rtc_.

[[rtcfile]]*rtcfile* _file_::
The *rtcfile* directive defines the name of the file in which *chronyd* can
save parameters associated with tracking the accuracy of the RTC.
+
An example of the directive is:
+
----
rtcfile @CHRONYVARDIR@/rtc
----
+
*chronyd* saves information in this file when it exits and when the *writertc*
command is issued in *chronyc*. The information saved is the RTC's error at
some epoch, that epoch (in seconds since January 1 1970), and the rate at which
the RTC gains or loses time.
+
So far, the support for real-time clocks is limited; their code is even more
system-specific than the rest of the software. You can only use the RTC
facilities (the <<rtcfile,*rtcfile*>> directive and the *-s* command-line
option to *chronyd*) if the following three conditions apply:
+
. You are running Linux.
. The kernel is compiled with extended real-time clock support (i.e. the
  _/dev/rtc_ device is capable of doing useful things).
. You do not have other applications that need to make use of _/dev/rtc_ at all.

[[rtconutc]]*rtconutc*::
*chronyd* assumes by default that the RTC keeps local time (including any
daylight saving changes). This is convenient on PCs running Linux which are
dual-booted with Windows.
+
If you keep the RTC on local time and your computer is off when daylight saving
(summer time) starts or ends, the computer's system time will be one hour in
error when you next boot and start chronyd.
+
An alternative is for the RTC to keep Universal Coordinated Time (UTC). This
does not suffer from the 1 hour problem when daylight saving starts or ends.
+
If the *rtconutc* directive appears, it means the RTC is required to keep UTC.
The directive takes no arguments. It is equivalent to specifying the *-u*
switch to the Linux *hwclock* program.
+
Note that this setting is overridden when the <<hwclockfile,*hwclockfile*>>
directive is specified.

[[rtcsync]]*rtcsync*::
The *rtcsync* directive enables a mode where the system time is periodically
copied to the RTC and *chronyd* does not try to track its drift. This directive
cannot be used with the <<rtcfile,*rtcfile*>> directive.
+
On Linux, the RTC copy is performed by the kernel every 11 minutes.
+
On Mac OS X, <<chronyd,*chronyd*>> will perform the RTC copy every 60 minutes
when the system clock is in a synchronised state.
+
On other systems this directive does nothing.

=== Logging

[[log]]*log* [_option_]...::
The *log* directive indicates that certain information is to be logged.
The log files are written to the directory specified by the <<logdir,*logdir*>>
directive. A banner is periodically written to the files to indicate the
meanings of the columns.
+
*measurements*:::
This option logs the raw NTP measurements and related information to a file
called _measurements.log_. An example line (which actually appears as a single
line in the file) from the log file is shown below.
+
----
2015-10-13 05:40:50 203.0.113.15    N  2 111 111 1111  10 10 1.0 \
   -4.966e-03  2.296e-01  1.577e-05  1.615e-01  7.446e-03
----
+
The columns are as follows (the quantities in square brackets are the values
from the example line above):
+
. Date [2015-10-13]
. Hour:Minute:Second. Note that the date-time pair is expressed in UTC, not the
  local time zone. [05:40:50]
. IP address of server or peer from which measurement came [203.0.113.15]
. Leap status (_N_ means normal, _+_ means that the last minute of the current
  month has 61 seconds, _-_ means that the last minute of the month has 59
  seconds, _?_ means the remote computer is not currently synchronised.) [N]
. Stratum of remote computer. [2]
. RFC 5905 tests 1 through 3 (1=pass, 0=fail) [111]
. RFC 5905 tests 5 through 7 (1=pass, 0=fail) [111]
. Tests for maximum delay, maximum delay ratio and maximum delay dev ratio,
  against defined parameters, and a test for synchronisation loop (1=pass,
  0=fail) [1111]
. Local poll [10]
. Remote poll [10]
. '`Score`' (an internal score within each polling level used to decide when to
  increase or decrease the polling level. This is adjusted based on number of
  measurements currently being used for the regression algorithm). [1.0]
. The estimated local clock error (_theta_ in RFC 5905). Positive indicates
  that the local clock is slow of the remote source. [-4.966e-03]
. The peer delay (_delta_ in RFC 5905). [2.296e-01]
. The peer dispersion (_epsilon_ in RFC 5905). [1.577e-05]
. The root delay (_DELTA_ in RFC 5905). [1.615e-01]
. The root dispersion (_EPSILON_ in RFC 5905). [7.446e-03]
+
*statistics*:::
This option logs information about the regression processing to a file called
_statistics.log_. An example line (which actually appears as a single line in
the file) from the log file is shown below.
+
----
2015-07-22 05:40:50 203.0.113.15     6.261e-03 -3.247e-03 \
     2.220e-03  1.874e-06  1.080e-06 7.8e-02  16   0   8
----
+
The columns are as follows (the quantities in square brackets are the values
from the example line above):
+
. Date [2015-07-22]
. Hour:Minute:Second. Note that the date-time pair is expressed in
  UTC, not the local time zone. [05:40:50]
. IP address of server or peer from which measurement comes [203.0.113.15]
. The estimated standard deviation of the measurements from the source (in
  seconds). [6.261e-03]
. The estimated offset of the source (in seconds, positive means the local
  clock is estimated to be fast, in this case). [-3.247e-03]
. The estimated standard deviation of the offset estimate (in seconds).
  [2.220e-03]
. The estimated rate at which the local clock is gaining or losing time
  relative to the source (in seconds per second, positive means the local clock
  is gaining). This is relative to the compensation currently being applied to
  the local clock, _not_ to the local clock without any compensation.
  [1.874e-06]
. The estimated error in the rate value (in seconds per second). [1.080e-06].
. The ratio of |old_rate - new_rate| / old_rate_error. Large values
  indicate the statistics are not modelling the source very well. [7.8e-02]
. The number of measurements currently being used for the regression
  algorithm. [16]
. The new starting index (the oldest sample has index 0; this is the method
  used to prune old samples when it no longer looks like the measurements fit a
  linear model). [0, i.e. no samples discarded this time]
. The number of runs. The number of runs of regression residuals with the same
  sign is computed. If this is too small it indicates that the measurements are
  no longer represented well by a linear model and that some older samples need
  to be discarded. The number of runs for the data that is being retained is
  tabulated. Values of approximately half the number of samples are expected.
  [8]
+
*tracking*:::
This option logs changes to the estimate of the system's gain or loss rate, and
any slews made, to a file called _tracking.log_. An example line (which
actually appears as a single line in the file) from the log file is shown
below.
+
----
2015-02-23 05:40:50 203.0.113.15     3    340.529      1.606  1.046e-03 N \
            4  6.849e-03 -4.670e-04
----
+
The columns are as follows (the quantities in square brackets are the
values from the example line above) :
+
. Date [2015-02-03]
. Hour:Minute:Second. Note that the date-time pair is expressed in UTC, not the
  local time zone. [05:40:50]
. The IP address of the server or peer to which the local system is synchronised.
  [203.0.113.15]
. The stratum of the local system. [3]
. The local system frequency (in ppm, positive means the local system runs fast
  of UTC). [340.529]
. The error bounds on the frequency (in ppm). [1.606]
. The estimated local offset at the epoch (which is rapidly corrected by
  slewing the local clock. (In seconds, positive indicates the local system
  is fast of UTC). [1.046e-3]
. Leap status (_N_ means normal, _+_ means that the last minute of this month
  has 61 seconds, _-_ means that the last minute of the month has 59 seconds,
  _?_ means the clock is not currently synchronised.) [N]
. The number of combined sources. [4]
. The estimated standard deviation of the combined offset (in seconds).
  [6.849e-03]
. The remaining offset correction from the previous update (in seconds,
  positive means the system clock is slow of UTC). [-4.670e-04]
+
*rtc*:::
This option logs information about the system's real-time clock. An example
line (which actually appears as a single line in the file) from the _rtc.log_
file is shown below.
+
----
2015-07-22 05:40:50     -0.037360 1       -0.037434\
          -37.948  12   5  120
----
+
The columns are as follows (the quantities in square brackets are the
values from the example line above):
+
. Date [2015-07-22]
. Hour:Minute:Second. Note that the date-time pair is expressed in UTC, not the
  local time zone. [05:40:50]
. The measured offset between the RTC and the system clock in seconds.
  Positive indicates that the RTC is fast of the system time [-0.037360].
. Flag indicating whether the regression has produced valid coefficients.
  (1 for yes, 0 for no). [1]
. Offset at the current time predicted by the regression process. A large
  difference between this value and the measured offset tends to indicate that
  the measurement is an outlier with a serious measurement error. [-0.037434]
. The rate at which the RTC is losing or gaining time relative to the system
  clock. In ppm, with positive indicating that the RTC is gaining time.
  [-37.948]
. The number of measurements used in the regression. [12]
. The number of runs of regression residuals of the same sign. Low values
  indicate that a straight line is no longer a good model of the measured data
  and that older measurements should be discarded. [5]
. The measurement interval used prior to the measurement being made (in
  seconds). [120]
+
*refclocks*:::
This option logs the raw and filtered reference clock measurements to a file
called _refclocks.log_. An example line (which actually appears as a single
line in the file) from the log file is shown below.
+
----
2009-11-30 14:33:27.000000 PPS2    7 N 1  4.900000e-07 -6.741777e-07  1.000e-06
----
+
The columns are as follows (the quantities in square brackets are the values
from the example line above):
+
. Date [2009-11-30]
. Hour:Minute:Second.Microsecond. Note that the date-time pair is expressed in
  UTC, not the local time zone. [14:33:27.000000]
. Reference ID of the reference clock from which the measurement came. [PPS2]
. Sequence number of driver poll within one polling interval for raw samples,
  or _-_ for filtered samples. [7]
. Leap status (_N_ means normal, _+_ means that the last minute of the current
  month has 61 seconds, _-_ means that the last minute of the month has 59
  seconds). [N]
. Flag indicating whether the sample comes from PPS source. (1 for yes,
  0 for no, or _-_ for filtered sample). [1]
. Local clock error measured by reference clock driver, or _-_ for filtered sample.
  [4.900000e-07]
. Local clock error with applied corrections. Positive indicates that the local
  clock is slow. [-6.741777e-07]
. Assumed dispersion of the sample. [1.000e-06]
+
*tempcomp*:::
This option logs the temperature measurements and system rate compensations to
a file called _tempcomp.log_. An example line (which actually appears as a
single line in the file) from the log file is shown below.
+
----
2015-04-19 10:39:48  2.8000e+04  3.6600e-01
----
+
The columns are as follows (the quantities in square brackets are the values
from the example line above):
+
. Date [2015-04-19]
. Hour:Minute:Second. Note that the date-time pair is expressed in UTC, not the
  local time zone. [10:39:48]
. Temperature read from the sensor. [2.8000e+04]
. Applied compensation in ppm, positive means the system clock is running
  faster than it would be without the compensation. [3.6600e-01]
+
::
An example of the directive is:
+
----
log measurements statistics tracking
----

[[logbanner]]*logbanner* _entries_::
A banner is periodically written to the log files enabled by the <<log,*log*>>
directive to indicate the meanings of the columns.
+
The *logbanner* directive specifies after how many entries in the log file
should be the banner written. The default is 32, and 0 can be used to disable
it entirely.

[[logchange]]*logchange* _threshold_::
This directive sets the threshold for the adjustment of the system clock that
will generate a syslog message. Clock errors detected via NTP packets,
reference clocks, or timestamps entered via the
<<chronyc.adoc#settime,*settime*>> command of *chronyc* are logged.
+
By default, the threshold is 1 second.
+
An example of the use is:
+
----
logchange 0.1
----
+
which would cause a syslog message to be generated if a system clock error of over
0.1 seconds starts to be compensated.

[[logdir]]*logdir* _directory_::
This directive allows the directory where log files are written to be
specified.
+
An example of the use of this directive is:
+
----
logdir /var/log/chrony
----

[[mailonchange]]*mailonchange* _email_ _threshold_::
This directive defines an email address to which mail should be sent if
*chronyd* applies a correction exceeding a particular threshold to the system
clock.
+
An example of the use of this directive is:
+
----
mailonchange root@localhost 0.5
----
+
This would send a mail message to root if a change of more than 0.5 seconds
were applied to the system clock.
+
This directive cannot be used when a system call filter is enabled by the *-F*
option as the *chronyd* process will not be allowed to fork and execute the
sendmail binary.

=== Miscellaneous

[[include]]*include* _pattern_::
The *include* directive includes a configuration file or multiple configuration
files if a wildcard pattern is specified. This can be useful when maintaining
configuration on multiple hosts to keep the differences in separate files.
+
An example of the directive is:
+
----
include @SYSCONFDIR@/chrony.d/*.conf
----

[[keyfile]]*keyfile* _file_::
This directive is used to specify the location of the file containing ID-key
pairs for authentication of NTP packets.
+
The format of the directive is shown in the example below:
+
----
keyfile @SYSCONFDIR@/chrony.keys
----
+
The argument is simply the name of the file containing the ID-key pairs. The
format of the file is shown below:
+
----
10 tulip
11 hyacinth
20 MD5 ASCII:crocus
25 SHA1 HEX:1dc764e0791b11fa67efc7ecbc4b0d73f68a070c
 ...
----
+
Each line consists of an ID, name of an authentication hash function (optional),
and a password. The ID can be any unsigned integer in the range 1 through
2^32-1. The default hash function is *MD5*. Depending on how *chronyd*
was compiled, other supported functions might be *SHA1*, *SHA256*, *SHA384*,
*SHA512*, *RMD128*, *RMD160*, *RMD256*, *RMD320*, *TIGER*, and *WHIRLPOOL*. The
password can be specified as a string of characters not containing white space
with an optional *ASCII:* prefix, or as a hexadecimal number with the *HEX:*
prefix. The maximum length of the line is 2047 characters.
+
The password is used with the hash function to generate and verify a message
authentication code (MAC) in NTP packets. It is recommended to use SHA1, or
stronger, hash function with random passwords specified in the hexadecimal
format that have at least 128 bits. *chronyd* will log a warning to
syslog on start if a source is specified in the configuration file with a key
that has password shorter than 80 bits.
+
The <<chronyc.adoc#keygen,*keygen*>> command of *chronyc* can be used to
generate random keys for the key file. By default, it generates 160-bit MD5 or
SHA1 keys.

[[lock_all]]*lock_all*::
The *lock_all* directive will lock chronyd into RAM so that it will never be
paged out. This mode is only supported on Linux. This directive uses the Linux
*mlockall()* system call to prevent *chronyd* from ever being swapped out. This
should result in lower and more consistent latency. It should not have
significant impact on performance as *chronyd's* memory usage is modest. The
*mlockall(2)* man page has more details.

[[pidfile]]*pidfile* _file_::
*chronyd* always writes its process ID (PID) to a file, and checks this file on
startup to see if another *chronyd* may already be running on the system. By
default, the file used is _/var/run/chronyd.pid_. The *pidfile* directive
allows the name to be changed, e.g.:
+
----
pidfile /run/chronyd.pid
----

[[sched_priority]]*sched_priority* _priority_::
On Linux, the *sched_priority* directive will select the SCHED_FIFO real-time
scheduler at the specified priority (which must be between 0 and 100). On Mac
OS X, this option must have either a value of 0 (the default) to disable the
thread time constraint policy or 1 for the policy to be enabled. Other systems
do not support this option.
+
On Linux, this directive uses the *sched_setscheduler()* system call to
instruct the kernel to use the SCHED_FIFO first-in, first-out real-time
scheduling policy for *chronyd* with the specified priority. This means that
whenever *chronyd* is ready to run it will run, interrupting whatever else is
running unless it is a higher priority real-time process. This should not
impact performance as *chronyd* resource requirements are modest, but it should
result in lower and more consistent latency since *chronyd* will not need to
wait for the scheduler to get around to running it. You should not use this
unless you really need it. The *sched_setscheduler(2)* man page has more
details.
+
On Mac OS X, this directive uses the *thread_policy_set()* kernel call to
specify real-time scheduling. As noted for Linux, you should not use this
directive unless you really need it.

[[user]]*user* _user_::
The *user* directive sets the name of the system user to which *chronyd* will
switch after start in order to drop root privileges.
+
On Linux, *chronyd* needs to be compiled with support for the *libcap* library.
On Mac OS X, FreeBSD, NetBSD and Solaris *chronyd* forks into two processes.
The child process retains root privileges, but can only perform a very limited
range of privileged system calls on behalf of the parent.
+
The compiled-in default value is _@DEFAULT_USER@_.

[[examples]]
== EXAMPLES

=== NTP client with permanent connection to NTP servers

This section shows how to configure *chronyd* for computers that are connected
to the Internet (or to any network containing true NTP servers which ultimately
derive their time from a reference clock) permanently or most of the time.

To operate in this mode, you will need to know the names of the NTP servers
you want to use. You might be able to find names of suitable servers by one of
the following methods:

* Your institution might already operate servers on its network.
  Contact your system administrator to find out.
* Your ISP probably has one or more NTP servers available for its
  customers.
* Somewhere under the NTP homepage there is a list of public
  stratum 1 and stratum 2 servers. You should find one or more servers that are
  near to you. Check that their access policy allows you to use their
  facilities.
* Use public servers from the http://www.pool.ntp.org/[pool.ntp.org] project.

Assuming that your NTP servers are called _foo.example.net_, _bar.example.net_
and _baz.example.net_, your _chrony.conf_ file could contain as a minimum:

----
server foo.example.net
server bar.example.net
server baz.example.net
----

However, you will probably want to include some of the other directives. The
<<driftfile,*driftfile*>>, <<makestep,*makestep*>> and <<rtcsync,*rtcsync*>>
might be particularly useful. Also, the *iburst* option of the
<<server,*server*>> directive is useful to speed up the initial
synchronisation. The smallest useful configuration file would look something
like:

----
server foo.example.net iburst
server bar.example.net iburst
server baz.example.net iburst
driftfile @CHRONYVARDIR@/drift
makestep 1.0 3
rtcsync
----

When using a pool of NTP servers (one name is used for multiple servers which
might change over time), it is better to specify them with the <<pool,*pool*>>
directive instead of multiple *server* directives. The configuration file could
in this case look like:

----
pool pool.ntp.org iburst
driftfile @CHRONYVARDIR@/drift
makestep 1.0 3
rtcsync
----

=== NTP client with infrequent connection to NTP servers

This section shows how to configure *chronyd* for computers that have
occasional connections to NTP servers. In this case, you will need some
additional configuration to tell *chronyd* when the connection goes up and
down. This saves the program from continuously trying to poll the servers when
they are inaccessible.

Again, assuming that your NTP servers are called _foo.example.net_,
_bar.example.net_ and _baz.example.net_, your _chrony.conf_ file would now
contain:

----
server foo.example.net offline
server bar.example.net offline
server baz.example.net offline
driftfile @CHRONYVARDIR@/drift
makestep 1.0 3
rtcsync
----

The *offline* keyword indicates that the servers start in an offline state, and
that they should not be contacted until *chronyd* receives notification from
*chronyc* that the link to the Internet is present. To tell *chronyd* when to
start and finish sampling the servers, the <<chronyc.adoc#online,*online*>> and
<<chronyc.adoc#offline,*offline*>> commands of *chronyc* need to be used.

To give an example of their use, assuming that *pppd* is the program being
used to connect to the Internet and that *chronyc* has been installed at
_@BINDIR@/chronyc_, the script _/etc/ppp/ip-up_ would include:

----
@BINDIR@/chronyc online
----

and the script _/etc/ppp/ip-down_ would include:

----
@BINDIR@/chronyc offline
----

*chronyd*'s polling of the servers would now only occur whilst the machine is
actually connected to the Internet.

=== Isolated networks

This section shows how to configure *chronyd* for computers that never have
network conectivity to any computer which ultimately derives its time from a
reference clock.

In this situation, one computer is selected to be the master timeserver. The
other computers are either direct clients of the master, or clients of clients.

The <<local,*local*>> directive enables a local reference mode, which allows
*chronyd* to appear synchronised even when it is not.

The rate value in the master's drift file needs to be set to the average rate
at which the master gains or loses time. *chronyd* includes support for this,
in the form of the <<manual,*manual*>> directive and the
<<chronyc.adoc#settime,*settime*>> command in the *chronyc* program.

If the master is rebooted, *chronyd* can re-read the drift rate from the drift
file. However, the master has no accurate estimate of the current time. To get
around this, the system can be configured so that the master can initially set
itself to a '`majority-vote`' of selected clients' times; this allows the
clients to '`flywheel`' the master while it is rebooting.

The <<smoothtime,*smoothtime*>> directive is useful when the clocks of the
clients need to stay close together when the local time is adjusted by the
<<chronyc.adoc#settime,*settime*>> command. The smoothing process needs to be
activated by the <<chronyc.adoc#smoothtime,*smoothtime activate*>> command when
the local time is ready to be served. After that point, any adjustments will be
smoothed out.

A typical configuration file for the master (called _master_) might be
(assuming the clients and the master are in the _192.168.165.x_ subnet):

----
initstepslew 1 client1 client3 client6
driftfile @CHRONYVARDIR@/drift
local stratum 8
manual
allow 192.168.165.0/24
smoothtime 400 0.01
rtcsync
----

For the clients that have to resynchronise the master when it restarts,
the configuration file might be:

----
server master iburst
driftfile @CHRONYVARDIR@/drift
allow 192.168.165.0/24
makestep 1.0 3
rtcsync
----

The rest of the clients would be the same, except that the *allow* directive is
not required.

If there is no suitable computer to be designated as the master, or there is a
requirement to keep the clients synchronised even when it fails, the *orphan*
option of the *local* directive enables a special mode where the master is
selected from multiple computers automatically. They all need to use the same
*local* configuration and poll one another. The server with the smallest
reference ID (which is based on its IP address) will take the role of the
master and others will be synchronised to it. When it fails, the server with
the second smallest reference ID will take over and so on.

A configuration file for the first server might be (assuming there are three
servers called _master1_, _master2_, and _master3_):

----
initstepslew 1 master2 master3
server master2
server master3
driftfile @CHRONYVARDIR@/drift
local stratum 8 orphan
manual
allow 192.168.165.0/24
rtcsync
----

The other servers would be the same, except the hostnames in the *initstepslew*
and *server* directives would be modified to specify the other servers. Their
clients might be configured to poll all three servers.

=== RTC tracking

This section considers a computer which has occasional connections to the
Internet and is turned off between '`sessions`'. In this case, *chronyd* relies
on the computer's RTC to maintain the time between the periods when it is
powered up. It assumes that Linux is run exclusively on the computer. Dual-boot
systems might work; it depends what (if anything) the other system does to the
RTC. On 2.6 and later kernels, if your motherboard has a HPET, you will need to
enable the *HPET_EMULATE_RTC* option in your kernel configuration. Otherwise,
*chronyd* will not be able to interact with the RTC device and will give up
using it.

When the computer is connected to the Internet, *chronyd* has access to
external NTP servers which it makes measurements from. These measurements are
saved, and straight-line fits are performed on them to provide an estimate of
the computer's time error and rate of gaining or losing time.

When the computer is taken offline from the Internet, the best estimate of the
gain or loss rate is used to free-run the computer until it next goes online.

Whilst the computer is running, *chronyd* makes measurements of the RTC (via
the _/dev/rtc_ interface, which must be compiled into the kernel). An estimate
is made of the RTC error at a particular RTC second, and the rate at which the
RTC gains or loses time relative to true time.

When the computer is powered down, the measurement histories for all the NTP
servers are saved to files (if the <<dumponexit,*dumponexit*>> directive is
specified in the configuration file), and the RTC tracking information is also
saved to a file (if the <<rtcfile,*rtcfile*>> directive has been specified).
These pieces of information are also saved if the <<chronyc.adoc#dump,*dump*>>
and <<chronyc.adoc#writertc,*writertc*>> commands respectively are issued
through *chronyc*.

When the computer is rebooted, *chronyd* reads the current RTC time and the RTC
information saved at the last shutdown. This information is used to set the
system clock to the best estimate of what its time would have been now, had it
been left running continuously. The measurement histories for the servers are
then reloaded.

The next time the computer goes online, the previous sessions' measurements can
contribute to the line-fitting process, which gives a much better estimate of
the computer's gain or loss rate.

One problem with saving the measurements and RTC data when the machine is shut
down is what happens if there is a power failure; the most recent data will not
be saved. Although *chronyd* is robust enough to cope with this, some
performance might be lost. (The main danger arises if the RTC has been changed
during the session, with the *trimrtc* command in *chronyc*. Because of this,
*trimrtc* will make sure that a meaningful RTC file is saved after the
change is completed).

The easiest protection against power failure is to put the *dump* and
*writertc* commands in the same place as the *offline* command is issued to
take *chronyd* offline; because *chronyd* free-runs between online sessions, no
parameters will change significantly between going offline from the Internet
and any power failure.

A final point regards computers which are left running for extended periods and
where it is desired to spin down the hard disc when it is not in use (e.g. when
not accessed for 15 minutes). *chronyd* has been planned so it supports such
operation; this is the reason why the RTC tracking parameters are not saved to
disc after every update, but only when the user requests such a write, or
during the shutdown sequence. The only other facility that will generate
periodic writes to the disc is the *log rtc* facility in the configuration
file; this option should not be used if you want your disc to spin down.

To illustrate how a computer might be configured for this case, example
configuration files are shown.

For the _chrony.conf_ file, the following can be used as an example.

----
server foo.example.net maxdelay 0.4 offline
server bar.example.net maxdelay 0.4 offline
server baz.example.net maxdelay 0.4 offline
logdir /var/log/chrony
log statistics measurements tracking
driftfile @CHRONYVARDIR@/drift
makestep 1.0 3
maxupdateskew 100.0
dumponexit
dumpdir @CHRONYVARDIR@
rtcfile @CHRONYVARDIR@/rtc
----

*pppd* is used for connecting to the Internet. This runs two scripts
_/etc/ppp/ip-up_ and _/etc/ppp/ip-down_ when the link goes online and offline
respectively.

The relevant part of the _/etc/ppp/ip-up_ file is:

----
@BINDIR@/chronyc online
----

and the relevant part of the _/etc/ppp/ip-down_ script is:

----
@BINDIR@/chronyc -m offline dump writertc
----

*chronyd* is started during the boot sequence with the *-r* and *-s* options.
It might need to be started before any software that depends on the system clock
not jumping or moving backwards, depending on the directives in *chronyd*'s
configuration file.

For the system shutdown, *chronyd* should receive a SIGTERM several seconds
before the final SIGKILL; the SIGTERM causes the measurement histories and RTC
information to be saved.

== SEE ALSO

<<chronyc.adoc#,*chronyc(1)*>>, <<chronyd.adoc#,*chronyd(8)*>>

== BUGS

For instructions on how to report bugs, please visit
https://chrony.tuxfamily.org/.

== AUTHORS

chrony was written by Richard Curnow, Miroslav Lichvar, and others.