Chrony is a software package for maintaining the accuracy of computer system clocks. It consists of a pair of programs :
chronyd. This is a daemon which runs in background on the
system. It obtains measurements (e.g. via the network) of the system's
offset relative to other systems, and adjusts the system time
accordingly. For isolated systems, the user can periodically enter the
correct time by hand (using chronyc). In either case,
chronyd determines the rate at which the computer gains or loses
time, and compensates for this.
chronyd can also act as an NTP server, and provide a time-of-day service
to other computers. A typical set-up is to run chronyd on a gateway
computer that has a dial-up link to the Internet, and use it to serve time to
computers on a private LAN sitting behind the gateway. The IP addresses that
can act as clients of chronyd can be tightly controlled. The default is
no client access.
chronyc. This is a command-line driven control and
monitoring program. An administrator can use this to fine-tune various
parameters within the daemon, add or delete servers etc whilst the
daemon is running.
The IP addresses from which chronyc clients may connect can be tightly
controlled. The default is just the computer that chronyd itself is
running on.
The chrony suite makes use of the algorithm known as RSA
Data Security, Inc. MD5 Message-Digest Algorithm for authenticating
messages between different machines on the network.
In writing the chronyd program, extensive use has been made of
RFC1305, written by David Mills. I have occasionally referred to the
xntp suite's source code to check details of the protocol that
the RFC did not make absolutely clear. The core algorithms in
chronyd are all completely distinct from xntp, however.
Links on the chrony home page describe how to obtain the software.
Although most of the program is portable between Unix-like systems, there are parts that have to be tailored to each specific vendor's system. These are the parts that interface with the operating system's facilities for adjusting the system clock; different operating systems may provide different function calls to achieve this, and even where the same function is used it may have different quirks in its behaviour.
The software is known to work in the following environments:
Closely related systems may work too, but they have not been tested.
Porting the software to other system (particularly to those supporting
an adjtime system call) should not be difficult, however it
requires access to such systems to test out the driver.
The `reference' implementation of the Network Time Protocol is the
program xntpd, available via
The NTP home page.
xntpd is designed to support all the operating modes defined by
RFC1305, and has driver support for a large number of reference clocks
(such as GPS receivers) that can be connected directly to a computer,
thereby providing a so-called 'stratum 1' server.
Things chronyd can do that xntpd can't:
chronyd can perform usefully in an environment where access to
the time reference is intermittent. chronyd estimates
both the current time offset and the rate at which the
computer's clock gains or loses time, and can use that rate estimate to
trim the clock after the reference disappears. xntpd corrects
any time offset by speeding up and slowing down the computer clock, and
so could be left with a significant rate error if the reference
disappears whilst it is trying to correct a big offset.
chronyd provides support for isolated networks whether the only
method of time correction is manual entry (e.g. by the administrator
looking at a clock). chronyd can look at the errors corrected at
different updates to work out the rate at which the computer gains or
loses time, and use this estimate to trim the computer clock
subsequently.
chronyd provides support to work out the gain or loss rate of the
`real-time clock', i.e. the clock that maintains the time when the
computer is turned off. It can use this data when the system boots to
set the system time from a corrected version of the real-time clock.
These real-time clock facilities are only available on certain releases
of Linux, so far.
xntpd program is supported by other programs to carry out
certain functions. ntpdate is used to provide an initial
correction to the system clock based on a `one-shot' sampling of other
NTP servers. tickadj is used to adjust certain operating system
parameters to make xntpd work better. All this functionality is
integrated into chronyd.
Things xntpd can do that chronyd can't:
xntpd supports effectively all of RFC1305, including broadcast /
multicast clients and extra encryption schemes for authenticating
data packets.
xntpd has been ported to more types of computer / operating
system (so far).
timed is a program that is part of the BSD networking suite. It
uses broadcast packets to find all machines running the daemon within a
subnet. The machines elect a master which periodically measures the
system clock offsets of the other computers using ICMP timestamps.
Corrections are sent to each member as a result of this process.
Problems that may arise with timed are :
timed does not seem to do this.
timed does not have any integrated capability for feeding
real-time into its estimates, or for estimating the average rate of time
loss/gain of the machines relative to real-time (unless one of the
computers in the group has access to an external reference and is always
appointed as the `master').
timed does have the benefit over chronyd that for isolated
networks of computers, they will track the `majority vote' time. For
such isolated networks, chronyd requires one computer to be the
`master' with the others slaved to it. If the master has a particular
defective clock, the whole set of computers will tend to slip relative
to real time (but they will stay accurate relative to one
another).
Chrony may be distributed in accordance with the GNU General Public License version 2, reproduced in See GPL.
If you think you've found a bug in chrony, or have a suggestion, please let us know. You can join chrony users mailing list by sending a message with the subject subscribe to chrony-users-request@chrony.tuxfamily.org. Only subscribers can post to the list.
When you are reporting a bug, please send us all the information you can. Unfortunately, chrony has proven to be one of those programs where it is very difficult to reproduce bugs in a different environment. So we may have to interact with you quite a lot to obtain enough extra logging and tracing to pin-point the problem in some cases. Please be patient and plan for this!
Of course, if you can debug the problem yourself and send us a source code patch to fix it, we will be very grateful!
Although chrony is now a fairly mature and established project, there are still areas that could be improved. If you can program in C and have some expertise in these areas, you might be able to fill the gaps.
Particular areas that need addressing are :
This involves creating equivalents of sys_solaris.c, sys_linux.c etc for the new system. Note, the Linux driver has been reported as working on a range of different architectures (Alpha, Sparc, MIPS as well as x86 of course).
I did a small amount of work on this under Cygwin. Only the sorting out of the include files has really been achieved so far. The two main areas still to address are
Currently, the RTC trimming mechanism is a manual operation, because there has to be a reasonable guarantee that the system will stay up for a reasonable length of time afterwards. (If it is shut down too soon, a poor characterisation of the RTC drift rate will be stored on disc, giving a bad system clock error when the system is next booted.)
To make chrony more automated for the non-expert user, it would be useful if this problem could be avoided so that trimrtc could be done automatically (e.g. in a crontab, or as part of the ip-up or ip-down scripts.)
The software is distributed as source code which has to be compiled. The source code is supplied in the form of a gzipped tar file, which unpacks to a subdirectory identifying the name and version of the program.
After unpacking the source code, change directory into it, and type
./configure
This is a shell script that automatically determines the system type.
There is a single optional parameter, --prefix which indicates
the directory tree where the software should be installed. For example,
./configure --prefix=/opt/free
will install the chronyd daemon into /opt/free/sbin and the
chronyc control program into /opt/free/bin. The default value for the
prefix is /usr/local.
The configure script assumes you want to use gcc as your compiler. If you want to use a different compiler, you can configure this way:
CC=cc CFLAGS=-O ./configure --prefix=/opt/free
for Bourne-family shells, or
setenv CC cc
setenv CFLAGS -O
./configure --prefix=/opt/free
for C-family shells.
If the software cannot (yet) be built on your system, an error message will be shown. Otherwise, Makefile will be generated.
If editline or readline library is available, chronyc will be built with line editing support. If you don't want this, specify the –disable-readline flag to configure. Please refer to see line editing support for more information.
If a timepps.h header is available, chronyd will be built with PPS API
reference clock driver. If the header is installed in a location that isn't
normally searched by the compiler, you can add it to the searched locations by
setting CPPFLAGS variable to -I/path/to/timepps.
Now type
make
to build the programs.
If you want to build the manual in plain text, HTML and info versions, type
make docs
Once the programs have been successfully compiled, they need to be installed in their target locations. This step normally needs to be performed by the superuser, and requires the following command to be entered.
make install
This will install the binaries, plain text manual and manpages.
To install the HTML and info versions of the manual as well, enter the command
make install-docs
If you want chrony to appear in the top level info directory listing, you need to run the install-info command manually after this step. install-info takes 2 arguments. The first is the path to the chrony.info file you have just installed. This will be the argument you gave to –prefix when you configured (/usr/local by default), with /share/info/chrony.info on the end. The second argument is the location of the file called dir. This will typically be /usr/share/info/dir. So the typical command line would be
install-info /usr/local/share/info/chrony.info /usr/share/info/dir
Now that the software is successfully installed, the next step is to set up a configuration file. The contents of this depend on the network environment in which the computer operates. Typical scenarios are described in the following section of the document.
Chronyc can be built with support for line editing, this allows you to use the cursor keys to replay and edit old commands. Two libraries are supported which provide such functionality, editline and GNU readline.
Please note that readline since version 6.0 is licensed under GPLv3+ which is incompatible with chrony's license GPLv2. You should use editline instead if you don't want to use older readline versions.
The configure script will automatically enable the line editing support if one of the supported libraries is available. If they are both available, the editline library will be used.
If you don't want to use it (in which case chronyc will use a minimal command line interface), invoke configure like this:
./configure --disable-readline other-options...
If you have editline, readline or ncurses installed in locations that aren't normally searched by the compiler and linker, you need to use extra options:
The configure and make procedures have some extra options that may be useful if you are building a distribution package for chrony.
The –infodir=DIR option to configure specifies an install directory for the info files. This overrides the info subdirectory of the argument to the –prefix option. For example, you might use
./configure --prefix=/usr --infodir=/usr/share/info
The –mandir=DIR option to configure specifies an install directory for the man pages. This overrides the man subdirectory of the argument to the –prefix option.
./configure --prefix=/usr --infodir=/usr/share/info --mandir=/usr/share/man
to set both options together.
The final option is the DESTDIR option to the make command. For example, you could use the commands
./configure --prefix=/usr --infodir=/usr/share/info --mandir=/usr/share/man
make all docs
make install DESTDIR=./tmp
cd tmp
tar cvf - . | gzip -9 > chrony.tar.gz
to build a package. When untarred within the root directory, this will install the files to the intended final locations.
In this section we discuss how to configure chrony for computers that have permanent connections to the internet (or to any network containing true NTP servers which ultimately derive their time from a reference clock).
To operate in this mode, you will need to know the names of the NTP server machines you wish to use. You may be able to find names of suitable servers by one of the following methods:
Assuming that you have found some servers, you need to set up a
configuration file to run chrony. The (compiled-in) default location
for this file is /etc/chrony.conf. Assuming that your ntp
servers are called a.b.c and d.e.f, your
chrony.conf file could contain as a minimum
server a.b.c
server d.e.f
server g.h.i
However, you will probably want to include some of the other directives
described later. The following directives will be particularly useful :
driftfile, commandkey, keyfile. The smallest
useful configuration file would look something like
server a.b.c
server d.e.f
server g.h.i
keyfile /etc/chrony.keys
commandkey 1
driftfile /etc/chrony.drift
In this section we discuss how to configure chrony for computers that have occasional connections to the internet.
As in the previous section, you will need access to NTP servers on the internet. The same remarks apply for how to find them.
In this case, you will need some additional configuration to tell
chronyd when the connection to the internet 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 a.b.c and
d.e.f, your chrony.conf file would need to contain
something like
server a.b.c
server d.e.f
server g.h.i
However, the following issues need to be addressed:
For this reason, it would be better to specify this part of your configuration file in the following way:
server 1.2.3.4 offline
server 5.6.7.8 offline
server 9.10.11.12 offline
Because numeric IP addresses have been used, the first problem is
overcome. The offline keyword indicates that the servers start
in an offline state, and that they should not be contacted until chronyd
receives notification that the link to the internet is present.
An alternative is to use the names of the NTP servers, and put entries for them into your /etc/hosts file. This will be OK as long as ‘files’ comes before ‘dns’ in the ‘hosts’ line of the /etc/nsswitch.conf file.
In order to notify chronyd of the presence of the link, you will need to
be able to log in to it with the program chronyc. To do this, chronyd
needs to be configured with an administrator password. To set up an
administrator password, you can create a file /etc/chrony.keys
containing a single line
1 xyzzy
and add the following line to /etc/chrony.conf (the order of the lines does not matter)
commandkey 1
The smallest useful configuration file would look something like
server 1.2.3.4 offline
server 5.6.7.8 offline
server 9.10.11.12 offline
keyfile /etc/chrony.keys
commandkey 1
driftfile /etc/chrony.drift
The next section describes how to tell chronyd when the internet link
goes up and down.
To use this option, you will need to configure a command key in
chronyd's configuration file /etc/chrony.conf, as described in
the previous section.
To tell chronyd when to start and finish sampling the servers, the
online and offline commands of chronyc need to be used.
To give an example of their use, we assume that pppd is the
program being used to connect to the internet, and that chronyc has been
installed at its default location /usr/local/bin/chronyc. We
also assume that the command key has been set up as described in the
previous section.
In the file /etc/ppp/ip-up we add the command sequence
/usr/local/bin/chronyc <<EOF
password xyzzy
online
EOF
and in the file /etc/ppp/ip-down we add the sequence
/usr/local/bin/chronyc <<EOF
password xyzzy
offline
EOF
chronyd's polling of the servers will now only occur whilst the
machine is actually connected to the Internet.
In this section we discuss how to configure chrony 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 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 directive in the
configuration file and the 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 across its outage.
A typical configuration file for the master (called master) might
be (assuming the clients are in the 192.168.165.x subnet and that the
master's address is 192.168.169.170)
driftfile /etc/chrony.drift
commandkey 25
keyfile /etc/chrony.keys
initstepslew 10 client1 client3 client6
local stratum 8
manual
allow 192.168.165
For the clients that have to resynchronise the master when it restarts, the configuration file might be
server master
driftfile /etc/chrony.drift
logdir /var/log/chrony
log measurements statistics tracking
keyfile /etc/chrony.keys
commandkey 24
local stratum 10
initstepslew 20 master
allow 192.168.169.170
The rest of the clients would be the same, except that the local
and allow directives are not required.
This section considers the home computer which has a dial-up connection. It assumes that Linux is run exclusively on the computer. Dual-boot systems may work; it depends what (if anything) the other system does to the system's real-time clock.
Much of the configuration for this case is discussed earlier (see Infrequent connection). This section addresses specifically the case of a computer which is turned off between 'sessions'.
In this case, chronyd relies on the computer's real-time clock
(RTC) to maintain the time between the periods when it is powered up.
The arrangement is shown in the figure below.
trim if required PSTN
+---------------------------+ +----------+
| | | |
v | | |
+---------+ +-------+ +-----+ +---+
| System's| measure error/ |chronyd| |modem| |ISP|
|real-time|------------------->| |-------| | | |
| clock | drift rate +-------+ +-----+ +---+
+---------+ ^ |
| | |
+---------------------------+ --o-----o---
set time at boot up |
+----------+
|NTP server|
+----------+
When the computer is connected to the Internet (via the modem),
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/losing time.
When the computer is taken offline from the Internet, the best estimate of the gain/loss rate is used to free-run the computer until it next goes online.
Whilst the computer is running, chronyd makes measurements of the
real-time clock (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.
The RTC is fully supported in 2.2, 2.4 and 2.6 kernels.
On 2.6 kernels, if your motherboard has a HPET, you need to enable the ‘HPET_EMULATE_RTC’ option in your kernel configuration. Otherwise, chrony will not be able to interact with the RTC device and will give up using it.
For kernels in the 2.0 series prior to 2.0.32, the kernel was set up to
trim the RTC every 11 minutes. This would be disasterous for
chronyd – there is no reliable way of synchronising with this
trimming. For this reason, chronyd only supports the RTC in 2.0
kernels from v2.0.32 onwards.
When the computer is powered down, the measurement histories for all the
NTP servers are saved to files (if the dumponexit directive is
specified in the configuration file), and the RTC tracking information
is also saved to a file (if the rtcfile directive has been
specified). These pieces of information are also saved if the
dump and 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/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 may 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 out
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 home 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 dial-up home computer might be configured, example configuration files are shown in this section.
For the /etc/chrony.conf file, the following can be used as an
example. NOTE : The server directives are only applicable
to customers of Demon Internet; users of other ISPs will need to use
their own ISP's NTP servers or public NTP servers.
server 158.152.1.65 minpoll 5 maxpoll 10 maxdelay 0.4 offline
server 158.152.1.76 minpoll 5 maxpoll 10 maxdelay 0.4 offline
server 194.159.253.2 minpoll 5 maxpoll 10 maxdelay 0.4 offline
logdir /var/log/chrony
log statistics measurements tracking
driftfile /etc/chrony.drift
keyfile /etc/chrony.keys
commandkey 25
maxupdateskew 100.0
dumponexit
dumpdir /var/log/chrony
rtcfile /etc/chrony.rtc
With Freeserve as the ISP, I use the following server lines :
server 194.152.64.68 minpoll 5 maxpoll 10 maxdelay 0.4 offline
server 194.152.64.35 minpoll 5 maxpoll 10 maxdelay 0.4 offline
server 194.152.64.34 minpoll 5 maxpoll 10 maxdelay 0.4 offline
I use pppd for connecting to my ISP. 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 (with a dummy password)
/usr/local/bin/chronyc <<EOF
password xxxxxxxx
online
EOF
and the relevant part of the /etc/ppp/ip-down script is
/usr/local/bin/chronyc <<EOF
password xxxxxxxx
offline
dump
writertc
EOF
(Because they have to contain the administrator password, it would be desirable to make the files readable only by root on a multiuser machine).
To start chronyd during the boot sequence, I have the following
in /etc/rc.d/rc.local (this is a Slackware system)
if [ -f /usr/local/sbin/chronyd -a -f /etc/chrony.conf ]; then
/usr/local/sbin/chronyd -r -s
echo "Start chronyd"
fi
The placement of this command may be important on some systems. In
particular, chronyd may need to be started several seconds (about
10 as a minimum) 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 out. There should be no need
to add anything to the shutdown sequence, unless (as my system had)
there is no pause between the SIGTERM and SIGKILL being delivered to the
remaining processes. So if you find something like
killall5 -15
killall5 -9
in your /etc/rc.d/rc.0 script, you will need to insert a sleep, e.g.
killall5 -15
sleep 5
killall5 -9
Otherwise, chronyd will not always save information on shutdown,
which could be a problem if you don't use dump and
writertc when you go offline.
The most common option to include in the configuration file is the
driftfile option. One of the major tasks of chronyd is to
work out how fast or how slow the system clock runs relative to real
time - e.g. in terms of seconds gained or lost per day. Measurements
over a long period are usually required to refine this estimate to an
acceptable degree of accuracy. Therefore, it would be bad if
chronyd had to work the value out each time it is restarted,
because the system clock would not run so accurately whilst the
determination is taking place.
To avoid this problem, chronyd allows the gain or loss rate to be
stored in a file, which can be read back in when the program is
restarted. This file is called the drift file, and might typically be
stored in /etc/chrony.drift. By specifying an option like the
following
driftfile /etc/chrony.drift
in the configuration file (/etc/chrony.conf), the drift file facility will be activated.
If chronyd has been installed to its default location
/usr/local/sbin/chronyd, starting it is simply a matter of
entering the command
/usr/local/sbin/chronyd
Information messages and warnings will be logged to syslog.
The command line options supported are as follows:
-d-f <conf-file>-rdump command in
chronyc, or by setting the dumponexit directive in the
configuration file. This option is useful if you want to stop and
restart chronyd briefly for any reason, e.g. to install a new
version. However, it only makes sense on systems where the kernel can
maintain clock compensation whilst not under chronyd's control.
The only version where this happens so far is Linux. On systems where
this is not the case, e.g. Solaris and SunOS the option should not be
used.
-sSupport for real-time clocks is limited at present - the criteria are
described in the section on the rtcfile directive (see rtcfile directive).
If chronyd cannot support the real time clock on your computer,
this option cannot be used and a warning message will be logged to the
syslog.
If used in conjunction with the `-r' flag, chronyd will attempt
to preserve the old samples after setting the system clock from the real
time clock. This can be used to allow chronyd to perform long
term averaging of the gain or loss rate across system reboots, and is
useful for dial-up systems that are shut down when not in use. For this
to work well, it relies on chronyd having been able to determine
accurate statistics for the difference between the real time clock and
system clock last time the computer was on.
-u <user>-vchronyd's version number to the terminal and
exits.
-P <priority>-m-4-6On systems that support an /etc/rc.local file for starting
programs at boot time, chronyd can be started from there.
On systems with a System V style initialisation (e.g. Solaris), a suitable start/stop script might be as shown below. This might be placed in the file /etc/rc2.d/S83chrony.
#!/bin/sh
# This file should have uid root, gid sys and chmod 744
#
killproc() { # kill the named process(es)
pid=`/usr/bin/ps -e |
/usr/bin/grep -w $1 |
/usr/bin/sed -e 's/^ *//' -e 's/ .*//'`
[ "$pid" != "" ] && kill $pid
}
case "$1" in
'start')
if [ -f /opt/free/sbin/chronyd -a -f /etc/chrony.conf ]; then
/opt/free/sbin/chronyd
fi
;;
'stop')
killproc chronyd
;;
*)
echo "Usage: /etc/rc2.d/S83chrony { start | stop }"
;;
esac
(In both cases, you may want to bear in mind that chronyd can
step the time when it starts. There may be other programs started at
boot time that could be upset by this, so you may need to consider the
ordering carefully. However, chronyd will need to start after
daemons providing services that it may require, e.g. the domain name
service.)
The configuration file is normally called /etc/chrony.conf; in fact, this is the compiled-in default. However, other locations can be specified with a command line option.
Each command in the configuration file is placed on a separate line. The following sections describe each of the commands in turn. The directives can occur in any order in the file.