Network Time Protocol daemon: Difference between revisions
(→Synchronize once per boot: rm gap) |
(→Usage: Paragraphs are enough for separation here) |
||
| Line 45: | Line 45: | ||
== Usage == | == Usage == | ||
The basic command to start the NTP daemon is: | The basic command to start the NTP daemon is: | ||
| Line 55: | Line 53: | ||
# ntpd -u ntp:ntp | # ntpd -u ntp:ntp | ||
Use ''ntpq'' to see the list of configured peers: | Use ''ntpq'' to see the list of configured peers: | ||
| Line 63: | Line 59: | ||
The delay, offset and jitter columns should be non-zero. The servers ''ntpd'' is synchronizing with are prefixed by an asterisk. It can take several minutes before ''ntpd'' selects a server to synchronize with; try checking after 17 minutes (1024 seconds). | The delay, offset and jitter columns should be non-zero. The servers ''ntpd'' is synchronizing with are prefixed by an asterisk. It can take several minutes before ''ntpd'' selects a server to synchronize with; try checking after 17 minutes (1024 seconds). | ||
System time can be synchronized without using the daemon as well. However, this would not be suitable for machines that run without rebooting for more than a few days. To synchronize the system clock just once without starting ''ntpd'' in the background, run: | System time can be synchronized without using the daemon as well. However, this would not be suitable for machines that run without rebooting for more than a few days. To synchronize the system clock just once without starting ''ntpd'' in the background, run: | ||
Revision as of 10:03, 3 June 2015
zh-CN:Network Time Protocol daemon
Network Time Protocol is the most common method to synchronize the software clock of a GNU/Linux system with internet time servers. It is designed to mitigate the effects of variable network latency and can usually maintain time to within tens of milliseconds over the public Internet. The accuracy on local area networks is even better, up to one millisecond.
The NTP Project provides a reference implementation of the protocol called simply NTP. An alternative to NTP is Chrony, a dial-up friendly and specifically designed for systems that are not online all the time, and OpenNTPD, part of the OpenBSD project.
This article further describes how to set up and run the NTP daemon, both as a client and as a server.
Installation
Install the ntp package. By default, ntpd works in client mode without further configuration.
Start the daemon at boot
timedatectl set-ntp 1 as root will inadvertedly stop a running ntpd.service.[1]The ntp package provides ntpd.service for systemd. Enable it with systemctl to start the daemon at boot.
Synchronize once per boot
The ntp package provides the ntpdate.service. Enable it with systemd to synchronize time once per boot. This oneshot service is the useful option, if the system does not require the daemon to remain active after one successful time synchronization.
If the synchronized time should be written to the hardware clock as well, configure the provided unit as described in systemd#Editing provided unit files:
/etc/systemd/system/ntpdate.service.d/hwclock.conf
[Service] ExecStart=/usr/bin/hwclock -w
before starting it.
Usage
The basic command to start the NTP daemon is:
# ntpd
However, it will run as root in the background. Hence, it should always be started specifying the user option:
# ntpd -u ntp:ntp
Use ntpq to see the list of configured peers:
$ ntpq -p
The delay, offset and jitter columns should be non-zero. The servers ntpd is synchronizing with are prefixed by an asterisk. It can take several minutes before ntpd selects a server to synchronize with; try checking after 17 minutes (1024 seconds).
System time can be synchronized without using the daemon as well. However, this would not be suitable for machines that run without rebooting for more than a few days. To synchronize the system clock just once without starting ntpd in the background, run:
# ntpd -q
Using the -q flag causes the ntpd daemon to set the time once and quit, i.e. not fork to the background.
If the operation is unsuccessful, your system clock will not be synchronized.
The system time also will not be synchronized if the ntp-server's time differs from the system clock by more than a given threshold (so-called panic-gate), in order to protect running system processes. However, the option -g can be used to disable the threshold and allow it to be exceeded, e.g. when the time is set for the first time or if the hardware clock is faulty.
Configuration
The main daemon is ntpd, which is configured in /etc/ntp.conf.
The following describes main configuration items to customize. Also refer to the manual pages: man ntp.conf and the related man {ntpd|ntp_auth|ntp_mon|ntp_acc|ntp_clock|ntp_misc}.
Configuring connection to NTP servers
If you want to configure /etc/ntp.conf manually, first thing you define is the servers your machine will synchronize to.
NTP servers are classified in a hierarchical system with many levels called strata: the devices which are considered independent time sources are classified as stratum 0 sources; the servers directly connected to stratum 0 devices are classified as stratum 1 sources; servers connected to stratum 1 sources are then classified as stratum 2 sources and so on.
It has to be understood that a server's stratum cannot be taken as an indication of its accuracy or reliability. Typically, stratum 2 servers are used for general synchronization purposes: if you do not already know the servers you are going to connect to, you should choose a server pool close to your location from the pool.ntp.org servers (alternative link).
Since ntp version 4.2.7.p465-2, Arch Linux uses its own default vendor pool of NTP servers provided by the NTP Pool Project (see FS#41700). Modify those to suit your needs, e.g. if you want to use your country's servers with an option:
/etc/ntp.conf
server 0.fr.pool.ntp.org iburst server 1.fr.pool.ntp.org iburst server 2.fr.pool.ntp.org iburst server 3.fr.pool.ntp.org iburst
The iburst option is recommended, and sends a burst of packets only if it cannot obtain a connection with the first attempt. The burst option always does this, even on the first attempt, and should never be used without explicit permission and may result in blacklisting.
Configuring your own NTP server
If setting up an NTP server, you need to add local clock as a server, so that, in case it loses internet access, it will continue serving time to the network; add local clock as a stratum 10 server (using the fudge command) (you can set up to stratum 15) so that it will never be used unless internet access is lost:
server 127.127.1.1 fudge 127.127.1.1 stratum 12
Next, define the rules that will allow clients to connect to your service (localhost is considered a client too) using the restrict command; you should already have a line like this in your file:
restrict default nomodify nopeer noquery
This restricts everyone from modifying anything and prevents everyone from querying the status of your time server: nomodify prevents reconfiguring ntpd (with ntpq or ntpdc), and noquery is important to prevent dumping status data from ntpd (also with ntpq or ntpdc).
You can also add other options:
restrict default kod nomodify notrap nopeer noquery
noserve to stop serving time. It will also block time synchronization since it blocks all packets except ntpq and ntpdc queries.If you want to change any of these, see the full docs for the "restrict" option in man ntp_acc, the detailed ntp instructions and #As a daemon.
Following this line, you need to tell ntpd what to allow through into your server; the following line is enough if you are not configuring an NTP server:
restrict 127.0.0.1
If you want to force DNS resolution to the IPv6 namespace, write -6 before the IP address or host name (-4 forces IPv4 instead), for example:
restrict -6 default kod nomodify notrap nopeer noquery restrict -6 ::1 # ::1 is the IPv6 equivalent for 127.0.0.1
Lastly, specify the drift file (which keeps track of your clock's time deviation) and optionally the log file location:
driftfile /var/lib/ntp/ntp.drift logfile /var/log/ntp.log
A very basic configuration file will look like this:
/etc/ntp.conf
server 0.pool.ntp.org iburst server 1.pool.ntp.org iburst server 2.pool.ntp.org iburst server 3.pool.ntp.org iburst restrict default kod nomodify notrap nopeer noquery restrict -6 default kod nomodify notrap nopeer noquery restrict 127.0.0.1 restrict -6 ::1 driftfile /var/lib/ntp/ntp.drift logfile /var/log/ntp.log
Tips and tricks
Start at network connection
ntpd can be started by your network manager, so that the daemon only runs when the computer is online.
Netctl
Append the following lines to your netctl profile:
ExecUpPost='/usr/bin/ntpd || true' ExecDownPre='killall ntpd || true'
NetworkManager
The ntpd daemon can be brought up/down along with a network connection through the use of NetworkManager's dispatcher scripts. The networkmanager-dispatcher-ntpd from the official repositories installs one, pre-configured to start and stop the ntpd service with a connection.
Wicd
For Wicd, create a start script in the postconnect directory and a stop script in the predisconnect directory. Remember to make them executable:
/etc/wicd/scripts/postconnect/ntpd
#!/bin/bash /usr/bin/ntpd & -or- systemctl start ntpd &
/etc/wicd/scripts/predisconnect/ntpd
#!/bin/bash killall ntpd & -or- systemctl stop ntpd &
See also Wicd#Scripts.
KDE
KDE can use NTP (ntp must be installed) by right clicking the clock and selecting Adjust date/time. However, this requires the ntp daemon to be disabled before configuring KDE to use NTP. [2]
Configuring NTPD to use your GPS
Most of the articles online about configuring ntpd to receive time from a GPS suggest to use the SHM (shared memory) method. However, at least since ntpd version 4.2.8 a much better method is available. It connects directly to gpsd, so gpsd needs to be installed.
Add these lines to your /etc/ntp.conf:
/etc/ntp.conf
#========================================================= # GPSD native ntpd driver #========================================================= # This driver exists from at least ntp version 4.2.8 # Details at # https://www.eecis.udel.edu/~mills/ntp/html/drivers/driver46.html server 127.127.46.0 fudge 127.127.46.0 time1 0.0 time2 0.0 refid GPS
This will work as long as you have gpsd working. It connects to gpsd via the local socket and queries the "gpsd_json" object that is returned.
To test the setup, do the following:
- First, make sure that gpsd is working by running:
$ cgps -s
- Second, make sure ntpd is talking to gpsd. To test this, wait a few minutes and run
ntpq -p. The output should be something like this:
$ ntpq -p
remote refid st t when poll reach delay offset jitter ================================================================================== *GPSD_JSON(0) .GPS. 0 l 55 64 377 0.000 2.556 14.109
Running in a chroot
Create a new directory /etc/systemd/system/ntpd.service.d/ if it does not exist and a file named customexec.conf inside with the following content:
[Service] ExecStart= ExecStart=/usr/bin/ntpd -g -i /var/lib/ntp -u ntp:ntp -p /run/ntpd.pid
Then, edit /etc/ntp.conf to change the driftfile path such that it is relative to the chroot directory, rather than to the real system root. Change:
driftfile /var/lib/ntp/ntp.drift
to
driftfile /ntp.drift
Create a suitable chroot environment so that getaddrinfo() will work by creating pertinent directories and files (as root):
# mkdir /var/lib/ntp/etc /var/lib/ntp/lib /var/lib/ntp/proc # touch /var/lib/ntp/etc/resolv.conf /var/lib/ntp/etc/services
and by bind-mounting the aformentioned files:
/etc/fstab
... #ntpd chroot mounts /etc/resolv.conf /var/lib/ntp/etc/resolv.conf none bind 0 0 /etc/services /var/lib/ntp/etc/services none bind 0 0 /lib /var/lib/ntp/lib none bind 0 0 /proc /var/lib/ntp/proc none bind 0 0
# mount -a
Finally, restart ntpd daemon again. Once it restarted you can verify that the daemon process is chrooted by checking where /proc/{PID}/root symlinks to:
# ps -C ntpd | awk '{print $1}' | sed 1d | while read -r PID; do ls -l /proc/$PID/root; done
should now link to /var/lib/ntp instead of /.
It is relatively difficult to be sure that your driftfile configuration is actually working without waiting a while, as ntpd does not read or write it very often. If you get it wrong, it will log an error; if you get it right, it will update the timestamp. If you do not see any errors about it after a full day of running, and the timestamp is updated, you should be confident of success.