Network Time Protocol daemon

From ArchWiki
Revision as of 18:08, 17 February 2014 by Indigo (Talk | contribs) (Running as a daemon: adding ML link to warning; https://mailman.archlinux.org/pipermail/arch-dev-public/2014-February/025872.html)

Jump to: navigation, search

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 and currently not maintained for Linux.

This article further describes how to set up and run the NTP daemon, both as a client and as a server.

Installation

Install ntp, available in the official repositories.

Configuration

The main daemon is ntpd, which is configured in /etc/ntp.conf.

The ntp package provides a default configuration file that should make ntpd work out of the box in client mode, without requiring custom configuration.

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 use the pool.ntp.org servers (alternative link) and choose the server pool that is closest to your location.

The following lines are just an example:

/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

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) so that it will never be used unless internet access is lost:

server 127.127.1.0
fudge  127.127.1.0 stratum 10

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 prevents dumping status data from ntpd (also with ntpq or ntpdc).

You can also add other options:

restrict default kod nomodify notrap nopeer noquery
Note: This still allows other people to query your time server. You need to add noserve to stop serving time. It will also block time synchronization since it blocks all packets except ntpq and ntpdc queries.

Full docs for the "restrict" option are in man ntp_acc. See https://support.ntp.org/bin/view/Support/AccessRestrictions for detailed instructions.

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
Note: Defining the log file is not mandatory, but it is always a good idea to have feedback for ntpd operations.

Other resources about NTP configuration

In conclusion, never forget manual pages: man ntp.conf is likely to answer any doubts you could still have (see also the related manual pages: man {ntpd|ntp_auth|ntp_mon|ntp_acc|ntp_clock|ntp_misc}).

Using without daemon

To synchronize your system clock just once without starting ntpd, run:

# ntpd -q

Using the -q flag causes ntpd to set the time once and quit, i.e. the daemon will not be started. If the operation is unsuccessful, your system clock will not be synchronized.

Note: This has the same effect as the now deprecated ntpdate.

After updating the system clock, store the time to the hardware clock so that it is preserved when rebooting:

# hwclock -w

Synchronize once per boot

Warning: Using this method is discouraged on servers, and in general on machines that run without rebooting for more than a few days.

Write a oneshot systemd unit:

/etc/systemd/system/ntp-once.service
[Unit]
Description=Network Time Service (once)
After=network.target nss-lookup.target 

[Service]
Type=oneshot
ExecStart=/usr/bin/ntpd -g -u ntp:ntp ; /usr/bin/hwclock -w

[Install]
WantedBy=multi-user.target

and enable it.

Note: A systemd unit of the type oneshot executes once only. Hence the ntpd -q option should not be used in this case.

Netctl

To synchronize your system clock along with a network connection through the use with Netctl. You can append the following line to your netctl profile.

ExecUpPost='/usr/bin/ntpd -gq || true'

Wicd

Executing ntpd once after successful connection of Wicd can be achieved by creating a script in the appropriate postconnect directory:

/etc/wicd/scripts/postconnect/ntpd
 #!/bin/bash
 
 /usr/bin/ntpd -gq 

Running as a daemon

Warning: As of February 2014, ntpd used as a daemon is vulnerable to DDoS attacks in some configurations. See here and the CVE report.

Start ntpd.service with systemctl. To have it started at boot also enable it, or alternatively use the command:

# timedatectl set-ntp 1

Check whether the daemon is synchronizing correctly

Use ntpq to see the list of configured peers:

$ ntpq -np

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).

NetworkManager

Note: ntpd should still be running when the network is down if the hwclock daemon is disabled, so you should not use this.

The ntpd daemon can be brought up/down along with a network connection through the use of NetworkManager's dispatcher scripts. You will need to install networkmanager-dispatcher-ntpd from the official repositories.

Running in a chroot

Tango-view-refresh-red.pngThis article or section is out of date.Tango-view-refresh-red.png

Reason: ntpd.service does not use /etc/conf.d/ntpd.conf anymore (Discuss in Talk:Network Time Protocol daemon#)
Note: ntpd should be run as non-root before attempting to jail it in a chroot (default in the vanilla Arch Linux package), since chroots are relatively useless at securing processes running as root.

Edit /etc/conf.d/ntpd.conf and change

NTPD_ARGS="-g -u ntp:ntp"

to

NTPD_ARGS="-g -i /var/lib/ntp -u ntp:ntp"

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.

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.

See also