Skip to content

Migrating from NTPsec

Both ntpsec and ntpd-rs can serve a similar role on Unix systems. This guide is aimed to help those migrating machines currently running ntpsec to ntpd-rs. We assume some experience with the ntpsec configuration format. A user with no or little ntpsec experience may be better of following our getting started guide.

Configuration for ntpsec uses a custom format that functions as a list of commands. In contrast, ntpd-rs uses a configuration file in the .toml format that gives values to properties. That means that in most cases fields cannot be repeated. Comments can be added by starting them with a #. The remainder of the line is then ignored.

An ntpd-rs configuration can be validated with ntp-ctl validate -c <path>. This will check all field names, and in some cases validates that a value is within the range of valid values for that property.

This guide will not go into detail on all of ntpsec's configuration directives, but rather focus on those most important for successful migration. If a particular directive is not mentioned here, there may still be ntpd-rs options in the configuration reference that achieve the desired effect. Note that not all functionality of ntpsec is currently supported, a short overview of major differences is given at the end of this document

Time sources

Server entries

The server and pool commands have a direct equivalent in ntpd-rs:

# ntpsec

# ntpd-rs
mode = "server"
address = ""

mode = "server"
address = ""

mode = "pool"
address = ""
count = 4

A source in pool mode must explicitly define an integer count, the maximum number of connections from this pool. The ntpd-rs daemon will actively try to keep the pool "filled": new connections will be spun up if a source from the pool is unreachable.

For server directives with NTS, these can be converted to

# ntpsec
server nts
server nts ca path/to/certificate/authority.pem

# ntpd-rs

certificate_authority = "path/to/certificate/authority.pem"

There is no direct equivalent of ntpsec's maxpoll and minpoll flags that can be configured on a per-source basis. Instead ntpd-rs defines poll interval bounds globally for all time sources:

poll-interval-limits = { min = <minpoll>, max = <maxpoll> }
initial-poll-interval = <desired initial poll interval>

There is no support for bursting in ntpd-rs yet, but the ntpd-rs algorithm is able to synchronize much more quickly (with fewer measurements) than ntpsec's algorithm. Therefore, if any bursting directive (burst or iburst) is present, these usually can be ignored when translating configurations. In some cases, if strict custom poll limits are in place, these may need to be relaxed.

Reference clocks

The current version of ntpd-rs does not yet support local reference clocks, but this feature is on our roadmap. If you are interested in migrating a configuration using local reference clocks, we would be interested in hearing the details. This information will help guide our implementation effort.

Time synchronization options

The minimum number of time sources needed for time synchronization in ntpd-rs is configured through minimum-agreeing-sources:

mininum-agreeing-sources = <minsources>
If fewer agreeing source are available, no synchronization is performed and the clock will drift. This option is a combination of ntpd's minclock and minsane. Its default value is 3, the recommended value from a security perspective. In ntpd, a default of 3 is used for minclock and 1 for minsane.

Through the tinker command's step and stepout flags, ntpsec allows limiting of the maximum change in time made. Although not entirely the same in functionality, ntpd-rs allows similar restrictions to be enforced through a number of panic thresholds. Steps at startup are controlled through the startup-panic-threshold, whilst steps during normal operation are controlled with single-step-panic-threshold and accumulated-step-panic-threshold.

single-step-panic-threshold = 1000
startup-step-panic-threshold = { forward="inf", backward = 86400 }
accumulated-step-panic-threshold = "inf"

ntpsec and ntpd-rs use different algorithms for synchronizing the time. This means that options for tuning filtering of the time differ significantly, and we cannot offer precise guidance on how to translate the ntpsec parameters to values for ntpd-rs. When migrating a configuration that tunes ntpsec's algorithm, one should take the intent of the tuning and use that as guidance when choosing which of ntpd-rs's time synchronization options to change.

Server Configuration & Access Control

The restrict command is used in ntpsec to deny requests from a client. In ntpsec this is a global setting. A flag configures what happens with connections from this client. For instance, ignore will silently ignore the request, while kod sends a response to the client that notifies it that its request is denied.

This logic is expressed differently in ntpd-rs. A specific server can be configured to have a denylist and an allowlist. The subnets to allow or deny must be specified in CIDR notation (an IP address followed by a slash and the number of masked bits, for example or

listen="<ip or [::]>:<port>"

filter = [
action = "ignore"

filter = [
action = "deny"

The allow and deny list configuration is optional in ntpd-rs. By default, if a server is configured it will accept traffic from anywhere. When configuring both allow and deny lists, ntpd-rs will first check if a remote is on the deny list. Only if this is not the case will the allow list be considered.

The allowlist.action and denylist.action properties can have two values:

  • ignore corresponds to ntpsec's ignore and silently ignores the request
  • deny corresponds to ntpsec's kod and sends a deny kiss-o'-death packet

The stratum can can be configured in ntpd-rs with the local-stratum key:

local-stratum = <stratum>


NTS can be enabled for a server by configuring an NTS key exchange server:

# ntpsec
nts key /etc/letsencrypt/live/
nts cert /etc/letsencrypt/live/

# ntpd-rs
listen = "<ip or [::]>:<port>
certificate-chain-path = "/etc/letsencrypt/live/"
private-key-path = "/etc/letsencrypt/live/"

Note that unlike ntpsec, ntpd-rs does not have a default ip address on which it listens for nts-ke traffic: it needs to be provided explicitly. The port is optional and defaults to the standard value 4460.

The keys used to sign the cookies is kept in memory, but can additionally be stored to a file (so they are preserved after a restart).

# ntpsec
nts cookie /var/lib/ntp/nts-keys

# ntpd-rs
key_storage_path = "/var/lib/ntp/nts-keys"

The key_storage_path requires a full path to a file, and there is no default path.

Keys are rotated to limit the damage when a key is leaked. By default, this occurs every 24 hours. At most 7 older keys are remembered to serve clients with older cookies. These numbers can be configured with the key-rotation-interval and stale-key-count parameters:

stale-key-count = <number of old keys to keep>
key-rotation-interval = <rotation interval in seconds>
Note that the defaults for these settings mean that cookies for the server are only valid for slightly more than 1 week.

Sharing the keys with which the nts cookies are encrypted between multiple ntpd-rs servers is not yet supported.

Unsupported features

Not all functionality in ntpsec currently has an equivalent in ntpd-rs. In particular, the following major features currently don't have good alternatives in ntpd-rs:

  • Local hardware devices as time sources.
  • Support for ntp mac authentication.
  • Marking subsets of sources as more trusted than others.
  • bursting

If any of these features are critical for your use case, ntpd-rs might not be an option for you yet. Please let us know if you miss these features or want to sponsor any of them, as this helps us prioritise our work.