Operation

The Knot DNS server part knotd can run either in the foreground, or in the background using the -d option. When run in the foreground, it doesn't create a PID file. Other than that, there are no differences and you can control both the same way.

The tool knotc is designed as a user front-end, making it easier to control a running server daemon. If you want to control the daemon directly, use SIGINT to quit the process or SIGHUP to reload the configuration.

If you pass neither configuration file (-c parameter) nor configuration database (-C parameter), the server will first attempt to use the default configuration database stored in /var/lib/knot/confdb or the default configuration file stored in /etc/knot/knot.conf. Both the default paths can be reconfigured with --with-storage=path or --with-configdir=path respectively.

Example of server start as a daemon:

$ knotd -d -c knot.conf

Example of server shutdown:

$ knotc -c knot.conf stop

For a complete list of actions refer to the program help (-h parameter) or to the corresponding manual page.

Also, the server needs to create rundir and storage directories in order to run properly.

Note

Avoid editing of or other manipulation with configuration file during start or reload of knotd or start of knotc and other utilities which use it. There is a risk of malfunction or a crash otherwise.

Configuration database

In the case of a huge configuration file, the configuration can be stored in a binary database. Such a database can be simply initialized:

$ knotc conf-init

or preloaded from a file:

$ knotc conf-import input.conf

Also the configuration database can be exported into a textual file:

$ knotc conf-export output.conf

Warning

The import and export commands access the configuration database directly, without any interaction with the server. Therefore, any data not yet committed to the database won't be exported. And the server won't reflect imported configuration correctly. So it is strictly recommended to import new configuration when the server is not running.

Dynamic configuration

The configuration database can be accessed using the server control interface while the server is running. To get the full power of the dynamic configuration, the server must be started with a specified configuration database location or with the default database initialized. Otherwise all the changes to the configuration will be temporary (until the server is stopped).

Note

The database can be imported in advance.

Most of the commands get an item name and value parameters. The item name is in the form of section[identifier].name. If the item is multivalued, more values can be specified as individual (command line) arguments.

Caution

Beware of the possibility of pathname expansion by the shell. For this reason, it is advisable to escape (with backslash) square brackets or to quote command parameters if not executed in the interactive mode.

To get the list of configuration sections or to get the list of section items:

$ knotc conf-list
$ knotc conf-list 'server'

To get the whole configuration or to get the whole configuration section or to get all section identifiers or to get a specific configuration item:

$ knotc conf-read
$ knotc conf-read 'remote'
$ knotc conf-read 'zone.domain'
$ knotc conf-read 'zone[example.com].master'

Warning

The following operations don't work on OpenBSD!

Modifying operations require an active configuration database transaction. Just one transaction can be active at a time. Such a transaction then can be aborted or committed. A semantic check is executed automatically before every commit:

$ knotc conf-begin
$ knotc conf-abort
$ knotc conf-commit

To set a configuration item value or to add more values or to add a new section identifier or to add a value to all identified sections:

$ knotc conf-set 'server.identity' 'Knot DNS'
$ knotc conf-set 'server.listen' '0.0.0.0@53' '::@53'
$ knotc conf-set 'zone[example.com]'
$ knotc conf-set 'zone.slave' 'slave2'

Note

Also the include operation can be performed. A non-absolute file location is relative to the server binary path, not to the control binary path!

$ knotc conf-set 'include' '/tmp/new_zones.conf'

To unset the whole configuration or to unset the whole configuration section or to unset an identified section or to unset an item or to unset a specific item value:

$ knotc conf-unset
$ knotc conf-unset 'zone'
$ knotc conf-unset 'zone[example.com]'
$ knotc conf-unset 'zone[example.com].master'
$ knotc conf-unset 'zone[example.com].master' 'remote2' 'remote5'

To get the change between the current configuration and the active transaction for the whole configuration or for a specific section or for a specific identified section or for a specific item:

$ knotc conf-diff
$ knotc conf-diff 'zone'
$ knotc conf-diff 'zone[example.com]'
$ knotc conf-diff 'zone[example.com].master'

Caution

While it is possible to change most of the configuration parameters dynamically or via configuration file reload, a few of the parameters in the section server require restarting the server, such that the changes take effect. These parameters are: rundir, user, pidfile, tcp-reuseport, udp-workers, tcp-workers, background-workers, and listen.

An example of possible configuration initialization:

$ knotc conf-begin
$ knotc conf-set 'server.listen' '0.0.0.0@53' '::@53'
$ knotc conf-set 'remote[master_server]'
$ knotc conf-set 'remote[master_server].address' '192.168.1.1'
$ knotc conf-set 'template[default]'
$ knotc conf-set 'template[default].storage' '/var/lib/knot/zones/'
$ knotc conf-set 'template[default].master' 'master_server'
$ knotc conf-set 'zone[example.com]'
$ knotc conf-diff
$ knotc conf-commit

Secondary (slave) mode

Running the server as a secondary is very straightforward as the zone is transfered automatically from a remote server. The received zone is usually stored in a zone file after the zonefile-sync period elapses. Zone differences are stored in the zone journal.

Primary (master) mode

If you just want to check the zone files before starting, you can use:

$ knotc zone-check example.com

Reading and editing zones

Knot DNS allows you to read or change zone contents online using the server control interface.

Warning

Avoid concurrent zone access from a third party software when a zone event (zone file load, refresh, DNSSEC signing, dynamic update) is in progress or pending. In such a case, zone events must be frozen before. For more information on how to freeze the zone read Reading and editing the zone file safely.

To get contents of all configured zones, or a specific zone contents, or zone records with a specific owner, or even with a specific record type:

$ knotc zone-read --
$ knotc zone-read example.com
$ knotc zone-read example.com ns1
$ knotc zone-read example.com ns1 NS

Note

If the record owner is not a fully qualified domain name, then it is considered as a relative name to the zone name.

To start a writing transaction on all zones or on specific zones:

$ knotc zone-begin --
$ knotc zone-begin example.com example.net

Now you can list all nodes within the transaction using the zone-get command, which always returns current data with all changes included. The command has the same syntax as zone-read.

Within the transaction, you can add a record to a specific zone or to all zones with an open transaction:

$ knotc zone-set example.com ns1 3600 A 192.168.0.1
$ knotc zone-set -- ns1 3600 A 192.168.0.1

To remove all records with a specific owner, or a specific rrset, or specific record data:

$ knotc zone-unset example.com ns1
$ knotc zone-unset example.com ns1 A
$ knotc zone-unset example.com ns1 A 192.168.0.2

To see the difference between the original zone and the current version:

$ knotc zone-diff example.com

Finally, either commit or abort your transaction:

$ knotc zone-commit example.com
$ knotc zone-abort example.com

A full example of setting up a completely new zone from scratch:

$ knotc conf-begin
$ knotc conf-set zone.domain example.com
$ knotc conf-commit
$ knotc zone-begin example.com
$ knotc zone-set example.com  @ 3600 SOA  ns admin 1 86400 900 691200 3600
$ knotc zone-set example.com  @ 3600 NS   ns
$ knotc zone-set example.com ns 3600 A    192.168.0.1
$ knotc zone-set example.com ns 3600 AAAA 2001:DB8::1
$ knotc zone-commit example.com

Note

If quotes are necessary for record data specification, remember to escape them:

$ knotc zone-set example.com @ 3600 TXT \"v=spf1 a:mail.example.com -all\"

Reading and editing the zone file safely

It's always possible to read and edit zone contents via zone file manipulation. It may lead to confusion or even a program crash, however, if the zone contents are continuously being changed by DDNS, DNSSEC signing and the like. In such a case, the safe way to modify the zone file is to freeze zone events first:

$ knotc -b zone-freeze example.com.
$ knotc -b zone-flush example.com.

After calling freeze on the zone, there still may be running zone operations (e.g. signing), causing freeze pending. Because of this, the blocking mode is used to ensure the operation was finished. Then the zone can be flushed to a file.

Now the zone file can be safely modified (e.g. using a text editor). If zonefile-load is not set to difference-no-serial, it's also necessary to increase SOA serial in this step to keep consistency. Finally, we can load the modified zone file and if successful, thaw the zone:

$ knotc -b zone-reload example.com.
$ knotc zone-thaw example.com.

Zone loading

The process of how the server loads a zone is influenced by the configuration of the zonefile-load and journal-content parameters (also DNSSEC signing applies), the existence of a zone file and journal (and their relative out-of-dateness), and whether it is a cold start of the server or a zone reload (e.g. invoked by the knotc interface). Please note that zone transfers are not taken into account here – they are planned after the zone is loaded (including zone bootstrap).

If the zone file exists and is not excluded by the configuration, it is first loaded and according to its SOA serial number, relevant journal changesets are applied. If this is a zone reload and we have zonefile-load set to difference, the difference between old and new contents is computed and stored in the journal like an update. The zone file should be either unchanged since last load or changed with incremented SOA serial. In the case of a decreased SOA serial, the load is interrupted with an error; if unchanged, it is increased by the server.

If the procedure described above succeeds without errors, the resulting zone contents are (after potential DNSSEC signing) used as the new zone.

The option journal-content set to all lets the server, beside better performance, keep track of the zone contents also across server restarts. It makes the cold start effectively work like a zone reload with the old contents loaded from the journal (unless this is the very first start with the zone not yet saved into the journal).

Journal behaviour

The zone journal keeps some history of changes made to the zone. It is useful for responding to IXFR queries. Also if zone file flush is disabled, the journal keeps the difference between the zone file and the current zone in case of server shutdown. The history is stored in changesets – differences of zone contents between two (usually subsequent) zone versions (specified by SOA serials).

Journals of all zones are stored in a common LMDB database. Huge changesets are split into 15-70 KiB [1] blocks to prevent fragmentation of the DB. The journal does each operation in one transaction to keep consistency of the DB and performance.

Each zone journal has its own occupation limits maximum usage and maximum depth. Changesets are stored in the journal one by one. When hitting any of the limits, the zone is flushed into the zone file if there are no redundant changesets to delete, and the oldest changesets are deleted. In the case of the size limit, twice [1] the needed amount of space is purged to prevent overly frequent deletes.

If zone file flush is disabled, then instead of flushing the zone, the journal tries to save space by merging the changesets into a special one. This approach is effective if the changes rewrite each other, e.g. periodically changing the same zone records, re-signing the whole zone etc. Thus the difference between the zone file and the zone is still preserved even if the journal deletes some older changesets.

If the journal is used to store both zone history and contents, a special changeset is present with zone contents. When the journal gets full, the changes are merged into this special changeset.

There is also a safety hard limit for overall journal database size, but it's strongly recommended to set the per-zone limits in a way to prevent hitting this one. For LMDB, it's hard to recover from the database-full state. For wiping one zone's journal, see knotc zone-purge +journal command.

Handling zone file, journal, changes, serials

Some configuration options regarding the zone file and journal, together with operation procedures, might lead to unexpected results. This chapter points out potential interference and both recommends and warns before some combinations thereof. Unfortunately, there is no optimal combination of configuration options, every approach has some disadvantages.

Example 1

Keep the zone file updated:

zonefile-sync: 0
zonefile-load: whole
journal-content: changes

These are default values. The user can always check the current zone contents in the zone file, and also modify it (recommended with server turned-off or taking the safe way). The journal serves here just as a source of history for secondary servers' IXFR. Some users dislike that the server overwrites their prettily prepared zone file.

Example 2

Zonefileless setup:

zonefile-sync: -1
zonefile-load: none
journal-content: all

Zone contents are stored only in the journal. The zone is updated by DDNS, zone transfer, or via the control interface. The user might have filled the zone contents initially from a zone file by setting zonefile-load to whole temporarily. It's also a good setup for secondary servers. Anyway, it's recommended to carefully tune the journal-size-related options to avoid surprises like the journal getting full (see Journal behaviour).

Example 3

Input-only zone file:

zonefile-sync: -1
zonefile-load: difference
journal-content: changes

The user can make changes to the zone by editing the zone file, and his pretty zone file is never overwritten or filled with DNSSEC-related autogenerated records – they are only stored in the journal.

Warning

The zone file's SOA serial must be properly set to a number which is higher than the current SOA serial in the zone (not in the zone file) if manually updated! This is important to ensure consistency of the journal and outgoing IXFR.

Note

This mode is not suitable if the zone can be modified externally (e.g. DDNS, knotc).

Example 4

Auto-increment SOA serial:

zonefile-sync: -1
zonefile-load: difference-no-serial
journal-content: all

This is similar to the previous setup, but the SOA serial is handled by the server automatically. So the user no longer needs to care about it in the zone file.

However, this requires setting journal-content to all so that the information about the last real SOA serial is preserved in case of server re-start. The sizing of journal limits needs to be taken into consideration (see Journal behaviour).

Note

This mode is not suitable if the zone can be modified externally (e.g. DDNS, knotc).

Zone bootstrapping on secondary

When zone refresh from the primary fails, the retry value from SOA is used as the interval between refresh attempts. In a case that SOA isn't known to the secondary (either because the zone hasn't been retrieved from the primary yet, or the zone has expired), a backoff is used for repeated retry attempts.

With every retry, the delay rises as a quadratic polynomial (5 * n^2, where n represents the sequence number of the retry attempt) up to two hours, each time with a random delay of 0 to 30 seconds added to spread the load on the primary. In each attempt, the retry interval is subject to retry-min-interval and retry-max-interval.

Until the refresh has been successfully completed, the backoff is restarted from the beginning by every zone-refresh or zone-retransfer of the zone triggered manually via knotc, by zone-purge or zone-restore of the zone's timers, or by a restart of knotd.

Zone expiration

On a primary, zone normally never expires. On a secondary, zone expiration results in removal of the current zone contents and a trigger of immediate zone refresh. The zone file and zone's journal are kept, but not used for answering requests until the refresh is successfully completed.

The zone expire timer is set according to the zone's SOA expire field. In addition to it, Knot DNS also supports EDNS EXPIRE extension of the expire timer in both primary and secondary roles as described in RFC 7314.

When Knot DNS is configured as a secondary, EDNS EXPIRE option present in a SOA, IXFR, or AFXR response from the primary is processed and used to update the zone timer when necessary. This functionality (together with requests of any other EDNS options) for a specified primary may be disabled using the no-edns configuration parameter.

If it's necessary, any zone may be expired manually using the zone-purge command of the knotc utility. Manual expiration is applicable to any zone, including a catalog zone or a zone on a primary. Beware, a manually expired zone on a primary or a manually expired catalog zone becomes valid again after a server configuration is reloaded or the knotd process is restarted, provided that the zone data hasn't been removed.

DNSSEC key states

During its lifetime, a DNSSEC key finds itself in different states. Most of the time it is used for signing the zone and published in the zone. In order to exchange the key, one type of a key rollover is necessary, and during this rollover, the key goes through various states with respect to the rollover type and also the state of the other key being rolled-over.

First, let's list the states of the key being rolled-in.

Standard states:

  • active — The key is used for signing.

  • published — The key is published in the zone, but not used for signing. If the key is a KSK or CSK, it is used for signing the DNSKEY RRSet.

  • ready (only for KSK) — The key is published in the zone and used for signing. The old key is still active, since we are waiting for the DS records in the parent zone to be updated (i.e. "KSK submission").

Special states for algorithm rollover:

  • pre-active — The key is not yet published in the zone, but it's used for signing the zone.

  • published — The key is published in the zone, and it's still used for signing since the pre-active state.

Second, we list the states of the key being rolled-out.

Standard states:

  • retire-active — The key is still used for signing, and is published in the zone, waiting for the updated DS records in parent zone to be acked by resolvers (KSK case) or synchronizing with KSK during algorithm rollover (ZSK case).

  • retired — The key is no longer used for signing. If ZSK, the key is still published in the zone.

  • removed — The key is not used in any way (in most cases such keys are deleted immediately).

Special states for algorithm rollover:

  • post-active — The key is no longer published in the zone, but still used for signing.

Special states for RFC 5011 trust anchor roll-over

  • revoke (only for KSK) — The key is published and used for signing, and the Revoked flag is set.

Note

Trust anchor roll-over is not implemented with automatic key management.

The revoke state can only be established using keymgr when using Manual key management.

The states listed above are relevant for keymgr operations like generating a key, setting its timers and listing KASP database.

Note that the key "states" displayed in the server log lines while zone signing are not according to those listed above, but just a hint as to what the key is currently used for (e.g. "public, active" = key is published in the zone and used for signing).

DNSSEC key rollovers

This section describes the process of DNSSEC key rollover and its implementation in Knot DNS, and how the operator might watch and check that it's working correctly. The prerequisite is automatic zone signing with enabled automatic key management.

The KSK and ZSK rollovers are triggered by the respective zone key getting old according to the settings (see KSK and ZSK lifetimes).

The algorithm rollover starts when the policy algorithm field is updated to a different value.

The signing scheme rollover happens when the policy signing scheme field is changed.

It's also possible to change the algorithm and signing scheme in one rollover.

The operator may check the next rollover phase time by watching the next zone signing time, either in the log or via knotc zone-status. There is no special log for finishing a rollover.

Note

There are never two key rollovers running in parallel for one zone. If a rollover is triggered while another is in progress, it waits until the first one is finished. Note that a rollover might be considered finished when the old key is retired or waiting to be deleted.

The ZSK rollover is performed with Pre-publish method, KSK rollover uses Double-Signature scheme, as described in RFC 6781.

Automatic KSK and ZSK rollovers example

Let's start with the following set of keys:

2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, key, tag 53594, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active

The last fields hint the key state: public denotes a key that will be presented as the DNSKEY record, ready means that CDS/CDNSKEY records were created, active tells us that the key is used for signing, while active+ is an active key undergoing a roll-over or roll-in.

For demonstration purposes, the following configuration is used:

submission:
 - id: test_submission
   check-interval: 2s
   parent: dnssec_validating_resolver

policy:
 - id: test_policy
   ksk-lifetime: 5m
   zsk-lifetime: 2m
   propagation-delay: 2s
   dnskey-ttl: 10s
   zone-max-ttl: 15s
   ksk-submission: test_submission

Upon the zone's KSK lifetime expiration, a new KSK is generated and the rollover continues along the lines of RFC 6781#section-4.1.2:

# KSK Rollover (53594 -> 3375)

2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, KSK rollover started
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, next key action, KSK tag 3375, submit at 2024-02-14T15:20:12+0100
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, key, tag 53594, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active+
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111204, new RRSIGs 3
2024-02-14T15:20:00+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:20:12+0100

... (propagation-delay + dnskey-ttl) ...

2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:20:12+0100 notice: [example.com.] DNSSEC, KSK submission, waiting for confirmation
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, key, tag 53594, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, ready, active+
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111205, new RRSIGs 6
2024-02-14T15:20:12+0100 info: [example.com.] DNSSEC, next signing at 2024-02-28T15:19:37+0100

At this point the new KSK has to be submitted to the parent zone. Knot detects the updated parent's DS record automatically (and waits for additional period of the DS's TTL before retiring the old key) if parent DS check is configured, otherwise the operator must confirm it manually (using knotc zone-ksk-submitted)

Note

A DS record for the new KSK can be generated using:

$ keymgr example.com ds 3375

2024-02-14T15:20:12+0100 info: [example.com.] DS check, outgoing, remote 127.0.0.1@5300 TCP, KSK submission check: negative
2024-02-14T15:20:14+0100 info: [example.com.] DS check, outgoing, remote 127.0.0.1@5300 TCP/pool, KSK submission check: negative
2024-02-14T15:20:16+0100 info: [example.com.] DS check, outgoing, remote 127.0.0.1@5300 TCP/pool, KSK submission check: positive
2024-02-14T15:20:16+0100 notice: [example.com.] DNSSEC, KSK submission, confirmed
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, key, tag 53594, algorithm ECDSAP256SHA256, KSK, public, active+
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111206, new RRSIGs 2
2024-02-14T15:20:16+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:20:23+0100

... (parent's DS TTL is 7 seconds) ...

2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, next key action, ZSK, generate at 2024-02-14T15:21:54+0100
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111207, new RRSIGs 2
2024-02-14T15:20:23+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:21:54+0100

Upon the zone's ZSK lifetime expiration, a new ZSK is generated and the rollover continues along the lines of RFC 6781#section-4.1.1:

# ZSK Rollover (36185 -> 38559)

2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, ZSK rollover started
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, next key action, ZSK tag 38559, replace at 2024-02-14T15:22:06+0100
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, key, tag 38559, algorithm ECDSAP256SHA256, public
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111208, new RRSIGs 2
2024-02-14T15:21:54+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:22:06+0100

... (propagation-delay + dnskey-ttl) ...

2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, next key action, ZSK tag 36185, remove at 2024-02-14T15:22:23+0100
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, key, tag 36185, algorithm ECDSAP256SHA256, public
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, key, tag 38559, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111209, new RRSIGs 14
2024-02-14T15:22:06+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:22:23+0100

... (propagation-delay + zone-max-ttl) ...

2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, next key action, ZSK, generate at 2024-02-14T15:24:06+0100
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, key, tag 38559, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111210, new RRSIGs 2
2024-02-14T15:22:23+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:24:06+0100

Further rollovers:

... (zsk-lifetime - propagation-delay - zone-max-ttl) ...

# Another ZSK Rollover (38559 -> 59825)

2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, ZSK rollover started
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, next key action, ZSK tag 59825, replace at 2024-02-14T15:24:18+0100
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, key, tag 38559, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, key, tag 59825, algorithm ECDSAP256SHA256, public
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111211, new RRSIGs 2
2024-02-14T15:24:06+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:24:18+0100

...

# Another KSK Rollover (3375 -> 50822)

2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, signing zone
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, KSK rollover started
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, next key action, KSK tag 50822, submit at 2024-02-14T15:25:12+0100
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, key, tag  3375, algorithm ECDSAP256SHA256, KSK, public, active
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, key, tag 59825, algorithm ECDSAP256SHA256, public, active
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, key, tag 50822, algorithm ECDSAP256SHA256, KSK, public, active+
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, signing started
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, successfully signed, serial 2010111214, new RRSIGs 3
2024-02-14T15:25:00+0100 info: [example.com.] DNSSEC, next signing at 2024-02-14T15:25:12+0100

...

Tip

If systemd is available, the KSK submission event is logged into journald in a structured way. The intended use case is to trigger a user-created script. Example:

journalctl -f -t knotd -o json | python3 -c '
import json, sys
for line in sys.stdin:
  k = json.loads(line);
  if "KEY_SUBMISSION" in k:
    print("%s, zone=%s, keytag=%s" % (k["__REALTIME_TIMESTAMP"], k["ZONE"], k["KEY_SUBMISSION"]))
'

Alternatively, the D-Bus signaling can be utilized for the same use.

DNSSEC shared KSK

Knot DNS allows, with automatic DNSSEC key management, to configure a shared KSK for multiple zones. By enabling ksk-shared, we tell Knot to share all newly-created KSKs among all the zones with the same DNSSEC signing policy assigned.

The feature works as follows. Each zone still manages its keys separately. If a new KSK shall be generated for the zone, it first checks if it can grab another zone's shared KSK instead - that is the last generated KSK in any of the zones with the same policy assigned. Anyway, only the cryptographic material is shared, the key may have different timers in each zone.

Consequences:

If we have an initial setting with brand new zones without any DNSSEC keys, the initial keys for all zones are generated. With shared KSK, they will all have the same KSK, but different ZSKs. The KSK rollovers may take place at slightly different times for each of the zones, but the resulting new KSK will be shared again among all of them.

If we have zones which already have their keys, turning on the shared KSK feature triggers no action. But when a KSK rollover takes place, they will use the same new key afterwards.

Warning

Changing the policy id must be done carefully if shared KSK is in use.

DNSSEC delete algorithm

This is how to "disconnect" a signed zone from a DNSSEC-aware parent zone. More precisely, we tell the parent zone to remove our zone's DS record by publishing a special formatted CDNSKEY and CDS record. This is mostly useful if we want to turn off DNSSEC on our zone so it becomes insecure, but not bogus.

With automatic DNSSEC signing and key management by Knot, this is as easy as configuring cds-cdnskey-publish option and reloading the configuration. We check if the special CDNSKEY and CDS records with the rdata "0 3 0 AA==" and "0 0 0 00", respectively, appeared in the zone.

After the parent zone notices and reflects the change, we wait for TTL expire (so all resolvers' caches get updated), and finally we may do anything with the zone, e.g. turning off DNSSEC, removing all the keys and signatures as desired.

DNSSEC Offline KSK

Knot DNS allows a special mode of operation where the private part of the Key Signing Key is not available to the daemon, but it is rather stored securely in an offline storage. This requires that the KSK/ZSK signing scheme is used (i.e. single-type-signing is off). The Zone Signing Key is always fully available to the daemon in order to sign common changes to the zone contents.

The server (or the "ZSK side") only uses ZSK to sign zone contents and its changes. Before performing a ZSK rollover, the DNSKEY records will be pre-generated and signed by the signer (the "KSK side"). Both sides exchange keys in the form of human-readable messages with the help of the keymgr utility.

Prerequisites

For the ZSK side (i.e. the operator of the DNS server), the zone has to be configured with:

For the KSK side (i.e. the operator of the KSK signer), the zone has to be configured with:

Generating and signing future ZSKs

  1. Use the keymgr pregenerate command on the ZSK side to prepare the ZSKs for a specified period of time in the future. The following example generates ZSKs for the example.com zone for 6 months ahead starting from now:

    $ keymgr -c /path/to/ZSK/side.conf example.com. pregenerate +6mo

    If the time period is selected as e.g. 2 x zsk-lifetime + 4 x propagation-delay, it will prepare roughly two complete future key rollovers. The newly-generated ZSKs remain in non-published state until their rollover starts, i.e. the time they would be generated in case of automatic key management.

  2. Use the keymgr generate-ksr command on the ZSK side to export the public parts of the future ZSKs in a form similar to DNSKEY records. You might use the same time period as in the first step:

    $ keymgr -c /path/to/ZSK/side.conf example.com. generate-ksr +0 +6mo > /path/to/ksr/file

    Save the output of the command (called the Key Signing Request or KSR) to a file and transfer it to the KSK side e.g. via e-mail.

  3. Use the keymgr sign-ksr command on the KSK side with the KSR file from the previous step as a parameter:

    $ keymgr -c /path/to/KSK/side.conf example.com. sign-ksr /path/to/ksr/file > /path/to/skr/file

    This creates all the future forms of the DNSKEY, CDNSKEY and CSK records and all the respective RRSIGs and prints them on output. Save the output of the command (called the Signed Key Response or SKR) to a file and transfer it back to the ZSK side.

  4. Use the keymgr import-skr command to import the records and signatures from the SKR file generated in the last step into the KASP DB on the ZSK side:

    $ keymgr -c /path/to/ZSK/side.conf example.com. import-skr /path/to/skr/file

  5. Use the knotc zone-keys-load command to trigger a zone re-sign on the ZSK-side and set up the future re-signing events correctly.:

    $ knotc -c /path/to/ZSK/side.conf zone-keys-load example.com.

  6. Now the future ZSKs and DNSKEY records with signatures are ready in KASP DB for later usage. Knot automatically uses them at the correct time intervals. The entire procedure must be repeated before the time period selected at the beginning passes, or whenever a configuration is changed significantly. Importing new SKR over some previously-imported one leads to deleting the old offline records.

Offline KSK and manual ZSK management

If the automatically preplanned ZSK roll-overs (first step) are not desired, just set the zsk-lifetime to zero, and manually pregenerate ZSK keys and set their timers. Then follow the steps generate-ksr — sign-ksr — import-skr — zone-keys-load and repeat the ceremony when necessary.

Offline KSK roll-over

The KSKs (on the KSK side) must be managed manually, but manual KSK roll-over is possible. Just plan the steps of the KSK roll-over in advance, and whenever the KSK set or timers are changed, re-perform the relevant rest of the ceremony sign-ksr — import-skr — zone-keys-load.

Emergency SKR

A general recommendation for large deployments is to have some backup pre-published keys, so that if the current ones are compromised, they can be rolled-over to the backup ones without any delay. But in the case of Offline KSK, according to the procedures above, both ZSK and KSK immediate rollovers require the KSR-SKR ceremony.

However, a trick can be done to achieve really immediate key substitution. This is no longer about Knot DNS functionality, just a hint for the operator.

The idea is to perform every KSR-SKR ceremony twice: once with normal state of the keys (the backup key is only published), and once with the keys already exchanged (the backup key is temporarily marked as active and the standard key temporarily as public only). The second (backup) SKR should be saved for emergency key replacement.

Summary of the steps:

  • Prepare KSK and ZSK side as usual, including public-only emergency key

  • Perform normal Offline KSK ceremony:

    • Pre-generate ZSKs (only in the case of automatic ZSK management)

    • Generate KSR

    • Sign KSR on the KSK side

    • Import SKR

    • Re-sign the zone

  • Freeze the zone on the ZSK side

  • Temporarily set the backup key as active and the normal key as publish-only

  • Perform backup Offline KSK ceremony:

    • Generate KSR (only if the backup key is a replacement for ZSK)

    • Sign the KSR on the KSK side

    • Save the SKR to a backup storage, don't import it yet

  • Return the keys to the previous state

  • Thaw the zone on the ZSK side

Emergency key replacement:

  • Import the backup SKR

  • Align the keys with the new states (backup key as active, compromised key as public)

  • Re-sign the zone

DNSSEC multi-signer

Multi-signer is a general term that refers to any mode of operation in which a DNS zone is signed by multiple servers (usually two) in parallel. Knot DNS offers various multi-signer modes, which are recommended for redundancy within an organization. For multi-signer operations involving multiple "DNSSEC providers" and the ability to switch between them, you can also refer to MUSIC.

Regardless of the chosen mode from the following options, any secondary that has multiple signers configured as primaries must prevent interchanged IXFR from them. This can be achieved either by setting master pinning on every secondary or by setting distinct serial-modulo on each signer. It is recommended to combine both approaches. Alternatively, if any of the secondaries is not Knot DNS, provide-ixfr can be disabled on the signers.

In order to prevent keytag conflicts, it is recommended that the keytags of keys generated by each signer are from distinct subset of possible values. With Knot DNS, this can be achieved using keytag-modulo option (e.g. for three signers, setting 0/3 on the first one, 1/3 on the second, and 2/3 on the third of them).

Sharing private keys, manual policy

When DNSSEC keys are shared among zone signing servers (signers), one challenge is automatic key management (roll-overs) and synchronization among the signers. In this example mode of operation, it is expected that key management is maintained outside of Knot, and the generated keys, including private keys and metadata (timers), are available in Bind9 format.

Every new key is then imported into each Knot using the keymgr import-bind command, after which knotc zone-keys-load is invoked. With manual policy configured, the signers simply follow prescribed key timers, maintaining the same key set at each signer. For more useful commands like list, set, and delete, refer to keymgr.

Sharing private keys, automatic policy

Knot handles automatic key management very well, but enabling it on multiple instances would lead to redundant key generation. However, it's possible to enable it on one signer and keep synchronizing the keys to all others. The managing signer shall be configured with automatic ZSK/KSK management, all others with manual policy.

The key set changes on the managing signer can be monitored by periodic queries with keymgr list, or by listening to D-Bus interface and watching for the keys_updated event.

Whenever the key set is changed, key synchronization can be safely performed with Data and metadata backup feature. Dump the KASP database on the managing signer with knotc zone-backup +kaspdb, transfer the backup directory to each other signer, and import the keys by knotc zone-restore +kaspdb, followed by zone-keys-load on them.

This way, the full key set, including private keys and all metadata, is always synchronized between signers. The method of transporting the backup directory is beyond the scope of Knot and this documentation. An eventual loss of the managing signer results in the automatic key management being halted, but DNSSEC signing continues to function. The synchronization delay for keys between the managing signer and other signers must be accounted for in propagation-delay.

Distinct keys, DNSKEY record synchronization

When the DNSSEC keys are not shared among signers, each server can manage its own keys separately. However, the DNSKEY (including CDNSKEY and CDS) records (with public keys) must be synchronized for full validity of the signed zone. Dynamic updates can be used to achieve this sharing.

The following configuration options should be used:

  • Set dnskey-management to incremental on each signer to ensure it retains the other's DNSKEY records in the zone during signing.

  • Set delete-delay to a reasonable time interval, which ensures that all signers get synchronized during this period.

  • Set cds-cdnskey-publish to either none or always, otherwise the parent DS record might configure itself to point only to one signer's KSK.

  • Configure dnskey-sync to all other signers so that this signer's public keys appear in each other's DNSKEY (also applies to CDNSKEY and CDS) RRSets.

  • Configure Access control list (ACL) so that DDNS from all other signers is allowed.

  • Set ddns-master to empty value ("") so that DDNS from other signers is not forwarded to the primary master if any.

  • Additionally, the synchronization delay between all signers must be accounted for in propagation-delay.

With careful configuration, all signers automatically synchronize their DNSKEY (and eventually CDNSKEY and CDS) RRSets, keeping them synchronized during roll-overs. Nevertheless, it is recommended to monitor their logs.

Note

It is highly recommended to use this mode with only two signers. With three or more signers, it often happens that they continuously overwrite each other's DNSKEYs for a long time before settling down. This can be mitigated by configuring dnskey-sync in a cyclic maner, such that they form a cycle (i.e. signer1 synchronizes only to signer2, signer2 to signer3 and so on). However, this in turn leads to a breakage in DNSKEY synchronization whenever any signer goes offline. A practical compromise is carefully configuring the order of each signer's dnskey-sync values in the way that the "cycling" signer is at the first position and the remaining signers follow.

An illustrative example of the second of three signers:

remote:
    - id: signer1
      address: 10.20.30.1
    - id: signer3
      address: 10.20.30.3

 acl:
    - id: signers
      remote: [ signer1, signer3 ]
      action: [ query, update ]
      # TODO configure TSIGs!

dnskey-sync:
    - id: sync
      remote: [ signer3, signer1 ] # the order matters here!

policy:
    - id: multisigner
      single-type-signing: on
      ksk-lifetime: 60d
      ksk-submission: ... # TODO see Automatic KSK management
      propagation-delay: 14h
      delete-delay: 2h
      cds-cdnskey-publish: always
      dnskey-management: incremental
      dnskey-sync: sync

zone:
    - domain: example.com.
      # TODO configure zonefile and journal
      # TODO configure transfers in/out: master, NOTIFY, ACLs...
      dnssec-signing: on
      dnssec-policy: multisigner
      ddns-master: ""
      serial-modulo: 1/3
      acl: signers

Distinct keys, DNSKEY at common unsigned primary

The same approach and configuration can be used, with the difference that the signers do not send updated DNSKEYs (along with CDNSKEYs and CDSs) to each other. Instead, they send the updates to their common primary, which holds the unsigned version of zone. The only configuration change involves redirecting dnskey-sync to the common primary and adjusting its ACL to allow DDNS from the signers.

It is also necessary to configure ixfr-benevolent on each signer so that they accept incremental zone transfers from the primary with additions (or removals) of their own's DNSKEYs.

This setup should work nicely with any number of signers, however, due to the size of DNSKEY RRSet, at most three are advisable.

DNSSEC keys import to HSM

Knot DNS stores DNSSEC keys in textual PEM format (RFC 7468), while many HSM management software require the keys for import to be in binary DER format (Rec. ITU-T X.690). Keys can be converted from one format to another by software tools such as certtool from GnuTLS suite or openssl from OpenSSL suite.

In the examples below, c4eae5dea3ee8c15395680085c515f2ad41941b6 is used as the key ID, c4eae5dea3ee8c15395680085c515f2ad41941b6.pem represents the filename of the key in PEM format as copied from the Knot DNS zone's KASP database directory, c4eae5dea3ee8c15395680085c515f2ad41941b6.priv.der represents the file containing the private key in DER format as generated by the conversion tool, and c4eae5dea3ee8c15395680085c515f2ad41941b6.pub.der represents the file containing the public key in DER format as generated by the conversion tool.

$ certtool -V -k --outder --infile c4eae5dea3ee8c15395680085c515f2ad41941b6.pem \
  --outfile c4eae5dea3ee8c15395680085c515f2ad41941b6.priv.der

$ certtool -V --pubkey-info --outder --load-privkey c4eae5dea3ee8c15395680085c515f2ad41941b6.pem \
  --outfile c4eae5dea3ee8c15395680085c515f2ad41941b6.pub.der

As an alternative, openssl can be used instead. It is necessary to specify either rsa or ec command according to the algorithm used by the key.

$ openssl rsa -outform DER -in c4eae5dea3ee8c15395680085c515f2ad41941b6.pem \
  -out c4eae5dea3ee8c15395680085c515f2ad41941b6.priv.der

$ openssl rsa -outform DER -in c4eae5dea3ee8c15395680085c515f2ad41941b6.pem \
  -out c4eae5dea3ee8c15395680085c515f2ad41941b6.pub.der -pubout

Actual import of keys (both public and private keys from the same key pair) to an HSM can be done via PKCS #11 interface, by pkcs11-tool from OpenSC toolkit for example. In the example below, /usr/local/lib/pkcs11.so is used as a name of the PKCS #11 library or module used for communication with the HSM.

$ pkcs11-tool --module /usr/local/lib/pkcs11.so --login \
  --write-object c4eae5dea3ee8c15395680085c515f2ad41941b6.priv.der --type privkey \
  --usage-sign --id c4eae5dea3ee8c15395680085c515f2ad41941b6

$ pkcs11-tool --module /usr/local/lib/pkcs11.so -login \
  --write-object c4eae5dea3ee8c15395680085c515f2ad41941b6.pub.der --type pubkey \
  --usage-sign --id c4eae5dea3ee8c15395680085c515f2ad41941b6

Daemon controls

Knot DNS was designed to allow server reconfiguration on-the-fly without interrupting its operation. Thus it is possible to change both configuration and zone files and also add or remove zones without restarting the server. This can be done with:

$ knotc reload

If you want to refresh the secondary zones, you can do this with:

$ knotc zone-refresh

Logging

Knot DNS supports logging to syslog or systemd-journald facility, to a specified file, to standard output, or to standard error output. Several different logging targets may be used in parallel.

If syslog or systemd-journald is used for logging, log rotation is handled by that logging facility. When logging to a specified file, log rotation should be done by moving the current log file followed by reopening of the log file with either knotc -b reload or by sending SIGHUP to the knotd process (see the pidfile).

Data and metadata backup

Some of the zone-related data, such as zone contents or DNSSEC signing keys, and metadata, like zone timers, might be worth backing up. For the sake of consistency, it's usually necessary to shut down the server, or at least freeze all the zones, before copying the data like zone files, KASP database, etc, to a backup location. To avoid this necessity, Knot DNS provides a feature to back up some or all of the zones seamlessly.

Online backup

While the server is running and the zones normally loaded (even when they are constantly/frequently being updated), the user can manually trigger the backup by calling:

$ knotc zone-backup +backupdir /path/of/backup

To back up just some of the zones (instead of all), the user might provide their list:

$ knotc zone-backup +backupdir /path/to/backup zone1.com. zone2.com. ...

The backup directory should be empty or non-existing and it must be accessible and writable for the user account under which knotd is running. The backup procedure will begin soon and will happen zone-by-zone (partially in parallel if more background-workers are configured). The user shall check the logs for the outcome of each zone's backup attempt. The knotc's -b parameter might be used if the user desires to wait until the backup work is done and a simple result status is printed out.

Tip

There is a plain ASCII text file in the backup directory, knot_backup.label, that contains some useful information about the backup, such as the backup creation date & time, the server identity, etc. Care must always be taken not to remove this file from the backup nor to damage it.

If a backup fails, the backup directory containing incomplete backup is retained. For repeated backup attempts to the same directory, it must be removed or renamed manually first.

Note

When backing up or restoring a catalog zone, it's necessary to make sure that the contents of the catalog doesn't change during the backup or restore. An easy solution is to use knotc zone-freeze on the catalog zone for the time of backup and restore.

Offline restore

If the Online backup was performed for all zones, it's possible to restore the backed up data by simply copying them to their normal locations, since they're simply copies. For example, the user can copy (overwrite) the backed up KASP database files to their configured location.

This restore of course must be done when the server is stopped. After starting up the server, it should run in the same state as at the time of backup.

This method is recommended in the case of complete data loss, for example physical server failure.

Note

The online backup procedure stores all zone files in a single directory using their default file names. If the original directory layout was different, then the required directory structure must be created manually for offline restore and zone files must be placed individually to their respective directories. If the zone file names don't follow the default pattern, they must be renamed manually to match the configuration. These limitations don't apply to the online restore procedure.

Online restore

This procedure is symmetrical to Online backup. By calling:

$ knotc zone-restore +backupdir /path/of/backup

the user triggers a one-by-one zone restore from the backup on a running server. Again, a subset of zones might be specified. It must be specified if the backup was created for only a subset of zones.

Note

For restore of backups that have been created by Knot DNS releases prior to 3.1, it's necessary to use the -f option. Since this option also turns off some verification checks, it shouldn't be used in other cases.

Note

For QUIC/TLS, only the auto-generated key is restored. The zone-restore command doesn't restore a user-defined QUIC/TLS key and certificate so as to avoid possible configuration management conflicts and they must be restored from the backup (its subdirectory quic) manually. In all cases, restart of the Knot server after the restore is necessary for the restored QUIC/TLS key/certificate to take effect.

Limitations

Neither configuration file nor Configuration database is backed up by zone backup. The configuration has to be synchronized before zone restore is performed!

If the private keys are stored in a HSM (anything using a PKCS#11 interface), they are not backed up. This includes internal metadata of the PKCS#11 provider software, such as key mappings, authentication information, and the configuration of the provider. Details are vendor-specific.

The restore procedure does not care for keys deleted after taking the snapshot. Thus, after restore, there might remain some redundant .pem files of obsolete signing keys.

Tip

In order to seamlessly deploy a restored backup of KASP DB with respect to a possibly ongoing DNSSEC key rollover, it's recommended to set propagation-delay to the sum of:

  • The maximum delay between beginning of the zone signing and publishing re-signed zone on all public secondary servers.

  • How long it takes for the backup server to start up with the restored data.

  • The period between taking backup snapshots of the live environment.

Statistics

The server provides some general statistics and optional query module statistics (see mod-stats).

Server statistics or global module statistics can be shown by:

$ knotc stats
$ knotc stats server             # Show all server counters
$ knotc stats mod-stats          # Show all mod-stats counters
$ knotc stats server.zone-count  # Show specific server counter

Per zone statistics can be shown by:

$ knotc zone-stats example.com.                       # Show all zone counters
$ knotc zone-stats example.com. mod-stats             # Show all zone mod-stats counters
$ knotc zone-stats example.com. mod-stats.query-type  # Show specific zone counter
$ knotc zone-stats --                                 # Show all zone counters for all zones
$ knotc zone-stats -- mod-stats.request-protocol      # Show specific zone counter for all zones

To show all supported counters even with 0 value, use the force option.

A simple periodic statistic dump to a YAML file can also be enabled. See statistics section for the configuration details.

As the statistics data can be accessed over the server control socket, it is possible to create an arbitrary script (Python is supported at the moment) which could, for example, publish the data in JSON format via HTTP(S) or upload the data to a more efficient time series database. Take a look into the python folder of the project for these scripts.

Mode XDP

Thanks to recent Linux kernel capabilities, namely eXpress Data Path and AF_XDP address family, Knot DNS offers a high-performance DNS over UDP packet processing mode. The basic idea is to filter DNS messages close to the network device and effectively forward them to the nameserver without touching the network stack of the operating system. Other messages (including DNS over TCP) are processed as usual.

If listen is configured, the server creates additional XDP workers, listening on specified interface(s) and port(s) for DNS over UDP queries. Each XDP worker handles one RX and TX network queue pair.

Pre-requisites

  • Linux kernel 4.18+ (5.x+ is recommended for optimal performance) compiled with the CONFIG_XDP_SOCKETS=y option. The XDP mode isn't supported in other operating systems.

  • A multiqueue network card, which offers enough Combined RX/TX channels, with native XDP support is highly recommended. Successfully tested cards:

    • NVIDIA (Mellanox) ConnectX-6 Dx (driver mlx5_core), maximum number of channels per interface is 63. Official drivers are recommended.

    • Intel series 700 (driver i40e), maximum number of channels per interface is 64. Linux kernel drivers are recommended.

    Cards with known instability issues:

    • Intel series E810 (driver ice).

    • Intel series 500 (driver ixgbe).

  • If the knotd service is not directly executed in the privileged mode, some additional Linux capabilities have to be set:

    Execute command:

    systemctl edit knot

    And insert these lines:

    [Service]
CapabilityBoundingSet=CAP_NET_RAW CAP_NET_ADMIN CAP_SYS_ADMIN CAP_IPC_LOCK CAP_SYS_RESOURCE
AmbientCapabilities=CAP_NET_RAW CAP_NET_ADMIN CAP_SYS_ADMIN CAP_IPC_LOCK CAP_SYS_RESOURCE

    The CAP_SYS_RESOURCE is needed on Linux < 5.11.

    All the capabilities are dropped upon the service is started.

  • For proper processing of VLAN traffic, VLAN offloading should be disabled. E.g.:

    ethtool -K <interface> tx-vlan-offload off rx-vlan-offload off

Optimizations

Some helpful commands:

ethtool -N <interface> rx-flow-hash udp4 sdfn
ethtool -N <interface> rx-flow-hash udp6 sdfn
ethtool -L <interface> combined <?>
ethtool -G <interface> rx <?> tx <?>
renice -n 19 -p $(pgrep '^ksoftirqd/[0-9]*$')

Limitations

  • Request and its response must go through the same physical network device.

  • Dynamic DNS over XDP is not supported.

  • MTU higher than 1790 bytes is not supported.

  • Multiple BPF filters per one network device are not supported.

  • Systems with big-endian byte ordering require special recompilation of the nameserver.

  • IPv4 header and UDP checksums are not verified on received DNS messages.

  • DNS over XDP traffic is not visible to common system tools (e.g. firewall, tcpdump etc.).

  • BPF filter is not automatically unloaded from the network device. Manual filter unload:

    ip link set dev <interface> xdp off