| Internet-Draft | dns-catalog-zones | October 2022 |
| van Dijk, et al. | Expires 22 April 2023 | [Page] |
This document describes a method for automatic DNS zone provisioning among DNS primary and secondary nameservers by storing and transferring the catalog of zones to be provisioned as one or more regular DNS zones.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 22 April 2023.¶
Copyright (c) 2022 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The content of a DNS zone is synchronized amongst its primary and secondary nameservers using AXFR and IXFR. However, the list of zones served by the primary (called a catalog in [RFC1035]) is not automatically synchronized with the secondaries. To add or remove a zone, the administrator of a DNS nameserver farm not only has to add or remove the zone from the primary, they must also add/remove the zone from all secondaries, either manually or via an external application. This can be both inconvenient and error-prone; it is also dependent on the nameserver implementation.¶
This document describes a method in which the catalog is represented as a regular DNS zone (called a "catalog zone" here), and transferred using DNS zone transfers. As zones are added to or removed from the catalog zone, these changes are distributed to the secondary nameservers in the normal way. The secondary nameservers then add/remove/modify the zones they serve in accordance with the changes to the catalog zone. Other use-cases of nameserver remote configuration by catalog zones are possible, where the catalog consumer might not be a secondary.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119][RFC8174] when, and only when, they appear in all capitals, as shown here.¶
$CATZ$OLDCATZ and $NEWCATZ are used to discuss migration of a member zone from one catalog zone $OLDCATZ to another catalog zone $NEWCATZ.¶
A catalog zone is a DNS zone whose contents are specially crafted. Its records primarily constitute a list of PTR records referencing other DNS zones (so-called "member zones"). The catalog zone may contain other records indicating additional metadata (so-called "properties") associated with these member zones.¶
Catalog consumers MUST ignore any RR in the catalog zone which is meaningless or useless to the implementation.¶
Authoritative servers may be pre-configured with multiple catalog zones, each associated with a different set of configurations.¶
Although the contents of a catalog zone are interpreted and acted upon by nameservers, a catalog zone is a regular DNS zone and so must adhere to the standards for such zones.¶
A catalog zone is primarily intended for the management of a farm of authoritative nameservers. The content of catalog zones may not be accessible from any recursive nameserver.¶
As with any other DNS zone, a catalog zone MUST have a SOA record and at least one NS record at its apex.¶
The SOA record's SERIAL, REFRESH, RETRY and EXPIRE fields [RFC1035] are used during zone transfer. A catalog zone's SOA SERIAL field MUST increase when an update is made to the catalog zone's contents as per serial number arithmetic defined in [RFC1982]. Otherwise, catalog consumers might not notice updates to the catalog zone's contents.¶
There is no requirement to be able to query the catalog zone via recursive nameservers. However, at least one NS RR is still required so that catalog zone is a syntactically correct DNS zone. A single NS RR with a NSDNAME field containing the absolute name "invalid." is RECOMMENDED [RFC2606][RFC6761].¶
The list of member zones is specified as a collection of member nodes, represented by domain names under the owner name "zones" where "zones" is a direct child domain of the catalog zone.¶
The names of member zones are represented on the RDATA side (instead of as a part of owner names) of a PTR record, so that all valid domain names may be represented regardless of their length [RFC1035]. This PTR record MUST be the only record in the PTR RRset with the same name. More than one record in the RRset denotes a broken catalog zone which MUST NOT be processed (see Section 5.1).¶
For example, if a catalog zone lists three zones "example.com.", "example.net." and "example.org.", the member node RRs would appear as follows:¶
<unique-1>.zones.$CATZ 0 IN PTR example.com. <unique-2>.zones.$CATZ 0 IN PTR example.net. <unique-3>.zones.$CATZ 0 IN PTR example.org.¶
where <unique-N> is a label that tags each record in the collection.
<unique-N> has an unique value in the collection.
When different <unique-N> labels hold the same PTR value (i.e. point to the same member zone), the catalog zone is broken and MUST NOT be processed (see Section 5.1).¶
Member node labels carry no informational meaning beyond labeling member zones. A changed label may indicate that the state for a zone needs to be reset (see Section 5.6).¶
Having the zones uniquely tagged with the <unique-N> label ensures that additional RRs can be added below the member node (see Section 4.3).¶
The CLASS field of every RR in a catalog zone MUST be IN (1).¶
The TTL field's value is not defined by this memo. Catalog zones are for authoritative nameserver management only and are not intended for general querying via recursive resolvers.¶
Catalog zone information is stored in the form of "properties".¶
Properties are identified by their name, which is used as an owner name prefix for one or more record sets underneath a member node, with type(s) as appropriate for the respective property.¶
Known properties with the correct RR type, but which are for some reason invalid (for example because of an impossible value or because of an illegal number of RRs in the RRset), denote a broken catalog zone which MUST NOT be processed (see Section 5.1).¶
This specification defines a number of so-called properties, as well as a mechanism to allow implementers to store additional information in the catalog zone with Custom properties, see Section 4.5. The meaning of such custom properties is determined by the implementation in question.¶
Some properties are defined at the global level; others are scoped to apply only to a specific member zone. This document defines a single mandatory global property in Section 4.3.1. Member-specific properties are described in Section 4.3.¶
More properties may be defined in future documents.¶
version property)
The catalog zone schema version is specified by an integer value embedded in a TXT RR named version.$CATZ.
All catalog zones MUST have a TXT RRset named version.$CATZ with exactly one RR.¶
Catalog consumers MUST NOT apply catalog zone processing to¶
version property¶
version property with more than one RR in the RRset¶
version property without an expected value in the
version.$CATZ TXT RR¶
These conditions signify a broken catalog zone which MUST NOT be processed (see Section 5.1).¶
For this memo, the value of the version.$CATZ TXT RR MUST be set to "2", i.e.:¶
version.$CATZ 0 IN TXT "2"¶
NB: Version 1 was used in a draft version of this memo and reflected the implementation first found in BIND 9.11.¶
Each member zone MAY have one or more additional properties, described in this chapter. The member properties described in this document are all optional and implementations MAY choose to implement one, all or none of them. Member zone properties are represented by RRsets below the corresponding member node.¶
coo property)
The coo property facilitates controlled migration of a member zone from one catalog to another.¶
A Change Of Ownership is signaled by the coo property in the catalog zone currently "owning" the zone.
The name of the new catalog is the value of a PTR record in the relevant coo property in the old catalog.
For example if member "example.com." will migrate from catalog zone $OLDCATZ to catalog zone $NEWCATZ, this appears in the $OLDCATZ catalog zone as follows:¶
<unique-N>.zones.$OLDCATZ 0 IN PTR example.com. coo.<unique-N>.zones.$OLDCATZ 0 IN PTR $NEWCATZ¶
The PTR RRset MUST consist of a single PTR record. More than one record in the RRset denotes a broken catalog zone which MUST NOT be processed (see Section 5.1).¶
When a consumer of a catalog zone $OLDCATZ receives an update which adds or changes a coo property for a member zone in $OLDCATZ, it does not migrate the member zone immediately.
The migration has to wait for an update of $NEWCATZ. in which the member zone is present. The consumer MUST verify, before the actual migration, that coo property pointing to $NEWCATZ is still present in $OLDCATZ.¶
Unless the member node label (i.e. <unique-N>) for the member is the same in $NEWCATZ, all its associated state for a just migrated zone MUST be reset (see Section 5.6).
Note that the owner of $OLDCATZ allows for the zone associated state to be taken over by the owner of $NEWCATZ by default.
To prevent the takeover of state, the owner of $OLDCATZ must remove this state by updating the associated properties or by performing a zone state reset (see Section 5.6) before or simultaneous with adding the coo property. (see also Section 7)¶
The old owner may remove the member zone containing the coo property from $OLDCATZ once it has been established that all its consumers have processed the Change of Ownership.¶
group property)
With a group property, consumer(s) can be signaled to treat some member zones within the catalog zone differently.¶
The consumer MAY apply different configuration options when processing member zones, based on the value of the group property.
The exact handling of configuration referred to by the group property value is left to the consumer's implementation and configuration.
The property is defined by a TXT record in the sub-node labeled group.¶
The producer MAY assign a group property to all, some, or none of the member zones within a catalog zone.
The producer MAY assign more than one group property to one member zone. This will make it possible to transfer group information for different consumer operators in a single catalog zone.
Consumer operators SHOULD namespace their group properties to limit risk of clashes.¶
The consumer MUST ignore group property values it does not understand.¶
When a consumer sees multiple values in a group property of a single member
zone that it does understand, it MAY choose to process multiple, any one or
none of them.
This is left to the implementation.¶
<unique-1>.zones.$CATZ 0 IN PTR example.com. group.<unique-1>.zones.$CATZ 0 IN TXT nodnssec <unique-2>.zones.$CATZ 0 IN PTR example.net. group.<unique-2>.zones.$CATZ 0 IN TXT operator-x-sign-with-nsec3 group.<unique-2>.zones.$CATZ 0 IN TXT operator-y-nsec3¶
The catalog zone (snippet) above is an example where the producer signals how the consumer(s) shall treat DNSSEC for the zones "example.net." and "example.com."¶
For "example.com.", the consumer might be implemented and configured in the way that the member zone will not be signed with DNSSEC. For "example.net.", the consumers, at two different operators, might be implemented and configured in the way that the member zone will be signed with a NSEC3 chain.¶
*.ext properties)
Implementations and operators of catalog zones may choose to provide their own properties.
Custom properties can occur both globally, or for a specific member zone.
To prevent a name clash with future properties, such properties MUST be represented below the label ext.¶
ext is not a placeholder. A custom property is named as follows:¶
; a global custom property: <your-property>.ext.$CATZ ; a member zone custom property: <your-property>.ext.<unique-N>.zones.$CATZ¶
<your-property> may consist of one or more labels.¶
Implementations SHOULD namespace their custom properties to limit risk of clashes with other implementations of catalog zones.
For example by including the name of the implementation in the property, e.g. like: <property-name>.<implementation-name>.ext.$CATZ.¶
Implementations MAY use such properties on the member zone level to store additional information about member zones, for example to flag them for specific treatment.¶
Further, implementations MAY use custom properties on the global level to store additional information about the catalog zone itself.
While there may be many use cases for this, a plausible one is to store default values for custom properties on the global level,
then overriding them using a property of the same name on the member level (= under the ext label of the member node) if so desired.
A property description should clearly say what semantics apply, and whether a property is global, member, or both.¶
The meaning of the custom properties described in this section is determined by the implementation alone, without expectation of interoperability.¶
As it is a regular DNS zone, a catalog zone can be transferred using DNS zone transfers among nameservers.¶
Catalog updates should be automatic, i.e., when a nameserver that supports catalog zones completes a zone transfer for a catalog zone, it SHOULD apply changes to the catalog within the running nameserver automatically without any manual intervention.¶
Nameservers MAY allow loading and transfer of broken zones with incorrect catalog zone syntax (as they are treated as regular zones). The reason a catalog zone is considered broken SHOULD be communicated clearly to the operator (e.g. through a log message).¶
When a previously correct catalog zone becomes a broken catalog zone, because of an update through an incremental transfer or otherwise, it loses its catalog meaning. No special processing occurs. Member zones previously configured by this catalog MUST NOT be removed or reconfigured in any way.¶
If a name server restarts with a broken catalog zone, the broken catalog SHOULD NOT prevent the name server from starting up and serving the member zones in the last valid version of the catalog zone.¶
Processing of a broken catalog SHALL start (or resume) when the catalog turns into a correct catalog zone, for example by an additional update (through zone transfer or updates) fixing the catalog zone.¶
Similarly, when a catalog zone expires, it loses its catalog meaning and MUST no longer be processed as such. No special processing occurs until the zone becomes fresh again.¶
If there is a clash between an existing zone's name (either from an existing member zone or otherwise configured zone) and an incoming member zone's name (via transfer or update), the new instance of the zone MUST be ignored and an error SHOULD be logged.¶
A clash between an existing member zone's name and an incoming member zone's name (via transfer or update), may be an attempt to migrate a zone to a different catalog, but should not be treated as one except as described in Section 4.4.1.¶
When a member zone is removed from a specific catalog zone, an authoritative server MUST NOT remove the zone and associated state data if the zone was not configured from that specific catalog zone. Only when the zone was configured from a specific catalog zone, and the zone is removed as a member from that specific catalog zone, the zone and associated state (such as zone data and DNSSEC keys) MUST be removed.¶
When via a single update or transfer, the member node's label value (<unique-N>) changes, catalog consumers MUST process this as a member zone removal including all the zone's associated state (as described in Section 5.3), immediately followed by processing the member as a newly to be configured zone in the same catalog.¶
If all consumers of the catalog zones involved support the coo property, it is RECOMMENDED to perform migration of a member zone by following the procedure described in Section 4.4.1.
Otherwise a migration of a member zone from a catalog zone $OLDCATZ to a catalog zone $NEWCATZ has to be done by: first removing the member zone from $OLDCATZ; second adding the member zone to $NEWCATZ.¶
If in the process of a migration some consumers of the involved catalog zones did not catch the removal of the member zone from $OLDCATZ yet (because of a lost packet or down time or otherwise), but did already see the update of $NEWCATZ, they may consider the update adding the member zone in $NEWCATZ to be a name clash (see Section 5.2) and as a consequence the member is not migrated to $NEWCATZ.
This possibility needs to be anticipated with a member zone migration.
Recovery from such a situation is out of the scope of this document.
It may for example entail a manually forced retransfer of $NEWCATZ to consumers after they have been detected to have received and processed the removal of the member zone from $OLDCATZ.¶
It may be desirable to reset state (such as zone data and DNSSEC keys) associated with a member zone.¶
A zone state reset may be performed by a change of the member node's name (see Section 5.4).¶
Although any valid domain name can be used for the catalog name $CATZ, it is RECOMMENDED to use either a domain name owned by the catalog producer, or to use a name under a suitable Special-Use Domain Name [RFC6761].¶
Catalog zones on secondary nameservers would have to be set up manually, perhaps as static configuration, similar to how ordinary DNS zones are configured when catalog zones or another automatic configuration mechanism is not in place. The secondary additionally needs to be configured as a catalog consumer for the catalog zone to enable processing of the member zones in the catalog, such as automatic synchronization of the member zones for secondary service.¶
Operators of catalog consumers should note that secondary name servers may receive DNS NOTIFY messages [RFC1996] for zones before they are seen as newly added member zones to the catalog from which that secondary is provisioned.¶
Although they are regular DNS zones, catalog zones contain only information for the management of a set of authoritative nameservers. For this reason, operators may want to limit the systems able to query these zones.¶
Querying/serving catalog zone contents may be inconvenient via DNS due to the nature of their representation. An administrator may therefore want to use a different method for looking at data inside the catalog zone. Typical queries might include dumping the list of member zones, dumping a member zone's effective configuration, querying a specific property value of a member zone, etc. Because of the structure of catalog zones, it may not be possible to perform these queries intuitively, or in some cases, at all, using DNS QUERY. For example, it is not possible to enumerate the contents of a multi-valued property (such as the list of member zones) with a single QUERY. Implementations are therefore advised to provide a tool that uses either the output of AXFR or an out-of-band method to perform queries on catalog zones.¶
Great power comes with great responsibility: Catalog zones simplify zone provisioning by orchestrating zones on secondary name servers from a single data source - the catalog. Hence, the catalog producer has great power and changes must be treated carefully. For example if the catalog is generated by some script and this script for whatever reason generates an empty catalog, millions of member zones may get deleted from their secondaries within seconds and all the affected domains may be offline in a blink.¶
As catalog zones are transmitted using DNS zone transfers, it is RECOMMENDED that catalog zone transfer are protected from unexpected modifications by way of authentication, for example by using TSIG [RFC8945], or Strict or Mutual TLS authentication with DNS Zone transfer over TLS [RFC9103].¶
Use of DNS UPDATE [RFC2136] to modify the content of catalog zones SHOULD similarly be authenticated.¶
Zone transfers of member zones SHOULD similarly be authenticated. TSIG shared secrets used for member zones SHOULD NOT be mentioned in the catalog zone data. However, key identifiers may be shared within catalog zones.¶
Catalog zones reveal the zones served by the consumers of the catalog zone. It is RECOMMENDED to limit the systems able to query these zones. It is RECOMMENDED to transfer catalog zones confidentially [RFC9103].¶
As with regular zones, primary and secondary nameservers for a catalog zone may be operated by different administrators. The secondary nameservers may be configured as a catalog consumer to synchronize catalog zones from the primary, but the primary's administrators may not have any administrative access to the secondaries.¶
Administrative control over what zones are served from the configured name servers shifts completely from the server operator (consumer) to the "owner" (producer) of the catalog zone content.¶
With migration of member zones between catalogs using the coo property, it is possible for the owner of the target catalog (i.e. $NEWCATZ) to take over all its associated state with the zone from the original owner (i.e. $OLDCATZ) by maintaining the same member node label (i.e. <unique-N>).
To prevent the takeover of the zone associated state, the original owner has to enforce a zone state reset by changing the member node label (see Section 5.6) before or simultaneously with adding the coo property.¶
Our deepest thanks and appreciation go to Stephen Morris, Ray Bellis and Witold Krecicki who initiated this draft and did the bulk of the work.¶
Catalog zones originated as the chosen method among various proposals that were evaluated at ISC for easy zone management. The chosen method of storing the catalog as a regular DNS zone was proposed by Stephen Morris.¶
The initial authors discovered that Paul Vixie's earlier [Metazones] proposal implemented a similar approach and reviewed it. Catalog zones borrow some syntax ideas from Metazones, as both share this scheme of representing the catalog as a regular DNS zone.¶
Thanks to Leo Vandewoestijne. Leo's presentation in the DNS devroom at the FOSDEM'20 [FOSDEM20] was one of the motivations to take up and continue the effort of standardizing catalog zones.¶
Thanks to Brian Conry, Klaus Darilion, Brian Dickson, Tony Finch, Evan Hunt, Shane Kerr, Patrik Lundin, Matthijs Mekking, Victoria Risk, Josh Soref, Petr Spacek, Michael StJohns, Carsten Strotmann and Tim Wicinski for reviewing draft proposals and offering comments and suggestions.¶
Note to the RFC Editor: please remove this entire appendix before publication.¶
In the following implementation status descriptions, "DNS Catalog Zones" refers to DNS Catalog Zones as described in this document.¶
Knot DNS 3.1 (released August 2, 2021) supports both producing and consuming of catalog zones, including the group property.¶
PowerDNS from version 4.7 (released October 3, 2022) supports both producing and consuming of catalog zones.¶
Proof of concept python scripts that can be used for both generating and consuming DNS Catalog Zones with NSD have been developed during the hackathon at the IETF-109.¶
BIND 9.18.3+ supports version 2 catalog zones as described in this document¶
Interoperability between the above implementations has been tested during the hackathon at the IETF-109.¶
Note to the RFC Editor: please remove this entire appendix before publication.¶
Initial public draft.¶
Added Witold, Ray as authors. Fixed typos, consistency issues. Fixed references. Updated Area. Removed newly introduced custom RR TYPEs. Changed schema version to 1. Changed TSIG requirement from MUST to SHOULD. Removed restrictive language about use of DNS QUERY. When zones are introduced into a catalog zone, a primary SHOULD first make the new zones available for transfers first (instead of MUST). Updated examples, esp. use IPv6 in examples per Fred Baker. Add catalog zone example.¶
Addressed some review comments by Patrik Lundin.¶
Revision bump.¶
Reordering of sections into more logical order. Separation of multi-valued properties into their own category.¶
New authors to pickup the editor pen on this draft¶
Remove data type definitions for zone properties Removing configuration of member zones through zone properties altogether¶
Remove Open issues and discussion Appendix, which was about zone options (including primary/secondary relationships) only.¶
Added a new section "The Serial Property", introducing a new mechanism which can help with disseminating zones from the primary to the secondary nameservers in a timely fashion more reliably.¶
Three different ways to provide a "serial" property with a member zone are offered to or the workgroup for discussion.¶
Added a new section "Implementation Status", listing production ready, upcoming and Proof of Concept implementations, and reporting on interoperability of the different implementations.¶
Adding the
cooproperty for zone migration in a controlled fashion¶Adding the
groupproperty for reconfigure settings of member zones in an atomic update¶Adding the
epochproperty to reset zone associated state in a controlled fashion¶
Big cleanup!¶
Introducing the terms catalog consumer and catalog producer¶
Reorganized topics to create a more coherent whole¶
Properties all have consistent format now¶
Try to assume the least possible from implementations w.r.t.:¶
1) Predictability of the <unique-N> IDs of member zones¶
2) Whether or not fallback catalog zones can be found for a member¶
3) Whether or not a catalog consumer can maintain state¶
Move Implementation status to appendix¶
Miscellaneous textual improvements¶
cooproperty points to$NEWCATZ(and notzones.$NEWCATZ)¶Remove suggestion to increase serial and remove member zone from
$OLDCATZafter migration¶More consistent usage of the terms catalog consumer and catalog producer throughout the document¶
Better (safer) description of resetting refresh timers of member zones with the
serialproperty¶Removing a member MUST remove zone associated state¶
Make authentication requirements a bit less prescriptive in security considerations¶
Updated implementation status for KnotDNS¶
Describe member node name changes and update "Zone associated state reset" to use that as the mechanism for it.¶
Add Peter Thomassen as co-author¶
Complete removal of the
epochproperty. We consider consumer optimizations with predictable member node labels (for example based on a hash) out of the scope of this document.¶Miscellaneous editorial improvements¶
Add Kees Monshouwer as co-author¶
Removed the "serial" property¶
Allow custom properties on the global level¶
Move administrative control explanation to Security Considerations¶
Move comment on query methods to Implementation Notes¶
Clarify what happens on expiry¶
Clarify catalog consumer behavior when MUST condition is violated¶
Better text on ordering of operations for Change of Ownership¶
Suggest to namespace custom properties¶
Clarify how to handle property record with wrong type¶
Cover the case of multiple different <unique-N>'s having the same value¶
Recommendations for naming catalog zones¶
Add and operational note about notifies for not yet existing zones¶
Add text about name server restarts with broken zones¶
Great power comes with great responsibility (Thanks Klaus!)¶
Mention the new BIND implementation¶
All invalid properties cause a broken catalog zone, including invalid
groupandversionproperties.¶Add Aram Sargsyan as author (he did the BIND9 implementation)¶
groupproperties can have more than one value¶
Some spelling fixes from Tim Wicinski and Josh Soref¶
Replace SHOULDs with MUSTs for ignoring things that are meaningless to a catalog consumer (Thanks Michael StJohns)¶
Update the list of people to thank in the Acknowledgements section¶
Mention PowerDNS support of catalog zones from version 4.7.0 onwards¶