| Internet-Draft | Generalized Notifications | July 2024 |
| Stenstam, et al. | Expires 9 January 2025 | [Page] |
This document extends the use of DNS NOTIFY [RFC1996] beyond conventional zone transfer hints, bringing the benefits of ad-hoc notifications to DNS delegation maintenance in general. Use cases include DNSSEC key rollovers hints, and quicker changes to a delegation's NS record set.¶
To enable this functionality, a method for discovering the receiver endpoint for such notification message is introduced, via the new NOTIFY record type.¶
TO BE REMOVED: This document is being collaborated on in Github at: https://github.com/peterthomassen/draft-ietf-dnsop-generalized-notify. The most recent working version of the document, open issues, etc. should all be available there. The authors (gratefully) accept pull requests.¶
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 9 January 2025.¶
Copyright (c) 2024 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.¶
Traditional DNS notifications [RFC1996], which are here referred to as "NOTIFY(SOA)", are sent from a primary server to a secondary server to minimize the latter's convergence time to a new version of the zone. This mechanism successfully addresses a significant inefficiency in the original protocol.¶
Today similar inefficiencies occur in new use cases, in particular delegation maintenance (DS and NS record updates). Just as in the NOTIFY(SOA) case, a new set of notification types will have a major positive benefit by allowing the DNS infrastructure to completely sidestep these inefficiencies. For additional context, see Appendix A.¶
No DNS protocol changes are introduced by this document. The mechanism instead makes use of a wider range of DNS messages allowed by the protocol. Future extension for further use cases (such as multi-signer key exchange) is possible.¶
Readers are expected to be familiar with DNSSEC, including [RFC4033], [RFC4034], [RFC4035], [RFC6781], [RFC7344], [RFC7477], [RFC7583], and [RFC8901].¶
When the parent is interested in notifications for delegation maintenance (such as for DS or NS updates), a service will need to be made available for accepting these notifications. Depending on the context, this service may be run by the parent zone operator themselves, or by a designated entity who is in charge of handling the domain's delegation data (such as a domain registrar).¶
It seems desirable to minimize the number of steps that the notification sender needs to figure out where to send the NOTIFY. This suggests that the lookup process be ignorant of the details of the parent-side relationships (e.g., whether there is a registrar or not). This is addressed by parameterizing the lookup with the name of the child. The parent may then (optionally) announce the notification endpoint in a delegation-specific way, that is, at a child-specific name. (A catch-all endpoint may be indicated by wildcarding.)¶
The solution proposed here is thus for the parent to publish the address where it listens for notifications, in a child-specific way (see Section 3.1). Potential senders, knowing the name of the parent zone, can then simply look up that information (see Section 4.1).¶
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.¶
The DSYNC RDATA wire format is encoded as follows:¶
1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| RRtype | Scheme | Port
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Target ... /
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-/
¶
The type of generalized NOTIFY that this DSYNC RR defines the desired target address for (see "Resource Record (RR) TYPEs" IANA registry). For now, only CDS and CSYNC are supported values, with the former indicating an updated CDS or CDNSKEY record set.¶
The scheme indicates the mode used for locating the desired notification address. This is an 8 bit unsigned integer. Records with value 0 (null scheme) are ignored by consumers. Value 1 is described in this document, and values 128-255 are reserved for private use. All other values are currently unspecified.¶
The port on the target host of the notification service. This is a 16 bit unsigned integer in network byte order.¶
The fully-qualified, uncompressed domain name of the target host providing the service of listening for generalized notifications of the specified type. This name MUST resolve to one or more address records.¶
For now, the only scheme defined is scheme=1 with the interpretation that when a new CDS/CDNSKEY (or CSYNC) is published, a NOTIFY(CDS) (or NOTIFY(CSYNC)) should be sent to the address and port listed in the corresponding NOTIFY RRset.¶
Example (for the owner names of these records, see Section 3.1):¶
IN DSYNC CDS 1 5359 cds-scanner.example.net. IN DSYNC CSYNC 1 5360 csync-scanner.example.net.¶
Should a need for other mechanisms arise, other schemes may be defined to deal with such requirements using alternative logic.¶
(RFC Editor: This subsection is to be removed before publication)¶
It may look like it's possible to store the same information in an SRV record. However, this would require indicating the RRtype via a label in the owner name, leading to name space pollution. It would also require changing the semantics of one of the integer fields of the SRV record.¶
Such overloading has not been a good idea in the past. Furthermore, as the generalized notifications are a new proposal with no prior deployments, there is an opportunity to avoid repeating mistakes.¶
The DSYNC record type also provides a cleaner solution for bundling all the new types of notification signaling in one RRset, like:¶
IN DSYNC CDS 1 59 scanner.example.net. IN DSYNC CSYNC 1 59 scanner.example.net.¶
For DSYNC records indicating CDS/CDNSKEY/CSYNC notification targets, no special processing needs to be applied by the authoritative nameserver upon insertion of a DSYNC record. The nameserver can thus be "unaware".¶
Future use cases (such as for multi-signer key exchange) may require the nameserver to trigger special operations, for example when a DSYNC record is inserted during onboarding of a new signer. It seems cleaner and easier that such processing be associated with the insertion of a record of a new type, not an existing type like SRV.¶
To use generalized notifications, it is necessary for the sender to know where to direct each NOTIFY message. This section describes the procedure for discovering that notification target.¶
Note that generalized NOTIFY messages are but one mechanism for improving the efficiency of automated delegation maintenance. Other alternatives alternatives, such as contacting the parent via an API or DNS Update ([RFC2136]), may (or may not) be more suitable in individual cases. Like generalized notifications, they similarly require a means for discovering where to send the API or DNS Update requests.¶
The scope for the publication mechanism is therefore wider than only to support generalized notifications, and a unified approach that works independently of the notification method is specified in this section.¶
Parents participating in the discovery scheme for the purpose of delegation maintenance notifications MUST publish endpoint information using the record type defined in Section 2. A parent MUST NOT publish more than one DSYNC record for each combination of rrtype and scheme.¶
If the parent itself performs CDS/CDNSKEY or CSYNC processing, or if the parent forwards the notifications internally to the designated party (such as as registrar), the following scheme is used:¶
*._dsync.example. IN DSYNC CDS scheme port target *._dsync.example. IN DSYNC CSYNC scheme port target¶
It is RECOMMENDED to secure the corresponding zone with DNSSEC.¶
It is also possible to publish child-specific records, where the wildcard label is replaced by the child's FQDN with the parent zone's labels stripped.¶
As an example, consider a registrar offering domains like
child.example, delegated from example zone. If the registrar
provides the notification endpoint, e.g., rr-endpoint.example:5300,
the parent may publish this information as follows:¶
child._dsync.example. IN DSYNC CDS 1 5300 rr-endpoint.example.¶
In case the parent does not need child-specificity, the wildcard label
may be dropped from the DSYNC owner name (i.e., it may be published at
the _dsync label instead). While this practice enables zone structures
without wildcards, it also requires an additional step during discovery
(see Section 4.1), and is therefore NOT RECOMMENDED.¶
For practical purposes, the parent MAY delegate the _dsync domain as a
separate zone, and/or synthesize records under it. If child-specificity
is not needed, the parent can publish a wildcard DSYNC record.¶
To accommodate indirect delegation management models, the parent's designated notification target may relay notifications to a third party (such as the registrar, in ICANN's model). The details of such arrangements are out of scope for this document.¶
Delegation maintenance notifications address the inefficiencies related to scanning child zones for CDS/CDNSKEY records [RFC7344]. (For an overview of the issues, see Appendix A.)¶
Delegation maintenance NOTIFY messages MUST be formatted as described in
[RFC1996], with the qtype field replaced as appropriate.¶
To address the CDS/CDNSKEY dichotomy, the NOTIFY(CDS) message (with
qtype=CDS) is defined to indicate any child-side changes pertaining
to an upcoming update of DS records. Upon receipt of NOTIFY(CDS), the
recipient (the parent registry or a registrar) SHOULD initiate the same DNS
lookups and verifications that would otherwise be triggered based on a
timer.¶
The CSYNC [RFC7477] inefficiency may be similarly treated, with the
child sending a NOTIFY(CSYNC) message (with qtype=CSYNC) to an address
where the parent (or a registrar) is listening to CSYNC notifications.¶
In both cases the notification will speed up processing times by providing the recipient with a hint that a particular child zone has published new CDS, CDNSKEY and/or CSYNC records.¶
To locate the target for outgoing delegation maintenance notifications, the notification sender MUST perform the following procedure:¶
Construct the lookup name, by injecting the _dsync label after the
first label of the delegation owner name.¶
Perform a lookup of type DSYNC for the lookup name, and validate the response if DNSSEC is enabled. If a DSYNC RRset results, return it.¶
If the query resulted in a negative response:¶
If the response's SOA record indicates that the parent is more than
one label away from the _dsync label, construct a new lookup name
by inserting the _dsync label into the delegation owner name just
before the parent zone labels inferred from the negative response,
and go to step 2.¶
For example, city.ise.mie.jp is delegated from jp (and not
from ise.mie.jp or mie.jp!). The initial DSYNC query relating
to it is thus directed at city._dsync.ise.mie.jp. This is
expected to result in a negative response from jp, and another
query for city.ise.mie._dsync.jp is then required;¶
Otherwise, if the lookup name has any labels in front of the
_dsync label, remove them to construct a new lookup name (such
as _dsync.jp), and go to step 2.
(This is to enable zone structures without wildcards.)¶
Otherwise, return null (no notification target available).¶
When changing a CDS/CDNSKEY/CSYNC RRset in the child zone, the DNS operator SHOULD send a suitable NOTIFY message to the endpoint located as described in the previous section.¶
A NOTIFY message can only carry information about changes concerning one child zone. When there are changes to several child zones, the sender MUST send a separate notification for each one.¶
When a primary name server publishes a new RRset in the child, there typically is a time delay until all publicly visible copies of the zone will have been updated. If the primary sends a NOTIFY at the exact time of publication of the new zone, there is a potential for the parent to attempt CDS/CDNSKEY/CSYNC processing before the updated zone is visible. In this case the parent may draw the wrong conclusion (“the CDS RRset has not been updated”).¶
It is therefore RECOMMENDED that the child delays sending NOTIFY messages to the recipient until a consistent public view of the pertinent records is ensured.¶
NOTIFY messages are expected to elicit a response from the recipient ([RFC1996] Section 4.7). If no response is received, senders SHOULD employ the same logic as for SOA notifications ([RFC1996] Sections 3.5 and 3.6).¶
The parent's attempt to act upon the delegation update request may fail for a variety of reasons (e.g., due to violation of the continuity requirement set forth in [RFC7344] Section 4.1). Such failures may occur asynchronously, even after the NOTIFY response has been sent.¶
In order to learn about such failures, senders MAY include an [RFC9567] EDNS0 Report-Channel option in the NOTIFY message to request the receiving side to report any errors by making a report query with an appropriate extended DNS error code as described in [RFC8914]. When including this EDNS0 option, its agent domain MUST be subordinate or equal to one of the NS hostnames, as listed in the child's delegation in the parent zone.¶
Because of the security model where a notification by itself never causes a change (it can only speed up the time until the next check for the same thing), the sender's identity is not crucial. This opens up the possibility of having an arbitrary party (e.g., a side-car service) send the notifications, enabling this functionality even before the emergence of native support in nameserver software.¶
While the CDS/CDNSKEY/CSYNC processing following the receipt of a NOTIFY will often be performed by the registry, the protocol anticipates that in some contexts (especially for ICANN gTLDs), registrars may take on the task. In such cases, the parent may publish the current registrar notification endpoint, enabling notifications to be directed to the appropriate target. The mechanics of how this is arranged between registry and registrar are out of scope for this document; the protocol only offers the possibility to arrange things such that from the child perspective, it is inconsequential how the parent-side parties are organized: notifications are simply sent to the published address.¶
In the general case, the child DNS operator is unaware of whether the parent consumes CDS records or prefers CDNSKEY, or when that policy changes. It therefore seems advisable to publish both types of records, preferably using automation features of common authoritative nameserver software for ensuring consistency.¶
(RFC Editor: This subsection is to be removed before publication)¶
In the most common cases of using generalized notifications the recipient is expected to not be a nameserver, but rather some other type of service, like a CDS/CSYNC scanner.¶
However, this will likely not always be true. In particular it seems likely that in cases where the parent is not a large delegation-centric zone like a TLD, but rather a smaller zone with a small number of delegations there will not be separate services for everything and the recipient of the NOTIFY(CDS) or NOTIFY(CSYNC) will be an authoritative nameserver for the parent zone.¶
For this reason it seems most reasonable to stay within the the well documented and already supported message format specified in RFC 1996 and delivered over normal DNS transport, although not necessarily to port 53.¶
NOTIFY(CDS) messages carrying notification payloads (records) for several child zones MUST be discarded, as sending them is an error.¶
Upon receipt of a (potentially forwarded) valid NOTIFY(CDS) message for a particular child zone at the published address for CDS notifications, the receiving side (parent registry or registrar) has two options:¶
Acknowledge receipt by sending a NOTIFY response as described in [RFC1996] Section 4.7 (identical to NOTIFY query, but with QR bit set) and schedule an immediate check of the CDS and CDNSKEY RRsets as published by that particular child zone.¶
If the NOTIFY message contains an [RFC9567] EDNS0 Report-Channel option with an agent domain subordinate or equal to one of the NS hostnames listed in the delegation, the processing party SHOULD report any errors occuring during CDS/CDNSKEY processing by sending a report query with an appropriate extended DNS error code as described in [RFC8914].¶
If the check finds that the CDS/CDNSKEY RRset has indeed changed, the parent MAY reset the scanning timer for children for which NOTIFY(CDS) is received, or reduce the periodic scanning frequency accordingly (e.g. to every two weeks). This will decrease the scanning effort for the parent. If a CDS/CDNSKEY change is then detected (without having received a notification), the parent SHOULD clear that state and revert to the default scanning schedule.¶
Parents introducing CDS/CDNSKEY scanning support at the same time as NOTIFY(CDS) support are not in danger of breaking children's scanning assumption, and MAY therefore use a low-frequency scanning schedule in default mode.¶
Do not act upon the notification. To prevent retries, recipients SHOULD acknowledge the notification by sending a NOTIFY response even when otherwise ignoring the request, combined with a report query if feasible (see above). One reason to do this may be a rate limit (see Section 5), in which case "Blocked" (15) maybe a suitable extended DNS error code.¶
If the parent implements the first option, the convergence time (time between publication of a new CDS/CDNSKEY record in the child and propagation of the resulting DS) will decrease significantly, thereby providing improved service to the child zone.¶
If the parent, in addition to scheduling an immediate check for the child zone of the notification, also choses to modify the scanning schedule (to be less frequent), the cost of providing the scanning service will be reduced.¶
Upon receipt of a NOTIFY(CSYNC) to the published address for CSYNC notifications, the same options and considerations apply as for the NOTIFY(CDS).¶
The original NOTIFY specification sidesteps most security issues by not relying on the information in the NOTIFY message in any way, and instead only using it to "enter the state it would if the zone's refresh timer had expired" (Section 4.7 of [RFC1996]).¶
This security model is reused for generalized NOTIFY messages. It therefore seems impossible to affect the behaviour of the recipient of the NOTIFY other than by hastening the timing for when different checks are initiated.¶
The receipt of a notification message will, in general, cause the receiving party to perform one or more outbound queries for the records of interest (for example, NOTIFY(CDS) will cause CDS/CDNSKEY queries). When done using standard DNS, the size of these queries is comparable to that of the NOTIFY messages themselves, rendering any amplification attempts futile. The number of queries triggered per notification is also limited by the requirement that a NOTIFY message can refer to one child only.¶
However, when the outgoing query occurs via encrypted transport, some amplification is possible, both with respect to bandwidth and computational burden. In this case, the usual principle of bounding the work, even under unreasonable events, applies.¶
Receivers therefore MUST implement rate limiting for notification processing. It is RECOMMENDED to configure rate limiting independently for both the notification's source IP address and the name of the zone that is conveyed in the NOTIFY message. Rate limiting also mitigates processing load from garbage notifications.¶
Alternative solutions (such as signing notifications and validating their signatures) appear significantly more expensive without tangible benefit.¶
In order to facilitate schemes that are authenticated outside of DNSSEC (such as via SIG(0)), zones containing DSYNC records are not required to be signed. Spoofed DSYNC responses would prevent notifications from reaching their legitimate target, and a different party may receive unsolicited notifications; both effects, however, can also be achieved in the presence of DNSSEC. The illegitimate target is also enabled to learn notification contents in real-time, which may be a privacy concern for the sender. If so, the sender may choose to ignore unsigned DSYNC records.¶
IANA is requested to update the "Resource Record (RR) TYPEs" registry under the "Domain Name System (DNS) Parameters" registry group as follows:¶
Per [RFC8552], IANA is requested to create a new registry on the "Domain Name System (DNS) Parameters" IANA web page as follows:¶
DSYNC: Location of Synchronization Endpoints¶
Expert Review¶
(this document)¶
| RRtype | Scheme | Purpose | Reference |
|---|---|---|---|
| 0 | Null scheme (no-op) | (this document) | |
| CDS | 1 | Delegation management | (this document) |
| CSYNC | 1 | Delegation management | (this document) |
| 2-127 | Unassigned | ||
| 128-255 | Reserved (private use) | (this document) |
In order of first contribution: Joe Abley, Mark Andrews, Christian Elmerot, Ólafur Guðmundsson, Paul Wouters, Brian Dickson, Warren Kumari, Patrick Mevzek, Tim Wicinski¶
[RFC1996] introduced the concept of a DNS Notify message which was used to improve the convergence time for secondary servers when a DNS zone had been updated in the primary. The basic idea was to augment the traditional "pull" mechanism (a periodic SOA query) with a "push" mechanism (a Notify) for a common case that was otherwise very inefficient (due to either slow convergence or wasteful overly frequent scanning of the primary for changes).¶
While it is possible to indicate how frequently checks should occur (via the SOA Refresh parameter), these checks did not allow catching zone changes that fall between checkpoints. [RFC1996] addressed the optimization of the time-and-cost trade-off between a seceondary checking frequently for new versions of a zone, and infrequent checking, by replacing scheduled scanning with the more efficient NOTIFY mechanism.¶
Today, we have similar issues with slow updates of DNS data in spite of the data having been published. The two most obvious cases are CDS and CSYNC scanners deployed in a growing number of TLD registries. Because of the large number of child delegations, scanning for CDS and CSYNC records is rather slow (as in infrequent).¶
It is only a very small number of the delegations that will have updated CDS or CDNSKEY record in between two scanning runs. However, frequent scanning for CDS and CDNSKEY records is costly, and infrequent scanning causes slower convergence (i.e., delay until the DS RRset is updated).¶
Unlike in the original case, where the primary is able to suggest the scanning interval via the SOA Refresh parameter, an equivalent mechanism does not exist for DS-related scanning.¶
All of this above also applies to parents that offer automated NS and glue record maintenance via CSYNC scanning [RFC7477]. Again, given that CSYNC records change only rarely, frequent scanning of a large number of delegations seems disproportionately costly, while infrequent scanning causes slower convergence (delay until the delegation is updated).¶
While use of the NOTIFY mechanism for coordinating the key exchange in multi-signer setups [I-D.ietf-dnsop-dnssec-automation] is conceivable, the detailed specification is left for future work.¶
draft-ietf-dnsop-generalized-notify-02¶
Nits by Tim Wicinski¶
Dnsdir feedback¶
Specify timeout and error handling¶
Editorial nits¶
Reserve scheme value 0¶
draft-ietf-dnsop-generalized-notify-01¶
Reserve scheme values 128-255¶
Rename NOTIFY rrtype to DSYNC (to distinguish from NOTIFY message)¶
Describe endpoint discovery¶
Discussion on garbage notifications¶
More discussion on amplification risks¶
Clean-up, editorial changes¶
draft-ietf-dnsop-generalized-notify-00¶
Revision after adoption.¶
draft-thomassen-dnsop-generalized-dns-notify-02¶
Add rationale for staying in band¶
Add John as an author¶
draft-thomassen-dnsop-generalized-dns-notify-01¶
Mention Ry-to-Rr forwarding to accommodate RRR model¶
Add port number flexiblity¶
Add scheme parameter¶
Drop SRV-based alternative in favour of new NOTIFY RR¶
Editorial improvements¶
draft-thomassen-dnsop-generalized-dns-notify-00¶
Initial public draft.¶