]> The Swedish Internet Foundation

Sweden johan.stenstam@internetstiftelsen.se

Internet DNSOP Working Group Internet-Draft Delegation information (i.e. the NS RRset, possible glue, possible DS records) should always be kept in sync between child zone and parent zone. However, in practice that is not always the case. When the delegation information is not in sync the child zone is usually working fine, but without the amount of redundancy that the zone owner likely expects to have. Hence, should any further problems ensue it could have catastropic consequences. The DNS name space has lived with this problem for decades and it never goes away. Or, rather, it will never go away until a fully automated mechanism for how to keep the information in sync automatically is deployed. This document proposes such a mechanism. TO BE REMOVED: This document is being collaborated on in Github at: https://github.com/johanix/draft-johani-dnsop-delegation-mgmt-via-ddns. The most recent working version of the document, open issues, etc, should all be available there. The author (gratefully) accept pull requests.

Introduction For DNSSEC signed child zones with TLD registries as parents there is an emerging trend towards running so-called "CDS scanners" and "CSYNC scanners" by the parent. These scanners detect publication of CDS records (the child signalling a desire for an update to the DS RRset in the parent) and/or a CSYNC record (the child signalling a desire for an update to the NS RRset or, possibly, in-bailiwick glue in the parent. The scanners have a number of drawbacks, including being inefficient and slow. They are only applicable to DNSSEC-signed child zones and they add significant complexity to the parent operations. But given that, they do work. I-D.ietf-dnsop-generalized-notify-01 proposes a method to alleviate the inefficiency and slowness of scanners. But the DNSSEC requirement and the complexity remain. This document proposes an alternative mechanism to automate the updating of delegation information in the parent zone for a child zone based on DNS Dynamic Updates secured with SIG(0) signatures. This alternative mechanism shares the property of being efficient and provide rapid convergence (similar to generalized notifications in conjuction with scanners). Furthermore, it has the advantages of not requiring any scanners in the parent at all and also not being dependent on the child (and parent) being DNSSEC-signed. Knowledge of DNS NOTIFY and DNS Dynamic Updates and is assumed. DNS SIG(0) transaction signatures are documented in . In addition this Internet-Draft borrows heavily from the thoughts and problem statement from the Internet-Draft on Generalized DNS Notifications (work in progress).

Requirements Notation 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 when, and only when, they appear in all capitals, as shown here.

Is there a Use Case? Because of the drawbacks of CDS and CSYNC scanners they are unlikely to be able to fully automate the maintenance of delegation information in all parent zones. The primary reasons are the hard requirement on DNSSEC in the child zones and the complexity cost of operating the scanner infrastructure. In practice, scanners are likely mostly realistic for parent zones that are operated by well-resourced registries. All the parts of the DNS name space where the parent is smaller and more resource constrained would be able to automate the delegation management via this mechanism without the requirement of operating scanners. Also all parts of the name space where there are child zones that are not DNSSEC-signed would be able to use this. Obviously, also well-resourced parent zones with DNSSEC-signed child zones would be able to use this DNS UPDATE-based mechanism, but in those cases scanners plus generalized notifications would also work.

DNS NOTIFY versus DNS UPDATE DNS NOTIFY and DNS UPDATE messages share several properties and are used to address similar issues.

Similarities between NOTIFY and UPDATE Both NOTIFY and UPDATE are "push" rather than "pull" messages and therefore very efficient. Both NOTIFY and UPDATE are messages, in DNS packet format. They are used by one party (the sender) to inform another party (the recipient) that some piece of DNS information has changed and that as a consequence of this change the recipient of the message may want to also make a change to its DNS data. A NOTIFY (as per ) is only a hint and the recipient may ignore it. But if the recipient does listen to the NOTIFY it should make its own lookups to verify what has changed and whether that should trigger any changes in the DNS data provided by the recipient.

Differencies between NOTIFY and UPDATE The difference between the UPDATE and the NOTIFY is that the UPDATE contains the exact change that should (in the opinion of the sender) be applied to the recipients DNS data. Furthermore, for secure Dynamic Updates, the message also contains proof why the update should be trusted (in the form of a digital signature by a key that the recipient trusts). In this document the term "Dynamic Update" or "DNS UPDATE" implies secure dynamic update. Furthermore this document implies that the signature algorithms are always based on asymmetric crypto keys, using the same algorithms as are being used for DNSSEC. I.e. by using the right algorithm the resulting signatures will be as strong as DNSSEC-signatures. DNS UPDATEs can be used to update any information in a zone (subject to the policy of the recipient). But in the special case where the data that is updated is the delegation information for a child zone and it is sent across a zone cut (i.e. the child sends it to the parent), it acts as a glorified generalized NOTIFY. The DNS UPDATE in this case is essentially a message that says:

Terminology

SIG(0)

An assymmetric signing algorithm that allows the recipient to only need to know the public key to verify a signature created by the senders private key.

Updating Delegation Information via DNS UPDATEs. This is not a new idea. There is lots of prior art and prior documents, including the expired I-D.andrews-dnsop-update-parent-zones-04. The functionality to update delegation information in the parent zone via DNS UPDATE has been available for years in a least one DNS implementation (BIND9). However, while DNS UPDATE is used extensively inside organisations it has not seen much use across organisational boundaries. And zone cuts, almost by definition, usually cuts across such boundaries. When sending a DNS UPDATE it is necessary to know where to send it. Inside an organisation this information is usually readily available. But outside the organisation it is not. And even if the sender would know where to send the update, it is not at all clear that the destination is reachable to the sender (the parent primary is likely to be protected by firewalls and other measures). This constitutes a problem for using DNS UPDATES across zone cuts. Another concern is that traditionally DNS UPDATEs are sent to a primary nameserver, and if the update signture verifies the update is automatically applied to the DNS zone. This is often not an acceptable mechanism. The recipient may, for good reason, require additional policy checks and likely an audit trail. Finally, the change should in many cases not be applied to the running zone but rather to some sort of provisioning system or a database. This creates another problem for using DNS UPDATEs for managing delegation information. Both problems are addressed by the proposed mechanism for locating the recipient of a generalized NOTIFY.

Locating the Target for a generalized NOTIFY and/or DNS UPDATE. Section 3 of I-D.ietf-dnsop-generalized-notify-01 proposes a new RR type, tentatively with the mnemonic DSYNC, that has the following format:

where {target} is the domain name of the recipient of the NOTIFY message. {RRtype} is typically "CDS" or "CSYNC" in the case where delegation information should be updated (there are also other uses of generalized notifications). Finally, {scheme} is an 8 bit unsigned integer to indicate the type of notification mechanism to use. Scheme=1 is defined "NOTIFY", as in "send a generalized NOTIFY to {target} on port {port}". This document proposes the definition of a new {scheme} for the same record that is used for generalized NOTIFY. Scheme=2 is here defined as "UPDATE", as in "send a DNS UPDATE to {target} on port {port}". When parsing or presenting DNS zone data the 8 bit unsigned integer "2" should be replaced by the string "UPDATE", as in the examples below. Apart from defining a new scheme to specify the mechanism "UPDATE" (rather than the mechanism "NOTIFY") this document does not say anything about what Qname to look up or what RR type. The UPDATE mechanism should use exactly the same method of locating the target of the UPDATE as is used for generalized NOTIFY. This lookup addresses the first issue with using DNS UPDATE across organizational boundaries. Example 1: a parent zone announces support for DNS UPDATE as a mechanism for delegation synchronization for all child zones:

Example 2: a parent zone announces support different DNS UPDATE targets on a per-child basis

The DSYNC RRset is looked up, typically by the child primary name server or by a separate agent for the child, at the time that the delegation information for the child zone changes in some way that would prompt an update in the parent zone. When the {scheme} is "UPDATE" (i.e. the number 2 in the wire protocol) the interpretation is: Send a DNS UPDATE to the IP address for the name {target} on port 5302, where {target} is the domain name in the right-hand side of the DSYNC record that matches the qname in the DNS query.

Limitation of Scope for the Proposed Mechanism DNS UPDATE is in wide use all over the world, for all sorts of purposes. It is not in wide use across organizational boundaries. This document only address the specific case of a child zone that makes a change in its DNS information that will require an update of the corresponding information in the parent zone. This includes: changes to the NS RRset changes to glue (if any) changes to the DS RRset (if any) Only for those specific cases is the descibed mechanism proposed.

The DNS UPDATE Receiver While the simplest design is to send the DNS UPDATEs to the primary name server of the parent it will in most cases be more interesting to send them to a separate UPDATE Receiver. To separate the primary name server from the UPDATE Receiver use a {target} with addresses separate from the addresses of the primary name server.

Processing the UPDATE in the DNS UPDATE Receiver The receiver of the DNS UPDATE messages should implement a suitably strict policy for what updates are accepted (typically only allowing updates to the NS RRset, glue and DS RRset). Furthermore, it is strongly recommended that the policy is further tightened by only allowing updates to the delegation information of a child zone with the exact same name as the name of the SIG(0) key the signed the UPDATE request. I.e. an UPDATE request for the delegation information for the zone child.parent. should only be processed if it is signed by a SIG(0) key with the name child.parent. and the signature verifies correctly. Once the DNS UPDATE message has been verified to be correctly signed by a known and trusted key with the correct name and also adhere to the update policy it should be subjected to the same set of correctness tests as CDS/CSYNC scanner would have performed. If these requirements are also fulfilled the change may be applied to the parent zone in whatever manner the parent zone is maintained (as a text file, data in a database, via and API, etc).

Interpretation of the response to the DNS UPDATE. All DNS transactions are designed as a pair of messages and this is true also for DNS UPDATE. The interpretation of the different responses to DNS UPDATE are fully documented in , section 2.2.

RCODE NOERROR A response with rcode=0 ("NOERROR") should be interpreted as a confirmation that the DNS UPDATE has been received and accepted. I.e. the change to the parent DNS data should be expected to be published in the parent zone at some future time.

RCODE REFUSED A response with rcode=5 ("REFUSED") should be interpreted as a permanent signal that DNS UPDATEs are not supported by the receiver. This would indicate a parent misconfiguration, as the UPDATE should not be sent unless the parent has announced support for DNS UPDATE via publication of an appropriate target location record.

RCODE BADKEY A response with rcode=17 ("BADKEY") should be interpreted as a definitive statement that the DNS UPDATE Receiver does not have access to the public SIG(0) key needed for signature verification. In this case the child should fall back to bootstrap of the SIG(0) public key into the DNS UPDATE Receiver, see below.

No response to a DNS UPDATE The case of no response is more complex, as it is not possible to know whether the DNS UPDATE actually reached the reciever (or was lost in transit) or the response was not sent (or lost in transit). For this reason it is suggested that a lack of response is left as implementation dependent. That way the implementation has sufficient freedom do chose a sensible approach. Eg. if the sender of the DNS UPDATE message (like the primary name server of the child zone) only serves a single child, then resending the DNS UPDATE once or twice may be ok (to ensure that the lack of response is not due to packets being lost in transit). On the other hand, if the sender serves a large number of child zones below the same parent zone, then it may already know that the receiver for the DNS UPDATEs is not responding for any of the child zones, and then resending the update immediately is likely pointless.

Management of SIG(0) Public Keys Only the child should have access to the SIG(0) private key. The corresponding SIG(0) public key should preferably be published in DNS, but it doesn't have have to be. The SIG(0) public key only needs to be available to the parent DNS UPDATE Receiver. Keeping all the public SIG(0) keys for different child zones in some sort of database is perfectly fine.

Bootstrapping the SIG(0) Public Key Into the DNS UPDATE Receiver Bootstrap is simpler if the child zone is signed. Therefore the signed and unsigned cases are described separately.

Child zone is DNSSEC-signed If the child zone is DNSSEC-signed, then the preferred mechanism is to publish the public SIG(0) key as a KEY record at the child apex. This can then be looked up and validated by the DNS UPDATE Receiver.

However, the receiver should have access to the key at the time of receiving the update, it should not have to do DNS lookups and DNSSEC validation in response to a DNS UPDATE message (that might open up for various types of attacks). Therefore the proposal is to trigger the parent reciver to lookup and validate the key by issuing a DNS UPDATE that only contains an addition (no delete) of the KEY record from the child zone:

When receiving such a message the reciever SHOULD put that key into a queue for later look up of the corresponding KEY record and validation of the DNSSEC-signature. In case of validation failure (or absence of a DNSSEC signature) the SIG(0) SHOULD NOT be added to the set of keys that the receiver knows and trust. If the validation succeeds the key should be added to the set of trusted public SIG(0) keys stored locally at the receiver.

Child zone is unsigned In the absence of a DNSSEC-based validation path some alternative mechanism will have ot be found. The primary audience for this DNS UPDATE based synchronization mechanism is "non-registries". In those cases there is by definition some mechanism in place to communicate information from the child to the parent, be it email, a web form, pieces of paper or something else. The same mechanism can be extended to also be used to communicate the initial SIG(0) public key from the child to the parent. It should also be noted that the bootstrap problem is essentially identical to the "automatic DNSSEC Bootstrap via CDS" service that multiple TLD registries provide today. Hence, to bootstrap the public SIG(0) key for child zone it is possible for the parent to scan the child zone for the KEY RRset. To make spoofed responses more difficult it is possible to also add a requirement to use a more stable transport, like TCP (or in the future DOT, DOH or DOQ once those become more generally available for DNS queries to authoritative servers). If the received KEY RRset is consistent from multiple vantage points and multiple times then it is considered authentic and stored by the parent's UPDATE Receiver as a trusted SIG(0) key for the child. Should a "registry" parent want to support this mechanism (as a service to its unsigned children) then the most likely model is that the target of the DNS UPDATE is operated by the registrar (or possibly that the DNS UPDATE is forwarded to the registrar). The registrar performs its normal verifications of a change and then transforms the update into an EPP transaction to communicate it to the registry.

Rolling the SIG(0) Key Once the parent (or registrar) DNS UPDATE Receiver has the key, the child can update it via a DNS UPDATE just like updating the NS RRset, the DS RRset or the glue in the parent zone (assuming a suitable DNS UPDATE policy in the parent). I.e. only the initial bootstrapping of the key is an issue. Note, however, that the alternative of re-bootstrapping (by whatever bootstrapping mechanism was used) in case of a key compromise may be a better alternative to the parent supporting key rollover for child SIG(0) keys. The decision of whether to allow SIG(0) key rollover via DNS UPDATE is left as parent-side policy.

Scalability Considerations The primary existing mechanism for automatic synchronization of DNS delegation information is based on parent-side "scanning" of the child zones for CDS and/or CSYNC RRsets, performing DNSSEC validation on the result and then, in the CSYNC case, based on the result, issue and validate a potentially large number of additional DNS queries, all of which must be DNSSEC validated. This makes a CDS/CSYNC scanner for a parent with a significant number of delegations a complex and resource consuming service. Among the issues are rate-limiting by large DNS operators and inherent difficulties in issuing millions of recursive DNS queries where all received data must be validated. By comparision, the DNS UPDATE based mechanism for automatic synchronization shifts most of the effort to the child side. It is the child that is responsible for detecting the need to update the delegation information in the parent zone (which makes sense as it is the child that has made a change and therefore, in many cases, already "knows"). It is the child rather than the parent that computes what records should be added or removed. All of this is good for scalability. Furthermore, the information collection and validation effort for the UPDATE Receiver is restricted to validation of a single DNS message, using a SIG(0) key that the UPDATE Receiver already has. Hence, as the data collection and validation is much simplified the task of the UPDATE Receiver is mostly focused on the policy issues of whether to approve the UPDATE or not (i.e. the same process that a CDS and/or CSYNC scanner follows).

Security Considerations. Any fully automatic mechanism to update the contents of a DNS zone opens up a potential vulnerability should the mechanism not be implemented correctly. In this case the definition of "correct" is a question for the receiver of the DNS UPDATE. The receiver should validate the authenticity of the DNS UPDATE and then do the same checks and verifications as a CDS or CSYNC scanner does. The difference from the scanner is only in the validation: single SIG(0) signature by a key that the UPDATE Receiver trusts vs DNSSEC signatures that chain back to a DNSSEC trust anchor that the validator trusts. Another issue of concern is whether a parent-side service that provides support for changes to child delegation information via DNS UPDATE is open for potential denial-of-service attacks. The answer is likely no, as it is possible to have a very strict rate-limiting policy based on the observation that no child zone should have a legitimate need to change its delegation information frequently. Furthermore, as the location of the UPDATE Receiver can be separated from any parent name server even in the worst case the only service that can be subject to an attack is the UPDATE Receiver itself, which is a service that previously did not exist.

IANA Considerations. IANA is requested to assign a new "scheme" value to the registry for "DSYNC Location of Synchronization Endpoints" as follows:

Reference

(this document)

RRtype Scheme Purpose Reference ANY 2 Delegation management (this document)

Acknowledgements. Peter Thomassen and I together came up with the location mechanism for the generalized notifications, which this draft relies upon. Mark Andrews provided the initial inspiration for writing some code to experiment with the combination of the location mechanism from the generalised notifications with DNS UPDATEs across zone cuts.

This memo describes the NOTIFY opcode for DNS, by which a master server advises a set of slave servers that the master's data has been changed and that a query should be initiated to discover the new data. [STANDARDS-TRACK] Using this specification of the UPDATE opcode, it is possible to add or delete RRs or RRsets from a specified zone. Prerequisites are specified separately from update operations, and can specify a dependency upon either the previous existence or nonexistence of an RRset, or the existence of a single RR. [STANDARDS-TRACK] This document proposes a method for performing secure Domain Name System (DNS) dynamic updates. [STANDARDS-TRACK] This document describes the minor but non-interoperable changes in Request and Transaction signature resource records ( SIG(0)s ) that implementation experience has deemed necessary. [STANDARDS-TRACK] In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. This document describes an application-layer client-server protocol for the provisioning and management of objects stored in a shared central repository. Specified in XML, the protocol defines generic object management operations and an extensible framework that maps protocol operations to objects. This document includes a protocol specification, an object mapping template, and an XML media type registration. This document obsoletes RFC 4930. [STANDARDS-TRACK]

Change History (to be removed before publication) draft-johani-dnsop-delegation-mgmt-via-ddns-03

• Update the draft based on the excellent dnsdir review of draft-ietf-dnsop-generalized-notify-01 by Patrick Mevsek.

• Expand the section on alternatives for initial bootstrap to suggest a mechanism similar to what is used for automatic bootstrap of DNSSEC signed delegations via multiple queries for child the CDS RRset.

• Added a section on scalability considerations.

• Expanded the Security Considerations section with a paragraph on the potential for DDOS attacks aimed at the UPDATE Receiver.

draft-johani-dnsop-delegation-mgmt-via-ddns-02

• Update the references to the (updated) I-D for generalized notifications.

• Remove duplicated descriptions between the two drafts. It is sufficient that the generalized notification draft describes the mechanics.

draft-johani-dnsop-delegation-mgmt-via-ddns-01

• Change the RRtype from the original "NOTIFY" to the proposed "DSYNC"

• Expand the description of how to interpret different RCODE responses to the UPDATE. Expand the description of bootstrapping alternatives. Change the mnemonic of the RR type used from "NOTIFY" to "DSYNC" in the examples.

draft-johani-dnsop-delegation-mgmt-via-ddns-00

• Initial public draft.