DNSSEC Operational Practices, Version 2
draft-ietf-dnsop-rfc4641bis-03

Abstract

This document describes a set of practices for operating the DNS with security extensions (DNSSEC). The target audience is zone administrators deploying DNSSEC.

The document discusses operational aspects of using keys and signatures in the DNS. It discusses issues of key generation, key storage, signature generation, key rollover, and related policies.

[When apporoved] This document obsoletes RFC 4641 as it covers more operational ground and gives more up-to-date requirements with respect to key sizes and the DNSSEC operations.

Status of This Memo

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 http://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 January 13, 2011.

Copyright Notice

Copyright (c) 2010 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 (http://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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.

Table of Contents

1.  Introduction
    1.1.  The Use of the Term 'key'
    1.2.  Time Definitions
2.  Keeping the Chain of Trust Intact
3.  Keys Generation and Storage
    3.1.  Operational motivation for ZSKs and KSKs
        3.1.1.  Motivations for the KSK and ZSK Separation
        3.1.2.  Practical concequences of KSK and ZSK Separation
            3.1.2.1.  Rolling a KSK that is not a trust-anchor
            3.1.2.2.  Rolling a KSK that is a trust-anchor
            3.1.2.3.  The use of the SEP flag
        3.1.3.  Key Effectivity Period
        3.1.4.  Cryptographic Considerations
            3.1.4.1.  Key Algorithm
            3.1.4.2.  Key Sizes
            3.1.4.3.  Private Key Storage
            3.1.4.4.  Key Generation
            3.1.4.5.  Differentiation for 'High-Level' Zones?
4.  Signature Generation, Key Rollover, and Related Policies
    4.1.  Key Rollovers
        4.1.1.  Zone Signing Key Rollovers
            4.1.1.1.  Pre-Publish Key Rollover
            4.1.1.2.  Double Signature Zone Signing Key Rollover
            4.1.1.3.  Pros and Cons of the Schemes
        4.1.2.  Key Signing Key Rollovers
        4.1.3.  Difference Between ZSK and KSK Rollovers
        4.1.4.  Rollover for a Single Type Signing Key rollover
        4.1.5.  Key algorithm rollover
        4.1.6.  Automated Key Rollovers
    4.2.  Planning for Emergency Key Rollover
        4.2.1.  KSK Compromise
            4.2.1.1.  Keeping the Chain of Trust Intact
            4.2.1.2.  Breaking the Chain of Trust
        4.2.2.  ZSK Compromise
        4.2.3.  Compromises of Keys Anchored in Resolvers
    4.3.  Parental Policies
        4.3.1.  Initial Key Exchanges and Parental Policies Considerations
        4.3.2.  Storing Keys or Hashes?
        4.3.3.  Security Lameness
        4.3.4.  DS Signature Validity Period
        4.3.5.  Changing DNS Operators
            4.3.5.1.  Cooperationg DNS operators
            4.3.5.2.  Non Cooperationg DNS Operators
    4.4.  Time in DNSSEC
        4.4.1.  Time Considerations
        4.4.2.  Signature Validation Periods
            4.4.2.1.  Maximum Value
            4.4.2.2.  Minimum Value
            4.4.2.3.  differentiation between RR sets
            4.4.2.4.  Other timing parameters in a zone
5.  Next Record type
    5.1.  Differences between NSEC and NSEC3
    5.2.  NSEC or NSEC3
    5.3.  NSEC3 parameters
        5.3.1.  NSEC3 Algorithm
        5.3.2.  NSEC3 Iterations
        5.3.3.  NSEC3 Salt
        5.3.4.  Opt-out
6.  Security Considerations
7.  IANA considerations
8.  Acknowledgments
9.  References
    9.1.  Normative References
    9.2.  Informative References
Appendix A.  Terminology
Appendix B.  Zone Signing Key Rollover How-To
Appendix C.  Typographic Conventions
Appendix D.  Document Editing History
    D.1.  draft-ietf-dnsop-rfc4641-00
    D.2.  version 0->1
    D.3.  version 1->2
    D.4.  version 2->3

1.  Introduction

This document describes how to run a DNS Security (DNSSEC)-enabled environment. It is intended for operators who have knowledge of the DNS (see RFC 1034 (Mockapetris, P., “Domain names - concepts and facilities,” November 1987.) [1] and RFC 1035 (Mockapetris, P., “Domain names - implementation and specification,” November 1987.) [2]) and want to deploy DNSSEC. The focus of the document is on serving authoritative DNS information and concerns zone owners, name server operators, registry, registrar and registrants whereby it is assumed that there is no direct relation between those entities and the operators of validating recursive name servers (validators). See RFC 4033 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “DNS Security Introduction and Requirements,” March 2005.) [3] for an introduction to DNSSEC, RFC 4034 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Resource Records for the DNS Security Extensions,” March 2005.) [4] for the newly introduced Resource Records (RRs), and RFC 4035 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Protocol Modifications for the DNS Security Extensions,” March 2005.) [5] for the protocol changes.

During workshops and early operational deployment, operators and system administrators have gained experience about operating the DNS with security extensions (DNSSEC). This document translates these experiences into a set of practices for zone administrators. At the time of writing -the root is being signed and the first secure delegations are provisioned- there exists relatively little experience with DNSSEC in production environments; this document should therefore explicitly not be seen as representing 'Best Current Practices'. The document therefore targets to provide a number of choices with the considerations that motivate a particular choice. It then gives the operational guidelines that relate to a choice. The document does not give strong recommendations, that may be subject for a future version of this document. [OK: This is really a straw-man and causes a difference in tone that I believe was the instruction of the WG during the IETF 77 meeting. The document could be made much shorter when particular recommendations are made? Is there a general consensus that we should currently not make particular recommendations?]

The procedures herein are focused on the maintenance of signed zones (i.e., signing and publishing zones on authoritative servers). It is intended that maintenance of zones such as re-signing or key rollovers be transparent to any verifying clients on the Internet.

The structure of this document is as follows. In Section 2 (Keeping the Chain of Trust Intact), we discuss the importance of keeping the "chain of trust" intact. Aspects of key generation and storage of private keys are discussed in Section 3 (Keys Generation and Storage); the focus in this section is mainly on the private part of the key(s). Section 4 (Signature Generation, Key Rollover, and Related Policies) describes considerations concerning the public part of the keys. Since these public keys appear in the DNS one has to take into account all kinds of timing issues, which are discussed in Section 4.4 (Time in DNSSEC). Section 4.1 (Key Rollovers) and Section 4.2 (Planning for Emergency Key Rollover) deal with the rollover, or supercession, of keys. Finally, Section 4.3 (Parental Policies) discusses considerations on how parents deal with their children's public keys in order to maintain chains of trust.

The typographic conventions used in this document are explained in Appendix C (Typographic Conventions).

Since this is a document with operational suggestions and there are no protocol specifications, the RFC 2119 (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.) [6] language does not apply.

This document [OK: when approved] obsoletes RFC 4641 (Kolkman, O. and R. Gieben, “DNSSEC Operational Practices,” September 2006.) [15].

[OK: Editorial comments and questions are indicated by square brackets and editor innitials]

1.1.  The Use of the Term 'key'

It is assumed that the reader is familiar with the concept of asymmetric keys on which DNSSEC is based (public key cryptography RFC4949 (Shirey, R., “Internet Security Glossary, Version 2,” August 2007.) [16]). Therefore, this document will use the term 'key' rather loosely. Where it is written that 'a key is used to sign data' it is assumed that the reader understands that it is the private part of the key pair that is used for signing. It is also assumed that the reader understands that the public part of the key pair is published in the DNSKEY Resource Record and that it is the public part that is used in key exchanges.

1.2.  Time Definitions

In this document, we will be using a number of time-related terms. The following definitions apply:

• "Signature validity period" The period that a signature is valid. It starts at the time specified in the signature inception field of the RRSIG RR and ends at the time specified in the expiration field of the RRSIG RR.
• "Signature publication period" Time after which a signature (made with a specific key) is replaced with a new signature (made with the same key). This replacement takes place by publishing the relevant RRSIG in the master zone file. After one stops publishing an RRSIG in a zone, it may take a while before the RRSIG has expired from caches and has actually been removed from the DNS.
• "Key effectivity period" The period during which a key pair is expected to be effective. This period is defined as the time between the first inception time stamp and the last expiration date of any signature made with this key, regardless of any discontinuity in the use of the key. The key effectivity period can span multiple signature validity periods.
• "Maximum/Minimum Zone Time to Live (TTL)" The maximum or minimum value of the TTLs from the complete set of RRs in a zone. Note that the minimum TTL is not the same as the MINIMUM field in the SOA RR. See RFC2308 (Andrews, M., “Negative Caching of DNS Queries (DNS NCACHE),” March 1998.) [9] for more information.

2.  Keeping the Chain of Trust Intact

Maintaining a valid chain of trust is important because broken chains of trust will result in data being marked as Bogus (as defined in RFC4033 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “DNS Security Introduction and Requirements,” March 2005.) [3] Section 5), which may cause entire (sub)domains to become invisible to verifying clients. The administrators of secured zones have to realize that their zone is, to verifying clients, part of a chain of trust.

As mentioned in the introduction, the procedures herein are intended to ensure that maintenance of zones, such as re-signing or key rollovers, will be transparent to the verifying clients on the Internet.

Administrators of secured zones will have to keep in mind that data published on an authoritative primary server will not be immediately seen by verifying clients; it may take some time for the data to be transferred to other secondary authoritative nameservers and clients may be fetching data from caching non-authoritative servers. In this light, note that the time for a zone transfer from master to slave is negligible when using NOTIFY [8] (Vixie, P., “A Mechanism for Prompt Notification of Zone Changes (DNS NOTIFY),” August 1996.) and incremental transfer (IXFR) [7] (Ohta, M., “Incremental Zone Transfer in DNS,” August 1996.). It increases when full zone transfers (AXFR) are used in combination with NOTIFY. It increases even more if you rely on full zone transfers based on only the SOA timing parameters for refresh.

For the verifying clients, it is important that data from secured zones can be used to build chains of trust regardless of whether the data came directly from an authoritative server, a caching nameserver, or some middle box. Only by carefully using the available timing parameters can a zone administrator ensure that the data necessary for verification can be obtained.

The responsibility for maintaining the chain of trust is shared by administrators of secured zones in the chain of trust. This is most obvious in the case of a 'key compromise' when a trade-off between maintaining a valid chain of trust and replacing the compromised keys as soon as possible must be made. Then zone administrators will have to make a trade-off, between keeping the chain of trust intact -- thereby allowing for attacks with the compromised key -- or deliberately breaking the chain of trust and making secured subdomains invisible to security-aware resolvers. Also see Section 4.2 (Planning for Emergency Key Rollover).

3.  Keys Generation and Storage

This section describes a number of considerations with respect to the use of keys. For the desgin of a operational procedure for key generation and storage the following decisions need to be taken in order:

• Does one differentiate between Zone Signing and Key Signing Keys or is the use of one type of key sufficient;
• Are Key Signing Keys (likely to be) in use as Trust Anchors;
• What are the timing parameters that are allowed by the operational requirements; and finally
• What are the cryptographic parameters that fit the operational need?

Below are the considerations that need to be taken into account when making decisions.

3.1.  Operational motivation for ZSKs and KSKs

The DNSSEC validation protocol does not distinguish between different types of DNSKEYs. All DNSKEYs can be used during the validation. The motivations to differentiate between keys are purely operational.

For operational reasons, motivated below, it is possible to designate one or more keys as Key Signing Keys (KSKs). These keys will only sign the apex DNSKEY RRSet in a zone. Other keys can be used to sign all the RRSets in a zone and are referred to as Zone Signing Keys (ZSKs). In case the differentiation between KSK and ZSK is not made we talk about a Single Type signing scheme

3.1.1.  Motivations for the KSK and ZSK Separation

Differentiating between the KSK and ZSK is mostly done for operational purposes whereby

If the two functions are separated then, for almost any method of key management and zone signing, the KSK is used less frequently then the ZSK. Once a key set is signed with the KSK, all the keys in the key set can be used as ZSKs. If there has been an event that increases the risk that a ZSK is compromised it can be simply dropped from the key set. The new key set is then re-signed with the KSK.

Changing a key with secure entry point (SEP) functionality can be relatively expensive as it involves interaction with 3rd parties: When a key is only pointed to by the parental DS one needs to complete the interaction with the parental registry and wait for the transaction to appear in the DNS. In the case that a SEP key that is in use as a trust-anchor one has to wait until one has sufficient confidence that all trust anchors have been replaced. In fact, it may be that one is not able to reach the complete user-base with information about the key rollover.

There is also a risk that keys are compromised through theft or loss. For keys that are needed to sign dynamically updated records and are therefore installed on file-systems of nameservers that are connected to the network that risk is relatively high. For keys that are stored on Hardware Signing Modules such risk is relatively low. By separating the KSK and ZSK functionality these risks can be managed while making the tradeoff against the costs involved. For example, a KSK can be stored off-line or with more limitation on access control than ZSKs which need to be readily available for operational purposes such as the addition or deletion of zone data. More concretely a KSK stored on a smartcard, that is kept in a safe, combinded with a ZSK stored on a filesystem accessible by operators for daily routine may provide more operational flexibility and higher computational performance than a single key (with combined KSK and ZSK functionality) stored on an HSM.

[OK: The following paragraph is new in version 3 of the draft, does it need to be more detailed e.g. by referring to RFC 3766? Also the argument is made on http://www.educatedguesswork.org/2009/10/on_the_security_of_zsk_rollove.html, not sure about how about providing a stable reference to that. Should we add an informative reference here? ]

Finally there is a risk of cryptanalysis of the key material. The costs of such analysis are correlated to the length of the key. However, cryptanalysis arguments provide no strong motivation for a KSK/ZSK split. Suppose one differentiates between a KSK and a ZSK whereby the KSK effectivity period is X times the ZSK effectivity period. Then, in order for the resistance to cryptanalysis to be the same for the KSK and the ZSK, the KSK needs to be X times stronger than the ZSK. Since for all practical purposes X will somewhere of the order of 10 to 100 the associated key sizes will vary only about a byte in size for symmetric keys. Which, translated to asymmetric keys, is still an insignificant enough size difference to warrant a key-split; not for packet size or signing speed arguments.

The arguments for differentiation between the ZSK and KSK are weakest when:

the exposure to risk is low (e.g. when keys are stored on HSMs);

if one can be certain that a key is not used as a trust-anchor;

maintenance of the various keys cannot be performed through tools (is prone to human error); and

the interaction through the registrar-registry provisioning chain -- in particular the timely appearance of a new DS record in the parent zone in emergency situations -- is predictable.

If the above holds then the costs of the operational complexity of a KSK-ZSK split may outweigh the costs of operational flexibility and choosing for a single type signing scheme is then a reasonable option. In other cases we advise that the separation between KSKs and ZSKs is made and that the SEP flag is exclusively set on KSKs.

3.1.2.  Practical concequences of KSK and ZSK Separation

Given the assumption that for KSKs the SEP flag is set, the KSK can be distinguished from a ZSK by examining the flag field in the DNSKEY RR: If the flag field is an odd number it is a KSK. If it is an even number it is a ZSK.

The Zone Signing Key can be used to sign all the data in a zone on a regular basis. When a Zone Signing Key is to be rolled, no interaction with the parent is needed. This allows for signature validity periods on the order of days.

The Key Signing Key is only to be used to sign the DNSKEY RRs in a zone. If a Key Signing Key is to be rolled over, there will be interactions with parties other than the zone administrator. If there is a parent zone, these can include the registry of the parent zone or administrators of verifying resolvers that have the particular key configured as secure entry points. If this is a trust anchor, everyone relying on the trust anchor needs to roll over to the new key. The latter may be subject to stability costs if automated trust-anchor rollover mechanisms (such as e.g. RFC5011 (StJohns, M., “Automated Updates of DNS Security (DNSSEC) Trust Anchors,” September 2007.) [17]) are not in place. Hence, the key effectivity period of these keys can and should be made much longer.

3.1.2.1.  Rolling a KSK that is not a trust-anchor

There are 3 schools of thought on rolling a KSK that is not a trust anchor:

• It should be done frequently and regularly (possibly every few months) so that a key rollover remains an operational routine.
• It should be done frequently but irregularly. Frequently meaning every few monthts, again based on the argument that a rollover is a practiced and common operational routine, but irregular, i.e. with a large jitter, so that 3rd parties do not start to rely on the key and will not be tempted to configure those as a trust-anchor.
• It should only be done when it is known or strongly suspected that the key can be or has been compromised.

There is no widespread agreement on which of these three schools of thought is better for different deployments of DNSSEC. There is a stability cost every time a non-anchor KSK is rolled over, but it is possibly low if the communication between the child and the parent is good. On the other hand, the only completely effective way to tell if the communication is good is to test it periodically. Thus, rolling a KSK with a parent is only done for two reasons: to test and verify the rolling system to prepare for an emergency, and in the case of (preventing) an actual emergency.

Finally, a zone ower can, in most cases, not be fully certain that the zone's KSK is not in use as a trust-anchor and while the configuration of trust-anchors is not the responsibility of the zone owner there may be stabilty costs for the validator administrator that (wrongfully) configured the trust-anchor when the zone owner roles a KSK.

3.1.2.2.  Rolling a KSK that is a trust-anchor

The same operational concerns apply to the rollover of KSKs that are used as trust-anchors. But remember: if a trust anchor replacement is done incorrectly, the entire zone that the trust anchor covers will become bogus until the trust anchor is corrected.

In a large number of cases it will be safe to work from the assumption that ones keys are not in use as trust-anchors. If a zone owner publishes a "DNSSEC Signing Policy and Practice Statement" [25] (Ljunggren, F., Eklund-Lowinder, A., and T. Okubo, “DNSSEC Signing Policy & Practice Statement Framework,” March 2010.) that should be explicit about the fact whether the existence of trust anchors will be taken into account in any way or not. There may be cases where local policies enforce the configuration of trust-anchors on zones which are mission critical (e.g. in enterprises where the trust-anchor for the enterprise domain is configured in the enterprise's validator) It is expected that the zone owners are aware of such circumstances.

One can argue that because of the difficulty of getting all users of a trust anchor to replace an old trust anchor with a new one, a KSK that is a trust anchor should never be rolled unless it is known or strongly suspected that the key has been compromised. In other words the costs of a KSK rollover are prohibitively high because some users cannot be reached.

However, the "operational habit" argument also applies to trust anchor reconfiguration at the clients validator. If a short key effectivity period is used and the trust anchor configuration has to be revisited on a regular basis, the odds that the configuration tends to be forgotten is smaller. In fact, the costs for those users can be minimized by automating the rollover RFC5011 (StJohns, M., “Automated Updates of DNS Security (DNSSEC) Trust Anchors,” September 2007.) [17] and by rolling the key regularly, and advertising such, so that the operators of recursive nameservers will put the appropriate mechanism in place to deal with these stability costs, or, in other words, budget for these costs instead of incuring them unexpectedly.

It is therefore recommended to roll KSKs (that are likely to be used as trust-anchors) if and only if those rollovers can be tracked using standardized (e.g. RFC5011) mechanisms.

3.1.2.3.  The use of the SEP flag

The so-called Secure Entry Point (SEP) [5] (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Protocol Modifications for the DNS Security Extensions,” March 2005.) flag can be used to distinguish between keys that are intended to be used as the secure entry point into the zone when building chains of trust, e.g they are (to be) pointed to by parental DS RRs or configured as a trust-anchor.

While the SEP flag does not play any role in the validation failure it is used in parctice for operational purposes such as for the rollover mechanism discribed in RFC5011 (StJohns, M., “Automated Updates of DNS Security (DNSSEC) Trust Anchors,” September 2007.) [17]. The comon convention is to set the SEP flag on any key that is used for key exchanges with the parent and potentially for configuration as trust anchors. Therefore it is recomended that the SEP flag is set on KSKs and not on ZSKs while in those cases that a distinction between KSK and ZSK is not made (i.e. for a Single Type signing scheme) the SEP flag is recommended be set on all keys.

Note that signing tools may assume a KSK/ZSK split and use the (non) presence of the SEP flag to determine which key is to be used for signing zone data, these tools may get confused when a single type signing scheme is used.

3.1.3.  Key Effectivity Period

In general the available key lenght sets an upper limit on the Key Effectivity Period. For all practical purposes it is sufficient to define the Key Effectivity Period based on purely operational requirements and match the keylength to that value; Ignoring the operational perspective, a reasonable effectivity period for KSKs that have a parent zone is of the order of 2 decades or longer. That is, if one does not plan to test the rollover procedure, the key should be effective essentially forever, and then only rolled over in case of emergency.

When one chooses for a regular key-rollover, a reasonable key effectivity period for KSKs that have a parent zone is 13 months, with the intent to replace them after 12 months. As argued above, this annual rollover gives operational practice to rollovers for both the zone as the validator administrators. Besided, in most environments a year is a timespan that is easily planned and communicated.

In case keys are stored on on-line systems and the exposure to various threats of compromise is fairly high an intended key effectivity period of a month is reasonable for Zone Signing Keys.

Key effectivity periods can be made very short, as in a few minutes. But when replacing keys one has to take the considerations from Section 4.4 (Time in DNSSEC) and Section 4.1 (Key Rollovers) into account.

The motivation for having the ZSK's effectivity period shorter than the KSK's effectivity period is rooted in the operational consideration that it is more likely that operators have more frequent read access to the ZSK than to the KSK. If ZSK's are maintained on cryptographic Hardware Security Modules (HSM) than the motivation to have different key effectivity periods is weakend.

However, if the risk of loss, theft or other compromise is the same for a zone and key signing key there are little arguments to choose different effectivity periods for ZSKs and KSKs. And in fact, when the split between ZSKs and KSKs is not made than all considerations above apply.

There are certainly cases (e.g. where the the costs and risk of compromise, and the costs and risks involved with having to perform an emergency roll are also low) that the use of a single type signing scheme with a long key effectivity period is a good choice.

3.1.4.  Cryptographic Considerations

3.1.4.1.  Key Algorithm

There are currently two types of signature algorithms that can be used in DNSSEC: RSA and DSA. Both are fully specified in many freely-available documents, and both are widely considered to be patent-free. The creation of signatures wiht RSA and DSA takes roughly the same time, but DSA is about ten times slower for signature verification.

We suggest the use of RSA/SHA-256 as the preferred signature algorithms and RSA/SHA-1 as an alternative. Both have advantages and disadvantages. RSA/SHA-1 has been deployed for many years, while RSA/SHA-256 has only begun to be deployed. On the other hand, it is expected that if effective attacks on either algorithm appeark, they will appear for RSA/SHA-1 first. RSA/MD5 should not be considered for use because RSA/MD5 will very likely be the first common-use signature algorithm to have an effective attack.

At the time of publication, it is known that the SHA-1 hash has cryptanalysis issues. There is work in progress on addressing these issues. We recommend the use of public key algorithms based on hashes stronger than SHA-1 (e.g., SHA-256), as soon as these algorithms are available in protocol specifications (see RFC5702 (Jansen, J., “Use of SHA-2 Algorithms with RSA in DNSKEY and RRSIG Resource Records for DNSSEC,” October 2009.) [23] and RFC4509 (Hardaker, W., “Use of SHA-256 in DNSSEC Delegation Signer (DS) Resource Records (RRs),” May 2006.) [20]) and implementations.

3.1.4.2.  Key Sizes

DNSSEC signing keys should be large enough to avoid all know cryptographic attacks during the lifetime of the key. To date, despite huge efforts, no one has broken a regular 1024-bit key; in fact, the best completed attack is estimated to be the equivalent of a 700-bit key. An attacker breaking a 1024-bit signing key would need expend phenominal amounts of networked computing power in a way that would not be detected in order to break a single key. Because of this, it is estimated that most zones can safely use 1024-bit keys for at least the next ten years. A 1024-bit asymmetric key has an approximate equivalent strength of a symmetric 80-bit key.

Keys that are used as extremely high value trust anchors, or non-anchor keys that may be difficult to roll over, may want to use lengths longer than 1024 bits. Typically, the next larger key size used is 2048 bits, which have the approximate equivalent strength of a symmetric 112-bit key. In a standard CPU, it takes about four times as long to sign or verify with a 2048-bit key as it does with a 1024-bit key.

Another way to decide on the size of key to use is to remember that the phenominal effort it takes for an attacker to break a 1024-bit key is the same regardless of how the key is used. If an attacker has the capability of breaking a 1024-bit DNSSEC key, he also has the capability of breaking one of the many 1024-bit TLS trust anchor keys that are installed with web browsers. If the value of a DNSSEC key is lower to the attacker than the value of a TLS trust anchor, the attacker will use the resources to attack the TLS trust anchor.

It is possible that there is a unexpected improvement in the ability for attackers to beak keys, and that such an attack would make it feasible to break 1024-bit keys but not 2048-bit keys. If such an improvement happens, it is likely that there will be a huge amount of publicity, particularly because of the large number of 1024-bit TLS trust anchors build into popular web browsers. At that time, all 1024-bit keys (both ones with parent zones and ones that are trust anchors) can be rolled over and replaced with larger keys.

Earlier documents (including the previous version of this document) urged the use of longer keys in situations where a particular key was "heavily used". That advice may have been true 15 years ago, but it is not true today when using RSA or DSA algorithms and keys of 1024 bits or higher.

3.1.4.3.  Private Key Storage

It is recommended that, where possible, zone private keys and the zone file master copy that is to be signed be kept and used in off-line, non-network-connected, physically secure machines only. Periodically, an application can be run to add authentication to a zone by adding RRSIG and NSEC RRs. Then the augmented file can be transferred.

When relying on dynamic update [10] (Wellington, B., “Secure Domain Name System (DNS) Dynamic Update,” November 2000.) to manage a signed zone, be aware that at least one private key of the zone will have to reside on the master server. This key is only as secure as the amount of exposure the server receives to unknown clients and the security of the host. Although not mandatory, one could administer the DNS in the following way. The master that processes the dynamic updates is unavailable from generic hosts on the Internet, it is not listed in the NS RRSet, although its name appears in the SOA RRs MNAME field. The nameservers in the NS RRSet are able to receive zone updates through NOTIFY, IXFR, AXFR, or an out-of-band distribution mechanism. This approach is known as the "hidden master" setup.

The ideal situation is to have a one-way information flow to the network to avoid the possibility of tampering from the network. Keeping the zone master file on-line on the network and simply cycling it through an off-line signer does not do this. The on-line version could still be tampered with if the host it resides on is compromised. For maximum security, the master copy of the zone file should be off-net and should not be updated based on an unsecured network mediated communication.

The ideal situation may not be achievable because of economic tradeoffs between risks and costs. For instance, keeping a zone file off-line is not practical and will increase the costs of operating a DNS zone. So in practice the machines on which zone files are maintained will be connected to a network. Operators are advised to take security measures to shield unauthorized access to the master copy in order to prevent modification of DNS data before its signed.

Similarly the choice for storing a private key in a HSM will be influenced by a tradeoff between various concerns:

• The risks that an unauthorized person has unnoticed read-access to the private key
• The remaining window of opportunity for the attacker.
• The economic impact of the possible attacks (for a TLD that impact will in most cases be higher than for an individual users).
• The costs of rolling the (compromised) keys: whereby the costs of roling a ZSK is lowest and the costs of rolling a KSK that is in wide use as a trust anchor is highest
• The costs of buying and maintaining an HSM.

For dynamically updated secured zones [10] (Wellington, B., “Secure Domain Name System (DNS) Dynamic Update,” November 2000.), both the master copy and the private key that is used to update signatures on updated RRs will need to be on-line.

3.1.4.4.  Key Generation

Careful generation of all keys is a sometimes overlooked but absolutely essential element in any cryptographically secure system. The strongest algorithms used with the longest keys are still of no use if an adversary can guess enough to lower the size of the likely key space so that it can be exhaustively searched. Technical suggestions for the generation of random keys will be found in RFC 4086 (Eastlake, D., Schiller, J., and S. Crocker, “Randomness Requirements for Security,” June 2005.) [13] and NIST SP 800-900 (Barker, E. and J. Kelsey, “Recommendation for Random Number Generation Using Deterministic Random Bit Generators (Revised),” March 2007.) [19]. One should carefully assess if the random number generator used during key generation adheres to these suggestions.

Keys with a long effectivity period are particularly sensitive as they will represent a more valuable target and be subject to attack for a longer time than short-period keys. It is strongly recommended that long-term key generation occur off-line in a manner isolated from the network via an air gap or, at a minimum, high-level secure hardware.

3.1.4.5.  Differentiation for 'High-Level' Zones?

In an earlier version of this document (RFC4641 (Kolkman, O. and R. Gieben, “DNSSEC Operational Practices,” September 2006.) [15]) we made a differentiation between KSKs used for zones that are high in the DNS hierarchy versus KSKs used for zones low in that hierarchy.

Longer keys are not useful because the crypto guidance is that everyone should use keys that no one can break. Also, it is impossible to judge which zones are more or less valuable to an attacker. An attack can only be used if the compromise is unnoticed and the attacker can act as an man-in-the-middle attack (MITM) in an unnoticed way. If example. is compromised and the attacker forges answers for somebank.example. and sends them out as an MITM, when the attack is discovered it will be simple to prove that example. has been compromised and the KSK will be rolled. Defining a long-term successful attack is difficult for keys at any level.

4.  Signature Generation, Key Rollover, and Related Policies

4.1.  Key Rollovers

Regardless of whether a zone uses periodic key rollovers in order to practice for emergencies, or only rolls over keys in an emergency, key rollovers are a fact of life when using DNSSEC. Zone administrators who are in the process of rolling their keys have to take into account that data published in previous versions of their zone still lives in caches. When deploying DNSSEC, this becomes an important consideration; ignoring data that may be in caches may lead to loss of service for clients.

The most pressing example of this occurs when zone material signed with an old key is being validated by a resolver that does not have the old zone key cached. If the old key is no longer present in the current zone, this validation fails, marking the data "Bogus". Alternatively, an attempt could be made to validate data that is signed with a new key against an old key that lives in a local cache, also resulting in data being marked "Bogus".

4.1.1.  Zone Signing Key Rollovers

If the choice for splitting zone and key signing keys has been made than those two types of keys can be rolled seperatly and zone signing keys can be rolled without taking into acount DS records from the parent or the configuration of such a key as trust-anchor.

For "Zone Signing Key rollovers", there are two ways to make sure that during the rollover data still cached can be verified with the new key sets or newly generated signatures can be verified with the keys still in caches. One schema, described in Section 4.1.1.2 (Double Signature Zone Signing Key Rollover), uses double signatures; the other uses key pre-publication (Section 4.1.1.1 (Pre-Publish Key Rollover)). The pros, cons, and recommendations are described in Section 4.1.1.3 (Pros and Cons of the Schemes).

4.1.1.1.  Pre-Publish Key Rollover

This section shows how to perform a ZSK rollover without the need to sign all the data in a zone twice -- the "pre-publish key rollover". This method has advantages in the case of a key compromise. If the old key is compromised, the new key has already been distributed in the DNS. The zone administrator is then able to quickly switch to the new key and remove the compromised key from the zone. Another major advantage is that the zone size does not double, as is the case with the double signature ZSK rollover. A small "how-to" for this kind of rollover can be found in Appendix B (Zone Signing Key Rollover How-To).

Pre-publish key rollover involves four stages as follows:

----------------------------------------------------------------
 initial         new DNSKEY       new RRSIGs      DNSKEY removal
----------------------------------------------------------------
 SOA0            SOA1             SOA2            SOA3
 RRSIG10(SOA0)   RRSIG10(SOA1)    RRSIG11(SOA2)   RRSIG11(SOA3)

 DNSKEY1         DNSKEY1          DNSKEY1         DNSKEY1
 DNSKEY10        DNSKEY10         DNSKEY10        DNSKEY11
                 DNSKEY11         DNSKEY11
 RRSIG1 (DNSKEY) RRSIG1 (DNSKEY)  RRSIG1(DNSKEY)  RRSIG1 (DNSKEY)
 RRSIG10(DNSKEY) RRSIG10(DNSKEY)  RRSIG11(DNSKEY) RRSIG11(DNSKEY)
----------------------------------------------------------------

Pre-Publish Key Rollover

initial:

Initial version of the zone: DNSKEY 1 is the Key Signing Key. DNSKEY 10 is used to sign all the data of the zone, the Zone Signing Key.

new DNSKEY:

DNSKEY 11 is introduced into the key set. Note that no signatures are generated with this key yet, but this does not secure against brute force attacks on the public key. The minimum duration of this pre-roll phase is the time it takes for the data to propagate to the authoritative servers plus TTL value of the key set.

new RRSIGs:

At the "new RRSIGs" stage (SOA serial 2), DNSKEY 11 is used to sign the data in the zone exclusively (i.e., all the signatures from DNSKEY 10 are removed from the zone). DNSKEY 10 remains published in the key set. This way data that was loaded into caches from version 1 of the zone can still be verified with key sets fetched from version 2 of the zone. The minimum time that the key set including DNSKEY 10 is to be published is the time that it takes for zone data from the previous version of the zone to expire from old caches, i.e., the time it takes for this zone to propagate to all authoritative servers plus the Maximum Zone TTL value of any of the data in the previous version of the zone.

DNSKEY removal:

DNSKEY 10 is removed from the zone. The key set, now only containing DNSKEY 1 and DNSKEY 11, is re-signed with the DNSKEY 1.

The above scheme can be simplified by always publishing the "future" key immediately after the rollover. The scheme would look as follows (we show two rollovers); the future key is introduced in "new DNSKEY" as DNSKEY 12 and again a newer one, numbered 13, in "new DNSKEY (II)":

    initial             new RRSIGs          new DNSKEY
   -----------------------------------------------------------------
    SOA0                SOA1                SOA2
    RRSIG10(SOA0)       RRSIG11(SOA1)       RRSIG11(SOA2)

    DNSKEY1             DNSKEY1             DNSKEY1
    DNSKEY10            DNSKEY10            DNSKEY11
    DNSKEY11            DNSKEY11            DNSKEY12
    RRSIG1(DNSKEY)      RRSIG1 (DNSKEY)     RRSIG1(DNSKEY)
    RRSIG10(DNSKEY)     RRSIG11(DNSKEY)     RRSIG11(DNSKEY)
    ----------------------------------------------------------------

    ----------------------------------------------------------------
    new RRSIGs (II)     new DNSKEY (II)
    ----------------------------------------------------------------
    SOA3                SOA4
    RRSIG12(SOA3)       RRSIG12(SOA4)

    DNSKEY1             DNSKEY1
    DNSKEY11            DNSKEY12
    DNSKEY12            DNSKEY13
    RRSIG1(DNSKEY)      RRSIG1(DNSKEY)
    RRSIG12(DNSKEY)     RRSIG12(DNSKEY)
    ----------------------------------------------------------------

Pre-Publish Key Rollover, Showing Two Rollovers

Note that the key introduced in the "new DNSKEY" phase is not used for production yet; the private key can thus be stored in a physically secure manner and does not need to be 'fetched' every time a zone needs to be signed.

4.1.1.2.  Double Signature Zone Signing Key Rollover

This section shows how to perform a ZSK key rollover using the double zone data signature scheme, aptly named "double signature rollover".

During the "new DNSKEY" stage the new version of the zone file will need to propagate to all authoritative servers and the data that exists in (distant) caches will need to expire, requiring at least the Maximum Zone TTL.

Double signature ZSK rollover involves three stages as follows:

   ----------------------------------------------------------------
   initial             new DNSKEY         DNSKEY removal
   ----------------------------------------------------------------
   SOA0                SOA1               SOA2
   RRSIG10(SOA0)       RRSIG10(SOA1)      RRSIG11(SOA2)
                       RRSIG11(SOA1)
   DNSKEY1             DNSKEY1            DNSKEY1
   DNSKEY10            DNSKEY10           DNSKEY11
                       DNSKEY11
   RRSIG1(DNSKEY)      RRSIG1(DNSKEY)     RRSIG1(DNSKEY)
   RRSIG10(DNSKEY)     RRSIG10(DNSKEY)    RRSIG11(DNSKEY)
                       RRSIG11(DNSKEY)
   ----------------------------------------------------------------

Double Signature Zone Signing Key Rollover

initial:

Initial Version of the zone: DNSKEY 1 is the Key Signing Key. DNSKEY 10 is used to sign all the data of the zone, the Zone Signing Key.

new DNSKEY:

At the "New DNSKEY" stage (SOA serial 1) DNSKEY 11 is introduced into the key set and all the data in the zone is signed with DNSKEY 10 and DNSKEY 11. The rollover period will need to continue until all data from version 0 of the zone has expired from remote caches. This will take at least the Maximum Zone TTL of version 0 of the zone.

DNSKEY removal:

DNSKEY 10 is removed from the zone. All the signatures from DNSKEY 10 are removed from the zone. The key set, now only containing DNSKEY 11, is re-signed with DNSKEY 1.

At every instance, RRSIGs from the previous version of the zone can be verified with the DNSKEY RRSet from the current version and the other way around. The data from the current version can be verified with the data from the previous version of the zone. The duration of the "new DNSKEY" phase and the period between rollovers should be at least the Maximum Zone TTL.

Making sure that the "new DNSKEY" phase lasts until the signature expiration time of the data in the initial version of the zone is recommended. This way all caches are cleared of the old signatures. However, this duration could be considerably longer than the Maximum Zone TTL, making the rollover a lengthy procedure.

Note that in this example we assumed that the zone was not modified during the rollover. New data can be introduced in the zone as long as it is signed with both keys.

4.1.1.3.  Pros and Cons of the Schemes

Pre-publish key rollover:

This rollover does not involve signing the zone data twice. Instead, before the actual rollover, the new key is published in the key set and thus is available for cryptanalysis attacks. A small disadvantage is that this process requires four steps. Also the pre-publish scheme involves more parental work when used for KSK rollovers as explained in Section 4.1.3 (Difference Between ZSK and KSK Rollovers).

Double signature ZSK rollover:

The drawback of this signing scheme is that during the rollover the number of signatures in your zone doubles; this may be prohibitive if you have very big zones. An advantage is that it only requires three steps.

4.1.2.  Key Signing Key Rollovers

For the rollover of a Key Signing Key, the same considerations as for the rollover of a Zone Signing Key apply. However, we can use a double signature scheme to guarantee that old data (only the apex key set) in caches can be verified with a new key set and vice versa. Since only the key set is signed with a KSK, zone size considerations do not apply.

--------------------------------------------------------------------
    initial        new DNSKEY        DS change       DNSKEY removal
--------------------------------------------------------------------
  Parent:
    SOA0           -------->         SOA1            -------->
    RRSIGpar(SOA0) -------->         RRSIGpar(SOA1)  -------->
    DS1            -------->         DS2             -------->
    RRSIGpar(DS)   -------->         RRSIGpar(DS)    -------->


  Child:
    SOA0            SOA1             -------->       SOA2
    RRSIG10(SOA0)   RRSIG10(SOA1)    -------->       RRSIG10(SOA2)
                                     -------->
    DNSKEY1         DNSKEY1          -------->       DNSKEY2
                    DNSKEY2          -------->
    DNSKEY10        DNSKEY10         -------->       DNSKEY10
    RRSIG1 (DNSKEY) RRSIG1 (DNSKEY)  -------->       RRSIG2 (DNSKEY)
                    RRSIG2 (DNSKEY)  -------->
    RRSIG10(DNSKEY) RRSIG10(DNSKEY)  -------->       RRSIG10(DNSKEY)
--------------------------------------------------------------------

Stages of Deployment for a Double Signature Key Signing Key Rollover

initial:

Initial version of the zone. The parental DS points to DNSKEY1. Before the rollover starts, the child will have to verify what the TTL is of the DS RR that points to DNSKEY1 -- it is needed during the rollover and we refer to the value as TTL_DS.

new DNSKEY:

During the "new DNSKEY" phase, the zone administrator generates a second KSK, DNSKEY2. The key is provided to the parent, and the child will have to wait until a new DS RR has been generated that points to DNSKEY2. After that DS RR has been published on all servers authoritative for the parent's zone, the zone administrator has to wait at least TTL_DS to make sure that the old DS RR has expired from caches.

DS change:

The parent replaces DS1 with DS2.

DNSKEY removal:

DNSKEY1 has been removed.

The scenario above puts the responsibility for maintaining a valid chain of trust with the child. It also is based on the premise that the parent only has one DS RR (per algorithm) per zone. An alternative mechanism has been considered. Using an established trust relation, the interaction can be performed in-band, and the removal of the keys by the child can possibly be signaled by the parent. In this mechanism, there are periods where there are two DS RRs at the parent. Since at the moment of writing the protocol for this interaction has not been developed, further discussion is out of scope for this document.

The scenario scetched above assumes that the KSK is not in use as a trust-anchor too. If that is the case then special care need to be taken. For instance, when RFC5011 type rollover is in use then the DNSKEY1 removal phase above is the moment that the revoke flag is set on DNSKEY1 while it is still published, at least as long as the RFC5011 holdback timer proscribes. Only after that timer expired DNSKEY1 can be removed.

4.1.3.  Difference Between ZSK and KSK Rollovers

Note that KSK rollovers and ZSK rollovers are different in the sense that a KSK rollover requires interaction with the parent (and possibly replacing of trust anchors) and the ensuing delay while waiting for it.

A zone key rollover can be handled in two different ways: pre-publish (Section 4.1.1.1 (Pre-Publish Key Rollover)) and double signature (Section 4.1.1.2 (Double Signature Zone Signing Key Rollover)).

As the KSK is used to validate the key set and because the KSK is not changed during a ZSK rollover, a cache is able to validate the new key set of the zone. The pre-publish method would also work for a KSK rollover. The records that are to be pre-published are the parental DS RRs. The pre-publish method has some drawbacks for KSKs. We first describe the rollover scheme and then indicate these drawbacks.

--------------------------------------------------------------------
  initial         new DS           new DNSKEY      DS/DNSKEY removal
--------------------------------------------------------------------
Parent:
  SOA0            SOA1             -------->       SOA2
  RRSIGpar(SOA0)  RRSIGpar(SOA1)   -------->       RRSIGpar(SOA2)
  DS1             DS1              -------->       DS2
                  DS2              -------->
  RRSIGpar(DS)    RRSIGpar(DS)     -------->       RRSIGpar(DS)

Child:
  SOA0            -------->        SOA1            SOA1
  RRSIG10(SOA0)   -------->        RRSIG10(SOA1)   RRSIG10(SOA1)
                  -------->
  DNSKEY1         -------->        DNSKEY2         DNSKEY2
                  -------->
  DNSKEY10        -------->        DNSKEY10        DNSKEY10
  RRSIG1 (DNSKEY) -------->        RRSIG2(DNSKEY)  RRSIG2 (DNSKEY)
  RRSIG10(DNSKEY) -------->        RRSIG10(DNSKEY) RRSIG10(DNSKEY)
--------------------------------------------------------------------

Stages of Deployment for a Pre-Publish Key Signing Key Rollover

When the child zone wants to roll, it notifies the parent during the "new DS" phase and submits the new key (or the corresponding DS) to the parent. The parent publishes DS1 and DS2, pointing to DNSKEY1 and DNSKEY2, respectively. During the rollover ("new DNSKEY" phase), which can take place as soon as the new DS set propagated through the DNS, the child replaces DNSKEY1 with DNSKEY2. Immediately after that ("DS/DNSKEY removal" phase), it can notify the parent that the old DS record can be deleted.

The drawbacks of this scheme are that during the "new DS" phase the parent cannot verify the match between the DS2 RR and DNSKEY2 using the DNS -- as DNSKEY2 is not yet published. Besides, we introduce a "security lame" key (see Section 4.3.3 (Security Lameness)). Finally, the child-parent interaction consists of two steps. The "double signature" method only needs one interaction.

4.1.4.  Rollover for a Single Type Signing Key rollover

The rollover of a DNSKEY when a Single Type Signing scheme is used is subject to the same requirement as the rollover of a KSK or ZSK: During any stage of the rollover the chain of trust needs to continue to validate for any combination of data in the zone as well as data that may still live in distant caches.

There are two variants for this rollover. Since the choice for a Single Type Signing scheme is motivated by operational simplicity we first describe the most straightforward rollover scheme first.

   -----------------------------------------------------------------
   initial           new DNSKEY       DS change      DNSKEY removal
   -----------------------------------------------------------------
 Parent:
   SOA0             -------->        SOA1            -------->
   RRSIGpar(SOA0)   -------->        RRSIGpar(SOA1)  -------->
   DS1              -------->        DS2             -------->
   RRSIGpar(DS)     -------->        RRSIGpar(DS)    -------->


 Child:
   SOA0             SOA1             ------------>    SOA2
   RRSIG1(SOA0)     RRSIG1(SOA1)     ------------>    RRSIG2(SOA2)
                    RRSIG2(SOA1)     ------------>
   DNSKEY1          DNSKEY1          ------------>    DNSKEY2
                    DNSKEY2          ------------>
   RRSIG1(DNSKEY)   RRSIG1(DNSKEY)   ------------>    RRSIG2(DNSKEY)
                    RRSIG2(DNSKEY)   ------------>
   -----------------------------------------------------------------

Stages of the Straightforward rollover in a Single Type Signing scheme.

initial:

Parental DS points to DNSKEY1. All RR sets in the zone are signed with DNSKEY1.

new DNSKEY:

A new key (DNSKEY2) is introduced and all the RR sets are signed with both DNSKEY1 and DNSKEY2.

DS change:

After the DNSKEY RRset with the two keys had time to propagate into distant caches (that is the key set exclusively containing DNSKEY1 has been expired) the parental DS record can be changed.

DNSKEY removal:

After the DS RRset containing DS1 has expired from distant caches DNSKEY1 can be removed from the DNSKEY RRset .

There is a second variety of this rollover during which one introduces a new DNSKEY into the key set and signs the keyset with both keys while signing the zone data with only the original DNSKEY1. One replaces the DNSKEY1 signatures with signatures made with DNSKEY2 at the moment of DNSKEY1 removal.

The second variety of this rollover can be considered when zone size considerations prevent the introduction of double signatures over all of the zone data although in that case choosing for a KSK/ZSK split may be a better option.

A double DS rollover scheme is compatible with a rollover using a Single Type signing scheme although in order to maintain a valid chain of trust the zone data would need to be published with a double signatures or a double key key set would need to be published. Since this leads to increase in zone and packet size at both child and parent there are little benefits to a double DS rollover with a Single Type signing scheme.

4.1.5.  Key algorithm rollover

A special class of keyrollover is the rollover of key algorithms (either adding a new algorithm, removing an old algorithm, or both), additional steps are needed to retain integrity during the rollover.

Because of the algorithm downgrade protection in RFC4035 section 2.2, you may not have a key of an algorithm for which you do not have signatures.

When adding a new algorithm, the signatures should be added first. After the TTL has expired, and caches have dropped the old data covered by those signatures, the DNSKEY with the new algorithm can be added. When removing an old algorithm, the DNSKEY should be removed first.

The following figure describes the steps. Whereby the trailing underscored number indicates the algorithm and ZSK and KSK indicate the obvious difference in key use. For example DNSKEY_KSK_1 is a the DNSKEY RR representing the public part of a key signing key of algorithm type 1 while RRSIG_ZSK_2(SOA3) is the RRSIG RR made with the private part of a zone signing key of algorithm type 2 over a SOA RR (that has serialnumber 3).

----------------------------------------------------------------
1 Initial            2 New RRSIGS         3 New DNSKEY
----------------------------------------------------------------
SOA0                 SOA1                 SOA2
RRSIG_ZSK_1(SOA0)    RRSIG_ZSK_1(SOA1)    RRSIG_ZSK_1(SOA2)
                     RRSIG_ZSK_2(SOA1)    RRSIG_ZSK_2(SOA2)

DNSKEY_KSK_1         DNSKEY_KSK_1         DNSKEY_KSK_1
DNSKEY_ZSK_1         DNSKEY_ZSK_1         DNSKEY_ZSK_1
RRSIG_KSK_1(DNSKEY)  RRSIG_KSK_1(DNSKEY)  DNSKEY_KSK_2
                     RRSIG_KSK_2(DNSKEY)  DNSKEY_ZKS_2
                                          RRSIG_KSK_1(DNSKEY)
                                          RRSIG_KSK_2(DNSKEY)
----------------------------------------------------------------
4 Remove DNSKEY      5 Remove RRSIGS
----------------------------------------------------------------
SOA3                 SOA4
RRSIG_ZSK_1(SOA3)    RRSIG_ZSK_2(SOA4)
RRSIG_ZSK_2(SOA3)

DNSKEY_KSK_2         DNSKEY_KSK_2
DNSKEY_ZSK_2         DNSKEY_ZSK_2
RRSIG_KSK_1(DNSKEY)  RRSIG_KSK_2(DNSKEY)
RRSIG_KSK_2(DNSKEY)
----------------------------------------------------------------

Stages of Deployment during an Algorithm Rollover.

In step 2, the signatures for the new key are added, but the key itself is not. While in theory, the signatures of the keyset should always be synchronized with the keyset itself, it can be possible that RRSIGS are requested separately, so it might be prudent to also sign the DNSKEY set with the new signature.

After the cache data has expired, the new key can be added to the zone, as done in step 3.

The next step is to remove the old algorithm. This time the key needs to be removed first, before removing the signatures. The key is removed in step 4, and after the cache data has expired, the signatures can be removed in step 5.

If a parent needs to roll a DS record that step can be done during stage 3. First one has to wait until caches are populated with the content of the DNSKEY RRset that was created at the beginning of stage 3, then the parent can swap the DS from pointing to DNSKEY_KSK_1 to DNSKEY_KSK_2, and then one has to wait for the old DS to expire from distant caches before taking step 4.

[OK:The following paragrapgh is provisional and needs careful review.]

A special case is the rollover from an NSEC signed zone to an NSEC3 signed zone. In this case algortihm numbers are used to signal support for NSEC3 but they do not mandate the use of NSEC3. Therefore NSEC records should remain to be served untill the rollover to a new algorithm has completed and the new DNSKEY RR set has populated distant caches(at least one TTL into stage 4, or at any time during stage 5). At that stage the validators that have not implemented NSEC3 will treat the zone as unsecured as soon as they follow the chain of trust to DS that points to a DNSKEY of the new algorithm while validators that support NSEC3 will happily validate using NSEC. Turning on NSEC3 can then be done when moving from one serial to another, realizing that that involves a resigning of the zone and the introduction of the NSECPARAM record in order to signal authoritative servers to start serving NSEC3 authenticated denial of existence.

4.1.6.  Automated Key Rollovers

As keys must be renewed periodically, there is some motivation to automate the rollover process. Consider the following:

• ZSK rollovers are easy to automate as only the child zone is involved.
• A KSK rollover needs interaction between parent and child. Data exchange is needed to provide the new keys to the parent; consequently, this data must be authenticated and integrity must be guaranteed in order to avoid attacks on the rollover.

4.2.  Planning for Emergency Key Rollover

This section deals with preparation for a possible key compromise. Our advice is to have a documented procedure ready for when a key compromise is suspected or confirmed.

When the private material of one of your keys is compromised it can be used for as long as a valid trust chain exists. A trust chain remains intact for

• as long as a signature over the compromised key in the trust chain is valid,
• as long as a parental DS RR (and signature) points to the compromised key,
• as long as the key is anchored in a resolver and is used as a starting point for validation (this is generally the hardest to update).

While a trust chain to your compromised key exists, your namespace is vulnerable to abuse by anyone who has obtained illegitimate possession of the key. Zone operators have to make a trade-off if the abuse of the compromised key is worse than having data in caches that cannot be validated. If the zone operator chooses to break the trust chain to the compromised key, data in caches signed with this key cannot be validated. However, if the zone administrator chooses to take the path of a regular rollover, the malicious key holder can spoof data so that it appears to be valid.

4.2.1.  KSK Compromise

A zone containing a DNSKEY RRSet with a compromised KSK is vulnerable as long as the compromised KSK is configured as trust anchor or a parental DS points to it.

A compromised KSK can be used to sign the key set of an attacker's zone. That zone could be used to poison the DNS.

Therefore, when the KSK has been compromised, the trust anchor or the parental DS should be replaced as soon as possible. It is local policy whether to break the trust chain during the emergency rollover. The trust chain would be broken when the compromised KSK is removed from the child's zone while the parent still has a DS pointing to the compromised KSK (the assumption is that there is only one DS at the parent. If there are multiple DSes this does not apply -- however the chain of trust of this particular key is broken).

Note that an attacker's zone still uses the compromised KSK and the presence of a parental DS would cause the data in this zone to appear as valid. Removing the compromised key would cause the attacker's zone to appear as valid and the child's zone as Bogus. Therefore, we advise not to remove the KSK before the parent has a DS to a new KSK in place.

4.2.1.1.  Keeping the Chain of Trust Intact

If we follow this advice, the timing of the replacement of the KSK is somewhat critical. The goal is to remove the compromised KSK as soon as the new DS RR is available at the parent. And also make sure that the signature made with a new KSK over the key set with the compromised KSK in it expires just after the new DS appears at the parent, thus removing the old cruft in one swoop.

The procedure is as follows:

1. Introduce a new KSK into the key set, keep the compromised KSK in the key set.
2. Sign the key set, with a short validity period. The validity period should expire shortly after the DS is expected to appear in the parent and the old DSes have expired from caches.
3. Upload the DS for this new key to the parent.
4. Follow the procedure of the regular KSK rollover: Wait for the DS to appear in the authoritative servers and then wait as long as the TTL of the old DS RRs. If necessary re-sign the DNSKEY RRSet and modify/extend the expiration time.
5. Remove the compromised DNSKEY RR from the zone and re-sign the key set using your "normal" validity interval.

An additional danger of a key compromise is that the compromised key could be used to facilitate a legitimate DNSKEY/DS rollover and/or nameserver changes at the parent. When that happens, the domain may be in dispute. An authenticated out-of-band and secure notify mechanism to contact a parent is needed in this case.

Note that this is only a problem when the DNSKEY and or DS records are used for authentication at the parent.

4.2.1.2.  Breaking the Chain of Trust

There are two methods to break the chain of trust. The first method causes the child zone to appear 'Bogus' to validating resolvers. The other causes the child zone to appear 'insecure'. These are described below.

In the method that causes the child zone to appear 'Bogus' to validating resolvers, the child zone replaces the current KSK with a new one and re-signs the key set. Next it sends the DS of the new key to the parent. Only after the parent has placed the new DS in the zone is the child's chain of trust repaired.

An alternative method of breaking the chain of trust is by removing the DS RRs from the parent zone altogether. As a result, the child zone would become insecure.

4.2.2.  ZSK Compromise

Primarily because there is no parental interaction required when a ZSK is compromised, the situation is less severe than with a KSK compromise. The zone must still be re-signed with a new ZSK as soon as possible. As this is a local operation and requires no communication between the parent and child, this can be achieved fairly quickly. However, one has to take into account that just as with a normal rollover the immediate disappearance of the old compromised key may lead to verification problems. Also note that as long as the RRSIG over the compromised ZSK is not expired the zone may be still at risk.

4.2.3.  Compromises of Keys Anchored in Resolvers

A key can also be pre-configured in resolvers. For instance, if DNSSEC is successfully deployed the root key may be pre-configured in most security aware resolvers.

If trust-anchor keys are compromised, the resolvers using these keys should be notified of this fact. Zone administrators may consider setting up a mailing list to communicate the fact that a SEP key is about to be rolled over. This communication will of course need to be authenticated, e.g., by using digital signatures.

End-users faced with the task of updating an anchored key should always validate the new key. New keys should be authenticated out-of-band, for example, through the use of an announcement website that is secured using secure sockets (TLS) [22] (Dierks, T. and E. Rescorla, “The Transport Layer Security (TLS) Protocol Version 1.2,” August 2008.).

4.3.  Parental Policies

4.3.1.  Initial Key Exchanges and Parental Policies Considerations

The initial key exchange is always subject to the policies set by the parent. The initial key exchange is always subject to the policies set by the parent. This is specifically important in a registry-registrar model where the key material is to be passed from the DNS operator, via a registrar to the (parent) registry, where both DNS operator and registrar is selected by the registrant and they might be different organisations. When designing a key exchange policy one should take into account that the authentication and authorization mechanisms used during a key exchange should be as strong as the authentication and authorization mechanisms used for the exchange of delegation information between parent and child. That is, there is no implicit need in DNSSEC to make the authentication process stronger than it was in DNS.

Using the DNS itself as the source for the actual DNSKEY material, with an out-of-band check on the validity of the DNSKEY, has the benefit that it reduces the chances of user error. A DNSKEY query tool can make use of the SEP bit [5] (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Protocol Modifications for the DNS Security Extensions,” March 2005.) to select the proper key from a DNSSEC key set, thereby reducing the chance that the wrong DNSKEY is sent. It can validate the self-signature over a key; thereby verifying the ownership of the private key material. Fetching the DNSKEY from the DNS ensures that the chain of trust remains intact once the parent publishes the DS RR indicating the child is secure.

Note: the out-of-band verification is still needed when the key material is fetched via the DNS. The parent can never be sure whether or not the DNSKEY RRs have been spoofed.

4.3.2.  Storing Keys or Hashes?

When designing a registry system one should consider which of the DNSKEYs and/or the corresponding DSes to store. Since a child zone might wish to have a DS published using a message digest algorithm not yet understood by the registry, the registry can't count on being able to generate the DS record from a raw DNSKEY. Thus, we recommend that registry systems at least support storing DS records.

It may also be useful to store DNSKEYs, since having them may help during troubleshooting and, as long as the child's chosen message digest is supported, the overhead of generating DS records from them is minimal. Having an out-of-band mechanism, such as a registry directory (e.g., Whois), to find out which keys are used to generate DS Resource Records for specific owners and/or zones may also help with troubleshooting.

The storage considerations also relate to the design of the customer interface and the method by which data is transferred between registrant and registry; Will the child zone administrator be able to upload DS RRs with unknown hash algorithms or does the interface only allow DNSKEYs? When Registries support the Extensible Provisioning Protocol (EPP) [14] (Hollenbeck, S., “Domain Name System (DNS) Security Extensions Mapping for the Extensible Provisioning Protocol (EPP),” December 2005.) then that can be used for registrar-registry interactions since that protocol allows the transfer of DS and optionally DNSKEY RRs. For moving data between DNS operator and registrar there is no standardized way for moving the data. Different registrars have different mechanisms, ranging from simple web interfaces to various APIs. In some cases the use of the DNSSEC extentions to EPP may be aplicable to.

4.3.3.  Security Lameness

Security lameness is defined as what happens when a parent has a DS RR pointing to a non-existing DNSKEY RR. When this happens, the child's zone may be marked "Bogus" by verifying DNS clients.

As part of a comprehensive delegation check, the parent could, at key exchange time, verify that the child's key is actually configured in the DNS. However, if a parent does not understand the hashing algorithm used by child, the parental checks are limited to only comparing the key id.

Child zones should be very careful in removing DNSKEY material, specifically SEP keys, for which a DS RR exists.

Once a zone is "security lame", a fix (e.g., removing a DS RR) will take time to propagate through the DNS.

4.3.4.  DS Signature Validity Period

Since the DS can be replayed as long as it has a valid signature, a short signature validity period over the DS minimizes the time a child is vulnerable in the case of a compromise of the child's KSK(s). A signature validity period that is too short introduces the possibility that a zone is marked "Bogus" in case of a configuration error in the signer. There may not be enough time to fix the problems before signatures expire. Something as mundane as operator unavailability during weekends shows the need for DS signature validity periods longer than 2 days. We recommend an absolute minimum for a DS signature validity period of a few days.

The maximum signature validity period of the DS record depends on how long child zones are willing to be vulnerable after a key compromise. On the other hand, shortening the DS signature validity interval increases the operational risk for the parent. Therefore, the parent may have policy to use a signature validity interval that is considerably longer than the child would hope for.

A compromise between the operational constraints of the parent and minimizing damage for the child may result in a DS signature validity period somewhere between a week and months.

In addition to the signature validity period, which sets a lower bound on the number of times the zone owner will need to sign the zone data and which sets an upper bound to the time a child is vulnerable after key compromise, there is the TTL value on the DS RRs. Shortening the TTL means that the authoritative servers will see more queries. But on the other hand, a short TTL lowers the persistence of DS RRSets in caches thereby increasing the speed with which updated DS RRSets propagate through the DNS.

4.3.5.  Changing DNS Operators

[OK: No feedback received between version 2 and 3. This is likely to remain]

4.3.5.1.  Cooperationg DNS operators

The parent-child relation is often described in terms of a (thin) registry model. Where a registry maintains the parent zone, and the registrant (the user of the child-domain name), deals with the registry through an intermediary called a registrar. (See [11] (Hollenbeck, S., “Generic Registry-Registrar Protocol Requirements,” September 2002.) for a comprehensive definition). Registrants may out-source the maintenance of their DNS system, including the maintenance of DNSSEC key material, to the registrar or to another third party, which we will call the DNS operator. The DNS operator that has control over the DNS zone and its keys may prevent the registrant to make a timely move to a different DNS operator.

Suppose that the registrant wants to move from a losing DNS operator (loosing operator for short) to a gaining DNS operator (gaining operator). Let us first look what would happen in a cooperative environment. The assumption is that the loosing operator will not hand off any private key material to the gaining operator, that would constitute a trivial case.

In a cooperating environment one could proceed with a pre-publish ZSK rollover whereby the loosing operator pre-publishes the ZSK of the gaining operator, combined with a double signature KSK rollover where the two registrars exchange public KSKs and independently generate a signature over those keysets that they combine and both publish in their copy of the zone. Once that is done they can use their own private keys to sign any of their zone content during the transfer.

 ............................................................
 intitial          |        pre-publish             |
 ............................................................
 Parent:
  NSA                            NSA
  DSA                            DSA

 ............................................................
Child at A:             Child at A:    Child at B:
  ZSKA                   ZSKA            ZSKA
  KSKA                   ZSKB            ZSKB
  RRSIGZA(DNSKEY)        KSKA            KSKA
  RRSIGKA(DNSKEY)        KSKB            KSKB
                         RRSIGZB         RRSIGZB
                         RRSIGKB         RRSIGKB
                         RRSIGZA         RRSIGZA
                         RRSIGKA         RRSIGKA


  SOAA                   SOAA             SOAB
  RRSIGZA(SOA)           RRSIGZA(SOA)     RRSIGZB(SOA)

  NSA                    NSA              NSB
  RRSIGZA(NS)            NSB              RRSIGZB(NS)
                         RRSIGZA(NS)
 ............................................................

 ............................................................
       Redelegation          |        post migration        |
 ............................................................

           NSB                             NSB
           DSB                             DSB

 ............................................................
 Child at A:    Child at B:             Child at B:
  ZSKA            ZSKA                   ZSKB
  ZSKB            ZSKB                   KSKB
  KSKA            KSKA                   RRSIGZB(DNSKEY)
  KSKB            KSKB                   RRSIGKB(DNSKEY)
  RRSIGZB         RRSIGZB
  RRSIGKB         RRSIGKB
  RRSIGZA         RRSIGZA
  RRSIGKA         RRSIGKA


  SOAA             SOAB                  SOAB
  RRSIGZA(SOA)     RRSIGZB(SOA)          RRSIGZB(SOA)

  NSA              NSB                   NSB
  NSB              RRSIGZB(NS)           RRSIGZB(NS)
  RRSIGZA(NS)
 ............................................................

In this figure A denotes the loosing DNS operator and B the gaining DNS operator. RRSIGZ is the RRSIG produced by a ZSK, RRSIGK is produced with a KSK, the appendend A or B indicate the producers of the key pair. Child at A is how the zone content is represented by the loosing DNS operator and Child at B is how the zone content is represented by the gaining DNS operator.

4.3.5.2.  Non Cooperationg DNS Operators

In the non-cooperative case matters are more complicated. The loosing operator may not cooperate and leave the data in the DNS as is. In the extreme case the loosing operator may become obstructive and publish a DNSKEY RR with a high TTL and corresponding signature validity so that registrar A's DNSKEY, would end up in caches for, in theory, tens of years.

The problem arises when a validator tries to validate with the loosing operator's key and there is no signature material produced with the loosing operator available in the delegation path after redelegation from the loosing operator to the gaiing operator has taken place. One could imagine a rollover scenario where the gaining operator pulls all RRSIGs created by the loosing operator and publishes those in conjunction with its own signatures, but that would not allow any changes in the zone content. Since a redelegation took place the NS RRset has -- per definition-- changed so such rollover scenario will not work. Besides if zone transfers are not allowed by the loosing operator and NSEC3 is deployed in the loosing operator's zone then the gainging operator's zone will not have certainty that all of A's RRSIGs are transfered.

The only viable option for the registrant is to publish its zone unsigned and ask the registry to remove the DS pointing to the loosing operator's DNSKEY for as long as the DNSKEY of the loosing operator, or any of the signatures produced by it are likely to disappear in caches, which as mentioned above could in theory be for tens of years.

Note that some [OK: most/all ?] implementations limit the time DNSKEYs that seem to be unable to validate signatures are cached and/or will try to recover from cases where DNSKEYs do not seem to be able to validate data. Although that is not a protocol requirement it seems that that practice may limit the impact of this problem the problem of non-cooperating registrars.

However, there is no operational methodology to work around this business issue, and proper contractual relationships between all involved parties seems to be the only solution to cope with these problems. It should be noted that in many cases, the problem with temporary broken delegations already exists when a zone changes from one DNS operator to another. Besides, it is often the case that when the DNS moves from one operator to another the services that that zone references also change operator, possibly involving some downtime.

In any case. To minimise such problems, the classic recommendation is to have relative short TTL on all involved resource records. That will solve many of the problems regarding changes to a zone regardless of whether DNSSEC is used.

4.4.  Time in DNSSEC

Without DNSSEC, all times in the DNS are relative. The SOA fields REFRESH, RETRY, and EXPIRATION are timers used to determine the time elapsed after a slave server synchronized with a master server. The Time to Live (TTL) value and the SOA RR minimum TTL parameter [9] (Andrews, M., “Negative Caching of DNS Queries (DNS NCACHE),” March 1998.) are used to determine how long a forwarder should cache data after it has been fetched from an authoritative server. By using a signature validity period, DNSSEC introduces the notion of an absolute time in the DNS. Signatures in DNSSEC have an expiration date after which the signature is marked as invalid and the signed data is to be considered Bogus.

The considerations in this section are all qualitative and focused on the operational and managerial issues. A more thourhough quantitative analysis of rollover timing parameters can be found in draft-ietf-dnsop-dnssec-key-timing (Morris, S., Ihren, J., and J. Dickinson, “DNSSEC Key Timing Considerations,” July 2010.) [24]

4.4.1.  Time Considerations

Because of the expiration of signatures, one should consider the following:

• We suggest the Maximum Zone TTL of your zone data to be a fraction of your signature validity period.

If the TTL would be of similar order as the signature validity period, then all RRSets fetched during the validity period would be cached until the signature expiration time. Section 7.1 of RFC4033 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “DNS Security Introduction and Requirements,” March 2005.) [3] suggests that "the resolver may use the time remaining before expiration of the signature validity period of a signed RRSet as an upper bound for the TTL". As a result, query load on authoritative servers would peak at signature expiration time, as this is also the time at which records simultaneously expire from caches.

To avoid query load peaks, we suggest the TTL on all the RRs in your zone to be at least a few times smaller than your signature validity period.

• We suggest the signature publication period to end at least one Maximum Zone TTL duration before the end of the signature validity period.

Re-signing a zone shortly before the end of the signature validity period may cause simultaneous expiration of data from caches. This in turn may lead to peaks in the load on authoritative servers. To avoid this schemes are deployed whereby the zone is periodically visited for a resigning operation and those signatures that are within a so called refresh intervall from signature expiration are recreated. Also see Section 4.4.2 (Signature Validation Periods) below.

• We suggest the Minimum Zone TTL to be long enough to both fetch and verify all the RRs in the trust chain. In workshop environments, it has been demonstrated [18] (Rose, S., “NIST DNSSEC workshop notes,” June 2001.) that a low TTL (under 5 to 10 minutes) caused disruptions because of the following two problems:

1. During validation, some data may expire before the validation is complete. The validator should be able to keep all data until it is completed. This applies to all RRs needed to complete the chain of trust: DSes, DNSKEYs, RRSIGs, and the final answers, i.e., the RRSet that is returned for the initial query.

2. Frequent verification causes load on recursive nameservers. Data at delegation points, DSes, DNSKEYs, and RRSIGs benefit from caching. The TTL on those should be relatively long.

• Slave servers will need to be able to fetch newly signed zones well before the RRSIGs in the zone served by the slave server pass their signature expiration time.

When a slave server is out of sync with its master and data in a zone is signed by expired signatures, it may be better for the slave server not to give out any answer.

Normally, a slave server that is not able to contact a master server for an extended period will expire a zone. When that happens, the server will respond differently to queries for that zone. Some servers issue SERVFAIL, whereas others turn off the 'AA' bit in the answers. The time of expiration is set in the SOA record and is relative to the last successful refresh between the master and the slave servers. There exists no coupling between the signature expiration of RRSIGs in the zone and the expire parameter in the SOA.

If the server serves a DNSSEC zone, then it may well happen that the signatures expire well before the SOA expiration timer counts down to zero. It is not possible to completely prevent this from happening by tweaking the SOA parameters.

However, the effects can be minimized where the SOA expiration time is equal to or shorter than the signature validity period.

The consequence of an authoritative server not being able to update a zone, whilst that zone includes expired signatures, is that non-secure resolvers will continue to be able to resolve data served by the particular slave servers while security-aware resolvers will experience problems because of answers being marked as Bogus.

We suggest the SOA expiration timer being approximately one third or one fourth of the signature validity period. It will allow problems with transfers from the master server to be noticed before the actual signature times out.

We also suggest that operators of nameservers that supply secondary services develop 'watch dogs' to spot upcoming signature expirations in zones they slave, and take appropriate action.

When determining the value for the expiration parameter one has to take the following into account: What are the chances that all my secondaries expire the zone? How quickly can I reach an administrator of secondary servers to load a valid zone? These questions are not DNSSEC specific but may influence the choice of your signature validity intervals.

4.4.2.  Signature Validation Periods

[OK: This section is newly introduced and needs a check on consistency with the rest of the document]

4.4.2.1.  Maximum Value

The first consideration for choosing a maximum signature validity period is the risk of a replay attack. For low-value, long-term stable resources the risks may be minimal and the signature validity period may be multiple months. Although signature validity periods of many years are allowed the same operational habit arguments as in Section 3.1.2.2 (Rolling a KSK that is a trust-anchor) play a role: when a zone is resigned with some regularity then operators remain conscious about the operational necessity of resigning.

4.4.2.2.  Minimum Value

The minimum value of the signature validity period is set by the time by which one would like to survive operational failure in provisioning: what is the time that a failure will be noticed, what is the time that action is expected to be taken? By answering these questions availability of operators during (long) weekends or access to backup media needs to be taken into account which could easily suggest a minimum Signature Validity period of a few days.

Note however, the argument above is assuming that zone data has just been signed and published when the problem occurred. In practice it may be that a zone is signed according to a frequency set by the Resign Period whereby the signer visits the zone content and only refreshes signatures that are close to expiring: the signer will only refresh signatures if they are within the Refresh Period from the signature expiration time. The Resign Period must be smaller than the Refresh Period in order for zone data to be signed timely.

If an operational problem occurs during resigning then the signatures in the zone wich will expire first are the ones that have been generated longest ago. And in the worst case these signatures are the Refresh Period minus the Resign Period away from signature validation.

In other words, the minimum Signature Validity intervall is set by first choosing the Refresh Period (usually a few days), then defining the Resign period in such a way that the Refresh Period minus the Resign period sets the time in which operational havoc can be resolved.

To make matters slightly mor complicated. Some signers apply some jitter to the signature expiry time so that not all signatures expire at the same time. The jitter should not influence your calculation as long as it is smaller than the refresh period and the resign period is at least half the refresh period [OK: The above needs careful review]

 Inception      Signing                                 Expiration
 time             time                                     time
 |                  |                              |       |       |
 |------------------|------------------------------|.......|.......|
 |                  |                              |       |       |
                                                       +/- jitter

 | Inception offset |         Validity Period              |
 | <--------------->|<------------------------------------>|




 Inception      Signing  reuse  reuse  reuse    new      Expiration
 time             time                          signature     time
 |                  |      |      |      |      |             |
 |------------------|-----------------------------------------|
 |                  |      |      |      |      |             |
                     <----> <----> <----> <---->
                 Resign Period

                                             |                  |
                                             |<-Refresh Period->|
                                             |                  |

4.4.2.3.  differentiation between RR sets

It is possible to vary signature validity periods between signatures over different RR sets in the zone. In practice this could be done when zones contain highly volatile data (which may be the case in dynamic update environments). Note however that the risk of replay (e.g. by stale secondary servers) is what should be leading in determining the signature validity period since the TTL on the data itself still are the primary parameter for cache expiry. [OK: are there strong arguments besides replay risks for varying signature validity]

In some cases the risk of replaying existing data might be different from the risk of replaying the denial of data. In those cases the signature validity period on NSEC or NSEC3 records may be tweaked accordingly.

When a zone contains secure delegations then a relatively short signature validity interval protects the child agains replay attacks, in the case the child's key is compromised (see Section 4.3.4 (DS Signature Validity Period)). Since there is a higher operational risk for the parent registry when choosing a short validity intervall and a higher operational risk for the child when choosing a long validity period some (price) differentiation may occur for validity periods between individual DS RRs in a single zone.

There seem to be no other arguments for differentiation in validity periods.

4.4.2.4.  Other timing parameters in a zone

[OK: Isn't the following not to vague? Is it sufficient?]

The arguments for tuning minimum signature validity period are remarkably similar to the arguments used to set the SOA expiration timer. It is advised to set the expiration time longer than the signature validity period.

5.  Next Record type

One of the design tradeofs made during the development of DNSSEC has been to seperate the signing and serving operations instead of performing cryptographic operations on the fly. It is therefore necessry to create records that cover the very large number of non-existent names that lie between the names that do exist.

There are two mechanisms to provide authenticated proof of non-existence of domain names in DNSSEC: clear text one and an obfuscated-data one. Each mechanism

• includes a list of all the RRTYPEs present which can be used to prove the non-existence of RRTYPEs at a certain name;
• stores only the name for which the zone is authoritative (that is, glue in the zone is omitted); and
• uses a specific RRTYPE to store information about the RRTYPEs present at the name: the clear-text mechanism uses NSEC, and the obfuscated-data mechanism uses NSEC3.

5.1.  Differences between NSEC and NSEC3

The clear text mechanism (NSEC) is implemented using a sorted linked list of names in the zone. The obfuscated-data mechanism (NSEC3) first hashes the names using a one-way hash function, and then sorts the resulting (hashed) strings.

The NSEC record requires no cryptographic operations aside from validating its associated signature record. It is human readable and can be used in manual queries to determine correct operation. The disadvantage is that it allows for "zone walking", where one can request all the entries of a zone by following the next RRlabel pointed to in each subsequent NSEC record.

Though all agree DNS data is accessible through query mechanisms, a side effect of NSEC is that it allows the contents of a zone file to be enumerated in full by sequential queries. Whilst for some operators this behaviour is acceptable or even desirable, for others it is undesirable for policy, regulatory or other reasons. This is the first difference between NSEC and NSEC3.

The second difference between NSEC and NSEC3 is that NSEC requires a signature over every RR in the zonefile, thereby ensuring that any denial of existence is cryptographically signed. However, in a large zonefile containing many delegations very few of which are to signed zones, this may produce unacceptable additional overhead especially where insecure delegations are subject to frequent update (a typical example might be a TLD operator with few registrants using secure delegations). NSEC3 allows intervals between two such delegations to "Opt-out" in which case they may contain one more more insecure delegations, thus reducing the size and cryptographic complexity of the zone at the expense of the ability to cryptographically deny the existence of names in a specific span.

The NSEC3 record uses a hashing method of the requested RRlabel. To increase the workload required to guess entries in the zone, the number of hashing iteration's can be specified in the NSEC3 record. Additionally, a salt can be specified that also modifies the hashes. Note that NSEC3 does not give full protection against information leakage from the zone.

5.2.  NSEC or NSEC3

The first motivation to deploy NSEC3, prevention of zone enumeration, only makes sense when zone content is not highly structured or trivially guessable. Highly structured zones such as the in-addr.arpa, ip6.arpa and e164.arpa can be trivially enumerated using ordinary DNS properties while for small zones that only contain contain records in the APEX and a few common RRlabels such as "www" or "mail" guessing zone content and proving completeness is also trivial when using NSEC3.

In those cases the use of NSEC is recommended to ease the work required by signers and validating resolvers.

For large zones where there is an implication of "not readily available" RRlabels, such as those where one has to sign an NDA before obtaining it, NSEC3 is recommended.

The considerations for the second reason to deploy NSEC3 are discussed below (Section 5.3.4 (Opt-out)).

5.3.  NSEC3 parameters

The NSEC3 hashing algorithm is performed on the Fully Qualified Domain Name (FQDN) in its uncompressed form. This ensures brute force work done by an attacker for one (FQDN) RRlabel cannot be re-used for another (FQDN) RRlabel attack, as these entries are per definition unique.

5.3.1.  NSEC3 Algorithm

At the moment of writing there is only one NSEC3 Hashing algorithm defined. [21] (Laurie, B., Sisson, G., Arends, R., and D. Blacka, “DNS Security (DNSSEC) Hashed Authenticated Denial of Existence,” March 2008.) specifically calls out that when a new hash algorithm for use with NSEC3 is specified, a transition mechanism MUST also be defined. Therefore this document does not considder NSEC3 hash algorithm transition.

5.3.2.  NSEC3 Iterations

One of the concerns with NSEC3 is that bad actors could perform a pre-calculated dictionary attack in order to assess if certain domain names exist within the zones or not. Two mechanisms are introduced in the NSEC3 specification to increase the costs of such dictionary attacks: Iterations and Salt.

RFC5155 (Laurie, B., Sisson, G., Arends, R., and D. Blacka, “DNS Security (DNSSEC) Hashed Authenticated Denial of Existence,” March 2008.) [21] considers the trade-offs between incuring cost during the signing process, imposing costs to the validating nameserver, while still providing a reasonable barrier against dictionary attacks. It provides useful limits of iterations compared to RSA key size. These are 150 iterations for 1024 bit keys, 500 iterations for 2048 bit keys and 2,500 iterations for 4096 bit keys. Choosing 2/3rd of the maximum is deemed to be a sufficiently costly yet not excessive value.

5.3.3.  NSEC3 Salt

While the NSEC3 iterations parameter increases the cost of hashing a dictionary word, the NSEC3 salt reduces the lifetime for which that calculated hash can be used. A change of the salt value by the zone owner would cause an attacker to lose all precalculated work for that zone.

The FQDN RRlabel, which is part of the value that is hashed, already ensures that brute force work for one RRlabel can not be re-used to attack other RRlabel (e.g. in other domains) due to their uniqueness.

The salt of all NSEC3 records in a zone needs to be the same. Since changing the salt requires all the NSEC3 records to be regenerated, and thus requires generating new RRSIG's over these NSEC3 records, it is recommended to align the change of the salt with a change of the Zone Signing Key, as that process in itself already requires all RRSIG's to be regenerated. If there is no critical dependency on incremental signing and the whole zone can be signed with little effort there is no need for such alignment. However, unlike Zone Signing Key changes, NSEC3 salt changes do not need special rollover procedures. It is possible to change the salt each time the zone is updated.

5.3.4.  Opt-out

The Opt-Out mechanism was introduced to allow for a gradual introduction of signed records in zones that contain mostly delegation records. The use of the OPT-OUT flag changes the meaning of the NSEC3 span from authoritative denial of the existence of names within the span to a proof that DNSSEC is not available for the delegations within the span. [Editors Note: One could make this construct more correct by talking about the hashed names and the hashed span, but I believe that is overkill]. This allows for the addition or removal of the delegations covered by the span without recalculating or re- signing RRs in the NSEC3 RR chain.

Opt-Out is specified to be used only over delegation points and will therefore only bring relieve in zones with a large number of zones and where the number of secure delegations is small. This consideration typically holds for large top-level-domains and similar zones, in most other circumstances Opt-Out should not be deployed. Further considerations can be found in RFC5155 section 12.2 (Laurie, B., Sisson, G., Arends, R., and D. Blacka, “DNS Security (DNSSEC) Hashed Authenticated Denial of Existence,” March 2008.) [21].

6.  Security Considerations

DNSSEC adds data integrity to the DNS. This document tries to assess the operational considerations to maintain a stable and secure DNSSEC service. Not taking into account the 'data propagation' properties in the DNS will cause validation failures and may make secured zones unavailable to security-aware resolvers.

7.  IANA considerations

There are no IANA considerations with respect to this document

8.  Acknowledgments

Most of the text of this document is copied from RFC4641 (Kolkman, O. and R. Gieben, “DNSSEC Operational Practices,” September 2006.) [15]. That document was edited by Olaf Kolkman and Miek Gieben. Other people that contributed or where otherwise involved in that work were in random order: Rip Loomis, Olafur Gudmundsson, Wesley Griffin, Michael Richardson, Scott Rose, Rick van Rein, Tim McGinnis, Gilles Guette Olivier Courtay, Sam Weiler, Jelte Jansen, Niall O'Reilly, Holger Zuleger, Ed Lewis, Hilarie Orman, Marcos Sanz, Peter Koch, Mike StJohns, Emma Bretherick, Adrian Bedford, and Lindy Foster, G. Guette, and O. Courtay.

For this version of the document we would like to acknowldge a few people for significant contributions:

• Paul Hoffman for his contribution on the choice of cryptographic paramenters and addressing some of the trust anchor issues.
• Jelte Jansen who provided the text in Section 4.1.5 (Key algorithm rollover)
• Paul Wouters who provided the initial text for Section 5 (Next Record type) and Alex Bligh who improved it.
• Erik Rescorla's blogpost on "the Security of ZSK rollovers" inspired text in Section 3.1.1 (Motivations for the KSK and ZSK Separation).
• The figure in Section 4.4.2 (Signature Validation Periods) was adapted from the OpenDNSSEC user documentation.

In addition valuable contributions in the form of text, comments, or review where provided by Mark Andrews, Patrik Faltstrom, Tony Finch, Olafur Gudmundsson, Alfred Hines, Bill Manning, Scott Rose.

[EDITOR NOTE: please let me know if there is an oversight here]

9.  References

9.1. Normative References

9.2. Informative References

Appendix A.  Terminology

In this document, there is some jargon used that is defined in other documents. In most cases, we have not copied the text from the documents defining the terms but have given a more elaborate explanation of the meaning. Note that these explanations should not be seen as authoritative.

Anchored key:

A DNSKEY configured in resolvers around the globe. This key is hard to update, hence the term anchored.

Bogus:

Also see Section 5 of RFC4033 (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “DNS Security Introduction and Requirements,” March 2005.) [3]. An RRSet in DNSSEC is marked "Bogus" when a signature of an RRSet does not validate against a DNSKEY.

Key Signing Key or KSK:

A Key Signing Key (KSK) is a key that is used exclusively for signing the apex key set. The fact that a key is a KSK is only relevant to the signing tool.

Key size:

The term 'key size' can be substituted by 'modulus size' throughout the document. It is mathematically more correct to use modulus size, but as this is a document directed at operators we feel more at ease with the term key size.

Private and public keys:

DNSSEC secures the DNS through the use of public key cryptography. Public key cryptography is based on the existence of two (mathematically related) keys, a public key and a private key. The public keys are published in the DNS by use of the DNSKEY Resource Record (DNSKEY RR). Private keys should remain private.

Key rollover:

A key rollover (also called key supercession in some environments) is the act of replacing one key pair with another at the end of a key effectivity period.

Refresh Period:

The time at the end of the Signature Validity Period during which signatures are refreshed.

Resinging frequency:

Frequency with wich a signing pass on the zone is performed. Alternatively expressed as "Resigning Period". It defines when the zone is exposed to the signer. During a signing pass not all signatures in the zone may be refreshed, that depend refresh frequency/intervall.

Secure Entry Point (SEP) key:

A KSK that has a parental DS record pointing to it or is configured as a trust anchor. Although not required by the protocol, we recommend that the SEP flag [5] (Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Protocol Modifications for the DNS Security Extensions,” March 2005.) is set on these keys.

Self-signature:

This only applies to signatures over DNSKEYs; a signature made with DNSKEY x, over DNSKEY x is called a self-signature. Note: without further information, self-signatures convey no trust. They are useful to check the authenticity of the DNSKEY, i.e., they can be used as a hash.

Signing Jitter:

Jitter applied to the signature validty intervall.

Signer:

The system that has access to the private key material and signs the Resource Record sets in a zone. A signer may be configured to sign only parts of the zone, e.g., only those RRSets for which existing signatures are about to expire.

Single Type Signing Scheme:

A signing scheme whereby the distinction between Zone Signing Keys and Key Singing Keys is not made.

Zone Signing Key (ZSK):

A key that is used for signing all data in a zone (except, perhaps, the DNSKEY RRSet). The fact that a key is a ZSK is only relevant to the signing tool.

Singing the zone file:

The term used for the event where an administrator joyfully signs its zone file while producing melodic sound patterns.

Zone administrator:

The 'role' that is responsible for signing a zone and publishing it on the primary authoritative server.

Appendix B.  Zone Signing Key Rollover How-To

Using the pre-published signature scheme and the most conservative method to assure oneself that data does not live in caches, here follows the "how-to".

Step 0:

The preparation: Create two keys and publish both in your key set. Mark one of the keys "active" and the other "published". Use the "active" key for signing your zone data. Store the private part of the "published" key, preferably off-line. The protocol does not provide for attributes to mark a key as active or published. This is something you have to do on your own, through the use of a notebook or key management tool.

Step 1:

Determine expiration: At the beginning of the rollover make a note of the highest expiration time of signatures in your zone file created with the current key marked as active. Wait until the expiration time marked in Step 1 has passed.

Step 2:

Then start using the key that was marked "published" to sign your data (i.e., mark it "active"). Stop using the key that was marked "active"; mark it "rolled".

Step 3:

It is safe to engage in a new rollover (Step 1) after at least one signature validity period.

Appendix C.  Typographic Conventions

The following typographic conventions are used in this document:

Key notation:

A key is denoted by DNSKEYx, where x is a number or an identifier, x could be thought of as the key id.

RRSet notations:

RRs are only denoted by the type. All other information -- owner, class, rdata, and TTL -- is left out. Thus: "example.com 3600 IN A 192.0.2.1" is reduced to "A". RRSets are a list of RRs. A example of this would be "A1, A2", specifying the RRSet containing two "A" records. This could again be abbreviated to just "A".

Signature notation:

Signatures are denoted as RRSIGx(RRSet), which means that RRSet is signed with DNSKEYx.

Zone representation:

Using the above notation we have simplified the representation of a signed zone by leaving out all unnecessary details such as the names and by representing all data by "SOAx"

SOA representation:

SOAs are represented as SOAx, where x is the serial number.

Using this notation the following signed zone:

example.net.      86400  IN SOA  ns.example.net. bert.example.net. (
                         2006022100   ; serial
                         86400        ; refresh (  24 hours)
                         7200         ; retry   (   2 hours)
                         3600000      ; expire  (1000 hours)
                         28800 )      ; minimum (   8 hours)
                  86400  RRSIG   SOA 5 2 86400 20130522213204 (
                               20130422213204 14 example.net.
                               cmL62SI6iAX46xGNQAdQ... )
                  86400  NS      a.example.net.
                  86400  NS      b.example.net.
                  86400  RRSIG   NS 5 2 86400 20130507213204 (
                               20130407213204 14 example.net.
                               SO5epiJei19AjXoUpFnQ ... )
                  86400  DNSKEY  256 3 5 (
                               EtRB9MP5/AvOuVO0I8XDxy0... ) ; id = 14
                  86400  DNSKEY  257 3 5 (
                               gsPW/Yy19GzYIY+Gnr8HABU... ) ; id = 15
                  86400  RRSIG   DNSKEY 5 2 86400 20130522213204 (
                               20130422213204 14 example.net.
                               J4zCe8QX4tXVGjV4e1r9... )
                  86400  RRSIG   DNSKEY 5 2 86400 20130522213204 (
                               20130422213204 15 example.net.
                               keVDCOpsSeDReyV6O... )
                  86400  RRSIG   NSEC 5 2 86400 20130507213204 (
                               20130407213204 14 example.net.
                               obj3HEp1GjnmhRjX... )
a.example.net.    86400  IN TXT  "A label"
                  86400  RRSIG   TXT 5 3 86400 20130507213204 (
                               20130407213204 14 example.net.
                               IkDMlRdYLmXH7QJnuF3v... )
                  86400  NSEC    b.example.com. TXT RRSIG NSEC
                  86400  RRSIG   NSEC 5 3 86400 20130507213204 (
                               20130407213204 14 example.net.
                               bZMjoZ3bHjnEz0nIsPMM... )
                  ...

is reduced to the following representation:

    SOA2006022100
    RRSIG14(SOA2006022100)
    DNSKEY14
    DNSKEY15

    RRSIG14(KEY)
    RRSIG15(KEY)

The rest of the zone data has the same signature as the SOA record, i.e., an RRSIG created with DNSKEY 14.

Appendix D.  Document Editing History

[To be removed prior to publication as an RFC]

D.1.  draft-ietf-dnsop-rfc4641-00

Version 0 was differs from RFC4641 in the following ways.

• Status of this memo appropriate for I-D
• TOC formatting differs.
• Whitespaces, linebreaks, and pagebreaks may be slightly different because of xml2rfc generation.
• References slightly reordered.
• Applied the errata from http://www.rfc-editor.org/errata_search.php?rfc=4641
• Inserted trivial "IANA considertations" section.

In other words it should not contain substantive changes in content as intended by the workinggroup for the original RFC4641.

D.2.  version 0->1

Cryptography details rewritten. (See http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/cryptography_flawed)

• Reference to NIST 800-90 added
• RSA/SHA256 is being recommended in addition to RSA/SHA1.
• Complete rewrite of Section 3.1.4.2 (Key Sizes) removing the table and suggesting a keysize of 1024 for keys in use for less than 8 years, issued up to at least 2015.
• Replaced the reference to Schneiers' applied cryptograpy with a reference to RFC4949.
• Removed the KSK for high level zones consideration

Applied some differentiation with respect of the use of a KSK for parent or trust-anchor relation http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/differentiation_trustanchor_parent

http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/rollover_assumptions

Added Section 4.1.5 (Key algorithm rollover) as suggested by Jelte Jansen in http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/Key_algorithm_roll

Added Section 4.3.5.1 (Cooperationg DNS operators) Issue identified by Antoin Verschuur http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/non-cooperative-registrars

In Appendix A (Terminology): ZSK does not nescessarily sign the DNSKEY RRset.

D.3.  version 1->2

• Significant rewrite of Section 3 (Keys Generation and Storage) whereby the argument is made that the timescakes for rollovers are made purely on operational arguments hopefully resolving http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/discussion_of_timescales
• Added Section 5 (Next Record type) based on http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/NSEC-NSEC3
• Added a reference to draft-morris-dnsop-dnssec-key-timing (Morris, S., Ihren, J., and J. Dickinson, “DNSSEC Key Timing Considerations,” July 2010.) [24] for the quantitative analysis on keyrolls
• Updated Section 4.3.5 (Changing DNS Operators) to reflect that the problem occurs when changing DNS operators, and not DNS registrars, also added the table indicating the redelegation procedure. Added text about the fact that implementations will dismiss keys that fail to validate at some point.
• Updated a number of references.

D.4.  version 2->3

• Added bulleted list to serve as an introduction on the decision tree in Section 3 (Keys Generation and Storage).
• In section Section 3.1.1 (Motivations for the KSK and ZSK Separation):
  • tried to motivate that keylength is not a strong motivation for KSK ZSK split (based on http://www.educatedguesswork.org/2009/10/on_the_security_of_zsk_rollove.html)
  • Introduced Common Signing Key terminology and made the arguments for the choice of a Common Signing Key more explicit.
  • Moved the SEP flag considerations to its own paragraph
• In a few places in the document, but section Section 4 (Signature Generation, Key Rollover, and Related Policies) in particular the comments from Patrik Faltstrom (On Mar 24, 2010) on the clarity on the roles of the registrant, dns operator, registrar and registry was addressed.
• Added some terms based on http://www.nlnetlabs.nl/svn/rfc4641bis/trunk/open-issues/timing_terminology
• Added paragrap 2 and clarified the second but last paragraph of Section 3.1.2.2 (Rolling a KSK that is a trust-anchor).
• Clarified the table and some text in Section 4.1.5 (Key algorithm rollover). Also added some text on what happens when the algorithm rollover also involves a roll from NSEC to NSEC3.
• Added a paragraph about rolling KSKs that are also configured as trust-anchors in Section 4.1.2 (Key Signing Key Rollovers)
• Added Section 4.1.4 (Rollover for a Single Type Signing Key rollover).
• Added Section 4.4.2 (Signature Validation Periods) to address issue "Signature_validity"

$Id: draft-ietf-dnsop-rfc4641bis-03.xml 60 2010-07-12 19:33:51Z olaf $

Author's Address