Guidance for NSEC3 parameter settings

1.1. Requirements notation

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

2.1. Algorithms

The algorithm field is not discussed by this document.¶

2.2. Flags

The NSEC3PARAM flags field currently contains no flags, but individual NSEC3 records contain the "Opt-Out" flag [RFC5155], which specifies whether or not that NSEC3 record provides proof of non-existence or not. In general, NSEC3 with the Opt-Out flag enabled should only be used in large, highly dynamic zones with a small percentage of signed delegations. Operationally, this allows for fewer signature creations when new delegations are inserted into a zone. This is typically only necessary for extremely large registration points providing zone updates faster than real-time signing allows or when using memory-constrained hardware. Smaller zones, or large but relatively static zones, are encouraged to use a Flags value of 0 (zero) and take advantage of DNSSEC's proof-of-non-existence support.¶

2.3. Iterations

NSEC3 records are created by first hashing the input domain and then repeating that hashing algorithm a number of times based on the iterations parameter in the NSEC3PARM and NSEC3 records. The first hash is typically sufficient to discourage zone enumeration performed by "zone walking" an NSEC or NSEC3 chain. Only determined parties with significant resources are likely to try and uncover hashed values, regardless of the number of additional iterations performed. If an adversary really wants to expend significant CPU resources to mount an offline dictionary attack on a zone's NSEC3 chain, they'll likely be able to find most of the "guessable" names despite any level of additional hashing iterations.¶

Most names published in the DNS are rarely secret or unpredictable. They are published to be memorable, used and consumed by humans. They are often recorded in many other network logs such as email logs, certificate transparency logs, web page links, intrusion detection systems, malware scanners, email archives, etc. Many times a simple dictionary of commonly used domain names prefixes (www, ftp, mail, imap, login, database, etc) can be used to quickly reveal a large number of labels within a zone. Because of this, there are increasing performance costs yet diminishing returns associated with applying additional hash iterations beyond the first.¶

Although Section 10.3 of [RFC5155] specifies upper bounds for the number of hash iterations to use, there is no published guidance for zone owners about good values to select. Because hashing provides only moderate protection, as shown recently in academic studies of NSEC3 protected zones [GPUNSEC3][ZONEENUM], this document recommends that zone owners SHOULD use an iteration value of 0 (zero), indicating that only the initial hash value should be placed into a DNS zone's NSEC3 records.¶

2.4. Salt

Operators are encouraged to forget the salt entirely by using a zero-length salt value instead (represented as a "-" in the presentation format).¶

NSEC3 records provide an an additional salt value, which can be combined with an FQDN to influence the resulting hash, but properties of this extra salt are complicated.¶

In cryptography, salts generally add a layer of protection against offline, stored dictionary attacks by combining the value to be hashed with a unique "salt" value. This prevents adversaries from building up and remembering a single dictionary of values that can translate a hash output back to the value that it derived from.¶

In the case of DNS, the situation is different because the hashed names placed in NSEC3 records are always implicitly "salted" by hashing the fully-qualified domain name from each zone. Thus, no single pre-computed table works to speed up dictionary attacks against multiple target zones. An attacker is always required to compute a complete dictionary per zone, which is expensive in both storage and CPU time.¶

To understand role of the additional NSEC3 salt field, we have to consider how a typical zone walking attack works. Typically the attack has two phases - online and offline. In the online phase, an attacker "walks the zone" by enumerating (almost) all hashes listed in NSEC3 records and storing them for the offline phase. Then, in the offline cracking phase, the attacker attempts to crack the underlying hash. In this phase, the additional salt value raises the cost of the attack only if the salt value changes during the online phase of the attack. In other words, an additional, constant salt value does not change cost of the attack.¶

Changing a zone's salt value requires the construction of a complete new NSEC3 chain. This is true both when resigning the entire zone at once, or when incrementally signing it in the background where the new salt is only activated once every name in the chain has been completed. As a result, re-salting a is very complex operation, with significant CPU time, memory, and bandwidth consumption. This makes very frequent re-salting unpractical, and renders the additional salt field functionally useless.¶

3.1. Best-practice for zone publishers

First, if the operational or security features of NSEC3 are not needed, then NSEC SHOULD be used in preference to NSEC3. NSEC3 requires greater computational power (see Appendix B) for both authoritative servers and validating clients. Specifically, there is a non trivial complexity in finding matching NSEC3 records to randomly generated prefixes within a DNS zone. NSEC mitigates this concern. If NSEC3 must be used, then an iterations count of 0 MUST be used to alleviate computational burdens. Please note that extra iteration counts other than 0 increase impact of resource CPU-exhausting DoS attacks, and also increase risk of interoperability problems.¶

Note that deploying NSEC with minimally covering NSEC records [RFC4470] also incurs a cost, and zone owners should measure the computational difference in deploying both RFC4470 or NSEC3.¶

In short, for all zones, the recommended NSEC3 parameters are as shown below:¶

; SHA-1, no extra iterations, empty salt:
;
bcp.example. IN NSEC3PARAM 1 0 0 -

For small zones, the use of opt-out based NSEC3 records is NOT RECOMMENDED.¶

For very large and sparsely signed zones, where the majority of the records are insecure delegations, opt-out MAY be used.¶

Since the NSEC3PARAM RR is not used by validating resolvers (see [RFC5155] section 4), the iterations and salt parameters can be changed without the need to wait for RRsets to expire from caches. A complete new NSEC3 chain needs to be constructed and the zone resigned.¶

3.2. Recommendation for validating resolvers

Because there has been a large growth of open (public) DNSSEC validating resolvers that are subject to compute resource constraints when handling requests from anonymous clients, this document recommends that validating resolvers change their behavior with respect to large iteration values. Specifically, validating resolver operators and validating resolver software implementers are encouraged to continue evaluating NSEC3 iteration count deployments and lower their default acceptable limits over time. Similarly, because treating a high iterations count as insecure leaves zones subject to attack, validating resolver operators and validating resolver software implementers are further encouraged to lower their default and acceptable limit for returning SERVFAIL when processing NSEC3 parameters containing large iteration count values. See Appendix A for measurements taken near the time of publication and potential starting points.¶

Validating resolvers MAY return an insecure response when processing NSEC3 records with iterations larger than 0. Validating resolvers MAY also return SERVFAIL when processing NSEC3 records with iterations larger than 0. This significantly decreases the requirements originally specified in Section 10.3 of [RFC5155]. See the Security Considerations for arguments on how to handle responses with non-zero iteration count.¶

Validating resolvers returning an insecure or SERVFAIL answer because of unsupported NSEC3 parameter values SHOULD return an Extended DNS Error (EDE) {RFC8914} EDNS0 option of value (RFC EDITOR: TBD).¶

Note that a validating resolver MUST still validate the signature over the NSEC3 record to ensure the iteration count was not altered since record publication (see [RFC5155] section 10.3).¶

3.3. Recommendation for Primary / Secondary relationships

Primary and secondary authoritative servers for a zone that are not being run by the same operational staff and/or using the same software and configuration must take into account the potential differences in NSEC3 iteration support.¶

Operators of secondary services should advertise the parameter limits that their servers support. Correspondingly, operators of primary servers need to ensure that their secondaries support the NSEC3 parameters they expect to use in their zones. To ensure reliability, after primaries change their iteration counts, they should query their secondaries with known non-existent labels to verify the secondary servers are responding as expected.¶

E.1. OpenDNSSEC

The OpenDNSSEC configuration checking utility will alert the user about nsec3 iteration values larger than 100.¶

E.2. PowerDNS

PowerDNS 4.5.2 changed the default value of nsec3-max-iterations to 150.¶

E.3. Knot DNS and Knot Resolver

Knot DNS 3.0.6 warns when signing with more than 20 NSEC3 iterations. Knot Resolver 5.3.1 treats NSEC3 iterations above 150 as insecure.¶

E.4. Google Public DNS Resolver

Google Public DNS treats NSEC3 iterations above 100 as insecure since September 2021.¶

E.5. Google Cloud DNS

Google Cloud DNS uses 1 iteration and 64-bits of fixed random salt for all zones using NSEC3. These parameters cannot be adjusted by users.¶