← Back to list
Security & Privacy
#DNSSEC#캐시포이즈닝#신뢰사슬#RRSIG#DS레코드
Last updated · 2026-11-14

DNSSEC (DNS Security Extensions)

1. Overview

DNSSEC (DNS Security Extensions) is an IETF standard extension that attaches public-key digital signatures to DNS response data so that a resolver can confirm the response came from a legitimately authoritative name server (origin authentication), was not altered in transit (integrity), and even proves the absence of a record that does not exist (authenticated denial of existence) (RFC 4033/4034/4035). The key point is that it targets data origin authentication and integrity, not confidentiality (encryption).

DNSSEC arose from a structural weakness in DNS: the protocol was inherently designed so that a received response is trusted without question (trust-by-default). When DNS was designed in 1983, in an environment of only a few hundred hosts worldwide, queries and responses were mostly exchanged as plaintext UDP datagrams, and the only devices for distinguishing a genuine response were a 16-bit transaction ID and the source port. This optimistic design made name resolution light and fast, but it was defenseless against cache poisoning, in which an attacker pushes a forged response into the resolver ahead of the legitimate name server and contaminates the cache. The attack disclosed by Dan Kaminsky in 2008 demonstrated how easily the entropy of the transaction ID and port can be guessed and exhausted, driving home that stopgaps such as added entropy (0x20 encoding, port randomization) alone cannot solve the root problem. In short, a cryptographic basis for verifying "who signed" a DNS response was needed, and DNSSEC is the answer.

The necessity of DNSSEC becomes clear on three axes. First is verifying the authenticity of the origin, replacing "just believe the response received" with "trust only responses signed by the private key of an authoritative zone." Second is automatically blocking tampering and forgery, filtering out bogus IP addresses inserted by on-path attacks or a poisoned cache through signature mismatch. Third is the connectivity of the chain of trust: rather than each zone merely signing on its own, the entire top-down delegation structure from the root (.) through the TLD to second-level domains is bound into a single verifiable chain, so that a parent vouches for the public key of its child. It is this third property that elevates DNSSEC from mere "per-zone signing" to a global trust infrastructure.

The essential characteristics that distinguish DNSSEC from other security technologies can be summarized as signing the data itself, a top-down delegated chain of trust, and an orientation toward authentication and integrity rather than confidentiality. Unlike TLS, which encrypts the transport segment, DNSSEC signs the data object (RRset) directly, so even a cached response is verified; trust flows from the root to the leaf along delegation, which differs from a horizontal web of trust; and because it does not hide content, it forms a mutually complementary relationship with transport-encryption technologies such as DoT/DoH discussed later. These characteristics are the key to simultaneously explaining DNSSEC's strengths and its limitations, such as "zone content disclosure (zone walking)" and "key-management complexity."

2. DNSSEC Components and Chain-of-Trust Structure

The key to understanding DNSSEC is the structural correspondence that "the DNS delegation hierarchy was laid directly onto a signing trust hierarchy." The fact that a domain's response is trustworthy is proven by the response being signed with that zone's key (RRSIG), that key being vouched for by the parent zone (DS), and the parent zone's key in turn being vouched for by the one above it, continuing unbroken up to the root. The overall system is divided into "the side where each zone signs and publishes its own data" and "the side where the resolver starts from the root's trust anchor and descends with validation." The overall structural diagram below shows both the vouching relationships among records (top) and how they form a chain along delegation (bottom).

flowchart TB
  subgraph REC["In-zone signing objects"]
    DNSKEY["DNSKEY: zone public keys (KSK/ZSK)"]
    RRSIG["RRSIG: digital signature over an RRset"]
    NSEC["NSEC/NSEC3: denial-of-existence records"]
    DNSKEY --> RRSIG
    DNSKEY --> NSEC
  end
  subgraph CHAIN["Top-down delegated chain of trust"]
    ROOT["Root (.) KSK = Trust Anchor"]
    TLD["TLD (.kr/.com) zone: DNSKEY + DS"]
    SLD["Second-level domain zone: DNSKEY + RRSIG"]
    ROOT -- "vouches via DS" --> TLD
    TLD -- "vouches via DS" --> SLD
    SLD --> RRSIG
  end

A. DNSKEY and Key Separation (KSK/ZSK)

The starting point of DNSSEC signing is that each zone publishes its own public key as a DNSKEY record. The resolver verifies signatures (RRSIG) with this public key. In operational reality, however, a dilemma arises: changing the key often reduces the risk of signature forgery, but frequent changes require re-updating the vouching relationship (DS) with the parent zone each time. DNSSEC resolves this tension by separating the key into two with different roles. One is the ZSK (Zone Signing Key), the "workhorse" key that signs the actual data records inside the zone (A, MX, TXT, etc.). The other is the KSK (Key Signing Key), the "seal" key that signs only the DNSKEY RRset (that is, the ZSK and the KSK itself).

The practical benefit of this separation is decisive. Because the ZSK bears a heavy signing load and has a wide exposure surface, it is rotated frequently on a short cycle (weeks to months), yet rotating it requires no touch to the parent zone at all. This is because what the parent zone vouches for (DS) is not the ZSK but the KSK. Conversely, the KSK is kept long and secure (often a larger key size, stored offline) and is rolled over only rarely. In the end, by decoupling the "frequently changed key (ZSK)" from the "stable key registered with the parent (KSK)," one obtains both the flexibility of routine key rotation and the stability of the delegation relationship. In small zones, the two are sometimes combined into a single CSK (Combined Signing Key), a choice that trades operational simplicity for rollover flexibility.

Concretely, for the root zone the KSK is KSK-2017 (generated in 2017, replacing KSK-2010 in the first rollover of October 2018), operated with RSA/SHA-256 (algorithm number 8) at 2048 bits. The Trust Anchor that the resolver takes as its starting point of trust is precisely the public key of this root KSK, embedded in operating systems and resolver software or distributed via the RFC 5011 automatic update procedure. In other words, the trust of all DNSSEC validation worldwide ultimately converges on this single root KSK. A brief look at the actual records makes it clear at a glance what information each object holds.

; DNSKEY: 257=KSK (SEP flag), 256=ZSK
example.kr.  IN DNSKEY 257 3 13 (ECDSA P-256 public key...)   ; KSK
example.kr.  IN DNSKEY 256 3 13 (ECDSA P-256 public key...)   ; ZSK
; RRSIG: signs type A, algorithm 13, validity (inception~expiration)
www.example.kr. IN RRSIG A 13 3 3600 20261130000000 20261116000000 ...
; DS: hash of the child KSK, published in the parent (.kr) zone
example.kr.  IN DS 12345 13 2 (SHA-256 digest...)

In this example, the DNSKEY flags 257 (KSK, SEP bit) and 256 (ZSK), the short validity of the RRSIG, and the fact that the DS is placed in the parent zone can be confirmed to correspond exactly to the structure described above.

B. RRSIG and RRset Signing

The unit actually signed in DNSSEC is not an individual record but an RRset, a bundle of records sharing the same name, type, and class. For example, if www.example.com has two A records, the two are bundled into a single RRset and signed together. The reason for bundling is to seal "the entire set" at once so that an attacker cannot perform manipulation that removes or inserts only some records of the RRset. This signing result is the RRSIG (Resource Record Signature) record, which also carries the type signed, the algorithm and key tag used, the signature validity (inception/expiration), the signer name, and so on.

The fact that an RRSIG explicitly states a validity period is an important design decision of DNSSEC. Unlike an ordinary TLS certificate, valid for months to years after issuance, an RRSIG is often set to expire within days to weeks and is re-signed periodically. This is to limit the useful lifetime of a forged signature even if the signing key leaks, and to maintain freshness with only "short expiration + re-signing" rather than a complex revocation mechanism. However, this choice means the name server must continuously perform re-signing work, and if an operator misses re-signing and an RRSIG expires, it leads to a major outage in which, from the perspective of a validating resolver, the entire domain becomes unresolvable (SERVFAIL). In reality, a substantial share of large-scale service outages originate from "DNSSEC signature expiration," which is the core of the operational risk discussed later in the considerations.

C. DS Records and the Delegation Link

No matter how completely signatures and keys are furnished inside a zone, if the parent zone does not vouch for the child zone's key, the trust remains an island. What connects this broken link is the DS (Delegation Signer) record. The DS carries the hash value of the child zone's KSK and is published in the parent zone, where the parent signs this DS with its own ZSK. That is, if the .com zone contains the DS of example.com and it is vouched for by .com's signature, the resolver can cryptographically confirm the fact that ".com recognizes this KSK of example.com."

Thanks to this DS mechanism, DNSSEC turns the chain of delegation running root → TLD → second-level domain directly into a chain of trust. Looking at the validation process in reverse, the resolver verifies the root's DNSKEY with the embedded root KSK (trust anchor), verifies .kr's KSK with the DS of .kr published by the root, and again verifies that zone's KSK with the DS of example.kr published by .kr, thereby passing trust downward one step at a time. An important operational implication is that it is not enough for a domain owner to sign their own zone; the chain is completed only when the owner registers the DS with the parent registry (usually via a registrar). Until the DS is registered, that zone, even if signed, remains an "island of security" cut off from the validation chain.

D. NSEC/NSEC3 and Authenticated Denial of Existence

One distinctive challenge of DNSSEC is that it must prove, un-forgeably, even the fact that something "does not exist." In ordinary DNS, one could simply return "none (NXDOMAIN)" for a missing name, but that is meaningless if an attacker can forge even that "none" response. Yet signing "none" on the fly would require the private key to reside online, which is dangerous. DNSSEC solves this with the NSEC (Next Secure) record. NSEC sorts the names in the zone and pre-signs adjacency relationships in the form "the name after A is C." If a queried name B falls between A and C but does not appear in NSEC, the resolver is certain that "B does not exist" as a pre-signed fact.

However, NSEC has a fatal side effect. Because it reveals adjacent names in plaintext, an attacker can follow the NSEC chain and collect every name in the zone (zone walking). Internal host names are thereby exposed as-is. To prevent this, NSEC3 was introduced, which signs adjacency relationships sorted by hash values rather than plaintext names, making it hard to immediately recover the original name. NSEC3 slows dictionary attacks with a salt and iterated hashing, but as seen in the deep-dive later, an excessive iteration count creates a DoS risk on the validating side, so recent standards have pivoted toward recommending an iteration count of effectively 0. In the latest operations, techniques such as NSEC3 white lies combined with online signing, or NSEC "black lies," are used to trade off denial of existence against performance and privacy.

3. The DNSSEC Validation Process

How the components interlock to complete a single validation becomes clear by following the resolver's sequence of operations. The validating resolver verifies the RRSIG attached to a response with the DNSKEY, while recursively confirming whether that DNSKEY is vouched for by the parent's DS, until it reaches the root's trust anchor. The detailed flow diagram below shows the typical path by which a www.example.kr query is validated through root → TLD → authoritative zone.

sequenceDiagram
  participant C as Client (Stub)
  participant R as Validating Resolver
  participant Root as Root name server (.)
  participant TLD as TLD server (.kr)
  participant Auth as Authoritative server (example.kr)
  C->>R: www.example.kr A query (DO=1)
  R->>Root: root DNSKEY / delegation query
  Root-->>R: .kr NS + DS(.kr) + RRSIG
  Note over R: verify DS(.kr) with embedded root KSK
  R->>TLD: example.kr delegation query
  TLD-->>R: example.kr DS + RRSIG
  Note over R: verify DS(example.kr) with .kr DNSKEY
  R->>Auth: www.example.kr A query
  Auth-->>R: A record + RRSIG
  Note over R: verify RRSIG with example.kr DNSKEY -> chain complete
  R-->>C: validated A response (AD=1) or SERVFAIL

The starting signal of validation is setting the DO (DNSSEC OK) bit in the query. Only with this bit does the server return signing records such as RRSIG and DNSKEY as well. The resolver verifies the RRSIG of the received A record with the authoritative zone's ZSK (DNSKEY), verifies that DNSKEY RRset with the KSK's RRSIG, and confirms whether that KSK matches the hash of the DS published by the parent. Repeating this at each delegation step, once it connects unbroken up to the root KSK, it sets the AD (Authenticated Data) bit on the response to mark it "validated." Conversely, if there is no signature (unsigned zone), the signature is broken, the DS mismatches, or the RRSIG has expired, the resolver judges the response untrustworthy and returns SERVFAIL. The important nuance here is that a DNSSEC validation failure is not "handing over a wrong IP" but "handing over no response at all." For that reason, a signing operation mistake converts directly into an availability outage of the entire domain.

One distinction to note is protection of the last segment (stub resolver ↔ validating resolver). Validation is usually performed by a recursive resolver such as an ISP or public resolver, and the end-user device's stub trusts only the AD bit. But if this last segment is in plaintext, it is still exposed to man-in-the-middle forgery, so in practice the device-resolver link is encrypted with DoT/DoH or a configuration that validates directly on the device is run in parallel. In other words, we again confirm that DNSSEC handles "the authenticity of the data" and transport encryption handles "the confidentiality and integrity of the segment," in a mutually complementary relationship.

Validation also hides a tension between performance and freshness. The resolver caches a validated response for the TTL to reduce re-queries, but if the RRSIG's signature expiration arrives before the record's TTL, the cached signature may be consumed while expired. Therefore an operator must set the RRSIG validity comfortably longer than the TTL so that the inversion "signature expiration < cache lifetime" does not occur, which means re-signing cycle, TTL, and DS change timing must be managed as a single consistency problem. So that the validating resolver does not reconstruct the full chain from the trust anchor every time, the optimization of reusing a parent zone's DNSKEY/DS through aggressive caching (aggressive NSEC caching, RFC 8198) to lower both query volume and validation cost is also widely used in practice.

4. Comparison and Application Cases

To place DNSSEC accurately, it is most effective to contrast "what it protects and what it does not" against neighboring technologies. It is often confused in particular with transport-encryption technologies (DoT/DoH), but the two are not substitutes; they are complements that protect different targets. The table below organizes them by protection goal, target, and limits, while the prose below the table explains "why such differences arise."

Category DNSSEC DoT/DoH (DNS over TLS/HTTPS) DANE
Primary goal origin authentication / integrity transport-segment confidentiality / integrity DNSSEC-based certificate binding
Protection target DNS data (RRset) resolver-device segment authenticity of TLS certificate
Confidentiality none (plaintext exposure) yes (encrypted) not applicable
Basis of trust root KSK trust anchor CA PKI (web) DNSSEC chain of trust
Protects cached responses yes (data signing) no (segment only) yes

The root of this difference lies in "whether what is signed is the data or the segment." DoT/DoH encrypt the link between resolver and device with TLS to prevent eavesdropping and segment tampering, but if the resolver itself has fetched a poisoned cache or a forged parent response, it merely encrypts and delivers that false data as-is. DNSSEC, by contrast, attaches a signature to the data object itself, so even a response cached through several hops is verified back to the original signer. That is why a mature configuration uses DNSSEC (authenticity) + DoT/DoH (confidentiality) together. Meanwhile, DANE (DNS-based Authentication of Named Entities) places a TLS certificate fingerprint atop DNSSEC's chain of trust—an application that seeks to supplement the web CA ecosystem's mis-issuance problem via DNS, and a representative case showing that DNSSEC can be reused as a trust anchor for other security systems.

Looking at the state of adoption in numbers reveals DNSSEC's asymmetric diffusion. The root and most TLDs have already finished signing, so the upper infrastructure is essentially complete, and Sweden (.se), the Netherlands (.nl), and the Czech Republic (.cz) have relatively high second-level-domain signing rates. By contrast, the worldwide validating-resolver ratio hovers around 30%, and the ratio of signed second-level domains stays in the single digits to low teens, so an imbalance of "ready to validate but few domains that sign" has persisted for a long time. Domestically as well, the .kr/.한국 zones are signed, but the signing and DS registration of ordinary corporate and institutional domains is still limited. This gradual diffusion is largely attributable to the operational complexity and outage risk discussed in the deep-dive and considerations that follow.

One obstacle to diffusion, the manual burden of DS registration and renewal, is being rapidly alleviated by automation standards. Traditionally a domain owner had to manually enter a new DS into the registrar portal each time a key was rolled over, and if this procedure was omitted or delayed, the chain broke and an outage ensued. To solve this, the CDS/CDNSKEY records (RFC 7344/8078) were introduced; when a child zone publishes, in signed form within the zone, a request to "change my parent DS to this value," the parent registry/registrar reads it periodically and automatically renews the DS. In fact, .cz, .ch, and others operate CDS-based automatic rollover, creating cases where key rotation is carried out safely without human intervention. This is an important advance that absorbs the "key-rollover operational risk," DNSSEC's biggest weakness, through protocol-level automation, and is cited as a factor that will accelerate future diffusion.

5. Deep Dive — Recent Trends and Security Issues

DNSSEC is not a "finished standard" but a technology continually refined according to the operational and threat environment. The biggest recent event is the KeyTrap vulnerability (CVE-2023-50387, CVSS 7.5), disclosed in February 2024. This is an attack in which an attacker crafts a malicious zone response deliberately containing many colliding keys and signatures, driving the validating resolver into a CPU spike that results in denial of service (DoS) as it searches for a valid combination. Interestingly, this stems not from an implementation bug but from the lenient design of the DNSSEC specification itself, which says "try validation with every possible key," so major resolvers such as BIND, Unbound, and Knot were affected nearly simultaneously. The co-disclosed CVE-2023-50868, related to excessive NSEC3 iteration, is a similar resource-exhaustion threat. Vendors distributed mitigation patches around February 2024, including BIND 9.18.24, Unbound 1.19.1, and Knot Resolver 5.7.1, but researchers pointed out that "patches only reduce the impact; complete resolution is difficult without changing the underlying design philosophy." This event is an important case reminding us that a structure like DNSSEC, in which the validator bears the cost of validation, can be inherently vulnerable to asymmetric DoS.

A second current is the generational shift of algorithms. Early DNSSEC mainly used RSA/SHA-1 and RSA/SHA-256, but their large signature and key sizes had the problem of bloating responses and enlarging the attack surface for UDP fragmentation and amplification. Accordingly, a transition to elliptic-curve signatures such as ECDSA P-256 (algorithm 13) and EdDSA Ed25519 (algorithm 15) is recommended, because at the same security strength the shorter signature length greatly reduces response size and fragmentation risk. A third is preparation for post-quantum (PQC): once a sufficiently powerful quantum computer appears, RSA and ECDSA signatures are threatened, so IETF-level research is underway to apply to DNSSEC a lightweight quantum-resistant signature (e.g., the FALCON family, whose signatures are relatively small) usable under DNS response-size constraints. However, since most PQC signatures are large and collide with DNS's UDP-friendliness, the size-security trade-off remains the core difficulty.

6. Considerations and Implications

Adopting DNSSEC should, from a professional engineer's perspective, be approached as a problem of balanced design between "security gain and operational risk." The four points below are the core axes of consideration.

First, managing availability risk is as important as the security gain. Because a DNSSEC validation failure results not in a wrong IP but in SERVFAIL (unresolvable), operational errors such as RRSIG expiration, key-rollover mistakes, and DS mismatch immediately become a whole-domain outage. Therefore one must design with automatic re-signing and key-rollover automation (e.g., auto-signing name servers), near-expiry monitoring, and consistency management between DS changes and resolver cache TTL as preconditions. One must face the paradox that "signing incorrectly" can be more dangerous than "not signing."

Second, the division of roles with transport encryption must be made clear. Because DNSSEC handles data authenticity while DoT/DoH and DNSCrypt handle segment confidentiality, neither can replace the other. In particular, since the device-resolver last segment retains plaintext-exposure and forgery risk with DNSSEC alone, a defense-in-depth layering of DNSSEC validation + encrypted transport + (where possible) device-side validation is desirable.

Third, algorithm and parameter choices that consider the performance/attack-surface trade-off are needed. Choosing ECDSA/EdDSA instead of RSA to reduce response size simultaneously mitigates fragmentation/amplification risk and validation cost. Keep the NSEC3 iteration count low (0 recommended) to reduce the resource-exhaustion attack surface of the KeyTrap type, and if zone walking is a concern, examine online-signing-based black/white lies. In short, DNSSEC requires not a binary "on/off" choice but fine-grained operational decisions at the parameter level.

Fourth, the perspective of governance and a transition roadmap. Because global trust converges on the single point of the root KSK, one must have a resolver-update system in step with IANA's root KSK rollover and trust-anchor distribution (RFC 5011). At the organizational level, one should make it a medium-to-long-term task to secure the parent-registry DS registration procedure, confirm whether the registrar supports DNSSEC and CDS/CDNSKEY automation, and secure crypto-agility in anticipation of a PQC transition. It is desirable to position DNSSEC, together with related technologies such as DANE, RPKI, and email authentication (DKIM/DMARC), as one pillar of a "verifiable internet trust infrastructure."

References


In one line: DNSSEC is an extension that signs DNS responses with public keys and validates them through a DS chain of trust descending from the root to confirm origin authentication, integrity, and denial of existence; it is complementary to transport encryption (DoT/DoH) and must be adopted while jointly considering operational risks and recent trends such as KeyTrap, algorithm migration, and PQC.