Cyber Security

DNS CAA Records - Narrow Certificate Issuance Without Mistaking It for a Lock

DNS CAA Records - Narrow Certificate Issuance Without Mistaking It for a Lock

DNS CAA Records - Narrow Certificate Issuance Without Mistaking It for a Lock

Controlling a domain does not automatically mean choosing the only certificate authority that may issue a public TLS certificate for it. A certificate authority still has to validate the request, but without an applicable Certification Authority Authorization record, CAA does not narrow the set of eligible public issuers.

That is the small problem CAA addresses. It lets a domain holder publish which certificate authorities are authorized to issue. The control can reduce the risk of unintended certificate mis-issuance, but the wording matters: it narrows an issuance decision; it does not lock an existing certificate, replace domain validation, or instruct a browser to reject a certificate.

This distinction makes CAA useful for a personal website or small server, provided the policy is based on the site's real certificate dependencies rather than copied from a convenient example.

CAA Acts Before a Certificate Is Issued

RFC 8659 defines CAA as a DNS resource record through which a domain holder identifies one or more CAs authorized to issue certificates for that name. The CA checks the applicable records before issuance. The current CA/Browser Forum Baseline Requirements require public CAs in their scope to retrieve and process CAA for the relevant DNS names in a certificate request.

This is not a second certificate-validation system in the browser. RFC 8659 explicitly says that relying parties must not use current CAA records to validate an already issued certificate. A policy may have changed after issuance, while the certificate can still be within its validity period. Deleting an authorization therefore does not revoke certificates that the CA issued earlier.

A useful mental model is a guest list at the certificate counter. CAA can name which issuers may approach the counter for a domain. The CA must still perform its other checks, and the list says nothing about a certificate that has already left the counter.

Reading a CAA Record

The presentation format has three fields after the record type:

example.com.  CAA 0 issue "letsencrypt.org"

Here, 0 is the flags value, issue is the property tag, and letsencrypt.org is an issuer identifier. The identifier is not something to infer from the CA's brand or website hostname. Check the CA's current documentation or Certificate Policy/Certification Practice Statement for the value it recognizes. For example, Let's Encrypt documents letsencrypt.org as its identifier.

The standard defines three tags that matter for a basic policy:

  • issue authorizes a CA to issue ordinary certificates. In the absence of an issuewild property, it also controls wildcard issuance.
  • issuewild controls wildcard certificates such as *.example.com. When at least one issuewild property is present, those properties take precedence over issue for wildcard names.
  • iodef provides a URL to which a CA may report a policy violation. The Baseline Requirements require CAs to process the tag, but do not require them to act on its contents. It is not a dependable alert channel by itself.

For a site that deliberately uses one CA for both ordinary and wildcard certificates, a minimal policy could be:

example.com.  CAA 0 issue "letsencrypt.org"
example.com.  CAA 0 issuewild "letsencrypt.org"

The second line is not required when wildcard authorization should be identical to the first. Writing it explicitly can still make the intended wildcard policy easier for a human to inspect. It also means future edits must keep both lines correct.

Authorizations Are Additive

A CAA RRset can contain more than one authorization. They are alternatives, not a priority list:

example.com.  CAA 0 issue "ca-one.example"
example.com.  CAA 0 issue "ca-two.example"

Either named issuer may be authorized. Adding another record broadens the policy; it does not create a fallback that is considered only after the first CA fails.

An empty issuer value requests that no CA issue:

no-certs.example.com.  CAA 0 issue ";"

There is a subtle edge here. An empty authorization does not override a non-empty authorization in the same relevant RRset. Because authorizations are additive, combining issue ";" with a valid issuer still permits that issuer. CAA is not an ordered firewall rule set with a final deny.

The Closest CAA RRset Wins

A CA does not only ask for a record at the registered domain. For a request concerning shop.eu.example.com, the RFC algorithm starts at that name and moves upward until it finds a CAA RRset:

shop.eu.example.com
eu.example.com
example.com

The search stops at the first CAA RRset found. A policy at example.com can therefore cover descendants that have no closer CAA records, while a record at eu.example.com can replace that inherited policy for names below it. The sets are not merged across levels.

This makes a parent-domain policy convenient, but it also creates an inventory problem. A forgotten subdomain may use a hosting platform, CDN, or managed certificate service with a different issuer. Publishing a restrictive parent policy before identifying that dependency can break its next issuance or renewal.

CNAMEs and Managed TLS Need Extra Attention

CAA lookup follows DNS aliases, so policy associated with a CNAME target can also affect issuance. The exact service arrangement matters more than the visible hostname alone. Cloudflare's documentation, for example, explains that its managed products may add CAA records or require authorization for partner CAs, and advises checking a CNAME target as well as the original name.

That is a provider-specific example, not a universal list to copy. A safe inventory asks who obtains each certificate, which names it covers, whether wildcard names are involved, whether a CNAME target contributes policy, and which issuer identifiers the provider currently requires.

Inspect Before Changing

The BIND 9 dig manual documents DNS queries in the form dig name type. A terse CAA lookup is:

dig example.com CAA +short
dig app.example.com CAA +short

No output can be a valid result: the queried name may have no CAA RRset. It is not the same as SERVFAIL, a timeout, or another lookup error. Those failures leave a CA unable to establish the same answer and can stop issuance.

A direct query only shows the answer visible through the selected resolver for that name. It does not, by itself, explain every parent lookup or every alias. For a delegated or aliased hostname, inspect the relevant parent names and CNAME target too:

dig app.example.com CNAME +short
dig target.hosting.example CAA +short

Before publishing a restrictive policy, compare the inventory with recent certificates and the automation that renews them. Then check current documentation for every CA or managed provider involved. Static CA lists age badly because providers can change their issuance chain or partner set.

A Cautious Rollout

  1. List every certificate path. Include the apex domain, www, APIs, mail-related hostnames, delegated subdomains, CDN hostnames, and wildcard certificates. The goal is not to enumerate DNS generally, but to find services that obtain public certificates.
  2. Identify the actual requesters and issuers. A certificate installed on a home server may be requested directly by an ACME client, while a proxied hostname may be certified by the proxy provider.
  3. Confirm recognized CAA values in primary documentation. Do not derive them from organization names or an old generator result.
  4. Start with the policy you can justify. Add every issuer that is genuinely required, but do not add CAs merely because a template includes them.
  5. Wait for DNS propagation and inspect from more than one resolver. Check the apex, exceptional subdomains, and CNAME targets.
  6. Exercise renewal safely. Use a CA's staging environment when available, or perform a controlled renewal check that does not consume production rate limits unnecessarily.
  7. Keep the policy with operational documentation. When a CDN, host, or ACME account changes, CAA belongs on the migration checklist rather than in forgotten DNS archaeology.

The sequence is intentionally slower than adding two records and waiting for a stronger padlock icon. CAA does not change that icon. Its value appears when an issuance request is evaluated, often months after the DNS edit during an automated renewal.

Failure Modes Are Part of the Control

RFC 8659 describes deployment failures such as blocked CAA queries, malformed responses, and DNSSEC problems. Let's Encrypt likewise documents SERVFAIL and timeout cases that can prevent it from determining whether issuance is allowed. A restrictive record can be perfectly spelled and still become an availability problem if authoritative DNS does not answer reliably.

DNS integrity also limits what CAA can promise. RFC 8659 strongly recommends DNSSEC because suppression or spoofing of CAA data could alter what an issuer observes. Current Baseline Requirements require DNSSEC validation for CAA lookups from a CA's primary network perspective. That does not mean a domain must deploy DNSSEC before CAA has any effect, but signed and correctly operated DNS gives the issuer a way to authenticate the answer or its absence.

Reliability matters in both directions. A forged permissive answer can undermine the restriction; a forged or broken restrictive answer can deny issuance. CAA makes DNS part of the certificate-issuance control plane, so nameserver monitoring and renewal alerts remain necessary.

What CAA Does Not Protect

  • It does not revoke existing certificates. Removing a CA from DNS changes current authorization; it does not invalidate certificates already issued.
  • It does not make browsers consult DNS before accepting a certificate. Browser certificate validation and CA issuance authorization are different stages.
  • It does not prevent mis-issuance by an authorized CA. RFC 8659 states this limitation directly. The authorized issuer still needs sound domain-validation controls.
  • It does not repair compromised DNS or an exposed ACME account key. Those need their own controls and incident response.
  • It does not guarantee a report. An iodef address expresses where a report may go; it is not a substitute for Certificate Transparency monitoring or renewal monitoring.

These limits do not make CAA pointless. They locate it correctly: one defense at one decision point. It reduces unnecessary issuer choice while leaving certificate monitoring, DNS security, account security, key protection, and revocation as separate jobs.

Conclusion

CAA is a modest control with a precise question: which certificate authorities may issue for this DNS name now? A useful policy answers that question from an inventory of real services, not from a generic list of familiar CA names.

The safest rollout is therefore less about clever DNS syntax and more about boundaries. Understand the closest-record rule, treat wildcard authorization separately when necessary, follow CNAMEs, include managed TLS providers, and test renewal after the change. Then keep the claim proportional: CAA narrows issuance risk; it does not turn DNS into a lock around every certificate.

References

  1. RFC 8659: DNS Certification Authority Authorization (CAA) Resource Record - RFC Editor
  2. Baseline Requirements for Publicly-Trusted TLS Server Certificates - CA/Browser Forum
  3. Certification Authority Authorization (CAA) - Let's Encrypt
  4. BIND 9 Manual Pages: dig DNS Lookup Utility - ISC
  5. Add CAA Records - Cloudflare Docs