Salesforce

shuque@gmail.com

OpenSSL Corporation

ietf-dane@dukhovni.org

General Internet Engineering Task Force Internet-Draft DANE DNSSEC Authentication Client Certificate X.509 Certificate Raw Public Key The DANE TLSA protocol describes how to publish Transport Layer Security (TLS) server certificates or public keys in the DNS. This document updates RFC 6698 and RFC 7671. It describes how to use the TLSA record to publish client certificates or public keys, and also the rules and considerations for using them with TLS. In addition, it defines a new TLS extension, DANE Client Identity, to convey the client's domain name identity to the server.

Introduction and Motivation The TLS and DTLS protocols optionally support the authentication of clients using X.509 certificates or raw public keys. TLS applications that perform DANE authentication of servers using TLSA records may also desire to authenticate clients using the same mechanism, especially if the client identity is in the form of or can be represented by a DNS domain name. Some design patterns from the Internet of Things (IoT) plan to make use of this form of authentication, where large networks of physical objects identified by DNS names may authenticate themselves using TLS to centralized device management and control platforms. Other potential applications include authenticating the client side of SMTP transport security. In this document, the term TLS is used generically to describe both the TLS and DTLS (Datagram Transport Layer Security) protocols. The protocol changes described can also be used with QUIC, since QUIC re-uses the TLS handshake. This specification requires TLS 1.3 or DTLS 1.3 (or later versions). It relies on extensions carried within the CertificateRequest and Certificate handshake messages, and such per-certificate extensions are not available in earlier versions of the protocols, whose Certificate message carries no extension fields.

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

Associating Client Identities in DNS TLSA Records Different applications may have quite different conventions for naming clients via domain names. This document thus does not prescribe a single format, but mentions a few that may have wide applicability. The name that the client conveys to the server (in the TLS DANE Client Identity extension, see ) is the complete owner name of the client's TLSA record, that is, the exact DNS name the server queries. Unlike server-side DANE , where the verifier derives the TLSA owner name from transport coordinates (the _port._proto prefix), client naming conventions are application-specific; this document describes several below and does not prescribe one. Because the TLS layer is general-purpose and has no knowledge of these conventions, the client conveys the full owner name and the server performs no construction of its own.

Format 1: Service Specific Client Identity In this format, the owner name of the client TLSA record has the following structure: The first label identifies the application service name. The remaining labels form the client domain name. Encoding the application service name into the owner name allows the same client domain name to have different authentication credentials for different application services. There is no need to encode the transport label - the same name form is usable with both TLS and DTLS. The _service label could be a custom string for an application, but more commonly is expected to be a service name registered in the IANA Service Name Registry. The RDATA or data field portion of the TLSA record is formed exactly as specified in and , and carries the same meaning.

Format 2: IoT Device Identity The Device Identity form of the TLSA record has the following structure: The "_device" label interposed between the client device name labels and the organization domain labels allows management of all client identities to be delegated to a subzone or to another party.

Format 3: Freeform Client Identity In this format, the owner name of the client TLSA record is an ordinary DNS domain name, with no service label or other structural label interposed: This form imposes no structure beyond a domain name that identifies the client. Because it need not contain any underscore-prefixed labels, it can be represented directly in the dNSName component of the Subject Alternative Name extension of an X.509 certificate. It is therefore the most suitable form for deployments whose client certificates are issued by a public Certification Authority, or that otherwise rely on the certificate to carry the client's name, since public CAs and the dNSName type do not generally permit underscore-prefixed labels.

Example TLSA Records for Clients The following examples are provided in the textual presentation format of the DNS TLSA record.

Format 1: Service Specific Client Identity An example TLSA record for the client "device1.example.com." and the application "smtp-client". This record specifies the SHA-256 hash of the subject public key component of the end-entity certificate corresponding to the client. The certificate usage for this record is 3 (DANE-EE) and thus is validated in accordance with .

Format 2: DevID An example TLSA record for the device named "sensor7" managed by the organization "example.com". This record specifies the SHA-512 hash of the subject public key component of an EE certificate corresponding to the client. The example below shows a wildcard TLSA record mapped to a TLSA record with a DANE-TA usage mode. This allows all client identifiers matching the wildcard to be authenticated by client certificates issued by an organization-managed Certification Authority. This example presumes an organization-managed CA willing to issue the underscore-prefixed device name; see .

Format 3: Freeform Client Identity An example TLSA record for the client "device73.example.com", whose certificate is issued by a public Certification Authority. This record specifies certificate usage 1 (PKIX-EE) and the SHA-256 hash of the subject public key component of the end-entity certificate corresponding to the client.

Authentication Model The authentication model assumed in this document is the following: The client is assigned an identity corresponding to a DNS domain name. The client has a private and public key pair. Where client certificates are being used, the client also has a certificate binding the name to its public key. The certificate or public key has a corresponding TLSA record published in the DNS, which allows it to be authenticated directly via the DNS (using the DANE-TA or DANE-EE certificate usage modes) or via a PKIX public CA system constraint if the client's certificate was issued by a public CA (using the PKIX-TA or PKIX-EE DANE usage modes).

Client Identifiers in X.509 Certificates The client conveys its DNS name to the server using the TLS DANE Client Identity extension (see ). With the DANE-EE(3) certificate usage, the presented certificate or public key is matched directly against the TLSA record, so the client's name need not be present in the certificate at all. With the DANE-TA(2), PKIX-TA(0), and PKIX-EE(1) certificate usages, authentication additionally requires the presented certificate to be matched to a reference identity, as specified in . In these cases the client's name, as conveyed in the TLS DANE Client Identity extension, MUST be present as a dNSName entry in the certificate's Subject Alternative Name extension, and the server MUST verify that it matches; if it does not, the server MUST abort the connection with a handshake_failure alert. The choice of naming convention () interacts with the certificate usage mode; see for guidance.

Signaling the Client's DANE Identity in TLS The client MUST explicitly signal that it has a DANE identity. The most important reason is that the server needs an explicit indication from the client that it has a DANE record, so as to avoid unnecessary DNS queries in-band with the TLS handshake. The DANE Client Identity TLS extension is used for this purpose. A conforming client always conveys its DANE client identity (i.e., domain name) in this extension (see ), so the server never needs to derive it from the certificate. This is particularly important when using TLS raw public key authentication, where there is no certificate from which an identity could be extracted, and when a client certificate contains multiple identities, only a specific one of which has a DANE record. The format of this extension and the detailed client and server behavior are specified in .

TLS DANE Client Identity Extension The DANE Client Identity extension type, "dane_clientid", will have a value assigned and registered in the IANA "TLS ExtensionType Values" registry. Its extension_data field always has the following format: ; ]]> The ClientName field contains the full owner name of the client's TLSA record in textual presentation format, as described in RFC 1035, omitting the trailing dot. This is the exact DNS name the server queries; the server performs no further construction of it (see ). The lower bound length value of 0 octets indicates that no client name is present (this form is used by the TLS server, in its CertificateRequest message, to advertise that it supports the DANE client authentication protocol). A DNS name is a sequence of octets rather than a Unicode string. Where the client's identity includes internationalized labels, those labels MUST be expressed in their A-label (ASCII-Compatible Encoding) form , which is the form present in the DNS and, per , the form required in a certificate's dNSName. U-labels (native Unicode) MUST NOT be used. The wire format of a domain name is limited to 255 octets. In keeping with the practice of most TLS extensions, this extension specifies the use of the textual presentation format of domain names instead. In theory, the presentation format can exceed 255 characters because it allows the expression of any arbitrary octet with the "\DDD" sequence of characters (where DDD is the decimal value). Applications using this extension (and the DANE TLSA Client Authentication protocol more generally) should ensure that client domain names being used do not need to resort to the \DDD syntax by limiting the alphabet suitably, such as only allowing letters, digits, hyphens, and underscores. This ensures that the presentation format client domain name will comfortably fit within the 255 octet limit. A TLS server implementing this specification MUST send the "dane_clientid" extension, with a zero-length ClientName, in its CertificateRequest message, to indicate that it understands the extension and is capable of performing DANE client authentication. The server supplies no name of its own. A TLS client implementing this specification and intending to use DANE client authentication with the TLS server MUST send an extension of type "dane_clientid", with a nonzero-length ClientName populated with the full owner name of the client's corresponding DNS TLSA record (see ), in its Certificate message. Per the TLS protocol, the client is only permitted to send the extension if it sees the corresponding extension in the server's CertificateRequest message.

Changes to Client and Server Behavior

Client Behavior A TLS client conforming to this specification MUST have a DNSSEC signed TLSA record published corresponding to its DNS name and X.509 certificate or public key. A client can only use this protocol with a server that supports it and advertises that support by including a DANE Client Identity extension (with a zero-length ClientName) in its CertificateRequest message. When the client receives such a CertificateRequest, and intends to authenticate using DANE, it presents its certificate or public key and MUST send the DANE Client Identity extension in its Certificate message, carrying a nonzero-length ClientName populated with the full owner name of its TLSA record, as specified in .

Server Behavior A TLS Server implementing this specification performs the following steps:

1. Request a client certificate in the TLS handshake's "Certificate Request" message, that includes a DANE Client Identity extension with a zero-length ClientName.
2. Receive the client's Certificate message, together with any DANE Client Identity extension it carries.
3. If the client sent a nonzero-length ClientName in the DANE Client Identity extension of its Certificate message, extract the client's domain name from it. If the client sent the extension with a zero-length ClientName, this is a protocol violation (a conforming client sends a nonzero-length name, per ); the server MUST abort the connection with a handshake_failure alert. If the client did not send the extension at all, it has not requested DANE client authentication, and this specification does not apply to the connection. A server that supports DANE client authentication alongside other modes (for example, traditional TLS client authentication, or accepting unauthenticated clients) then handles the connection according to its general client-authentication policy; a server that requires DANE client authentication MUST abort the connection with a handshake_failure alert.
4. Look up the TLSA record set in the DNS, using the client's domain name obtained in the previous step directly as the DNS query name. The response MUST be cryptographically validated using DNSSEC. The server could perform DNSSEC validation itself, authenticating the full chain back to a configured trust anchor (normally the DNS root). Alternatively, it could also be configured to trust responses obtained via a validating resolver to which it has a secure connection, by requiring the Authenticated Data (AD) bit to be set in the responses. If DNSSEC validation fails, the server MUST either abort the connection with a handshake_failure TLS alert, or treat the client as unauthenticated, if TLS server policy allows.
5. Extract the RDATA of the TLSA records and match them to the presented client certificate according to the rules specified in the DANE TLS protocol . If successfully matched, the client is authenticated and the TLS session proceeds. If unsuccessful, the server MUST either abort the connection with a handshake_failure TLS alert, or treat the client as unauthenticated, if TLS server policy allows.
6. If there are multiple records in the TLSA record set, then the client is authenticated as long as at least one of the TLSA records matches, subject to digest agility, which SHOULD be implemented.

When the client's name is expected to appear in the certificate, and the certificate contains multiple dNSName identities, the ClientName sent in the TLS DANE Client Identity extension unambiguously indicates which one is the client's name. If the name in the TLS DANE Client Identity extension does not match one of the dNSNames in the certificate, then the server MUST abort the connection with a handshake_failure TLS alert. Servers may have their own allowlisting and authorization rules for which certificates they accept. For example a TLS server may be configured to only allow TLS sessions from clients with certificate identities within a specific domain or set of domains. If such rules are not met, the TLS server MUST either abort the connection with a handshake_failure TLS alert, or treat the client as unauthenticated, if TLS server policy allows.

Raw Public Keys When using raw public keys in TLS, there is no certificate to carry the client's name, so the DANE Client Identity extension (which a conforming client always sends; see ) is the only means by which the client's identity is conveyed. The associated DANE TLSA records employ only certificate usage 3 (DANE-EE) and a selector value of 1 (SPKI), as described in .

Operational Considerations The naming conventions described in interact with the DANE certificate usage mode a deployment chooses, because of the way a client's name may need to appear in its certificate. requires that a dNSName in the Subject Alternative Name extension be in the "preferred name syntax", which permits only letters, digits, and hyphens within a label. Underscore-prefixed labels, such as the _service label of the service-specific format or the _device label of the device-identity format (), therefore cannot appear in a conformant dNSName. This restriction applies to any dNSName, independent of the DANE certificate usage mode. With the DANE-EE(3) usage the client's name need not appear in the certificate (), and with raw public keys there is no certificate at all (). Deployments using these can use any of the naming formats in , including those with underscore-prefixed labels, since the name is conveyed only in the TLS DANE Client Identity extension and matched directly against the TLSA record. With the DANE-TA(2), PKIX-TA(0), and PKIX-EE(1) usages the client's name must appear as a dNSName in the certificate and be matched against it (), and must therefore be in the preferred name syntax. Whether an underscore-bearing name can be used in these latter cases depends in practice on which certification authority issues the client certificate. The preferred name syntax restriction is not enforced by many TLS and X.509 toolkits, which will create, encode, and verify certificates containing underscore-prefixed dNSNames. An organization operating its own certification authority can therefore issue such certificates for use with the DANE-TA usage, as in the wildcard example of . Commercial certification authorities, however, enforce the restriction and will not issue certificates containing underscore-prefixed dNSNames; deployments relying on them, typically with the PKIX-TA and PKIX-EE usages, cannot use the underscore-bearing formats and should adopt a naming convention that avoids such labels, for example the freeform identity of . It is expected that most deployments of this protocol will use the DANE certificate usages (DANE-EE(3) or DANE-TA(2)), often with an organization-managed CA. As described above, the naming conventions of , including the underscore-bearing formats, work in practice with these usages. The considerations concerning the preferred name syntax apply primarily to the comparatively less common deployments that rely on the PKIX-TA and PKIX-EE usages with certificates from a commercial certification authority.

Acknowledgements For detailed reviews and helpful comments, the authors would like to thank Wes Hardaker, Joey Salazar, Eric Rescorla, Paul Wouters, Ash Wilson, Robert Moskowitz, Bill Woodcock, Olle Johansson, Michael Richardson, Sandoche Balakrichenan, Rick van Rein, Mike Ounsworth, Deb Cooley, Rich Salz, and other members of the IETF DANCE and TLS working groups.

Security Considerations This document updates RFC 6698 and RFC 7671 by defining a new way of using DANE for TLS client authentication. Placing client identities in the DNS may pose privacy issues for certain applications, depending on the nature of the clients, the structure and content of the client names, and the mechanisms by which they are queried. In particular, client names that correspond to human persons may pose a graver privacy risk than machine identities. Applications employing this protocol should carefully assess those potential issues, as described below. A design goal of TLS 1.3 is that the client identity is encrypted in the Certificate message, and thus protected from disclosure on the wire. DANE client authentication however relies on the peer (the TLS server in this case) subsequently looking up the client's DANE record in the DNS. Although protocol specifications and implementations to encrypt DNS transport exist, they are very far from ubiquitously deployed. Deployers of DANCE client authentication should thus evaluate the risks of the client name being leaked in this manner by the server, until encrypted DNS transport becomes the norm. A possible way to avoid the TLS server looking up the client's DANE record in the DNS is described in , but it has a number of significant challenges. The service specific client identity form lends itself to a structure that might make it easy for the same client to have multiple identities corresponding to different applications using the same public key (e.g., by using wildcards and DANE-EE mode), which could make this protocol susceptible to cross-protocol attacks where traffic is redirected from one service to another. Deployers of this protocol should avoid this by not sharing per client credentials across distinct applications. Because this mechanism cannot operate over TLS 1.2 or earlier (see ), a server that requires DANE client authentication MUST NOT negotiate a protocol version below TLS 1.3 (or DTLS 1.3) for connections where such authentication is mandatory; otherwise the client would be unable to convey its DANE identity and authentication could be silently bypassed. This specification uses a single, generic TLS alert (handshake_failure) for the various client authentication failure conditions it describes, to avoid leaking unnecessary information to potential adversaries.

IANA Considerations IANA is requested to create the following entry in the "TLS ExtensionType Values" registry: Extension Name "dane_clientid" with value TBD, "TLS 1.3" column values set to "CR, CT", "DTLS-Only" column set to "N", and "Recommended" column set to "N". The 'N' designation in the "Recommended" column is because this extension has very specific use cases.

References Normative References Informative References

Possible Future Work One possible way to address the TLS server-side client identity leak is to suppress the need for the TLS server to lookup the client's DANE record, and instead to have the client supply it. The client could query its own DANE record and the corresponding full DNSSEC authentication chain, and assemble this in a new Certificate Extension that is sent to the TLS server within the handshake. The TLS server would then authenticate this full chain. A specification to do this in the other direction (from the server to the client) already exists: "TLS DNSSEC Chain Extension". However, there are several challenges when considering such an approach from the client end. It is quite a heavyweight operation that some constrained clients may have challenges with (for example LoRaWAN clients). In order to construct the DANE authentication chain, the client would need to perform DNS queries which would still leak its identity to the local network environment without encrypted DNS. Lastly, there may be client-side network impediments to making this work, e.g., middleboxes that prevent DNSSEC-enabled queries from succeeding - one of the original motivations for RFC 9102 in the first place. Nevertheless, if appetite to implement this mechanism exists, a future version of this specification could define the details.