]> Penango, Inc.

11400 West Olympic Boulevard Suite 1500 Los AngelesCA90064 USA dev+ietf@seantek.com http://www.penango.com/

Applications Network Working Group CERT X.509 PKIX Certificate URN URI Digital certificates are used in many systems and protocols to identify and authenticate parties. This document describes a Uniform Resource Name (URN) namespace that identifies certificates. These URNs can be used when certificates need to be identified by value or reference.

Digital certificates are used in many systems and protocols to identify and authenticate parties. Security considerations frequently require that the certificate must be identified with certainty, because selecting the wrong certificate will lead to validation errors (resulting in denial of service), or in improper credential selection (resulting in unwanted disclosure or substitution attacks). The goal of this namespace is to provide a uniform syntax for identifying certificates with precision in Uniform Resource Identifiers (URIs), specifically Uniform Resource Names (URNs). Using this syntax, any protocol or system that refers to a certificate in a textual format can unambiguously identify that certificate by value or reference. Implementers that parse these URNs can resolve them into actual certificates. Examples include:

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.

Although certificates have diverse applications, there has been no uniform way to refer to a certificate in text. De-facto standards such as PEM and PKIX text encoding are used to include whole certificates in textual formats, but this practice is impractical for a variety of use cases. Certificates that identify long public keys (e.g., 2048-bit RSA keys) and that contain required and recommended PKIX extensions can easily exceed many kilobytes in length. The purpose of this document is to provide a uniform textual format for identifying individual certificates. Certificate specifications, or "certspecs", are not designed or intended to provide a search tool or query language to match multiple certificates; the goal is to replace data elements that would otherwise have to include whole certificates in order to identify them. When a URN resolver resolves a "certspec", the resolver's output is expected to be a single certificate or nothing.

Identifying a specific certificate by reference or value allows diverse applications to have a common syntax. For example, applications can store certspecs as local or shared preferences, so that users can edit them without resorting to application-specific storage formats or relying on the availability of particular protocols represented by URLs (such as http:, ldap:, file:, or ni: schemes). When conveyed in protocol, a certspec can identify a specific certificate to a client or server using text-based formats such as YAML, XML, JSON, and others. The format described in this document is intended to be readily reproducible by users using common certificate processing tools, so that users can easily create, recognize, compare, and reproduce them at a glance. For example, the hash-based identifications use hexadecimal encoding so that a user can easily compose or compare an URN with a simple copy-and-paste operation. Accordingly, some tradeoffs have been made in favor of human usability.

When the certificate represented by a certspec needs to be resolved, an application can resort to any number of schemes. For example, when the certificate is identified by hash, the application can resolve the certspec to a Named Information (ni:) URI for further processing. When the certificate is identified by issuer and serial number, the application can resolve the certspec to an LDAP service (for example, ldap:/&wj;/&wj;/&wj;cn=ExampleCA,o=ExampleCo,st=California,c=US ).

A certspec is intended to identify a single certificate unambiguously. A certificate has no more than one corresponding certspec per certspec type; however, a certificate is expected to have an array of certspecs that identify the certificate. The choice of which certspec to use in a given situation is context-specific.

A certspec is a URN that complies with the modern URN syntax, with a few exceptions for usability. Following , NID is "cert", and the Namespace Specific String (NSS) has the ABNF below. The query and fragment productions are relevant to certspecs; these are discussed in .

A certspec NSS is comprised of two parts: certspec-type and certspec-value. The certspec-type identifies the certificate specification type. The acceptable characters for spec-type are the same as those in an URI scheme name (Section 3.1 in ); types are compared case-insensitively. The certspec-value identifies the certificate specification value. The acceptable characters for spec-value depend on the spec-type, but are never more than pchar except for relaxed human usability reasons in a few cases discussed below. In several cases, characters are significantly restricted, to the point that percent-encoding is prohibited. In such cases, a generator MUST NOT generate percent-encoded values, and a parser MUST treat the production as an error. The reasons are to simplify processing in such cases, and to improve human usability. Several certspecs use hexadecimal encodings of octets. Generally: if the hex octets are malformed (whether in the source material, such as the corresponding certificate element, or in the hex text), the certspec is invalid.

Standard certificate specifications are intended for interchange as durable, persistent, unique, and intuitive (to users and developers) identifiers for individual certificates--the exact criteria for URNs. This section provides four cryptographic hash-based certspecs, two content-based certspecs, and two element-based certspecs.

A cryptographic hash or "fingerprint" of a certificate uniquely identifies that certificate. For hash-based certspecs, the hash is computed over the octets of the DER encoding of the certificate, namely, the Certificate type in Section 4.1 of . The certspec-value is the hexadecimal encoding of the hash value octets. For example, a 256-bit SHA-256 hash is represented by exactly 32 hex octets, or 64 hex characters. Lexical equivalence of two hash-based certspecs that have the same certspec-type SHALL be determined by a case-insensitive comparison of certspec-values, or by converting the hexadecimal certspec-values to octets and comparing exact equivalence of the octets. A conforming implementation MUST reject values that contain non-hex digits, such as spaces, tabs, hyphens, percent-encoded characters, or anything else. Conforming implementations to this Internet-Draft MUST process these hash-based certspecs, unless security considerations dictate otherwise. Acceptable reasons for refusing to process a certspec include a) the local policy prohibits use of the hash, or b) the hash has known cryptographic weaknesses, such as a preimage attacks, which weaken the cryptographic uniqueness guarantees of the hash.

The certspec-type is "SHA-1". The hash is computed using SHA-1 .

The certspec-type is "SHA-256". The hash is computed using SHA-256 .

The certspec-type is "SHA-384". The hash is computed using SHA-384 .

The certspec-type is "SHA-512". The hash is computed using SHA-512 .

A certificate may be identified reflexively by its constituent octets. For small-to-medium certificates, identifying the certificate by embedding it in the certspec will be computationally efficient and resistant to denial-of-service attacks (by always being available). A conforming implementation MUST implement base64 and hex specs. The octets of a certificate are the octets of the DER encoding of the certificate, namely, the Certificate type in Section 4.1 of . The DER encoding includes tag and length octets, so it always starts with 30h (the tag for SEQUENCE). Lexical equivalence of two certspecs that are value-based SHALL be determined by decoding the certspec-value to certificate octets, and comparing the octets for strict equivalence. Accordingly, it is possible that base64 and hex certspecs are lexically equivalent. Because users may end up copying and pasting base64 or hex-encoded certificates into certspecs, and because these certspecs will routinely exceed 72 characters, a production might contain embedded whitespace. If there are contexts where line breaks or other whitespace must be allowed for practical reasons, the implementation should consider the URN in context as "a URN, possibly with embedded whitespace (which is ignored)".

The certspec-type is "base64". The certspec-value is the base64url encoding of the certificate octets (Section 5 of ), but MAY be relaxed as follows. Unlike the data: URL, URN NSS productions are not supposed to have the "/" character, which is integral to standard base64. On the other hand, it is anticipated that users will want to copy-and-paste base64 encoded certificates--such as those produced by PKIX text encodings--directly into base64 certspecs. Generators of base64 certspecs SHOULD emit base64url-encoded data, where the characters '-' and '_' refer to values 62 and 63, respectively, and where the trailing equal signs '=' are absent. Alternatively, generators MAY emit base64 data with precent-encoding for the non-pchar conformant characters (specifically "/"). In any event, generators MUST NOT generate non-pchar conformant characters (specifically "/"). Parsers of base64 certspecs that are not under strict URN conformance constraints MUST also accept '+' and '/' as values 62 and 63, respectively, and MUST accept trailing '=' characters in conformance with standard base64. None of '+', '/', or '=' have reserved meanings in this certspec-type. This relaxed parsing rule is reflected in the base64relaxed production of . Similarly, states that non-reserved characters (in this case, alphanumerics) must not be "%"-encoded, but a lenient implementation MAY decode these "%"-encoded characters anyway. This document neither recommends nor discourages such leniency, but implementors should weigh the benefits and risks as discussed further in the Security Considerations (). Overall, percent-encoding in base64 certspecs is permissible because unlike most of the other certspecs, the complete base64 encoding is not expected to be human-readable or identifiable at a glance.

The certspec-type is "hex". The certspec-value is the hexadecimal encoding of the certificate octets. Percent-encoding is not allowed; implementations MUST NOT process percent-encoded values. The reasons are because percent-encoding would reduce the human readability of the certspec, and (marginally) increase the complexity of certspec parsers.

A certificate may be identified by certain data elements contained within it. The following certspecs reflect the traditional reliance of PKIX and CMS on a certificate's issuer distinguished name and serial number, or a certificate's subject key identifier. If some of an element-based certspec is based on the DER encoded part of a certificate, and if the encodings are incorrect, the URN is invalid.

The certspec-type is "issuersn". The distinguishedNameUC production encodes the certificate's issuer distingiushed name (DN) field in LDAP string format, whose characters are subsequently percent-encoded to conform to URN NSS syntax. The <distinguishedName> on which distinguishedNameUC is based is defined in , and <SEMI> is defined in . no longer separates relative distinguished names (RDNs) by semicolons, as required by its predecessor, . Accordingly, ';' is used to separate the issuer's DN from the subject's serial number. Care should be taken in escaping and percent-encoding the relevant characters. In particular: "?" is permitted in a distinguishedName, but MUST NOT appear in a URN unless it delimits the query component (see ). Any question marks in distinguished names MUST be percent-encoded when placed in the certspec-value. "#" is used as a token at the beginning of the hexstring production for attributeValue data, but MUST NOT appear in a URN unless it delimits the fragment component (see ). Any "#" characters in distinguished names MUST be percent-encoded when placed in the certspec-value. Due to the nature of distinguished names, it is likely that a given LDAP string will contain spaces (for example, between words). Additionally, implementations might consider emitting spacing between name components for readability. For human usability, it is anticipated that parsers would encounter such whitespace, even though those characters are supposed to be percent-encoded in URNs. Parsers SHOULD accept spaces when parsing this certspec; generators MAY emit spaces when strict conformance to URN syntax is less important than human readability (for example, when the URN is rendered for display, or in cases where the URN is expected to be handled by humans). These considerations are reflected in the distinguishedNameUCrelaxed production. For reference, the following characters are permitted in distinguished names in the issuer production: [[TODO put table here of valid characters, and spaces]] The serialNumber production is the hexadecimal encoding of the contents octets of the DER encoding of the CertificateSerialNumber ::= INTEGER as specified in Section 4.1 of . A conforming implementation SHOULD implement this issuersn certspec. If the implementation implements it, the implementation MUST process serial numbers up to the same length as required by Section 4.1.2.2 of (20 octets), and MUST process distinguished name strings as required by , including the table of minimum AttributeType name strings that MUST be recognized. Additionally, implementations MUST process attribute descriptors specified in (MUST or SHOULD), and (specifically: E, email, emailAddress). Implementations are encouraged to recognize additional attribute descriptors where possible. A complete list of attribute descriptors for reference is provided in . Lexical equivalence of two issuersn certspecs SHALL be determined by comparing the serialNumbers for exact equivalence, and comparing the issuer distinguished names for a match. The lexical equivalence of serialNumbers SHALL be determined by a case-insensitive comparison of them, or by converting the hexadecimal text to octets and comparing exact equivalence of the octets. A conforming implementation MUST reject values that contain non-hex digits, such as spaces, tabs, hyphens, percent-encoded characters, or anything else. Distinguished names match if they satisfy the name matching requirements of and [TODO: or matching rules in LDAP RFCs].

The certspec-type is "ski". The certspec-value is the hexadecimal encoding of the certificate's subject key identifier, which is recorded in the certificate's Subject Key Identifier extension (Section 4.2.1.2 of ). The octets are the DER-encoded contents octets of the SubjectKeyIdentifier (OCTET STRING) extension value. A certificate that lacks a subject key identifier cannot and MUST NOT be identified using this spec. Lexical equivalence of two ski certspecs SHALL be determined by a case-insensitive comparison of certspec-values, or by converting the hexadecimal certspec-values to octets and comparing exact equivalence of the octets. A conforming implementation MUST reject values that contain non-hex digits, such as spaces, tabs, hyphens, percent-encoded characters, or anything else. A conforming implementation MAY implement this ski spec.

The additional certificate specifications in this section are provided for applications to use as local identifiers that are useful, intuitive, or supportive of legacy systems or overriding design goals. These certspecs SHOULD NOT be used for interchange.

The certspec-type is "data". This document reserves this spec-type for future use. An implementation may embed the contents of a data URL (data URI) into the certspec-value. Specifically:

See . In such a case, the mediatype SHOULD be "application/pkix-cert" since the data URL components identify a certificate; however, an implementation MAY be able to support other media types so long as a single certificate is extractable from the data production. Data URLs containing certificates generally will not conform to URN syntax "as-is". The considerations of stuffing base64-encoded content into URNs discussed in apply to this certspec as well, bearing in mind that data URLs only contain traditional base64 (not base64url)-encoded data, or binary percent-encoded data. Because this certspec is content-based, an implementation can determine lexical equivalence with other content-based certspecs.

The spec-type is "dbkey". This document reserves this spec-type for future use.

The certspec-type is "subject". The certspec-value is the RFC 4514 LDAP string encoding of the certificate's subject distinguished name. Characters MAY be percent-encoded; implementations MUST process the percent-encoded characters in the certspec-value before further LDAP string processing. All the considerations of encoding the issuer field in apply to this type.

clarifies that the query and fragment productions of apply to URNs. This document provides semantics for these productions, as applied to certificates.

As a certspec identifies a single certificate, two certspecs are identical lexically or semantically if the NSS parts identify the same certificate. The query and fragment productions do not affect this equivalence.

A certspec URN can have attributes (i.e., metadata) that are associated with--but not instrinsic to--the certificate or its identifiers. The syntax is intended primarily to convey certificate metadata such as attributes found in PKCS #9, PKCS #11, PKCS #12, and particular implementations of cryptographic libraries. This document does not further define certattrs; the characters of certattrs can be any valid query character from .

A certspec can include a fragment that identifies a part of interest within the identified certificate. For example, a user agent may wish to draw attention to the notAfter time for an expired certificate. This document defines the following fragments ("certparts"): Identifier Certificate Part (ASN.1 identifier) vtbsCertificate.version sntbsCertificate.serialNumber sigtbsCertificate.signature; also signatureAlgorithm issuertbsCertificate.issuer notBeforetbsCertificate.validity.notBefore notAftertbsCertificate.validity.notAfter subjecttbsCertificate.subject spkitbsCertificate.subjectPublicKeyInfo exttbsCertificate.extensions ext:tbsCertificate.extensions {Extension matching extoid == extnID}* sigvalsignatureValue * The particular extension in the Extensions SEQUENCE is identified by OID only; there are no textual identifiers. The certparts defined in the table above are case-insensitive. Should additional certparts be required, a future document may specify additional certparts that match the other-certpart production.

certspec is useful wherever a system may need to include or refer to a certificate. Some implementations may wish to refer to a certificate without enabling all of the expressive power (and security considerations) of URIs. Accordingly, this section provides a uniform method for using a certspec outside of a URN. Examples:

To use certspec outside of a URI (URN) context, simply omit the prefix "urn:cert:". All other lexical rules apply, including percent-encoding, query (certattrs), and fragment (certparts). Care should be taken to process "?" and "#" in particular, since they delimit the attributes and parts. A conforming implementation of raw certspecs MUST permit the prefix "urn:cert:" in addition to the raw certspec. Additionally, this document guarantees that the the certspec-types "urn" and "cert" are RESERVED and will never be used. However, implementors must take note that a raw certspec is not a valid URI, because certspec-types are not registered URI schemes and do not have the same semantics as URIs.

This document requests the assignment of formal URN namespace ID "cert". [[TODO: Consider...This document requests the creation of a registry to record specs.]] New certspec types shall be ratified by the IETF consensus process. [[Some commenters have suggested the creation of a registry for certspec types. This is under consideration. One drawback is that it is desirable to limit the certspec types for interoperability and recognizability reasons--probably the only reason to include more types is for using new hashes as old hash algorithms become cryptanalyzed.]]

Digital certificates are important building blocks for authentication, integrity, authorization, and (occasionally) confidentiality services. Accordingly, identifying digital certificates incorrectly can have significant security ramifications. When using hash-based certspecs, the cryptographic hash algorithm MUST be implemented properly and SHOULD have no known attack vectors. For this reason, algorithms that are considered "broken" as of the date of this Internet-Draft, such as MD5, are precluded from being valid certspecs. The registration of a particular algorithm spec in this namespace does NOT mean that it is acceptable or safe for every usage, even though this Internet-Draft requires that a conforming implementation MUST implement certain specs. When using content-based certspecs, the implementation MUST be prepared to process URNs of arbitrary length. As of this writing, useful certificates rarely exceed 10KB, and most implementations are concerned with keeping certificate sizes down rather than up [CITE: Google SSL overclocking, etc.]. However, a pathological or malicious certificate could easily exceed these metrics. If an URN resolver cannot process a URN's full length, it MUST reject the certspec. When using element-based certspecs, the implementation MUST be prepared to deal with multiple found certificates that contain the same certificate data, but are not the same certificate. In such a case, the implementation MUST segregate these certificates so that it only resolves the URN to certificates that it considers valid or trustworthy (as discussed further below). If, despite this segregation, multiple valid or trustworthy certificates match the certspec, the certspec MUST be rejected, because a certspec is meant to identify exactly one certificate (not a family of certificates). Certificates identified by certspecs should only be used with an analysis of their validity, such as by computing the Certification Path Validation Algorithm ([RFC5280 sec. 6]) or by other means. For example, if a certificate database contains a set of certificates that it considers inherently trustworthy, then the inclusion of a certificate in that set makes it trustworthy, regardless of the results of the Certification Path Validation Algorithm. Such a database is frequently used for "Root CA" lists.

A Uniform Resource Name (URN) is a Uniform Resource Identifier (URI) that is intended to serve as a persistent, location-independent resource identifier. This document defines the canonical syntax for URIs under the "urn" scheme, guidelines for URN namespaces, requirements for URN presentation and transmission, and methods for determining URN equivalence. This document obsoletes RFC 2141. National Institute of Standards and Technology &RFC2119; &RFC2141; &RFC2397; &RFC3406; &RFC3986; &RFC4512; &RFC4514; &RFC4648; &RFC5280; &RFC5750; This document describes and discuss the text encodings of Public-Key Infrastructure using X.509 (PKIX) Certificates, PKIX Certificate Revocation Lists (CRLs), PKCS #10 Certification Request Syntax, PKCS #7 structures, Cryptographic Message Syntax (CMS), PKCS #8 Private- Key Information Syntax, and Attribute Certificates. The text encodings are well-known, are implemented by several applications and libraries, and are widely deployed. This document is intended to articulate the de-facto rules that existing implementations operate by, and to give recommendations that will promote interoperability going forward. &RFC1421; &RFC2253; &RFC3187; &RFC5652; &RFC6151; &RFC6920;

As per , attribute descriptors case-insensitive. [TODO: add table.]