Constrained Resource Identifiers

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶

This Internet-Draft will expire on 11 January 2024.¶

Copyright Notice

Copyright (c) 2023 IETF Trust and the persons identified as the document authors. All rights reserved.¶

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶

Table of Contents

1. Introduction

The Uniform Resource Identifier (URI) [RFC3986] and its most common usage, the URI reference, are the Internet standard for linking to resources in hypertext formats such as HTML [W3C.REC-html52-20171214] or the HTTP "Link" header field [RFC8288].¶

A URI reference is a sequence of characters chosen from the repertoire of US-ASCII characters. The individual components of a URI reference are delimited by a number of reserved characters, which necessitates the use of a character escape mechanism called "percent-encoding" when these reserved characters are used in a non-delimiting function. The resolution of URI references involves parsing a character sequence into its components, combining those components with the components of a base URI, merging path components, removing dot-segments, and recomposing the result back into a character sequence.¶

Overall, the proper handling of URI references is quite intricate. This can be a problem especially in constrained environments [RFC7228], where nodes often have severe code size and memory size limitations. As a result, many implementations in such environments support only an ad-hoc, informally-specified, bug-ridden, non-interoperable subset of half of RFC 3986.¶

This document defines the Constrained Resource Identifier (CRI) by constraining URIs to a simplified subset and representing their components in Concise Binary Object Representation (CBOR) [RFC8949] instead of a sequence of characters. This allows typical operations on URI references such as parsing, comparison and reference resolution (including all corner cases) to be implemented in a comparatively small amount of code.¶

As a result of simplification, however, CRIs are not capable of expressing all URIs permitted by the generic syntax of RFC 3986 (hence the "constrained" in "Constrained Resource Identifier"). The supported subset includes all URIs of the Constrained Application Protocol (CoAP) [RFC7252], most URIs of the Hypertext Transfer Protocol (HTTP) [RFC9110], Uniform Resource Names (URNs) [RFC8141], and other similar URIs. The exact constraints are defined in Section 2.¶

2. Constraints

A Constrained Resource Identifier consists of the same five components as a URI: scheme, authority, path, query, and fragment. The components are subject to the following constraints:¶

C1.

The scheme name can be any Unicode string (see Definition D80 in [Unicode]) that matches the syntax of a URI scheme (see Section 3.1 of [RFC3986], which constrains schemes to ASCII) and is lowercase (see Definition D139 in [Unicode]). The scheme is always present.¶

C2.

An authority is always a host identified by an IP address or registered name, along with optional port information, and optionally preceded by user information.¶

Alternatively, the authority can be absent; the two cases for this defined in Section 3.3 of [RFC3986] are modeled by two different values used in place of an absent authority:¶

• the path can be root-based (zero or more path components that are each started in the URI with "/", as when the authority is present), or¶
• the path can be rootless, which requires at least one path component.¶

(Note that in Figure 1, no-authority is marked as a feature, as not all CRI implementations will support authority-less URIs.)¶

C3.

A userinfo is a text string built out of unreserved characters (Section 2.3 of [RFC3986]) or "sub-delims" (Section 2.2 of [RFC3986]); any other character needs to be percent-encoded (Section 7.1). Note that this excludes the ":" character, which is commonly deprecated as a way to delimit a cleartext password in a userinfo.¶

C4.

An IP address can be either an IPv4 address or an IPv6 address, optionally with a zone identifier [RFC6874]. Future versions of IP are not supported (it is likely that a binary mapping would be strongly desirable, and that cannot be designed ahead of time, so these versions need to be added as a future extension if needed).¶

C5.

A registered name is a sequence of one or more labels, which, when joined with dots (".") in between them, result in a Unicode string that is lowercase and in Unicode Normalization Form C (NFC) (see Definition D120 in [Unicode]). (The syntax may be further restricted by the scheme. As per Section 3.2.2 of [RFC3986], a registered name can be empty, for which case a scheme can define a default for the host.)¶

C6.

A port is always an integer in the range from 0 to 65535. Ports outside this range, empty ports (port subcomponents with no digits, see Section 3.2.3 of [RFC3986]), or ports with redundant leading zeros, are not supported.¶

C7.

The port is omitted if and only if the port would be the same as the scheme's default port (provided the scheme is defining such a default port) or the scheme is not using ports.¶

C8.

A path consists of zero or more path segments. Note that a path of just a single zero-length path segment is allowed — this is considered equivalent to a path of zero path segments by HTTP and CoAP, but this equivalence does not hold for CRIs in general as they only perform normalization on the Syntax-Based Normalization level (Section 6.2.2 of [RFC3986], not on the scheme-specific Scheme-Based Normalization level (Section 6.2.3 of [RFC3986]).¶

(A CRI implementation may want to offer scheme-cognizant interfaces, performing this scheme-specific normalization for schemes it knows. The interface could assert which schemes the implementation knows and provide pre-normalized CRIs. This can also relieve the application from removing a lone zero-length path segment before putting path segments into CoAP Options, i.e., from performing the check and jump in item 8 of Section 6.4 of [RFC7252]. See also SP1 in Appendix B.)¶

C9.

A path segment can be any Unicode string that is in NFC, with the exception of the special "." and ".." complete path segments. Note that this includes the zero-length string.¶

If no authority is present in a CRI, the leading path segment cannot be empty. (See also SP1 in Appendix B.)¶

C10.

A query always consists of one or more query parameters. A query parameter can be any Unicode string that is in NFC. It is often in the form of a "key=value" pair. When converting a CRI to a URI, query parameters are separated by an ampersand ("&") character. (This matches the structure and encoding of the target URI in CoAP requests.) Queries are optional; there is a difference between an absent query and a single query parameter that is the empty string.¶

C11.

A fragment identifier can be any Unicode string that is in NFC. Fragment identifiers are optional; there is a difference between an absent fragment identifier and a fragment identifier that is the empty string.¶

C12.

The syntax of registered names, path segments, query parameters, and fragment identifiers may be further restricted and sub-structured by the scheme. There is no support, however, for escaping sub-delimiters that are not intended to be used in a delimiting function.¶

C13.

When converting a CRI to a URI, any character that is outside the allowed character range or is a delimiter in the URI syntax is percent-encoded. For CRIs, percent-encoding always uses the UTF-8 encoding form (see Definition D92 in [Unicode]) to convert the character to a sequence of bytes (that is then converted to a sequence of %HH triplets).¶

Examples for URIs at or beyond the boundaries of these constraints are in SP2 in Appendix B.¶

3. Creation and Normalization

In general, resource identifiers are created on the initial creation of a resource with a certain resource identifier, or the initial exposition of a resource under a particular resource identifier.¶

A Constrained Resource Identifier SHOULD be created by the naming authority that governs the namespace of the resource identifier (see also [RFC8820]). For example, for the resources of an HTTP origin server, that server is responsible for creating the CRIs for those resources.¶

The naming authority MUST ensure that any CRI created satisfies the constraints defined in Section 2. The creation of a CRI fails if the CRI cannot be validated to satisfy all of the constraints.¶

If a naming authority creates a CRI from user input, it MAY apply the following (and only the following) normalizations to get the CRI more likely to validate:¶

• map the scheme name to lowercase (C1);¶
• map the registered name to NFC (C5) and split it on embedded dots;¶
• elide the port if it is the default port for the scheme (C7);¶
• map path segments, query parameters and the fragment identifier to NFC form (C9, C10, C11).¶

Once a CRI has been created, it can be used and transferred without further normalization. All operations that operate on a CRI SHOULD rely on the assumption that the CRI is appropriately pre-normalized. (This does not contradict the requirement that when CRIs are transferred, recipients must operate on as-good-as untrusted input and fail gracefully in the face of malicious inputs.)¶

4. Comparison

One of the most common operations on CRIs is comparison: determining whether two CRIs are equivalent, without dereferencing the CRIs (using them to access their respective resource(s)).¶

Determination of equivalence or difference of CRIs is based on simple component-wise comparison. If two CRIs are identical component-by-component (using code-point-by-code-point comparison for components that are Unicode strings) then it is safe to conclude that they are equivalent.¶

This comparison mechanism is designed to minimize false negatives while strictly avoiding false positives. The constraints defined in Section 2 imply the most common forms of syntax- and scheme-based normalizations in URIs, but do not comprise protocol-based normalizations that require accessing the resources or detailed knowledge of the scheme's dereference algorithm. False negatives can be caused, for example, by CRIs that are not appropriately pre-normalized and by resource aliases.¶

When CRIs are compared to select (or avoid) a network action, such as retrieval of a representation, fragment components (if any) should be excluded from the comparison.¶

5. CRI References

The most common usage of a Constrained Resource Identifier is to embed it in resource representations, e.g., to express a hyperlink between the represented resource and the resource identified by the CRI.¶

This section defines the representation of CRIs in Concise Binary Object Representation (CBOR) [RFC8949]. When reduced representation size is desired, CRIs are not represented directly. Instead, CRIs are indirectly referenced through CRI references. These take advantage of hierarchical locality and provide a very compact encoding. The CBOR representation of CRI references is specified in Section 5.1.¶

The only operation defined on a CRI reference is reference resolution: the act of transforming a CRI reference into a CRI. An application MUST implement this operation by applying the algorithm specified in Section 5.3 (or any algorithm that is functionally equivalent to it).¶

The reverse operation of transforming a CRI into a CRI reference is not specified in detail in this document; implementations are free to use any algorithm as long as reference resolution of the resulting CRI reference yields the original CRI. Notably, a CRI reference is not required to satisfy all of the constraints of a CRI; the only requirement on a CRI reference is that reference resolution MUST yield the original CRI.¶

When testing for equivalence or difference, applications SHOULD NOT directly compare CRI references; the references should be resolved to their respective CRI before comparison.¶

; not expressed in this CDDL spec: trailing nulls to be left off

RFC-XXXX-Definitions = [CRI, CRI-Reference]

CRI = [
  scheme,
  authority / no-authority,
  path / null,
  query / null,
  fragment / null
]


CRI-Reference = [
  ((scheme / null, authority / no-authority)
   // discard),                 ; relative reference
  path / null,
  query / [] / null,            ; [] is explicit unset
  fragment / null
]

scheme      = scheme-name / scheme-id
scheme-name = text .regexp "[a-z][a-z0-9+.-]*"
scheme-id   = nint

no-authority = NOAUTH-ROOTBASED / NOAUTH-ROOTLESS
NOAUTH-ROOTBASED = null .feature "no-authority"
NOAUTH-ROOTLESS = true .feature "no-authority"

authority   = [?userinfo, host, ?port]
userinfo    = (false, text .feature "userinfo")
host        = (host-ip // host-name)
host-name   = (*text) ; lowercase, NFC labels
host-ip     = (bytes .size 4 //
               (bytes .size 16, ?zone-id))
zone-id     = text
port        = 0..65535

discard     = DISCARD-ALL / 0..127
DISCARD-ALL = true
path        = [*text]
query       = [+text]
fragment    = text

[-1,             / scheme -- equivalent to "coap" /
 [h'C6336401',   / host /
  61616],        / port /
 [".well-known", / path /
  "core"]
]

[true,                  / discard /
 [".well-known",        / path /
  "core"],
 ["rt=temperature-c"]]  / query /

[-6,             / scheme -- equivalent to "did" /
 true,           / authority = NOAUTH-NOSLASH /
 ["web:alice:bob"] / path /
]

6. Relationship between CRIs, URIs and IRIs

CRIs are meant to replace both Uniform Resource Identifiers (URIs) [RFC3986] and Internationalized Resource Identifiers (IRIs) [RFC3987] in constrained environments [RFC7228]. Applications in these environments may never need to use URIs and IRIs directly, especially when the resource identifier is used simply for identification purposes or when the CRI can be directly converted into a CoAP request.¶

However, it may be necessary in other environments to determine the associated URI or IRI of a CRI, and vice versa. Applications can perform these conversions as follows:¶

CRI to URI

A CRI is converted to a URI as specified in Section 6.1.¶

URI to CRI

The method of converting a URI to a CRI is unspecified; implementations are free to use any algorithm as long as converting the resulting CRI back to a URI yields an equivalent URI.¶

Note that CRIs are defined to enable implementing conversions from or to URIs analogously to processing URIs into CoAP Options and back, with the exception that item 8 of Section 6.4 of [RFC7252] and item 7 of Section 6.5 of [RFC7252] do not apply to CRI processing. See SP1 in Appendix B for more details.¶

CRI to IRI

A CRI can be converted to an IRI by first converting it to a URI as specified in Section 6.1, and then converting the URI to an IRI as described in Section 3.2 of [RFC3987].¶

IRI to CRI

An IRI can be converted to a CRI by first converting it to a URI as described in Section 3.1 of [RFC3987], and then converting the URI to a CRI as described above.¶

Everything in this section also applies to CRI references, URI references and IRI references.¶

7. Extending CRIs

CRIs have been designed to relieve implementations operating on CRIs from string scanning, which both helps constrained implementations and implementations that need to achieve high throughput.¶

The CRI structure described up to this point is termed the Basic CRI. It should be sufficient for all applications that use the CoAP protocol, as well as most other protocols employing URIs.¶

However, Basic CRIs have one limitation: They do not support URI components that require percent-encoding (Section 2.1 of [RFC3986]) to represent them in the URI syntax, except where that percent-encoding is used to escape the main delimiter in use.¶

E.g., the URI¶

https://alice/3%2f4-inch

is represented by the basic CRI¶

[-4, ["alice"], ["3/4-inch"]]

However, percent-encoding that is used at the application level is not supported by basic CRIs:¶

did:web:alice:7%3A1-balun

Extended forms of CRIs may be defined to enable these applications. They will generally extend the potential values of text components of URIs, such as userinfo, hostnames, paths, queries, and fragments.¶

One such extended form is described in the following Section 7.1. Consumers of CRIs will generally notice when an extended form is in use, by finding structures that do not match the CDDL rules given in Figure 1. Future definitions of extended forms need to strive to be distinguishable in their structures from the extended form presented here as well as other future forms.¶

Extensions to CRIs MUST NOT allow indefinite length items. This provision ensures that recipients o CRIs can deal with unprocessable CRIs as described in Section 5.2.1.¶

userinfo    = (false, text .feature "userinfo")
host-name   = (*text)
path        = [*text]
query       = [+text]
fragment    = text

userinfo    = (false, text-or-pet .feature "userinfo")
host-name   = (*text-or-pet)
path        = [*text-or-pet]
query       = [+text-or-pet]
fragment    = text-or-pet

text-or-pet = text /
    text-pet-sequence .feature "extended-cri"

; text1 and pet1 alternating, at least one pet1:
text-pet-sequence = [?text1, ((+(pet1, text1), ?pet1) // pet1)]
; pet is percent-encoded bytes
pet1 = bytes .ne ''
text1 = text .ne ""

[-6, true, [["web:alice:7", ':', "1-balun"]]]

[-6, true, [["web:alice:", '7:', "1-balun"]]]
[-6, true, [["web:alice:7", ':1', "-balun"]]]

8. Implementation Status

With the exception of the authority=true fix, host-names split into labels, and Section 7.1, CRIs are implemented in https://gitlab.com/chrysn/micrurus. A golang implementation of version -10 of this document is found at: https://github.com/thomas-fossati/href¶

9. Security Considerations

Parsers of CRI references must operate on input that is assumed to be untrusted. This means that parsers MUST fail gracefully in the face of malicious inputs. Additionally, parsers MUST be prepared to deal with resource exhaustion (e.g., resulting from the allocation of big data items) or exhaustion of the call stack (stack overflow). See Section 10 of [RFC8949] for additional security considerations relating to CBOR.¶

The security considerations discussed in Section 7 of [RFC3986] and Section 8 of [RFC3987] for URIs and IRIs also apply to CRIs.¶

10. IANA Considerations

Appendix A. Mapping Scheme Numbers to Scheme Names

Table 1 defines the initial mapping from CRI scheme numbers to URI scheme names.¶

The assignments from this table can be extracted from the XML form of this document (when stored in a file "this.xml") into CSV form [RFC4180] using this short Ruby program:¶

require 'rexml/document'; include REXML
XPath.each(Document.new(File.read("this.xml")),"//tr") {|row|
  puts XPath.each(row,"td").map{|d|d.text()}[0..1].join(",")}

Appendix B. The Small Print

This appendix lists a few corner cases of URI semantics that implementers of CRIs need to be aware of, but that are not representative of the normal operation of CRIs.¶

SP1.

Initial (Lone/Leading) Empty Path Segments:¶

• Lone empty path segments: As per [RFC3986], s://x is distinct from s://x/ -- i.e., a URI with an empty path ([] in CRI) is different from one with a lone empty path segment ([""]). However, in HTTP and CoAP, they are implicitly aliased (for CoAP, in item 8 of Section 6.4 of [RFC7252]). As per item 7 of Section 6.5 of [RFC7252], recomposition of a URI without Uri-Path Options from the other URI-related CoAP Options produces s://x/, not s://x -- CoAP prefers the lone empty path segment form. Similarly, after discussing HTTP semantics, Section 6.2.3 of [RFC3986] states:¶

In general, a URI that uses the generic syntax for authority with an empty path should be normalized to a path of "/".¶

• Leading empty path segments without authority: Somewhat related, note also that URIs and URI references that do not carry an authority cannot represent initial empty path segments (i.e., that are followed by further path segments): s://x//foo works, but in a s://foo URI or an (absolute-path) URI reference of the form //foo the double slash would be mis-parsed as leading in to an authority.¶

SP2.

Constraints (Section 2) of CRIs/basic CRIs¶

While most URIs in everyday use can be converted to CRIs and back to URIs matching the input after syntax-based normalization of the URI, these URIs illustrate the constraints by example:¶

• https://host%ffname, https://example.com/x?data=%ff¶

  All URI components must, after percent decoding, be valid UTF-8 encoded text. Bytes that are not valid UTF-8 show up, for example, in BitTorrent web seeds.¶

• https://example.com/component%3bone;component%3btwo, http://example.com/component%3dequals¶

  While delimiters can be used in an escaped and unescaped form in URIs with generally distinct meanings, basic CRIs (i.e., without percent-encoded text Section 7.1) only support one escapable delimiter character per component, which is the delimiter by which the component is split up in the CRI.¶

  Note that the separators . (for authority parts), / (for paths), & (for query parameters) are special in that they are syntactic delimiters of their respective components in CRIs. Thus, the following examples are convertible to basic CRIs:¶

  https://interior%2edot/¶

  https://example.com/path%2fcomponent/second-component¶

  https://example.com/x?ampersand=%26&questionmark=?¶

• https://[email protected]/¶

  The user information can be expressed in CRIs if the "userinfo" feature is present. The URI https://@example.com is represented as [-4, [false, "", "example", "com"]]; the false serves as a marker that the next element is the userinfo.¶

  The rules do not cater for unencoded ":" in userinfo, which is commonly considered a deprecated inclusion of a literal password.¶

Appendix C. Change Log

This section is to be removed before publishing as an RFC.¶

Changes from -08 to -09¶

• Identify more esoteric features with a CDDL ".feature".¶
• Clarify that well-formedness requires removing trailing nulls.¶
• Fragments can contain PET.¶
• Percent-encoded text in PET is treated as byte strings.¶

• URIs with an authority but a completely empty path (e.g., http://example.com): CRIs with an authority component no longer always produce at least a slash in the path component.¶

  For generic schemes, the conversion of scheme://example.com to a CRI is now possible because CRI produces a URI with an authority not followed by a slash following the updated rules of Section 6.1. Schemes like http and coap do not distinguish between the empty path and the path containing a single slash when an authority is set (as recommended in [RFC3986]). For these schemes, that equivalence allows implementations to convert the just-a-slash URI to a CRI with a zero length path array (which, however, when converted back, does not produce a slash after the authority).¶

  (Add an appendix "the small print" for more detailed discussion of pesky corner cases like this.)¶

Changes from -07 to -08¶

• Fix the encoding of NOAUTH-NOSLASH / NOAUTH-LEADINGSLASH¶
• Add URN and DID schemes, add example.¶
• Add PET¶
• Remove hopeless attempt to encode "remote trailing nulls" rule in CDDL (which is not a transformation language).¶

Changes from -06 to -07¶

• More explicitly discuss constraints (Section 2), add examples (Appendix B, Paragraph 6, Item 1).¶
• Make CDDL more explicit about special simple values.¶
• Lots of gratuitous changes from XML2RFC redefinition of <tt> semantics.¶

Changes from -05 to -06¶

• rework authority:¶

  • split reg-names at dots;¶
  • add optional zone identifiers [RFC6874] to IP addresses¶

Changes from -04 to -05¶

• Simplify CBOR structure.¶
• Add implementation status section.¶

Changes from -03 to -04:¶

• Minor editorial improvements.¶
• Renamed path.type/path-type to discard.¶
• Renamed option to section, substructured into items.¶
• Simplified the table "resolution-variables".¶
• Use the CBOR structure inspired by Jim Schaad's proposals.¶

Changes from -02 to -03:¶

• Expanded the set of supported schemes (#3).¶
• Specified creation, normalization and comparison (#9).¶
• Clarified the default value of the path.type option (#33).¶
• Removed the append-relation path.type option (#41).¶
• Renumbered the remaining path.types.¶
• Renumbered the option numbers.¶
• Restructured the document.¶
• Minor editorial improvements.¶

Changes from -01 to -02:¶

• Changed the syntax of schemes to exclude upper case characters (#13).¶
• Minor editorial improvements (#34 #37).¶

Changes from -00 to -01:¶

• None.¶

Acknowledgements

CRIs were developed by Klaus Hartke for use in the Constrained RESTful Application Language (CoRAL). The current author team is completing this work with a view to achieve good integration with the potential use cases, both inside and outside of CoRAL.¶

Thanks to Christian Amsüss, Thomas Fossati, Ari Keränen, Jim Schaad, Dave Thaler and Marco Tiloca for helpful comments and discussions that have shaped the document.¶

Contributors

Authors' Addresses