Concise Binary Object Representation (CBOR) Tags for Typed Arrays
draft-ietf-cbor-array-tags-08

Abstract

The Concise Binary Object Representation (CBOR, RFC 7049) is a data format whose design goals include the possibility of extremely small code size, fairly small message size, and extensibility without the need for version negotiation.

The present document makes use of this extensibility to define a number of CBOR tags for typed arrays of numeric data, as well as two additional tags for multi-dimensional and homogeneous arrays. It is intended as the reference document for the IANA registration of the CBOR tags defined.

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 April 10, 2020.

Copyright Notice

Copyright (c) 2019 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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

• 1. Introduction
• 1.1. Terminology
• 2. Typed Arrays
• 2.1. Types of numbers
• 3. Additional Array Tags
• 3.1. Multi-dimensional Array
• 3.1.1. Row-major Order
• 3.1.2. Column-Major order
• 3.2. Homogeneous Array
• 4. Discussion
• 5. CDDL typenames
• 6. IANA Considerations
• 7. Security Considerations
• 8. References
• 8.1. Normative References
• 8.2. Informative References
• Contributors
• Acknowledgements
• Author's Address

1. Introduction

The Concise Binary Object Representation (CBOR, [RFC7049]) provides for the interchange of structured data without a requirement for a pre-agreed schema. RFC 7049 defines a basic set of data types, as well as a tagging mechanism that enables extending the set of data types supported via an IANA registry.

Recently, a simple form of typed arrays of numeric data has received interest both in the Web graphics community [TypedArray] and in the JavaScript specification [TypedArrayES6], as well as in corresponding implementations [ArrayBuffer].

Since these typed arrays may carry significant amounts of data, there is interest in interchanging them in CBOR without the need of lengthy conversion of each number in the array. This can also save space overhead with encoding a type for each element of an array.

This document defines a number of interrelated CBOR tags that cover these typed arrays, as well as two additional tags for multi-dimensional and homogeneous arrays. It is intended as the reference document for the IANA registration of the tags defined.

Note that an application that generates CBOR with these tags has considerable freedom in choosing variants, e.g., with respect to endianness, embedded type (signed vs. unsigned), and number of bits per element, or whether a tag defined in this specification is used at all instead of more basic CBOR. In contrast to representation variants of single CBOR numbers, there is no representation that could be identified as “preferred”. If deterministic encoding is desired in a CBOR-based protocol making use of these tags, the protocol has to define which of the encoding variants are used in which case.

1.1. Terminology

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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

The term “byte” is used in its now customary sense as a synonym for “octet”. Where bit arithmetic is explained, this document uses the notation familiar from the programming language C [C] (including C++14’s 0bnnn binary literals [Cplusplus]), except that the operator “**” stands for exponentiation.

The term “array” is used in a general sense in this document, unless further specified. The term “classical CBOR array” describes an array represented with CBOR major type 4. A “homogeneous array” is an array of elements that are all of the same type (the term is neutral as to whether that is a representation type or an application data model type).

The terms “big endian” and “little endian” are used to indicate a most significant byte first (MSB first) representation of integers, and a least significant byte first (LSB first) representation, respectively.

2. Typed Arrays

Typed arrays are homogeneous arrays of numbers, all of which are encoded in a single form of binary representation. The concatenation of these representations is encoded as a single CBOR byte string (major type 2), enclosed by a single tag indicating the type and encoding of all the numbers represented in the byte string.

2.1. Types of numbers

Three classes of numbers are of interest: unsigned integers (uint), signed integers (two’s complement, sint), and IEEE 754 binary floating point numbers (which are always signed). For each of these classes, there are multiple representation lengths in active use:

Here, sintN stands for a signed integer of exactly N bits (for instance, sint16), and uintN stands for an unsigned integer of exactly N bits (for instance, uint32). The name binaryN stands for the number form of the same name defined in IEEE 754 [IEEE754].

Since one objective of these tags is to be able to directly ship the ArrayBuffers underlying the Typed Arrays without re-encoding them, and these may be either in big endian (network byte order) or in little endian form, we need to define tags for both variants.

In total, this leads to 24 variants. In the tag, we need to express the choice between integer and floating point, the signedness (for integers), the endianness, and one of the four length values.

In order to simplify implementation, a range of tags is being allocated that allows retrieving all this information from the bits of the tag: Tag values from 64 to 87.

The value is split up into 5 bit fields: 0b010_f_s_e_ll, as detailed in Table 2.

The number of bytes in each array element can then be calculated by 2**(f + ll) (or 1 << (f + ll) in a typical programming language). (Notice that 0f and ll are the two least significant bits, respectively, of each nibble (4bit) in the byte.)

In the CBOR representation, the total number of elements in the array is not expressed explicitly, but implied from the length of the byte string and the length of each representation. It can be computed from the length, in bytes, of the byte string comprising the representation of the array by inverting the previous formula: bytelength >> (f + ll).

For the uint8/sint8 values, the endianness is redundant. Only the tag for the big endian variant is used and assigned as such. The Tag that would signify the little endian variant of sint8 MUST NOT be used, its tag number is marked as reserved. As a special case, the Tag that would signify the little endian variant of uint8 is instead assigned to signify that the numbers in the array are using clamped conversion from integers, as described in more detail in Section 7.1.11 (ToUint8Clamp) of the ES6 JavaScript specification [TypedArrayES6]; the assumption here is that a program-internal representation of this array after decoding would be marked this way for further processing, providing “roundtripping” of JavaScript typed arrays through CBOR.

IEEE 754 binary floating numbers are always signed. Therefore, for the float variants (f == 1), there is no need to distinguish between signed and unsigned variants; the s bit is always zero. The Tag numbers where s would be one (which would have Tag values 88 to 95) remain free to use by other specifications.

3. Additional Array Tags

This specification defines three additional array tags. The Multi-dimensional Array tags can be combined with classical CBOR arrays as well as with Typed Arrays in order to build multi-dimensional arrays with constant numbers of elements in the sub-arrays. The Homogeneous Array tag can be used as a signal by an application to identify a classical CBOR array as a homogeneous array, even when a Typed Array does not apply.

3.1. Multi-dimensional Array

A multi-dimensional array is represented as a tagged array that contains two (one-dimensional) arrays. The first array defines the dimensions of the multi-dimensional array (in the sequence of outer dimensions towards inner dimensions) while the second array represents the contents of the multi-dimensional array. If the second array is itself tagged as a Typed Array then the element type of the multi-dimensional array is known to be the same type as that of the Typed Array.

Two tags are defined by this document, one for elements arranged in row-major order, and one for column-major order [RowColMajor].

3.1.1. Row-major Order

Tag:

40

Data Item:

array (major type 4) of two arrays, one array (major type 4) of dimensions, which are unsigned integers distinct from zero, and one array (either a CBOR array of major type 4, or a Typed Array, or a Homogeneous Array) of elements

Data in the second array consists of consecutive values where the last dimension is considered contiguous (row-major order).

Figure 1 shows a declaration of a two-dimensional array in the C language, a representation of that in CBOR using both a multidimensional array tag and a typed array tag.

uint16_t a[2][3] = {
  {2, 4, 8},   /* row 0 */
  {4, 16, 256},
};

<Tag 40> # multi-dimensional array tag
   82       # array(2)
     82      # array(2)
       02     # unsigned(2) 1st Dimension
       03     # unsigned(3) 2nd Dimension
     <Tag 65> # uint16 array
       4c     # byte string(12)
         0002 # unsigned(2)
         0004 # unsigned(4)
         0008 # unsigned(8)
         0004 # unsigned(4)
         0010 # unsigned(16)
         0100 # unsigned(256)

Figure 1: Multi-dimensional array in C and CBOR

Figure 2 shows the same two-dimensional array using the multidimensional array tag in conjunction with a basic CBOR array (which, with the small numbers chosen for the example, happens to be shorter).

<Tag 40> # multi-dimensional array tag
   82       # array(2)
     82      # array(2)
       02     # unsigned(2) 1st Dimension
       03     # unsigned(3) 2nd Dimension
     86     # array(6)
       02      # unsigned(2)
       04      # unsigned(4)
       08      # unsigned(8)
       04      # unsigned(4)
       10      # unsigned(16)
       19 0100 # unsigned(256)

Figure 2: Multi-dimensional array using basic CBOR array

3.1.2. Column-Major order

The multidimensional arrays specified in the previous sub-subsection are in “row major” order, which is the preferred order for the purposes of this specification. An analogous representation that uses “column major” order arrays is provided in this subsection under the tag 1040, as illustrated in Figure 3.

Tag:

1040

Data Item:

as with tag 40, except that the data in the second array consists of consecutive values where the first dimension is considered contiguous (column-major order).

<Tag 1040> # multi-dimensional array tag, column major order
   82       # array(2)
     82      # array(2)
       02     # unsigned(2) 1st Dimension
       03     # unsigned(3) 2nd Dimension
     86     # array(6)
       02      # unsigned(2)
       04      # unsigned(4)
       04      # unsigned(4)
       10      # unsigned(16)
       08      # unsigned(8)
       19 0100 # unsigned(256)

Figure 3: Multi-dimensional array using basic CBOR array, column major order

3.2. Homogeneous Array

Tag:

41

Data Item:

array (major type 4)

This tag identifies the classical CBOR array (a one-dimensional array) tagged by it as a homogeneous array, that is, it has elements that are all of the same application model data type. The element type of the array is thus determined by the application model data type of the first array element.

This can be used in application data models that apply specific semantics to homogeneous arrays. Also, in certain cases, implementations in strongly typed languages may be able to create native homogeneous arrays of specific types instead of ordered lists while decoding. Which CBOR data items constitute elements of the same application type is specific to the application.

Figure 4 shows an example for a homogeneous array of booleans in C++ [Cplusplus] and CBOR.

bool boolArray[2] = { true, false };

<Tag 41>  # Homogeneous Array Tag
   82           #array(2)
      F5        # true
      F4        # false

Figure 4: Homogeneous array in C++ and CBOR

Figure 5 extends the example with a more complex structure.

typedef struct {
  bool active;
  int value;
} foo;
foo myArray[2] = { {true, 3}, {true, -4} };

<Tag 41>
    82  # array(2)
       82  #  array(2)
             F5  # true
             03  # 3
       82 # array(2)
             F5  # true
             23  # -4

Figure 5: Homogeneous array in C++ and CBOR

4. Discussion

Support for both little- and big-endian representation may seem out of character with CBOR, which is otherwise fully big endian. This support is in line with the intended use of the typed arrays and the objective not to require conversion of each array element.

This specification allocates a sizable chunk out of the single-byte tag space. This use of code point space is justified by the wide use of typed arrays in data interchange.

Providing a column-major order variant of the multi-dimensional array may seem superfluous to some, and useful to others. It is cheap to define the additional tag so it is available when actually needed. Allocating it out of a different number space makes the preference for row-major evident.

Applying a Homogeneous Array tag to a Typed Array would usually be redundant and is therefore not provided by the present specification.

5. CDDL typenames

For the use with CDDL [RFC8610], the typenames defined in Figure 6 are recommended:

ta-uint8 = #6.64(bstr)
ta-uint16be = #6.65(bstr)
ta-uint32be = #6.66(bstr)
ta-uint64be = #6.67(bstr)
ta-uint8-clamped = #6.68(bstr)
ta-uint16le = #6.69(bstr)
ta-uint32le = #6.70(bstr)
ta-uint64le = #6.71(bstr)
ta-sint8 = #6.72(bstr)
ta-sint16be = #6.73(bstr)
ta-sint32be = #6.74(bstr)
ta-sint64be = #6.75(bstr)
; reserved: #6.76(bstr)
ta-sint16le = #6.77(bstr)
ta-sint32le = #6.78(bstr)
ta-sint64le = #6.79(bstr)
ta-float16be = #6.80(bstr)
ta-float32be = #6.81(bstr)
ta-float64be = #6.82(bstr)
ta-float128be = #6.83(bstr)
ta-float16le = #6.84(bstr)
ta-float32le = #6.85(bstr)
ta-float64le = #6.86(bstr)
ta-float128le = #6.87(bstr)
homogeneous<array> = #6.41(array)
multi-dim<dim, array> = #6.40([dim, array])
multi-dim-column-major<dim, array> = #6.1040([dim, array])

Figure 6: Recommended typenames for CDDL

6. IANA Considerations

IANA has allocated the tags in Table 3, with the present document as the specification reference. (The reserved value is reserved for a future revision of typed array tags.)

The allocations came out of the “specification required” space (24..255), with the exception of 1040, which came out of the “first come first served” space (256..).

*) 40 or 1040 data item: second element of outer array in data item is native CBOR array (major type 4) or Typed Array (one of Tag 64..87)

7. Security Considerations

The security considerations of RFC 7049 apply; special attention is drawn to the second paragraph of Section 8 of RFC 7049.

The Tag for homogeneous arrays makes a promise about its tagged data item that a maliciously constructed CBOR input can then choose to ignore. As always, the decoder therefore has to ensure that it is not driven into an undefined state by array elements that do not fulfill the promise and that it does continue to fulfill its API contract in this case as well.

As with all formats that are used for data interchange, an attacker may have control over the shape of the data delivered as input to the application, which therefore needs to validate that shape before it makes it the basis of its further processing. One unique aspect that typed arrays add to this is that an attacker might substitute a Uint8ClampedArray for where the application expects a Uint8Array, or vice versa, potentially leading to very different (and unexpected) processing semantics of the in-memory data structures constructed. Applications that could be affected by this therefore will need to be careful about making this distinction in their input validation.

8. References

8.1. Normative References

8.2. Informative References

Contributors

The initial draft for this specification was written by Johnathan Roatch (roatch@gmail.com). Many thanks for getting this ball rolling.

Glenn Engel suggested the tags for multi-dimensional arrays and homogeneous arrays.

Acknowledgements

Jim Schaad provided helpful comments and reminded us that column-major order still is in use. Jeffrey Yaskin helped improve the definition of homogeneous arrays. IANA helped correct an error in a previous version. Francesca Palombini acted as a shepherd, and Alexey Melnikov as responsible area director. Elwyn Davies as Gen-ART reviewer and IESG members Martin Vigoureux, Adam Roach, Roman Danyliw, and Benjamin Kaduk helped finding further improvements of the text; thanks also to the other reviewers.

Author's Address