= Project Connected Home over IP
:toc: macro
:toclevels: 7

[discrete]
[.text-center]
== TLV Format

[.text-center]
_Revision 6_ +
_2020-05-08_

[.text-center]
*Status:* [red]*Final* / [red]*Active*

toc::[]

== Introduction

This document describes the CHIP TLV (_Tag-Length-Value_) format. CHIP
TLV is a generalized encoding method for simple structured data. It
shares many properties with the commonly used JSON serialization format
while being considerably more compact over the wire.

== TLV Elements and Encodings

Values in the CHIP TLV format are encoded as _TLV elements_. Each TLV
element has a type. Element types fall into two categories: _primitive
types_ and _container types_. Primitive types convey fundamental data
values such as integers and strings. Container types convey collections
of elements that themselves are either primitives or containers. The
CHIP TLV format supports three different container types: structures,
arrays and lists.

All valid _TLV encodings_ consist of a single top-level element. This
value can be either a primitive type or a container type.

== Tags

A TLV element includes an optional numeric tag that identifies its
purpose. Two categories of tags are defined: _profile-specific _and
_context-specific_. A TLV element without a tag is called an _anonymous_
element.

=== Profile-Specific Tags

Profile-specific tags identify elements globally. A profile-specific tag
is a 64-bit number composed of the following fields:

* 16-bit vendor id
* 16-bit profile number
* 32-bit tag number

Profile-specific tags are defined either by CHIP or by external vendors.
Additionally the CHIP Common Profile includes a set of predefined
profile-specific tags that can be used across organizations.

=== Context-Specific Tags

Context-specific tags identify elements within the context of a
containing structure element. A context-specific tag consists of a
single 8-bit tag number. The meaning of a context specific tag derives
from the structure it resides in, implying that the same tag number may
have different meanings in the context of different structures.
Effectively, the interpretation of a context specific tag depends on the
tag attached to the containing element. Because structures themselves
can be assigned context-specific tags, the interpretation of a
context-specific tag may ultimately depend on a nested chain of such
tags.

Context-specific tags can only be assigned to elements that are
immediately within a structure. This implies that an element with a
context-specific tag cannot appear as the outermost element of a TLV
encoding.

=== Anonymous Tags

A special anonymous tag is used to denote TLV elements that lack a tag
value. Such a TLV element is referred to as an anonymous element.  

=== Canonical Ordering of Tags

Where a distinguished ordering of tags is required (e.g. for the purposes
of generating a hash or cryptographic signature of elements within a
structure), the following ordering rules shall be used:

* Anonymous tags shall be ordered before all other tags.
* Context specific tags shall be ordered before profile specific tags.
* Context specific tags with numerically lesser tag values shall be ordered
before those with higher tag values.
* Profile specific tags with numerically lesser vendor ids shall be ordered
before those with higher vendor ids.
* Profile specific tags with the same vendor id, but numerically lesser
profile numbers shall be ordered before those with higher numbers.

== Lengths

Depending on its type, a TLV element may contain a length field that
gives the length, in bytes, of the element’s value field. A length field
is only present for string types (character and byte strings). Other
element types either have a predetermined length or are encoded with a
marker that identifies their end.

== Primitive Types

The CHIP TLV format supports the following primitive types:

* Signed integers
* Unsigned integers
* UTF-8 Strings
* Byte Strings
* Single or double-precision floating point numbers (IEEE 754-1985
format)
* Booleans
* Nulls

Of the primitive types, integers, floating point numbers, booleans and
nulls have a predetermined length specified by their type. Byte strings
and UTF-8 strings include a length field that gives their lengths in
bytes.

== Container Types

The CHIP TLV format supports the following container types:

* Structures
* Arrays
* Lists

Each of the container types is a form of element collection that can
contain primitive types and/or other container types. The elements
appearing immediately within a container type are called its _members_.
A container type can contain any number of member elements, including
none. Container types can be nested to any depth and in any combination.
The end of a container type is denoted by a special element called the
‘end-of-container’ element. Although encoded as a member, conceptually
the end-of-container element is not included in the members of the
containing type.

=== Structures

A structure is a collection of member elements that each have a distinct
meaning. All member elements within a structure must have a unique tag
as compared to the other members of the structure. Member elements
without tags (anonymous elements) are not allowed in structures. The
encoded ordering of members in a structure may or may not be important
depending on the intent of the sender or the expectations of the
receiver. For example, in some situations, senders and receivers may
agree on a particular ordering of elements to make encoding and decoding
easier.

Where a distinguished ordering of members is required (for example, for
the purposes of generating a hash or cryptographic signature of the
structure), the members of the structure shall be encoded in order of
their tag values, according to the rules for the canonical ordering of tags.

=== Arrays

An array is an ordered collection of member elements that either do not
have distinct meanings, or whose meanings are implied by their encoded
positions in the array. An array can contain any type of element,
including other arrays. All member elements of an array must be anonymous
elements–that is, they must be encoded with an anonymous tag.

=== Lists

A list is an ordered collection of member elements, each of which may be
encoded with a tag. The meanings of member elements in a list are denoted
by their position within the list in conjunction with any associated tag
value they may have.

A list can contain any type of element, including other lists. The members
of a list may be encoded with any form of tag, including an anonymous tag.
The tags within a list needn’t be unique with respect to other members of
the list.

|===
**Note:** In a previous version of this specification, the list type was
referred to as a “path”. It was subsequently renamed to better reflect its
use in practice.
|===

== Element Encoding

A TLV element is encoded a single control byte, followed by a sequence
of tag, length and value bytes. Depending on the nature of the element,
any of the tag, length or value fields may be omitted.

[options="header"]
|===
|Control Byte|Tag|Length|Value
|1 byte|0 to 8 bytes|0 to 8 bytes|Variable
|===

== Control Byte Encoding

The control byte specifies the type of a TLV element and how its tag,
length and value fields are encoded. The control byte consists of two
subfields: an _element type field_ which occupies the lower 5 bits, and
a _tag control field_ which occupies the upper 3 bits.

=== Element Type Field

The element type field encodes the element’s type as well as how the
corresponding length and value fields are encoded. In the case of
Booleans and the null value, the element type field also encodes the
value itself.

[cols="^1,^1,^1,^1,^1,^1,^1,^1,<8",options="header"]
|===
8+h|Control Byte
.3+h|Description
3+h|Tag Control
5+h|Element Type
h|7
h|6
h|5
h|4
h|3
h|2
h|1
h|0
|x|x|x|0|0|0|0|0|Signed Integer, 1-byte value
|x|x|x|0|0|0|0|1|Signed Integer, 2-byte value
|x|x|x|0|0|0|1|0|Signed Integer, 4-byte value
|x|x|x|0|0|0|1|1|Signed Integer, 8-byte value
|x|x|x|0|0|1|0|0|Unsigned Integer, 1-byte value
|x|x|x|0|0|1|0|1|Unsigned Integer, 2-byte value
|x|x|x|0|0|1|1|0|Unsigned Integer, 4-byte value
|x|x|x|0|0|1|1|1|Unsigned Integer, 8-byte value
|x|x|x|0|1|0|0|0|Boolean False
|x|x|x|0|1|0|0|1|Boolean True
|x|x|x|0|1|0|1|0|Floating Point Number, 4-byte value
|x|x|x|0|1|0|1|1|Floating Point Number, 8-byte value
|x|x|x|0|1|1|0|0|UTF-8 String, 1-byte length
|x|x|x|0|1|1|0|1|UTF-8 String, 2-byte length
|x|x|x|0|1|1|1|0|UTF-8 String, 4-byte length
|x|x|x|0|1|1|1|1|UTF-8 String, 8-byte length
|x|x|x|1|0|0|0|0|Byte String, 1-byte length
|x|x|x|1|0|0|0|1|Byte String, 2-byte length
|x|x|x|1|0|0|1|0|Byte String, 4-byte length
|x|x|x|1|0|0|1|1|Byte String, 8-byte length
|x|x|x|1|0|1|0|0|Null
|x|x|x|1|0|1|0|1|Structure
|x|x|x|1|0|1|1|0|Array
|x|x|x|1|0|1|1|1|List
|x|x|x|1|1|0|0|0|End of Container
|x|x|x|1|1|0|0|1|Reserved
|x|x|x|1|1|0|1|0|Reserved
|x|x|x|1|1|0|1|1|Reserved
|x|x|x|1|1|1|0|0|Reserved
|x|x|x|1|1|1|0|1|Reserved
|x|x|x|1|1|1|1|0|Reserved
|x|x|x|1|1|1|1|1|Reserved
|===

For types that have varying length or value fields, the bottom two bits
of the element type field signal the width of the corresponding field as
follows:

* 00 -- 1 byte
* 01 -- 2 bytes
* 10 -- 4 bytes
* 11 -- 8 bytes

=== Tag Control Field

The tag control field identifies the form of tag assigned to the element
(including none) as well as the encoding of the tag bytes.

[cols="^1,^1,^1,^1,^1,^1,^1,^1,<8",options="header"]
|===
8+h|Control Byte
.3+h|Description
3+h|Tag Control
5+h|Element Type
h|7
h|6
h|5
h|4
h|3
h|2
h|1
h|0
|0|0|0|x|x|x|x|x|Anonymous, 0 bytes
|0|0|1|x|x|x|x|x|Context-specific Tag, 1 byte
|0|1|0|x|x|x|x|x|Common Profile Tag, 2 bytes
|0|1|1|x|x|x|x|x|Common Profile Tag, 4 bytes
|1|0|0|x|x|x|x|x|Implicit Profile Tag, 2 bytes
|1|0|1|x|x|x|x|x|Implicit Profile Tag, 4 bytes
|1|1|0|x|x|x|x|x|Fully-qualified Tag, 6 bytes
|1|1|1|x|x|x|x|x|Fully-qualified Tag, 8 bytes
|===

== Tag Encoding

Tags are encoded in 0, 1, 2, 4, 6 or 8 byte widths as specified by the
tag control field. Tags consist of up to three numeric fields: a _vendor
id field_, a _profile number field_, and a _tag number field_. All
fields are encoded in little-endian order.

=== Fully-Qualified Form

A profile-specific tag can be encoded in _fully-qualified form_, where
the encoding includes all three tag components (vendor id, profile
number and tag number). Two variants of this form are supported, one
with a 16-bit tag number and one with a 32-bit tag number. The 16-bit
variant must be used with tag numbers < 65536, while the 32-bit variant
must be used with tag numbers >= 65536.

[options="header"]
|===
|Tag Control|Vendor Id Size|Profile Number Size|Tag Number Size|
|C0h|2 bytes|2 bytes|2 bytes|For tag numbers < 65536
|E0h|2 bytes|2 bytes|4 bytes|For tag numbers >= 65535
|===

=== Implicit Form

A profile-specific tag can also be encoded in _implicit form_, where the
encoding includes only the tag number, and the vendor id and profile
number are inferred from the protocol context in which the TLV encoding
is communicated. This form also has two variants based on the magnitude
of the tag number.

[options="header"]
|===
|Tag Control|Tag Number Size|
|80h|2 bytes|For tag numbers < 65536
|A0h|4 bytes|For tag numbers >= 65535
|===

=== Common Profile Form

A special encoding exists for profile-specific tags that are defined by
the CHIP Common Profile. These are encoded in the same manner as
implicit tags except that they are identified as common profile tags,
rather than implicit profile tags in the tag control field.

[options="header"]
|===
|Tag Control|Tag Number Size|
|40h|2 bytes|For tag numbers < 65536
|60h|4 bytes|For tag numbers >= 65535
|===

=== Context-Specific Form

Context-specific tags are encoded as a single byte conveying the tag
number.

[options="header"]
|===
|Tag Control|Tag Number Size|
|20h|1 bytes|All tag numbers 0 - 255
|===

=== Anonymous

Anonymous elements do not encode any tag bytes.

[options="header"]
|===
|Tag Control|Tag Size|
|00h|0 bytes|No data encoded.
|===

== Length Encoding

Length fields are encoded in 0, 1, 2 or 4 byte widths, as specified by
the element type field. Length fields of more than one byte are encoded
in little-endian order. The choice of width for the length field is up
to the discretion of the sender, implying that a sender can choose to
send more length bytes than strictly necessary to encode the value.

== End of Container Encoding

The end of a container type is marked with a special element called the
end-of-container element. The end-of-container element is encoded as a
single control byte with the value 18h. The tag control bits within the
control byte must be set to zero, implying that end-of-container element
can never have a tag.

[options="header"]
|===
|Control Byte
|1 byte
|===

== Value Encodings

=== Integers

An integer element is encoded as follows:

[options="header"]
|===
|Control Byte|Tag|Value
|1 byte|0 to 8 bytes|1, 2, 4 or 8 bytes
|===

The number of bytes in the value field is indicated by the element type
field within the control byte. The choice of value byte count is at the
sender’s discretion, implying that a sender is free to send more bytes
than strictly necessary to encode the value. Within the value bytes, the
integer value is encoded in little-endian two’s complement format.

=== UTF-8 and Byte Strings

UTF-8 and byte strings are encoded as follows:

[options="header"]
|===
|Control Byte|Tag|Length|Value
|1 byte|0 to 8 bytes|1 to 4 bytes|0 to 2^32^-1 bytes
|===

The length field of a UTF-8 or byte string encodes the number of bytes
(not characters) present in the value field. The number of bytes in the
length field is implied by the type specified in the element type field
(within the control byte).

For UTF-8 strings, the value bytes must encode a valid UTF-8 character
sequence. Senders *should not* include a terminating null character to
mark the end of a string. For byte strings, the value can be any
arbitrary sequence of bytes.

=== Booleans

Boolean elements are encoded as follows:

[options="header"]
|===
|Control Byte|Tag
|1 byte|0 to 8 bytes
|===

The value of a Boolean element (true or false) is implied by the type
indicated in the element type field.

=== Arrays, Structures and Lists

Array, structure and list elements are encoded as follows:

[options="header"]
|===
|Control Byte|Tag|Value|End-of-Container
|1 byte|0 to 8 bytes|_Variable_|1-byte
|===

The value field of an array/structure/list element is a sequence of
encoded TLV elements that constitute the members of the element,
followed by an end-of-container element. The end-of-container element
must always be present, even in cases where the end of the
array/structure/list element could be inferred by other means (e.g. the
length of the packet containing the TLV encoding).

=== Floating Point Numbers

A floating point number is encoded as follows:

[options="header"]
|===
|Control Byte|Tag|Value
|1 byte|0 to 8 bytes|4 or 8 bytes
|===

The value field of a floating point element contains an IEEE 754-1985
single or double precision floating point number encoded in
little-endian format (specifically, the reverse of the order described
in External Data Representation, RFC 4506). The choice of precision is
implied by the type specified in the element type field (within the
control byte). The sender is free to choose either precision at their
discretion.

=== Nulls

A null value is encoded as follows:

[options="header"]
|===
|Control Byte|Tag
|1 byte|0 to 8 bytes
|===

[discrete]
== Revision History

[cols="^1,^1,<1,<4",options="header"]
|===
|Revision |Date |Modified By |Description
|6|2020-05-08|Jay Logue|Renamed path type to list and clarified text in corresponding section.

Added section on canonical ordering of tags.

Added section describing anonymous tags.

Clarified descriptions of structure and array types.

Clarified encoding of floating point values with reference to RFC-4506.
|5|2020-04-23|Grant Erickson|Import into Project Connected Home over IP.
|4|2013-05-20|Jay Logue|Fixed incorrect control byte value for end of container encoding.
|3|2013-04-22|Jay Logue|Renamed dictionary to structure.
|2|2013-04-17|Jay Logue|Normalized the naming for ‘container’ types.
|1|2013-04-15|Jay Logue|Initial revision.
|===

[.text-center]
_Project Connect Home over IP Public Information_