Protocol-independent Content Negotiation Framework
A shared vocabulary and framework for agreeing on which version of a resource to send
Klyne, G., "Protocol-independent Content Negotiation Framework", RFC 2703, DOI 10.17487/RFC2703, September 1999, <https://www.rfc-editor.org/info/rfc2703>.
@misc{rfc2703,
series = {Request for Comments},
number = 2703,
howpublished = {RFC 2703},
publisher = {RFC Editor},
doi = {10.17487/RFC2703},
url = {https://www.rfc-editor.org/info/rfc2703},
author = {G. Klyne},
title = {{Protocol-independent Content Negotiation Framework}},
pagetotal = 20,
year = 1999,
month = sep,
} RFC text
Contents
Network Working Group G. Klyne
Request for Comments: 2703 5GM/Content Technologies
Category: Informational September 1999
Protocol-independent Content Negotiation Framework
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (1999). All Rights Reserved.
Abstract
A number of Internet application protocols have a need to provide
content negotiation for the resources with which they interact. MIME
media types [1,2] provide a standard method for handling one major
axis of variation, but resources also vary in ways which cannot be
expressed using currently available MIME headers.
This memo sets out terminology, an abstract framework and goals for
protocol-independent content negotiation, and identifies some
technical issues which may need to be addressed.
The abstract framework does not attempt to specify the content
negotiation process, but gives an indication of the anticipated scope
and form of any such specification. The goals set out the desired
properties of a content negotiation mechanism.
Table of Contents
1. Introduction.............................................2
1.1 Structure of this document ...........................3
1.2 Discussion of this document ..........................3
2. Terminology and definitions..............................3
3. Framework................................................7
3.1 Abstract framework for content negotiation ...........8
3.1.1 The negotiation process..........................9
3.2 Abstract model for negotiation metadata .............10
3.3 Text representation for negotiation metadata ........11
3.4 ASN.1 description of negotiation metadata ...........11
3.5 Protocol binding guidelines .........................11
4. Goals...................................................12
4.1 Generic framework and metadata goals ................12
4.2 Protocol-specific deployment goals ..................12
5. Technical issues........................................14
5.1 Non-message resource transfers ......................14
5.2 End-to-end vs hop-by-hop negotiations ...............14
5.3 Third-party negotiation .............................15
5.4 Use of generic directory and resolution services ....15
5.5 Billing issues ......................................15
5.6 Performance considerations ..........................15
5.7 Confidence levels in negotiated options .............16
6. Security Considerations.................................16
6.1 Privacy .............................................16
6.2 Denial of service attacks ...........................17
6.3 Mailing list interactions ...........................17
6.4 Use of security services ............................17
6.5 Disclosure of security weaknesses ...................18
6.5.1 User agent identification.......................18
6.5.2 Macro viruses...................................18
6.5.3 Personal vulnerability..........................18
6.6 Problems of negotiating security ....................18
7. Acknowledgements........................................18
8. References..............................................19
9. Author's Address........................................19
10. Full Copyright Statement...............................20
1. Introduction
A number of Internet application protocols have a need to provide
content negotiation for the resources with which they interact.
While MIME media types [1, 2] provide a standard method for handling
one major axis of variation, resources also vary in ways which cannot
be expressed using currently available MIME headers.
This memo sets out terminology, a framework and some goals for a
protocol-independent content negotiation framework, and identifies
some technical issues which may need to be addressed.
The framework does not attempt to specify the content negotiation
process; rather it gives an indication of the anticipated scope and
form of any such specifications.
The statement of goals is intended to set out the desired properties
of a content negotiation framework, while trying to avoid any
assumption of the form that framework may take.
1.1 Structure of this document
The main part of this memo addresses four main areas:
Section 2 defines some of the terms which are used with special
meaning.
Section 3 outlines a proposed framework for describing protocol-
independent content negotiation.
Section 4 describes various goals for content negotiation.
Section 5 discusses some of the technical issues which are raised by
this document, with cross-references to other work where appropriate.
1.2 Discussion of this document
Discussion of this document should take place on the content
negotiation and media feature registration mailing list hosted by the
Internet Mail Consortium (IMC).
Please send comments regarding this document to:
ietf-medfree@imc.org
To subscribe to this list, send a message with the body 'subscribe'
to "ietf-medfree-request@imc.org".
To see what has gone on before you subscribed, please see the mailing
list archive at:
http://www.imc.org/ietf-medfree/
2. Terminology and definitions
This section introduces a number of terms which are used with
specific meaning in the content negotiation documents. Many of these
have been copied and adapted from [5].
The terms are listed in alphabetical order.
Capability
An attribute of a sender or receiver (often the receiver)
which indicates an ability to generate or process a
particular type of message content.
Characteristic
Some description of a sender or receiver which indicates a
possible capability or preference.
Choice message
A choice message returns a representation of some selected
variant or variants, together with the variant list of the
negotiable resource. It can be generated when the sender
has sufficient information to select a variant for the
receiver, and also requires to inform the receiver about
the other variants available.
Connected mode
A mode of operation in which sender and receiver are
directly connected, and hence are not prevented from
definitively determining each other's capabilities. (See
also: Session mode)
Content feature
(see Feature)
Content negotiation
An exchange of information (negotiation metadata) which
leads to selection of the appropriate representation
(variant) when transferring a data resource.
Data resource
A network data object that can be transferred. Data
resources may be available in multiple representations
(e.g. multiple languages, data formats, size, resolutions)
or vary in other ways. (See also: Message, Resource)
Feature A piece of information about the media handling properties
of a message passing system component or of a data
resource.
Feature tag
A name that identifies a "feature".
Feature set
Information about a sender, recipient, data file or other
participant in a message transfer which describes the set
of features that it can handle.
Where a 'feature' describes a single identified attribute
of a resource, a 'feature set' describes full set of
possible attributes.
List message
A list message sends the variant list of a negotiable
resource, but no variant data. It can be generated when
the sender does not want to, or is not allowed to, send a
particular variant.
Media feature
information that indicates facilities assumed to be
available for the message content to be properly rendered
or otherwise presented. Media features are not intended to
include information that affects message transmission.
Message Data which is transmitted from a sender to a receiver,
together with any encapsulation which may be applied.
Where a data resource is the original data which may be
available in a number of representations, a message
contains those representation(s) which are actually
transmitted. Negotiation metadata is not generally
considered to be part of a message.
Message data is distinguished from other transmitted data
by the fact that its content is fully determined before the
start of transmission.
Negotiated content
Message content which has been selected by content
negotiation.
Negotiation
(See: content negotiation)
Negotiable resource
A data resource which has multiple representations
(variants) associated with it. Selection of an appropriate
variant for transmission in a message is accomplished by
content negotiation between the sender and recipient.
Negotiation metadata
Information which is exchanged between the sender and
receiver of a message by content negotiation in order to
determine the variant which should be transferred.
Neighbouring variant
A particular representation (variant) of a variant resource
which can safely be assumed to be subject to the same
access controls as the variant resource itself. Not all
variants of a given variant resource are necessarily
neighbouring variants. The fact that a particular variant
is or is not a neighbouring variant has implications for
security considerations when determining whether that
variant can be sent to a receiver in place of the
corresponding variant resource. It may also have
implications when determining whether or not a sender is
authorized to transmit a particular variant.
Preference
An attribute of a sender or receiver (often the receiver)
which indicates an preference to generate or process one
particular type of message content over another, even if
both are possible.
Receiver A system component (device or program) which receives a
message.
Receiver-initiated transmission
A message transmission which is requested by the eventual
receiver of the message. Sometimes described as 'pull'
messaging. E.g. an HTTP GET operation.
Resource A document, data file or facility which is accessed or
transmitted across a network. (See also: Data resource)
Sender A system component (device or program) which transmits a
message.
Sender-initiated transmission
A message transmission which is invoked by the sender of
the message. Sometimes described as 'push' messaging. E.g.
sending an e-mail.
Session mode
A mode of message transmission in which confirmation of
message delivery is received by the sender in the same
application session (usually the same transport connection)
that is used to transmit the message. (See also: connected
mode, store and forward mode)
Store and forward mode
A mode of message transmission in which the message is held
in storage for an unknown period of time on message
transfer agents before being delivered.
Syntax The form used to express some value; especially the format
used to express a media feature value, or a feature set.
(See also: feature value, feature set, type.)
Transmission
The process of transferring a message from a sender to a
receiver. This may include content negotiation.
Type The range of values that can be indicated by some
identifier of variable; especially the range of values
that can be indicated by a feature tag. (See also:
feature, syntax.)
NOTE: this differs from usage employed by the LDAP/X.500
directory community, who use the terms "attribute type" to
describe an identifier for a value in a directory entry,
and "attribute syntax" to describe a range of allowed
attribute values.
User agent
A system component which prepares and transmits a message,
or receives a message and displays, prints or otherwise
processes its contents.
Variant One of several possible representations of a data
resource.
Variant list
A list containing variant descriptions, which can be bound
to a negotiable resource.
Variant description
A machine-readable description of a variant resource,
usually found in a variant list. A variant description
contains a variant resource identifier and various
attributes which describe properties of the variant.
Variant resource
A data resource for which multiple representations
(variants) are available.
3. Framework
For the purposes of this document, message transmission protocol
capabilities are explicitly disregarded: it is presumed that these
will be dealt with separately by some orthogonal mechanism.
Content negotiation covers three elements:
1. expressing the capabilities of the sender and the data resource to
be transmitted (as far as a particular message is concerned),
2. expressing the capabilities of a receiver (in advance of the
transmission of the message), and
3. a protocol by which capabilities are exchanged.
These negotiation elements are addressed by a negotiation framework
incorporating a number of design elements with dependencies shown:
[ Abstract ] [ Abstract ]
[negotiation] [ negotiation ]
[ process ] [ metadata ]
| |
V V
[Negotiation] [ Negotiation ]
[ protocol ] [ metadata ]
[ binding ] [representation]
| |
------- -------
| |
V V
[Application protocol]
[ incorporating ]
[content negotiation ]
Within this overall framework, expressing the capabilities of sender
and receiver is covered by negotiation metadata. The protocol for
exchanging capabilities is covered by the abstract negotiation
framework and its binding to a specific application protocol.
Application protocol independence is addressed by separating the
abstract negotiation process and metadata from concrete
representations and protocol bindings.
3.1 Abstract framework for content negotiation
The negotiation framework provides for an exchange of negotiation
metadata between the sender and receiver of a message which leads to
determination of a data format which the sender can provide and the
recipient can process. Thus, there are three main elements which are
the subjects of the negotiation process and whose capabilities are
described by the negotiation metadata: the sender, the transmitted
data file format and the receiver.
The life of a data resource may be viewed as:
(C) (T) (F)
[A]-->--[S]-->--[R]-->--[U]
where:
[A] = author of document
(C) = original document content
[S] = message sending system
(T) = transmitted data file (representation of (C))
[R] = receiving system
(F) = formatted (rendered) document data (presentation of (C))
[U] = user or consumer of a document
Here, it is [S] and [R] who exchange negotiation metadata to decide
the form of (T), so these elements are the focus of our attention.
Negotiation metadata provided by [S] would take account of available
document content (C) (e.g. availability of resource variants) as well
as its own possible ability to offer that content in a variety of
formats.
Negotiation metadata provided by [R] would similarly take account of
the needs and preferences of its user [U] as well as its own
capabilities to process and render received data.
3.1.1 The negotiation process
Negotiation between the sender [S] and the receiver [R] consists of a
series of negotiation metadata exchanges that proceeds until either
party determines a specific data file (T) to be transmitted. If the
sender makes the final determination, it can send the file directly.
Otherwise the receiver must communicate its selection to the sender
who sends the indicated file.
This process implies an open-ended exchange of information between
sender and receiver. Not every implementation is expected to
implement this scheme with the full generality thus implied. Rather,
it is expected that every concrete negotiation can be viewed as a
subset of this process.
For example, Transparent Content Negotiation (TCN) [5] uses a model
in which one of the following happens:
o The recipient requests a resource with no variants, in which case
the sender simply sends what is available.
o A variant resource is requested, in which case the server replies
with a list of available variants, and the client chooses one
variant from those offered.
o The recipient requests a variant resource, and also provides
negotiation metadata (in the form 'Accept' headers) which allows
the server to make a choice on the client's behalf.
Another, simpler example is that of fax negotiation: in this case
the intended recipient declares its capabilities, and the sender
chooses a message variant to match.
Each of these can be viewed as a particular case of the general
negotiation process described above. Similar observations can be
made regarding the use of directory services or MIME '
Multipart/alternative' in conjunction with e-mail message
transmission.
3.2 Abstract model for negotiation metadata
A simple but general negotiation framework has been described, which
is based on the exchange of negotiation metadata between sender and
recipient. The mechanism by which data is exchanged is not important
to the abstract negotiation framework, but something does need to be
said about the general form of the metadata.
The terminology and definitions section of this document places
constraints on the form of negotiation metadata, and the descriptions
that follow should be read in conjunction with the definitions to
which they refer.
Negotiation metadata needs to encompass the following elements:
o Media feature: a way to describe attributes of a data resource.
o Feature set: a description of a range of possible media feature
combinations which can be: offered by a sender; represented by a
data file format; or processed by a receiver.
o One or more naming schemes for labelling media features and
feature sets. These should be backed up by some kind of
registration process to ensure uniqueness of names and to
encourage a common vocabulary for commonly used features.
o A framework of data types for media features, indicating the range
and properties of value types which can be represented.
o A way to combine media features into feature sets, capable of
expressing feature dependencies within a feature set (e.g.
640x480 pixel size and 256 colours, or 800x600 pixel size and 16
colours).
o Some way to rank feature sets based upon sender and receiver
preferences for different feature values.
3.3 Text representation for negotiation metadata
A concrete textual representation for media feature values and
feature set descriptions would provide a common vocabulary for
feature data in text-based protocols like HTTP and SMTP.
In defining a textual representation, the issue of allowable
character sets needs to be addressed. Whether or not negotiation
metadata needs to support a full gamut of international characters
will depend upon the framework of data types adopted for media
features. As negotiation metadata would be used as a protocol
element (not directly visible to the user) rather than part of the
message content, support for extended character sets may be not
required.
A textual representation for negotiation metadata would imply a
textual representation for media feature names, and also for
expressions of the media feature combining algebra.
3.4 ASN.1 description of negotiation metadata
For use with non-text-based protocols, an ASN.1 description and
encoding designation for negotiation metadata could be helpful for
incorporating the common negotiation framework into ASN.1-derived
protocols like X.400, X.500, LDAP and SNMP.
An ASN.1 description of negotiation metadata formats suggests that
separate media feature naming scheme based on ISO object identifiers
would be valuable.
3.5 Protocol binding guidelines
Specific protocol bindings will be needed to use the abstract
framework for negotiation.
Details of protocol bindings would be beyond the scope of this work,
but guidelines maybe not. (SASL might provide a useful model here.)
4. Goals
These goals are presented in two categories:
1. Negotiation framework and metadata goals which address the broad
goals of negotiation in a protocol-independent fashion.
2. Specific goals which relate to the deployment of negotiation in
the context of a specific protocol (e.g. relation to HTTP protocol
operations, cache interactions, security issues, existing HTTP
negotiation mechanisms, application to variant selection, etc.).
These would be addressed by a specific protocol binding for the
negotiation framework.
4.1 Generic framework and metadata goals
o A common vocabulary for designating features and feature sets.
o A stable reference for commonly used features.
o An extensible framework, to allow rapid and easy adoption of new
features.
o Permit an indication of quality or preference.
o Capture dependencies between feature values
o A uniform framework mechanism for exchanging negotiation metadata
should be defined that can encompass existing negotiable features
and is extensible to future (unanticipated) features.
o Efficient negotiation should be possible in both receiver
initiated ('pull') and sender initiated ('push') message
transfers.
o The structure of the negotiation procedure framework should stand
independently of any particular message transfer protocol.
o Be capable of addressing the role of content negotiation in
fulfilling the communication needs of less able computer users.
4.2 Protocol-specific deployment goals
o A negotiation should generally result in identification of a
mutually acceptable form of message data to be transferred.
o If capabilities are being sent at times other than the time of
message transmission, then they should include sufficient
information to allow them to be verified and authenticated.
o A capability assertion should clearly identify the party to whom
the capabilities apply, the party to whom they are being sent, and
some indication of their date/time or range of validity. To be
secure, capability assertions should be protected against
interception and substitution of valid data by invalid data.
o A request for capability information, if sent other than in
response to delivery of a message, should clearly identify the
requester, the party whose capabilities are being requested, and
the time of the request. It should include sufficient information
to allow the request to be authenticated.
o In the context of a given application, content negotiation may use
one or several methods for transmission, storage, or distribution
of capabilities.
o The negotiation mechanism should include a standardized method for
associating features with resource variants.
o Negotiation should provide a way to indicate provider and
recipient preferences for specific features.
o Negotiation should have the minimum possible impact on network
resource consumption, particularly in terms of bandwidth and
number of protocol round-trips required.
o Systems should protect the privacy of users' profiles and
providers' inventories of variants.
o Protocol specifications should identify and permit mechanisms to
verify the reasonable accuracy of any capability data provided.
o Negotiation must not significantly jeopardize the overall
operation or integrity of any system in the face of erroneous
capability data, whether accidentally or maliciously provided.
o Intelligent gateways, proxies, or caches should be allowed to
participate in the negotiation.
o Negotiation metadata should be regarded as cacheable, and explicit
cache control mechanisms provided to forestall the introduction of
ad-hoc cache-busting techniques.
o Automatic negotiation should not pre-empt a user's ability to
choose a document format from those available.
5. Technical issues
5.1 Non-message resource transfers
The ideas for generic content negotiation have been conceived and
developed in the context of message-oriented data transmissions.
Message data is defined elsewhere as a data whose entire content is
decided before the start of data transmission. The following are
examples of non-message data transfers.
o streamed data,
o interactive computations,
o real-time data acquisition,
Does a proposed approach to negotiation based on message data
reasonably extend to streamed data (e.g. data whose content is not
fully determined by the time the first data items are transmitted)?
It may be that the metadata will be applicable, but the abstract
negotiation process framework may be insufficient to these more
demanding circumstances.
5.2 End-to-end vs hop-by-hop negotiations
Could this distinction place any special demands or constraints on a
generic negotiation framework, or is this simply a protocol issue?
o End-to-end negotiation gives greatest confidence in the outcome.
o Hop-by-hop may have advantages in a network of occasionally-
connected systems, but will place additional demands on
intervening message transmission agents.
Hop-by-hop negotiation implies that negotiation responses are not
necessarily a definitive indication of an endpoint system's
capabilities. This in turn implies a possible need for time-to-live
and re-verification mechanisms to flush out stale negotiation data.
Note that one of the stated goals is to allow proxies and caches to
participate in the negotiation process, as appropriate.
5.3 Third-party negotiation
An extension of the hop-by-hop vs. end-to-end negotiation theme is to
consider the implications of allowing any system other than an
endpoint participant in the message transmission to supply
negotiation metadata.
Any use of a third party in the negotiation process inevitably
increases the possibilities for introducing errors into the
negotiation metadata.
One particular example of a third party participant in a negotiation
process that is frequently suggested is the use of a directory
service using LDAP or similar protocols. What additional steps need
to be taken to ensure reasonable reliability of negotiation metadata
supplied by this means?
5.4 Use of generic directory and resolution services
It is clearly helpful to use existing protocols such as LDAP to
exchange content negotiation metadata.
To achieve this, it be necessary to define directory or other schema
elements which are specific to content negotiation. For example, an
LDAP attribute type for a media feature set.
5.5 Billing issues
Negotiation may raise some billing-related issues in some contexts
because it potentially incurs a two-way exchange of data not
necessarily completed during a single connection. There is an issue
of who pays for return messages, etc., in a non-connected environment
like e-mail or fax.
5.6 Performance considerations
Negotiation can impact performance in both positive and negative
ways.
The obvious negative impact arises from the exchange of additional
data which necessarily consumes some additional bandwidth. There is
also an issue of round-trip or third-party query delays while
negotiation metadata is being exchanged before transmission of the
message itself is commenced.
Over the Internet, there are some bandwidth/latency trade-offs which
can be made. For example, in Internet e-mail the MIME type '
multipart/alternative' can be used to send multiple versions of a
resource: this preserves latency by using additional bandwidth to
send a greater volume of data. On the other hand, HTTP [7] suggests
a negotiation mechanism which preserves bandwidth at the cost of
introducing a round-trip delay (section 12.2, Agent-driven
negotiation).
To set against the negative performance impact of content
negotiation, it is to be hoped that overall network efficiency is to
be improved if it results in the most useful data format being
delivered to its intended recipient, first time, almost every time.
5.7 Confidence levels in negotiated options
In some cases (e.g. when there has been a direct exchange of
information with the remote system) the communicating parties will
have a high degree of confidence in the outcome of a negotiation.
Here, a data exchange can be performed without need for subsequent
confirmation that the options used were acceptable.
In other cases, the options will be a best-guess, and it may be
necessary to make provision for parties to reject the options
actually used in preference for some other set.
This consideration is likely to interact with performance
considerations.
A useful pattern, adopted by TCN [5], is to define a negotiation
procedure which guarantees a correct outcome. This forms the
foundation for a procedure which attempts to use easily-obtained but
less reliable information in an attempt to optimize the negotiation
process but that contains checks to guarantee the final result will
be the same as would have been obtained by the full negotiation
procedure. Such procedures sometimes have to resort to the original
"full cycle" negotiation procedure, but in a majority of cases are
expected to reach their conclusion by an optimized route.
6. Security Considerations
The purposes of this section is to identify and catalogue some
security issues that feature negotiation protocols should consider.
6.1 Privacy
Privacy may be adversely affected by:
o Unintended disclosure of personal information.
o Spoofed requests for negotiation data simply for the purposes of
gathering information, and not as part of a bona fide message
transmission.
6.2 Denial of service attacks
Service denial may be caused by:
o Injection of false negotiation data.
o Excessive requests for negotiation data
6.3 Mailing list interactions
Content negotiation with final recipients is somewhat at odds with
normal practice for maintaining lists for redistribution of Internet
mail.
It may be appropriate for a sender to negotiate data formats with a
list manager, and for a list manager to negotiate with message
recipients. But the common practice of keeping confidential the
identities and addresses of mailing list subscribers suggests that
end-to-end negotiation through a mailing list is not consistent with
good security practice.
6.4 Use of security services
Protocols that employ security services for message transfer should
also apply those services to content negotiation:
o Authenticated requests for negotiation metadata provide a means
for a potential recipient to moderate the distribution of media
capability information.
o Authentication of negotiation metadata provides a means for
potential message senders to avoid using incorrect information
injected by some other party.
o Encryption of negotiation data may help to prevent disclosure of
sensitive capability-related information to snoopers.
o Conducting a negotiation exchange over an authenticated or
encrypted protocol session (e.g. SASL), transport connection or
network path (e.g. TLS, IPSEC) can provide for mutual
authentication of both parties in an exchange of negotiation data.
6.5 Disclosure of security weaknesses
6.5.1 User agent identification
Disclosure of capability information may allow a potential attacker
to deduce what message handling agent is used, and hence may lead to
the exploitation of known security weaknesses in that agent.
6.5.2 Macro viruses
Macro viruses are a widespread problem among applications such as
word processors and spreadsheets. Knowing which applications a
recipient employs (e.g. by file format) may assist in a malicious
attack. However, such viruses can be spread easily without such
knowledge by sending multiple messages, where each message infects a
specific application version.
6.5.3 Personal vulnerability
One application of content negotiation is to enable the delivery of
message content that meets specific requirements of less able people.
Disclosure of this information may make such people potential targets
for attacks that play on their personal vulnerabilities.
6.6 Problems of negotiating security
If feature negotiation is used to decide upon security-related
features to be used, some special problems may be created if the
negotiation procedure can be subverted to prevent the selection of
effective security procedures.
The security considerations section of GSS-API negotiation [8]
discusses the use of integrity protecting mechanisms with security
negotiation.
7. Acknowledgements
Some material in this memo has been derived from earlier memos by
Koen Holtman, Andrew Mutz, Ted Hardie, Larry Masinter, Dan Wing, Neil
Joffe. Matters relating to the importance and relevance of content
negotiation to less-able users were raised by Al Gilman.
This memo has also been informed by the debates of the IETF "conneg"
working group.
8. References
[1] Freed, N. and N. Borenstein, "Multipurpose Internet Mail
Extensions (MIME) Part 1: Format of Internet message bodies",
RFC 2045, November 1996.
[2] Freed, N. and N. Borenstein, "Multipurpose Internet Mail
Extensions (MIME) Part 2: Media Types", RFC 2046, November 1996.
[3] Holtman, K., et al., "The Alternates Header Field", Work in
Progress.
[4] Hardie, T., "Scenarios for the Delivery of Negotiated Content",
Work in Progress.
[5] Holtman, K. and A. Mutz, "Transparent Content Negotiation in
HTTP", RFC 2295, March 1998.
[6] Wing, D., "Indicating Supported Media Features Using Extensions
to DSN and MDN", RFC 2530, March 1999.
[7] Fielding, R., Gettys, J., Mogul, J., Frytyk, H. and T. Berners-
Lee, "Hyptertext Transfer Protocol -- HTTP/1.1", RFC 2068,
January 1997.
[8] Blaize, E. and D. Pinkas, "The Simple and Protected GSS-API
Negotiation Mechanism", RFC 2478, December 1998.
9. Author's Address
Graham Klyne
5th Generation Messaging Ltd. Content Technologies Ltd.
5 Watlington Street 1220 Parkview, Arlington Business Park
Nettlebed Theale
Henley-on-Thames, RG9 5AB Reading, RG7 4SA
United Kingdom United Kingdom.
Phone: +44 1491 641 641 +44 118 930 1300
Fax: +44 1491 641 611 +44 118 930 1301
EMail: GK@ACM.ORG
10. Full Copyright Statement
Copyright (C) The Internet Society (1999). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
ISO Plain Language
On this page
§1Why Protocol-Independent Content Negotiation
Many Internet application protocols need to negotiate content for the resources they work with. MIME media types give a standard way to handle one major axis of variation, but resources also vary in ways that current MIME headers cannot express.
The memo therefore sets out terminology, an abstract framework and goals for protocol-independent content negotiation, and points out technical issues that may need attention. It does not try to specify the negotiation process itself; it indicates the scope and form that such a specification might take. The goals are the desired properties of a negotiation mechanism, written so that no particular shape of framework is assumed.
Discussion of the memo was to take place on the content negotiation and media feature registration mailing list hosted by the Internet Mail Consortium, and the memo says it was informed by the debates of the IETF conneg working group.
When one computer can send a file in several forms, sender and receiver have to agree on which form to use. RFC 2703 gives the vocabulary for that agreement and a general shape for it, so that the idea can work the same way in e-mail, in the web and in directory lookups.
§2The Vocabulary
The memo's second section defines the terms used with a special meaning, many of them copied and adapted from the Transparent Content Negotiation work, and lists them alphabetically.
The core of the vocabulary is the variant: one of several possible representations of a data resource. A data resource may be available in several representations, for example multiple languages, data formats, sizes or resolutions. A variant resource is a data resource for which several representations are available, and a negotiable resource is one where the variant to transmit is selected by negotiating between sender and recipient. The variant list holds descriptions of the variants, and a variant description contains a variant resource identifier plus attributes describing the properties of that variant.
On the capabilities side, a feature is a piece of information about the media handling properties of a component or a resource, and a feature tag is the name that identifies it. A feature set is the full set of features that a sender, a recipient or a data file can handle. A capability is an attribute indicating an ability to generate or process a particular type of content, a characteristic indicates a possible capability or preference, and a preference says which of two possible contents is wanted. Negotiation metadata is the information that sender and receiver exchange in order to work out which variant should be transferred.
Two kinds of message are named. A list message sends the variant list but no variant data, and can be generated when the sender does not want to, or is not allowed to, send a particular variant. A choice message returns a representation of the selected variant, or variants, together with the variant list, and needs the receiver to be told what else was on offer. The memo also distinguishes sender-initiated transmission, which is push, from receiver-initiated transmission, which is pull, giving an HTTP GET operation as an example of the latter.
The words matter here. A variant is one version of a resource, such as the same report in two languages. A feature set is the list of things a computer can handle, and negotiation metadata is the information both sides swap to agree on a version. The memo also names the two obvious moves: a list message that offers the versions without sending one, and a choice message that sends a version and still says what else there was.
- VariantOne of several possible representations of a data resource.
- Feature setThe full set of features a sender, a recipient or a file can handle.
- CapabilityAn attribute saying a party can generate or process a type of content.
- Negotiation metadataThe information exchanged to decide which variant to transfer.
- List messageSends the variant list but no variant data.
- Choice messageSends a variant and also tells the receiver what else was available.
§3The Abstract Framework
The framework covers three things: expressing the capabilities of the sender and of the data resource to be transmitted, expressing the capabilities of a receiver before the transmission, and a protocol by which capabilities are exchanged.
The memo draws the design elements and their dependencies. An abstract negotiation process and an abstract negotiation metadata model sit above a negotiation protocol binding and a negotiation metadata representation, and all of it is used by an application protocol that incorporates content negotiation. Protocol independence comes from keeping the abstract process and the metadata apart from concrete representations and protocol bindings. The capabilities of the transmission protocol itself are deliberately disregarded, on the assumption that they are handled separately by some orthogonal mechanism.
To show where the negotiation sits, the memo gives the life of a data resource: an author produces original document content, a sending system transmits a data file that is a representation of that content, a receiving system renders it into formatted document data, and a user consumes the result. It is the sending system and the receiving system that exchange negotiation metadata to decide the form of the data file. The sender's metadata reflects which variants of the content exist and what formats it can offer them in; the receiver's metadata reflects the needs and preferences of its user and what it is able to process.
The negotiation itself is a series of metadata exchanges that continues until either side settles on a specific data file. If the sender makes the final choice it can send the file directly; otherwise the receiver must communicate its selection and the sender sends what was chosen. Every concrete negotiation is expected to be a subset of that general process: Transparent Content Negotiation, for example, covers a request with no variants, a request that gets a list of variants back so the client can choose, and a request that carries the client's own metadata so the server can choose on its behalf. Fax negotiation is a simpler case where the recipient declares its capabilities and the sender picks a variant to match.
The metadata model then has to cover six things: a way to describe attributes of a data resource, called a media feature; a description of a range of possible media feature combinations, called a feature set; one or more naming schemes for labelling media features and feature sets, backed by a registration process that keeps names unique and encourages a shared vocabulary; a framework of data types giving the range and properties of the values that can be represented; a way to combine media features into feature sets that can express dependencies inside a set, such as 640x480 pixels with 256 colours or 800x600 pixels with 16 colours; and some way of ranking feature sets by the preferences of sender and receiver. A textual representation would give text-based protocols such as HTTP and SMTP a shared vocabulary, and an ASN.1 description would help protocols built on ASN.1, such as X.400, X.500, LDAP and SNMP; separate media feature naming based on ISO object identifiers would be valuable for that. Concrete protocol bindings will be needed, though their details are out of scope, and the memo notes that SASL might provide a useful model.
The framework has three moving parts: what the sender can offer, what the receiver can accept, and the protocol used to swap that information. Above them sits the idea of an abstract negotiation process, which can be bound to any application protocol. The memo shows where the talking happens by drawing the life of a document: an author writes it, a sending system picks a form, a receiving system renders it, and a user reads it. The two systems in the middle are the ones that negotiate.
- ORIGINAL CONTENTThe document as written.
- NEGOTIATION METADATAWhat the receiver can process, exchanged before the file is sent.
- DATA FILEThe negotiated representation of the original content.
- FORMATTED DOCUMENTThe content rendered for the user to consume.
§4Goals for a Negotiation Framework
The goals fall into two groups. The generic goals address negotiation in a protocol-independent way: a common vocabulary for designating features and feature sets; a stable reference for commonly used features; an extensible framework so new features can be adopted quickly; a way to indicate quality or preference; a way to capture dependencies between feature values; a uniform mechanism for exchanging metadata that covers existing features and can grow to meet unanticipated ones; efficient negotiation in both pull and push transfers; a structure that stands independently of any particular message transfer protocol; and the ability to serve the communication needs of less able computer users.
The protocol-specific goals are for whoever writes the binding for a particular protocol. A negotiation should generally end with a form of data that both sides accept. Capabilities sent outside a transfer should carry enough information to be verified and authenticated, and a capability assertion should name the party the capabilities apply to, the party they are sent to, and their date or range of validity, protected against interception and substitution. A request for capabilities should name the requester, the party whose capabilities are asked for and the time of the request. There should be a standardised way to associate features with resource variants, and a way to show provider and recipient preferences. Negotiation should have as little effect as possible on network resources, especially bandwidth and round-trips. Systems should protect the privacy of user profiles and provider inventories, and protocols should allow the accuracy of capability data to be checked. Negotiation must not jeopardise the operation or integrity of a system when the capability data is wrong, whether that happens by accident or attack. Gateways, proxies and caches should be allowed to take part, and the metadata should be regarded as cacheable with explicit cache control provided, so that ad-hoc cache-busting techniques do not creep in. Finally, automatic negotiation should not take away the user's ability to choose a document format from those available.
The goals read like a wish list. There should be one shared vocabulary for features, new features should be easy to add, and the whole thing should work the same whether the transfer is push or pull. It should cost as little bandwidth and round-trips as possible, it must not break a system when the data it is given is wrong, and it must not quietly decide the format for the user.
§5Open Technical Issues
The memo lists questions it does not answer. The ideas for generic content negotiation were conceived in the context of message-oriented transfers, where the content is settled before transmission starts; streamed data, interactive computations and real-time data acquisition are not, and it may be that the metadata still applies but the abstract process is not enough for them.
There is also the question of end-to-end against hop-by-hop negotiation. End-to-end gives the greatest confidence in the outcome. Hop-by-hop may have advantages in a network of occasionally connected systems, but puts additional demands on the intervening message transmission agents, and a hop-by-hop response is not necessarily a definitive indication of the endpoint's capabilities, which implies a possible need for time-to-live and re-verification mechanisms to flush out stale negotiation data. Allowing a third party, such as a directory service using LDAP, to supply negotiation metadata increases the ways errors can be introduced, and steps are needed to keep it reliable; using LDAP to carry the metadata also means defining schema elements specific to content negotiation, for instance an attribute type for a media feature set.
Two further issues are raised. Negotiation may raise billing questions, because it can involve a two-way exchange of data that is not completed within a single connection, so somebody has to pay for the return messages, as in e-mail or fax. On performance, negotiation can help or hurt: the extra data costs bandwidth, and the metadata exchange adds round-trip or third-party query delays before transmission of the message itself starts, but in Internet e-mail the multipart/alternative type buys latency with bandwidth by sending several versions of a resource, while HTTP suggests a mechanism that buys bandwidth at the cost of a round-trip. And on confidence, the parties may be certain of the outcome after a direct exchange, or working from a best guess that the recipient should be able to reject. The memo points to the pattern used by Transparent Content Negotiation, where a procedure guarantees a correct outcome and an optimisation uses easily obtained but less reliable information while keeping checks that guarantee the same final result.
Several questions are left open. The framework was built around messages, where everything is settled before sending starts, so streamed or live data may need something more. Negotiating end to end is the most reliable, but letting each router in between help out saves work and makes the answer less definite. Using a directory service to carry the negotiation details is convenient but adds a new way for things to go wrong.
| Approach | What it gives | What it costs |
|---|---|---|
| End-to-end negotiation | The greatest confidence in the outcome | The two end systems must deal with each other directly |
| Hop-by-hop negotiation | Advantages in a network of occasionally connected systems | Extra demands on the agents in between, and a response is not a definitive statement of the endpoint's capabilities |
§6Security Considerations
The memo catalogues security issues that negotiation protocols should consider. Privacy can be harmed by the unintended disclosure of personal information, and by spoofed requests for negotiation data made only to gather information rather than as part of a real transfer. Service denial can be caused by injecting false negotiation data or by making excessive requests for it.
Content negotiation with final recipients is somewhat at odds with the normal practice of keeping the identities and addresses of mailing list subscribers confidential, so end-to-end negotiation through a mailing list is not consistent with good security practice. Protocols that use security services for message transfer should apply those services to negotiation as well: authenticated requests let a potential recipient moderate the distribution of media capability information, authentication of the metadata stops a sender being misled by data injected by another party, encryption may prevent the disclosure of sensitive capability information to snoopers, and conducting the exchange over an authenticated or encrypted protocol session or network path can give both parties mutual authentication.
Three disclosure risks are named. Capability information may let an attacker deduce which message handling agent is in use and so lead to the exploitation of known weaknesses in that agent. Knowing which applications a recipient employs, for example from the file format, may assist an attack, although macro viruses can be spread without such knowledge by sending multiple messages, each of which infects a specific application version. And because one use of content negotiation is to deliver content meeting the requirements of less able people, disclosing that information may make those people targets for attacks that play on their personal vulnerabilities. Finally, if feature negotiation is used to decide which security-related features to use, a subverted negotiation procedure could prevent effective security procedures from being selected; the memo points to the security considerations of GSS-API negotiation for the use of integrity protecting mechanisms with security negotiation.
Negotiation can leak information about the person on the other end, so requests for it should be authenticated and the answers protected. It can also be abused: false data or too many requests can tie up a server. Telling a network what software someone runs may help an attacker, and so may revealing that someone needs an accessible format. And if negotiation is allowed to choose the security features, a manipulated negotiation could pick weak ones.