Request for Comments: 3930
The Protocol versus Document Points of View in Computer Protocols
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 © The Internet Society (2004).
This document contrasts two points of view: the "document" point of view, where digital objects of interest are like pieces of paper written and viewed by people, and the "protocol" point of view where objects of interest are composite dynamic network messages. Although each point of view has a place, adherence to a document point of view can be damaging to protocol design. By understanding both points of view, conflicts between them may be clarified and reduced.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2 2. Points of View . . . . . . . . . . . . . . . . . . . . . . . . 2 2.1. The Basic Points of View . . . . . . . . . . . . . . . . 3 2.2. Questions of Meaning . . . . . . . . . . . . . . . . . . 3 2.2.1. Core Meaning . . . . . . . . . . . . . . . . . . 3 2.2.2. Adjunct Meaning. . . . . . . . . . . . . . . . . 4 2.3. Processing Models. . . . . . . . . . . . . . . . . . . . 5 2.3.1. Amount of Processing . . . . . . . . . . . . . . 5 2.3.2. Granularity of Processing. . . . . . . . . . . . 5 2.3.3. Extensibility of Processing. . . . . . . . . . . 6 2.4. Security and Canonicalization. . . . . . . . . . . . . . 6 2.4.1. Canonicalization . . . . . . . . . . . . . . . . 6 2.4.2. Digital Authentication . . . . . . . . . . . . . 8 2.4.3. Canonicalization and Digital Authentication. . . 8 2.4.4. Encryption . . . . . . . . . . . . . . . . . . . 9 2.5. Unique Internal Labels . . . . . . . . . . . . . . . . . 10 3. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4. Resolution of the Points of View . . . . . . . . . . . . . . . 11 5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 12 6. Security Considerations. . . . . . . . . . . . . . . . . . . . 12 Informative References . . . . . . . . . . . . . . . . . . . . . . 12 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . 14 Full Copyright Statement . . . . . . . . . . . . . . . . . . . . . 15
This document contrasts: the "document" point of view, where digital objects of interest are thought of as pieces of paper written and viewed by people, and the "protocol" point of view, where objects of interest are composite dynamic network messages. Those accustomed to one point of view frequently have great difficulty appreciating the other: Even after they understand it, they almost always start by considering things from their accustomed point of view, assume that most of the universe of interest is best viewed from their perspective, and commonly slip back into thinking about things entirely from that point of view. Although each point of view has a place, adherence to a document point of view can be damaging to protocol design. By understanding both points of view, conflicts between them may be clarified and reduced.
Much of the IETF's traditional work has concerned low level binary protocol constructs. These are almost always viewed from the protocol point of view. But as higher level application constructs and syntaxes are involved in the IETF and other standards processes, difficulties can arise due to participants who have the document point of view. These two different points of view defined and explored in section 2 below.
Section 3 gives some examples. Section 4 tries to synthesize the views and give general design advice in areas that can reasonably be viewed either way.
2. Points of View
The following subsections contrast the document and protocol points of view. Each viewpoint is EXAGGERATED for effect.
The document point of view is indicated in paragraphs headed "DOCUM", and the protocol point of view is indicated in paragraphs headed "PROTO".
2.1. The Basic Points of View
DOCUM: What is important are complete (digital) documents, analogous
to pieces of paper, viewed by people. A major concern is to be able to present such documents as directly as possible to a court or other third party. Because what is presented to the person is all that is important, anything that can effect this, such as a "style sheet" [CSS], MUST be considered part of the document. Sometimes it is forgotten that the "document" originates in a computer, may travel over, be processed in, and be stored in computer systems, and is viewed on a computer, and that such operations may involve transcoding, enveloping, or data reconstruction.
PROTO: What is important are bits on the wire generated and consumed
by well-defined computer protocol processes. No person ever sees the full messages as such; it is only viewed as a whole by geeks when debugging, and even then they only see some translated visible form. If one actually ever has to demonstrate something about such a message in a court or to a third party, there isn't any way to avoid having computer experts interpret it. Sometimes it is forgotten that pieces of such messages may end up being included in or influencing data displayed to a person.
2.2. Questions of Meaning
The document and protocol points of view have radically different concepts of the "meaning" of data. The document oriented tend to consider "meaning" to a human reader extremely important, but this is something the protocol oriented rarely think about at all.
This difference in point of view extends beyond the core meaning to the meaning of addenda to data. Both core and addenda meaning are discussed below.
2.2.1. Core Meaning
DOCUM: The "meaning" of a document is a deep and interesting human
question related to volition. It is probably necessary for the document to include or reference human language policy and/or warranty/disclaimer information. At an absolute minimum, some sort of semantic labelling is required. The assumed situation is always a person interpreting the whole "document" without other context. Thus it is reasonable to consult attorneys during message design, to require that human-readable statements be "within the four corners" of the document, etc.
PROTO: The "meaning" of a protocol message should be clear and
unambiguous from the protocol specification. It is frequently defined in terms of the state machines of the sender and recipient processes and may have only the most remote connection with human volition. Such processes have additional context, and the message is usually only meaningful with that additional context. Adding any human-readable text that is not functionally required is silly. Consulting attorneys during design is a bad idea that complicates the protocol and could tie a design effort in knots.
2.2.2. Adjunct Meaning
Adjunct items can be added or are logical addenda to a message.
DOCUM: From a document point of view, at the top level is a person
looking at a document. So adjunct items such as digital signatures, person's names, dates, etc., must be carefully labeled as to meaning. Thus a digital signature needs to include, in more or less human-readable form, what that signature means (is the signer a witness, author, guarantor, authorizer, or what?). Similarly, a person's name needs to be accompanied by that person's role, such as editor, author, subject, or contributor. As another example, a date needs to be accompanied by the significance of the date, such as date of creation, modification, distribution, or some other event. Given the unrestrained scope of what can be documented, there is a risk of trying to enumerate and standardize all possible "semantic tags" for each kind of adjunct data during in the design process. This can be a difficult, complex, and essentially infinite task (i.e., a rat hole).
PROTO: From a protocol point of view, the semantics of the message
and every adjunct in it are defined in the protocol specification. Thus, if there is a slot for a digital signature, person's name, a date, or whatever, the party who is to enter that data, the party or parties who are to read it, and its meaning are all pre- defined. Even if there are several possible meanings, the specific meaning that applies can be specified by a separate enumerated type field. There is no reason for such a field to be directly human readable. Only the "meanings" directly relevant to the particular protocol need be considered. Another way to look at this is that the "meaning" of each adjunct, instead of being pushed into and coupled with the adjunct itself, as the document point of view encourages, is commonly promoted to the level of the protocol specification, resulting in simpler adjuncts.
2.3. Processing Models
The document oriented and protocol oriented have very different views on what is likely to happen to an object.
2.3.1. Amount of Processing
DOCUM: The model is of a quasi-static object like a piece of paper.
About all one does to pieces of paper is transfer them as a whole, from one storage area to another, or add signatures, date stamps, or similar adjuncts. (Possibly one might want an extract from a document or to combine multiple documents into a summary, but this isn't the common case.)
PROTO: The standard model of a protocol message is as an ephemeral
composite, multi-level object created by a source process and consumed by a destination process. Such a message is constructed from information contained in previously received messages, locally stored information, local calculations, etc. Quite complex processing is normal.
2.3.2. Granularity of Processing
DOCUM: The document view is generally of uniform processing or
evaluation of the object being specified. There may be an allowance for attachments or addenda, but, if so, they would probably be simple, one level, self documenting attachments or addenda. (Separate processing of an attachment or addenda is possible but not usual.)
PROTO: Processing is complex and almost always affects different
pieces of the message differently. Some pieces may be intended for use only by the destination process and may be extensively processed there. Others may be present so that the destination process can, at some point, do minimal processing and forward them in other messages to yet more processes. The object's structure can be quite rich and have multilevel or recursive aspects. Because messages are processed in a local context, elements of the message may include items like a signature that covers multiple data elements, some of which are in the message, some received in previous messages, and some locally calculated.
2.3.3. Extensibility of Processing
DOCUM: The document oriented don't usually think of extensibility as
a major problem. They assume that their design, perhaps with some simple version scheme, will meet all requirements. Or, coming from an SGML/DTD world of closed systems, they may assume that knowledge of new versions or extensions can be easily and synchronously distributed to all participating sites.
PROTO: Those who are protocol oriented assume that protocols will
always need to be extended and that it will not be possible to update all implementations as such extensions are deployed and/or retired. This is a difficult problem but those from the protocol point of view try to provide the tools needed. For example, they specify carefully defined versioning and extension/feature labelling, including the ability to negotiate versions and features where possible and at least a specification of how parties running different levels should interact, providing length/delimiting information for all data so that it can be skipped if not understood, and providing destination labelling so that a process can tell that it should ignore data except for passing it through to a later player.
2.4. Security and Canonicalization
Security is a subtle area. Sometime problems can be solved in a way that is effective across many applications. Those solutions are typically incorporated into standard security syntaxes such as those for ASN.1 [RFC3852] and XML [RFC3275, XMLENC]. But there are almost always application specific questions, particularly the question of exactly what information needs to be authenticated or encrypted.
Questions of exactly what needs to be secured and how to do so robustly are deeply entwined with canonicalization. They are also somewhat different for authentication and encryption, as discussed below.
Canonicalization is the transformation of the "significant" information in a message into a "standard" form, discarding "insignificant" information, for example, encoding into a standard character set or changing line endings into a standard encoding and discarding the information about the original character set or line ending encodings. Obviously, what is "significant" and what is "insignificant" varies with the application or protocol and can be tricky to determine. However, it is common that for each particular syntax, such as ASCII [ASCII], ASN.1 [ASN.1], or XML [XML], a standard canonicalization (or canonicalizations) is specified or developed through practice. This leads to the design of applications that assume one of such standard canonicalizations, thus reducing the need for per-application canonicalization. (See also [RFC3076, RFC3741].)
DOCUM: From the document point of view, canonicalization is suspect
if not outright evil. After all, if you have a piece of paper with writing on it, any modification to "standardize" its format can be an unauthorized change in the original message as created by the "author", who is always visualized as a person. Digital signatures are like authenticating signatures or seals or time stamps on the bottom of the "piece of paper". They do not justify and should not depend on changes in the message appearing above them. Similarly, encryption is just putting the "piece of paper" in a vault that only certain people can open and does not justify any standardization or canonicalization of the message.
PROTO: From the protocol point of view, canonicalization is simply a
necessity. It is just a question of exactly what canonicalization or canonicalizations to apply to a pattern of bits that are calculated, processed, stored, communicated, and finally parsed and acted on. Most of these bits have never been seen and never will be seen by a person. In fact, many of the parts of the message will be artifacts of encoding, protocol structure, and computer representation rather than anything intended for a person to see. Perhaps in theory, the "original", idiosyncratic form of any digitally signed part could be conveyed unchanged through the computer process, storage, and communications channels that implement the protocol and could be usefully signed in that form. But in practical systems of any complexity, this is unreasonably difficult, at least for most parts of messages. And if it were possible, it would be virtually useless, because to authenticate messages you would still have to determine their equivalence with the preserved original form. Thus, signed data must be canonicalized as part of signing and verification to compensate for insignificant changes made in processing, storage, and communication. Even if, miraculously, an initial system design avoids all cases of signed message reconstruction based on processed data or re-encoding based on character set or line ending or capitalization or numeric representation or time zones or whatever, later protocol revisions and extensions are certain to require such reconstruction and/or re-encoding eventually. If such "insignificant" changes are not ameliorated by canonicalization, signatures won't work, as discussed in more detail in 2.4.3 below.
2.4.2. Digital Authentication
DOCUM: The document-oriented view on authentication tends to be a
"digital signature" and "forms" point of view. (The "forms" point of view is a subset of the document point of view that believes that a principal activity is presenting forms to human beings so that they can fill out and sign portions of those forms [XForms]). Since the worry is always about human third parties and viewing the document in isolation, those who are document oriented always want "digital signature" (asymmetric key) authentication, with its characteristics of "non-repudiability", etc. As a result, they reject secret key based message authentication codes, which provide the verifier with the capability of forging an authentication code, as useless. (See any standard reference on the subject for the usual meaning of these terms.) From their point of view, you have a piece of paper or form which a person signs. Sometimes a signature covers only part of a form, but that's usually because a signature can only cover data that is already there. And normally at least one signature covers the "whole" document/form. Thus the document oriented want to be able to insert digital signatures into documents without changing the document type and even "inside" the data being signed, which requires a mechanism to skip the signature so that it does not try to sign itself.
PROTO: From a protocol point of view, the right kind of
authentication to use, whether "digital signature" or symmetric keyed authentication code (or biometric or whatever), is just another engineering decision affected by questions of efficiency, desired security model, etc. Furthermore, the concept of signing a "whole" message seems very peculiar (unless it is a copy being saved for archival purposes, in which case you might be signing a whole archive at once anyway). Typical messages are made up of various pieces with various destinations, sources, and security requirements. Furthermore, there are common fields that it is rarely useful to sign because they change as the message is communicated and processed. Examples include hop counts, routing history, and local forwarding tags.
2.4.3. Canonicalization and Digital Authentication
For authenticating protocol system messages of practical complexity, you are faced with the choice of doing
(1) "too little canonicalization" and having brittle authentication,
useless due to verification failures caused by surface representation changes without significance,
(2) the sometimes difficult and tricky work of selecting or designing
an appropriate canonicalization or canonicalizations to be used as part of authentication generation and verification, producing robust and useful authentication, or
(3) "too much canonicalization" and having insecure authentication,
useless because it still verifies even when significant changes are made in the signed data.
The only useful option above is number 2.
In terms of processing, transmission, and storage, encryption turns out to be much easier to get working than signatures. Why? Because the output of encryption is essentially random bits. It is clear from the beginning that those bits need to be transferred to the destination in some absolutely clean way that does not change even one bit. Because the encrypted bits are meaningless to a human being, there is no temptation among the document oriented to try to make them more "readable". So appropriate techniques of encoding at the source, such as Base64 [RFC2045], and decoding at the destination, are always incorporated to protect or "armor" the encrypted data.
Although the application of canonicalization is more obvious with digital signatures, it may also apply to encryption, particularly encryption of parts of a message. Sometimes elements of the environment where the plain text data is found may affect its interpretation. For example, interpretation can be affected by the character encoding or bindings of dummy symbols. When the data is decrypted, it may be into an environment with a different character encoding or dummy symbol bindings. With a plain text message part, it is usually clear which of these environmental elements need to be incorporated in or conveyed with the message. But an encrypted message part is opaque. Thus some canonical representation that incorporates such environmental factors may be needed.
DOCUM: Encryption of the entire document is usually what is
considered. Because signatures are always thought of as human assent, people with a document point of view tend to vehemently assert that encrypted data should never be signed unless the plain text of it is known.
PROTO: Messages are complex composite multi-level structures, some
pieces of which are forwarded multiple hops. Thus the design question is what fields should be encrypted by what techniques to what destination or destinations and with what canonicalization.
It sometimes makes perfect sense to sign encrypted data you don't understand; for example, the signature could just be for integrity protection or for use as a time stamp, as specified in the protocol.
2.5. Unique Internal Labels
It is desirable to be able to reference parts of structured messages or objects by some sort of "label" or "id" or "tag". The idea is that this forms a fixed "anchor" that can be used "globally", at least within an application domain, to reference the tagged part.
DOCUM: From the document point of view, it seems logical just to
provide for a text tag. Users or applications could easily come up with short readable tags. These would probably be meaningful to a person if humanly generated (e.g., "Susan") and at least fairly short and systematic if automatically generated (e.g., "A123"). The ID attribute type in XML [XML] appears to have been thought of this way, although it can be used in other ways.
PROTO: From a protocol point of view, unique internal labels look
very different than they do from a document point of view. Since this point of view assumes that pieces of different protocol messages will later be combined in a variety of ways, previously unique labels can conflict. There are really only three possibilities if such tags are needed, as follows:
(1) Have a system for dynamically rewriting such tags to maintain uniqueness. This is usually a disaster, as it (a) invalidates any stored copies of the tags that are not rewritten, and it is usually impossible to be sure there aren't more copies lurking somewhere you failed to update, and (b) invalidates digital signatures that cover a changed tag. (2) Use some form of hierarchical qualified tags. Thus the total tag can remain unique even if a part is moved, because its qualification changes. This avoids the digital signature problems described above. But it destroys the concept of a globally-unique anchor embedded in and moving with the data. And stored tags may still be invalidated by data moves. Nevertheless, within the scope of a particular carefully designed protocol, such as IOTP [RFC2801], this can work. (3) Construct a lengthy globally-unique tag string. This can be done successfully by using a good enough random number generator and big enough random tags (perhaps about 24 characters) sequentially, as in the way email messages IDs are created [RFC2822].
Thus, from a protocol point of view, such tags are difficult but if they are needed, choice 3 works best.
IETF protocols are replete with examples of the protocol viewpoint such as TCP [RFC793], IPSEC [RFC2411], SMTP [RFC2821], and IOTP [RFC2801, RFC2802].
The eXtensible Markup Language [XML] is an example of something that can easily be viewed both ways and where the best results frequently require attention to both the document and the protocol points of view.
Computerized court documents, human-to-human email, and the X.509v3 Certificate [X509v3], particularly the X509v3 policy portion, are examples primarily designed from the document point of view.
4. Resolution of the Points of View
There is some merit to each point of view. Certainly the document point of view has some intuitive simplicity and appeal and is OK for applications where it meets needs.
The protocol point of view can come close to encompassing the document point of view as a limiting case. In particular, it does so under the following circumstances:
- As the complexity of messages declines to a single payload (perhaps with a few attachments).
- As the mutability of the payload declines to some standard format that needs little or no canonicalization.
- As the number of parties and amount of processing declines as messages are transferred.
- As the portion of the message intended for more or less direct human consumption increases.
Under the above circumstances, the protocol point of view would be narrowed to something quite close to the document point of view. Even when the document point of view is questionable, the addition of a few options to a protocol will usually mollify the perceived needs of those looking at things from that point of view. For example, adding optional non-canonicalization or an optional policy statement, or inclusion of semantic labels, or the like.
On the other hand, the document point of view is hard to stretch to encompass the protocol case. From a strict piece of paper perspective, canonicalization is wrong; inclusion of human language policy text within every significant object and a semantic tag with every adjunct should be mandatory; and so on. Objects designed in this way are rarely suitable for protocol use, as they tend to be improperly structured to accommodate hierarchy and complexity, inefficient (due to unnecessary text and self-documenting inclusions), and insecure (due to brittle signatures).
Thus, to produce usable protocols, it is best to start with the protocol point of view and add document point of view items as necessary to achieve consensus.
I hope that this document will help explain to those of either point of view where those with the other view are coming from. It is my hope that this will decrease conflict, shed some light -- in particular on the difficulties of security design -- and lead to better protocol designs.
6. Security Considerations
This document considers the security implications of the Document and Protocol points of view, as defined in Sections 2.1 and 2.2, and warns of the security defects in the Document view. Most of these security considerations appear in Section 2.4 but they are also touched on elsewhere in Section 2 which should be read in its entirety.
[ASCII] "USA Standard Code for Information Interchange", X3.4, American National Standards Institute: New York, 1968. [ASN.1] ITU-T Recommendation X.680 (1997) | ISO/IEC 8824-1:1998, "Information Technology - Abstract Syntax Notation One (ASN.1): Specification of Basic Notation". ITU-T Recommendation X.690 (1997) | ISO/IEC 8825-1:1998, "Information Technology - ASN.1 Encoding Rules: Specification of Basic Encoding Rules (BER), Canonical Encoding Rules (CER) and Distinguished Encoding Rules (DER)". <http://www.itu.int/ITU- T/studygroups/com17/languages/index.html>. [CSS] "Cascading Style Sheets, level 2 revision 1 CSS 2.1 Specification", B. Bos, T. Gelik, I. Hickson, H. Lie, W3C Candidate Recommendation, 25 February 2004. <http://www.w3.org/TR/CSS21> [RFC793] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981. [RFC2045] Freed, N. and N. Borenstein, "Multipurpose Internet Mail Extensions (MIME) Part One: Format of Internet Message Bodies", RFC 2045, November 1996. [RFC2411] Thayer, R., Doraswamy, N., and R. Glenn, "IP Security Document Roadmap", RFC 2411, November 1998. [RFC3852] Housley, R., "Cryptographic Message Syntax (CMS)", RFC 3852, July 2004. [RFC2801] Burdett, D., "Internet Open Trading Protocol - IOTP Version 1.0", RFC 2801, April 2000. [RFC2802] Davidson, K. and Y. Kawatsura, "Digital Signatures for the v1.0 Internet Open Trading Protocol (IOTP)", RFC 2802, April 2000. [RFC2821] Klensin, J., "Simple Mail Transfer Protocol", RFC 2821, April 2001. [RFC2822] Resnick, P., "Internet Message Format", RFC 2822, April 2001. [RFC3076] Boyer, J., "Canonical XML Version 1.0", RFC 3076, March 2001. [RFC3275] Eastlake 3rd, D., Reagle, J., and D. Solo, "(Extensible Markup Language) XML-Signature Syntax and Processing", RFC 3275, March 2002. [RFC3741] Berger, L., "Generalized Multi-Protocol Label Switching (GMPLS) Signaling Functional Description", RFC 3471, January 2003. [X509v3] "ITU-T Recommendation X.509 version 3 (1997), Information Technology - Open Systems Interconnection - The Directory Authentication Framework", ISO/IEC 9594- 8:1997. [XForms] "XForms 1.0", M. Dubinko, L. Klotz, R. Merrick, T. Raman, W3C Recommendation 14 October 2003. <http://www.w3.org/TR/xforms/> [XML] "Extensible Markup Language (XML) 1.0 Recommendation (2nd Edition)". T. Bray, J. Paoli, C. M. Sperberg- McQueen, E. Maler, October 2000. <http://www.w3.org/TR/2000/REC-xml-20001006> [XMLENC] "XML Encryption Syntax and Processing", J. Reagle, D. Eastlake, December 2002. <http://www.w3.org/TR/2001/RED-xmlenc-core-20021210/>
Donald E. Eastlake 3rd
155 Beaver Street
Milford, MA 01757 USA
Phone: +1 508-786-7554 (w) +1 508-634-2066 (h) Fax: +1 508-786-7501 (w) EMail: Donald.Eastlake@motorola.com
Full Copyright Statement
Copyright © The Internet Society (2004).
This document is subject to the rights, licenses and restrictions contained in BCP 78, and at www.rfc-editor.org, and except as set forth therein, the authors retain all their rights.
This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM 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.
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the ISOC's procedures with respect to rights in ISOC Documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf- email@example.com.
Funding for the RFC Editor function is currently provided by the Internet Society.