RFC 3978 Update

This document discuss two problems with BCP 78. Familiarity with that document is assumed. Several terms defined in that document are used below. In particular, this document use the term "Contributions" to denote material to which rights are granted. After describing the problems, we conclude with an update to BCP 78 to address the problems.

The license in RFC 2026 gave third parties the right to produce unrestricted derivative works, under some conditions. That right has widely been used to adapt RFCs into online help, reference manuals, source code comments and source code. BCP 78 does not grant any rights to permit this usage. Examples of common usage that is not permitted through BCP 78 include: Extracting parts of RFCs into source code comments, where the source code is licensed under a license that permits modifications. Line breaks and other simple re-formatting is common. Further re-formatting or expansion, to help explain the source code is not uncommon. Material needed in source code may include large tables, see StringPrep . It is customary to modify these tables and schemas, without changing the semantic nor affecting protocol interoperability, when implementing these IETF standards. The modifications are because the table as-is is not valid program code. Examples of deployed programs using tables derived from RFC 3454 include GNU Libc/Libidn, VeriSign's XCODE, and IBM's ICU. ASN.1 schemas from RFCs are often incorporated into products. For example, see the Kerberos 5 ASN.1 schema. These schemas are frequently modified to be usable within particular implementations. Examples of deployed programs using modified ASN.1 schemas for Kerberos 5 include GNU Shishi, MIT Kerberos and Heimdal. Some RFCs include source code or header files that is meant to be included in software packages, and in practice they are frequently modified after inclusion in the software package. For example, GSS-API C bindings, GSS-API Java bindings, SHA-1 hash algorithm, SCTP checksum, Punycode, getaddrinfo DNS resolver API, multicast socket API. Examples of deployed projects here include GSS-API libraries, libc libraries, and cryptographic packages. Material used in manuals or online help ("man" pages) may include protocol overviews (such as in SASL) or API function documentation (such as in GSS-API). For example, several vendors, including commercial companies like Sun, ship a man page for getaddrinfo that is derived from RFC 2133. Educational use. One example that has been presented is to use RFCs in teaching, and make them part of the course material. Further, the RFCs may be modified as part of excercises, e.g., asking students to implement a modified protocol as a lab excercise. There is a desire to include that modified RFC in the course material, which can be distributed freely over the Internet. Development of protocol extensions. When IETF protocols evolve, that is sometimes done by creating a modified version of an existing implementation to experiment with how well the idea works. For example, the NSEC3 work done in the DNSEXT WG break compliance with DNSSEC. Yet, patches are provided for several DNS implementations that include portions of the RFCs in the source code and documentation. That means that it would be difficult to license a RFC under a "field of use" permission that only allow use for implementation a RFC compliant product. Experiments for protocol extensions may thus be illegal. This list is far from conclusive, it is meant to illustrate some of the problems the license in BCP 78 causes. We argue that the usage explained in these examples, and others, should be permitted, and that BCP 78 should be updated to resolve this matter. Adding specific text for each kind of usage is problematic, because we do not have a complete list of "good" usages at this point. We believe the license text should be general enough to enable many different kind of usages. We believe that the license text chosen to resolve this matter should be compatible with licenses typically used for implementions of IETF standards. These range from proprietary closed source licenses, such as Microsoft's End User License Agreement (EULA) or Apple's Mac OS X Software License, over permissive licenses such as the MIT license, to copy-left licenses such as the GNU General Public License . Further, we also argue that the license text should be aligned with the policy used in popular distribution mechanisms of IETF implementations, such as Debian and their Free Software Guidelines . The revised section in BCP 78 below is believed to resolve this problem.

Traditionally, third parties have been permitted to copy and distribute Contributions. These rights were granted by the license text in RFC 2026. Reading BCP 78 reveal that it does not grant any rights to third parties that permit this usage. To explain how this is so, we will review the rights granted by BCP 78, and explain where it is failing, and reference discussion of its interpretation. All rights granted to Contributions, in section 3.3 of BCP 78, begin with (emphasis mine):

a. To the extent that a Contribution or any portion thereof is protected by copyright and other rights of authorship, the Contributor, and each named co-Contributor, and the organization he or she represents or is sponsored by (if any) grant a perpetual, irrevocable, non-exclusive, royalty- free, world-wide right and license to the ISOC and the IETF ------------------------ under all intellectual property rights in the Contribution: That means that all the rights given in that section are given to the IETF and the ISOC, but not to third parties. Similar wording occur in section 4.2 on rights granted for RFC Editor contributions. The document does not grant any rights elsewhere. Scott Bradner claims that section 7.1 gives you these rights. However, the entire section 7 is titled:

Exposition of why these procedures are the way they are We believe section 7.1 can thus not be reasonably said to be included in the actual license statement. Rather, it is only part of the considerations that went into the process that ended up with the current license statement. Another argument that were put forward was that the note "distribution of this memo is unlimited" in RFCs give third parties the necessary rights. However, there is no requirement for that text in RFCs, and it also appear unclear why the RFC Editor still adds this text . This problem was acknowledged through changes between version 00 and version 01 of ; changes that, at least partially, address this concern. The revised section below addresses this concern.

A fear has been expressed that if modification of IETF Contributions is permitted, it may lead to the distribution of fake works. That is, a work that is based on a IETF published document but modified in a subtle way that cannot easily be identified. The work would be easy to confuse with the original, causing confusion for the community. David Black proposed to require that modified works would be forced to remove the boilerplate that is used at the beginning of RFCs. The theory is that the fake is not credible if the endorsement by the IETF and the boilerplate is absent. The license text below incorporate this idea.

This memo adds the following as paragraph 3.3(c) of RFC 3978:

c. The Contributor grants third parties the irrevocable right to copy, use and distribute the Contribution, with or without modification, in any medium, without royalty, provided that, unless separate permission is granted, redistributed modified works: (a) do not contain misleading author, version, name of work, or endorsement information, and (b) do not claim endorsement of the modified work by the Contributor, or any organization the Contributor belongs to, the Internet Engineering Task Force (IETF), Internet Research Task Force (IRTF), Internet Engineering Steering Group (IESG), Internet Architecture Board (IAB), Internet Assigned Numbers Authority (IANA), Internet Society (ISOC), Request For Comments (RFC) Editor, or any combination or variation of such terms (including without limitation the IETF "4 diamonds" logo), or any terms that are confusingly similar thereto, and (c) remove any claims of status as an Internet Standard, including without limitation removing the RFC boilerplate. The IETF suggests that any citation or excerpt of unmodified text reference the RFC or other document from which the text is derived.

To further protect against works claiming to be the original work, it has been suggested that the license should require that people who modify the contents should be forced to add a "warning label". There are two ways to achieve this. The first is to explicitly suggest the text that is to be added. The second is to implicitly require that text is added, by stating that the derivative works must not be confusable with the original. The first approach of using a particular warning label may lead to problems when an implementation support many RFCs. If the particular warning label include the RFC number, there may be a potentially huge list of warning labels that only differ in the RFC number. Even if the specific RFC number is not part of the warning label, the text of the warning label may seem inappropriate and confusing in some modified uses, making the reader question what the warning label is adressing. We believe the second approach, of requiring that implicit warning labels protect against the same abuse but leads to more readable end products. A license that require warning labels to be added may be incompatible with certain free software licenses, and depending on the language used, it may even make the license non-free. We believe further review by the free software community is required if this solution is to be considered further. As a starting point for further research, we propose to add the following implicit requirement for warning labels to the license below:

(d) clearly label a derivative work as such, to avoid similarity between the original work and the derivative work.

It has been suggested that the license should be separated to cover code and text separately. The intention is supposedly that code can use a free and GPL/BSD-compatible license, whereas the text portion will not. We argue that this solution is sub-optimal, and in particular it would prevent scenarios including the following: Excerpts from RFCs included in source code comments. This is a common example, many free IETF implementations embed excerpts from the RFC text in source code comments. Teaching material derived from RFC texts. University teacher may want to modify RFCs to illustrate a point and as an implementation exercises, and include the resulting work in freely available online course material. Documentation for IETF implementations. Both FreeBSD and Sun has shipped a getaddrinfo man page that is derived from the RFC. I believe being able to inherit that text from the RFC helped getting getaddrinfo implemented and utilized. If implementations are not permitted to do this, adoption of IETF protocols will likely be slower. If we believe the IETF is producing standards that, if adopted, would benefit the IETF, I believe we should make sure they can be implemented as fast and as widely as possible. For these reasons, we argue that the same license should be used for both code and text.

The reader should be aware that another BCP 78 Update proposal has been published . The change from version 00 and version 01 of that document was to discuss our second problem, i.e., that third parties have no rights to distribute copies of RFCs. However, it does not discuss nor address our first problem.

The author acknowledges contributions from: David Black, Todd Glassey, Ted Hardie, M. Warner Losh, Francesco Poli, Chuck Powers, MJ Ray, Branden Robinson, Joe Smith, Florian Weimer. Special acknowledgment is given to Joost van Baal, Markus Demleitner, Nathanael Nerode, Nikos Mavrogiannopoulos, and Wytze van der Raay of Stichting NLNet. Contributions of many members of the IPR mailing list are gratefully acknowledged.