Internet-Draft cbor-file-magic April 2021
Richardson Expires 23 October 2021 [Page]
CBOR Working Group
Intended Status:
Best Current Practice
M. Richardson
Sandelman Software Works

On storing CBOR encoded items on stable storage


This document proposes an on-disk format for CBOR objects that is friendly to common on-disk recognition systems like the Unix file(1) command.

This document is being discussed at:

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on 23 October 2021.

Table of Contents

1. Introduction

Since very early in computing, operating systems have sought ways to mark which files could be processed by which programs.

For instance, the Unix file(1) command, which has existed since 1973 [file], has been able to identify many file formats for decades. Many systems (Linux, MacOS, Windows) will select the correct application based upon the file contents, if the system can not determine it by other means: for instance, the classic MacOS maintained a resource fork that includes media type ("MIME type") information and therefore ideally never needs to know what anything about the file. Other systems do this by file extensions.

While having a media type associated with the file is a better solution in general, when files become disconnected from their type information, such as when attempting to do forensics on a damaged system, then being able to identify a file type can become very important.

It is noted that in the media type registration, that a magic number is asked for, if available, as is a file extension.

A challenge for the file(1) program is often that it can be confused by the encoding vs the content. For instance, an Android "apk" used to transfer and store an application may be identified as a ZIP file. Both OpenOffice or MSOffice files are ZIP files of XML files. (Unless OpenOffice files are flat (fodp) files, in which case they may appear to be generic XML files.)

As CBOR becomes a more and more common encoding for a wide variety of artifacts, identifying them as just "CBOR" is probably not sufficient. This document provides a way to encode a magic number into the beginning of a CBOR format file. Two options are presented: typically a CBOR Protocol author will specify one.

A CBOR Protocol is a specification which uses CBOR as its encoding. Examples of CBOR Protocols currently under development include CoSWID [I-D.ietf-sacm-coswid], and EAT [I-D.ietf-rats-eat]. COSE itself [RFC8152] is considered infrastructure, however the encoding of public keys in CBOR as described in [I-D.mattsson-cose-cbor-cert-compress] would be an identified CBOR Protocol.

A major inspiration for this document is observing the mess in ASN.1 based systems where most files are PEM encoded, identified by the extension "pem", confusing public keys, private keys, certificate requests and SIME content.

These proposals are invasive to how CBOR protocols are written to disk, but in both cases, the proposed envelope does not require that the tag be transfered on the wire.

In addition to the on-disk identification aspects, there are some protocols which may benefit from having such a magic number on the wire if they presently using a different (legacy) encoding scheme. The presence of the identifiable magic sequence signals that CBOR is being used or a legacy scheme.

2. Requirements for a Magic Number

A magic number is ideally a unique fingerprint, present in the first 4 or 8 bytes of the file, which does not change when the contents change, and does not depend upon the length of the file.

Less ideal solutions have a pattern that needs to be matched, but in which some bytes need to be ignored. While the Unix file(1) command can be told to ignore bytes, this can lead to ambiguities.

3. Protocol

There are two variations of this practice. Both use CBOR Tags in a way that results in a deterministic first 8 to 12 bytes.

3.1. The CBOR Protocol Specific Tag

CBOR Protocol designers should obtain a tag for each major type of object that they might store on disk. As there are more than 4 million available 4-byte tags, there should be little issue in allocating a few to each available CBOR Protocol.

The policy is First Come First Served, so all that is required is an email to IANA, having filled in the small template provided in section 9.2 of [RFC8949].

This tag should be allocated by the author of the CBOR Protocol, and to be in the four-byte range, it should be at least 0x01000000 (decimal 16777216) in value.

The use of a sequence of four US-ASCII codes which are mnemonic to the protocol is encouraged, but not required.

3.2. CBOR Tag Wrapped

This proposal starts with the Self-described CBOR tag, 55799, as described in [RFC8949] section 3.4.6.

A second CBOR Tag is then allocated to describe the specific Protocol involved, as described above.

This proposal wraps the CBOR value as tags usually do. Applications that need to send the CBOR value across a constrained link may wish to remove the two tags if the use is implicitly understood. This is a decision of the CBOR Protocol specification.

3.3. CBOR Tag Sequence

This proposal makes use of CBOR Sequences as described in [RFC8742].

This proposal consists of two tags and a constant string for a total of 12 bytes.

  1. The file shall start with the Self-described CBOR Sequence tag, 55800.
  2. The file shall continue with a CBOR tag, from the First Come First Served space, which uniquely identifies the CBOR Protocol. The use of a four-byte tag is encouraged.
  3. The three byte CBOR byte string containing 0x42_4F_52. When encoded it shows up as "CBOR"

The first part identifies the file as being CBOR, and does so with all the desirable properties explained in [RFC8949] section 3.4.6. Specifically, it does not seem to conflict with any known file types, and it is not valid Unicode in any Unicode encoding.

The second part identifies which CBOR Protocol is used, as described above.

The third part is a constant value 0x43_42_4f_52, "CBOR". That is, it the three byte sequence 0x42_4f_52 ("BOR"). This is the data item that is tagged.

The actual CBOR Protocol value then follows as the next data item(s) in the CBOR sequence, without a need for any further specific tag. The use of a CBOR Sequence allows the application to trivially remove the first item with the two tags.

This means that should a file be reviewed by a human (directly in an editor, or in a hexdump display), it will include the string "CBOR" prominently. This value is also included simply because the two tags need to tag something.

4. Security Considerations

This document provides a way to identify CBOR Protocol objects. Clearly identifying CBOR contents on disk may have a variety of impacts.

The most obvious is that it may allow malware to identify interesting objects on disk, and then corrupt them.

5. IANA Considerations

There are no IANA actions. This section documents the allocation that was done.

5.1. CBOR Sequence Tag

IANA has allocated tag 55800 as the CBOR Sequence tag. This tag is from the First Come/First Served area.

The value has been picked to have properties similiar to the 55799 tag.

The hexadecimal representation is: 0xd9_\d9_f8.

This is not valid UTF-8: the first 0xd9 puts the value into the three-byte value of UTF-8, but the 0xd9 as the second value is not a valid second byte for UTF-8.

This is not valid UTF-16: the byte sequence 0xd9d9 (in either endian order), puts this value into the UTF-16 high-half zone, which would signal that this a 32-bit Unicode value. However, the following 16-bit big-endian value 0xf8.. is not a valid second sequence according to [RFC2781]. On a little-endian system, it would be necessary to examine the fourth byte to determine if it is valid. That next byte is determined by the subsequent encoding, and [RFC8949] section 3.4.6 has already determined that no valid CBOR encodings result in a valid UTF-16.

Data Item: byte string
Semantics: indicates that the file contains CBOR Sequences

6. Acknowledgements

The CBOR WG brainstormed this protocol on January 20, 2021.

7. Changelog

8. References

8.1. Normative References

Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <>.
Bormann, C., "Concise Binary Object Representation (CBOR) Sequences", RFC 8742, DOI 10.17487/RFC8742, , <>.
Bormann, C. and P. Hoffman, "Concise Binary Object Representation (CBOR)", STD 94, RFC 8949, DOI 10.17487/RFC8949, , <>.

8.2. Informative References

Wikipedia, "file (command)", , <>.
Mandyam, G., Lundblade, L., Ballesteros, M., and J. O'Donoghue, "The Entity Attestation Token (EAT)", Work in Progress, Internet-Draft, draft-ietf-rats-eat-09, , <>.
Birkholz, H., Fitzgerald-McKay, J., Schmidt, C., and D. Waltermire, "Concise Software Identification Tags", Work in Progress, Internet-Draft, draft-ietf-sacm-coswid-17, , <>.
Raza, S., Höglund, J., Selander, G., Mattsson, J. P., and M. Furuhed, "CBOR Encoded X.509 Certificates (C509 Certificates)", Work in Progress, Internet-Draft, draft-mattsson-cose-cbor-cert-compress-08, , <>.
Wikipedia, "Interleaved BitMap", , <>.
Hoffman, P. and F. Yergeau, "UTF-16, an encoding of ISO 10646", RFC 2781, DOI 10.17487/RFC2781, , <>.
Schaad, J., "CBOR Object Signing and Encryption (COSE)", RFC 8152, DOI 10.17487/RFC8152, , <>.

Appendix A. Example from Openswan

The Openswan IPsec project has a daemon ("pluto"), and two control programs ("addconn", and "whack"). They communicate via a Unix-domain socket, over which a C-structure containing pointers to strings is serialized using a bespoke mechanism. This is normally not a problem as the structure is compiled by the same compiler; but when there are upgrades it is possible for the daemon and the control programs to get out of sync by the bespoke serialization. As a result, there are extra compensations to deal with shutting the daemon down. During testing it is sometimes the case that upgrades are backed out.

In addition, when doing unit testing, the easiest way to load policy is to use the normal policy reading process, but that is not normally loaded in the daemon. Instead the IPC that is normally sent across the wire is compiled/serialized and placed in a file. The above magic number is included in the file, and also on the IPC in order to distinguish the "shutdown" command CBOR operation.

In order to reduce the problems due to serialization, the serialization is being changed to CBOR. Additionally, this change allows the IPC to be described by CDDL, and for any language that encode to CBOR can be used.

IANA has allocated the tag 1330664270, or 0x4f_50_ 53_ 4e for this purpose. As a result, each file and each IPC is prefixed with:

In diagnostic notation: ~~~~ 55800(1330664270(h'424F52')) ~~~~

Or in hex: ~~~~ 00000000 d9 d9 f9 da 4f 50 53 4e 43 42 4f 52 |....OPSNCBOR| ~~~~


Carsten Bormann
Josef 'Jeff' Sipek

Author's Address

Michael Richardson
Sandelman Software Works