Home

Awesome

Binary Modular Crypt Format

Binary Modular Crypt Format (BMCF) is the binary counterpart to Modular Crypt Format (MCF). MCF is a de facto standard for encoding password hashes in an ASCII string, and it allows hashes generated with different hash algorithms to be stored alongside one another compatibly. MCF is not formally defined, but it has been documented by the authors of the Passlib password hashing library for Python.

Modular Crypt Format originated with the crypt(3) Unix C library function, and has since been implemented in several other libraries. Historically, MCF was used to encode the password hashes of operating system users for storage in a text file. Nowadays, MCF is also used in systems which store password hashes in database management systems. On large systems with many users, it is often desirable to maximize storage efficiency. To that end, I present Binary Modular Crypt Format as a compact binary serialization of MCF hashes. Depending on the hash scheme, BMCF reduces size by approximately 25–33%.

Terminology

MCF Example

A typical MCF hash has several components and looks like this:

$2y$14$i5btSOiulHhaPHPbgNUGdObga/GC.AVG/y5HHY1ra7L0C9dpCaw8u

MCF hashes start with <code>$</code>, which is also used to delimit fields. The field following the first <code>$</code> identifies the hash scheme that was used. Beyond that, field definitions vary by hash scheme, but generally include algorithm parameters, a salt, and the actual digest.

The example above is a Bcrypt hash, identified by the <code>2y</code> scheme identifier. The next field, <code>14</code>, is the Bcrypt cost parameter. The first 22 characters of the final field represent the 128-bit salt, <code>i5btSOiulHhaPHPbgNUGdO</code>. The remaining 31 characters represent the 184-bit Bcrypt hash digest, <code>bga/GC.AVG/y5HHY1ra7L0C9dpCaw8u</code>. The salt and digest are base-64 encoded using a non-standard alphabet.

Note that while Bcrypt puts the salt and digest in the same field, other schemes may use separate fields, usually to accommodate variable-length salts.

Hash sizes vary by algorithm and hash scheme. A Bcrypt hash, as in the example above, is 60 characters in MCF, which of course requires 60 bytes of storage in any single-byte character encoding. A Bcrypt hash in BMCF requires 40 bytes: <code>(128+184)/8 + 1</code>.

General Specification

So far, BMCF is only fully defined for Bcrypt. BMCF definitions for additional algorithms can be added as needed.

Binary MCF hashes start with a header octet. This first byte employs its three most significant bits for the scheme identifier as follows:

0x00 [Reserved; perhaps for the pre-MCF implicit DES schemes]
0x20 $2$    Bcrypt
0x40 $2a$   Bcrypt
0x60 $2x$   Bcrypt
0x80 $2y$   Bcrypt
0xA0 $2b$   Bcrypt
0xC0 [Reserved]
0xE0 [Reserved to expand the identifier to the next 5 bits]

Usage of the remaining five bits of the header octet (and all subsequent octets) is specified by the BMCF definition for the hash scheme. The only exception is the special case <code>0xE0</code>, which is used to indicate that the remaining five bits (the least significant bits) are used for the hash scheme identifier. If needed, usage of this space will be defined later to accommodate additional hash schemes and indicate overflow of the identifier to subsequent octets.

Most MCF hash schemes encode the password digest and salt using some form of base-64 binary-to-text encoding. For conversion to BMCF, such fields are decoded back to their binary forms with the reverse text-to-binary process. Typically, the encoding used by crypt(3) hash schemes does not conform to RFC 4648 Base64. Most schemes differ only in the selection of the base-64 alphabet, but others differ in the bit ordering used during the encoding process. For use with BMCF, it is important to decode using precisely the base-64 process that is prescribed by the hash scheme, including bit ordering, alphabet selection, and alphabet letter ordering.

With regard to the hash as a whole, to decode is to convert a hash from its textual notation (MCF) to its binary serialization (BMCF).

BMCF = decode(MCF)

To encode is to convert a hash from binary form to textual MCF notation.

MCF = encode(BMCF)

BMCF definitions for specific hash schemes must ensure data integrity, meaning they perform lossless decode/encode roundtrips for valid hashes. Given an invalid hash, behavior may vary by hash scheme or code implementation. An error may be raised; an erroneous hash component may be corrected, breaking data integrity; or the invalid hash component may be encoded and decoded as-is, maintaining data integrity of the invalid hash.

Bcrypt BMCF Definition

A Brcypt MCF hash looks like this:

$2y$14$i5btSOiulHhaPHPbgNUGdObga/GC.AVG/y5HHY1ra7L0C9dpCaw8u

Decoded to BMCF, the hash requires 40 bytes, numbered 0–39.

Bcrypt hashes have four components in three fields. The salt and digest share the same field; they are not separated by the <code>$</code> delimiter.

$<scheme>$<cost>$<salt><digest>

There are five Bcrypt scheme identifiers.

0x20 $2$
0x40 $2a$
0x60 $2x$
0x80 $2y$
0xA0 $2b$

To decode, we start with the scheme identifier in the header octet, byte 0. For the <code>2y</code> scheme in our example, this is <code>0x80</code>. The algorithm identifier uses only the three most significant bits. In the lower five bits we store the cost parameter, which Bcrypt requires to be between 04 and 31 (with the leading zero in MCF). With five bits we can store any value between 00 and 31. Adding <code>14</code> to the example header octet, we now have <code>0x8E</code>.

The 128-bit salt and 184-bit digest in the MCF representation are base-64 encoded. Bcrypt base 64 can be encoded and decoded as described in RFC 4648, but using the following alphabet, and omitting the <code>=</code> padding:

./ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789

The Bcrypt salt is the first 22 characters from the salt/digest field, and the digest is the final 31 characters. The salt and digest must be base-64 decoded separately. Decoding the salt yields 16 bytes, appended to the header octet in positions 1–16. Decoding the digest yields 23 bytes to occupy positions 17–39, completing the decoded hash.

Code Implementations

There is a PHP implementation of BMCF.