This is a C/C++ implementation of the zxcvbn password strength estimation.
The code is intended to be included as part of the source of a C/C++ program. Like the original this code is for character sets which use single byte characters primarily in the code range 0x20 to 0x7E.
The original CoffeeScript version is available at https://github.com/lowe/zxcvbn
An article on the reasons for zxcvbn is at https://blogs.dropbox.com/tech/2012/04/zxcvbn-realistic-password-strength-estimation/
The makefile will build several test programs to test the code. It shows the steps needed to use the code in C and C++ programs, using the dictionary data read from file or included within the program executable. The makefile has only been tried on Linux using GCC version 4.8.4, but should be fairly portable to other systems.
When dictionary data is included in your program's executable, the files zxcvbn.c ,
zxcvbn.h , dict-src.h are used in your program.
When dictionary data is read from file, the files zxcvbn.c , zxcvbn.h , dict-crc.h
and zxcvbn.dict are used in your program, compiled with #define USE_DICT_FILE. The CRC
of the dictionary data file is written to dict-crc.h so your executable can detect
corruption of the data.
Rename zxcvbn.c to zxcvbn.cpp (or whatever your compiler uses) to compile as C++.
The dict*.h and zxcvbn.dict files are generated by the dictgen program compiled from
dict-generate.cpp (see makefile for details).
Initially call ZxcvbnInit() with the pathname of the zxcvbn.dict file. This can be
omitted when dictionary data is included in the executable.
Call ZxcvbnMatch() with the password and optional user dictionary to get the entropy
estimation and optional information on the password parts (which will need freeing with
ZxcvbnFreeInfo() after use). Do this for each password to be tested, or as each character
of it is entered into your program. The optional user dictionary can change between each
call.
Finally call ZxcvbnUninit() to free the dictionary data from read from file. This can be
omitted when dictionary data is included in the executable.
Review the test program in test.c for an example.
The entropy calculated will sometimes differ from the original because of
- The UK keyboard layout is also included, so there are additional spacial sequences, e.g. ;'# is a spacial sequence.
- The different character classes in a password are taken into account when calculating the strength of brute-force matches.
- Dijkstra's path searching algorithm is used to combine parts of the entered password. This can result in the found parts of the password being combined differently than the original CoffeeScript. E.g. the password passwordassword is combined by the original CoffeeScript as p (3.5 bits) + asswordassword (12.6 bits) + multiple part allowance (1.0bit) to give total entropy of 17.1 bits. This implementation combines it as password (1.0 bit) + assword (11.6 bits) + multiple part allowance (1.0bit) to give 13.6 bits.
- For multi-part passwords the original CoffeeScript version multiplies the number of guesses needed by the factorial of the number of parts. This is not possible in this version as Dijkstra's algorithm is used. Instead one bit entropy is added for the part at the end of the password, 1.7 bits for each part in the middle of a password and nothing for the part at the beginning. This gives similar results compared to the CoffeeScript version when there are 4 or less parts, but will differ significantly when there are many parts (which is likely to be a rare occurrence).
The original CoffeeScript version is available at https://github.com/lowe/zxcvbn
The dictionary words are taken from the original CoffeeScript version.
Dictionary trie encoding (used for by the word lookup code) based on idea from the Caroline Word Graph from http://www.pathcom.com/~vadco/cwg.html
MIT License