Markov-chain passwords generator

The Markov-chain password generator is one of the password guessing methods. The trained model represents the probability of transition between specific characters. Password guessing attacks using n-grams (i.e., substrings of length n appearing in a training set) have been originally proposed by Narayanan and Shmatikov.

There is another implementation by brannondorsey here. It has written in Python, and It is very slow for experimentation purposes. This implementation is written in Go and is much faster.

License

markov-passwords is licensed under the MIT license.

Tests

Speed

Trained on Top304Thousand-probable-v2 password list.

The pv results:

./bin/mkpass -sample top304K.json -m 100000000 | pv -lta > /dev/null
00:30 [1.44M/s]

Python implementation:

python sample.py | pv -lta > /dev/null
00:30 [5.30k/s]

Unique Passwords

The sort results:

./bin/mkpass -sample top304K.json -m 1000000 | sort -u | wc -l
786577

Python implementation:

python sample.py | head -1000000 | sort -u | wc -l
541842

Usage

Get and Build

go get github.com/bujimuji/markov-passwords/...
cd $GOPATH/src/github.com/bujimuji/markov-passwords
make

Train and Generate Passwords

Usage of ./bin/mkpass:
  -infile string
    	read training password list from file (- for stdin)
  -j int
    	number of channels for fan-out (default 64)
  -m uint
    	the amount of passwords created (default 1000000)
  -ngram int
    	number of ngrams to train markov model (default 3)
  -sample string
    	model path for generating new passwords
  -train string
    	model path for saving trained model in json format and exit

Example:

./bin/mkpass -infile assets/Top304Thousand-probable-v2.txt -m 100

Only Train

Example:

./bin/mkpass -infile assets/Top304Thousand-probable-v2.txt -train top304k.json

Only Generate

Example:

./bin/mkpass -sample top304K.json -m 1000