An implementation, in Python3, of the Kravatte pseudo-random function and associated modes based on the Farfalle PRF system. At its core, Kravatte accepts a user defined secret key and a sequence of input bytes to generate pseudo-random output of arbitrary size. From this primitive, a number of authenticated encryption modes can be built.
Kravatte makes use of the Keccak permutation, most notably used in NIST’s FIPS 202 SHA-3 algorithm. Because the underlying structure of Keccak function works on a three-dimensional state of 1600 bits, it maps well to a 5x5 matrix of 64-bit unsigned integers. As such, the NumPy computational library is a natural fit to quickly manipulate such a structure and thus is a hard requirement.
This implementation reflects the updated, more secure Kravatte
Achouffe released in late 2017. The older Kravatte 6644 logic is
available within this repo as well.
Also supported are the Kravatte-SANE and Kravatte-SANSE
session based modes. These modes replace the deprecated Kravatte-SAE and Kravatte-SIV modes and
utilizes the Deck-SANE and Deck-SANSE modes described in the Xoodoo Cookbook.
Kravatte can be easily installed from pypi via pip:
$ pip install kravatteIf pip is unavailable, this repo can be cloned and setup can be done manually:
$ python setup.py installThe basic Kravatte object operates on two Keccak-1600 state matrices;
the collector state and the key state. Instantiating a Kravatte object
initializes the key state with provided user key and sets the collector
state to zeros.
In [1]: from kravatte import Kravatte
In [2]: my_krav = Kravatte(b'1234567890')The newly initialized Kravatte object is now ready to accept input
strings of bytes for absorption into the collector state via the
collect_message method. Repeated calls to collect_message are
equivalent to B ◦ A sequences as described in the the Farfalle
spec:
In [3]: input_a = b'The quick brown fox jumps over the lazy dog'
In [4]: my_krav.collect_message(input_a)
In [5]: input_b = b'3533392d36302d35313235'
In [6]: my_krav.collect_message(input_b)Once absorbing message strings is complete, the Kravatte object can
produce an arbitrary number of pseudo-random output bytes via the
generate_digest method. Those bytes are then available in the
digest attribute:
In [7]: output_bytes = 64
In [8]: my_krav.generate_digest(output_bytes)
In [9]: from binascii import hexlify
In [10]: hexlify(my_krav.digest)
Out[10]: b'8a0fc89899e058dedd368b60111bf4958f4f24216bbac76936471e6f7c3958b881c38c8e829ff07bf137701917b3e49ab392e93f3b2abfc714f90c0ca023124d'The absorb/output sequence can be restarted with another call to
collect_message. This clears the collector state and resets the key
state to its initialized value. Alternatively, the user may change to a
new secret key with the update_key method to reinitialize the key
state used at the start of message absorption.
When a Kravatte object has reached the end of its usable lifetime, the scrub method
can be used to try and cleanup interim state and key data in resident memory. This method is executed by default
on the standalone functions mac, siv_wrap, siv_unwrap, and is available in all Kravatte derived classes.
NOTE This method only clears the sensitive attributes collector, kra_key, and roll_key and shouldn't be
considered applicable when using the multi-process accelerated mode.
The most basic mode of Kravatte is an authenticated pseudo-random
function (PRF). Kravatte can absorb an arbitrary sized user message and
key, and output an arbitrary collection of pseudo-random bytes that can
act as a message authentication code. The mac does this in a single step:
In [1] from kravatte import mac
In [2] from binascii import hexlify
In [3] message = b'Attack at Dawn!'
In [4] key = b'something_secret'
In [5] mac_size = 64
In [6] g = mac(key, message, mac_size)
In [7] hexlify(g)
Out[7] b'24f61fc5fd38fef7f3d799ed72b24578c4479e1c035c70d8bc55ce23d74124255d5e8a0c5dd33aa36d5289f1e4e995a19be804d97bb338fa875e01e3c2d2dd51'Kravatte-SANE mode is a session based method of AEAD. Given a random
nonce and secret key, this mode encrypts a sequence of plaintext
messages and/or metadata into appropriately sized ciphertexts and a validation
tags. The sequence of plaintext/metadata is tracked as a history that
builds a chain of authentication from message to message and requires
all generated ciphertexts to be processed to fully decrypt and verify.
A separate KravatteSANE class is provided that adds the history
tracking for each encryption operation done via the wrap method.
In [1]: from os import urandom
In [2]: from binascii import hexlify
In [3]: from time import monotonic
In [4]: my_nonce=urandom(32)
In [5]: hexlify(my_nonce)
Out[5]: b'41c48803e34eefd9ac1d39d3412d3e32592173fbcdd1b60d85dc177ae7156733'
In [6]: message1=b'Nice List:'
In [7]: meta1=str(monotonic()).encode()
In [8]: message2=b'Alice,Bob'
In [9]: meta2=str(monotonic()).encode()
In [10]: message3=b'Naughty List:'
In [11]: meta3=str(monotonic()).encode()
In [12]: message4=b'Chuck, Eve'
In [13]: meta4=str(monotonic()).encode()
In [14]: my_sane = KravatteSANE(my_nonce,my_key)
In [15]: ctext_1, tag_1 = my_sane.wrap(message1, meta1)
In [16]: hexlify(ctext_1)
Out[16]: b'4b42fef9cb5a6ce69d78'
In [17]: hexlify(tag_1)
Out[17]: b'169e7eb0f63cebd70efb779ff45a67f0'
In [18]: ctext_2, tag_2 = my_sane.wrap(message2, meta2)
In [19]: ctext_3, tag_3 = my_sane.wrap(message3, meta3)
In [20]: ctext_4, tag_4 = my_sane.wrap(message4, meta4)For decryption and validation, the unwrap method accepts the
ciphertext, original metadata, and validation tag to not only decrypt
the plaintext, but return a boolean if the decrypted plaintext is valid
within the chain of messages.
In [21]: decrypt_sane = KravatteSANE(my_nonce,my_key)
In [22]: ptext_1, tag_valid1 = decrypt_sane.unwrap(ctext_1, meta1, tag_1)
In [23]: ptext_1
Out[23]: b'Nice List:'
In [24]: tag_valid1
Out[24]: True
In [25]: ptext_2, tag_valid2 = decrypt_sane.unwrap(ctext_2, meta2, tag_2)
In [26]: tag_valid2
Out[26]: True
In [27]: ptext_2
Out[27]: b'Alice,Bob'
In [28]: ptext_3, tag_valid3 = decrypt_sane.unwrap(ctext_3, meta3, tag_3)
In [29]: ptext_3
Out[29]: b'Naughty List:'
In [30]: tag_valid3
Out[30]: True
In [31]: ptext_4, tag_valid4 = decrypt_sane.unwrap(ctext_4, meta4, tag_4)
In [32]: ptext_4
Out[32]: b'Chuck, Eve'
In [33]: tag_valid4
Out[33]: TrueKravatte-SANSE mode is session based method of authenticated encryption with
associated metadata (AEAD) that allows for encrypting a provided
plaintext with a secret shared key and an arbitrary metadata value.
This mode does not require a nonce as it operates with a
Synthetic Initialization Vector (SIV)
Encryption generates an equal length ciphertext and fixed length tag
that can be used to validate the plaintext at decryption. Metadata
values can be shared for different key/message combinations with
understanding that the more a value is used, the greater the chance of a
tag collision. This mode replaces Kravatte-SIV
A KravatteSANSE class is provided that adds the history
tracking for each encryption operation done via the wrap method.
In [1]: from binascii import hexlify
In [2]: from kravatte import KravatteSANSE
In [3]: my_message = b'And yet it moves'
In [4]: my_key = b'name of childhood pet'
In [5]: metadata_1 = b'1024x768'
In [6]: another_message = b'The present is theirs; the future, for which I really worked, is mine.'
In [7]: metadata_2 = b'7680x4320'
In [8]: my_sanse = KravatteSANSE(my_key)
In [9]: ctext_1, tag_1 = my_sanse.wrap(my_message, metadata_1)
In [10]: hexlify(ctext_1)
Out[10]: b'79e4773536a2ac4b4ec9e93583a817a5'
In [11]: hexlify(tag_1)
Out[11]: b'eaa50cb8a02e3238aa8dd5d1186ec0a87ebf6fe71b6fd89bea20b2001fef6810'
In [12]: ctext_2, tag_2 = my_sanse.wrap(another_message, metadata_2)In [13]: decrypt_sanse = KravatteSANSE(my_key)
In [14]: ptext_1, tag_valid_1 = decrypt_sanse.unwrap(ctext_1, metadata_1, tag_1)
In [15]: ptext_1
Out[15]: b'And yet it moves'
In [16]: tag_valid_1
Out[16]: True
In [17]: ptext_2, tag_valid_2 = decrypt_sanse.unwrap(ctext_2, metadata_2, tag_2)
In [18]: ptext_2
Out[18]: b'The present is theirs; the future, for which I really worked, is mine.'
In [19]: tag_valid_2
Out[19]: TrueKravatte Wide Block Cipher mode is a symmetric block cipher mode where the user can specify
the size of the block, an arbitrary tweak value input, and arbitrary secret key. The KravatteWBC
object, once initialized, can encrypt/decrypt messages of the given block size (or smaller). KravatteWBC
splits messages into left and right components and uses a 4-stage Feistel sequence to encrypt/decrypt.
In [1]: from kravatte import KravatteWBC
In [2]: block_size = 64
In [3]: my_tweak = b'tweak can be anything'
In [4]: my_key = b'\x00' * 24
In [5]: my_wbc = KravatteWBC(block_size, my_tweak, my_key)
In [6]: c_block = my_wbc.encrypt(b'This is some random 64-byte text string to use in this example!!')
In [7]: from binascii import hexlify
In [8]: hexlify(c_block)
Out[8]: b'2368fae1271e5c784537df331586d5d4daeeb34a6fe4ebea03cc1df7f9c0d79fcc709a9ff2199514f431da685e27658dbf6c5afed11ce5c8172f7615c19db1b9'
In [9]: my_wbc.decrypt(c_block)
Out[9]: b'This is some random 64-byte text string to use in this example!!'KravatteWBC-AE is a variant of KravatteWBC that extends the desired block size by 16 bytes and
embeds authentication data. The tweak is replaced with arbitrary associated metadata. When the
block is decrypted it is also validated as being encrypted with same secret key.
In [1]: from datetime import datetime
In [2]: from binascii import hexlify
In [3]: my_key = b"Doesn't look like anything to me"
In [4]: metadata = str(datetime.now()).encode()
In [5]: message = b'These violent delights have violent ends'
In [6]: len(message)
Out[6]: 40
In [7]: my_WBC_AE = KravatteWBC_AE(40, my_key)
In [8]: ctext_ae = my_WBC_AE.wrap(message, metadata)
In [9]: len(ctext_ae)
Out[9]: 56
In [10]: hexlify(ctext_ae)
Out[10]: b'388623f7a7d3c044cda574063b4ff16edbdfc95cb449f335a1c5ad5ed37897aa2470f3575825a55df04cc1dab34b4feb03aa6d35f6190d62'
In [11]: plaintext, validated = my_WBC_AE.unwrap(ctext_ae, metadata)
In [12]: plaintext
Out[12]: b'These violent delights have violent ends'
In [13]: validated
Out[13]: TrueKravatteOracle is a simple pseudo-random number generator built from the Kravatte PRF primitive. Initialized
with an authentication key, the KravatteOracle object absorbs an arbitrarily sized seed value into the
collector state. From there, streams of random bytes can be generated on demand via the random method.
The generator can be re-seeded at any point with the seed_generator method.
In [1]: my_psrng = KravatteOracle(my_seed, my_key)
In [2]: my_key = b'1234'
In [3]: my_seed = b'watermelon'
In [4]: my_psrng = KravatteOracle(my_seed, my_key)
In [5]: random_bytes = my_psrng.random(24)
In [6]: hexlify(random_bytes)
Out[6]: b'14a42ab5756efe61eae73893570b6736b392d0031a87e36d'
In [7]: random_bytes = my_psrng.random(42)
In [8]: hexlify(random_bytes)
Out[8]: b'77d6308e18d57fb124e75602ced2e863e7de34c69ea57bec47efae84e85d0075c3ebbf7e535ec0fb096f'In [9]: my_psrng.seed_generator(b'apple')
In [10]: random_bytes = my_psrng.random(18)
In [11]: hexlify(random_bytes)
Out[11]: b'3e108c3f627f561943893b6a3184e5b76472'Kravatte-SIV mode is a method of authenticated encryption with
associated metadata (AEAD) that allows for encrypting a provided
plaintext with a secret shared key and an arbitrary metadata value.
Encryption generates an equal length ciphertext and fixed length tag
that can be used to validate the plaintext at decryption. Metadata
values can be shared for different key/message combinations with
understanding that the more a value is used, the greater the chance of a
tag collision. Deprecated in favor of Kravatte-SANSE
In [1] from kravatte import siv_wrap, siv_unwrap
In [2] from binascii import hexlify
In [3] from datetime import datetime
In [4] message = b'Attack at Dawn!'
In [5] key = b'something_secret'
In [6] metadata = str(datetime.now()).encode()
In [7] ciphertext, tag = siv_wrap(key, message, metadata)
In [8] hexlify(ciphertext)
Out[8] b'79f7bd89a7cb7af1892ea51c531f4b'
In [9] hexlify(tag)
Out[9] b'37c7e11f0c9c744e7c113590fdfba7737cb38b629ef6901df22d6994340e89eas'In [10] plaintext, tag_valid = siv_unwrap(key, ciphertext, tag, metadata)
In [11] plaintext
Out[11] b'Attack at Dawn!'
In [12] tag_valid
Out[12] TrueKravatte-SAE mode is a session based method of AEAD. Given a random
nonce and secret key, this mode encrypts a sequence of plaintext
messages and/or metadata into equal size ciphertexts and a validation
tag. The sequence of plaintext/metadata is tracked as a history that
builds a chain of authentication from message to message and requires
all generated ciphertexts to be processed to fully decrypt and verify.
Deprecated in favor of Kravatte-SANE
A separate KravatteSAE class is provided that adds the history
tracking for each encryption operation done via the sae_wrap method.
In [1]: from kravatte import KravatteSAE
In [2]: from datetime import datetime
In [3]: from binascii import hexlify
In [4]: message_1 = b'Directions to my house:'
In [5]: metadata_1 = str(datetime.now()).encode()
In [6]: message_2 = b'Turn right on main street'
In [7]: metadata_2 = str(datetime.now()).encode()
In [8]: message_3 = b'Continue straight for 3500 miles'
In [9]: metadata_3 = str(datetime.now()).encode()
In [10]: message_4 = b'You have arrived at your destination'
In [11]: metadata_4 = str(datetime.now()).encode()
In [12]: nonce = b'a well chosen random number'
In [13]: key = b'an even better random number'
In [14]: KravSAE_wrapper = KravatteSAE(nonce, key)
In [15]: ciphertext_1, tag_1 = KravSAE_wrapper.sae_wrap(message_1, metadata_1)
In [16]: hexlify(ciphertext_1)
Out[16]: b'7b8932a1c3673fcfe752631ef5b867843951514335de61'
In [17]: hexlify(tag_1)
Out[17]: b'3384885ca293925cc65a03fa10790420'
In [18]: ciphertext_2, tag_2 = KravSAE_wrapper.sae_wrap(message_2, metadata_2)
In [19]: hexlify(ciphertext_2)
Out[19]: b'ab48882d4339c6def9d5d06f608db5318a87a417566c0b20bd'
In [20]: hexlify(tag_2)
Out[20]: b'347f5a152dcc9ccc3c19fa936067c3d2'
In [21]: ciphertext_3, tag_3 = KravSAE_wrapper.sae_wrap(message_3, metadata_3)
In [22]: hexlify(ciphertext_3)
Out[22]: b'bc461f40db74705c10b1400b6a9967dd7164cbf774c196d5b649faf2bd792339'
In [23]: hexlify(tag_3)
Out[23]: b'6ba2faee4d2aa5654a054222a049d926'
In [24]: ciphertext_4, tag_4 = KravSAE_wrapper.sae_wrap(message_4, metadata_4)
In [25]: hexlify(ciphertext_4)
Out[25]: b'1f451f51d9882f9f7674c37dace4036efd9efe39d6b58ccdf6b012ef988e4e1f2617479f'
In [26]: hexlify(tag_4)
Out[26]: b'5f3511f140b4ea36412c0e4b22d1c218'For decryption and validation, the sae_unwrap method accepts the
ciphertext, original metadata, and validation tag to not only decrypt
the plaintext, but return a boolean if the decrypted plaintext is valid
within the chain of messages.
In [27]: KravSAE_unwrapper = KravatteSAE(nonce, key)
In [28]: plaintext_1, check_tag_1 = KravSAE_unwrapper.sae_unwrap(ciphertext_1, metadata_1, tag_1)
In [29]: plaintext_1
Out[29]: b'Directions to my house:'
In [30]: check_tag_1
Out[30]: True
In [31]: plaintext_2, check_tag_2 = KravSAE_unwrapper.sae_unwrap(ciphertext_2, metadata_2, tag_2)
In [32]: plaintext_2
Out[32]: b'Turn right on main street'
In [33]: check_tag_2
Out[33]: True
In [34]: plaintext_3, check_tag_3 = KravSAE_unwrapper.sae_unwrap(ciphertext_3, metadata_3, tag_3)
In [35]: plaintext_3
Out[35]: b'Continue straight for 3500 miles'
In [36]: check_tag_3
Out[36]: True
In [37]: plaintext_4, check_tag_4 = KravSAE_unwrapper.sae_unwrap(ciphertext_4, metadata_4, tag_4)
In [38]: plaintext_4
Out[38]: b'You have arrived at your destination'
In [39]: check_tag_4
Out[39]: TrueThe Farfalle PRF allows for significant parallelism in both the compression and expansion phases when
consuming or generating large numbers of blocks. We can exploit that fact for increased performance
via Python's multiprocessing module.
This allows us to spawn any number of identical worker subprocesses that can consume additional
CPU core resources. Enabling the multi-process mode is done at object creation time for Kravatte,
or any of its operating modes, with the workers arguments:
In [1]: new_kravatte = Kravatte(my_key, workers=8)
In [2]: my_kra_mac = mac(my_key, my_message, my_output_size, workers=16)
In [3]: my_wbc = KravatteWBC(block_size, my_tweak, my_key, workers=4)For optimal performance, the number of workers should match the number of CPU cores reported by
os.cpu_count. This is set automatically if workers is set to 0:
# Equivalent objects
In [4]: my_psrng = KravatteOracle(my_seed, my_key, workers=0)
In [5]: my_psrng = KravatteOracle(my_seed, my_key, workers=os.cpu_count())Multi-process mode can be explicitly disabled by setting workers to None:
In [6]: my_psrng = KravatteOracle(my_seed, my_key, workers=None)There is a non-trivial performance cost associated with generating new Python processes. For small, generally < 100KB, inputs and outputs, it can be faster to use the single process variant.
For asymmetrically sized workloads, such a generating a MAC on a multi-megabyte input and only
generating a few dozen bytes of output, multiprocessing mode can be explicitly enabled or disabled with the
mp_input and mp_output arguments. These booleans are available for Kravatte and all derived classes/functions.
# Enable Multiprocessing acceleration only for processing of input bytes
In [7]: my_psrng = KravatteOracle(my_seed, my_key, workers=16, mp_input=True, mp_output=False)A full test suite is available in the test/ dir. Tests can be invoked with pytest:
$ cd path/to/cloned/kravatte/
$ pytest -xvvvThe same tests are run against the standard codepath and the multiprocess code path utilizing all available CPU cores. Test vectors were generated using the KeccakTools C++ library available from the Keccak Team
- Being a Python implementation, performance on large files or data sets may be inadequate (even with multi-processing enabled).
- The inputs and outputs of this implementation are limited to byte (8-bit) divisible sizes
- While security was top of mind during development, this implementation has not been fully audited for timing attacks, side channel attacks or other vulnerabilities. Other bugs not caught by the test cases may be present. Use in a production environment is not encouraged.
If any of above are of concern, please check out the official KeccakTools and Keccak Code Package
- Add Kravatte-SANE Support [Calvin McCoy]
- Add Kravatte-SANSE Support [Calvin McCoy]
- Refactor tests into seperate classes [Calvin McCoy]
- Add TravisCI Testing [Calvin McCoy]
- Add memory scrub functionality. [Calvin McCoy]
- Optimizations to Keccak, expand permutation, and compress permutations [Calvin McCoy]
- Added ability to enable Multi-processing for just input or output [Calvin McCoy]
- Fixed Typos and Added Some Extra Comments [Calvin McCoy]
- Added Multi-processing mode [Calvin McCoy]
- Cleanup for 1.0 release [Calvin McCoy]
- Add KravatteOracle pseudo-random generator [Calvin McCoy]
- Add type hinting [Calvin McCoy]
- Fix Typos [Calvin McCoy]
- General package cleanup and fix typos [Calvin McCoy]
- Initial Commit [Calvin McCoy]