derivepassphrase.git

1) # SPDX-FileCopyrightText: 2024 Marco Ricci <m@the13thletter.info>
2) #
3) # SPDX-License-Identifier: MIT

4) 

5) """Work-alike of vault(1) – a deterministic, stateless password manager"""  # noqa: RUF002

6) 
7) from __future__ import annotations
8) 

9) import base64

10) import collections
11) import hashlib
12) import math

13) import unicodedata

14) 

15) from typing_extensions import assert_type

16) 
17) import sequin
18) import ssh_agent_client
19) 

20) __author__ = 'Marco Ricci <m@the13thletter.info>'

21) __version__ = '0.1.2'

22) 

23) 

24) class AmbiguousByteRepresentationError(ValueError):
25)     """The object has an ambiguous byte representation."""

26)     def __init__(self):
27)         super().__init__('text string has ambiguous byte representation')
28) 

29) 

30) _CHARSETS = collections.OrderedDict([
31)     ('lower', b'abcdefghijklmnopqrstuvwxyz'),
32)     ('upper', b'ABCDEFGHIJKLMNOPQRSTUVWXYZ'),
33)     ('alpha', b''),  # Placeholder.
34)     ('number', b'0123456789'),
35)     ('alphanum', b''),  # Placeholder.
36)     ('space', b' '),
37)     ('dash', b'-_'),
38)     ('symbol', b'!"#$%&\'()*+,./:;<=>?@[\\]^{|}~-_'),
39)     ('all', b''),  # Placeholder.
40) ])
41) _CHARSETS['alpha'] = _CHARSETS['lower'] + _CHARSETS['upper']
42) _CHARSETS['alphanum'] = _CHARSETS['alpha'] + _CHARSETS['number']

43) _CHARSETS['all'] = (
44)     _CHARSETS['alphanum'] + _CHARSETS['space'] + _CHARSETS['symbol']
45) )
46) 

47) 

48) class Vault:
49)     """A work-alike of James Coglan's vault.
50) 
51)     Store settings for generating (actually: deriving) passphrases for
52)     named services, with various constraints, given only a master
53)     passphrase.  Also, actually generate the passphrase.  The derivation
54)     is deterministic and non-secret; only the master passphrase need be
55)     kept secret.  The implementation is compatible with [vault][].
56) 
57)     [James Coglan explains the passphrase derivation algorithm in great
58)     detail][ALGORITHM] in his blog post on said topic: A principally
59)     infinite bit stream is obtained by running a key-derivation function
60)     on the master passphrase and the service name, then this bit stream

61)     is fed into a [Sequin][sequin.Sequin] to generate random numbers in
62)     the correct range, and finally these random numbers select
63)     passphrase characters until the desired length is reached.

64) 
65)     [vault]: https://getvau.lt
66)     [ALGORITHM]: https://blog.jcoglan.com/2012/07/16/designing-vaults-generator-algorithm/
67) 
68)     """

69) 

70)     _UUID = b'e87eb0f4-34cb-46b9-93ad-766c5ab063e7'
71)     """A tag used by vault in the bit stream generation."""

72)     _CHARSETS = _CHARSETS

73)     """
74)         Known character sets from which to draw passphrase characters.
75)         Relies on a certain, fixed order for their definition and their
76)         contents.
77) 
78)     """
79) 
80)     def __init__(

81)         self,
82)         *,
83)         phrase: bytes | bytearray | str = b'',
84)         length: int = 20,
85)         repeat: int = 0,
86)         lower: int | None = None,
87)         upper: int | None = None,
88)         number: int | None = None,
89)         space: int | None = None,
90)         dash: int | None = None,

91)         symbol: int | None = None,
92)     ) -> None:
93)         """Initialize the Vault object.
94) 
95)         Args:
96)             phrase:
97)                 The master passphrase from which to derive the service

98)                 passphrases.  If a text string, then the byte
99)                 representation must be unique.

100)             length:
101)                 Desired passphrase length.
102)             repeat:
103)                 The maximum number of immediate character repetitions
104)                 allowed in the passphrase.  Disabled if set to 0.
105)             lower:

106)                 Optional constraint on ASCII lowercase characters.  If

107)                 positive, include this many lowercase characters
108)                 somewhere in the passphrase.  If 0, avoid lowercase
109)                 characters altogether.
110)             upper:

111)                 Same as `lower`, but for ASCII uppercase characters.

112)             number:
113)                 Same as `lower`, but for ASCII digits.
114)             space:
115)                 Same as `lower`, but for the space character.
116)             dash:
117)                 Same as `lower`, but for the hyphen-minus and underscore
118)                 characters.
119)             symbol:
120)                 Same as `lower`, but for all other hitherto unlisted
121)                 ASCII printable characters (except backquote).
122) 

123)         Raises:
124)             AmbiguousByteRepresentationError:
125)                 The phrase is a text string with differing NFC- and
126)                 NFD-normalized UTF-8 byte representations.
127) 

128)         """

129)         self._phrase = self._get_binary_string(phrase)

130)         self._length = length
131)         self._repeat = repeat
132)         self._allowed = bytearray(self._CHARSETS['all'])
133)         self._required: list[bytes] = []

134) 

135)         def subtract_or_require(
136)             count: int | None, characters: bytes | bytearray
137)         ) -> None:
138)             if not isinstance(count, int):
139)                 return

140)             if count <= 0:

141)                 self._allowed = self._subtract(characters, self._allowed)
142)             else:
143)                 for _ in range(count):
144)                     self._required.append(characters)

145) 

146)         subtract_or_require(lower, self._CHARSETS['lower'])
147)         subtract_or_require(upper, self._CHARSETS['upper'])
148)         subtract_or_require(number, self._CHARSETS['number'])
149)         subtract_or_require(space, self._CHARSETS['space'])
150)         subtract_or_require(dash, self._CHARSETS['dash'])
151)         subtract_or_require(symbol, self._CHARSETS['symbol'])

152)         if len(self._required) > self._length:

153)             msg = 'requested passphrase length too short'
154)             raise ValueError(msg)

155)         if not self._allowed:

156)             msg = 'no allowed characters left'
157)             raise ValueError(msg)

158)         for _ in range(len(self._required), self._length):
159)             self._required.append(bytes(self._allowed))
160) 

161)     def _entropy(self) -> float:

162)         """Estimate the passphrase entropy, given the current settings.
163) 
164)         The entropy is the base 2 logarithm of the amount of

165)         possibilities.  We operate directly on the logarithms, and use
166)         sorting and [`math.fsum`][] to keep high accuracy.
167) 
168)         Note:
169)             We actually overestimate the entropy here because of poor
170)             handling of character repetitions.  In the extreme, assuming
171)             that only one character were allowed, then because there is
172)             only one possible string of each given length, the entropy
173)             of that string `s` is always be zero.  However, we calculate
174)             the entropy as `math.log2(math.factorial(len(s)))`, i.e. we
175)             assume the characters at the respective string position are
176)             distinguishable from each other.
177) 
178)         Returns:
179)             A valid (and somewhat close) upper bound to the entropy.

180) 
181)         """
182)         factors: list[int] = []

183)         if not self._required or any(not x for x in self._required):
184)             return float('-inf')

185)         for i, charset in enumerate(self._required):

186)             factors.extend([i + 1, len(charset)])

187)         factors.sort()
188)         return math.fsum(math.log2(f) for f in factors)
189) 
190)     def _estimate_sufficient_hash_length(

191)         self,
192)         safety_factor: float = 2.0,

193)     ) -> int:
194)         """Estimate the sufficient hash length, given the current settings.
195) 
196)         Using the entropy (via `_entropy`) and a safety factor, give an
197)         initial estimate of the length to use for `create_hash` such
198)         that using a `Sequin` with this hash will not exhaust it during
199)         passphrase generation.
200) 
201)         Args:
202)             safety_factor: The safety factor.  Must be at least 1.
203) 
204)         Returns:
205)             The estimated sufficient hash length.
206) 
207)         Warning:
208)             This is a heuristic, not an exact computation; it may
209)             underestimate the true necessary hash length.  It is
210)             intended as a starting point for searching for a sufficient
211)             hash length, usually by doubling the hash length each time
212)             it does not yet prove so.
213) 
214)         """
215)         try:
216)             safety_factor = float(safety_factor)
217)         except TypeError as e:

218)             msg = f'invalid safety factor: not a float: {safety_factor!r}'
219)             raise TypeError(msg) from e

220)         if not math.isfinite(safety_factor) or safety_factor < 1.0:

221)             msg = f'invalid safety factor {safety_factor!r}'
222)             raise ValueError(msg)

223)         # Ensure the bound is strictly positive.
224)         entropy_bound = max(1, self._entropy())
225)         return int(math.ceil(safety_factor * entropy_bound / 8))

226) 

227)     @staticmethod
228)     def _get_binary_string(s: bytes | bytearray | str, /) -> bytes:
229)         """Convert the input string to a read-only, binary string.
230) 
231)         If it is a text string, then test for an unambiguous UTF-8
232)         representation, otherwise abort.  (That is, check whether the
233)         NFC and NFD forms of the string coincide.)
234) 
235)         Args:
236)             s: The string to (check and) convert.
237) 
238)         Returns:
239)             A read-only, binary copy of the string.
240) 
241)         Raises:
242)             AmbiguousByteRepresentationError:
243)                 The text string has differing NFC- and NFD-normalized
244)                 UTF-8 byte representations.
245) 
246)         """
247)         if isinstance(s, str):
248)             norm = unicodedata.normalize
249)             if norm('NFC', s) != norm('NFD', s):

250)                 raise AmbiguousByteRepresentationError

251)             return s.encode('UTF-8')
252)         return bytes(s)
253) 

254)     @classmethod
255)     def create_hash(

256)         cls,
257)         phrase: bytes | bytearray | str,
258)         service: bytes | bytearray,
259)         *,
260)         length: int = 32,

261)     ) -> bytes:

262)         r"""Create a pseudorandom byte stream from phrase and service.
263) 
264)         Create a pseudorandom byte stream from `phrase` and `service` by
265)         feeding them into the key-derivation function PBKDF2
266)         (8 iterations, using SHA-1).

267) 
268)         Args:

269)             phrase:
270)                 A master passphrase, or sometimes an SSH signature.
271)                 Used as the key for PBKDF2, the underlying cryptographic
272)                 primitive.

273) 
274)                 If a text string, then the byte representation must be
275)                 unique.

276)             service:
277)                 A vault service name.  Will be suffixed with
278)                 `Vault._UUID`, and then used as the salt value for
279)                 PBKDF2.

280)             length:
281)                 The length of the byte stream to generate.
282) 
283)         Returns:
284)             A pseudorandom byte string of length `length`.
285) 

286)         Raises:
287)             AmbiguousByteRepresentationError:
288)                 The phrase is a text string with differing NFC- and
289)                 NFD-normalized UTF-8 byte representations.
290) 

291)         Note:
292)             Shorter values returned from this method (with the same key
293)             and message) are prefixes of longer values returned from
294)             this method.  (This property is inherited from the
295)             underlying PBKDF2 function.)  It is thus safe (if slow) to
296)             call this method with the same input with ever-increasing
297)             target lengths.
298) 

299)         Examples:

300)             >>> # See also Vault.phrase_from_key examples.

301)             >>> phrase = bytes.fromhex('''
302)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
303)             ... 00 00 00 40
304)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
305)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
306)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
307)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
308)             ... ''')
309)             >>> Vault.create_hash(phrase, b'some_service', length=4)
310)             b'M\xb1<S'
311)             >>> Vault.create_hash(phrase, b'some_service', length=16)
312)             b'M\xb1<S\x827E\xd1M\xaf\xf8~\xc8n\x10\xcc'
313)             >>> Vault.create_hash(phrase, b'NOSUCHSERVICE', length=16)
314)             b'\x1c\xc3\x9c\xd9\xb6\x1a\x99CS\x07\xc41\xf4\x85#s'
315) 

316)         """

317)         phrase = cls._get_binary_string(phrase)
318)         assert not isinstance(phrase, str)

319)         salt = bytes(service) + cls._UUID

320)         return hashlib.pbkdf2_hmac(
321)             hash_name='sha1',
322)             password=phrase,
323)             salt=salt,
324)             iterations=8,
325)             dklen=length,
326)         )

327) 
328)     def generate(

329)         self,
330)         service_name: str | bytes | bytearray,
331)         /,
332)         *,

333)         phrase: bytes | bytearray | str = b'',

334)     ) -> bytes:
335)         r"""Generate a service passphrase.

336) 
337)         Args:
338)             service_name:
339)                 The service name.
340)             phrase:
341)                 If given, override the passphrase given during
342)                 construction.
343) 

344)                 If a text string, then the byte representation must be
345)                 unique.
346) 
347)         Returns:
348)             The service passphrase.
349) 
350)         Raises:
351)             AmbiguousByteRepresentationError:
352)                 The phrase is a text string with differing NFC- and
353)                 NFD-normalized UTF-8 byte representations.
354) 

355)         Examples:
356)             >>> phrase = b'She cells C shells bye the sea shoars'
357)             >>> # Using default options in constructor.
358)             >>> Vault(phrase=phrase).generate(b'google')
359)             b': 4TVH#5:aZl8LueOT\\{'
360)             >>> # Also possible:
361)             >>> Vault().generate(b'google', phrase=phrase)
362)             b': 4TVH#5:aZl8LueOT\\{'
363) 

364)         """

365)         hash_length = self._estimate_sufficient_hash_length()
366)         assert hash_length >= 1

367)         # Ensure the phrase is a bytes object.  Needed later for safe
368)         # concatenation.
369)         if isinstance(service_name, str):
370)             service_name = service_name.encode('utf-8')
371)         elif not isinstance(service_name, bytes):
372)             service_name = bytes(service_name)
373)         assert_type(service_name, bytes)

374)         if not phrase:
375)             phrase = self._phrase

376)         phrase = self._get_binary_string(phrase)

377)         # Repeat the passphrase generation with ever-increasing hash
378)         # lengths, until the passphrase can be formed without exhausting
379)         # the sequin.  See the guarantee in the create_hash method for
380)         # why this works.
381)         while True:
382)             try:
383)                 required = self._required[:]

384)                 seq = sequin.Sequin(
385)                     self.create_hash(
386)                         phrase=phrase, service=service_name, length=hash_length
387)                     )
388)                 )

389)                 result = bytearray()
390)                 while len(result) < self._length:
391)                     pos = seq.generate(len(required))
392)                     charset = required.pop(pos)
393)                     # Determine if an unlucky choice right now might
394)                     # violate the restriction on repeated characters.
395)                     # That is, check if the current partial passphrase
396)                     # ends with r - 1 copies of the same character
397)                     # (where r is the repeat limit that must not be
398)                     # reached), and if so, remove this same character
399)                     # from the current character's allowed set.

400)                     if self._repeat and result:
401)                         bad_suffix = bytes(result[-1:]) * (self._repeat - 1)
402)                         if result.endswith(bad_suffix):

403)                             charset = self._subtract(
404)                                 bytes(result[-1:]), charset
405)                             )

406)                     pos = seq.generate(len(charset))

407)                     result.extend(charset[pos : pos + 1])

408)             except sequin.SequinExhaustedError:

409)                 hash_length *= 2
410)             else:

411)                 return bytes(result)

412) 

413)     @staticmethod
414)     def _is_suitable_ssh_key(key: bytes | bytearray, /) -> bool:
415)         """Check whether the key is suitable for passphrase derivation.
416) 
417)         Currently, this only checks whether signatures with this key
418)         type are deterministic.
419) 
420)         Args:
421)             key: SSH public key to check.
422) 
423)         Returns:
424)             True if and only if the key is suitable for use in deriving
425)             a passphrase deterministically.
426) 
427)         """
428)         deterministic_signature_types = {

429)             'ssh-ed25519': lambda k: k.startswith(
430)                 b'\x00\x00\x00\x0bssh-ed25519'
431)             ),
432)             'ssh-ed448': lambda k: k.startswith(b'\x00\x00\x00\x09ssh-ed448'),
433)             'ssh-rsa': lambda k: k.startswith(b'\x00\x00\x00\x07ssh-rsa'),

434)         }
435)         return any(v(key) for v in deterministic_signature_types.values())
436) 

437)     @classmethod

438)     def phrase_from_key(cls, key: bytes | bytearray, /) -> bytes:

439)         """Obtain the master passphrase from a configured SSH key.
440) 
441)         vault allows the usage of certain SSH keys to derive a master
442)         passphrase, by signing the vault UUID with the SSH key.  The key
443)         type must ensure that signatures are deterministic.
444) 
445)         Args:
446)             key: The (public) SSH key to use for signing.
447) 
448)         Returns:

449)             The signature of the vault UUID under this key, unframed but
450)             encoded in base64.

451) 
452)         Raises:
453)             ValueError:
454)                 The SSH key is principally unsuitable for this use case.
455)                 Usually this means that the signature is not
456)                 deterministic.
457) 

458)         Examples:

459)             >>> import base64
460)             >>> # Actual Ed25519 test public key.

461)             >>> public_key = bytes.fromhex('''
462)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
463)             ... 00 00 00 20
464)             ... 81 78 81 68 26 d6 02 48 5f 0f ff 32 48 6f e4 c1
465)             ... 30 89 dc 1c 6a 45 06 09 e9 09 0f fb c2 12 69 76
466)             ... ''')

467)             >>> expected_sig_raw = bytes.fromhex('''

468)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
469)             ... 00 00 00 40
470)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
471)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
472)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
473)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
474)             ... ''')

475)             >>> # Raw Ed25519 signatures are 64 bytes long.
476)             >>> signature_blob = expected_sig_raw[-64:]
477)             >>> phrase = base64.standard_b64encode(signature_blob)
478)             >>> Vault.phrase_from_key(phrase) == expected  # doctest:+SKIP

479)             True
480) 

481)         """

482)         if not cls._is_suitable_ssh_key(key):

483)             msg = (
484)                 'unsuitable SSH key: bad key, or '
485)                 'signature not deterministic'
486)             )

487)             raise ValueError(msg)

488)         with ssh_agent_client.SSHAgentClient() as client:

489)             raw_sig = client.sign(key, cls._UUID)

490)         _keytype, trailer = client.unstring_prefix(raw_sig)

491)         signature_blob = client.unstring(trailer)
492)         return bytes(base64.standard_b64encode(signature_blob))

493) 

494)     @staticmethod

495)     def _subtract(

496)         charset: bytes | bytearray,
497)         allowed: bytes | bytearray,

498)     ) -> bytearray:
499)         """Remove the characters in charset from allowed.
500) 
501)         This preserves the relative order of characters in `allowed`.
502) 
503)         Args:

504)             charset:
505)                 Characters to remove.  Must not contain duplicate
506)                 characters.
507)             allowed:
508)                 Character set to remove the other characters from.  Must
509)                 not contain duplicate characters.

510) 
511)         Returns:

512)             The pruned "allowed" character set.

513) 
514)         Raises:

515)             ValueError:
516)                 `allowed` or `charset` contained duplicate characters.

517) 
518)         """

519)         allowed = (
520)             allowed if isinstance(allowed, bytearray) else bytearray(allowed)
521)         )

522)         assert_type(allowed, bytearray)

523)         msg_dup_characters = 'duplicate characters in set'

524)         if len(frozenset(allowed)) != len(allowed):

525)             raise ValueError(msg_dup_characters)

526)         if len(frozenset(charset)) != len(charset):

527)             raise ValueError(msg_dup_characters)

528)         for c in charset:
529)             try:
530)                 pos = allowed.index(c)

531)             except ValueError:

532)                 pass
533)             else:

534)                 allowed[pos : pos + 1] = []