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
6) 

7) """  # noqa: RUF002

8) 
9) from __future__ import annotations
10) 

11) import base64

12) import collections
13) import hashlib
14) import math

15) import unicodedata

16) 

17) from typing_extensions import assert_type

18) 
19) import sequin
20) import ssh_agent_client
21) 

22) __author__ = "Marco Ricci <m@the13thletter.info>"

23) __version__ = "0.1.1"

24) 

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

27)     def __init__(self):
28)         super().__init__('text string has ambiguous byte representation')
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'] = (_CHARSETS['alphanum'] + _CHARSETS['space']
44)                     + _CHARSETS['symbol'])

45) 

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

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

62) 
63)     [vault]: https://getvau.lt
64)     [ALGORITHM]: https://blog.jcoglan.com/2012/07/16/designing-vaults-generator-algorithm/
65) 
66)     """
67)     _UUID = b'e87eb0f4-34cb-46b9-93ad-766c5ab063e7'
68)     """A tag used by vault in the bit stream generation."""

69)     _CHARSETS = _CHARSETS

70)     """
71)         Known character sets from which to draw passphrase characters.
72)         Relies on a certain, fixed order for their definition and their
73)         contents.
74) 
75)     """
76) 
77)     def __init__(

78)         self, *, phrase: bytes | bytearray | str = b'',
79)         length: int = 20, repeat: int = 0, lower: int | None = None,

80)         upper: int | None = None, number: int | None = None,
81)         space: int | None = None, dash: int | None = None,
82)         symbol: int | None = None,
83)     ) -> None:
84)         """Initialize the Vault object.
85) 
86)         Args:
87)             phrase:
88)                 The master passphrase from which to derive the service

89)                 passphrases.  If a text string, then the byte
90)                 representation must be unique.

91)             length:
92)                 Desired passphrase length.
93)             repeat:
94)                 The maximum number of immediate character repetitions
95)                 allowed in the passphrase.  Disabled if set to 0.
96)             lower:

97)                 Optional constraint on ASCII lowercase characters.  If

98)                 positive, include this many lowercase characters
99)                 somewhere in the passphrase.  If 0, avoid lowercase
100)                 characters altogether.
101)             upper:

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

103)             number:
104)                 Same as `lower`, but for ASCII digits.
105)             space:
106)                 Same as `lower`, but for the space character.
107)             dash:
108)                 Same as `lower`, but for the hyphen-minus and underscore
109)                 characters.
110)             symbol:
111)                 Same as `lower`, but for all other hitherto unlisted
112)                 ASCII printable characters (except backquote).
113) 

114)         Raises:
115)             AmbiguousByteRepresentationError:
116)                 The phrase is a text string with differing NFC- and
117)                 NFD-normalized UTF-8 byte representations.
118) 

119)         """

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

121)         self._length = length
122)         self._repeat = repeat
123)         self._allowed = bytearray(self._CHARSETS['all'])
124)         self._required: list[bytes] = []
125)         def subtract_or_require(
126)             count: int | None, characters: bytes | bytearray
127)         ) -> None:
128)             if not isinstance(count, int):
129)                 return

130)             if count <= 0:

131)                 self._allowed = self._subtract(characters, self._allowed)
132)             else:
133)                 for _ in range(count):
134)                     self._required.append(characters)
135)         subtract_or_require(lower, self._CHARSETS['lower'])
136)         subtract_or_require(upper, self._CHARSETS['upper'])
137)         subtract_or_require(number, self._CHARSETS['number'])
138)         subtract_or_require(space, self._CHARSETS['space'])
139)         subtract_or_require(dash, self._CHARSETS['dash'])
140)         subtract_or_require(symbol, self._CHARSETS['symbol'])

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

142)             msg = 'requested passphrase length too short'
143)             raise ValueError(msg)

144)         if not self._allowed:

145)             msg = 'no allowed characters left'
146)             raise ValueError(msg)

147)         for _ in range(len(self._required), self._length):
148)             self._required.append(bytes(self._allowed))
149) 

150)     def _entropy(self) -> float:

151)         """Estimate the passphrase entropy, given the current settings.
152) 
153)         The entropy is the base 2 logarithm of the amount of

154)         possibilities.  We operate directly on the logarithms, and use
155)         sorting and [`math.fsum`][] to keep high accuracy.
156) 
157)         Note:
158)             We actually overestimate the entropy here because of poor
159)             handling of character repetitions.  In the extreme, assuming
160)             that only one character were allowed, then because there is
161)             only one possible string of each given length, the entropy
162)             of that string `s` is always be zero.  However, we calculate
163)             the entropy as `math.log2(math.factorial(len(s)))`, i.e. we
164)             assume the characters at the respective string position are
165)             distinguishable from each other.
166) 
167)         Returns:
168)             A valid (and somewhat close) upper bound to the entropy.

169) 
170)         """
171)         factors: list[int] = []

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

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

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

176)         factors.sort()
177)         return math.fsum(math.log2(f) for f in factors)
178) 
179)     def _estimate_sufficient_hash_length(
180)         self, safety_factor: float = 2.0,
181)     ) -> int:
182)         """Estimate the sufficient hash length, given the current settings.
183) 
184)         Using the entropy (via `_entropy`) and a safety factor, give an
185)         initial estimate of the length to use for `create_hash` such
186)         that using a `Sequin` with this hash will not exhaust it during
187)         passphrase generation.
188) 
189)         Args:
190)             safety_factor: The safety factor.  Must be at least 1.
191) 
192)         Returns:
193)             The estimated sufficient hash length.
194) 
195)         Warning:
196)             This is a heuristic, not an exact computation; it may
197)             underestimate the true necessary hash length.  It is
198)             intended as a starting point for searching for a sufficient
199)             hash length, usually by doubling the hash length each time
200)             it does not yet prove so.
201) 
202)         """
203)         try:
204)             safety_factor = float(safety_factor)
205)         except TypeError as e:

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

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

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

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

214) 

215)     @staticmethod
216)     def _get_binary_string(s: bytes | bytearray | str, /) -> bytes:
217)         """Convert the input string to a read-only, binary string.
218) 
219)         If it is a text string, then test for an unambiguous UTF-8
220)         representation, otherwise abort.  (That is, check whether the
221)         NFC and NFD forms of the string coincide.)
222) 
223)         Args:
224)             s: The string to (check and) convert.
225) 
226)         Returns:
227)             A read-only, binary copy of the string.
228) 
229)         Raises:
230)             AmbiguousByteRepresentationError:
231)                 The text string has differing NFC- and NFD-normalized
232)                 UTF-8 byte representations.
233) 
234)         """
235)         if isinstance(s, str):
236)             norm = unicodedata.normalize
237)             if norm('NFC', s) != norm('NFD', s):

238)                 raise AmbiguousByteRepresentationError

239)             return s.encode('UTF-8')
240)         return bytes(s)
241) 

242)     @classmethod
243)     def create_hash(

244)         cls, phrase: bytes | bytearray | str,
245)         service: bytes | bytearray, *, length: int = 32,

246)     ) -> bytes:

247)         r"""Create a pseudorandom byte stream from phrase and service.
248) 
249)         Create a pseudorandom byte stream from `phrase` and `service` by
250)         feeding them into the key-derivation function PBKDF2
251)         (8 iterations, using SHA-1).

252) 
253)         Args:

254)             phrase:
255)                 A master passphrase, or sometimes an SSH signature.
256)                 Used as the key for PBKDF2, the underlying cryptographic
257)                 primitive.

258) 
259)                 If a text string, then the byte representation must be
260)                 unique.

261)             service:
262)                 A vault service name.  Will be suffixed with
263)                 `Vault._UUID`, and then used as the salt value for
264)                 PBKDF2.

265)             length:
266)                 The length of the byte stream to generate.
267) 
268)         Returns:
269)             A pseudorandom byte string of length `length`.
270) 

271)         Raises:
272)             AmbiguousByteRepresentationError:
273)                 The phrase is a text string with differing NFC- and
274)                 NFD-normalized UTF-8 byte representations.
275) 

276)         Note:
277)             Shorter values returned from this method (with the same key
278)             and message) are prefixes of longer values returned from
279)             this method.  (This property is inherited from the
280)             underlying PBKDF2 function.)  It is thus safe (if slow) to
281)             call this method with the same input with ever-increasing
282)             target lengths.
283) 

284)         Examples:

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

286)             >>> phrase = bytes.fromhex('''
287)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
288)             ... 00 00 00 40
289)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
290)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
291)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
292)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
293)             ... ''')
294)             >>> Vault.create_hash(phrase, b'some_service', length=4)
295)             b'M\xb1<S'
296)             >>> Vault.create_hash(phrase, b'some_service', length=16)
297)             b'M\xb1<S\x827E\xd1M\xaf\xf8~\xc8n\x10\xcc'
298)             >>> Vault.create_hash(phrase, b'NOSUCHSERVICE', length=16)
299)             b'\x1c\xc3\x9c\xd9\xb6\x1a\x99CS\x07\xc41\xf4\x85#s'
300) 

301)         """

302)         phrase = cls._get_binary_string(phrase)
303)         assert not isinstance(phrase, str)

304)         salt = bytes(service) + cls._UUID
305)         return hashlib.pbkdf2_hmac(hash_name='sha1', password=phrase,
306)                                    salt=salt, iterations=8, dklen=length)

307) 
308)     def generate(

309)         self, service_name: str | bytes | bytearray, /, *,

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

311)     ) -> bytes:
312)         r"""Generate a service passphrase.

313) 
314)         Args:
315)             service_name:
316)                 The service name.
317)             phrase:
318)                 If given, override the passphrase given during
319)                 construction.
320) 

321)                 If a text string, then the byte representation must be
322)                 unique.
323) 
324)         Returns:
325)             The service passphrase.
326) 
327)         Raises:
328)             AmbiguousByteRepresentationError:
329)                 The phrase is a text string with differing NFC- and
330)                 NFD-normalized UTF-8 byte representations.
331) 

332)         Examples:
333)             >>> phrase = b'She cells C shells bye the sea shoars'
334)             >>> # Using default options in constructor.
335)             >>> Vault(phrase=phrase).generate(b'google')
336)             b': 4TVH#5:aZl8LueOT\\{'
337)             >>> # Also possible:
338)             >>> Vault().generate(b'google', phrase=phrase)
339)             b': 4TVH#5:aZl8LueOT\\{'
340) 

341)         """

342)         hash_length = self._estimate_sufficient_hash_length()
343)         assert hash_length >= 1

344)         # Ensure the phrase is a bytes object.  Needed later for safe
345)         # concatenation.
346)         if isinstance(service_name, str):
347)             service_name = service_name.encode('utf-8')
348)         elif not isinstance(service_name, bytes):
349)             service_name = bytes(service_name)
350)         assert_type(service_name, bytes)

351)         if not phrase:
352)             phrase = self._phrase

353)         phrase = self._get_binary_string(phrase)

354)         # Repeat the passphrase generation with ever-increasing hash
355)         # lengths, until the passphrase can be formed without exhausting
356)         # the sequin.  See the guarantee in the create_hash method for
357)         # why this works.
358)         while True:
359)             try:
360)                 required = self._required[:]
361)                 seq = sequin.Sequin(self.create_hash(

362)                     phrase=phrase, service=service_name, length=hash_length))

363)                 result = bytearray()
364)                 while len(result) < self._length:
365)                     pos = seq.generate(len(required))
366)                     charset = required.pop(pos)
367)                     # Determine if an unlucky choice right now might
368)                     # violate the restriction on repeated characters.
369)                     # That is, check if the current partial passphrase
370)                     # ends with r - 1 copies of the same character
371)                     # (where r is the repeat limit that must not be
372)                     # reached), and if so, remove this same character
373)                     # from the current character's allowed set.

374)                     if self._repeat and result:
375)                         bad_suffix = bytes(result[-1:]) * (self._repeat - 1)
376)                         if result.endswith(bad_suffix):
377)                             charset = self._subtract(bytes(result[-1:]),
378)                                                      charset)

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

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

381)             except sequin.SequinExhaustedError:

382)                 hash_length *= 2
383)             else:

384)                 return bytes(result)

385) 

386)     @staticmethod
387)     def _is_suitable_ssh_key(key: bytes | bytearray, /) -> bool:
388)         """Check whether the key is suitable for passphrase derivation.
389) 
390)         Currently, this only checks whether signatures with this key
391)         type are deterministic.
392) 
393)         Args:
394)             key: SSH public key to check.
395) 
396)         Returns:
397)             True if and only if the key is suitable for use in deriving
398)             a passphrase deterministically.
399) 
400)         """
401)         deterministic_signature_types = {
402)             'ssh-ed25519':
403)                 lambda k: k.startswith(b'\x00\x00\x00\x0bssh-ed25519'),
404)             'ssh-ed448':
405)                 lambda k: k.startswith(b'\x00\x00\x00\x09ssh-ed448'),
406)             'ssh-rsa':
407)                 lambda k: k.startswith(b'\x00\x00\x00\x07ssh-rsa'),
408)         }
409)         return any(v(key) for v in deterministic_signature_types.values())
410) 

411)     @classmethod

412)     def phrase_from_key(

413)         cls, key: bytes | bytearray, /

414)     ) -> bytes:

415)         """Obtain the master passphrase from a configured SSH key.
416) 
417)         vault allows the usage of certain SSH keys to derive a master
418)         passphrase, by signing the vault UUID with the SSH key.  The key
419)         type must ensure that signatures are deterministic.
420) 
421)         Args:
422)             key: The (public) SSH key to use for signing.
423) 
424)         Returns:

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

427) 
428)         Raises:
429)             ValueError:
430)                 The SSH key is principally unsuitable for this use case.
431)                 Usually this means that the signature is not
432)                 deterministic.
433) 

434)         Examples:

435)             >>> import base64
436)             >>> # Actual Ed25519 test public key.

437)             >>> public_key = bytes.fromhex('''
438)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
439)             ... 00 00 00 20
440)             ... 81 78 81 68 26 d6 02 48 5f 0f ff 32 48 6f e4 c1
441)             ... 30 89 dc 1c 6a 45 06 09 e9 09 0f fb c2 12 69 76
442)             ... ''')

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

444)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
445)             ... 00 00 00 40
446)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
447)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
448)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
449)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
450)             ... ''')

451)             >>> # Raw Ed25519 signatures are 64 bytes long.
452)             >>> signature_blob = expected_sig_raw[-64:]
453)             >>> phrase = base64.standard_b64encode(signature_blob)
454)             >>> Vault.phrase_from_key(phrase) == expected  # doctest:+SKIP

455)             True
456) 

457)         """

458)         if not cls._is_suitable_ssh_key(key):

459)             msg = ('unsuitable SSH key: bad key, or '
460)                    'signature not deterministic')
461)             raise ValueError(msg)

462)         with ssh_agent_client.SSHAgentClient() as client:

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

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

465)         signature_blob = client.unstring(trailer)
466)         return bytes(base64.standard_b64encode(signature_blob))

467) 

468)     @staticmethod

469)     def _subtract(

470)         charset: bytes | bytearray, allowed: bytes | bytearray,

471)     ) -> bytearray:
472)         """Remove the characters in charset from allowed.
473) 
474)         This preserves the relative order of characters in `allowed`.
475) 
476)         Args:

477)             charset:
478)                 Characters to remove.  Must not contain duplicate
479)                 characters.
480)             allowed:
481)                 Character set to remove the other characters from.  Must
482)                 not contain duplicate characters.

483) 
484)         Returns:

485)             The pruned "allowed" character set.

486) 
487)         Raises:

488)             ValueError:
489)                 `allowed` or `charset` contained duplicate characters.

490) 
491)         """
492)         allowed = (allowed if isinstance(allowed, bytearray)
493)                    else bytearray(allowed))
494)         assert_type(allowed, bytearray)

495)         msg_dup_characters = 'duplicate characters in set'

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

497)             raise ValueError(msg_dup_characters)

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

499)             raise ValueError(msg_dup_characters)

500)         for c in charset:
501)             try:
502)                 pos = allowed.index(c)

503)             except ValueError:

504)                 pass
505)             else:

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