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) """
8) 
9) from __future__ import annotations
10) 

11) import base64

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

15) import unicodedata

16) import warnings
17) 
18) from typing import assert_type, reveal_type
19) 
20) import sequin
21) import ssh_agent_client
22) 

23) __author__ = "Marco Ricci <m@the13thletter.info>"
24) __version__ = "0.1.0"
25) 

26) class AmbiguousByteRepresentationError(ValueError):
27)     """The object has an ambiguous byte representation."""
28) 

29) class Vault:
30)     """A work-alike of James Coglan's vault.
31) 
32)     Store settings for generating (actually: deriving) passphrases for
33)     named services, with various constraints, given only a master
34)     passphrase.  Also, actually generate the passphrase.  The derivation
35)     is deterministic and non-secret; only the master passphrase need be
36)     kept secret.  The implementation is compatible with [vault][].
37) 
38)     [James Coglan explains the passphrase derivation algorithm in great
39)     detail][ALGORITHM] in his blog post on said topic: A principally
40)     infinite bit stream is obtained by running a key-derivation function
41)     on the master passphrase and the service name, then this bit stream

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

45) 
46)     [vault]: https://getvau.lt
47)     [ALGORITHM]: https://blog.jcoglan.com/2012/07/16/designing-vaults-generator-algorithm/
48) 
49)     """
50)     _UUID = b'e87eb0f4-34cb-46b9-93ad-766c5ab063e7'
51)     """A tag used by vault in the bit stream generation."""
52)     _CHARSETS: collections.OrderedDict[str, bytes]
53)     """
54)         Known character sets from which to draw passphrase characters.
55)         Relies on a certain, fixed order for their definition and their
56)         contents.
57) 
58)     """
59)     _CHARSETS = collections.OrderedDict([
60)         ('lower', b'abcdefghijklmnopqrstuvwxyz'),
61)         ('upper', b'ABCDEFGHIJKLMNOPQRSTUVWXYZ'),
62)         ('alpha', b''),  # Placeholder.
63)         ('number', b'0123456789'),
64)         ('alphanum', b''),  # Placeholder.
65)         ('space', b' '),
66)         ('dash', b'-_'),
67)         ('symbol', b'!"#$%&\'()*+,./:;<=>?@[\\]^{|}~-_'),
68)         ('all', b''),  # Placeholder.
69)     ])
70)     _CHARSETS['alpha'] = _CHARSETS['lower'] + _CHARSETS['upper']
71)     _CHARSETS['alphanum'] = _CHARSETS['alpha'] + _CHARSETS['number']
72)     _CHARSETS['all'] = (_CHARSETS['alphanum'] + _CHARSETS['space']
73)                         + _CHARSETS['symbol'])
74) 
75)     def __init__(

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

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

87)                 passphrases.  If a text string, then the byte
88)                 representation must be unique.

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

95)                 Optional constraint on ASCII lowercase characters.  If

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

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

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

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

117)         """

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

119)         self._length = length
120)         self._repeat = repeat
121)         self._allowed = bytearray(self._CHARSETS['all'])
122)         self._required: list[bytes] = []
123)         def subtract_or_require(
124)             count: int | None, characters: bytes | bytearray
125)         ) -> None:
126)             if not isinstance(count, int):
127)                 return
128)             elif count <= 0:
129)                 self._allowed = self._subtract(characters, self._allowed)
130)             else:
131)                 for _ in range(count):
132)                     self._required.append(characters)
133)         subtract_or_require(lower, self._CHARSETS['lower'])
134)         subtract_or_require(upper, self._CHARSETS['upper'])
135)         subtract_or_require(number, self._CHARSETS['number'])
136)         subtract_or_require(space, self._CHARSETS['space'])
137)         subtract_or_require(dash, self._CHARSETS['dash'])
138)         subtract_or_require(symbol, self._CHARSETS['symbol'])

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

140)             raise ValueError('requested passphrase length too short')
141)         if not self._allowed:
142)             raise ValueError('no allowed characters left')
143)         for _ in range(len(self._required), self._length):
144)             self._required.append(bytes(self._allowed))
145) 

146)     def _entropy(self) -> float:

147)         """Estimate the passphrase entropy, given the current settings.
148) 
149)         The entropy is the base 2 logarithm of the amount of

150)         possibilities.  We operate directly on the logarithms, and use
151)         sorting and [`math.fsum`][] to keep high accuracy.
152) 
153)         Note:
154)             We actually overestimate the entropy here because of poor
155)             handling of character repetitions.  In the extreme, assuming
156)             that only one character were allowed, then because there is
157)             only one possible string of each given length, the entropy
158)             of that string `s` is always be zero.  However, we calculate
159)             the entropy as `math.log2(math.factorial(len(s)))`, i.e. we
160)             assume the characters at the respective string position are
161)             distinguishable from each other.
162) 
163)         Returns:
164)             A valid (and somewhat close) upper bound to the entropy.

165) 
166)         """
167)         factors: list[int] = []

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

170)         for i, charset in enumerate(self._required):
171)             factors.append(i + 1)
172)             factors.append(len(charset))

173)         factors.sort()
174)         return math.fsum(math.log2(f) for f in factors)
175) 
176)     def _estimate_sufficient_hash_length(
177)         self, safety_factor: float = 2.0,
178)     ) -> int:
179)         """Estimate the sufficient hash length, given the current settings.
180) 
181)         Using the entropy (via `_entropy`) and a safety factor, give an
182)         initial estimate of the length to use for `create_hash` such
183)         that using a `Sequin` with this hash will not exhaust it during
184)         passphrase generation.
185) 
186)         Args:
187)             safety_factor: The safety factor.  Must be at least 1.
188) 
189)         Returns:
190)             The estimated sufficient hash length.
191) 
192)         Warning:
193)             This is a heuristic, not an exact computation; it may
194)             underestimate the true necessary hash length.  It is
195)             intended as a starting point for searching for a sufficient
196)             hash length, usually by doubling the hash length each time
197)             it does not yet prove so.
198) 
199)         """
200)         try:
201)             safety_factor = float(safety_factor)
202)         except TypeError as e:
203)             raise TypeError(f'invalid safety factor: not a float: '
204)                             f'{safety_factor!r}') from e
205)         if not math.isfinite(safety_factor) or safety_factor < 1.0:
206)             raise ValueError(f'invalid safety factor {safety_factor!r}')
207)         # Ensure the bound is strictly positive.
208)         entropy_bound = max(1, self._entropy())
209)         return int(math.ceil(safety_factor * entropy_bound / 8))

210) 

211)     @staticmethod
212)     def _get_binary_string(s: bytes | bytearray | str, /) -> bytes:
213)         """Convert the input string to a read-only, binary string.
214) 
215)         If it is a text string, then test for an unambiguous UTF-8
216)         representation, otherwise abort.  (That is, check whether the
217)         NFC and NFD forms of the string coincide.)
218) 
219)         Args:
220)             s: The string to (check and) convert.
221) 
222)         Returns:
223)             A read-only, binary copy of the string.
224) 
225)         Raises:
226)             AmbiguousByteRepresentationError:
227)                 The text string has differing NFC- and NFD-normalized
228)                 UTF-8 byte representations.
229) 
230)         """
231)         if isinstance(s, str):
232)             norm = unicodedata.normalize
233)             if norm('NFC', s) != norm('NFD', s):
234)                 raise AmbiguousByteRepresentationError(
235)                     'text string has ambiguous byte representation')
236)             return s.encode('UTF-8')
237)         return bytes(s)
238) 

239)     @classmethod
240)     def create_hash(

241)         cls, phrase: bytes | bytearray | str,
242)         service: bytes | bytearray, *, length: int = 32,

243)     ) -> bytes:

244)         r"""Create a pseudorandom byte stream from phrase and service.
245) 
246)         Create a pseudorandom byte stream from `phrase` and `service` by
247)         feeding them into the key-derivation function PBKDF2
248)         (8 iterations, using SHA-1).

249) 
250)         Args:

251)             phrase:
252)                 A master passphrase, or sometimes an SSH signature.
253)                 Used as the key for PBKDF2, the underlying cryptographic
254)                 primitive.

255) 
256)                 If a text string, then the byte representation must be
257)                 unique.

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

262)             length:
263)                 The length of the byte stream to generate.
264) 
265)         Returns:
266)             A pseudorandom byte string of length `length`.
267) 

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

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

281)         Examples:

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

283)             >>> phrase = bytes.fromhex('''
284)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
285)             ... 00 00 00 40
286)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
287)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
288)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
289)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
290)             ... ''')
291)             >>> Vault.create_hash(phrase, b'some_service', length=4)
292)             b'M\xb1<S'
293)             >>> Vault.create_hash(phrase, b'some_service', length=16)
294)             b'M\xb1<S\x827E\xd1M\xaf\xf8~\xc8n\x10\xcc'
295)             >>> Vault.create_hash(phrase, b'NOSUCHSERVICE', length=16)
296)             b'\x1c\xc3\x9c\xd9\xb6\x1a\x99CS\x07\xc41\xf4\x85#s'
297) 

298)         """

299)         phrase = cls._get_binary_string(phrase)
300)         assert not isinstance(phrase, str)

301)         salt = bytes(service) + cls._UUID
302)         return hashlib.pbkdf2_hmac(hash_name='sha1', password=phrase,
303)                                    salt=salt, iterations=8, dklen=length)

304) 
305)     def generate(

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

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

308)     ) -> bytes:
309)         r"""Generate a service passphrase.

310) 
311)         Args:
312)             service_name:
313)                 The service name.
314)             phrase:
315)                 If given, override the passphrase given during
316)                 construction.
317) 

318)                 If a text string, then the byte representation must be
319)                 unique.
320) 
321)         Returns:
322)             The service passphrase.
323) 
324)         Raises:
325)             AmbiguousByteRepresentationError:
326)                 The phrase is a text string with differing NFC- and
327)                 NFD-normalized UTF-8 byte representations.
328) 

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

338)         """

339)         hash_length = self._estimate_sufficient_hash_length()
340)         assert hash_length >= 1

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

348)         if not phrase:
349)             phrase = self._phrase

350)         phrase = self._get_binary_string(phrase)

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

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

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

371)                     if self._repeat and result:
372)                         bad_suffix = bytes(result[-1:]) * (self._repeat - 1)
373)                         if result.endswith(bad_suffix):
374)                             charset = self._subtract(bytes(result[-1:]),
375)                                                      charset)

376)                     pos = seq.generate(len(charset))
377)                     result.extend(charset[pos:pos+1])

378)             except sequin.SequinExhaustedException:

379)                 hash_length *= 2
380)             else:

381)                 return bytes(result)

382) 

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

408)     @classmethod

409)     def phrase_from_key(

410)         cls, key: bytes | bytearray, /

411)     ) -> bytes:

412)         """Obtain the master passphrase from a configured SSH key.
413) 
414)         vault allows the usage of certain SSH keys to derive a master
415)         passphrase, by signing the vault UUID with the SSH key.  The key
416)         type must ensure that signatures are deterministic.
417) 
418)         Args:
419)             key: The (public) SSH key to use for signing.
420) 
421)         Returns:

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

424) 
425)         Raises:
426)             ValueError:
427)                 The SSH key is principally unsuitable for this use case.
428)                 Usually this means that the signature is not
429)                 deterministic.
430) 

431)         Examples:

432)             >>> import base64
433)             >>> # Actual Ed25519 test public key.

434)             >>> public_key = bytes.fromhex('''
435)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
436)             ... 00 00 00 20
437)             ... 81 78 81 68 26 d6 02 48 5f 0f ff 32 48 6f e4 c1
438)             ... 30 89 dc 1c 6a 45 06 09 e9 09 0f fb c2 12 69 76
439)             ... ''')

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

441)             ... 00 00 00 0b 73 73 68 2d 65 64 32 35 35 31 39
442)             ... 00 00 00 40
443)             ... f0 98 19 80 6c 1a 97 d5 26 03 6e cc e3 65 8f 86
444)             ... 66 07 13 19 13 09 21 33 33 f9 e4 36 53 1d af fd
445)             ... 0d 08 1f ec f8 73 9b 8c 5f 55 39 16 7c 53 54 2c
446)             ... 1e 52 bb 30 ed 7f 89 e2 2f 69 51 55 d8 9e a6 02
447)             ... ''')

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

452)             True
453) 

454)         """

455)         if not cls._is_suitable_ssh_key(key):

456)             raise ValueError(
457)                 'unsuitable SSH key: bad key, or signature not deterministic')
458)         with ssh_agent_client.SSHAgentClient() as client:

459)             raw_sig = client.sign(key, cls._UUID)
460)         keytype, trailer = client.unstring_prefix(raw_sig)
461)         signature_blob = client.unstring(trailer)
462)         return bytes(base64.standard_b64encode(signature_blob))

463) 

464)     @staticmethod

465)     def _subtract(

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

467)     ) -> bytearray:
468)         """Remove the characters in charset from allowed.
469) 
470)         This preserves the relative order of characters in `allowed`.
471) 
472)         Args:

473)             charset:
474)                 Characters to remove.  Must not contain duplicate
475)                 characters.
476)             allowed:
477)                 Character set to remove the other characters from.  Must
478)                 not contain duplicate characters.

479) 
480)         Returns:

481)             The pruned "allowed" character set.

482) 
483)         Raises:

484)             ValueError:
485)                 `allowed` or `charset` contained duplicate characters.

486) 
487)         """
488)         allowed = (allowed if isinstance(allowed, bytearray)
489)                    else bytearray(allowed))
490)         assert_type(allowed, bytearray)

491)         if len(frozenset(allowed)) != len(allowed):
492)             raise ValueError('duplicate characters in set')

493)         if len(frozenset(charset)) != len(charset):
494)             raise ValueError('duplicate characters in set')
495)         for c in charset:
496)             try:
497)                 pos = allowed.index(c)

498)             except ValueError: