Source code

Revision control

Line Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

/* Utilities for hashing. */

/*
 * This file exports functions for hashing data down to a uint32_t (a.k.a.
 * mozilla::HashNumber), including:
 *
 *  - HashString    Hash a char* or char16_t/wchar_t* of known or unknown
 *                  length.
 *
 *  - HashBytes     Hash a byte array of known length.
 *
 *  - HashGeneric   Hash one or more values.  Currently, we support uint32_t,
 *                  types which can be implicitly cast to uint32_t, data
 *                  pointers, and function pointers.
 *
 *  - AddToHash     Add one or more values to the given hash.  This supports the
 *                  same list of types as HashGeneric.
 *
 *
 * You can chain these functions together to hash complex objects.  For example:
 *
 *  class ComplexObject
 *  {
 *    char* mStr;
 *    uint32_t mUint1, mUint2;
 *    void (*mCallbackFn)();
 *
 *  public:
 *    HashNumber hash()
 *    {
 *      HashNumber hash = HashString(mStr);
 *      hash = AddToHash(hash, mUint1, mUint2);
 *      return AddToHash(hash, mCallbackFn);
 *    }
 *  };
 *
 * If you want to hash an nsAString or nsACString, use the HashString functions
 * in nsHashKeys.h.
 */

#ifndef mozilla_HashFunctions_h
#define mozilla_HashFunctions_h

#include "mozilla/Assertions.h"
#include "mozilla/Attributes.h"
#include "mozilla/Char16.h"
#include "mozilla/MathAlgorithms.h"
#include "mozilla/Types.h"
#include "mozilla/WrappingOperations.h"

#include <stdint.h>

namespace mozilla {

using HashNumber = uint32_t;
static const uint32_t kHashNumberBits = 32;

/**
 * The golden ratio as a 32-bit fixed-point value.
 */
static const HashNumber kGoldenRatioU32 = 0x9E3779B9U;

/*
 * Given a raw hash code, h, return a number that can be used to select a hash
 * bucket.
 *
 * This function aims to produce as uniform an output distribution as possible,
 * especially in the most significant (leftmost) bits, even though the input
 * distribution may be highly nonrandom, given the constraints that this must
 * be deterministic and quick to compute.
 *
 * Since the leftmost bits of the result are best, the hash bucket index is
 * computed by doing ScrambleHashCode(h) / (2^32/N) or the equivalent
 * right-shift, not ScrambleHashCode(h) % N or the equivalent bit-mask.
 *
 * FIXME: OrderedHashTable uses a bit-mask; see bug 775896.
 */
constexpr HashNumber
ScrambleHashCode(HashNumber h)
{
  /*
   * Simply returning h would not cause any hash tables to produce wrong
   * answers. But it can produce pathologically bad performance: The caller
   * right-shifts the result, keeping only the highest bits. The high bits of
   * hash codes are very often completely entropy-free. (So are the lowest
   * bits.)
   *
   * So we use Fibonacci hashing, as described in Knuth, The Art of Computer
   * Programming, 6.4. This mixes all the bits of the input hash code h.
   *
   * The value of goldenRatio is taken from the hex expansion of the golden
   * ratio, which starts 1.9E3779B9.... This value is especially good if
   * values with consecutive hash codes are stored in a hash table; see Knuth
   * for details.
   */
  return mozilla::WrappingMultiply(h, kGoldenRatioU32);
}

namespace detail {

MOZ_NO_SANITIZE_UNSIGNED_OVERFLOW
constexpr HashNumber
RotateLeft5(HashNumber aValue)
{
  return (aValue << 5) | (aValue >> 27);
}

constexpr HashNumber
AddU32ToHash(HashNumber aHash, uint32_t aValue)
{
  /*
   * This is the meat of all our hash routines.  This hash function is not
   * particularly sophisticated, but it seems to work well for our mostly
   * plain-text inputs.  Implementation notes follow.
   *
   * Our use of the golden ratio here is arbitrary; we could pick almost any
   * number which:
   *
   *  * is odd (because otherwise, all our hash values will be even)
   *
   *  * has a reasonably-even mix of 1's and 0's (consider the extreme case
   *    where we multiply by 0x3 or 0xeffffff -- this will not produce good
   *    mixing across all bits of the hash).
   *
   * The rotation length of 5 is also arbitrary, although an odd number is again
   * preferable so our hash explores the whole universe of possible rotations.
   *
   * Finally, we multiply by the golden ratio *after* xor'ing, not before.
   * Otherwise, if |aHash| is 0 (as it often is for the beginning of a
   * message), the expression
   *
   *   mozilla::WrappingMultiply(kGoldenRatioU32, RotateLeft5(aHash))
   *   |xor|
   *   aValue
   *
   * evaluates to |aValue|.
   *
   * (Number-theoretic aside: Because any odd number |m| is relatively prime to
   * our modulus (2**32), the list
   *
   *    [x * m (mod 2**32) for 0 <= x < 2**32]
   *
   * has no duplicate elements.  This means that multiplying by |m| does not
   * cause us to skip any possible hash values.
   *
   * It's also nice if |m| has large-ish order mod 2**32 -- that is, if the
   * smallest k such that m**k == 1 (mod 2**32) is large -- so we can safely
   * multiply our hash value by |m| a few times without negating the
   * multiplicative effect.  Our golden ratio constant has order 2**29, which is
   * more than enough for our purposes.)
   */
  return mozilla::WrappingMultiply(kGoldenRatioU32,
                                   RotateLeft5(aHash) ^ aValue);
}

/**
 * AddUintptrToHash takes sizeof(uintptr_t) as a template parameter.
 */
template<size_t PtrSize>
constexpr HashNumber
AddUintptrToHash(HashNumber aHash, uintptr_t aValue)
{
  return AddU32ToHash(aHash, static_cast<uint32_t>(aValue));
}

template<>
inline HashNumber
AddUintptrToHash<8>(HashNumber aHash, uintptr_t aValue)
{
  uint32_t v1 = static_cast<uint32_t>(aValue);
  uint32_t v2 = static_cast<uint32_t>(static_cast<uint64_t>(aValue) >> 32);
  return AddU32ToHash(AddU32ToHash(aHash, v1), v2);
}

} /* namespace detail */

/**
 * AddToHash takes a hash and some values and returns a new hash based on the
 * inputs.
 *
 * Currently, we support hashing uint32_t's, values which we can implicitly
 * convert to uint32_t, data pointers, and function pointers.
 */
template<typename T,
         bool TypeIsNotIntegral = !mozilla::IsIntegral<T>::value,
         typename U = typename mozilla::EnableIf<TypeIsNotIntegral>::Type>
MOZ_MUST_USE inline HashNumber
AddToHash(HashNumber aHash, T aA)
{
  /*
   * Try to convert |A| to uint32_t implicitly.  If this works, great.  If not,
   * we'll error out.
   */
  return detail::AddU32ToHash(aHash, aA);
}

template<typename A>
MOZ_MUST_USE inline HashNumber
AddToHash(HashNumber aHash, A* aA)
{
  /*
   * You might think this function should just take a void*.  But then we'd only
   * catch data pointers and couldn't handle function pointers.
   */

  static_assert(sizeof(aA) == sizeof(uintptr_t), "Strange pointer!");

  return detail::AddUintptrToHash<sizeof(uintptr_t)>(aHash, uintptr_t(aA));
}

// We use AddUintptrToHash() for hashing all integral types.  8-byte integral types
// are treated the same as 64-bit pointers, and smaller integral types are first
// implicitly converted to 32 bits and then passed to AddUintptrToHash() to be hashed.
template<typename T,
         typename U = typename mozilla::EnableIf<mozilla::IsIntegral<T>::value>::Type>
MOZ_MUST_USE constexpr HashNumber
AddToHash(HashNumber aHash, T aA)
{
  return detail::AddUintptrToHash<sizeof(T)>(aHash, aA);
}

template<typename A, typename... Args>
MOZ_MUST_USE HashNumber
AddToHash(HashNumber aHash, A aArg, Args... aArgs)
{
  return AddToHash(AddToHash(aHash, aArg), aArgs...);
}

/**
 * The HashGeneric class of functions let you hash one or more values.
 *
 * If you want to hash together two values x and y, calling HashGeneric(x, y) is
 * much better than calling AddToHash(x, y), because AddToHash(x, y) assumes
 * that x has already been hashed.
 */
template<typename... Args>
MOZ_MUST_USE inline HashNumber
HashGeneric(Args... aArgs)
{
  return AddToHash(0, aArgs...);
}

namespace detail {

template<typename T>
constexpr HashNumber
HashUntilZero(const T* aStr)
{
  HashNumber hash = 0;
  for (; T c = *aStr; aStr++) {
    hash = AddToHash(hash, c);
  }
  return hash;
}

template<typename T>
HashNumber
HashKnownLength(const T* aStr, size_t aLength)
{
  HashNumber hash = 0;
  for (size_t i = 0; i < aLength; i++) {
    hash = AddToHash(hash, aStr[i]);
  }
  return hash;
}

} /* namespace detail */

/**
 * The HashString overloads below do just what you'd expect.
 *
 * If you have the string's length, you might as well call the overload which
 * includes the length.  It may be marginally faster.
 */
MOZ_MUST_USE inline HashNumber
HashString(const char* aStr)
{
  return detail::HashUntilZero(reinterpret_cast<const unsigned char*>(aStr));
}

MOZ_MUST_USE inline HashNumber
HashString(const char* aStr, size_t aLength)
{
  return detail::HashKnownLength(reinterpret_cast<const unsigned char*>(aStr), aLength);
}

MOZ_MUST_USE
inline HashNumber
HashString(const unsigned char* aStr, size_t aLength)
{
  return detail::HashKnownLength(aStr, aLength);
}

// You may need to use the
// MOZ_{PUSH,POP}_DISABLE_INTEGRAL_CONSTANT_OVERFLOW_WARNING macros if you use
// this function. See the comment on those macros' definitions for more detail.
MOZ_MUST_USE constexpr HashNumber
HashString(const char16_t* aStr)
{
  return detail::HashUntilZero(aStr);
}

MOZ_MUST_USE inline HashNumber
HashString(const char16_t* aStr, size_t aLength)
{
  return detail::HashKnownLength(aStr, aLength);
}

/*
 * On Windows, wchar_t is not the same as char16_t, even though it's
 * the same width!
 */
#ifdef WIN32
MOZ_MUST_USE inline HashNumber
HashString(const wchar_t* aStr)
{
  return detail::HashUntilZero(aStr);
}

MOZ_MUST_USE inline HashNumber
HashString(const wchar_t* aStr, size_t aLength)
{
  return detail::HashKnownLength(aStr, aLength);
}
#endif

/**
 * Hash some number of bytes.
 *
 * This hash walks word-by-word, rather than byte-by-byte, so you won't get the
 * same result out of HashBytes as you would out of HashString.
 */
MOZ_MUST_USE extern MFBT_API HashNumber
HashBytes(const void* bytes, size_t aLength);

/**
 * A pseudorandom function mapping 32-bit integers to 32-bit integers.
 *
 * This is for when you're feeding private data (like pointer values or credit
 * card numbers) to a non-crypto hash function (like HashBytes) and then using
 * the hash code for something that untrusted parties could observe (like a JS
 * Map). Plug in a HashCodeScrambler before that last step to avoid leaking the
 * private data.
 *
 * By itself, this does not prevent hash-flooding DoS attacks, because an
 * attacker can still generate many values with exactly equal hash codes by
 * attacking the non-crypto hash function alone. Equal hash codes will, of
 * course, still be equal however much you scramble them.
 *
 * The algorithm is SipHash-1-3. See <https://131002.net/siphash/>.
 */
class HashCodeScrambler
{
  struct SipHasher;

  uint64_t mK0, mK1;

public:
  /** Creates a new scrambler with the given 128-bit key. */
  constexpr HashCodeScrambler(uint64_t aK0, uint64_t aK1) : mK0(aK0), mK1(aK1) {}

  /**
   * Scramble a hash code. Always produces the same result for the same
   * combination of key and hash code.
   */
  HashNumber scramble(HashNumber aHashCode) const
  {
    SipHasher hasher(mK0, mK1);
    return HashNumber(hasher.sipHash(aHashCode));
  }

private:
  struct SipHasher
  {
    SipHasher(uint64_t aK0, uint64_t aK1)
    {
      // 1. Initialization.
      mV0 = aK0 ^ UINT64_C(0x736f6d6570736575);
      mV1 = aK1 ^ UINT64_C(0x646f72616e646f6d);
      mV2 = aK0 ^ UINT64_C(0x6c7967656e657261);
      mV3 = aK1 ^ UINT64_C(0x7465646279746573);
    }

    uint64_t sipHash(uint64_t aM)
    {
      // 2. Compression.
      mV3 ^= aM;
      sipRound();
      mV0 ^= aM;

      // 3. Finalization.
      mV2 ^= 0xff;
      for (int i = 0; i < 3; i++)
        sipRound();
      return mV0 ^ mV1 ^ mV2 ^ mV3;
    }

    void sipRound()
    {
      mV0 = WrappingAdd(mV0, mV1);
      mV1 = RotateLeft(mV1, 13);
      mV1 ^= mV0;
      mV0 = RotateLeft(mV0, 32);
      mV2 = WrappingAdd(mV2, mV3);
      mV3 = RotateLeft(mV3, 16);
      mV3 ^= mV2;
      mV0 = WrappingAdd(mV0, mV3);
      mV3 = RotateLeft(mV3, 21);
      mV3 ^= mV0;
      mV2 = WrappingAdd(mV2, mV1);
      mV1 = RotateLeft(mV1, 17);
      mV1 ^= mV2;
      mV2 = RotateLeft(mV2, 32);
    }

    uint64_t mV0, mV1, mV2, mV3;
  };
};

} /* namespace mozilla */

#endif /* mozilla_HashFunctions_h */