summaryrefslogtreecommitdiff
path: root/src/crypto/fipsmodule/bn/mul.c
blob: 4a0711d8995b800e33b6a8eed58f9855ea426ad8 (plain)
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
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
 * All rights reserved.
 *
 * This package is an SSL implementation written
 * by Eric Young (eay@cryptsoft.com).
 * The implementation was written so as to conform with Netscapes SSL.
 *
 * This library is free for commercial and non-commercial use as long as
 * the following conditions are aheared to.  The following conditions
 * apply to all code found in this distribution, be it the RC4, RSA,
 * lhash, DES, etc., code; not just the SSL code.  The SSL documentation
 * included with this distribution is covered by the same copyright terms
 * except that the holder is Tim Hudson (tjh@cryptsoft.com).
 *
 * Copyright remains Eric Young's, and as such any Copyright notices in
 * the code are not to be removed.
 * If this package is used in a product, Eric Young should be given attribution
 * as the author of the parts of the library used.
 * This can be in the form of a textual message at program startup or
 * in documentation (online or textual) provided with the package.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *    "This product includes cryptographic software written by
 *     Eric Young (eay@cryptsoft.com)"
 *    The word 'cryptographic' can be left out if the rouines from the library
 *    being used are not cryptographic related :-).
 * 4. If you include any Windows specific code (or a derivative thereof) from
 *    the apps directory (application code) you must include an acknowledgement:
 *    "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
 *
 * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * The licence and distribution terms for any publically available version or
 * derivative of this code cannot be changed.  i.e. this code cannot simply be
 * copied and put under another distribution licence
 * [including the GNU Public Licence.] */

#include <openssl/bn.h>

#include <assert.h>
#include <string.h>

#include <openssl/err.h>
#include <openssl/mem.h>
#include <openssl/type_check.h>

#include "internal.h"
#include "../../internal.h"


#define BN_MUL_RECURSIVE_SIZE_NORMAL 16
#define BN_SQR_RECURSIVE_SIZE_NORMAL BN_MUL_RECURSIVE_SIZE_NORMAL


static void bn_abs_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
                             size_t num, BN_ULONG *tmp) {
  BN_ULONG borrow = bn_sub_words(tmp, a, b, num);
  bn_sub_words(r, b, a, num);
  bn_select_words(r, 0 - borrow, r /* tmp < 0 */, tmp /* tmp >= 0 */, num);
}

static void bn_mul_normal(BN_ULONG *r, const BN_ULONG *a, size_t na,
                          const BN_ULONG *b, size_t nb) {
  if (na < nb) {
    size_t itmp = na;
    na = nb;
    nb = itmp;
    const BN_ULONG *ltmp = a;
    a = b;
    b = ltmp;
  }
  BN_ULONG *rr = &(r[na]);
  if (nb == 0) {
    OPENSSL_memset(r, 0, na * sizeof(BN_ULONG));
    return;
  }
  rr[0] = bn_mul_words(r, a, na, b[0]);

  for (;;) {
    if (--nb == 0) {
      return;
    }
    rr[1] = bn_mul_add_words(&(r[1]), a, na, b[1]);
    if (--nb == 0) {
      return;
    }
    rr[2] = bn_mul_add_words(&(r[2]), a, na, b[2]);
    if (--nb == 0) {
      return;
    }
    rr[3] = bn_mul_add_words(&(r[3]), a, na, b[3]);
    if (--nb == 0) {
      return;
    }
    rr[4] = bn_mul_add_words(&(r[4]), a, na, b[4]);
    rr += 4;
    r += 4;
    b += 4;
  }
}

#if !defined(OPENSSL_X86) || defined(OPENSSL_NO_ASM)
// Here follows specialised variants of bn_add_words() and bn_sub_words(). They
// have the property performing operations on arrays of different sizes. The
// sizes of those arrays is expressed through cl, which is the common length (
// basicall, min(len(a),len(b)) ), and dl, which is the delta between the two
// lengths, calculated as len(a)-len(b). All lengths are the number of
// BN_ULONGs...  For the operations that require a result array as parameter,
// it must have the length cl+abs(dl). These functions should probably end up
// in bn_asm.c as soon as there are assembler counterparts for the systems that
// use assembler files.

static BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a,
                                  const BN_ULONG *b, int cl, int dl) {
  BN_ULONG c, t;

  assert(cl >= 0);
  c = bn_sub_words(r, a, b, cl);

  if (dl == 0) {
    return c;
  }

  r += cl;
  a += cl;
  b += cl;

  if (dl < 0) {
    for (;;) {
      t = b[0];
      r[0] = 0 - t - c;
      if (t != 0) {
        c = 1;
      }
      if (++dl >= 0) {
        break;
      }

      t = b[1];
      r[1] = 0 - t - c;
      if (t != 0) {
        c = 1;
      }
      if (++dl >= 0) {
        break;
      }

      t = b[2];
      r[2] = 0 - t - c;
      if (t != 0) {
        c = 1;
      }
      if (++dl >= 0) {
        break;
      }

      t = b[3];
      r[3] = 0 - t - c;
      if (t != 0) {
        c = 1;
      }
      if (++dl >= 0) {
        break;
      }

      b += 4;
      r += 4;
    }
  } else {
    int save_dl = dl;
    while (c) {
      t = a[0];
      r[0] = t - c;
      if (t != 0) {
        c = 0;
      }
      if (--dl <= 0) {
        break;
      }

      t = a[1];
      r[1] = t - c;
      if (t != 0) {
        c = 0;
      }
      if (--dl <= 0) {
        break;
      }

      t = a[2];
      r[2] = t - c;
      if (t != 0) {
        c = 0;
      }
      if (--dl <= 0) {
        break;
      }

      t = a[3];
      r[3] = t - c;
      if (t != 0) {
        c = 0;
      }
      if (--dl <= 0) {
        break;
      }

      save_dl = dl;
      a += 4;
      r += 4;
    }
    if (dl > 0) {
      if (save_dl > dl) {
        switch (save_dl - dl) {
          case 1:
            r[1] = a[1];
            if (--dl <= 0) {
              break;
            }
            OPENSSL_FALLTHROUGH;
          case 2:
            r[2] = a[2];
            if (--dl <= 0) {
              break;
            }
            OPENSSL_FALLTHROUGH;
          case 3:
            r[3] = a[3];
            if (--dl <= 0) {
              break;
            }
        }
        a += 4;
        r += 4;
      }
    }

    if (dl > 0) {
      for (;;) {
        r[0] = a[0];
        if (--dl <= 0) {
          break;
        }
        r[1] = a[1];
        if (--dl <= 0) {
          break;
        }
        r[2] = a[2];
        if (--dl <= 0) {
          break;
        }
        r[3] = a[3];
        if (--dl <= 0) {
          break;
        }

        a += 4;
        r += 4;
      }
    }
  }

  return c;
}
#else
// On other platforms the function is defined in asm.
BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
                           int cl, int dl);
#endif

// bn_abs_sub_part_words computes |r| = |a| - |b|, storing the absolute value
// and returning a mask of all ones if the result was negative and all zeros if
// the result was positive. |cl| and |dl| follow the |bn_sub_part_words| calling
// convention.
//
// TODO(davidben): Make this take |size_t|. The |cl| + |dl| calling convention
// is confusing. The trouble is 32-bit x86 implements |bn_sub_part_words| in
// assembly, but we can probably just delete it?
static BN_ULONG bn_abs_sub_part_words(BN_ULONG *r, const BN_ULONG *a,
                                      const BN_ULONG *b, int cl, int dl,
                                      BN_ULONG *tmp) {
  BN_ULONG borrow = bn_sub_part_words(tmp, a, b, cl, dl);
  bn_sub_part_words(r, b, a, cl, -dl);
  int r_len = cl + (dl < 0 ? -dl : dl);
  borrow = 0 - borrow;
  bn_select_words(r, borrow, r /* tmp < 0 */, tmp /* tmp >= 0 */, r_len);
  return borrow;
}

int bn_abs_sub_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
                         BN_CTX *ctx) {
  int cl = a->width < b->width ? a->width : b->width;
  int dl = a->width - b->width;
  int r_len = a->width < b->width ? b->width : a->width;
  BN_CTX_start(ctx);
  BIGNUM *tmp = BN_CTX_get(ctx);
  int ok = tmp != NULL &&
           bn_wexpand(r, r_len) &&
           bn_wexpand(tmp, r_len);
  if (ok) {
    bn_abs_sub_part_words(r->d, a->d, b->d, cl, dl, tmp->d);
    r->width = r_len;
  }
  BN_CTX_end(ctx);
  return ok;
}

// Karatsuba recursive multiplication algorithm
// (cf. Knuth, The Art of Computer Programming, Vol. 2)

// bn_mul_recursive sets |r| to |a| * |b|, using |t| as scratch space. |r| has
// length 2*|n2|, |a| has length |n2| + |dna|, |b| has length |n2| + |dnb|, and
// |t| has length 4*|n2|. |n2| must be a power of two. Finally, we must have
// -|BN_MUL_RECURSIVE_SIZE_NORMAL|/2 <= |dna| <= 0 and
// -|BN_MUL_RECURSIVE_SIZE_NORMAL|/2 <= |dnb| <= 0.
//
// TODO(davidben): Simplify and |size_t| the calling convention around lengths
// here.
static void bn_mul_recursive(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
                             int n2, int dna, int dnb, BN_ULONG *t) {
  // |n2| is a power of two.
  assert(n2 != 0 && (n2 & (n2 - 1)) == 0);
  // Check |dna| and |dnb| are in range.
  assert(-BN_MUL_RECURSIVE_SIZE_NORMAL/2 <= dna && dna <= 0);
  assert(-BN_MUL_RECURSIVE_SIZE_NORMAL/2 <= dnb && dnb <= 0);

  // Only call bn_mul_comba 8 if n2 == 8 and the
  // two arrays are complete [steve]
  if (n2 == 8 && dna == 0 && dnb == 0) {
    bn_mul_comba8(r, a, b);
    return;
  }

  // Else do normal multiply
  if (n2 < BN_MUL_RECURSIVE_SIZE_NORMAL) {
    bn_mul_normal(r, a, n2 + dna, b, n2 + dnb);
    if (dna + dnb < 0) {
      OPENSSL_memset(&r[2 * n2 + dna + dnb], 0,
                     sizeof(BN_ULONG) * -(dna + dnb));
    }
    return;
  }

  // Split |a| and |b| into a0,a1 and b0,b1, where a0 and b0 have size |n|.
  // Split |t| into t0,t1,t2,t3, each of size |n|, with the remaining 4*|n| used
  // for recursive calls.
  // Split |r| into r0,r1,r2,r3. We must contribute a0*b0 to r0,r1, a0*a1+b0*b1
  // to r1,r2, and a1*b1 to r2,r3. The middle term we will compute as:
  //
  //   a0*a1 + b0*b1 = (a0 - a1)*(b1 - b0) + a1*b1 + a0*b0
  //
  // Note that we know |n| >= |BN_MUL_RECURSIVE_SIZE_NORMAL|/2 above, so
  // |tna| and |tnb| are non-negative.
  int n = n2 / 2, tna = n + dna, tnb = n + dnb;

  // t0 = a0 - a1 and t1 = b1 - b0. The result will be multiplied, so we XOR
  // their sign masks, giving the sign of (a0 - a1)*(b1 - b0). t0 and t1
  // themselves store the absolute value.
  BN_ULONG neg = bn_abs_sub_part_words(t, a, &a[n], tna, n - tna, &t[n2]);
  neg ^= bn_abs_sub_part_words(&t[n], &b[n], b, tnb, tnb - n, &t[n2]);

  // Compute:
  // t2,t3 = t0 * t1 = |(a0 - a1)*(b1 - b0)|
  // r0,r1 = a0 * b0
  // r2,r3 = a1 * b1
  if (n == 4 && dna == 0 && dnb == 0) {
    bn_mul_comba4(&t[n2], t, &t[n]);

    bn_mul_comba4(r, a, b);
    bn_mul_comba4(&r[n2], &a[n], &b[n]);
  } else if (n == 8 && dna == 0 && dnb == 0) {
    bn_mul_comba8(&t[n2], t, &t[n]);

    bn_mul_comba8(r, a, b);
    bn_mul_comba8(&r[n2], &a[n], &b[n]);
  } else {
    BN_ULONG *p = &t[n2 * 2];
    bn_mul_recursive(&t[n2], t, &t[n], n, 0, 0, p);
    bn_mul_recursive(r, a, b, n, 0, 0, p);
    bn_mul_recursive(&r[n2], &a[n], &b[n], n, dna, dnb, p);
  }

  // t0,t1,c = r0,r1 + r2,r3 = a0*b0 + a1*b1
  BN_ULONG c = bn_add_words(t, r, &r[n2], n2);

  // t2,t3,c = t0,t1,c + neg*t2,t3 = (a0 - a1)*(b1 - b0) + a1*b1 + a0*b0.
  // The second term is stored as the absolute value, so we do this with a
  // constant-time select.
  BN_ULONG c_neg = c - bn_sub_words(&t[n2 * 2], t, &t[n2], n2);
  BN_ULONG c_pos = c + bn_add_words(&t[n2], t, &t[n2], n2);
  bn_select_words(&t[n2], neg, &t[n2 * 2], &t[n2], n2);
  OPENSSL_COMPILE_ASSERT(sizeof(BN_ULONG) <= sizeof(crypto_word_t),
                         crypto_word_t_too_small);
  c = constant_time_select_w(neg, c_neg, c_pos);

  // We now have our three components. Add them together.
  // r1,r2,c = r1,r2 + t2,t3,c
  c += bn_add_words(&r[n], &r[n], &t[n2], n2);

  // Propagate the carry bit to the end.
  for (int i = n + n2; i < n2 + n2; i++) {
    BN_ULONG old = r[i];
    r[i] = old + c;
    c = r[i] < old;
  }

  // The product should fit without carries.
  assert(c == 0);
}

// bn_mul_part_recursive sets |r| to |a| * |b|, using |t| as scratch space. |r|
// has length 4*|n|, |a| has length |n| + |tna|, |b| has length |n| + |tnb|, and
// |t| has length 8*|n|. |n| must be a power of two. Additionally, we must have
// 0 <= tna < n and 0 <= tnb < n, and |tna| and |tnb| must differ by at most
// one.
//
// TODO(davidben): Make this take |size_t| and perhaps the actual lengths of |a|
// and |b|.
static void bn_mul_part_recursive(BN_ULONG *r, const BN_ULONG *a,
                                  const BN_ULONG *b, int n, int tna, int tnb,
                                  BN_ULONG *t) {
  // |n| is a power of two.
  assert(n != 0 && (n & (n - 1)) == 0);
  // Check |tna| and |tnb| are in range.
  assert(0 <= tna && tna < n);
  assert(0 <= tnb && tnb < n);
  assert(-1 <= tna - tnb && tna - tnb <= 1);

  int n2 = n * 2;
  if (n < 8) {
    bn_mul_normal(r, a, n + tna, b, n + tnb);
    OPENSSL_memset(r + n2 + tna + tnb, 0, n2 - tna - tnb);
    return;
  }

  // Split |a| and |b| into a0,a1 and b0,b1, where a0 and b0 have size |n|. |a1|
  // and |b1| have size |tna| and |tnb|, respectively.
  // Split |t| into t0,t1,t2,t3, each of size |n|, with the remaining 4*|n| used
  // for recursive calls.
  // Split |r| into r0,r1,r2,r3. We must contribute a0*b0 to r0,r1, a0*a1+b0*b1
  // to r1,r2, and a1*b1 to r2,r3. The middle term we will compute as:
  //
  //   a0*a1 + b0*b1 = (a0 - a1)*(b1 - b0) + a1*b1 + a0*b0

  // t0 = a0 - a1 and t1 = b1 - b0. The result will be multiplied, so we XOR
  // their sign masks, giving the sign of (a0 - a1)*(b1 - b0). t0 and t1
  // themselves store the absolute value.
  BN_ULONG neg = bn_abs_sub_part_words(t, a, &a[n], tna, n - tna, &t[n2]);
  neg ^= bn_abs_sub_part_words(&t[n], &b[n], b, tnb, tnb - n, &t[n2]);

  // Compute:
  // t2,t3 = t0 * t1 = |(a0 - a1)*(b1 - b0)|
  // r0,r1 = a0 * b0
  // r2,r3 = a1 * b1
  if (n == 8) {
    bn_mul_comba8(&t[n2], t, &t[n]);
    bn_mul_comba8(r, a, b);

    bn_mul_normal(&r[n2], &a[n], tna, &b[n], tnb);
    // |bn_mul_normal| only writes |tna| + |tna| words. Zero the rest.
    OPENSSL_memset(&r[n2 + tna + tnb], 0, sizeof(BN_ULONG) * (n2 - tna - tnb));
  } else {
    BN_ULONG *p = &t[n2 * 2];
    bn_mul_recursive(&t[n2], t, &t[n], n, 0, 0, p);
    bn_mul_recursive(r, a, b, n, 0, 0, p);

    OPENSSL_memset(&r[n2], 0, sizeof(BN_ULONG) * n2);
    if (tna < BN_MUL_RECURSIVE_SIZE_NORMAL &&
        tnb < BN_MUL_RECURSIVE_SIZE_NORMAL) {
      bn_mul_normal(&r[n2], &a[n], tna, &b[n], tnb);
    } else {
      int i = n;
      for (;;) {
        i /= 2;
        if (i < tna || i < tnb) {
          // E.g., n == 16, i == 8 and tna == 11. |tna| and |tnb| are within one
          // of each other, so if |tna| is larger and tna > i, then we know
          // tnb >= i, and this call is valid.
          bn_mul_part_recursive(&r[n2], &a[n], &b[n], i, tna - i, tnb - i, p);
          break;
        }
        if (i == tna || i == tnb) {
          // If there is only a bottom half to the number, just do it. We know
          // the larger of |tna - i| and |tnb - i| is zero. The other is zero or
          // -1 by because of |tna| and |tnb| differ by at most one.
          bn_mul_recursive(&r[n2], &a[n], &b[n], i, tna - i, tnb - i, p);
          break;
        }

        // This loop will eventually terminate when |i| falls below
        // |BN_MUL_RECURSIVE_SIZE_NORMAL| because we know one of |tna| and |tnb|
        // exceeds that.
      }
    }
  }

  // t0,t1,c = r0,r1 + r2,r3 = a0*b0 + a1*b1
  BN_ULONG c = bn_add_words(t, r, &r[n2], n2);

  // t2,t3,c = t0,t1,c + neg*t2,t3 = (a0 - a1)*(b1 - b0) + a1*b1 + a0*b0.
  // The second term is stored as the absolute value, so we do this with a
  // constant-time select.
  BN_ULONG c_neg = c - bn_sub_words(&t[n2 * 2], t, &t[n2], n2);
  BN_ULONG c_pos = c + bn_add_words(&t[n2], t, &t[n2], n2);
  bn_select_words(&t[n2], neg, &t[n2 * 2], &t[n2], n2);
  OPENSSL_COMPILE_ASSERT(sizeof(BN_ULONG) <= sizeof(crypto_word_t),
                         crypto_word_t_too_small);
  c = constant_time_select_w(neg, c_neg, c_pos);

  // We now have our three components. Add them together.
  // r1,r2,c = r1,r2 + t2,t3,c
  c += bn_add_words(&r[n], &r[n], &t[n2], n2);

  // Propagate the carry bit to the end.
  for (int i = n + n2; i < n2 + n2; i++) {
    BN_ULONG old = r[i];
    r[i] = old + c;
    c = r[i] < old;
  }

  // The product should fit without carries.
  assert(c == 0);
}

// bn_mul_impl implements |BN_mul| and |bn_mul_consttime|. Note this function
// breaks |BIGNUM| invariants and may return a negative zero. This is handled by
// the callers.
static int bn_mul_impl(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
                       BN_CTX *ctx) {
  int al = a->width;
  int bl = b->width;
  if (al == 0 || bl == 0) {
    BN_zero(r);
    return 1;
  }

  int ret = 0;
  BIGNUM *rr;
  BN_CTX_start(ctx);
  if (r == a || r == b) {
    rr = BN_CTX_get(ctx);
    if (r == NULL) {
      goto err;
    }
  } else {
    rr = r;
  }
  rr->neg = a->neg ^ b->neg;

  int i = al - bl;
  if (i == 0) {
    if (al == 8) {
      if (!bn_wexpand(rr, 16)) {
        goto err;
      }
      rr->width = 16;
      bn_mul_comba8(rr->d, a->d, b->d);
      goto end;
    }
  }

  int top = al + bl;
  static const int kMulNormalSize = 16;
  if (al >= kMulNormalSize && bl >= kMulNormalSize) {
    if (-1 <= i && i <= 1) {
      // Find the larger power of two less than or equal to the larger length.
      int j;
      if (i >= 0) {
        j = BN_num_bits_word((BN_ULONG)al);
      } else {
        j = BN_num_bits_word((BN_ULONG)bl);
      }
      j = 1 << (j - 1);
      assert(j <= al || j <= bl);
      BIGNUM *t = BN_CTX_get(ctx);
      if (t == NULL) {
        goto err;
      }
      if (al > j || bl > j) {
        // We know |al| and |bl| are at most one from each other, so if al > j,
        // bl >= j, and vice versa. Thus we can use |bn_mul_part_recursive|.
        assert(al >= j && bl >= j);
        if (!bn_wexpand(t, j * 8) ||
            !bn_wexpand(rr, j * 4)) {
          goto err;
        }
        bn_mul_part_recursive(rr->d, a->d, b->d, j, al - j, bl - j, t->d);
      } else {
        // al <= j && bl <= j. Additionally, we know j <= al or j <= bl, so one
        // of al - j or bl - j is zero. The other, by the bound on |i| above, is
        // zero or -1. Thus, we can use |bn_mul_recursive|.
        if (!bn_wexpand(t, j * 4) ||
            !bn_wexpand(rr, j * 2)) {
          goto err;
        }
        bn_mul_recursive(rr->d, a->d, b->d, j, al - j, bl - j, t->d);
      }
      rr->width = top;
      goto end;
    }
  }

  if (!bn_wexpand(rr, top)) {
    goto err;
  }
  rr->width = top;
  bn_mul_normal(rr->d, a->d, al, b->d, bl);

end:
  if (r != rr && !BN_copy(r, rr)) {
    goto err;
  }
  ret = 1;

err:
  BN_CTX_end(ctx);
  return ret;
}

int BN_mul(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx) {
  if (!bn_mul_impl(r, a, b, ctx)) {
    return 0;
  }

  // This additionally fixes any negative zeros created by |bn_mul_impl|.
  bn_set_minimal_width(r);
  return 1;
}

int bn_mul_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx) {
  // Prevent negative zeros.
  if (a->neg || b->neg) {
    OPENSSL_PUT_ERROR(BN, BN_R_NEGATIVE_NUMBER);
    return 0;
  }

  return bn_mul_impl(r, a, b, ctx);
}

int bn_mul_small(BN_ULONG *r, size_t num_r, const BN_ULONG *a, size_t num_a,
                 const BN_ULONG *b, size_t num_b) {
  if (num_r != num_a + num_b) {
    OPENSSL_PUT_ERROR(BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
    return 0;
  }
  // TODO(davidben): Should this call |bn_mul_comba4| too? |BN_mul| does not
  // hit that code.
  if (num_a == 8 && num_b == 8) {
    bn_mul_comba8(r, a, b);
  } else {
    bn_mul_normal(r, a, num_a, b, num_b);
  }
  return 1;
}

// tmp must have 2*n words
static void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, size_t n,
                          BN_ULONG *tmp) {
  if (n == 0) {
    return;
  }

  size_t max = n * 2;
  const BN_ULONG *ap = a;
  BN_ULONG *rp = r;
  rp[0] = rp[max - 1] = 0;
  rp++;

  // Compute the contribution of a[i] * a[j] for all i < j.
  if (n > 1) {
    ap++;
    rp[n - 1] = bn_mul_words(rp, ap, n - 1, ap[-1]);
    rp += 2;
  }
  if (n > 2) {
    for (size_t i = n - 2; i > 0; i--) {
      ap++;
      rp[i] = bn_mul_add_words(rp, ap, i, ap[-1]);
      rp += 2;
    }
  }

  // The final result fits in |max| words, so none of the following operations
  // will overflow.

  // Double |r|, giving the contribution of a[i] * a[j] for all i != j.
  bn_add_words(r, r, r, max);

  // Add in the contribution of a[i] * a[i] for all i.
  bn_sqr_words(tmp, a, n);
  bn_add_words(r, r, tmp, max);
}

// bn_sqr_recursive sets |r| to |a|^2, using |t| as scratch space. |r| has
// length 2*|n2|, |a| has length |n2|, and |t| has length 4*|n2|. |n2| must be
// a power of two.
static void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, size_t n2,
                             BN_ULONG *t) {
  // |n2| is a power of two.
  assert(n2 != 0 && (n2 & (n2 - 1)) == 0);

  if (n2 == 4) {
    bn_sqr_comba4(r, a);
    return;
  }
  if (n2 == 8) {
    bn_sqr_comba8(r, a);
    return;
  }
  if (n2 < BN_SQR_RECURSIVE_SIZE_NORMAL) {
    bn_sqr_normal(r, a, n2, t);
    return;
  }

  // Split |a| into a0,a1, each of size |n|.
  // Split |t| into t0,t1,t2,t3, each of size |n|, with the remaining 4*|n| used
  // for recursive calls.
  // Split |r| into r0,r1,r2,r3. We must contribute a0^2 to r0,r1, 2*a0*a1 to
  // r1,r2, and a1^2 to r2,r3.
  size_t n = n2 / 2;
  BN_ULONG *t_recursive = &t[n2 * 2];

  // t0 = |a0 - a1|.
  bn_abs_sub_words(t, a, &a[n], n, &t[n]);
  // t2,t3 = t0^2 = |a0 - a1|^2 = a0^2 - 2*a0*a1 + a1^2
  bn_sqr_recursive(&t[n2], t, n, t_recursive);

  // r0,r1 = a0^2
  bn_sqr_recursive(r, a, n, t_recursive);

  // r2,r3 = a1^2
  bn_sqr_recursive(&r[n2], &a[n], n, t_recursive);

  // t0,t1,c = r0,r1 + r2,r3 = a0^2 + a1^2
  BN_ULONG c = bn_add_words(t, r, &r[n2], n2);
  // t2,t3,c = t0,t1,c - t2,t3 = 2*a0*a1
  c -= bn_sub_words(&t[n2], t, &t[n2], n2);

  // We now have our three components. Add them together.
  // r1,r2,c = r1,r2 + t2,t3,c
  c += bn_add_words(&r[n], &r[n], &t[n2], n2);

  // Propagate the carry bit to the end.
  for (size_t i = n + n2; i < n2 + n2; i++) {
    BN_ULONG old = r[i];
    r[i] = old + c;
    c = r[i] < old;
  }

  // The square should fit without carries.
  assert(c == 0);
}

int BN_mul_word(BIGNUM *bn, BN_ULONG w) {
  if (!bn->width) {
    return 1;
  }

  if (w == 0) {
    BN_zero(bn);
    return 1;
  }

  BN_ULONG ll = bn_mul_words(bn->d, bn->d, bn->width, w);
  if (ll) {
    if (!bn_wexpand(bn, bn->width + 1)) {
      return 0;
    }
    bn->d[bn->width++] = ll;
  }

  return 1;
}

int bn_sqr_consttime(BIGNUM *r, const BIGNUM *a, BN_CTX *ctx) {
  int al = a->width;
  if (al <= 0) {
    r->width = 0;
    r->neg = 0;
    return 1;
  }

  int ret = 0;
  BN_CTX_start(ctx);
  BIGNUM *rr = (a != r) ? r : BN_CTX_get(ctx);
  BIGNUM *tmp = BN_CTX_get(ctx);
  if (!rr || !tmp) {
    goto err;
  }

  int max = 2 * al;  // Non-zero (from above)
  if (!bn_wexpand(rr, max)) {
    goto err;
  }

  if (al == 4) {
    bn_sqr_comba4(rr->d, a->d);
  } else if (al == 8) {
    bn_sqr_comba8(rr->d, a->d);
  } else {
    if (al < BN_SQR_RECURSIVE_SIZE_NORMAL) {
      BN_ULONG t[BN_SQR_RECURSIVE_SIZE_NORMAL * 2];
      bn_sqr_normal(rr->d, a->d, al, t);
    } else {
      // If |al| is a power of two, we can use |bn_sqr_recursive|.
      if (al != 0 && (al & (al - 1)) == 0) {
        if (!bn_wexpand(tmp, al * 4)) {
          goto err;
        }
        bn_sqr_recursive(rr->d, a->d, al, tmp->d);
      } else {
        if (!bn_wexpand(tmp, max)) {
          goto err;
        }
        bn_sqr_normal(rr->d, a->d, al, tmp->d);
      }
    }
  }

  rr->neg = 0;
  rr->width = max;

  if (rr != r && !BN_copy(r, rr)) {
    goto err;
  }
  ret = 1;

err:
  BN_CTX_end(ctx);
  return ret;
}

int BN_sqr(BIGNUM *r, const BIGNUM *a, BN_CTX *ctx) {
  if (!bn_sqr_consttime(r, a, ctx)) {
    return 0;
  }

  bn_set_minimal_width(r);
  return 1;
}

int bn_sqr_small(BN_ULONG *r, size_t num_r, const BN_ULONG *a, size_t num_a) {
  if (num_r != 2 * num_a || num_a > BN_SMALL_MAX_WORDS) {
    OPENSSL_PUT_ERROR(BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
    return 0;
  }
  if (num_a == 4) {
    bn_sqr_comba4(r, a);
  } else if (num_a == 8) {
    bn_sqr_comba8(r, a);
  } else {
    BN_ULONG tmp[2 * BN_SMALL_MAX_WORDS];
    bn_sqr_normal(r, a, num_a, tmp);
    OPENSSL_cleanse(tmp, 2 * num_a * sizeof(BN_ULONG));
  }
  return 1;
}