/
p384.c
1483 lines (1328 loc) · 57.8 KB
/
p384.c
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
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
------------------------------------------------------------------------------------
Copyright Amazon.com Inc. or its affiliates. All Rights Reserved.
SPDX-License-Identifier: Apache-2.0 OR ISC
------------------------------------------------------------------------------------
*/
#include <openssl/bn.h>
#include <openssl/ec.h>
#include <openssl/err.h>
#include <openssl/mem.h>
#include "../bn/internal.h"
#include "../cpucap/internal.h"
#include "../delocate.h"
#include "internal.h"
// We have two implementations of the field arithmetic for P-384 curve:
// - Fiat-crypto
// - s2n-bignum
// Both Fiat-crypto and s2n-bignum implementations are formally verified.
// Fiat-crypto implementation is fully portable C code, while s2n-bignum
// implements the operations in assembly for x86_64 and aarch64 platforms.
// All the P-384 field operations supported by Fiat-crypto are supported
// by s2n-bignum as well, so s2n-bignum can be used as a drop-in replacement
// when appropriate. To do that we define macros for the functions.
// For example, field addition macro is either defined as
// #define p384_felem_add(out, in0, in1) fiat_p384_add(out, in0, in1)
// when Fiat-crypto is used, or as:
// #define p384_felem_add(out, in0, in1) bignum_add_p384(out, in0, in1)
// when s2n-bignum is used.
//
#if !defined(OPENSSL_NO_ASM) && \
(defined(OPENSSL_LINUX) || defined(OPENSSL_APPLE)) && \
(defined(OPENSSL_X86_64) || defined(OPENSSL_AARCH64)) && \
!defined(MY_ASSEMBLER_IS_TOO_OLD_FOR_AVX)
# include "../../../third_party/s2n-bignum/include/s2n-bignum_aws-lc.h"
# define P384_USE_S2N_BIGNUM_FIELD_ARITH 1
# define P384_USE_64BIT_LIMBS_FELEM 1
#else
# if defined(BORINGSSL_HAS_UINT128)
# include "../../../third_party/fiat/p384_64.h"
# define P384_USE_64BIT_LIMBS_FELEM 1
# else
# include "../../../third_party/fiat/p384_32.h"
# endif
#endif
#if defined(P384_USE_64BIT_LIMBS_FELEM)
#define P384_NLIMBS (6)
typedef uint64_t p384_limb_t;
typedef uint64_t p384_felem[P384_NLIMBS];
static const p384_felem p384_felem_one = {
0xffffffff00000001, 0xffffffff, 0x1, 0x0, 0x0, 0x0};
#else // 64BIT; else 32BIT
#define P384_NLIMBS (12)
typedef uint32_t p384_limb_t;
typedef uint32_t p384_felem[P384_NLIMBS];
static const p384_felem p384_felem_one = {
0x1, 0xffffffff, 0xffffffff, 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
#endif // 64BIT
#if defined(P384_USE_S2N_BIGNUM_FIELD_ARITH)
#if defined(OPENSSL_X86_64)
// On x86_64 platforms s2n-bignum uses bmi2 and adx instruction sets
// for some of the functions. These instructions are not supported by
// every x86 CPU so we have to check if they are available and in case
// they are not we fallback to slightly slower but generic implementation.
static inline uint8_t p384_use_s2n_bignum_alt(void) {
return (!CRYPTO_is_BMI2_capable() || !CRYPTO_is_ADX_capable());
}
#else
// On aarch64 platforms s2n-bignum has two implementations of certain
// functions -- the default one and the alternative (suffixed _alt).
// Depending on the architecture one version is faster than the other.
// Generally, the "_alt" functions are faster on architectures with higher
// multiplier throughput, for example, Graviton 3, Apple's M1 and iPhone chips.
static inline uint8_t p384_use_s2n_bignum_alt(void) {
return CRYPTO_is_ARMv8_wide_multiplier_capable();
}
#endif
#define p384_felem_add(out, in0, in1) bignum_add_p384(out, in0, in1)
#define p384_felem_sub(out, in0, in1) bignum_sub_p384(out, in0, in1)
#define p384_felem_opp(out, in0) bignum_neg_p384(out, in0)
#define p384_felem_to_bytes(out, in0) bignum_tolebytes_6(out, in0)
#define p384_felem_from_bytes(out, in0) bignum_fromlebytes_6(out, in0)
// The following four functions need bmi2 and adx support.
#define p384_felem_mul(out, in0, in1) \
if (p384_use_s2n_bignum_alt()) bignum_montmul_p384_alt(out, in0, in1); \
else bignum_montmul_p384(out, in0, in1);
#define p384_felem_sqr(out, in0) \
if (p384_use_s2n_bignum_alt()) bignum_montsqr_p384_alt(out, in0); \
else bignum_montsqr_p384(out, in0);
#define p384_felem_to_mont(out, in0) \
if (p384_use_s2n_bignum_alt()) bignum_tomont_p384_alt(out, in0); \
else bignum_tomont_p384(out, in0);
#define p384_felem_from_mont(out, in0) \
if (p384_use_s2n_bignum_alt()) bignum_deamont_p384_alt(out, in0); \
else bignum_deamont_p384(out, in0);
static p384_limb_t p384_felem_nz(const p384_limb_t in1[P384_NLIMBS]) {
return bignum_nonzero_6(in1);
}
#else // P384_USE_S2N_BIGNUM_FIELD_ARITH
// Fiat-crypto implementation of field arithmetic
#define p384_felem_add(out, in0, in1) fiat_p384_add(out, in0, in1)
#define p384_felem_sub(out, in0, in1) fiat_p384_sub(out, in0, in1)
#define p384_felem_opp(out, in0) fiat_p384_opp(out, in0)
#define p384_felem_mul(out, in0, in1) fiat_p384_mul(out, in0, in1)
#define p384_felem_sqr(out, in0) fiat_p384_square(out, in0)
#define p384_felem_to_mont(out, in0) fiat_p384_to_montgomery(out, in0)
#define p384_felem_from_mont(out, in0) fiat_p384_from_montgomery(out, in0)
#define p384_felem_to_bytes(out, in0) fiat_p384_to_bytes(out, in0)
#define p384_felem_from_bytes(out, in0) fiat_p384_from_bytes(out, in0)
static p384_limb_t p384_felem_nz(const p384_limb_t in1[P384_NLIMBS]) {
p384_limb_t ret;
fiat_p384_nonzero(&ret, in1);
return ret;
}
#endif // P384_USE_S2N_BIGNUM_FIELD_ARITH
static void p384_felem_copy(p384_limb_t out[P384_NLIMBS],
const p384_limb_t in1[P384_NLIMBS]) {
for (size_t i = 0; i < P384_NLIMBS; i++) {
out[i] = in1[i];
}
}
static void p384_felem_cmovznz(p384_limb_t out[P384_NLIMBS],
p384_limb_t t,
const p384_limb_t z[P384_NLIMBS],
const p384_limb_t nz[P384_NLIMBS]) {
p384_limb_t mask = constant_time_is_zero_w(t);
for (size_t i = 0; i < P384_NLIMBS; i++) {
out[i] = constant_time_select_w(mask, z[i], nz[i]);
}
}
// NOTE: the input and output are in little-endian representation.
static void p384_from_generic(p384_felem out, const EC_FELEM *in) {
p384_felem_from_bytes(out, in->bytes);
}
// NOTE: the input and output are in little-endian representation.
static void p384_to_generic(EC_FELEM *out, const p384_felem in) {
// This works because 384 is a multiple of 64, so there are no excess bytes to
// zero when rounding up to |BN_ULONG|s.
OPENSSL_STATIC_ASSERT(
384 / 8 == sizeof(BN_ULONG) * ((384 + BN_BITS2 - 1) / BN_BITS2),
p384_felem_to_bytes_leaves_bytes_uninitialized);
p384_felem_to_bytes(out->bytes, in);
}
// p384_inv_square calculates |out| = |in|^{-2}
//
// Based on Fermat's Little Theorem:
// a^p = a (mod p)
// a^{p-1} = 1 (mod p)
// a^{p-3} = a^{-2} (mod p)
// p = 2^384 - 2^128 - 2^96 + 2^32 - 1
// Hexadecimal representation of p − 3:
// p-3 = ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff fffffffe
// ffffffff 00000000 00000000 fffffffc
static void p384_inv_square(p384_felem out,
const p384_felem in) {
// This implements the addition chain described in
// https://briansmith.org/ecc-inversion-addition-chains-01#p384_field_inversion
// The side comments show the value of the exponent:
// squaring the element => doubling the exponent
// multiplying by an element => adding to the exponent the power of that element
p384_felem x2, x3, x6, x12, x15, x30, x60, x120;
p384_felem_sqr(x2, in); // 2^2 - 2^1
p384_felem_mul(x2, x2, in); // 2^2 - 2^0
p384_felem_sqr(x3, x2); // 2^3 - 2^1
p384_felem_mul(x3, x3, in); // 2^3 - 2^0
p384_felem_sqr(x6, x3);
for (int i = 1; i < 3; i++) {
p384_felem_sqr(x6, x6);
} // 2^6 - 2^3
p384_felem_mul(x6, x6, x3); // 2^6 - 2^0
p384_felem_sqr(x12, x6);
for (int i = 1; i < 6; i++) {
p384_felem_sqr(x12, x12);
} // 2^12 - 2^6
p384_felem_mul(x12, x12, x6); // 2^12 - 2^0
p384_felem_sqr(x15, x12);
for (int i = 1; i < 3; i++) {
p384_felem_sqr(x15, x15);
} // 2^15 - 2^3
p384_felem_mul(x15, x15, x3); // 2^15 - 2^0
p384_felem_sqr(x30, x15);
for (int i = 1; i < 15; i++) {
p384_felem_sqr(x30, x30);
} // 2^30 - 2^15
p384_felem_mul(x30, x30, x15); // 2^30 - 2^0
p384_felem_sqr(x60, x30);
for (int i = 1; i < 30; i++) {
p384_felem_sqr(x60, x60);
} // 2^60 - 2^30
p384_felem_mul(x60, x60, x30); // 2^60 - 2^0
p384_felem_sqr(x120, x60);
for (int i = 1; i < 60; i++) {
p384_felem_sqr(x120, x120);
} // 2^120 - 2^60
p384_felem_mul(x120, x120, x60); // 2^120 - 2^0
p384_felem ret;
p384_felem_sqr(ret, x120);
for (int i = 1; i < 120; i++) {
p384_felem_sqr(ret, ret);
} // 2^240 - 2^120
p384_felem_mul(ret, ret, x120); // 2^240 - 2^0
for (int i = 0; i < 15; i++) {
p384_felem_sqr(ret, ret);
} // 2^255 - 2^15
p384_felem_mul(ret, ret, x15); // 2^255 - 2^0
// Why (1 + 30) in the loop?
// This is as expressed in:
// https://briansmith.org/ecc-inversion-addition-chains-01#p384_field_inversion
// My guess is to say that we're going to shift 31 bits, but this time we
// won't add x31 to make all the new bits 1s, as was done in previous steps,
// but we're going to add x30 so there will be 255 1s, then a 0, then 30 1s
// to form this pattern:
// ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff fffffffe ffffffff
// (the last 2 1s are appended in the following step).
for (int i = 0; i < (1 + 30); i++) {
p384_felem_sqr(ret, ret);
} // 2^286 - 2^31
p384_felem_mul(ret, ret, x30); // 2^286 - 2^30 - 2^0
p384_felem_sqr(ret, ret);
p384_felem_sqr(ret, ret); // 2^288 - 2^32 - 2^2
p384_felem_mul(ret, ret, x2); // 2^288 - 2^32 - 2^0
// Why not 94 instead of (64 + 30) in the loop?
// Similarly to the comment above, there is a shift of 94 bits
// but what will be added is x30, which will cause 64 of those bits
// to be 64 0s and 30 1s to complete the pattern above with:
// 00000000 00000000 fffffffc
// (the last 2 0s are appended by the last 2 shifts).
for (int i = 0; i < (64 + 30); i++) {
p384_felem_sqr(ret, ret);
} // 2^382 - 2^126 - 2^94
p384_felem_mul(ret, ret, x30); // 2^382 - 2^126 - 2^94 + 2^30 - 2^0
p384_felem_sqr(ret, ret);
p384_felem_sqr(out, ret); // 2^384 - 2^128 - 2^96 + 2^32 - 2^2 = p - 3
}
// Group operations
// ----------------
//
// Building on top of the field operations we have the operations on the
// elliptic curve group itself. Points on the curve are represented in Jacobian
// coordinates.
//
// p384_point_double calculates 2*(x_in, y_in, z_in)
//
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
//
// Coq transcription and correctness proof:
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L93>
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L201>
// Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed;
// while x_out == y_in is not (maybe this works, but it's not tested).
static void p384_point_double(p384_felem x_out,
p384_felem y_out,
p384_felem z_out,
const p384_felem x_in,
const p384_felem y_in,
const p384_felem z_in) {
p384_felem delta, gamma, beta, ftmp, ftmp2, tmptmp, alpha, fourbeta;
// delta = z^2
p384_felem_sqr(delta, z_in);
// gamma = y^2
p384_felem_sqr(gamma, y_in);
// beta = x*gamma
p384_felem_mul(beta, x_in, gamma);
// alpha = 3*(x-delta)*(x+delta)
p384_felem_sub(ftmp, x_in, delta);
p384_felem_add(ftmp2, x_in, delta);
p384_felem_add(tmptmp, ftmp2, ftmp2);
p384_felem_add(ftmp2, ftmp2, tmptmp);
p384_felem_mul(alpha, ftmp, ftmp2);
// x' = alpha^2 - 8*beta
p384_felem_sqr(x_out, alpha);
p384_felem_add(fourbeta, beta, beta);
p384_felem_add(fourbeta, fourbeta, fourbeta);
p384_felem_add(tmptmp, fourbeta, fourbeta);
p384_felem_sub(x_out, x_out, tmptmp);
// z' = (y + z)^2 - gamma - delta
// The following calculation differs from that in p256.c:
// an add is replaced with a sub. This saves us 5 cmovznz operations
// when Fiat-crypto implementation of felem_add and felem_sub is used,
// and also a certain number of intructions when s2n-bignum is used.
p384_felem_add(ftmp, y_in, z_in);
p384_felem_sqr(z_out, ftmp);
p384_felem_sub(z_out, z_out, gamma);
p384_felem_sub(z_out, z_out, delta);
// y' = alpha*(4*beta - x') - 8*gamma^2
p384_felem_sub(y_out, fourbeta, x_out);
p384_felem_add(gamma, gamma, gamma);
p384_felem_sqr(gamma, gamma);
p384_felem_mul(y_out, alpha, y_out);
p384_felem_add(gamma, gamma, gamma);
p384_felem_sub(y_out, y_out, gamma);
}
// p384_point_add calculates (x1, y1, z1) + (x2, y2, z2)
//
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian.html#addition-add-2007-bl
// adapted for mixed addition (z2 = 1, or z2 = 0 for the point at infinity).
//
// Coq transcription and correctness proof:
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L467>
// <https://github.com/davidben/fiat-crypto/blob/c7b95f62b2a54b559522573310e9b487327d219a/src/Curves/Weierstrass/Jacobian.v#L544>
static void p384_point_add(p384_felem x3, p384_felem y3, p384_felem z3,
const p384_felem x1,
const p384_felem y1,
const p384_felem z1,
const int mixed,
const p384_felem x2,
const p384_felem y2,
const p384_felem z2) {
p384_felem x_out, y_out, z_out;
p384_limb_t z1nz = p384_felem_nz(z1);
p384_limb_t z2nz = p384_felem_nz(z2);
// z1z1 = z1**2
p384_felem z1z1;
p384_felem_sqr(z1z1, z1);
p384_felem u1, s1, two_z1z2;
if (!mixed) {
// z2z2 = z2**2
p384_felem z2z2;
p384_felem_sqr(z2z2, z2);
// u1 = x1*z2z2
p384_felem_mul(u1, x1, z2z2);
// two_z1z2 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2
p384_felem_add(two_z1z2, z1, z2);
p384_felem_sqr(two_z1z2, two_z1z2);
p384_felem_sub(two_z1z2, two_z1z2, z1z1);
p384_felem_sub(two_z1z2, two_z1z2, z2z2);
// s1 = y1 * z2**3
p384_felem_mul(s1, z2, z2z2);
p384_felem_mul(s1, s1, y1);
} else {
// We'll assume z2 = 1 (special case z2 = 0 is handled later).
// u1 = x1*z2z2
p384_felem_copy(u1, x1);
// two_z1z2 = 2z1z2
p384_felem_add(two_z1z2, z1, z1);
// s1 = y1 * z2**3
p384_felem_copy(s1, y1);
}
// u2 = x2*z1z1
p384_felem u2;
p384_felem_mul(u2, x2, z1z1);
// h = u2 - u1
p384_felem h;
p384_felem_sub(h, u2, u1);
p384_limb_t xneq = p384_felem_nz(h);
// z_out = two_z1z2 * h
p384_felem_mul(z_out, h, two_z1z2);
// z1z1z1 = z1 * z1z1
p384_felem z1z1z1;
p384_felem_mul(z1z1z1, z1, z1z1);
// s2 = y2 * z1**3
p384_felem s2;
p384_felem_mul(s2, y2, z1z1z1);
// r = (s2 - s1)*2
p384_felem r;
p384_felem_sub(r, s2, s1);
p384_felem_add(r, r, r);
p384_limb_t yneq = p384_felem_nz(r);
// This case will never occur in the constant-time |ec_GFp_mont_mul|.
p384_limb_t is_nontrivial_double = constant_time_is_zero_w(xneq | yneq) &
~constant_time_is_zero_w(z1nz) &
~constant_time_is_zero_w(z2nz);
if (is_nontrivial_double) {
p384_point_double(x3, y3, z3, x1, y1, z1);
return;
}
// I = (2h)**2
p384_felem i;
p384_felem_add(i, h, h);
p384_felem_sqr(i, i);
// J = h * I
p384_felem j;
p384_felem_mul(j, h, i);
// V = U1 * I
p384_felem v;
p384_felem_mul(v, u1, i);
// x_out = r**2 - J - 2V
p384_felem_sqr(x_out, r);
p384_felem_sub(x_out, x_out, j);
p384_felem_sub(x_out, x_out, v);
p384_felem_sub(x_out, x_out, v);
// y_out = r(V-x_out) - 2 * s1 * J
p384_felem_sub(y_out, v, x_out);
p384_felem_mul(y_out, y_out, r);
p384_felem s1j;
p384_felem_mul(s1j, s1, j);
p384_felem_sub(y_out, y_out, s1j);
p384_felem_sub(y_out, y_out, s1j);
p384_felem_cmovznz(x_out, z1nz, x2, x_out);
p384_felem_cmovznz(x3, z2nz, x1, x_out);
p384_felem_cmovznz(y_out, z1nz, y2, y_out);
p384_felem_cmovznz(y3, z2nz, y1, y_out);
p384_felem_cmovznz(z_out, z1nz, z2, z_out);
p384_felem_cmovznz(z3, z2nz, z1, z_out);
}
// OPENSSL EC_METHOD FUNCTIONS
// Takes the Jacobian coordinates (X, Y, Z) of a point and returns:
// (X', Y') = (X/Z^2, Y/Z^3).
static int ec_GFp_nistp384_point_get_affine_coordinates(
const EC_GROUP *group, const EC_RAW_POINT *point,
EC_FELEM *x_out, EC_FELEM *y_out) {
if (ec_GFp_simple_is_at_infinity(group, point)) {
OPENSSL_PUT_ERROR(EC, EC_R_POINT_AT_INFINITY);
return 0;
}
p384_felem z1, z2;
p384_from_generic(z1, &point->Z);
p384_inv_square(z2, z1);
if (x_out != NULL) {
p384_felem x;
p384_from_generic(x, &point->X);
p384_felem_mul(x, x, z2);
p384_to_generic(x_out, x);
}
if (y_out != NULL) {
p384_felem y;
p384_from_generic(y, &point->Y);
p384_felem_sqr(z2, z2); // z^-4
p384_felem_mul(y, y, z1); // y * z
p384_felem_mul(y, y, z2); // y * z^-3
p384_to_generic(y_out, y);
}
return 1;
}
static void ec_GFp_nistp384_add(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *a, const EC_RAW_POINT *b) {
p384_felem x1, y1, z1, x2, y2, z2;
p384_from_generic(x1, &a->X);
p384_from_generic(y1, &a->Y);
p384_from_generic(z1, &a->Z);
p384_from_generic(x2, &b->X);
p384_from_generic(y2, &b->Y);
p384_from_generic(z2, &b->Z);
p384_point_add(x1, y1, z1, x1, y1, z1, 0 /* both Jacobian */, x2, y2, z2);
p384_to_generic(&r->X, x1);
p384_to_generic(&r->Y, y1);
p384_to_generic(&r->Z, z1);
}
static void ec_GFp_nistp384_dbl(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *a) {
p384_felem x, y, z;
p384_from_generic(x, &a->X);
p384_from_generic(y, &a->Y);
p384_from_generic(z, &a->Z);
p384_point_double(x, y, z, x, y, z);
p384_to_generic(&r->X, x);
p384_to_generic(&r->Y, y);
p384_to_generic(&r->Z, z);
}
// The calls to from/to_generic are needed for the case
// when BORINGSSL_HAS_UINT128 is undefined, i.e. p384_32.h fiat code is used;
// while OPENSSL_64_BIT is defined, i.e. BN_ULONG is uint64_t
static void ec_GFp_nistp384_mont_felem_to_bytes(
const EC_GROUP *group, uint8_t *out, size_t *out_len, const EC_FELEM *in) {
size_t len = BN_num_bytes(&group->field);
EC_FELEM felem_tmp;
p384_felem tmp;
p384_from_generic(tmp, in);
p384_felem_from_mont(tmp, tmp);
p384_to_generic(&felem_tmp, tmp);
// Convert to a big-endian byte array.
for (size_t i = 0; i < len; i++) {
out[i] = felem_tmp.bytes[len - 1 - i];
}
*out_len = len;
}
static int ec_GFp_nistp384_mont_felem_from_bytes(
const EC_GROUP *group, EC_FELEM *out, const uint8_t *in, size_t len) {
EC_FELEM felem_tmp;
p384_felem tmp;
// This function calls bn_cmp_words_consttime
if (!ec_GFp_simple_felem_from_bytes(group, &felem_tmp, in, len)) {
return 0;
}
p384_from_generic(tmp, &felem_tmp);
p384_felem_to_mont(tmp, tmp);
p384_to_generic(out, tmp);
return 1;
}
static int ec_GFp_nistp384_cmp_x_coordinate(const EC_GROUP *group,
const EC_RAW_POINT *p,
const EC_SCALAR *r) {
if (ec_GFp_simple_is_at_infinity(group, p)) {
return 0;
}
// We wish to compare X/Z^2 with r. This is equivalent to comparing X with
// r*Z^2. Note that X and Z are represented in Montgomery form, while r is
// not.
p384_felem Z2_mont;
p384_from_generic(Z2_mont, &p->Z);
p384_felem_mul(Z2_mont, Z2_mont, Z2_mont);
p384_felem r_Z2;
p384_felem_from_bytes(r_Z2, r->bytes); // r < order < p, so this is valid.
p384_felem_mul(r_Z2, r_Z2, Z2_mont);
p384_felem X;
p384_from_generic(X, &p->X);
p384_felem_from_mont(X, X);
if (OPENSSL_memcmp(&r_Z2, &X, sizeof(r_Z2)) == 0) {
return 1;
}
// During signing the x coefficient is reduced modulo the group order.
// Therefore there is a small possibility, less than 2^189/2^384 = 1/2^195,
// that group_order < p.x < p.
// In that case, we need not only to compare against |r| but also to
// compare against r+group_order.
assert(group->field.width == group->order.width);
if (bn_less_than_words(r->words, group->field_minus_order.words,
group->field.width)) {
// We can ignore the carry because: r + group_order < p < 2^384.
EC_FELEM tmp;
bn_add_words(tmp.words, r->words, group->order.d, group->order.width);
p384_from_generic(r_Z2, &tmp);
p384_felem_mul(r_Z2, r_Z2, Z2_mont);
if (OPENSSL_memcmp(&r_Z2, &X, sizeof(r_Z2)) == 0) {
return 1;
}
}
return 0;
}
// ----------------------------------------------------------------------------
// SCALAR MULTIPLICATION OPERATIONS
// ----------------------------------------------------------------------------
//
// The method for computing scalar products in functions:
// - |ec_GFp_nistp384_point_mul|,
// - |ec_GFp_nistp384_point_mul_base|,
// - |ec_GFp_nistp384_point_mul_public|,
// is adapted from ECCKiila project (https://arxiv.org/abs/2007.11481).
//
// One difference from the processing in the ECCKiila project is the order of
// the digit processing in |ec_GFp_nistp384_point_mul_base|, where we end the
// processing with the least significant digit to be able to apply the
// analysis results detailed at the bottom of this file. In
// |ec_GFp_nistp384_point_mul_base| and |ec_GFp_nistp384_point_mul|, we
// considered using window size 7 based on that same analysis. However, the
// table size and performance measurements were more preferable for window
// size 5. The potential issue with different window sizes is that for some
// sizes, a scalar can be found such that a case of point doubling instead of
// point addition happens in the scalar multiplication. This would make
// the multiplication non constant-time. To the best of our knowledge this
// timing leak is not an exploitable issue because the only scalar for which
// the leak can happen is already known by the attacker. This is also provided
// that this recoding and window size are only used with ECDH and ECDSA
// protocols. Any other use would need to be analyzed to determine whether it is
// secure and the user should be aware of this side channel of a particular
// scalar value.
//
// OpenSSL has a similar analysis for P-521 implementation:
// https://github.com/openssl/openssl/blob/e9492d1cecf459261f1f5ac0eb03e9c631600537/crypto/ec/ecp_nistp521.c#L1318
//
// For detailed analysis of different window sizes see the bottom of this file.
// p384_get_bit returns the |i|-th bit in |in|
static crypto_word_t p384_get_bit(const uint8_t *in, int i) {
if (i < 0 || i >= 384) {
return 0;
}
return (in[i >> 3] >> (i & 7)) & 1;
}
// Constants for scalar encoding in the scalar multiplication functions.
#define P384_MUL_WSIZE (5) // window size w
// Assert the window size is 5 because the pre-computed table in |p384_table.h|
// is generated for window size 5.
OPENSSL_STATIC_ASSERT(P384_MUL_WSIZE == 5,
p384_scalar_mul_window_size_is_not_equal_to_five)
#define P384_MUL_TWO_TO_WSIZE (1 << P384_MUL_WSIZE)
#define P384_MUL_WSIZE_MASK ((P384_MUL_TWO_TO_WSIZE << 1) - 1)
// Number of |P384_MUL_WSIZE|-bit windows in a 384-bit value
#define P384_MUL_NWINDOWS ((384 + P384_MUL_WSIZE - 1)/P384_MUL_WSIZE)
// For the public point in |ec_GFp_nistp384_point_mul_public| function
// we use window size w = 5.
#define P384_MUL_PUB_WSIZE (5)
// We keep only odd multiples in tables, hence the table size is (2^w)/2
#define P384_MUL_TABLE_SIZE (P384_MUL_TWO_TO_WSIZE >> 1)
#define P384_MUL_PUB_TABLE_SIZE (1 << (P384_MUL_PUB_WSIZE - 1))
// Compute "regular" wNAF representation of a scalar, see
// Joye, Tunstall, "Exponent Recoding and Regular Exponentiation Algorithms",
// AfricaCrypt 2009, Alg 6.
// It forces an odd scalar and outputs digits in
// {\pm 1, \pm 3, \pm 5, \pm 7, \pm 9, ...}
// i.e. signed odd digits with _no zeroes_ -- that makes it "regular".
static void p384_felem_mul_scalar_rwnaf(int16_t *out, const unsigned char *in) {
int16_t window, d;
window = (in[0] & P384_MUL_WSIZE_MASK) | 1;
for (size_t i = 0; i < P384_MUL_NWINDOWS - 1; i++) {
d = (window & P384_MUL_WSIZE_MASK) - P384_MUL_TWO_TO_WSIZE;
out[i] = d;
window = (window - d) >> P384_MUL_WSIZE;
for (size_t j = 1; j <= P384_MUL_WSIZE; j++) {
window += p384_get_bit(in, (i + 1) * P384_MUL_WSIZE + j) << j;
}
}
out[P384_MUL_NWINDOWS - 1] = window;
}
// p384_select_point selects the |idx|-th projective point from the given
// precomputed table and copies it to |out| in constant time.
static void p384_select_point(p384_felem out[3],
size_t idx,
p384_felem table[][3],
size_t table_size) {
OPENSSL_memset(out, 0, sizeof(p384_felem) * 3);
for (size_t i = 0; i < table_size; i++) {
p384_limb_t mismatch = i ^ idx;
p384_felem_cmovznz(out[0], mismatch, table[i][0], out[0]);
p384_felem_cmovznz(out[1], mismatch, table[i][1], out[1]);
p384_felem_cmovznz(out[2], mismatch, table[i][2], out[2]);
}
}
// p384_select_point_affine selects the |idx|-th affine point from
// the given precomputed table and copies it to |out| in constant-time.
static void p384_select_point_affine(p384_felem out[2],
size_t idx,
const p384_felem table[][2],
size_t table_size) {
OPENSSL_memset(out, 0, sizeof(p384_felem) * 2);
for (size_t i = 0; i < table_size; i++) {
p384_limb_t mismatch = i ^ idx;
p384_felem_cmovznz(out[0], mismatch, table[i][0], out[0]);
p384_felem_cmovznz(out[1], mismatch, table[i][1], out[1]);
}
}
// Multiplication of a point by a scalar, r = [scalar]P.
// The product is computed with the use of a small table generated on-the-fly
// and the scalar recoded in the regular-wNAF representation.
//
// The precomputed (on-the-fly) table |p_pre_comp| holds 16 odd multiples of P:
// [2i + 1]P for i in [0, 15].
// Computing the negation of a point P = (x, y) is relatively easy:
// -P = (x, -y).
// So we may assume that instead of the above-mentioned 64, we have 128 points:
// [\pm 1]P, [\pm 3]P, [\pm 5]P, ..., [\pm 31]P.
//
// The 384-bit scalar is recoded (regular-wNAF encoding) into 77 signed digits
// each of length 5 bits, as explained in the |p384_felem_mul_scalar_rwnaf|
// function. Namely,
// scalar' = s_0 + s_1*2^5 + s_2*2^10 + ... + s_76*2^380,
// where digits s_i are in [\pm 1, \pm 3, ..., \pm 31]. Note that for an odd
// scalar we have that scalar = scalar', while in the case of an even
// scalar we have that scalar = scalar' - 1.
//
// The required product, [scalar]P, is computed by the following algorithm.
// 1. Initialize the accumulator with the point from |p_pre_comp|
// corresponding to the most significant digit s_76 of the scalar.
// 2. For digits s_i starting from s_75 down to s_0:
// 3. Double the accumulator 5 times. (note that doubling a point [a]P
// seven times results in [2^5*a]P).
// 4. Read from |p_pre_comp| the point corresponding to abs(s_i),
// negate it if s_i is negative, and add it to the accumulator.
//
// Note: this function is constant-time.
static void ec_GFp_nistp384_point_mul(const EC_GROUP *group, EC_RAW_POINT *r,
const EC_RAW_POINT *p,
const EC_SCALAR *scalar) {
p384_felem res[3] = {{0}, {0}, {0}}, tmp[3] = {{0}, {0}, {0}}, ftmp;
// Table of multiples of P: [2i + 1]P for i in [0, 15].
p384_felem p_pre_comp[P384_MUL_TABLE_SIZE][3];
// Set the first point in the table to P.
p384_from_generic(p_pre_comp[0][0], &p->X);
p384_from_generic(p_pre_comp[0][1], &p->Y);
p384_from_generic(p_pre_comp[0][2], &p->Z);
// Compute tmp = [2]P.
p384_point_double(tmp[0], tmp[1], tmp[2],
p_pre_comp[0][0], p_pre_comp[0][1], p_pre_comp[0][2]);
// Generate the remaining 15 multiples of P.
for (size_t i = 1; i < P384_MUL_TABLE_SIZE; i++) {
p384_point_add(p_pre_comp[i][0], p_pre_comp[i][1], p_pre_comp[i][2],
tmp[0], tmp[1], tmp[2], 0 /* both Jacobian */,
p_pre_comp[i - 1][0],
p_pre_comp[i - 1][1],
p_pre_comp[i - 1][2]);
}
// Recode the scalar.
int16_t rnaf[P384_MUL_NWINDOWS] = {0};
p384_felem_mul_scalar_rwnaf(rnaf, scalar->bytes);
// Initialize the accumulator |res| with the table entry corresponding to
// the most significant digit of the recoded scalar (note that this digit
// can't be negative).
int16_t idx = rnaf[P384_MUL_NWINDOWS - 1] >> 1;
p384_select_point(res, idx, p_pre_comp, P384_MUL_TABLE_SIZE);
// Process the remaining digits of the scalar.
for (int i = P384_MUL_NWINDOWS - 2; i >= 0; i--) {
// Double |res| 5 times in each iteration.
for (size_t j = 0; j < P384_MUL_WSIZE; j++) {
p384_point_double(res[0], res[1], res[2], res[0], res[1], res[2]);
}
int16_t d = rnaf[i];
// is_neg = (d < 0) ? 1 : 0
int16_t is_neg = (d >> 15) & 1;
// d = abs(d)
d = (d ^ -is_neg) + is_neg;
idx = d >> 1;
// Select the point to add, in constant time.
p384_select_point(tmp, idx, p_pre_comp, P384_MUL_TABLE_SIZE);
// Negate y coordinate of the point tmp = (x, y); ftmp = -y.
p384_felem_opp(ftmp, tmp[1]);
// Conditionally select y or -y depending on the sign of the digit |d|.
p384_felem_cmovznz(tmp[1], is_neg, tmp[1], ftmp);
// Add the point to the accumulator |res|.
p384_point_add(res[0], res[1], res[2], res[0], res[1], res[2],
0 /* both Jacobian */, tmp[0], tmp[1], tmp[2]);
}
// Conditionally subtract P if the scalar is even, in constant-time.
// First, compute |tmp| = |res| + (-P).
p384_felem_copy(tmp[0], p_pre_comp[0][0]);
p384_felem_opp(tmp[1], p_pre_comp[0][1]);
p384_felem_copy(tmp[2], p_pre_comp[0][2]);
p384_point_add(tmp[0], tmp[1], tmp[2], res[0], res[1], res[2],
0 /* both Jacobian */, tmp[0], tmp[1], tmp[2]);
// Select |res| or |tmp| based on the |scalar| parity, in constant-time.
p384_felem_cmovznz(res[0], scalar->bytes[0] & 1, tmp[0], res[0]);
p384_felem_cmovznz(res[1], scalar->bytes[0] & 1, tmp[1], res[1]);
p384_felem_cmovznz(res[2], scalar->bytes[0] & 1, tmp[2], res[2]);
// Copy the result to the output.
p384_to_generic(&r->X, res[0]);
p384_to_generic(&r->Y, res[1]);
p384_to_generic(&r->Z, res[2]);
}
// Include the precomputed table for the based point scalar multiplication.
#include "p384_table.h"
// Multiplication of the base point G of P-384 curve with the given scalar.
// The product is computed with the Comb method using the precomputed table
// |p384_g_pre_comp| from |p384_table.h| file and the regular-wNAF scalar
// encoding.
//
// The |p384_g_pre_comp| table has 20 sub-tables each holding 16 points:
// 0 : [1]G, [3]G, ..., [31]G
// 1 : [1*2^20]G, [3*2^20]G, ..., [31*2^20]G
// ...
// i : [1*2^20i]G, [3*2^20i]G, ..., [31*2^20i]G
// ...
// 19 : [2^380]G, [3*2^380]G, ..., [31*2^380]G.
// Computing the negation of a point P = (x, y) is relatively easy:
// -P = (x, -y).
// So we may assume that for each sub-table we have 32 points instead of 16:
// [\pm 1*2^20i]G, [\pm 3*2^20i]G, ..., [\pm 31*2^20i]G.
//
// The 384-bit |scalar| is recoded (regular-wNAF encoding) into 77 signed
// digits, each of length 5 bits, as explained in the
// |p384_felem_mul_scalar_rwnaf| function. Namely,
// scalar' = s_0 + s_1*2^5 + s_2*2^10 + ... + s_76*2^380,
// where digits s_i are in [\pm 1, \pm 3, ..., \pm 31]. Note that for an odd
// scalar we have that scalar = scalar', while in the case of an even
// scalar we have that scalar = scalar' - 1.
//
// To compute the required product, [scalar]G, we may do the following.
// Group the recoded digits of the scalar in 4 groups:
// | corresponding multiples in
// digits | the recoded representation
// -------------------------------------------------------------------------
// (0): {s_0, s_4, s_8, ..., s_72, s_76} | { 2^0, 2^20, ..., 2^360, 2^380}
// (1): {s_1, s_5, s_9, ..., s_73} | { 2^5, 2^25, ..., 2^365}
// (2): {s_2, s_6, s_10, ..., s_74} | {2^10, 2^30, ..., 2^370}
// (3): {s_3, s_7, s_11, ..., s_75} | {2^15, 2^35, ..., 2^375}
// corresponding sub-table lookup | { T0, T1, ..., T18, T19}
//
// The group (0) digits correspond precisely to the multiples of G that are
// held in the 20 precomputed sub-tables, so we may simply read the appropriate
// points from the sub-tables and sum them all up (negating if needed, i.e., if
// a digit s_i is negative, we read the point corresponding to the abs(s_i) and
// negate it before adding it to the sum).
// The remaining three groups (1), (2), and (3), correspond to the multiples
// of G from the sub-tables multiplied additionally by 2^5, 2^10, and 2^15,
// respectively. Therefore, for these groups we may read the appropriate points
// from the table, double them 5, 10, or 15 times, respectively, and add them
// to the final result.
//
// To minimize the number of required doubling operations we process the digits
// of the scalar from left to right. In other words, the algorithm is:
// 1. Read the points corresponding to the group (3) digits from the table
// and add them to an accumulator.
// 2. Double the accumulator 5 times.
// 3. Repeat steps 1. and 2. for groups (2) and (1),
// and perform step 1. for group (0).
// 4. If the scalar is even subtract G from the accumulator.
//
// Note: this function is constant-time.
static void ec_GFp_nistp384_point_mul_base(const EC_GROUP *group,
EC_RAW_POINT *r,
const EC_SCALAR *scalar) {
p384_felem res[3] = {{0}, {0}, {0}}, tmp[3] = {{0}, {0}, {0}}, ftmp;
int16_t rnaf[P384_MUL_NWINDOWS] = {0};
// Recode the scalar.
p384_felem_mul_scalar_rwnaf(rnaf, scalar->bytes);
// Process the 4 groups of digits starting from group (3) down to group (0).
for (int i = 3; i >= 0; i--) {
// Double |res| 5 times in each iteration, except in the first one.
for (int j = 0; i != 3 && j < P384_MUL_WSIZE; j++) {
p384_point_double(res[0], res[1], res[2], res[0], res[1], res[2]);
}
// Process the digits in the current group from the most to the least
// significant one (this is a requirement to ensure that the case of point
// doubling can't happen).
// For group (3) we process digits s_75 to s_3, for group (2) s_74 to s_2,
// group (1) s_73 to s_1, and for group (0) s_76 to s_0.
const size_t start_idx = ((P384_MUL_NWINDOWS - i - 1)/4)*4 + i;
for (int j = start_idx; j >= 0; j -= 4) {
// For each digit |d| in the current group read the corresponding point
// from the table and add it to |res|. If |d| is negative, negate
// the point before adding it to |res|.
int16_t d = rnaf[j];
// is_neg = (d < 0) ? 1 : 0
int16_t is_neg = (d >> 15) & 1;
// d = abs(d)
d = (d ^ -is_neg) + is_neg;
int16_t idx = d >> 1;
// Select the point to add, in constant time.
p384_select_point_affine(tmp, idx, p384_g_pre_comp[j / 4],
P384_MUL_TABLE_SIZE);
// Negate y coordinate of the point tmp = (x, y); ftmp = -y.
p384_felem_opp(ftmp, tmp[1]);
// Conditionally select y or -y depending on the sign of the digit |d|.
p384_felem_cmovznz(tmp[1], is_neg, tmp[1], ftmp);
// Add the point to the accumulator |res|.
// Note that the points in the pre-computed table are given with affine
// coordinates. The point addition function computes a sum of two points,
// either both given in projective, or one in projective and the other one
// in affine coordinates. The |mixed| flag indicates the latter option,
// in which case we set the third coordinate of the second point to one.
p384_point_add(res[0], res[1], res[2], res[0], res[1], res[2],
1 /* mixed */, tmp[0], tmp[1], p384_felem_one);
}
}
// Conditionally subtract G if the scalar is even, in constant-time.
// First, compute |tmp| = |res| + (-G).
p384_felem_copy(tmp[0], p384_g_pre_comp[0][0][0]);
p384_felem_opp(tmp[1], p384_g_pre_comp[0][0][1]);
p384_point_add(tmp[0], tmp[1], tmp[2], res[0], res[1], res[2],
1 /* mixed */, tmp[0], tmp[1], p384_felem_one);
// Select |res| or |tmp| based on the |scalar| parity.
p384_felem_cmovznz(res[0], scalar->bytes[0] & 1, tmp[0], res[0]);
p384_felem_cmovznz(res[1], scalar->bytes[0] & 1, tmp[1], res[1]);
p384_felem_cmovznz(res[2], scalar->bytes[0] & 1, tmp[2], res[2]);
// Copy the result to the output.
p384_to_generic(&r->X, res[0]);
p384_to_generic(&r->Y, res[1]);
p384_to_generic(&r->Z, res[2]);
}
// Computes [g_scalar]G + [p_scalar]P, where G is the base point of the P-384
// curve, and P is the given point |p|.
//
// Both scalar products are computed by the same "textbook" wNAF method,
// with w = 5 for g_scalar and w = 5 for p_scalar.
// For the base point G product we use the first sub-table of the precomputed
// table |p384_g_pre_comp| from |p384_table.h| file, while for P we generate
// |p_pre_comp| table on-the-fly. The tables hold the first 16 odd multiples
// of G or P:
// g_pre_comp = {[1]G, [3]G, ..., [31]G},
// p_pre_comp = {[1]P, [3]P, ..., [31]P}.
// Computing the negation of a point P = (x, y) is relatively easy:
// -P = (x, -y).
// So we may assume that we also have the negatives of the points in the tables.
//
// The 384-bit scalars are recoded by the textbook wNAF method to 385 digits,
// where a digit is either a zero or an odd integer in [-31, 31]. The method
// guarantees that each non-zero digit is followed by at least four
// zeroes.
//
// The result [g_scalar]G + [p_scalar]P is computed by the following algorithm:
// 1. Initialize the accumulator with the point-at-infinity.
// 2. For i starting from 384 down to 0:
// 3. Double the accumulator (doubling can be skipped while the
// accumulator is equal to the point-at-infinity).