1 /* Copyright (c) 2015, Google Inc.
2  *
3  * Permission to use, copy, modify, and/or distribute this software for any
4  * purpose with or without fee is hereby granted, provided that the above
5  * copyright notice and this permission notice appear in all copies.
6  *
7  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10  * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12  * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13  * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14 
15 /* A 64-bit implementation of the NIST P-256 elliptic curve point
16  * multiplication
17  *
18  * OpenSSL integration was taken from Emilia Kasper's work in ecp_nistp224.c.
19  * Otherwise based on Emilia's P224 work, which was inspired by my curve25519
20  * work which got its smarts from Daniel J. Bernstein's work on the same. */
21 
22 #include <openssl/base.h>
23 
24 #if defined(OPENSSL_64_BIT) && !defined(OPENSSL_WINDOWS)
25 
26 #include <openssl/bn.h>
27 #include <openssl/ec.h>
28 #include <openssl/err.h>
29 #include <openssl/mem.h>
30 #include <openssl/obj.h>
31 
32 #include <string.h>
33 
34 #include "internal.h"
35 
36 
37 typedef uint8_t u8;
38 typedef uint64_t u64;
39 typedef int64_t s64;
40 typedef __uint128_t uint128_t;
41 typedef __int128_t int128_t;
42 
43 /* The underlying field. P256 operates over GF(2^256-2^224+2^192+2^96-1). We
44  * can serialise an element of this field into 32 bytes. We call this an
45  * felem_bytearray. */
46 typedef u8 felem_bytearray[32];
47 
48 /* These are the parameters of P256, taken from FIPS 186-3, page 86. These
49  * values are big-endian. */
50 static const felem_bytearray nistp256_curve_params[5] = {
51     {0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x01, /* p */
52      0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
53      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
54     {0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x01, /* a = -3 */
55      0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
56      0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
57      0xfc}, /* b */
58     {0x5a, 0xc6, 0x35, 0xd8, 0xaa, 0x3a, 0x93, 0xe7, 0xb3, 0xeb, 0xbd, 0x55,
59      0x76, 0x98, 0x86, 0xbc, 0x65, 0x1d, 0x06, 0xb0, 0xcc, 0x53, 0xb0, 0xf6,
60      0x3b, 0xce, 0x3c, 0x3e, 0x27, 0xd2, 0x60, 0x4b},
61     {0x6b, 0x17, 0xd1, 0xf2, 0xe1, 0x2c, 0x42, 0x47, /* x */
62      0xf8, 0xbc, 0xe6, 0xe5, 0x63, 0xa4, 0x40, 0xf2, 0x77, 0x03, 0x7d, 0x81,
63      0x2d, 0xeb, 0x33, 0xa0, 0xf4, 0xa1, 0x39, 0x45, 0xd8, 0x98, 0xc2, 0x96},
64     {0x4f, 0xe3, 0x42, 0xe2, 0xfe, 0x1a, 0x7f, 0x9b, /* y */
65      0x8e, 0xe7, 0xeb, 0x4a, 0x7c, 0x0f, 0x9e, 0x16, 0x2b, 0xce, 0x33, 0x57,
66      0x6b, 0x31, 0x5e, 0xce, 0xcb, 0xb6, 0x40, 0x68, 0x37, 0xbf, 0x51, 0xf5}};
67 
68 /* The representation of field elements.
69  * ------------------------------------
70  *
71  * We represent field elements with either four 128-bit values, eight 128-bit
72  * values, or four 64-bit values. The field element represented is:
73  *   v[0]*2^0 + v[1]*2^64 + v[2]*2^128 + v[3]*2^192  (mod p)
74  * or:
75  *   v[0]*2^0 + v[1]*2^64 + v[2]*2^128 + ... + v[8]*2^512  (mod p)
76  *
77  * 128-bit values are called 'limbs'. Since the limbs are spaced only 64 bits
78  * apart, but are 128-bits wide, the most significant bits of each limb overlap
79  * with the least significant bits of the next.
80  *
81  * A field element with four limbs is an 'felem'. One with eight limbs is a
82  * 'longfelem'
83  *
84  * A field element with four, 64-bit values is called a 'smallfelem'. Small
85  * values are used as intermediate values before multiplication. */
86 
87 #define NLIMBS 4
88 
89 typedef uint128_t limb;
90 typedef limb felem[NLIMBS];
91 typedef limb longfelem[NLIMBS * 2];
92 typedef u64 smallfelem[NLIMBS];
93 
94 /* This is the value of the prime as four 64-bit words, little-endian. */
95 static const u64 kPrime[4] = {0xfffffffffffffffful, 0xffffffff, 0,
96                               0xffffffff00000001ul};
97 static const u64 bottom63bits = 0x7ffffffffffffffful;
98 
99 /* bin32_to_felem takes a little-endian byte array and converts it into felem
100  * form. This assumes that the CPU is little-endian. */
bin32_to_felem(felem out,const u8 in[32])101 static void bin32_to_felem(felem out, const u8 in[32]) {
102   out[0] = *((u64 *)&in[0]);
103   out[1] = *((u64 *)&in[8]);
104   out[2] = *((u64 *)&in[16]);
105   out[3] = *((u64 *)&in[24]);
106 }
107 
108 /* smallfelem_to_bin32 takes a smallfelem and serialises into a little endian,
109  * 32 byte array. This assumes that the CPU is little-endian. */
smallfelem_to_bin32(u8 out[32],const smallfelem in)110 static void smallfelem_to_bin32(u8 out[32], const smallfelem in) {
111   *((u64 *)&out[0]) = in[0];
112   *((u64 *)&out[8]) = in[1];
113   *((u64 *)&out[16]) = in[2];
114   *((u64 *)&out[24]) = in[3];
115 }
116 
117 /* To preserve endianness when using BN_bn2bin and BN_bin2bn. */
flip_endian(u8 * out,const u8 * in,unsigned len)118 static void flip_endian(u8 *out, const u8 *in, unsigned len) {
119   unsigned i;
120   for (i = 0; i < len; ++i) {
121     out[i] = in[len - 1 - i];
122   }
123 }
124 
125 /* BN_to_felem converts an OpenSSL BIGNUM into an felem. */
BN_to_felem(felem out,const BIGNUM * bn)126 static int BN_to_felem(felem out, const BIGNUM *bn) {
127   if (BN_is_negative(bn)) {
128     OPENSSL_PUT_ERROR(EC, BN_to_felem, EC_R_BIGNUM_OUT_OF_RANGE);
129     return 0;
130   }
131 
132   felem_bytearray b_out;
133   /* BN_bn2bin eats leading zeroes */
134   memset(b_out, 0, sizeof(b_out));
135   unsigned num_bytes = BN_num_bytes(bn);
136   if (num_bytes > sizeof(b_out)) {
137     OPENSSL_PUT_ERROR(EC, BN_to_felem, EC_R_BIGNUM_OUT_OF_RANGE);
138     return 0;
139   }
140 
141   felem_bytearray b_in;
142   num_bytes = BN_bn2bin(bn, b_in);
143   flip_endian(b_out, b_in, num_bytes);
144   bin32_to_felem(out, b_out);
145   return 1;
146 }
147 
148 /* felem_to_BN converts an felem into an OpenSSL BIGNUM. */
smallfelem_to_BN(BIGNUM * out,const smallfelem in)149 static BIGNUM *smallfelem_to_BN(BIGNUM *out, const smallfelem in) {
150   felem_bytearray b_in, b_out;
151   smallfelem_to_bin32(b_in, in);
152   flip_endian(b_out, b_in, sizeof(b_out));
153   return BN_bin2bn(b_out, sizeof(b_out), out);
154 }
155 
156 /* Field operations. */
157 
smallfelem_one(smallfelem out)158 static void smallfelem_one(smallfelem out) {
159   out[0] = 1;
160   out[1] = 0;
161   out[2] = 0;
162   out[3] = 0;
163 }
164 
smallfelem_assign(smallfelem out,const smallfelem in)165 static void smallfelem_assign(smallfelem out, const smallfelem in) {
166   out[0] = in[0];
167   out[1] = in[1];
168   out[2] = in[2];
169   out[3] = in[3];
170 }
171 
felem_assign(felem out,const felem in)172 static void felem_assign(felem out, const felem in) {
173   out[0] = in[0];
174   out[1] = in[1];
175   out[2] = in[2];
176   out[3] = in[3];
177 }
178 
179 /* felem_sum sets out = out + in. */
felem_sum(felem out,const felem in)180 static void felem_sum(felem out, const felem in) {
181   out[0] += in[0];
182   out[1] += in[1];
183   out[2] += in[2];
184   out[3] += in[3];
185 }
186 
187 /* felem_small_sum sets out = out + in. */
felem_small_sum(felem out,const smallfelem in)188 static void felem_small_sum(felem out, const smallfelem in) {
189   out[0] += in[0];
190   out[1] += in[1];
191   out[2] += in[2];
192   out[3] += in[3];
193 }
194 
195 /* felem_scalar sets out = out * scalar */
felem_scalar(felem out,const u64 scalar)196 static void felem_scalar(felem out, const u64 scalar) {
197   out[0] *= scalar;
198   out[1] *= scalar;
199   out[2] *= scalar;
200   out[3] *= scalar;
201 }
202 
203 /* longfelem_scalar sets out = out * scalar */
longfelem_scalar(longfelem out,const u64 scalar)204 static void longfelem_scalar(longfelem out, const u64 scalar) {
205   out[0] *= scalar;
206   out[1] *= scalar;
207   out[2] *= scalar;
208   out[3] *= scalar;
209   out[4] *= scalar;
210   out[5] *= scalar;
211   out[6] *= scalar;
212   out[7] *= scalar;
213 }
214 
215 #define two105m41m9 (((limb)1) << 105) - (((limb)1) << 41) - (((limb)1) << 9)
216 #define two105 (((limb)1) << 105)
217 #define two105m41p9 (((limb)1) << 105) - (((limb)1) << 41) + (((limb)1) << 9)
218 
219 /* zero105 is 0 mod p */
220 static const felem zero105 = {two105m41m9, two105, two105m41p9, two105m41p9};
221 
222 /* smallfelem_neg sets |out| to |-small|
223  * On exit:
224  *   out[i] < out[i] + 2^105 */
smallfelem_neg(felem out,const smallfelem small)225 static void smallfelem_neg(felem out, const smallfelem small) {
226   /* In order to prevent underflow, we subtract from 0 mod p. */
227   out[0] = zero105[0] - small[0];
228   out[1] = zero105[1] - small[1];
229   out[2] = zero105[2] - small[2];
230   out[3] = zero105[3] - small[3];
231 }
232 
233 /* felem_diff subtracts |in| from |out|
234  * On entry:
235  *   in[i] < 2^104
236  * On exit:
237  *   out[i] < out[i] + 2^105. */
felem_diff(felem out,const felem in)238 static void felem_diff(felem out, const felem in) {
239   /* In order to prevent underflow, we add 0 mod p before subtracting. */
240   out[0] += zero105[0];
241   out[1] += zero105[1];
242   out[2] += zero105[2];
243   out[3] += zero105[3];
244 
245   out[0] -= in[0];
246   out[1] -= in[1];
247   out[2] -= in[2];
248   out[3] -= in[3];
249 }
250 
251 #define two107m43m11 (((limb)1) << 107) - (((limb)1) << 43) - (((limb)1) << 11)
252 #define two107 (((limb)1) << 107)
253 #define two107m43p11 (((limb)1) << 107) - (((limb)1) << 43) + (((limb)1) << 11)
254 
255 /* zero107 is 0 mod p */
256 static const felem zero107 = {two107m43m11, two107, two107m43p11, two107m43p11};
257 
258 /* An alternative felem_diff for larger inputs |in|
259  * felem_diff_zero107 subtracts |in| from |out|
260  * On entry:
261  *   in[i] < 2^106
262  * On exit:
263  *   out[i] < out[i] + 2^107. */
felem_diff_zero107(felem out,const felem in)264 static void felem_diff_zero107(felem out, const felem in) {
265   /* In order to prevent underflow, we add 0 mod p before subtracting. */
266   out[0] += zero107[0];
267   out[1] += zero107[1];
268   out[2] += zero107[2];
269   out[3] += zero107[3];
270 
271   out[0] -= in[0];
272   out[1] -= in[1];
273   out[2] -= in[2];
274   out[3] -= in[3];
275 }
276 
277 /* longfelem_diff subtracts |in| from |out|
278  * On entry:
279  *   in[i] < 7*2^67
280  * On exit:
281  *   out[i] < out[i] + 2^70 + 2^40. */
longfelem_diff(longfelem out,const longfelem in)282 static void longfelem_diff(longfelem out, const longfelem in) {
283   static const limb two70m8p6 =
284       (((limb)1) << 70) - (((limb)1) << 8) + (((limb)1) << 6);
285   static const limb two70p40 = (((limb)1) << 70) + (((limb)1) << 40);
286   static const limb two70 = (((limb)1) << 70);
287   static const limb two70m40m38p6 = (((limb)1) << 70) - (((limb)1) << 40) -
288                                     (((limb)1) << 38) + (((limb)1) << 6);
289   static const limb two70m6 = (((limb)1) << 70) - (((limb)1) << 6);
290 
291   /* add 0 mod p to avoid underflow */
292   out[0] += two70m8p6;
293   out[1] += two70p40;
294   out[2] += two70;
295   out[3] += two70m40m38p6;
296   out[4] += two70m6;
297   out[5] += two70m6;
298   out[6] += two70m6;
299   out[7] += two70m6;
300 
301   /* in[i] < 7*2^67 < 2^70 - 2^40 - 2^38 + 2^6 */
302   out[0] -= in[0];
303   out[1] -= in[1];
304   out[2] -= in[2];
305   out[3] -= in[3];
306   out[4] -= in[4];
307   out[5] -= in[5];
308   out[6] -= in[6];
309   out[7] -= in[7];
310 }
311 
312 #define two64m0 (((limb)1) << 64) - 1
313 #define two110p32m0 (((limb)1) << 110) + (((limb)1) << 32) - 1
314 #define two64m46 (((limb)1) << 64) - (((limb)1) << 46)
315 #define two64m32 (((limb)1) << 64) - (((limb)1) << 32)
316 
317 /* zero110 is 0 mod p. */
318 static const felem zero110 = {two64m0, two110p32m0, two64m46, two64m32};
319 
320 /* felem_shrink converts an felem into a smallfelem. The result isn't quite
321  * minimal as the value may be greater than p.
322  *
323  * On entry:
324  *   in[i] < 2^109
325  * On exit:
326  *   out[i] < 2^64. */
felem_shrink(smallfelem out,const felem in)327 static void felem_shrink(smallfelem out, const felem in) {
328   felem tmp;
329   u64 a, b, mask;
330   s64 high, low;
331   static const u64 kPrime3Test = 0x7fffffff00000001ul; /* 2^63 - 2^32 + 1 */
332 
333   /* Carry 2->3 */
334   tmp[3] = zero110[3] + in[3] + ((u64)(in[2] >> 64));
335   /* tmp[3] < 2^110 */
336 
337   tmp[2] = zero110[2] + (u64)in[2];
338   tmp[0] = zero110[0] + in[0];
339   tmp[1] = zero110[1] + in[1];
340   /* tmp[0] < 2**110, tmp[1] < 2^111, tmp[2] < 2**65 */
341 
342   /* We perform two partial reductions where we eliminate the high-word of
343    * tmp[3]. We don't update the other words till the end. */
344   a = tmp[3] >> 64; /* a < 2^46 */
345   tmp[3] = (u64)tmp[3];
346   tmp[3] -= a;
347   tmp[3] += ((limb)a) << 32;
348   /* tmp[3] < 2^79 */
349 
350   b = a;
351   a = tmp[3] >> 64; /* a < 2^15 */
352   b += a;           /* b < 2^46 + 2^15 < 2^47 */
353   tmp[3] = (u64)tmp[3];
354   tmp[3] -= a;
355   tmp[3] += ((limb)a) << 32;
356   /* tmp[3] < 2^64 + 2^47 */
357 
358   /* This adjusts the other two words to complete the two partial
359    * reductions. */
360   tmp[0] += b;
361   tmp[1] -= (((limb)b) << 32);
362 
363   /* In order to make space in tmp[3] for the carry from 2 -> 3, we
364    * conditionally subtract kPrime if tmp[3] is large enough. */
365   high = tmp[3] >> 64;
366   /* As tmp[3] < 2^65, high is either 1 or 0 */
367   high <<= 63;
368   high >>= 63;
369   /* high is:
370    *   all ones   if the high word of tmp[3] is 1
371    *   all zeros  if the high word of tmp[3] if 0 */
372   low = tmp[3];
373   mask = low >> 63;
374   /* mask is:
375    *   all ones   if the MSB of low is 1
376    *   all zeros  if the MSB of low if 0 */
377   low &= bottom63bits;
378   low -= kPrime3Test;
379   /* if low was greater than kPrime3Test then the MSB is zero */
380   low = ~low;
381   low >>= 63;
382   /* low is:
383    *   all ones   if low was > kPrime3Test
384    *   all zeros  if low was <= kPrime3Test */
385   mask = (mask & low) | high;
386   tmp[0] -= mask & kPrime[0];
387   tmp[1] -= mask & kPrime[1];
388   /* kPrime[2] is zero, so omitted */
389   tmp[3] -= mask & kPrime[3];
390   /* tmp[3] < 2**64 - 2**32 + 1 */
391 
392   tmp[1] += ((u64)(tmp[0] >> 64));
393   tmp[0] = (u64)tmp[0];
394   tmp[2] += ((u64)(tmp[1] >> 64));
395   tmp[1] = (u64)tmp[1];
396   tmp[3] += ((u64)(tmp[2] >> 64));
397   tmp[2] = (u64)tmp[2];
398   /* tmp[i] < 2^64 */
399 
400   out[0] = tmp[0];
401   out[1] = tmp[1];
402   out[2] = tmp[2];
403   out[3] = tmp[3];
404 }
405 
406 /* smallfelem_expand converts a smallfelem to an felem */
smallfelem_expand(felem out,const smallfelem in)407 static void smallfelem_expand(felem out, const smallfelem in) {
408   out[0] = in[0];
409   out[1] = in[1];
410   out[2] = in[2];
411   out[3] = in[3];
412 }
413 
414 /* smallfelem_square sets |out| = |small|^2
415  * On entry:
416  *   small[i] < 2^64
417  * On exit:
418  *   out[i] < 7 * 2^64 < 2^67 */
smallfelem_square(longfelem out,const smallfelem small)419 static void smallfelem_square(longfelem out, const smallfelem small) {
420   limb a;
421   u64 high, low;
422 
423   a = ((uint128_t)small[0]) * small[0];
424   low = a;
425   high = a >> 64;
426   out[0] = low;
427   out[1] = high;
428 
429   a = ((uint128_t)small[0]) * small[1];
430   low = a;
431   high = a >> 64;
432   out[1] += low;
433   out[1] += low;
434   out[2] = high;
435 
436   a = ((uint128_t)small[0]) * small[2];
437   low = a;
438   high = a >> 64;
439   out[2] += low;
440   out[2] *= 2;
441   out[3] = high;
442 
443   a = ((uint128_t)small[0]) * small[3];
444   low = a;
445   high = a >> 64;
446   out[3] += low;
447   out[4] = high;
448 
449   a = ((uint128_t)small[1]) * small[2];
450   low = a;
451   high = a >> 64;
452   out[3] += low;
453   out[3] *= 2;
454   out[4] += high;
455 
456   a = ((uint128_t)small[1]) * small[1];
457   low = a;
458   high = a >> 64;
459   out[2] += low;
460   out[3] += high;
461 
462   a = ((uint128_t)small[1]) * small[3];
463   low = a;
464   high = a >> 64;
465   out[4] += low;
466   out[4] *= 2;
467   out[5] = high;
468 
469   a = ((uint128_t)small[2]) * small[3];
470   low = a;
471   high = a >> 64;
472   out[5] += low;
473   out[5] *= 2;
474   out[6] = high;
475   out[6] += high;
476 
477   a = ((uint128_t)small[2]) * small[2];
478   low = a;
479   high = a >> 64;
480   out[4] += low;
481   out[5] += high;
482 
483   a = ((uint128_t)small[3]) * small[3];
484   low = a;
485   high = a >> 64;
486   out[6] += low;
487   out[7] = high;
488 }
489 
490 /*felem_square sets |out| = |in|^2
491  * On entry:
492  *   in[i] < 2^109
493  * On exit:
494  *   out[i] < 7 * 2^64 < 2^67. */
felem_square(longfelem out,const felem in)495 static void felem_square(longfelem out, const felem in) {
496   u64 small[4];
497   felem_shrink(small, in);
498   smallfelem_square(out, small);
499 }
500 
501 /* smallfelem_mul sets |out| = |small1| * |small2|
502  * On entry:
503  *   small1[i] < 2^64
504  *   small2[i] < 2^64
505  * On exit:
506  *   out[i] < 7 * 2^64 < 2^67. */
smallfelem_mul(longfelem out,const smallfelem small1,const smallfelem small2)507 static void smallfelem_mul(longfelem out, const smallfelem small1,
508                            const smallfelem small2) {
509   limb a;
510   u64 high, low;
511 
512   a = ((uint128_t)small1[0]) * small2[0];
513   low = a;
514   high = a >> 64;
515   out[0] = low;
516   out[1] = high;
517 
518   a = ((uint128_t)small1[0]) * small2[1];
519   low = a;
520   high = a >> 64;
521   out[1] += low;
522   out[2] = high;
523 
524   a = ((uint128_t)small1[1]) * small2[0];
525   low = a;
526   high = a >> 64;
527   out[1] += low;
528   out[2] += high;
529 
530   a = ((uint128_t)small1[0]) * small2[2];
531   low = a;
532   high = a >> 64;
533   out[2] += low;
534   out[3] = high;
535 
536   a = ((uint128_t)small1[1]) * small2[1];
537   low = a;
538   high = a >> 64;
539   out[2] += low;
540   out[3] += high;
541 
542   a = ((uint128_t)small1[2]) * small2[0];
543   low = a;
544   high = a >> 64;
545   out[2] += low;
546   out[3] += high;
547 
548   a = ((uint128_t)small1[0]) * small2[3];
549   low = a;
550   high = a >> 64;
551   out[3] += low;
552   out[4] = high;
553 
554   a = ((uint128_t)small1[1]) * small2[2];
555   low = a;
556   high = a >> 64;
557   out[3] += low;
558   out[4] += high;
559 
560   a = ((uint128_t)small1[2]) * small2[1];
561   low = a;
562   high = a >> 64;
563   out[3] += low;
564   out[4] += high;
565 
566   a = ((uint128_t)small1[3]) * small2[0];
567   low = a;
568   high = a >> 64;
569   out[3] += low;
570   out[4] += high;
571 
572   a = ((uint128_t)small1[1]) * small2[3];
573   low = a;
574   high = a >> 64;
575   out[4] += low;
576   out[5] = high;
577 
578   a = ((uint128_t)small1[2]) * small2[2];
579   low = a;
580   high = a >> 64;
581   out[4] += low;
582   out[5] += high;
583 
584   a = ((uint128_t)small1[3]) * small2[1];
585   low = a;
586   high = a >> 64;
587   out[4] += low;
588   out[5] += high;
589 
590   a = ((uint128_t)small1[2]) * small2[3];
591   low = a;
592   high = a >> 64;
593   out[5] += low;
594   out[6] = high;
595 
596   a = ((uint128_t)small1[3]) * small2[2];
597   low = a;
598   high = a >> 64;
599   out[5] += low;
600   out[6] += high;
601 
602   a = ((uint128_t)small1[3]) * small2[3];
603   low = a;
604   high = a >> 64;
605   out[6] += low;
606   out[7] = high;
607 }
608 
609 /* felem_mul sets |out| = |in1| * |in2|
610  * On entry:
611  *   in1[i] < 2^109
612  *   in2[i] < 2^109
613  * On exit:
614  *   out[i] < 7 * 2^64 < 2^67 */
felem_mul(longfelem out,const felem in1,const felem in2)615 static void felem_mul(longfelem out, const felem in1, const felem in2) {
616   smallfelem small1, small2;
617   felem_shrink(small1, in1);
618   felem_shrink(small2, in2);
619   smallfelem_mul(out, small1, small2);
620 }
621 
622 /* felem_small_mul sets |out| = |small1| * |in2|
623  * On entry:
624  *   small1[i] < 2^64
625  *   in2[i] < 2^109
626  * On exit:
627  *   out[i] < 7 * 2^64 < 2^67 */
felem_small_mul(longfelem out,const smallfelem small1,const felem in2)628 static void felem_small_mul(longfelem out, const smallfelem small1,
629                             const felem in2) {
630   smallfelem small2;
631   felem_shrink(small2, in2);
632   smallfelem_mul(out, small1, small2);
633 }
634 
635 #define two100m36m4 (((limb)1) << 100) - (((limb)1) << 36) - (((limb)1) << 4)
636 #define two100 (((limb)1) << 100)
637 #define two100m36p4 (((limb)1) << 100) - (((limb)1) << 36) + (((limb)1) << 4)
638 
639 /* zero100 is 0 mod p */
640 static const felem zero100 = {two100m36m4, two100, two100m36p4, two100m36p4};
641 
642 /* Internal function for the different flavours of felem_reduce.
643  * felem_reduce_ reduces the higher coefficients in[4]-in[7].
644  * On entry:
645  *   out[0] >= in[6] + 2^32*in[6] + in[7] + 2^32*in[7]
646  *   out[1] >= in[7] + 2^32*in[4]
647  *   out[2] >= in[5] + 2^32*in[5]
648  *   out[3] >= in[4] + 2^32*in[5] + 2^32*in[6]
649  * On exit:
650  *   out[0] <= out[0] + in[4] + 2^32*in[5]
651  *   out[1] <= out[1] + in[5] + 2^33*in[6]
652  *   out[2] <= out[2] + in[7] + 2*in[6] + 2^33*in[7]
653  *   out[3] <= out[3] + 2^32*in[4] + 3*in[7] */
felem_reduce_(felem out,const longfelem in)654 static void felem_reduce_(felem out, const longfelem in) {
655   int128_t c;
656   /* combine common terms from below */
657   c = in[4] + (in[5] << 32);
658   out[0] += c;
659   out[3] -= c;
660 
661   c = in[5] - in[7];
662   out[1] += c;
663   out[2] -= c;
664 
665   /* the remaining terms */
666   /* 256: [(0,1),(96,-1),(192,-1),(224,1)] */
667   out[1] -= (in[4] << 32);
668   out[3] += (in[4] << 32);
669 
670   /* 320: [(32,1),(64,1),(128,-1),(160,-1),(224,-1)] */
671   out[2] -= (in[5] << 32);
672 
673   /* 384: [(0,-1),(32,-1),(96,2),(128,2),(224,-1)] */
674   out[0] -= in[6];
675   out[0] -= (in[6] << 32);
676   out[1] += (in[6] << 33);
677   out[2] += (in[6] * 2);
678   out[3] -= (in[6] << 32);
679 
680   /* 448: [(0,-1),(32,-1),(64,-1),(128,1),(160,2),(192,3)] */
681   out[0] -= in[7];
682   out[0] -= (in[7] << 32);
683   out[2] += (in[7] << 33);
684   out[3] += (in[7] * 3);
685 }
686 
687 /* felem_reduce converts a longfelem into an felem.
688  * To be called directly after felem_square or felem_mul.
689  * On entry:
690  *   in[0] < 2^64, in[1] < 3*2^64, in[2] < 5*2^64, in[3] < 7*2^64
691  *   in[4] < 7*2^64, in[5] < 5*2^64, in[6] < 3*2^64, in[7] < 2*64
692  * On exit:
693  *   out[i] < 2^101 */
felem_reduce(felem out,const longfelem in)694 static void felem_reduce(felem out, const longfelem in) {
695   out[0] = zero100[0] + in[0];
696   out[1] = zero100[1] + in[1];
697   out[2] = zero100[2] + in[2];
698   out[3] = zero100[3] + in[3];
699 
700   felem_reduce_(out, in);
701 
702   /* out[0] > 2^100 - 2^36 - 2^4 - 3*2^64 - 3*2^96 - 2^64 - 2^96 > 0
703    * out[1] > 2^100 - 2^64 - 7*2^96 > 0
704    * out[2] > 2^100 - 2^36 + 2^4 - 5*2^64 - 5*2^96 > 0
705    * out[3] > 2^100 - 2^36 + 2^4 - 7*2^64 - 5*2^96 - 3*2^96 > 0
706    *
707    * out[0] < 2^100 + 2^64 + 7*2^64 + 5*2^96 < 2^101
708    * out[1] < 2^100 + 3*2^64 + 5*2^64 + 3*2^97 < 2^101
709    * out[2] < 2^100 + 5*2^64 + 2^64 + 3*2^65 + 2^97 < 2^101
710    * out[3] < 2^100 + 7*2^64 + 7*2^96 + 3*2^64 < 2^101 */
711 }
712 
713 /* felem_reduce_zero105 converts a larger longfelem into an felem.
714  * On entry:
715  *   in[0] < 2^71
716  * On exit:
717  *   out[i] < 2^106 */
felem_reduce_zero105(felem out,const longfelem in)718 static void felem_reduce_zero105(felem out, const longfelem in) {
719     out[0] = zero105[0] + in[0];
720     out[1] = zero105[1] + in[1];
721     out[2] = zero105[2] + in[2];
722     out[3] = zero105[3] + in[3];
723 
724     felem_reduce_(out, in);
725 
726     /* out[0] > 2^105 - 2^41 - 2^9 - 2^71 - 2^103 - 2^71 - 2^103 > 0
727      * out[1] > 2^105 - 2^71 - 2^103 > 0
728      * out[2] > 2^105 - 2^41 + 2^9 - 2^71 - 2^103 > 0
729      * out[3] > 2^105 - 2^41 + 2^9 - 2^71 - 2^103 - 2^103 > 0
730      *
731      * out[0] < 2^105 + 2^71 + 2^71 + 2^103 < 2^106
732      * out[1] < 2^105 + 2^71 + 2^71 + 2^103 < 2^106
733      * out[2] < 2^105 + 2^71 + 2^71 + 2^71 + 2^103 < 2^106
734      * out[3] < 2^105 + 2^71 + 2^103 + 2^71 < 2^106 */
735 }
736 
737 /* subtract_u64 sets *result = *result - v and *carry to one if the
738  * subtraction underflowed. */
subtract_u64(u64 * result,u64 * carry,u64 v)739 static void subtract_u64(u64 *result, u64 *carry, u64 v) {
740   uint128_t r = *result;
741   r -= v;
742   *carry = (r >> 64) & 1;
743   *result = (u64)r;
744 }
745 
746 /* felem_contract converts |in| to its unique, minimal representation. On
747  * entry: in[i] < 2^109. */
felem_contract(smallfelem out,const felem in)748 static void felem_contract(smallfelem out, const felem in) {
749   u64 all_equal_so_far = 0, result = 0;
750 
751   felem_shrink(out, in);
752   /* small is minimal except that the value might be > p */
753 
754   all_equal_so_far--;
755   /* We are doing a constant time test if out >= kPrime. We need to compare
756    * each u64, from most-significant to least significant. For each one, if
757    * all words so far have been equal (m is all ones) then a non-equal
758    * result is the answer. Otherwise we continue. */
759   unsigned i;
760   for (i = 3; i < 4; i--) {
761     u64 equal;
762     uint128_t a = ((uint128_t)kPrime[i]) - out[i];
763     /* if out[i] > kPrime[i] then a will underflow and the high 64-bits
764      * will all be set. */
765     result |= all_equal_so_far & ((u64)(a >> 64));
766 
767     /* if kPrime[i] == out[i] then |equal| will be all zeros and the
768      * decrement will make it all ones. */
769     equal = kPrime[i] ^ out[i];
770     equal--;
771     equal &= equal << 32;
772     equal &= equal << 16;
773     equal &= equal << 8;
774     equal &= equal << 4;
775     equal &= equal << 2;
776     equal &= equal << 1;
777     equal = ((s64)equal) >> 63;
778 
779     all_equal_so_far &= equal;
780   }
781 
782   /* if all_equal_so_far is still all ones then the two values are equal
783    * and so out >= kPrime is true. */
784   result |= all_equal_so_far;
785 
786   /* if out >= kPrime then we subtract kPrime. */
787   u64 carry;
788   subtract_u64(&out[0], &carry, result & kPrime[0]);
789   subtract_u64(&out[1], &carry, carry);
790   subtract_u64(&out[2], &carry, carry);
791   subtract_u64(&out[3], &carry, carry);
792 
793   subtract_u64(&out[1], &carry, result & kPrime[1]);
794   subtract_u64(&out[2], &carry, carry);
795   subtract_u64(&out[3], &carry, carry);
796 
797   subtract_u64(&out[2], &carry, result & kPrime[2]);
798   subtract_u64(&out[3], &carry, carry);
799 
800   subtract_u64(&out[3], &carry, result & kPrime[3]);
801 }
802 
smallfelem_square_contract(smallfelem out,const smallfelem in)803 static void smallfelem_square_contract(smallfelem out, const smallfelem in) {
804   longfelem longtmp;
805   felem tmp;
806 
807   smallfelem_square(longtmp, in);
808   felem_reduce(tmp, longtmp);
809   felem_contract(out, tmp);
810 }
811 
smallfelem_mul_contract(smallfelem out,const smallfelem in1,const smallfelem in2)812 static void smallfelem_mul_contract(smallfelem out, const smallfelem in1,
813                                     const smallfelem in2) {
814   longfelem longtmp;
815   felem tmp;
816 
817   smallfelem_mul(longtmp, in1, in2);
818   felem_reduce(tmp, longtmp);
819   felem_contract(out, tmp);
820 }
821 
822 /* felem_is_zero returns a limb with all bits set if |in| == 0 (mod p) and 0
823  * otherwise.
824  * On entry:
825  *   small[i] < 2^64 */
smallfelem_is_zero(const smallfelem small)826 static limb smallfelem_is_zero(const smallfelem small) {
827   limb result;
828   u64 is_p;
829 
830   u64 is_zero = small[0] | small[1] | small[2] | small[3];
831   is_zero--;
832   is_zero &= is_zero << 32;
833   is_zero &= is_zero << 16;
834   is_zero &= is_zero << 8;
835   is_zero &= is_zero << 4;
836   is_zero &= is_zero << 2;
837   is_zero &= is_zero << 1;
838   is_zero = ((s64)is_zero) >> 63;
839 
840   is_p = (small[0] ^ kPrime[0]) | (small[1] ^ kPrime[1]) |
841          (small[2] ^ kPrime[2]) | (small[3] ^ kPrime[3]);
842   is_p--;
843   is_p &= is_p << 32;
844   is_p &= is_p << 16;
845   is_p &= is_p << 8;
846   is_p &= is_p << 4;
847   is_p &= is_p << 2;
848   is_p &= is_p << 1;
849   is_p = ((s64)is_p) >> 63;
850 
851   is_zero |= is_p;
852 
853   result = is_zero;
854   result |= ((limb)is_zero) << 64;
855   return result;
856 }
857 
smallfelem_is_zero_int(const smallfelem small)858 static int smallfelem_is_zero_int(const smallfelem small) {
859   return (int)(smallfelem_is_zero(small) & ((limb)1));
860 }
861 
862 /* felem_inv calculates |out| = |in|^{-1}
863  *
864  * Based on Fermat's Little Theorem:
865  *   a^p = a (mod p)
866  *   a^{p-1} = 1 (mod p)
867  *   a^{p-2} = a^{-1} (mod p) */
felem_inv(felem out,const felem in)868 static void felem_inv(felem out, const felem in) {
869   felem ftmp, ftmp2;
870   /* each e_I will hold |in|^{2^I - 1} */
871   felem e2, e4, e8, e16, e32, e64;
872   longfelem tmp;
873   unsigned i;
874 
875   felem_square(tmp, in);
876   felem_reduce(ftmp, tmp); /* 2^1 */
877   felem_mul(tmp, in, ftmp);
878   felem_reduce(ftmp, tmp); /* 2^2 - 2^0 */
879   felem_assign(e2, ftmp);
880   felem_square(tmp, ftmp);
881   felem_reduce(ftmp, tmp); /* 2^3 - 2^1 */
882   felem_square(tmp, ftmp);
883   felem_reduce(ftmp, tmp); /* 2^4 - 2^2 */
884   felem_mul(tmp, ftmp, e2);
885   felem_reduce(ftmp, tmp); /* 2^4 - 2^0 */
886   felem_assign(e4, ftmp);
887   felem_square(tmp, ftmp);
888   felem_reduce(ftmp, tmp); /* 2^5 - 2^1 */
889   felem_square(tmp, ftmp);
890   felem_reduce(ftmp, tmp); /* 2^6 - 2^2 */
891   felem_square(tmp, ftmp);
892   felem_reduce(ftmp, tmp); /* 2^7 - 2^3 */
893   felem_square(tmp, ftmp);
894   felem_reduce(ftmp, tmp); /* 2^8 - 2^4 */
895   felem_mul(tmp, ftmp, e4);
896   felem_reduce(ftmp, tmp); /* 2^8 - 2^0 */
897   felem_assign(e8, ftmp);
898   for (i = 0; i < 8; i++) {
899     felem_square(tmp, ftmp);
900     felem_reduce(ftmp, tmp);
901   } /* 2^16 - 2^8 */
902   felem_mul(tmp, ftmp, e8);
903   felem_reduce(ftmp, tmp); /* 2^16 - 2^0 */
904   felem_assign(e16, ftmp);
905   for (i = 0; i < 16; i++) {
906     felem_square(tmp, ftmp);
907     felem_reduce(ftmp, tmp);
908   } /* 2^32 - 2^16 */
909   felem_mul(tmp, ftmp, e16);
910   felem_reduce(ftmp, tmp); /* 2^32 - 2^0 */
911   felem_assign(e32, ftmp);
912   for (i = 0; i < 32; i++) {
913     felem_square(tmp, ftmp);
914     felem_reduce(ftmp, tmp);
915   } /* 2^64 - 2^32 */
916   felem_assign(e64, ftmp);
917   felem_mul(tmp, ftmp, in);
918   felem_reduce(ftmp, tmp); /* 2^64 - 2^32 + 2^0 */
919   for (i = 0; i < 192; i++) {
920     felem_square(tmp, ftmp);
921     felem_reduce(ftmp, tmp);
922   } /* 2^256 - 2^224 + 2^192 */
923 
924   felem_mul(tmp, e64, e32);
925   felem_reduce(ftmp2, tmp); /* 2^64 - 2^0 */
926   for (i = 0; i < 16; i++) {
927     felem_square(tmp, ftmp2);
928     felem_reduce(ftmp2, tmp);
929   } /* 2^80 - 2^16 */
930   felem_mul(tmp, ftmp2, e16);
931   felem_reduce(ftmp2, tmp); /* 2^80 - 2^0 */
932   for (i = 0; i < 8; i++) {
933     felem_square(tmp, ftmp2);
934     felem_reduce(ftmp2, tmp);
935   } /* 2^88 - 2^8 */
936   felem_mul(tmp, ftmp2, e8);
937   felem_reduce(ftmp2, tmp); /* 2^88 - 2^0 */
938   for (i = 0; i < 4; i++) {
939     felem_square(tmp, ftmp2);
940     felem_reduce(ftmp2, tmp);
941   } /* 2^92 - 2^4 */
942   felem_mul(tmp, ftmp2, e4);
943   felem_reduce(ftmp2, tmp); /* 2^92 - 2^0 */
944   felem_square(tmp, ftmp2);
945   felem_reduce(ftmp2, tmp); /* 2^93 - 2^1 */
946   felem_square(tmp, ftmp2);
947   felem_reduce(ftmp2, tmp); /* 2^94 - 2^2 */
948   felem_mul(tmp, ftmp2, e2);
949   felem_reduce(ftmp2, tmp); /* 2^94 - 2^0 */
950   felem_square(tmp, ftmp2);
951   felem_reduce(ftmp2, tmp); /* 2^95 - 2^1 */
952   felem_square(tmp, ftmp2);
953   felem_reduce(ftmp2, tmp); /* 2^96 - 2^2 */
954   felem_mul(tmp, ftmp2, in);
955   felem_reduce(ftmp2, tmp); /* 2^96 - 3 */
956 
957   felem_mul(tmp, ftmp2, ftmp);
958   felem_reduce(out, tmp); /* 2^256 - 2^224 + 2^192 + 2^96 - 3 */
959 }
960 
smallfelem_inv_contract(smallfelem out,const smallfelem in)961 static void smallfelem_inv_contract(smallfelem out, const smallfelem in) {
962   felem tmp;
963 
964   smallfelem_expand(tmp, in);
965   felem_inv(tmp, tmp);
966   felem_contract(out, tmp);
967 }
968 
969 /* Group operations
970  * ----------------
971  *
972  * Building on top of the field operations we have the operations on the
973  * elliptic curve group itself. Points on the curve are represented in Jacobian
974  * coordinates. */
975 
976 /* point_double calculates 2*(x_in, y_in, z_in)
977  *
978  * The method is taken from:
979  *   http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
980  *
981  * Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed.
982  * while x_out == y_in is not (maybe this works, but it's not tested). */
point_double(felem x_out,felem y_out,felem z_out,const felem x_in,const felem y_in,const felem z_in)983 static void point_double(felem x_out, felem y_out, felem z_out,
984                          const felem x_in, const felem y_in, const felem z_in) {
985   longfelem tmp, tmp2;
986   felem delta, gamma, beta, alpha, ftmp, ftmp2;
987   smallfelem small1, small2;
988 
989   felem_assign(ftmp, x_in);
990   /* ftmp[i] < 2^106 */
991   felem_assign(ftmp2, x_in);
992   /* ftmp2[i] < 2^106 */
993 
994   /* delta = z^2 */
995   felem_square(tmp, z_in);
996   felem_reduce(delta, tmp);
997   /* delta[i] < 2^101 */
998 
999   /* gamma = y^2 */
1000   felem_square(tmp, y_in);
1001   felem_reduce(gamma, tmp);
1002   /* gamma[i] < 2^101 */
1003   felem_shrink(small1, gamma);
1004 
1005   /* beta = x*gamma */
1006   felem_small_mul(tmp, small1, x_in);
1007   felem_reduce(beta, tmp);
1008   /* beta[i] < 2^101 */
1009 
1010   /* alpha = 3*(x-delta)*(x+delta) */
1011   felem_diff(ftmp, delta);
1012   /* ftmp[i] < 2^105 + 2^106 < 2^107 */
1013   felem_sum(ftmp2, delta);
1014   /* ftmp2[i] < 2^105 + 2^106 < 2^107 */
1015   felem_scalar(ftmp2, 3);
1016   /* ftmp2[i] < 3 * 2^107 < 2^109 */
1017   felem_mul(tmp, ftmp, ftmp2);
1018   felem_reduce(alpha, tmp);
1019   /* alpha[i] < 2^101 */
1020   felem_shrink(small2, alpha);
1021 
1022   /* x' = alpha^2 - 8*beta */
1023   smallfelem_square(tmp, small2);
1024   felem_reduce(x_out, tmp);
1025   felem_assign(ftmp, beta);
1026   felem_scalar(ftmp, 8);
1027   /* ftmp[i] < 8 * 2^101 = 2^104 */
1028   felem_diff(x_out, ftmp);
1029   /* x_out[i] < 2^105 + 2^101 < 2^106 */
1030 
1031   /* z' = (y + z)^2 - gamma - delta */
1032   felem_sum(delta, gamma);
1033   /* delta[i] < 2^101 + 2^101 = 2^102 */
1034   felem_assign(ftmp, y_in);
1035   felem_sum(ftmp, z_in);
1036   /* ftmp[i] < 2^106 + 2^106 = 2^107 */
1037   felem_square(tmp, ftmp);
1038   felem_reduce(z_out, tmp);
1039   felem_diff(z_out, delta);
1040   /* z_out[i] < 2^105 + 2^101 < 2^106 */
1041 
1042   /* y' = alpha*(4*beta - x') - 8*gamma^2 */
1043   felem_scalar(beta, 4);
1044   /* beta[i] < 4 * 2^101 = 2^103 */
1045   felem_diff_zero107(beta, x_out);
1046   /* beta[i] < 2^107 + 2^103 < 2^108 */
1047   felem_small_mul(tmp, small2, beta);
1048   /* tmp[i] < 7 * 2^64 < 2^67 */
1049   smallfelem_square(tmp2, small1);
1050   /* tmp2[i] < 7 * 2^64 */
1051   longfelem_scalar(tmp2, 8);
1052   /* tmp2[i] < 8 * 7 * 2^64 = 7 * 2^67 */
1053   longfelem_diff(tmp, tmp2);
1054   /* tmp[i] < 2^67 + 2^70 + 2^40 < 2^71 */
1055   felem_reduce_zero105(y_out, tmp);
1056   /* y_out[i] < 2^106 */
1057 }
1058 
1059 /* point_double_small is the same as point_double, except that it operates on
1060  * smallfelems. */
point_double_small(smallfelem x_out,smallfelem y_out,smallfelem z_out,const smallfelem x_in,const smallfelem y_in,const smallfelem z_in)1061 static void point_double_small(smallfelem x_out, smallfelem y_out,
1062                                smallfelem z_out, const smallfelem x_in,
1063                                const smallfelem y_in, const smallfelem z_in) {
1064   felem felem_x_out, felem_y_out, felem_z_out;
1065   felem felem_x_in, felem_y_in, felem_z_in;
1066 
1067   smallfelem_expand(felem_x_in, x_in);
1068   smallfelem_expand(felem_y_in, y_in);
1069   smallfelem_expand(felem_z_in, z_in);
1070   point_double(felem_x_out, felem_y_out, felem_z_out, felem_x_in, felem_y_in,
1071                felem_z_in);
1072   felem_shrink(x_out, felem_x_out);
1073   felem_shrink(y_out, felem_y_out);
1074   felem_shrink(z_out, felem_z_out);
1075 }
1076 
1077 /* copy_conditional copies in to out iff mask is all ones. */
copy_conditional(felem out,const felem in,limb mask)1078 static void copy_conditional(felem out, const felem in, limb mask) {
1079   unsigned i;
1080   for (i = 0; i < NLIMBS; ++i) {
1081     const limb tmp = mask & (in[i] ^ out[i]);
1082     out[i] ^= tmp;
1083   }
1084 }
1085 
1086 /* copy_small_conditional copies in to out iff mask is all ones. */
copy_small_conditional(felem out,const smallfelem in,limb mask)1087 static void copy_small_conditional(felem out, const smallfelem in, limb mask) {
1088   unsigned i;
1089   const u64 mask64 = mask;
1090   for (i = 0; i < NLIMBS; ++i) {
1091     out[i] = ((limb)(in[i] & mask64)) | (out[i] & ~mask);
1092   }
1093 }
1094 
1095 /* point_add calcuates (x1, y1, z1) + (x2, y2, z2)
1096  *
1097  * The method is taken from:
1098  *   http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl,
1099  * adapted for mixed addition (z2 = 1, or z2 = 0 for the point at infinity).
1100  *
1101  * This function includes a branch for checking whether the two input points
1102  * are equal, (while not equal to the point at infinity). This case never
1103  * happens during single point multiplication, so there is no timing leak for
1104  * ECDH or ECDSA signing. */
point_add(felem x3,felem y3,felem z3,const felem x1,const felem y1,const felem z1,const int mixed,const smallfelem x2,const smallfelem y2,const smallfelem z2)1105 static void point_add(felem x3, felem y3, felem z3, const felem x1,
1106                       const felem y1, const felem z1, const int mixed,
1107                       const smallfelem x2, const smallfelem y2,
1108                       const smallfelem z2) {
1109   felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, ftmp6, x_out, y_out, z_out;
1110   longfelem tmp, tmp2;
1111   smallfelem small1, small2, small3, small4, small5;
1112   limb x_equal, y_equal, z1_is_zero, z2_is_zero;
1113 
1114   felem_shrink(small3, z1);
1115 
1116   z1_is_zero = smallfelem_is_zero(small3);
1117   z2_is_zero = smallfelem_is_zero(z2);
1118 
1119   /* ftmp = z1z1 = z1**2 */
1120   smallfelem_square(tmp, small3);
1121   felem_reduce(ftmp, tmp);
1122   /* ftmp[i] < 2^101 */
1123   felem_shrink(small1, ftmp);
1124 
1125   if (!mixed) {
1126     /* ftmp2 = z2z2 = z2**2 */
1127     smallfelem_square(tmp, z2);
1128     felem_reduce(ftmp2, tmp);
1129     /* ftmp2[i] < 2^101 */
1130     felem_shrink(small2, ftmp2);
1131 
1132     felem_shrink(small5, x1);
1133 
1134     /* u1 = ftmp3 = x1*z2z2 */
1135     smallfelem_mul(tmp, small5, small2);
1136     felem_reduce(ftmp3, tmp);
1137     /* ftmp3[i] < 2^101 */
1138 
1139     /* ftmp5 = z1 + z2 */
1140     felem_assign(ftmp5, z1);
1141     felem_small_sum(ftmp5, z2);
1142     /* ftmp5[i] < 2^107 */
1143 
1144     /* ftmp5 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2 */
1145     felem_square(tmp, ftmp5);
1146     felem_reduce(ftmp5, tmp);
1147     /* ftmp2 = z2z2 + z1z1 */
1148     felem_sum(ftmp2, ftmp);
1149     /* ftmp2[i] < 2^101 + 2^101 = 2^102 */
1150     felem_diff(ftmp5, ftmp2);
1151     /* ftmp5[i] < 2^105 + 2^101 < 2^106 */
1152 
1153     /* ftmp2 = z2 * z2z2 */
1154     smallfelem_mul(tmp, small2, z2);
1155     felem_reduce(ftmp2, tmp);
1156 
1157     /* s1 = ftmp2 = y1 * z2**3 */
1158     felem_mul(tmp, y1, ftmp2);
1159     felem_reduce(ftmp6, tmp);
1160     /* ftmp6[i] < 2^101 */
1161   } else {
1162     /* We'll assume z2 = 1 (special case z2 = 0 is handled later). */
1163 
1164     /* u1 = ftmp3 = x1*z2z2 */
1165     felem_assign(ftmp3, x1);
1166     /* ftmp3[i] < 2^106 */
1167 
1168     /* ftmp5 = 2z1z2 */
1169     felem_assign(ftmp5, z1);
1170     felem_scalar(ftmp5, 2);
1171     /* ftmp5[i] < 2*2^106 = 2^107 */
1172 
1173     /* s1 = ftmp2 = y1 * z2**3 */
1174     felem_assign(ftmp6, y1);
1175     /* ftmp6[i] < 2^106 */
1176   }
1177 
1178   /* u2 = x2*z1z1 */
1179   smallfelem_mul(tmp, x2, small1);
1180   felem_reduce(ftmp4, tmp);
1181 
1182   /* h = ftmp4 = u2 - u1 */
1183   felem_diff_zero107(ftmp4, ftmp3);
1184   /* ftmp4[i] < 2^107 + 2^101 < 2^108 */
1185   felem_shrink(small4, ftmp4);
1186 
1187   x_equal = smallfelem_is_zero(small4);
1188 
1189   /* z_out = ftmp5 * h */
1190   felem_small_mul(tmp, small4, ftmp5);
1191   felem_reduce(z_out, tmp);
1192   /* z_out[i] < 2^101 */
1193 
1194   /* ftmp = z1 * z1z1 */
1195   smallfelem_mul(tmp, small1, small3);
1196   felem_reduce(ftmp, tmp);
1197 
1198   /* s2 = tmp = y2 * z1**3 */
1199   felem_small_mul(tmp, y2, ftmp);
1200   felem_reduce(ftmp5, tmp);
1201 
1202   /* r = ftmp5 = (s2 - s1)*2 */
1203   felem_diff_zero107(ftmp5, ftmp6);
1204   /* ftmp5[i] < 2^107 + 2^107 = 2^108 */
1205   felem_scalar(ftmp5, 2);
1206   /* ftmp5[i] < 2^109 */
1207   felem_shrink(small1, ftmp5);
1208   y_equal = smallfelem_is_zero(small1);
1209 
1210   if (x_equal && y_equal && !z1_is_zero && !z2_is_zero) {
1211     point_double(x3, y3, z3, x1, y1, z1);
1212     return;
1213   }
1214 
1215   /* I = ftmp = (2h)**2 */
1216   felem_assign(ftmp, ftmp4);
1217   felem_scalar(ftmp, 2);
1218   /* ftmp[i] < 2*2^108 = 2^109 */
1219   felem_square(tmp, ftmp);
1220   felem_reduce(ftmp, tmp);
1221 
1222   /* J = ftmp2 = h * I */
1223   felem_mul(tmp, ftmp4, ftmp);
1224   felem_reduce(ftmp2, tmp);
1225 
1226   /* V = ftmp4 = U1 * I */
1227   felem_mul(tmp, ftmp3, ftmp);
1228   felem_reduce(ftmp4, tmp);
1229 
1230   /* x_out = r**2 - J - 2V */
1231   smallfelem_square(tmp, small1);
1232   felem_reduce(x_out, tmp);
1233   felem_assign(ftmp3, ftmp4);
1234   felem_scalar(ftmp4, 2);
1235   felem_sum(ftmp4, ftmp2);
1236   /* ftmp4[i] < 2*2^101 + 2^101 < 2^103 */
1237   felem_diff(x_out, ftmp4);
1238   /* x_out[i] < 2^105 + 2^101 */
1239 
1240   /* y_out = r(V-x_out) - 2 * s1 * J */
1241   felem_diff_zero107(ftmp3, x_out);
1242   /* ftmp3[i] < 2^107 + 2^101 < 2^108 */
1243   felem_small_mul(tmp, small1, ftmp3);
1244   felem_mul(tmp2, ftmp6, ftmp2);
1245   longfelem_scalar(tmp2, 2);
1246   /* tmp2[i] < 2*2^67 = 2^68 */
1247   longfelem_diff(tmp, tmp2);
1248   /* tmp[i] < 2^67 + 2^70 + 2^40 < 2^71 */
1249   felem_reduce_zero105(y_out, tmp);
1250   /* y_out[i] < 2^106 */
1251 
1252   copy_small_conditional(x_out, x2, z1_is_zero);
1253   copy_conditional(x_out, x1, z2_is_zero);
1254   copy_small_conditional(y_out, y2, z1_is_zero);
1255   copy_conditional(y_out, y1, z2_is_zero);
1256   copy_small_conditional(z_out, z2, z1_is_zero);
1257   copy_conditional(z_out, z1, z2_is_zero);
1258   felem_assign(x3, x_out);
1259   felem_assign(y3, y_out);
1260   felem_assign(z3, z_out);
1261 }
1262 
1263 /* point_add_small is the same as point_add, except that it operates on
1264  * smallfelems. */
point_add_small(smallfelem x3,smallfelem y3,smallfelem z3,smallfelem x1,smallfelem y1,smallfelem z1,smallfelem x2,smallfelem y2,smallfelem z2)1265 static void point_add_small(smallfelem x3, smallfelem y3, smallfelem z3,
1266                             smallfelem x1, smallfelem y1, smallfelem z1,
1267                             smallfelem x2, smallfelem y2, smallfelem z2) {
1268   felem felem_x3, felem_y3, felem_z3;
1269   felem felem_x1, felem_y1, felem_z1;
1270   smallfelem_expand(felem_x1, x1);
1271   smallfelem_expand(felem_y1, y1);
1272   smallfelem_expand(felem_z1, z1);
1273   point_add(felem_x3, felem_y3, felem_z3, felem_x1, felem_y1, felem_z1, 0, x2,
1274             y2, z2);
1275   felem_shrink(x3, felem_x3);
1276   felem_shrink(y3, felem_y3);
1277   felem_shrink(z3, felem_z3);
1278 }
1279 
1280 /* Base point pre computation
1281  * --------------------------
1282  *
1283  * Two different sorts of precomputed tables are used in the following code.
1284  * Each contain various points on the curve, where each point is three field
1285  * elements (x, y, z).
1286  *
1287  * For the base point table, z is usually 1 (0 for the point at infinity).
1288  * This table has 2 * 16 elements, starting with the following:
1289  * index | bits    | point
1290  * ------+---------+------------------------------
1291  *     0 | 0 0 0 0 | 0G
1292  *     1 | 0 0 0 1 | 1G
1293  *     2 | 0 0 1 0 | 2^64G
1294  *     3 | 0 0 1 1 | (2^64 + 1)G
1295  *     4 | 0 1 0 0 | 2^128G
1296  *     5 | 0 1 0 1 | (2^128 + 1)G
1297  *     6 | 0 1 1 0 | (2^128 + 2^64)G
1298  *     7 | 0 1 1 1 | (2^128 + 2^64 + 1)G
1299  *     8 | 1 0 0 0 | 2^192G
1300  *     9 | 1 0 0 1 | (2^192 + 1)G
1301  *    10 | 1 0 1 0 | (2^192 + 2^64)G
1302  *    11 | 1 0 1 1 | (2^192 + 2^64 + 1)G
1303  *    12 | 1 1 0 0 | (2^192 + 2^128)G
1304  *    13 | 1 1 0 1 | (2^192 + 2^128 + 1)G
1305  *    14 | 1 1 1 0 | (2^192 + 2^128 + 2^64)G
1306  *    15 | 1 1 1 1 | (2^192 + 2^128 + 2^64 + 1)G
1307  * followed by a copy of this with each element multiplied by 2^32.
1308  *
1309  * The reason for this is so that we can clock bits into four different
1310  * locations when doing simple scalar multiplies against the base point,
1311  * and then another four locations using the second 16 elements.
1312  *
1313  * Tables for other points have table[i] = iG for i in 0 .. 16. */
1314 
1315 /* gmul is the table of precomputed base points */
1316 static const smallfelem gmul[2][16][3] = {
1317     {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
1318      {{0xf4a13945d898c296, 0x77037d812deb33a0, 0xf8bce6e563a440f2,
1319        0x6b17d1f2e12c4247},
1320       {0xcbb6406837bf51f5, 0x2bce33576b315ece, 0x8ee7eb4a7c0f9e16,
1321        0x4fe342e2fe1a7f9b},
1322       {1, 0, 0, 0}},
1323      {{0x90e75cb48e14db63, 0x29493baaad651f7e, 0x8492592e326e25de,
1324        0x0fa822bc2811aaa5},
1325       {0xe41124545f462ee7, 0x34b1a65050fe82f5, 0x6f4ad4bcb3df188b,
1326        0xbff44ae8f5dba80d},
1327       {1, 0, 0, 0}},
1328      {{0x93391ce2097992af, 0xe96c98fd0d35f1fa, 0xb257c0de95e02789,
1329        0x300a4bbc89d6726f},
1330       {0xaa54a291c08127a0, 0x5bb1eeada9d806a5, 0x7f1ddb25ff1e3c6f,
1331        0x72aac7e0d09b4644},
1332       {1, 0, 0, 0}},
1333      {{0x57c84fc9d789bd85, 0xfc35ff7dc297eac3, 0xfb982fd588c6766e,
1334        0x447d739beedb5e67},
1335       {0x0c7e33c972e25b32, 0x3d349b95a7fae500, 0xe12e9d953a4aaff7,
1336        0x2d4825ab834131ee},
1337       {1, 0, 0, 0}},
1338      {{0x13949c932a1d367f, 0xef7fbd2b1a0a11b7, 0xddc6068bb91dfc60,
1339        0xef9519328a9c72ff},
1340       {0x196035a77376d8a8, 0x23183b0895ca1740, 0xc1ee9807022c219c,
1341        0x611e9fc37dbb2c9b},
1342       {1, 0, 0, 0}},
1343      {{0xcae2b1920b57f4bc, 0x2936df5ec6c9bc36, 0x7dea6482e11238bf,
1344        0x550663797b51f5d8},
1345       {0x44ffe216348a964c, 0x9fb3d576dbdefbe1, 0x0afa40018d9d50e5,
1346        0x157164848aecb851},
1347       {1, 0, 0, 0}},
1348      {{0xe48ecafffc5cde01, 0x7ccd84e70d715f26, 0xa2e8f483f43e4391,
1349        0xeb5d7745b21141ea},
1350       {0xcac917e2731a3479, 0x85f22cfe2844b645, 0x0990e6a158006cee,
1351        0xeafd72ebdbecc17b},
1352       {1, 0, 0, 0}},
1353      {{0x6cf20ffb313728be, 0x96439591a3c6b94a, 0x2736ff8344315fc5,
1354        0xa6d39677a7849276},
1355       {0xf2bab833c357f5f4, 0x824a920c2284059b, 0x66b8babd2d27ecdf,
1356        0x674f84749b0b8816},
1357       {1, 0, 0, 0}},
1358      {{0x2df48c04677c8a3e, 0x74e02f080203a56b, 0x31855f7db8c7fedb,
1359        0x4e769e7672c9ddad},
1360       {0xa4c36165b824bbb0, 0xfb9ae16f3b9122a5, 0x1ec0057206947281,
1361        0x42b99082de830663},
1362       {1, 0, 0, 0}},
1363      {{0x6ef95150dda868b9, 0xd1f89e799c0ce131, 0x7fdc1ca008a1c478,
1364        0x78878ef61c6ce04d},
1365       {0x9c62b9121fe0d976, 0x6ace570ebde08d4f, 0xde53142c12309def,
1366        0xb6cb3f5d7b72c321},
1367       {1, 0, 0, 0}},
1368      {{0x7f991ed2c31a3573, 0x5b82dd5bd54fb496, 0x595c5220812ffcae,
1369        0x0c88bc4d716b1287},
1370       {0x3a57bf635f48aca8, 0x7c8181f4df2564f3, 0x18d1b5b39c04e6aa,
1371        0xdd5ddea3f3901dc6},
1372       {1, 0, 0, 0}},
1373      {{0xe96a79fb3e72ad0c, 0x43a0a28c42ba792f, 0xefe0a423083e49f3,
1374        0x68f344af6b317466},
1375       {0xcdfe17db3fb24d4a, 0x668bfc2271f5c626, 0x604ed93c24d67ff3,
1376        0x31b9c405f8540a20},
1377       {1, 0, 0, 0}},
1378      {{0xd36b4789a2582e7f, 0x0d1a10144ec39c28, 0x663c62c3edbad7a0,
1379        0x4052bf4b6f461db9},
1380       {0x235a27c3188d25eb, 0xe724f33999bfcc5b, 0x862be6bd71d70cc8,
1381        0xfecf4d5190b0fc61},
1382       {1, 0, 0, 0}},
1383      {{0x74346c10a1d4cfac, 0xafdf5cc08526a7a4, 0x123202a8f62bff7a,
1384        0x1eddbae2c802e41a},
1385       {0x8fa0af2dd603f844, 0x36e06b7e4c701917, 0x0c45f45273db33a0,
1386        0x43104d86560ebcfc},
1387       {1, 0, 0, 0}},
1388      {{0x9615b5110d1d78e5, 0x66b0de3225c4744b, 0x0a4a46fb6aaf363a,
1389        0xb48e26b484f7a21c},
1390       {0x06ebb0f621a01b2d, 0xc004e4048b7b0f98, 0x64131bcdfed6f668,
1391        0xfac015404d4d3dab},
1392       {1, 0, 0, 0}}},
1393     {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
1394      {{0x3a5a9e22185a5943, 0x1ab919365c65dfb6, 0x21656b32262c71da,
1395        0x7fe36b40af22af89},
1396       {0xd50d152c699ca101, 0x74b3d5867b8af212, 0x9f09f40407dca6f1,
1397        0xe697d45825b63624},
1398       {1, 0, 0, 0}},
1399      {{0xa84aa9397512218e, 0xe9a521b074ca0141, 0x57880b3a18a2e902,
1400        0x4a5b506612a677a6},
1401       {0x0beada7a4c4f3840, 0x626db15419e26d9d, 0xc42604fbe1627d40,
1402        0xeb13461ceac089f1},
1403       {1, 0, 0, 0}},
1404      {{0xf9faed0927a43281, 0x5e52c4144103ecbc, 0xc342967aa815c857,
1405        0x0781b8291c6a220a},
1406       {0x5a8343ceeac55f80, 0x88f80eeee54a05e3, 0x97b2a14f12916434,
1407        0x690cde8df0151593},
1408       {1, 0, 0, 0}},
1409      {{0xaee9c75df7f82f2a, 0x9e4c35874afdf43a, 0xf5622df437371326,
1410        0x8a535f566ec73617},
1411       {0xc5f9a0ac223094b7, 0xcde533864c8c7669, 0x37e02819085a92bf,
1412        0x0455c08468b08bd7},
1413       {1, 0, 0, 0}},
1414      {{0x0c0a6e2c9477b5d9, 0xf9a4bf62876dc444, 0x5050a949b6cdc279,
1415        0x06bada7ab77f8276},
1416       {0xc8b4aed1ea48dac9, 0xdebd8a4b7ea1070f, 0x427d49101366eb70,
1417        0x5b476dfd0e6cb18a},
1418       {1, 0, 0, 0}},
1419      {{0x7c5c3e44278c340a, 0x4d54606812d66f3b, 0x29a751b1ae23c5d8,
1420        0x3e29864e8a2ec908},
1421       {0x142d2a6626dbb850, 0xad1744c4765bd780, 0x1f150e68e322d1ed,
1422        0x239b90ea3dc31e7e},
1423       {1, 0, 0, 0}},
1424      {{0x78c416527a53322a, 0x305dde6709776f8e, 0xdbcab759f8862ed4,
1425        0x820f4dd949f72ff7},
1426       {0x6cc544a62b5debd4, 0x75be5d937b4e8cc4, 0x1b481b1b215c14d3,
1427        0x140406ec783a05ec},
1428       {1, 0, 0, 0}},
1429      {{0x6a703f10e895df07, 0xfd75f3fa01876bd8, 0xeb5b06e70ce08ffe,
1430        0x68f6b8542783dfee},
1431       {0x90c76f8a78712655, 0xcf5293d2f310bf7f, 0xfbc8044dfda45028,
1432        0xcbe1feba92e40ce6},
1433       {1, 0, 0, 0}},
1434      {{0xe998ceea4396e4c1, 0xfc82ef0b6acea274, 0x230f729f2250e927,
1435        0xd0b2f94d2f420109},
1436       {0x4305adddb38d4966, 0x10b838f8624c3b45, 0x7db2636658954e7a,
1437        0x971459828b0719e5},
1438       {1, 0, 0, 0}},
1439      {{0x4bd6b72623369fc9, 0x57f2929e53d0b876, 0xc2d5cba4f2340687,
1440        0x961610004a866aba},
1441       {0x49997bcd2e407a5e, 0x69ab197d92ddcb24, 0x2cf1f2438fe5131c,
1442        0x7acb9fadcee75e44},
1443       {1, 0, 0, 0}},
1444      {{0x254e839423d2d4c0, 0xf57f0c917aea685b, 0xa60d880f6f75aaea,
1445        0x24eb9acca333bf5b},
1446       {0xe3de4ccb1cda5dea, 0xfeef9341c51a6b4f, 0x743125f88bac4c4d,
1447        0x69f891c5acd079cc},
1448       {1, 0, 0, 0}},
1449      {{0xeee44b35702476b5, 0x7ed031a0e45c2258, 0xb422d1e7bd6f8514,
1450        0xe51f547c5972a107},
1451       {0xa25bcd6fc9cf343d, 0x8ca922ee097c184e, 0xa62f98b3a9fe9a06,
1452        0x1c309a2b25bb1387},
1453       {1, 0, 0, 0}},
1454      {{0x9295dbeb1967c459, 0xb00148833472c98e, 0xc504977708011828,
1455        0x20b87b8aa2c4e503},
1456       {0x3063175de057c277, 0x1bd539338fe582dd, 0x0d11adef5f69a044,
1457        0xf5c6fa49919776be},
1458       {1, 0, 0, 0}},
1459      {{0x8c944e760fd59e11, 0x3876cba1102fad5f, 0xa454c3fad83faa56,
1460        0x1ed7d1b9332010b9},
1461       {0xa1011a270024b889, 0x05e4d0dcac0cd344, 0x52b520f0eb6a2a24,
1462        0x3a2b03f03217257a},
1463       {1, 0, 0, 0}},
1464      {{0xf20fc2afdf1d043d, 0xf330240db58d5a62, 0xfc7d229ca0058c3b,
1465        0x15fee545c78dd9f6},
1466       {0x501e82885bc98cda, 0x41ef80e5d046ac04, 0x557d9f49461210fb,
1467        0x4ab5b6b2b8753f81},
1468       {1, 0, 0, 0}}}};
1469 
1470 /* select_point selects the |idx|th point from a precomputation table and
1471  * copies it to out. */
select_point(const u64 idx,unsigned int size,const smallfelem pre_comp[16][3],smallfelem out[3])1472 static void select_point(const u64 idx, unsigned int size,
1473                          const smallfelem pre_comp[16][3], smallfelem out[3]) {
1474   unsigned i, j;
1475   u64 *outlimbs = &out[0][0];
1476   memset(outlimbs, 0, 3 * sizeof(smallfelem));
1477 
1478   for (i = 0; i < size; i++) {
1479     const u64 *inlimbs = (u64 *)&pre_comp[i][0][0];
1480     u64 mask = i ^ idx;
1481     mask |= mask >> 4;
1482     mask |= mask >> 2;
1483     mask |= mask >> 1;
1484     mask &= 1;
1485     mask--;
1486     for (j = 0; j < NLIMBS * 3; j++) {
1487       outlimbs[j] |= inlimbs[j] & mask;
1488     }
1489   }
1490 }
1491 
1492 /* get_bit returns the |i|th bit in |in| */
get_bit(const felem_bytearray in,int i)1493 static char get_bit(const felem_bytearray in, int i) {
1494   if (i < 0 || i >= 256) {
1495     return 0;
1496   }
1497   return (in[i >> 3] >> (i & 7)) & 1;
1498 }
1499 
1500 /* Interleaved point multiplication using precomputed point multiples: The
1501  * small point multiples 0*P, 1*P, ..., 17*P are in pre_comp[], the scalars
1502  * in scalars[]. If g_scalar is non-NULL, we also add this multiple of the
1503  * generator, using certain (large) precomputed multiples in g_pre_comp.
1504  * Output point (X, Y, Z) is stored in x_out, y_out, z_out. */
batch_mul(felem x_out,felem y_out,felem z_out,const felem_bytearray scalars[],const unsigned num_points,const u8 * g_scalar,const int mixed,const smallfelem pre_comp[][17][3],const smallfelem g_pre_comp[2][16][3])1505 static void batch_mul(felem x_out, felem y_out, felem z_out,
1506                       const felem_bytearray scalars[],
1507                       const unsigned num_points, const u8 *g_scalar,
1508                       const int mixed, const smallfelem pre_comp[][17][3],
1509                       const smallfelem g_pre_comp[2][16][3]) {
1510   int i, skip;
1511   unsigned num, gen_mul = (g_scalar != NULL);
1512   felem nq[3], ftmp;
1513   smallfelem tmp[3];
1514   u64 bits;
1515   u8 sign, digit;
1516 
1517   /* set nq to the point at infinity */
1518   memset(nq, 0, 3 * sizeof(felem));
1519 
1520   /* Loop over all scalars msb-to-lsb, interleaving additions of multiples
1521    * of the generator (two in each of the last 32 rounds) and additions of
1522    * other points multiples (every 5th round). */
1523 
1524   skip = 1; /* save two point operations in the first
1525              * round */
1526   for (i = (num_points ? 255 : 31); i >= 0; --i) {
1527     /* double */
1528     if (!skip) {
1529       point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1530     }
1531 
1532     /* add multiples of the generator */
1533     if (gen_mul && i <= 31) {
1534       /* first, look 32 bits upwards */
1535       bits = get_bit(g_scalar, i + 224) << 3;
1536       bits |= get_bit(g_scalar, i + 160) << 2;
1537       bits |= get_bit(g_scalar, i + 96) << 1;
1538       bits |= get_bit(g_scalar, i + 32);
1539       /* select the point to add, in constant time */
1540       select_point(bits, 16, g_pre_comp[1], tmp);
1541 
1542       if (!skip) {
1543         /* Arg 1 below is for "mixed" */
1544         point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1, tmp[0], tmp[1],
1545                   tmp[2]);
1546       } else {
1547         smallfelem_expand(nq[0], tmp[0]);
1548         smallfelem_expand(nq[1], tmp[1]);
1549         smallfelem_expand(nq[2], tmp[2]);
1550         skip = 0;
1551       }
1552 
1553       /* second, look at the current position */
1554       bits = get_bit(g_scalar, i + 192) << 3;
1555       bits |= get_bit(g_scalar, i + 128) << 2;
1556       bits |= get_bit(g_scalar, i + 64) << 1;
1557       bits |= get_bit(g_scalar, i);
1558       /* select the point to add, in constant time */
1559       select_point(bits, 16, g_pre_comp[0], tmp);
1560       /* Arg 1 below is for "mixed" */
1561       point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1, tmp[0], tmp[1],
1562                 tmp[2]);
1563     }
1564 
1565     /* do other additions every 5 doublings */
1566     if (num_points && (i % 5 == 0)) {
1567       /* loop over all scalars */
1568       for (num = 0; num < num_points; ++num) {
1569         bits = get_bit(scalars[num], i + 4) << 5;
1570         bits |= get_bit(scalars[num], i + 3) << 4;
1571         bits |= get_bit(scalars[num], i + 2) << 3;
1572         bits |= get_bit(scalars[num], i + 1) << 2;
1573         bits |= get_bit(scalars[num], i) << 1;
1574         bits |= get_bit(scalars[num], i - 1);
1575         ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
1576 
1577         /* select the point to add or subtract, in constant time. */
1578         select_point(digit, 17, pre_comp[num], tmp);
1579         smallfelem_neg(ftmp, tmp[1]); /* (X, -Y, Z) is the negative
1580                                        * point */
1581         copy_small_conditional(ftmp, tmp[1], (((limb)sign) - 1));
1582         felem_contract(tmp[1], ftmp);
1583 
1584         if (!skip) {
1585           point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], mixed, tmp[0],
1586                     tmp[1], tmp[2]);
1587         } else {
1588           smallfelem_expand(nq[0], tmp[0]);
1589           smallfelem_expand(nq[1], tmp[1]);
1590           smallfelem_expand(nq[2], tmp[2]);
1591           skip = 0;
1592         }
1593       }
1594     }
1595   }
1596   felem_assign(x_out, nq[0]);
1597   felem_assign(y_out, nq[1]);
1598   felem_assign(z_out, nq[2]);
1599 }
1600 
1601 /* Precomputation for the group generator. */
1602 typedef struct {
1603   smallfelem g_pre_comp[2][16][3];
1604 } NISTP256_PRE_COMP;
1605 
1606 /******************************************************************************/
1607 /*
1608  * OPENSSL EC_METHOD FUNCTIONS
1609  */
1610 
ec_GFp_nistp256_group_init(EC_GROUP * group)1611 int ec_GFp_nistp256_group_init(EC_GROUP *group) {
1612   int ret = ec_GFp_simple_group_init(group);
1613   group->a_is_minus3 = 1;
1614   return ret;
1615 }
1616 
ec_GFp_nistp256_group_set_curve(EC_GROUP * group,const BIGNUM * p,const BIGNUM * a,const BIGNUM * b,BN_CTX * ctx)1617 int ec_GFp_nistp256_group_set_curve(EC_GROUP *group, const BIGNUM *p,
1618                                     const BIGNUM *a, const BIGNUM *b,
1619                                     BN_CTX *ctx) {
1620   int ret = 0;
1621   BN_CTX *new_ctx = NULL;
1622   BIGNUM *curve_p, *curve_a, *curve_b;
1623 
1624   if (ctx == NULL) {
1625     if ((ctx = new_ctx = BN_CTX_new()) == NULL) {
1626       return 0;
1627     }
1628   }
1629   BN_CTX_start(ctx);
1630   if (((curve_p = BN_CTX_get(ctx)) == NULL) ||
1631       ((curve_a = BN_CTX_get(ctx)) == NULL) ||
1632       ((curve_b = BN_CTX_get(ctx)) == NULL)) {
1633     goto err;
1634   }
1635   BN_bin2bn(nistp256_curve_params[0], sizeof(felem_bytearray), curve_p);
1636   BN_bin2bn(nistp256_curve_params[1], sizeof(felem_bytearray), curve_a);
1637   BN_bin2bn(nistp256_curve_params[2], sizeof(felem_bytearray), curve_b);
1638   if (BN_cmp(curve_p, p) ||
1639       BN_cmp(curve_a, a) ||
1640       BN_cmp(curve_b, b)) {
1641     OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_group_set_curve,
1642                       EC_R_WRONG_CURVE_PARAMETERS);
1643     goto err;
1644   }
1645   ret = ec_GFp_simple_group_set_curve(group, p, a, b, ctx);
1646 
1647 err:
1648   BN_CTX_end(ctx);
1649   BN_CTX_free(new_ctx);
1650   return ret;
1651 }
1652 
1653 /* Takes the Jacobian coordinates (X, Y, Z) of a point and returns (X', Y') =
1654  * (X/Z^2, Y/Z^3). */
ec_GFp_nistp256_point_get_affine_coordinates(const EC_GROUP * group,const EC_POINT * point,BIGNUM * x,BIGNUM * y,BN_CTX * ctx)1655 int ec_GFp_nistp256_point_get_affine_coordinates(const EC_GROUP *group,
1656                                                  const EC_POINT *point,
1657                                                  BIGNUM *x, BIGNUM *y,
1658                                                  BN_CTX *ctx) {
1659   felem z1, z2, x_in, y_in;
1660   smallfelem x_out, y_out;
1661   longfelem tmp;
1662 
1663   if (EC_POINT_is_at_infinity(group, point)) {
1664     OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_point_get_affine_coordinates,
1665                       EC_R_POINT_AT_INFINITY);
1666     return 0;
1667   }
1668   if (!BN_to_felem(x_in, &point->X) ||
1669       !BN_to_felem(y_in, &point->Y) ||
1670       !BN_to_felem(z1, &point->Z)) {
1671     return 0;
1672   }
1673   felem_inv(z2, z1);
1674   felem_square(tmp, z2);
1675   felem_reduce(z1, tmp);
1676   felem_mul(tmp, x_in, z1);
1677   felem_reduce(x_in, tmp);
1678   felem_contract(x_out, x_in);
1679   if (x != NULL && !smallfelem_to_BN(x, x_out)) {
1680     OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_point_get_affine_coordinates,
1681                       ERR_R_BN_LIB);
1682     return 0;
1683   }
1684   felem_mul(tmp, z1, z2);
1685   felem_reduce(z1, tmp);
1686   felem_mul(tmp, y_in, z1);
1687   felem_reduce(y_in, tmp);
1688   felem_contract(y_out, y_in);
1689   if (y != NULL && !smallfelem_to_BN(y, y_out)) {
1690     OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_point_get_affine_coordinates,
1691                       ERR_R_BN_LIB);
1692     return 0;
1693   }
1694   return 1;
1695 }
1696 
1697 /* points below is of size |num|, and tmp_smallfelems is of size |num+1| */
make_points_affine(size_t num,smallfelem points[][3],smallfelem tmp_smallfelems[])1698 static void make_points_affine(size_t num, smallfelem points[][3],
1699                                smallfelem tmp_smallfelems[]) {
1700   /* Runs in constant time, unless an input is the point at infinity (which
1701    * normally shouldn't happen). */
1702   ec_GFp_nistp_points_make_affine_internal(
1703       num, points, sizeof(smallfelem), tmp_smallfelems,
1704       (void (*)(void *))smallfelem_one,
1705       (int (*)(const void *))smallfelem_is_zero_int,
1706       (void (*)(void *, const void *))smallfelem_assign,
1707       (void (*)(void *, const void *))smallfelem_square_contract,
1708       (void (*)(void *, const void *, const void *))smallfelem_mul_contract,
1709       (void (*)(void *, const void *))smallfelem_inv_contract,
1710       /* nothing to contract */
1711       (void (*)(void *, const void *))smallfelem_assign);
1712 }
1713 
1714 /* Computes scalar*generator + \sum scalars[i]*points[i], ignoring NULL
1715  * values Result is stored in r (r can equal one of the inputs). */
ec_GFp_nistp256_points_mul(const EC_GROUP * group,EC_POINT * r,const BIGNUM * scalar,size_t num,const EC_POINT * points[],const BIGNUM * scalars[],BN_CTX * ctx)1716 int ec_GFp_nistp256_points_mul(const EC_GROUP *group, EC_POINT *r,
1717                                const BIGNUM *scalar, size_t num,
1718                                const EC_POINT *points[],
1719                                const BIGNUM *scalars[], BN_CTX *ctx) {
1720   int ret = 0;
1721   int j;
1722   int mixed = 0;
1723   BN_CTX *new_ctx = NULL;
1724   BIGNUM *x, *y, *z, *tmp_scalar;
1725   felem_bytearray g_secret;
1726   felem_bytearray *secrets = NULL;
1727   smallfelem(*pre_comp)[17][3] = NULL;
1728   smallfelem *tmp_smallfelems = NULL;
1729   felem_bytearray tmp;
1730   unsigned i, num_bytes;
1731   int have_pre_comp = 0;
1732   size_t num_points = num;
1733   smallfelem x_in, y_in, z_in;
1734   felem x_out, y_out, z_out;
1735   const smallfelem(*g_pre_comp)[16][3] = NULL;
1736   EC_POINT *generator = NULL;
1737   const EC_POINT *p = NULL;
1738   const BIGNUM *p_scalar = NULL;
1739 
1740   if (ctx == NULL) {
1741     ctx = new_ctx = BN_CTX_new();
1742     if (ctx == NULL) {
1743       return 0;
1744     }
1745   }
1746 
1747   BN_CTX_start(ctx);
1748   if ((x = BN_CTX_get(ctx)) == NULL ||
1749       (y = BN_CTX_get(ctx)) == NULL ||
1750       (z = BN_CTX_get(ctx)) == NULL ||
1751       (tmp_scalar = BN_CTX_get(ctx)) == NULL) {
1752     goto err;
1753   }
1754 
1755   if (scalar != NULL) {
1756     /* try to use the standard precomputation */
1757     g_pre_comp = &gmul[0];
1758     generator = EC_POINT_new(group);
1759     if (generator == NULL) {
1760       goto err;
1761     }
1762     /* get the generator from precomputation */
1763     if (!smallfelem_to_BN(x, g_pre_comp[0][1][0]) ||
1764         !smallfelem_to_BN(y, g_pre_comp[0][1][1]) ||
1765         !smallfelem_to_BN(z, g_pre_comp[0][1][2])) {
1766       OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_points_mul, ERR_R_BN_LIB);
1767       goto err;
1768     }
1769     if (!ec_point_set_Jprojective_coordinates_GFp(group, generator, x, y, z,
1770                                                   ctx)) {
1771       goto err;
1772     }
1773     if (0 == EC_POINT_cmp(group, generator, group->generator, ctx)) {
1774       /* precomputation matches generator */
1775       have_pre_comp = 1;
1776     } else {
1777       /* we don't have valid precomputation: treat the generator as a
1778        * random point. */
1779       num_points++;
1780     }
1781   }
1782 
1783   if (num_points > 0) {
1784     if (num_points >= 3) {
1785       /* unless we precompute multiples for just one or two points,
1786        * converting those into affine form is time well spent */
1787       mixed = 1;
1788     }
1789     secrets = OPENSSL_malloc(num_points * sizeof(felem_bytearray));
1790     pre_comp = OPENSSL_malloc(num_points * 17 * 3 * sizeof(smallfelem));
1791     if (mixed) {
1792       tmp_smallfelems =
1793           OPENSSL_malloc((num_points * 17 + 1) * sizeof(smallfelem));
1794     }
1795     if (secrets == NULL || pre_comp == NULL ||
1796         (mixed && tmp_smallfelems == NULL)) {
1797       OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_points_mul, ERR_R_MALLOC_FAILURE);
1798       goto err;
1799     }
1800 
1801     /* we treat NULL scalars as 0, and NULL points as points at infinity,
1802      * i.e., they contribute nothing to the linear combination. */
1803     memset(secrets, 0, num_points * sizeof(felem_bytearray));
1804     memset(pre_comp, 0, num_points * 17 * 3 * sizeof(smallfelem));
1805     for (i = 0; i < num_points; ++i) {
1806       if (i == num) {
1807         /* we didn't have a valid precomputation, so we pick the generator. */
1808         p = EC_GROUP_get0_generator(group);
1809         p_scalar = scalar;
1810       } else {
1811         /* the i^th point */
1812         p = points[i];
1813         p_scalar = scalars[i];
1814       }
1815       if (p_scalar != NULL && p != NULL) {
1816         /* reduce scalar to 0 <= scalar < 2^256 */
1817         if (BN_num_bits(p_scalar) > 256 || BN_is_negative(p_scalar)) {
1818           /* this is an unusual input, and we don't guarantee
1819            * constant-timeness. */
1820           if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx)) {
1821             OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_points_mul, ERR_R_BN_LIB);
1822             goto err;
1823           }
1824           num_bytes = BN_bn2bin(tmp_scalar, tmp);
1825         } else {
1826           num_bytes = BN_bn2bin(p_scalar, tmp);
1827         }
1828         flip_endian(secrets[i], tmp, num_bytes);
1829         /* precompute multiples */
1830         if (!BN_to_felem(x_out, &p->X) ||
1831             !BN_to_felem(y_out, &p->Y) ||
1832             !BN_to_felem(z_out, &p->Z)) {
1833           goto err;
1834         }
1835         felem_shrink(pre_comp[i][1][0], x_out);
1836         felem_shrink(pre_comp[i][1][1], y_out);
1837         felem_shrink(pre_comp[i][1][2], z_out);
1838         for (j = 2; j <= 16; ++j) {
1839           if (j & 1) {
1840             point_add_small(pre_comp[i][j][0], pre_comp[i][j][1],
1841                             pre_comp[i][j][2], pre_comp[i][1][0],
1842                             pre_comp[i][1][1], pre_comp[i][1][2],
1843                             pre_comp[i][j - 1][0], pre_comp[i][j - 1][1],
1844                             pre_comp[i][j - 1][2]);
1845           } else {
1846             point_double_small(pre_comp[i][j][0], pre_comp[i][j][1],
1847                                pre_comp[i][j][2], pre_comp[i][j / 2][0],
1848                                pre_comp[i][j / 2][1], pre_comp[i][j / 2][2]);
1849           }
1850         }
1851       }
1852     }
1853     if (mixed) {
1854       make_points_affine(num_points * 17, pre_comp[0], tmp_smallfelems);
1855     }
1856   }
1857 
1858   /* the scalar for the generator */
1859   if (scalar != NULL && have_pre_comp) {
1860     memset(g_secret, 0, sizeof(g_secret));
1861     /* reduce scalar to 0 <= scalar < 2^256 */
1862     if (BN_num_bits(scalar) > 256 || BN_is_negative(scalar)) {
1863       /* this is an unusual input, and we don't guarantee
1864        * constant-timeness. */
1865       if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx)) {
1866         OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_points_mul, ERR_R_BN_LIB);
1867         goto err;
1868       }
1869       num_bytes = BN_bn2bin(tmp_scalar, tmp);
1870     } else {
1871       num_bytes = BN_bn2bin(scalar, tmp);
1872     }
1873     flip_endian(g_secret, tmp, num_bytes);
1874     /* do the multiplication with generator precomputation */
1875     batch_mul(x_out, y_out, z_out, (const felem_bytearray(*))secrets,
1876               num_points, g_secret, mixed, (const smallfelem(*)[17][3])pre_comp,
1877               g_pre_comp);
1878   } else {
1879     /* do the multiplication without generator precomputation */
1880     batch_mul(x_out, y_out, z_out, (const felem_bytearray(*))secrets,
1881               num_points, NULL, mixed, (const smallfelem(*)[17][3])pre_comp,
1882               NULL);
1883   }
1884 
1885   /* reduce the output to its unique minimal representation */
1886   felem_contract(x_in, x_out);
1887   felem_contract(y_in, y_out);
1888   felem_contract(z_in, z_out);
1889   if (!smallfelem_to_BN(x, x_in) ||
1890       !smallfelem_to_BN(y, y_in) ||
1891       !smallfelem_to_BN(z, z_in)) {
1892     OPENSSL_PUT_ERROR(EC, ec_GFp_nistp256_points_mul, ERR_R_BN_LIB);
1893     goto err;
1894   }
1895   ret = ec_point_set_Jprojective_coordinates_GFp(group, r, x, y, z, ctx);
1896 
1897 err:
1898   BN_CTX_end(ctx);
1899   EC_POINT_free(generator);
1900   BN_CTX_free(new_ctx);
1901   OPENSSL_free(secrets);
1902   OPENSSL_free(pre_comp);
1903   OPENSSL_free(tmp_smallfelems);
1904   return ret;
1905 }
1906 
EC_GFp_nistp256_method(void)1907 const EC_METHOD *EC_GFp_nistp256_method(void) {
1908   static const EC_METHOD ret = {
1909       EC_FLAGS_DEFAULT_OCT,
1910       ec_GFp_nistp256_group_init,
1911       ec_GFp_simple_group_finish,
1912       ec_GFp_simple_group_clear_finish,
1913       ec_GFp_simple_group_copy, ec_GFp_nistp256_group_set_curve,
1914       ec_GFp_simple_group_get_curve, ec_GFp_simple_group_get_degree,
1915       ec_GFp_simple_group_check_discriminant, ec_GFp_simple_point_init,
1916       ec_GFp_simple_point_finish, ec_GFp_simple_point_clear_finish,
1917       ec_GFp_simple_point_copy, ec_GFp_simple_point_set_to_infinity,
1918       ec_GFp_simple_set_Jprojective_coordinates_GFp,
1919       ec_GFp_simple_get_Jprojective_coordinates_GFp,
1920       ec_GFp_simple_point_set_affine_coordinates,
1921       ec_GFp_nistp256_point_get_affine_coordinates,
1922       0 /* point_set_compressed_coordinates */, 0 /* point2oct */,
1923       0 /* oct2point */, ec_GFp_simple_add, ec_GFp_simple_dbl,
1924       ec_GFp_simple_invert, ec_GFp_simple_is_at_infinity,
1925       ec_GFp_simple_is_on_curve, ec_GFp_simple_cmp, ec_GFp_simple_make_affine,
1926       ec_GFp_simple_points_make_affine, ec_GFp_nistp256_points_mul,
1927       0 /* precompute_mult */, 0 /* have_precompute_mult */,
1928       ec_GFp_simple_field_mul, ec_GFp_simple_field_sqr, 0 /* field_div */,
1929       0 /* field_encode */, 0 /* field_decode */, 0 /* field_set_to_one */
1930   };
1931 
1932   return &ret;
1933 }
1934 
1935 #endif  /* 64_BIT && !WINDOWS */
1936