Loading...
1/*
2 * Copyright (c) 2013, 2014 Kenneth MacKay. All rights reserved.
3 * Copyright (c) 2019 Vitaly Chikunov <vt@altlinux.org>
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met:
8 * * Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
15 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
16 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
17 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
18 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
20 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
24 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <linux/module.h>
28#include <linux/random.h>
29#include <linux/slab.h>
30#include <linux/swab.h>
31#include <linux/fips.h>
32#include <crypto/ecdh.h>
33#include <crypto/rng.h>
34#include <asm/unaligned.h>
35#include <linux/ratelimit.h>
36
37#include "ecc.h"
38#include "ecc_curve_defs.h"
39
40typedef struct {
41 u64 m_low;
42 u64 m_high;
43} uint128_t;
44
45static inline const struct ecc_curve *ecc_get_curve(unsigned int curve_id)
46{
47 switch (curve_id) {
48 /* In FIPS mode only allow P256 and higher */
49 case ECC_CURVE_NIST_P192:
50 return fips_enabled ? NULL : &nist_p192;
51 case ECC_CURVE_NIST_P256:
52 return &nist_p256;
53 default:
54 return NULL;
55 }
56}
57
58static u64 *ecc_alloc_digits_space(unsigned int ndigits)
59{
60 size_t len = ndigits * sizeof(u64);
61
62 if (!len)
63 return NULL;
64
65 return kmalloc(len, GFP_KERNEL);
66}
67
68static void ecc_free_digits_space(u64 *space)
69{
70 kzfree(space);
71}
72
73static struct ecc_point *ecc_alloc_point(unsigned int ndigits)
74{
75 struct ecc_point *p = kmalloc(sizeof(*p), GFP_KERNEL);
76
77 if (!p)
78 return NULL;
79
80 p->x = ecc_alloc_digits_space(ndigits);
81 if (!p->x)
82 goto err_alloc_x;
83
84 p->y = ecc_alloc_digits_space(ndigits);
85 if (!p->y)
86 goto err_alloc_y;
87
88 p->ndigits = ndigits;
89
90 return p;
91
92err_alloc_y:
93 ecc_free_digits_space(p->x);
94err_alloc_x:
95 kfree(p);
96 return NULL;
97}
98
99static void ecc_free_point(struct ecc_point *p)
100{
101 if (!p)
102 return;
103
104 kzfree(p->x);
105 kzfree(p->y);
106 kzfree(p);
107}
108
109static void vli_clear(u64 *vli, unsigned int ndigits)
110{
111 int i;
112
113 for (i = 0; i < ndigits; i++)
114 vli[i] = 0;
115}
116
117/* Returns true if vli == 0, false otherwise. */
118bool vli_is_zero(const u64 *vli, unsigned int ndigits)
119{
120 int i;
121
122 for (i = 0; i < ndigits; i++) {
123 if (vli[i])
124 return false;
125 }
126
127 return true;
128}
129EXPORT_SYMBOL(vli_is_zero);
130
131/* Returns nonzero if bit bit of vli is set. */
132static u64 vli_test_bit(const u64 *vli, unsigned int bit)
133{
134 return (vli[bit / 64] & ((u64)1 << (bit % 64)));
135}
136
137static bool vli_is_negative(const u64 *vli, unsigned int ndigits)
138{
139 return vli_test_bit(vli, ndigits * 64 - 1);
140}
141
142/* Counts the number of 64-bit "digits" in vli. */
143static unsigned int vli_num_digits(const u64 *vli, unsigned int ndigits)
144{
145 int i;
146
147 /* Search from the end until we find a non-zero digit.
148 * We do it in reverse because we expect that most digits will
149 * be nonzero.
150 */
151 for (i = ndigits - 1; i >= 0 && vli[i] == 0; i--);
152
153 return (i + 1);
154}
155
156/* Counts the number of bits required for vli. */
157static unsigned int vli_num_bits(const u64 *vli, unsigned int ndigits)
158{
159 unsigned int i, num_digits;
160 u64 digit;
161
162 num_digits = vli_num_digits(vli, ndigits);
163 if (num_digits == 0)
164 return 0;
165
166 digit = vli[num_digits - 1];
167 for (i = 0; digit; i++)
168 digit >>= 1;
169
170 return ((num_digits - 1) * 64 + i);
171}
172
173/* Set dest from unaligned bit string src. */
174void vli_from_be64(u64 *dest, const void *src, unsigned int ndigits)
175{
176 int i;
177 const u64 *from = src;
178
179 for (i = 0; i < ndigits; i++)
180 dest[i] = get_unaligned_be64(&from[ndigits - 1 - i]);
181}
182EXPORT_SYMBOL(vli_from_be64);
183
184void vli_from_le64(u64 *dest, const void *src, unsigned int ndigits)
185{
186 int i;
187 const u64 *from = src;
188
189 for (i = 0; i < ndigits; i++)
190 dest[i] = get_unaligned_le64(&from[i]);
191}
192EXPORT_SYMBOL(vli_from_le64);
193
194/* Sets dest = src. */
195static void vli_set(u64 *dest, const u64 *src, unsigned int ndigits)
196{
197 int i;
198
199 for (i = 0; i < ndigits; i++)
200 dest[i] = src[i];
201}
202
203/* Returns sign of left - right. */
204int vli_cmp(const u64 *left, const u64 *right, unsigned int ndigits)
205{
206 int i;
207
208 for (i = ndigits - 1; i >= 0; i--) {
209 if (left[i] > right[i])
210 return 1;
211 else if (left[i] < right[i])
212 return -1;
213 }
214
215 return 0;
216}
217EXPORT_SYMBOL(vli_cmp);
218
219/* Computes result = in << c, returning carry. Can modify in place
220 * (if result == in). 0 < shift < 64.
221 */
222static u64 vli_lshift(u64 *result, const u64 *in, unsigned int shift,
223 unsigned int ndigits)
224{
225 u64 carry = 0;
226 int i;
227
228 for (i = 0; i < ndigits; i++) {
229 u64 temp = in[i];
230
231 result[i] = (temp << shift) | carry;
232 carry = temp >> (64 - shift);
233 }
234
235 return carry;
236}
237
238/* Computes vli = vli >> 1. */
239static void vli_rshift1(u64 *vli, unsigned int ndigits)
240{
241 u64 *end = vli;
242 u64 carry = 0;
243
244 vli += ndigits;
245
246 while (vli-- > end) {
247 u64 temp = *vli;
248 *vli = (temp >> 1) | carry;
249 carry = temp << 63;
250 }
251}
252
253/* Computes result = left + right, returning carry. Can modify in place. */
254static u64 vli_add(u64 *result, const u64 *left, const u64 *right,
255 unsigned int ndigits)
256{
257 u64 carry = 0;
258 int i;
259
260 for (i = 0; i < ndigits; i++) {
261 u64 sum;
262
263 sum = left[i] + right[i] + carry;
264 if (sum != left[i])
265 carry = (sum < left[i]);
266
267 result[i] = sum;
268 }
269
270 return carry;
271}
272
273/* Computes result = left + right, returning carry. Can modify in place. */
274static u64 vli_uadd(u64 *result, const u64 *left, u64 right,
275 unsigned int ndigits)
276{
277 u64 carry = right;
278 int i;
279
280 for (i = 0; i < ndigits; i++) {
281 u64 sum;
282
283 sum = left[i] + carry;
284 if (sum != left[i])
285 carry = (sum < left[i]);
286 else
287 carry = !!carry;
288
289 result[i] = sum;
290 }
291
292 return carry;
293}
294
295/* Computes result = left - right, returning borrow. Can modify in place. */
296u64 vli_sub(u64 *result, const u64 *left, const u64 *right,
297 unsigned int ndigits)
298{
299 u64 borrow = 0;
300 int i;
301
302 for (i = 0; i < ndigits; i++) {
303 u64 diff;
304
305 diff = left[i] - right[i] - borrow;
306 if (diff != left[i])
307 borrow = (diff > left[i]);
308
309 result[i] = diff;
310 }
311
312 return borrow;
313}
314EXPORT_SYMBOL(vli_sub);
315
316/* Computes result = left - right, returning borrow. Can modify in place. */
317static u64 vli_usub(u64 *result, const u64 *left, u64 right,
318 unsigned int ndigits)
319{
320 u64 borrow = right;
321 int i;
322
323 for (i = 0; i < ndigits; i++) {
324 u64 diff;
325
326 diff = left[i] - borrow;
327 if (diff != left[i])
328 borrow = (diff > left[i]);
329
330 result[i] = diff;
331 }
332
333 return borrow;
334}
335
336static uint128_t mul_64_64(u64 left, u64 right)
337{
338 uint128_t result;
339#if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
340 unsigned __int128 m = (unsigned __int128)left * right;
341
342 result.m_low = m;
343 result.m_high = m >> 64;
344#else
345 u64 a0 = left & 0xffffffffull;
346 u64 a1 = left >> 32;
347 u64 b0 = right & 0xffffffffull;
348 u64 b1 = right >> 32;
349 u64 m0 = a0 * b0;
350 u64 m1 = a0 * b1;
351 u64 m2 = a1 * b0;
352 u64 m3 = a1 * b1;
353
354 m2 += (m0 >> 32);
355 m2 += m1;
356
357 /* Overflow */
358 if (m2 < m1)
359 m3 += 0x100000000ull;
360
361 result.m_low = (m0 & 0xffffffffull) | (m2 << 32);
362 result.m_high = m3 + (m2 >> 32);
363#endif
364 return result;
365}
366
367static uint128_t add_128_128(uint128_t a, uint128_t b)
368{
369 uint128_t result;
370
371 result.m_low = a.m_low + b.m_low;
372 result.m_high = a.m_high + b.m_high + (result.m_low < a.m_low);
373
374 return result;
375}
376
377static void vli_mult(u64 *result, const u64 *left, const u64 *right,
378 unsigned int ndigits)
379{
380 uint128_t r01 = { 0, 0 };
381 u64 r2 = 0;
382 unsigned int i, k;
383
384 /* Compute each digit of result in sequence, maintaining the
385 * carries.
386 */
387 for (k = 0; k < ndigits * 2 - 1; k++) {
388 unsigned int min;
389
390 if (k < ndigits)
391 min = 0;
392 else
393 min = (k + 1) - ndigits;
394
395 for (i = min; i <= k && i < ndigits; i++) {
396 uint128_t product;
397
398 product = mul_64_64(left[i], right[k - i]);
399
400 r01 = add_128_128(r01, product);
401 r2 += (r01.m_high < product.m_high);
402 }
403
404 result[k] = r01.m_low;
405 r01.m_low = r01.m_high;
406 r01.m_high = r2;
407 r2 = 0;
408 }
409
410 result[ndigits * 2 - 1] = r01.m_low;
411}
412
413/* Compute product = left * right, for a small right value. */
414static void vli_umult(u64 *result, const u64 *left, u32 right,
415 unsigned int ndigits)
416{
417 uint128_t r01 = { 0 };
418 unsigned int k;
419
420 for (k = 0; k < ndigits; k++) {
421 uint128_t product;
422
423 product = mul_64_64(left[k], right);
424 r01 = add_128_128(r01, product);
425 /* no carry */
426 result[k] = r01.m_low;
427 r01.m_low = r01.m_high;
428 r01.m_high = 0;
429 }
430 result[k] = r01.m_low;
431 for (++k; k < ndigits * 2; k++)
432 result[k] = 0;
433}
434
435static void vli_square(u64 *result, const u64 *left, unsigned int ndigits)
436{
437 uint128_t r01 = { 0, 0 };
438 u64 r2 = 0;
439 int i, k;
440
441 for (k = 0; k < ndigits * 2 - 1; k++) {
442 unsigned int min;
443
444 if (k < ndigits)
445 min = 0;
446 else
447 min = (k + 1) - ndigits;
448
449 for (i = min; i <= k && i <= k - i; i++) {
450 uint128_t product;
451
452 product = mul_64_64(left[i], left[k - i]);
453
454 if (i < k - i) {
455 r2 += product.m_high >> 63;
456 product.m_high = (product.m_high << 1) |
457 (product.m_low >> 63);
458 product.m_low <<= 1;
459 }
460
461 r01 = add_128_128(r01, product);
462 r2 += (r01.m_high < product.m_high);
463 }
464
465 result[k] = r01.m_low;
466 r01.m_low = r01.m_high;
467 r01.m_high = r2;
468 r2 = 0;
469 }
470
471 result[ndigits * 2 - 1] = r01.m_low;
472}
473
474/* Computes result = (left + right) % mod.
475 * Assumes that left < mod and right < mod, result != mod.
476 */
477static void vli_mod_add(u64 *result, const u64 *left, const u64 *right,
478 const u64 *mod, unsigned int ndigits)
479{
480 u64 carry;
481
482 carry = vli_add(result, left, right, ndigits);
483
484 /* result > mod (result = mod + remainder), so subtract mod to
485 * get remainder.
486 */
487 if (carry || vli_cmp(result, mod, ndigits) >= 0)
488 vli_sub(result, result, mod, ndigits);
489}
490
491/* Computes result = (left - right) % mod.
492 * Assumes that left < mod and right < mod, result != mod.
493 */
494static void vli_mod_sub(u64 *result, const u64 *left, const u64 *right,
495 const u64 *mod, unsigned int ndigits)
496{
497 u64 borrow = vli_sub(result, left, right, ndigits);
498
499 /* In this case, p_result == -diff == (max int) - diff.
500 * Since -x % d == d - x, we can get the correct result from
501 * result + mod (with overflow).
502 */
503 if (borrow)
504 vli_add(result, result, mod, ndigits);
505}
506
507/*
508 * Computes result = product % mod
509 * for special form moduli: p = 2^k-c, for small c (note the minus sign)
510 *
511 * References:
512 * R. Crandall, C. Pomerance. Prime Numbers: A Computational Perspective.
513 * 9 Fast Algorithms for Large-Integer Arithmetic. 9.2.3 Moduli of special form
514 * Algorithm 9.2.13 (Fast mod operation for special-form moduli).
515 */
516static void vli_mmod_special(u64 *result, const u64 *product,
517 const u64 *mod, unsigned int ndigits)
518{
519 u64 c = -mod[0];
520 u64 t[ECC_MAX_DIGITS * 2];
521 u64 r[ECC_MAX_DIGITS * 2];
522
523 vli_set(r, product, ndigits * 2);
524 while (!vli_is_zero(r + ndigits, ndigits)) {
525 vli_umult(t, r + ndigits, c, ndigits);
526 vli_clear(r + ndigits, ndigits);
527 vli_add(r, r, t, ndigits * 2);
528 }
529 vli_set(t, mod, ndigits);
530 vli_clear(t + ndigits, ndigits);
531 while (vli_cmp(r, t, ndigits * 2) >= 0)
532 vli_sub(r, r, t, ndigits * 2);
533 vli_set(result, r, ndigits);
534}
535
536/*
537 * Computes result = product % mod
538 * for special form moduli: p = 2^{k-1}+c, for small c (note the plus sign)
539 * where k-1 does not fit into qword boundary by -1 bit (such as 255).
540
541 * References (loosely based on):
542 * A. Menezes, P. van Oorschot, S. Vanstone. Handbook of Applied Cryptography.
543 * 14.3.4 Reduction methods for moduli of special form. Algorithm 14.47.
544 * URL: http://cacr.uwaterloo.ca/hac/about/chap14.pdf
545 *
546 * H. Cohen, G. Frey, R. Avanzi, C. Doche, T. Lange, K. Nguyen, F. Vercauteren.
547 * Handbook of Elliptic and Hyperelliptic Curve Cryptography.
548 * Algorithm 10.25 Fast reduction for special form moduli
549 */
550static void vli_mmod_special2(u64 *result, const u64 *product,
551 const u64 *mod, unsigned int ndigits)
552{
553 u64 c2 = mod[0] * 2;
554 u64 q[ECC_MAX_DIGITS];
555 u64 r[ECC_MAX_DIGITS * 2];
556 u64 m[ECC_MAX_DIGITS * 2]; /* expanded mod */
557 int carry; /* last bit that doesn't fit into q */
558 int i;
559
560 vli_set(m, mod, ndigits);
561 vli_clear(m + ndigits, ndigits);
562
563 vli_set(r, product, ndigits);
564 /* q and carry are top bits */
565 vli_set(q, product + ndigits, ndigits);
566 vli_clear(r + ndigits, ndigits);
567 carry = vli_is_negative(r, ndigits);
568 if (carry)
569 r[ndigits - 1] &= (1ull << 63) - 1;
570 for (i = 1; carry || !vli_is_zero(q, ndigits); i++) {
571 u64 qc[ECC_MAX_DIGITS * 2];
572
573 vli_umult(qc, q, c2, ndigits);
574 if (carry)
575 vli_uadd(qc, qc, mod[0], ndigits * 2);
576 vli_set(q, qc + ndigits, ndigits);
577 vli_clear(qc + ndigits, ndigits);
578 carry = vli_is_negative(qc, ndigits);
579 if (carry)
580 qc[ndigits - 1] &= (1ull << 63) - 1;
581 if (i & 1)
582 vli_sub(r, r, qc, ndigits * 2);
583 else
584 vli_add(r, r, qc, ndigits * 2);
585 }
586 while (vli_is_negative(r, ndigits * 2))
587 vli_add(r, r, m, ndigits * 2);
588 while (vli_cmp(r, m, ndigits * 2) >= 0)
589 vli_sub(r, r, m, ndigits * 2);
590
591 vli_set(result, r, ndigits);
592}
593
594/*
595 * Computes result = product % mod, where product is 2N words long.
596 * Reference: Ken MacKay's micro-ecc.
597 * Currently only designed to work for curve_p or curve_n.
598 */
599static void vli_mmod_slow(u64 *result, u64 *product, const u64 *mod,
600 unsigned int ndigits)
601{
602 u64 mod_m[2 * ECC_MAX_DIGITS];
603 u64 tmp[2 * ECC_MAX_DIGITS];
604 u64 *v[2] = { tmp, product };
605 u64 carry = 0;
606 unsigned int i;
607 /* Shift mod so its highest set bit is at the maximum position. */
608 int shift = (ndigits * 2 * 64) - vli_num_bits(mod, ndigits);
609 int word_shift = shift / 64;
610 int bit_shift = shift % 64;
611
612 vli_clear(mod_m, word_shift);
613 if (bit_shift > 0) {
614 for (i = 0; i < ndigits; ++i) {
615 mod_m[word_shift + i] = (mod[i] << bit_shift) | carry;
616 carry = mod[i] >> (64 - bit_shift);
617 }
618 } else
619 vli_set(mod_m + word_shift, mod, ndigits);
620
621 for (i = 1; shift >= 0; --shift) {
622 u64 borrow = 0;
623 unsigned int j;
624
625 for (j = 0; j < ndigits * 2; ++j) {
626 u64 diff = v[i][j] - mod_m[j] - borrow;
627
628 if (diff != v[i][j])
629 borrow = (diff > v[i][j]);
630 v[1 - i][j] = diff;
631 }
632 i = !(i ^ borrow); /* Swap the index if there was no borrow */
633 vli_rshift1(mod_m, ndigits);
634 mod_m[ndigits - 1] |= mod_m[ndigits] << (64 - 1);
635 vli_rshift1(mod_m + ndigits, ndigits);
636 }
637 vli_set(result, v[i], ndigits);
638}
639
640/* Computes result = product % mod using Barrett's reduction with precomputed
641 * value mu appended to the mod after ndigits, mu = (2^{2w} / mod) and have
642 * length ndigits + 1, where mu * (2^w - 1) should not overflow ndigits
643 * boundary.
644 *
645 * Reference:
646 * R. Brent, P. Zimmermann. Modern Computer Arithmetic. 2010.
647 * 2.4.1 Barrett's algorithm. Algorithm 2.5.
648 */
649static void vli_mmod_barrett(u64 *result, u64 *product, const u64 *mod,
650 unsigned int ndigits)
651{
652 u64 q[ECC_MAX_DIGITS * 2];
653 u64 r[ECC_MAX_DIGITS * 2];
654 const u64 *mu = mod + ndigits;
655
656 vli_mult(q, product + ndigits, mu, ndigits);
657 if (mu[ndigits])
658 vli_add(q + ndigits, q + ndigits, product + ndigits, ndigits);
659 vli_mult(r, mod, q + ndigits, ndigits);
660 vli_sub(r, product, r, ndigits * 2);
661 while (!vli_is_zero(r + ndigits, ndigits) ||
662 vli_cmp(r, mod, ndigits) != -1) {
663 u64 carry;
664
665 carry = vli_sub(r, r, mod, ndigits);
666 vli_usub(r + ndigits, r + ndigits, carry, ndigits);
667 }
668 vli_set(result, r, ndigits);
669}
670
671/* Computes p_result = p_product % curve_p.
672 * See algorithm 5 and 6 from
673 * http://www.isys.uni-klu.ac.at/PDF/2001-0126-MT.pdf
674 */
675static void vli_mmod_fast_192(u64 *result, const u64 *product,
676 const u64 *curve_prime, u64 *tmp)
677{
678 const unsigned int ndigits = 3;
679 int carry;
680
681 vli_set(result, product, ndigits);
682
683 vli_set(tmp, &product[3], ndigits);
684 carry = vli_add(result, result, tmp, ndigits);
685
686 tmp[0] = 0;
687 tmp[1] = product[3];
688 tmp[2] = product[4];
689 carry += vli_add(result, result, tmp, ndigits);
690
691 tmp[0] = tmp[1] = product[5];
692 tmp[2] = 0;
693 carry += vli_add(result, result, tmp, ndigits);
694
695 while (carry || vli_cmp(curve_prime, result, ndigits) != 1)
696 carry -= vli_sub(result, result, curve_prime, ndigits);
697}
698
699/* Computes result = product % curve_prime
700 * from http://www.nsa.gov/ia/_files/nist-routines.pdf
701 */
702static void vli_mmod_fast_256(u64 *result, const u64 *product,
703 const u64 *curve_prime, u64 *tmp)
704{
705 int carry;
706 const unsigned int ndigits = 4;
707
708 /* t */
709 vli_set(result, product, ndigits);
710
711 /* s1 */
712 tmp[0] = 0;
713 tmp[1] = product[5] & 0xffffffff00000000ull;
714 tmp[2] = product[6];
715 tmp[3] = product[7];
716 carry = vli_lshift(tmp, tmp, 1, ndigits);
717 carry += vli_add(result, result, tmp, ndigits);
718
719 /* s2 */
720 tmp[1] = product[6] << 32;
721 tmp[2] = (product[6] >> 32) | (product[7] << 32);
722 tmp[3] = product[7] >> 32;
723 carry += vli_lshift(tmp, tmp, 1, ndigits);
724 carry += vli_add(result, result, tmp, ndigits);
725
726 /* s3 */
727 tmp[0] = product[4];
728 tmp[1] = product[5] & 0xffffffff;
729 tmp[2] = 0;
730 tmp[3] = product[7];
731 carry += vli_add(result, result, tmp, ndigits);
732
733 /* s4 */
734 tmp[0] = (product[4] >> 32) | (product[5] << 32);
735 tmp[1] = (product[5] >> 32) | (product[6] & 0xffffffff00000000ull);
736 tmp[2] = product[7];
737 tmp[3] = (product[6] >> 32) | (product[4] << 32);
738 carry += vli_add(result, result, tmp, ndigits);
739
740 /* d1 */
741 tmp[0] = (product[5] >> 32) | (product[6] << 32);
742 tmp[1] = (product[6] >> 32);
743 tmp[2] = 0;
744 tmp[3] = (product[4] & 0xffffffff) | (product[5] << 32);
745 carry -= vli_sub(result, result, tmp, ndigits);
746
747 /* d2 */
748 tmp[0] = product[6];
749 tmp[1] = product[7];
750 tmp[2] = 0;
751 tmp[3] = (product[4] >> 32) | (product[5] & 0xffffffff00000000ull);
752 carry -= vli_sub(result, result, tmp, ndigits);
753
754 /* d3 */
755 tmp[0] = (product[6] >> 32) | (product[7] << 32);
756 tmp[1] = (product[7] >> 32) | (product[4] << 32);
757 tmp[2] = (product[4] >> 32) | (product[5] << 32);
758 tmp[3] = (product[6] << 32);
759 carry -= vli_sub(result, result, tmp, ndigits);
760
761 /* d4 */
762 tmp[0] = product[7];
763 tmp[1] = product[4] & 0xffffffff00000000ull;
764 tmp[2] = product[5];
765 tmp[3] = product[6] & 0xffffffff00000000ull;
766 carry -= vli_sub(result, result, tmp, ndigits);
767
768 if (carry < 0) {
769 do {
770 carry += vli_add(result, result, curve_prime, ndigits);
771 } while (carry < 0);
772 } else {
773 while (carry || vli_cmp(curve_prime, result, ndigits) != 1)
774 carry -= vli_sub(result, result, curve_prime, ndigits);
775 }
776}
777
778/* Computes result = product % curve_prime for different curve_primes.
779 *
780 * Note that curve_primes are distinguished just by heuristic check and
781 * not by complete conformance check.
782 */
783static bool vli_mmod_fast(u64 *result, u64 *product,
784 const u64 *curve_prime, unsigned int ndigits)
785{
786 u64 tmp[2 * ECC_MAX_DIGITS];
787
788 /* Currently, both NIST primes have -1 in lowest qword. */
789 if (curve_prime[0] != -1ull) {
790 /* Try to handle Pseudo-Marsenne primes. */
791 if (curve_prime[ndigits - 1] == -1ull) {
792 vli_mmod_special(result, product, curve_prime,
793 ndigits);
794 return true;
795 } else if (curve_prime[ndigits - 1] == 1ull << 63 &&
796 curve_prime[ndigits - 2] == 0) {
797 vli_mmod_special2(result, product, curve_prime,
798 ndigits);
799 return true;
800 }
801 vli_mmod_barrett(result, product, curve_prime, ndigits);
802 return true;
803 }
804
805 switch (ndigits) {
806 case 3:
807 vli_mmod_fast_192(result, product, curve_prime, tmp);
808 break;
809 case 4:
810 vli_mmod_fast_256(result, product, curve_prime, tmp);
811 break;
812 default:
813 pr_err_ratelimited("ecc: unsupported digits size!\n");
814 return false;
815 }
816
817 return true;
818}
819
820/* Computes result = (left * right) % mod.
821 * Assumes that mod is big enough curve order.
822 */
823void vli_mod_mult_slow(u64 *result, const u64 *left, const u64 *right,
824 const u64 *mod, unsigned int ndigits)
825{
826 u64 product[ECC_MAX_DIGITS * 2];
827
828 vli_mult(product, left, right, ndigits);
829 vli_mmod_slow(result, product, mod, ndigits);
830}
831EXPORT_SYMBOL(vli_mod_mult_slow);
832
833/* Computes result = (left * right) % curve_prime. */
834static void vli_mod_mult_fast(u64 *result, const u64 *left, const u64 *right,
835 const u64 *curve_prime, unsigned int ndigits)
836{
837 u64 product[2 * ECC_MAX_DIGITS];
838
839 vli_mult(product, left, right, ndigits);
840 vli_mmod_fast(result, product, curve_prime, ndigits);
841}
842
843/* Computes result = left^2 % curve_prime. */
844static void vli_mod_square_fast(u64 *result, const u64 *left,
845 const u64 *curve_prime, unsigned int ndigits)
846{
847 u64 product[2 * ECC_MAX_DIGITS];
848
849 vli_square(product, left, ndigits);
850 vli_mmod_fast(result, product, curve_prime, ndigits);
851}
852
853#define EVEN(vli) (!(vli[0] & 1))
854/* Computes result = (1 / p_input) % mod. All VLIs are the same size.
855 * See "From Euclid's GCD to Montgomery Multiplication to the Great Divide"
856 * https://labs.oracle.com/techrep/2001/smli_tr-2001-95.pdf
857 */
858void vli_mod_inv(u64 *result, const u64 *input, const u64 *mod,
859 unsigned int ndigits)
860{
861 u64 a[ECC_MAX_DIGITS], b[ECC_MAX_DIGITS];
862 u64 u[ECC_MAX_DIGITS], v[ECC_MAX_DIGITS];
863 u64 carry;
864 int cmp_result;
865
866 if (vli_is_zero(input, ndigits)) {
867 vli_clear(result, ndigits);
868 return;
869 }
870
871 vli_set(a, input, ndigits);
872 vli_set(b, mod, ndigits);
873 vli_clear(u, ndigits);
874 u[0] = 1;
875 vli_clear(v, ndigits);
876
877 while ((cmp_result = vli_cmp(a, b, ndigits)) != 0) {
878 carry = 0;
879
880 if (EVEN(a)) {
881 vli_rshift1(a, ndigits);
882
883 if (!EVEN(u))
884 carry = vli_add(u, u, mod, ndigits);
885
886 vli_rshift1(u, ndigits);
887 if (carry)
888 u[ndigits - 1] |= 0x8000000000000000ull;
889 } else if (EVEN(b)) {
890 vli_rshift1(b, ndigits);
891
892 if (!EVEN(v))
893 carry = vli_add(v, v, mod, ndigits);
894
895 vli_rshift1(v, ndigits);
896 if (carry)
897 v[ndigits - 1] |= 0x8000000000000000ull;
898 } else if (cmp_result > 0) {
899 vli_sub(a, a, b, ndigits);
900 vli_rshift1(a, ndigits);
901
902 if (vli_cmp(u, v, ndigits) < 0)
903 vli_add(u, u, mod, ndigits);
904
905 vli_sub(u, u, v, ndigits);
906 if (!EVEN(u))
907 carry = vli_add(u, u, mod, ndigits);
908
909 vli_rshift1(u, ndigits);
910 if (carry)
911 u[ndigits - 1] |= 0x8000000000000000ull;
912 } else {
913 vli_sub(b, b, a, ndigits);
914 vli_rshift1(b, ndigits);
915
916 if (vli_cmp(v, u, ndigits) < 0)
917 vli_add(v, v, mod, ndigits);
918
919 vli_sub(v, v, u, ndigits);
920 if (!EVEN(v))
921 carry = vli_add(v, v, mod, ndigits);
922
923 vli_rshift1(v, ndigits);
924 if (carry)
925 v[ndigits - 1] |= 0x8000000000000000ull;
926 }
927 }
928
929 vli_set(result, u, ndigits);
930}
931EXPORT_SYMBOL(vli_mod_inv);
932
933/* ------ Point operations ------ */
934
935/* Returns true if p_point is the point at infinity, false otherwise. */
936static bool ecc_point_is_zero(const struct ecc_point *point)
937{
938 return (vli_is_zero(point->x, point->ndigits) &&
939 vli_is_zero(point->y, point->ndigits));
940}
941
942/* Point multiplication algorithm using Montgomery's ladder with co-Z
943 * coordinates. From http://eprint.iacr.org/2011/338.pdf
944 */
945
946/* Double in place */
947static void ecc_point_double_jacobian(u64 *x1, u64 *y1, u64 *z1,
948 u64 *curve_prime, unsigned int ndigits)
949{
950 /* t1 = x, t2 = y, t3 = z */
951 u64 t4[ECC_MAX_DIGITS];
952 u64 t5[ECC_MAX_DIGITS];
953
954 if (vli_is_zero(z1, ndigits))
955 return;
956
957 /* t4 = y1^2 */
958 vli_mod_square_fast(t4, y1, curve_prime, ndigits);
959 /* t5 = x1*y1^2 = A */
960 vli_mod_mult_fast(t5, x1, t4, curve_prime, ndigits);
961 /* t4 = y1^4 */
962 vli_mod_square_fast(t4, t4, curve_prime, ndigits);
963 /* t2 = y1*z1 = z3 */
964 vli_mod_mult_fast(y1, y1, z1, curve_prime, ndigits);
965 /* t3 = z1^2 */
966 vli_mod_square_fast(z1, z1, curve_prime, ndigits);
967
968 /* t1 = x1 + z1^2 */
969 vli_mod_add(x1, x1, z1, curve_prime, ndigits);
970 /* t3 = 2*z1^2 */
971 vli_mod_add(z1, z1, z1, curve_prime, ndigits);
972 /* t3 = x1 - z1^2 */
973 vli_mod_sub(z1, x1, z1, curve_prime, ndigits);
974 /* t1 = x1^2 - z1^4 */
975 vli_mod_mult_fast(x1, x1, z1, curve_prime, ndigits);
976
977 /* t3 = 2*(x1^2 - z1^4) */
978 vli_mod_add(z1, x1, x1, curve_prime, ndigits);
979 /* t1 = 3*(x1^2 - z1^4) */
980 vli_mod_add(x1, x1, z1, curve_prime, ndigits);
981 if (vli_test_bit(x1, 0)) {
982 u64 carry = vli_add(x1, x1, curve_prime, ndigits);
983
984 vli_rshift1(x1, ndigits);
985 x1[ndigits - 1] |= carry << 63;
986 } else {
987 vli_rshift1(x1, ndigits);
988 }
989 /* t1 = 3/2*(x1^2 - z1^4) = B */
990
991 /* t3 = B^2 */
992 vli_mod_square_fast(z1, x1, curve_prime, ndigits);
993 /* t3 = B^2 - A */
994 vli_mod_sub(z1, z1, t5, curve_prime, ndigits);
995 /* t3 = B^2 - 2A = x3 */
996 vli_mod_sub(z1, z1, t5, curve_prime, ndigits);
997 /* t5 = A - x3 */
998 vli_mod_sub(t5, t5, z1, curve_prime, ndigits);
999 /* t1 = B * (A - x3) */
1000 vli_mod_mult_fast(x1, x1, t5, curve_prime, ndigits);
1001 /* t4 = B * (A - x3) - y1^4 = y3 */
1002 vli_mod_sub(t4, x1, t4, curve_prime, ndigits);
1003
1004 vli_set(x1, z1, ndigits);
1005 vli_set(z1, y1, ndigits);
1006 vli_set(y1, t4, ndigits);
1007}
1008
1009/* Modify (x1, y1) => (x1 * z^2, y1 * z^3) */
1010static void apply_z(u64 *x1, u64 *y1, u64 *z, u64 *curve_prime,
1011 unsigned int ndigits)
1012{
1013 u64 t1[ECC_MAX_DIGITS];
1014
1015 vli_mod_square_fast(t1, z, curve_prime, ndigits); /* z^2 */
1016 vli_mod_mult_fast(x1, x1, t1, curve_prime, ndigits); /* x1 * z^2 */
1017 vli_mod_mult_fast(t1, t1, z, curve_prime, ndigits); /* z^3 */
1018 vli_mod_mult_fast(y1, y1, t1, curve_prime, ndigits); /* y1 * z^3 */
1019}
1020
1021/* P = (x1, y1) => 2P, (x2, y2) => P' */
1022static void xycz_initial_double(u64 *x1, u64 *y1, u64 *x2, u64 *y2,
1023 u64 *p_initial_z, u64 *curve_prime,
1024 unsigned int ndigits)
1025{
1026 u64 z[ECC_MAX_DIGITS];
1027
1028 vli_set(x2, x1, ndigits);
1029 vli_set(y2, y1, ndigits);
1030
1031 vli_clear(z, ndigits);
1032 z[0] = 1;
1033
1034 if (p_initial_z)
1035 vli_set(z, p_initial_z, ndigits);
1036
1037 apply_z(x1, y1, z, curve_prime, ndigits);
1038
1039 ecc_point_double_jacobian(x1, y1, z, curve_prime, ndigits);
1040
1041 apply_z(x2, y2, z, curve_prime, ndigits);
1042}
1043
1044/* Input P = (x1, y1, Z), Q = (x2, y2, Z)
1045 * Output P' = (x1', y1', Z3), P + Q = (x3, y3, Z3)
1046 * or P => P', Q => P + Q
1047 */
1048static void xycz_add(u64 *x1, u64 *y1, u64 *x2, u64 *y2, u64 *curve_prime,
1049 unsigned int ndigits)
1050{
1051 /* t1 = X1, t2 = Y1, t3 = X2, t4 = Y2 */
1052 u64 t5[ECC_MAX_DIGITS];
1053
1054 /* t5 = x2 - x1 */
1055 vli_mod_sub(t5, x2, x1, curve_prime, ndigits);
1056 /* t5 = (x2 - x1)^2 = A */
1057 vli_mod_square_fast(t5, t5, curve_prime, ndigits);
1058 /* t1 = x1*A = B */
1059 vli_mod_mult_fast(x1, x1, t5, curve_prime, ndigits);
1060 /* t3 = x2*A = C */
1061 vli_mod_mult_fast(x2, x2, t5, curve_prime, ndigits);
1062 /* t4 = y2 - y1 */
1063 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1064 /* t5 = (y2 - y1)^2 = D */
1065 vli_mod_square_fast(t5, y2, curve_prime, ndigits);
1066
1067 /* t5 = D - B */
1068 vli_mod_sub(t5, t5, x1, curve_prime, ndigits);
1069 /* t5 = D - B - C = x3 */
1070 vli_mod_sub(t5, t5, x2, curve_prime, ndigits);
1071 /* t3 = C - B */
1072 vli_mod_sub(x2, x2, x1, curve_prime, ndigits);
1073 /* t2 = y1*(C - B) */
1074 vli_mod_mult_fast(y1, y1, x2, curve_prime, ndigits);
1075 /* t3 = B - x3 */
1076 vli_mod_sub(x2, x1, t5, curve_prime, ndigits);
1077 /* t4 = (y2 - y1)*(B - x3) */
1078 vli_mod_mult_fast(y2, y2, x2, curve_prime, ndigits);
1079 /* t4 = y3 */
1080 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1081
1082 vli_set(x2, t5, ndigits);
1083}
1084
1085/* Input P = (x1, y1, Z), Q = (x2, y2, Z)
1086 * Output P + Q = (x3, y3, Z3), P - Q = (x3', y3', Z3)
1087 * or P => P - Q, Q => P + Q
1088 */
1089static void xycz_add_c(u64 *x1, u64 *y1, u64 *x2, u64 *y2, u64 *curve_prime,
1090 unsigned int ndigits)
1091{
1092 /* t1 = X1, t2 = Y1, t3 = X2, t4 = Y2 */
1093 u64 t5[ECC_MAX_DIGITS];
1094 u64 t6[ECC_MAX_DIGITS];
1095 u64 t7[ECC_MAX_DIGITS];
1096
1097 /* t5 = x2 - x1 */
1098 vli_mod_sub(t5, x2, x1, curve_prime, ndigits);
1099 /* t5 = (x2 - x1)^2 = A */
1100 vli_mod_square_fast(t5, t5, curve_prime, ndigits);
1101 /* t1 = x1*A = B */
1102 vli_mod_mult_fast(x1, x1, t5, curve_prime, ndigits);
1103 /* t3 = x2*A = C */
1104 vli_mod_mult_fast(x2, x2, t5, curve_prime, ndigits);
1105 /* t4 = y2 + y1 */
1106 vli_mod_add(t5, y2, y1, curve_prime, ndigits);
1107 /* t4 = y2 - y1 */
1108 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1109
1110 /* t6 = C - B */
1111 vli_mod_sub(t6, x2, x1, curve_prime, ndigits);
1112 /* t2 = y1 * (C - B) */
1113 vli_mod_mult_fast(y1, y1, t6, curve_prime, ndigits);
1114 /* t6 = B + C */
1115 vli_mod_add(t6, x1, x2, curve_prime, ndigits);
1116 /* t3 = (y2 - y1)^2 */
1117 vli_mod_square_fast(x2, y2, curve_prime, ndigits);
1118 /* t3 = x3 */
1119 vli_mod_sub(x2, x2, t6, curve_prime, ndigits);
1120
1121 /* t7 = B - x3 */
1122 vli_mod_sub(t7, x1, x2, curve_prime, ndigits);
1123 /* t4 = (y2 - y1)*(B - x3) */
1124 vli_mod_mult_fast(y2, y2, t7, curve_prime, ndigits);
1125 /* t4 = y3 */
1126 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1127
1128 /* t7 = (y2 + y1)^2 = F */
1129 vli_mod_square_fast(t7, t5, curve_prime, ndigits);
1130 /* t7 = x3' */
1131 vli_mod_sub(t7, t7, t6, curve_prime, ndigits);
1132 /* t6 = x3' - B */
1133 vli_mod_sub(t6, t7, x1, curve_prime, ndigits);
1134 /* t6 = (y2 + y1)*(x3' - B) */
1135 vli_mod_mult_fast(t6, t6, t5, curve_prime, ndigits);
1136 /* t2 = y3' */
1137 vli_mod_sub(y1, t6, y1, curve_prime, ndigits);
1138
1139 vli_set(x1, t7, ndigits);
1140}
1141
1142static void ecc_point_mult(struct ecc_point *result,
1143 const struct ecc_point *point, const u64 *scalar,
1144 u64 *initial_z, const struct ecc_curve *curve,
1145 unsigned int ndigits)
1146{
1147 /* R0 and R1 */
1148 u64 rx[2][ECC_MAX_DIGITS];
1149 u64 ry[2][ECC_MAX_DIGITS];
1150 u64 z[ECC_MAX_DIGITS];
1151 u64 sk[2][ECC_MAX_DIGITS];
1152 u64 *curve_prime = curve->p;
1153 int i, nb;
1154 int num_bits;
1155 int carry;
1156
1157 carry = vli_add(sk[0], scalar, curve->n, ndigits);
1158 vli_add(sk[1], sk[0], curve->n, ndigits);
1159 scalar = sk[!carry];
1160 num_bits = sizeof(u64) * ndigits * 8 + 1;
1161
1162 vli_set(rx[1], point->x, ndigits);
1163 vli_set(ry[1], point->y, ndigits);
1164
1165 xycz_initial_double(rx[1], ry[1], rx[0], ry[0], initial_z, curve_prime,
1166 ndigits);
1167
1168 for (i = num_bits - 2; i > 0; i--) {
1169 nb = !vli_test_bit(scalar, i);
1170 xycz_add_c(rx[1 - nb], ry[1 - nb], rx[nb], ry[nb], curve_prime,
1171 ndigits);
1172 xycz_add(rx[nb], ry[nb], rx[1 - nb], ry[1 - nb], curve_prime,
1173 ndigits);
1174 }
1175
1176 nb = !vli_test_bit(scalar, 0);
1177 xycz_add_c(rx[1 - nb], ry[1 - nb], rx[nb], ry[nb], curve_prime,
1178 ndigits);
1179
1180 /* Find final 1/Z value. */
1181 /* X1 - X0 */
1182 vli_mod_sub(z, rx[1], rx[0], curve_prime, ndigits);
1183 /* Yb * (X1 - X0) */
1184 vli_mod_mult_fast(z, z, ry[1 - nb], curve_prime, ndigits);
1185 /* xP * Yb * (X1 - X0) */
1186 vli_mod_mult_fast(z, z, point->x, curve_prime, ndigits);
1187
1188 /* 1 / (xP * Yb * (X1 - X0)) */
1189 vli_mod_inv(z, z, curve_prime, point->ndigits);
1190
1191 /* yP / (xP * Yb * (X1 - X0)) */
1192 vli_mod_mult_fast(z, z, point->y, curve_prime, ndigits);
1193 /* Xb * yP / (xP * Yb * (X1 - X0)) */
1194 vli_mod_mult_fast(z, z, rx[1 - nb], curve_prime, ndigits);
1195 /* End 1/Z calculation */
1196
1197 xycz_add(rx[nb], ry[nb], rx[1 - nb], ry[1 - nb], curve_prime, ndigits);
1198
1199 apply_z(rx[0], ry[0], z, curve_prime, ndigits);
1200
1201 vli_set(result->x, rx[0], ndigits);
1202 vli_set(result->y, ry[0], ndigits);
1203}
1204
1205/* Computes R = P + Q mod p */
1206static void ecc_point_add(const struct ecc_point *result,
1207 const struct ecc_point *p, const struct ecc_point *q,
1208 const struct ecc_curve *curve)
1209{
1210 u64 z[ECC_MAX_DIGITS];
1211 u64 px[ECC_MAX_DIGITS];
1212 u64 py[ECC_MAX_DIGITS];
1213 unsigned int ndigits = curve->g.ndigits;
1214
1215 vli_set(result->x, q->x, ndigits);
1216 vli_set(result->y, q->y, ndigits);
1217 vli_mod_sub(z, result->x, p->x, curve->p, ndigits);
1218 vli_set(px, p->x, ndigits);
1219 vli_set(py, p->y, ndigits);
1220 xycz_add(px, py, result->x, result->y, curve->p, ndigits);
1221 vli_mod_inv(z, z, curve->p, ndigits);
1222 apply_z(result->x, result->y, z, curve->p, ndigits);
1223}
1224
1225/* Computes R = u1P + u2Q mod p using Shamir's trick.
1226 * Based on: Kenneth MacKay's micro-ecc (2014).
1227 */
1228void ecc_point_mult_shamir(const struct ecc_point *result,
1229 const u64 *u1, const struct ecc_point *p,
1230 const u64 *u2, const struct ecc_point *q,
1231 const struct ecc_curve *curve)
1232{
1233 u64 z[ECC_MAX_DIGITS];
1234 u64 sump[2][ECC_MAX_DIGITS];
1235 u64 *rx = result->x;
1236 u64 *ry = result->y;
1237 unsigned int ndigits = curve->g.ndigits;
1238 unsigned int num_bits;
1239 struct ecc_point sum = ECC_POINT_INIT(sump[0], sump[1], ndigits);
1240 const struct ecc_point *points[4];
1241 const struct ecc_point *point;
1242 unsigned int idx;
1243 int i;
1244
1245 ecc_point_add(&sum, p, q, curve);
1246 points[0] = NULL;
1247 points[1] = p;
1248 points[2] = q;
1249 points[3] = ∑
1250
1251 num_bits = max(vli_num_bits(u1, ndigits),
1252 vli_num_bits(u2, ndigits));
1253 i = num_bits - 1;
1254 idx = (!!vli_test_bit(u1, i)) | ((!!vli_test_bit(u2, i)) << 1);
1255 point = points[idx];
1256
1257 vli_set(rx, point->x, ndigits);
1258 vli_set(ry, point->y, ndigits);
1259 vli_clear(z + 1, ndigits - 1);
1260 z[0] = 1;
1261
1262 for (--i; i >= 0; i--) {
1263 ecc_point_double_jacobian(rx, ry, z, curve->p, ndigits);
1264 idx = (!!vli_test_bit(u1, i)) | ((!!vli_test_bit(u2, i)) << 1);
1265 point = points[idx];
1266 if (point) {
1267 u64 tx[ECC_MAX_DIGITS];
1268 u64 ty[ECC_MAX_DIGITS];
1269 u64 tz[ECC_MAX_DIGITS];
1270
1271 vli_set(tx, point->x, ndigits);
1272 vli_set(ty, point->y, ndigits);
1273 apply_z(tx, ty, z, curve->p, ndigits);
1274 vli_mod_sub(tz, rx, tx, curve->p, ndigits);
1275 xycz_add(tx, ty, rx, ry, curve->p, ndigits);
1276 vli_mod_mult_fast(z, z, tz, curve->p, ndigits);
1277 }
1278 }
1279 vli_mod_inv(z, z, curve->p, ndigits);
1280 apply_z(rx, ry, z, curve->p, ndigits);
1281}
1282EXPORT_SYMBOL(ecc_point_mult_shamir);
1283
1284static inline void ecc_swap_digits(const u64 *in, u64 *out,
1285 unsigned int ndigits)
1286{
1287 int i;
1288
1289 for (i = 0; i < ndigits; i++)
1290 out[i] = __swab64(in[ndigits - 1 - i]);
1291}
1292
1293static int __ecc_is_key_valid(const struct ecc_curve *curve,
1294 const u64 *private_key, unsigned int ndigits)
1295{
1296 u64 one[ECC_MAX_DIGITS] = { 1, };
1297 u64 res[ECC_MAX_DIGITS];
1298
1299 if (!private_key)
1300 return -EINVAL;
1301
1302 if (curve->g.ndigits != ndigits)
1303 return -EINVAL;
1304
1305 /* Make sure the private key is in the range [2, n-3]. */
1306 if (vli_cmp(one, private_key, ndigits) != -1)
1307 return -EINVAL;
1308 vli_sub(res, curve->n, one, ndigits);
1309 vli_sub(res, res, one, ndigits);
1310 if (vli_cmp(res, private_key, ndigits) != 1)
1311 return -EINVAL;
1312
1313 return 0;
1314}
1315
1316int ecc_is_key_valid(unsigned int curve_id, unsigned int ndigits,
1317 const u64 *private_key, unsigned int private_key_len)
1318{
1319 int nbytes;
1320 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1321
1322 nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1323
1324 if (private_key_len != nbytes)
1325 return -EINVAL;
1326
1327 return __ecc_is_key_valid(curve, private_key, ndigits);
1328}
1329EXPORT_SYMBOL(ecc_is_key_valid);
1330
1331/*
1332 * ECC private keys are generated using the method of extra random bits,
1333 * equivalent to that described in FIPS 186-4, Appendix B.4.1.
1334 *
1335 * d = (c mod(n–1)) + 1 where c is a string of random bits, 64 bits longer
1336 * than requested
1337 * 0 <= c mod(n-1) <= n-2 and implies that
1338 * 1 <= d <= n-1
1339 *
1340 * This method generates a private key uniformly distributed in the range
1341 * [1, n-1].
1342 */
1343int ecc_gen_privkey(unsigned int curve_id, unsigned int ndigits, u64 *privkey)
1344{
1345 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1346 u64 priv[ECC_MAX_DIGITS];
1347 unsigned int nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1348 unsigned int nbits = vli_num_bits(curve->n, ndigits);
1349 int err;
1350
1351 /* Check that N is included in Table 1 of FIPS 186-4, section 6.1.1 */
1352 if (nbits < 160 || ndigits > ARRAY_SIZE(priv))
1353 return -EINVAL;
1354
1355 /*
1356 * FIPS 186-4 recommends that the private key should be obtained from a
1357 * RBG with a security strength equal to or greater than the security
1358 * strength associated with N.
1359 *
1360 * The maximum security strength identified by NIST SP800-57pt1r4 for
1361 * ECC is 256 (N >= 512).
1362 *
1363 * This condition is met by the default RNG because it selects a favored
1364 * DRBG with a security strength of 256.
1365 */
1366 if (crypto_get_default_rng())
1367 return -EFAULT;
1368
1369 err = crypto_rng_get_bytes(crypto_default_rng, (u8 *)priv, nbytes);
1370 crypto_put_default_rng();
1371 if (err)
1372 return err;
1373
1374 /* Make sure the private key is in the valid range. */
1375 if (__ecc_is_key_valid(curve, priv, ndigits))
1376 return -EINVAL;
1377
1378 ecc_swap_digits(priv, privkey, ndigits);
1379
1380 return 0;
1381}
1382EXPORT_SYMBOL(ecc_gen_privkey);
1383
1384int ecc_make_pub_key(unsigned int curve_id, unsigned int ndigits,
1385 const u64 *private_key, u64 *public_key)
1386{
1387 int ret = 0;
1388 struct ecc_point *pk;
1389 u64 priv[ECC_MAX_DIGITS];
1390 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1391
1392 if (!private_key || !curve || ndigits > ARRAY_SIZE(priv)) {
1393 ret = -EINVAL;
1394 goto out;
1395 }
1396
1397 ecc_swap_digits(private_key, priv, ndigits);
1398
1399 pk = ecc_alloc_point(ndigits);
1400 if (!pk) {
1401 ret = -ENOMEM;
1402 goto out;
1403 }
1404
1405 ecc_point_mult(pk, &curve->g, priv, NULL, curve, ndigits);
1406 if (ecc_point_is_zero(pk)) {
1407 ret = -EAGAIN;
1408 goto err_free_point;
1409 }
1410
1411 ecc_swap_digits(pk->x, public_key, ndigits);
1412 ecc_swap_digits(pk->y, &public_key[ndigits], ndigits);
1413
1414err_free_point:
1415 ecc_free_point(pk);
1416out:
1417 return ret;
1418}
1419EXPORT_SYMBOL(ecc_make_pub_key);
1420
1421/* SP800-56A section 5.6.2.3.4 partial verification: ephemeral keys only */
1422int ecc_is_pubkey_valid_partial(const struct ecc_curve *curve,
1423 struct ecc_point *pk)
1424{
1425 u64 yy[ECC_MAX_DIGITS], xxx[ECC_MAX_DIGITS], w[ECC_MAX_DIGITS];
1426
1427 if (WARN_ON(pk->ndigits != curve->g.ndigits))
1428 return -EINVAL;
1429
1430 /* Check 1: Verify key is not the zero point. */
1431 if (ecc_point_is_zero(pk))
1432 return -EINVAL;
1433
1434 /* Check 2: Verify key is in the range [1, p-1]. */
1435 if (vli_cmp(curve->p, pk->x, pk->ndigits) != 1)
1436 return -EINVAL;
1437 if (vli_cmp(curve->p, pk->y, pk->ndigits) != 1)
1438 return -EINVAL;
1439
1440 /* Check 3: Verify that y^2 == (x^3 + a·x + b) mod p */
1441 vli_mod_square_fast(yy, pk->y, curve->p, pk->ndigits); /* y^2 */
1442 vli_mod_square_fast(xxx, pk->x, curve->p, pk->ndigits); /* x^2 */
1443 vli_mod_mult_fast(xxx, xxx, pk->x, curve->p, pk->ndigits); /* x^3 */
1444 vli_mod_mult_fast(w, curve->a, pk->x, curve->p, pk->ndigits); /* a·x */
1445 vli_mod_add(w, w, curve->b, curve->p, pk->ndigits); /* a·x + b */
1446 vli_mod_add(w, w, xxx, curve->p, pk->ndigits); /* x^3 + a·x + b */
1447 if (vli_cmp(yy, w, pk->ndigits) != 0) /* Equation */
1448 return -EINVAL;
1449
1450 return 0;
1451}
1452EXPORT_SYMBOL(ecc_is_pubkey_valid_partial);
1453
1454int crypto_ecdh_shared_secret(unsigned int curve_id, unsigned int ndigits,
1455 const u64 *private_key, const u64 *public_key,
1456 u64 *secret)
1457{
1458 int ret = 0;
1459 struct ecc_point *product, *pk;
1460 u64 priv[ECC_MAX_DIGITS];
1461 u64 rand_z[ECC_MAX_DIGITS];
1462 unsigned int nbytes;
1463 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1464
1465 if (!private_key || !public_key || !curve ||
1466 ndigits > ARRAY_SIZE(priv) || ndigits > ARRAY_SIZE(rand_z)) {
1467 ret = -EINVAL;
1468 goto out;
1469 }
1470
1471 nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1472
1473 get_random_bytes(rand_z, nbytes);
1474
1475 pk = ecc_alloc_point(ndigits);
1476 if (!pk) {
1477 ret = -ENOMEM;
1478 goto out;
1479 }
1480
1481 ecc_swap_digits(public_key, pk->x, ndigits);
1482 ecc_swap_digits(&public_key[ndigits], pk->y, ndigits);
1483 ret = ecc_is_pubkey_valid_partial(curve, pk);
1484 if (ret)
1485 goto err_alloc_product;
1486
1487 ecc_swap_digits(private_key, priv, ndigits);
1488
1489 product = ecc_alloc_point(ndigits);
1490 if (!product) {
1491 ret = -ENOMEM;
1492 goto err_alloc_product;
1493 }
1494
1495 ecc_point_mult(product, pk, priv, rand_z, curve, ndigits);
1496
1497 ecc_swap_digits(product->x, secret, ndigits);
1498
1499 if (ecc_point_is_zero(product))
1500 ret = -EFAULT;
1501
1502 ecc_free_point(product);
1503err_alloc_product:
1504 ecc_free_point(pk);
1505out:
1506 return ret;
1507}
1508EXPORT_SYMBOL(crypto_ecdh_shared_secret);
1509
1510MODULE_LICENSE("Dual BSD/GPL");
1/*
2 * Copyright (c) 2013, 2014 Kenneth MacKay. All rights reserved.
3 * Copyright (c) 2019 Vitaly Chikunov <vt@altlinux.org>
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met:
8 * * Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
15 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
16 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
17 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
18 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
20 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
24 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <crypto/ecc_curve.h>
28#include <linux/module.h>
29#include <linux/random.h>
30#include <linux/slab.h>
31#include <linux/swab.h>
32#include <linux/fips.h>
33#include <crypto/ecdh.h>
34#include <crypto/rng.h>
35#include <asm/unaligned.h>
36#include <linux/ratelimit.h>
37
38#include "ecc.h"
39#include "ecc_curve_defs.h"
40
41typedef struct {
42 u64 m_low;
43 u64 m_high;
44} uint128_t;
45
46/* Returns curv25519 curve param */
47const struct ecc_curve *ecc_get_curve25519(void)
48{
49 return &ecc_25519;
50}
51EXPORT_SYMBOL(ecc_get_curve25519);
52
53const struct ecc_curve *ecc_get_curve(unsigned int curve_id)
54{
55 switch (curve_id) {
56 /* In FIPS mode only allow P256 and higher */
57 case ECC_CURVE_NIST_P192:
58 return fips_enabled ? NULL : &nist_p192;
59 case ECC_CURVE_NIST_P256:
60 return &nist_p256;
61 case ECC_CURVE_NIST_P384:
62 return &nist_p384;
63 default:
64 return NULL;
65 }
66}
67EXPORT_SYMBOL(ecc_get_curve);
68
69static u64 *ecc_alloc_digits_space(unsigned int ndigits)
70{
71 size_t len = ndigits * sizeof(u64);
72
73 if (!len)
74 return NULL;
75
76 return kmalloc(len, GFP_KERNEL);
77}
78
79static void ecc_free_digits_space(u64 *space)
80{
81 kfree_sensitive(space);
82}
83
84static struct ecc_point *ecc_alloc_point(unsigned int ndigits)
85{
86 struct ecc_point *p = kmalloc(sizeof(*p), GFP_KERNEL);
87
88 if (!p)
89 return NULL;
90
91 p->x = ecc_alloc_digits_space(ndigits);
92 if (!p->x)
93 goto err_alloc_x;
94
95 p->y = ecc_alloc_digits_space(ndigits);
96 if (!p->y)
97 goto err_alloc_y;
98
99 p->ndigits = ndigits;
100
101 return p;
102
103err_alloc_y:
104 ecc_free_digits_space(p->x);
105err_alloc_x:
106 kfree(p);
107 return NULL;
108}
109
110static void ecc_free_point(struct ecc_point *p)
111{
112 if (!p)
113 return;
114
115 kfree_sensitive(p->x);
116 kfree_sensitive(p->y);
117 kfree_sensitive(p);
118}
119
120static void vli_clear(u64 *vli, unsigned int ndigits)
121{
122 int i;
123
124 for (i = 0; i < ndigits; i++)
125 vli[i] = 0;
126}
127
128/* Returns true if vli == 0, false otherwise. */
129bool vli_is_zero(const u64 *vli, unsigned int ndigits)
130{
131 int i;
132
133 for (i = 0; i < ndigits; i++) {
134 if (vli[i])
135 return false;
136 }
137
138 return true;
139}
140EXPORT_SYMBOL(vli_is_zero);
141
142/* Returns nonzero if bit of vli is set. */
143static u64 vli_test_bit(const u64 *vli, unsigned int bit)
144{
145 return (vli[bit / 64] & ((u64)1 << (bit % 64)));
146}
147
148static bool vli_is_negative(const u64 *vli, unsigned int ndigits)
149{
150 return vli_test_bit(vli, ndigits * 64 - 1);
151}
152
153/* Counts the number of 64-bit "digits" in vli. */
154static unsigned int vli_num_digits(const u64 *vli, unsigned int ndigits)
155{
156 int i;
157
158 /* Search from the end until we find a non-zero digit.
159 * We do it in reverse because we expect that most digits will
160 * be nonzero.
161 */
162 for (i = ndigits - 1; i >= 0 && vli[i] == 0; i--);
163
164 return (i + 1);
165}
166
167/* Counts the number of bits required for vli. */
168static unsigned int vli_num_bits(const u64 *vli, unsigned int ndigits)
169{
170 unsigned int i, num_digits;
171 u64 digit;
172
173 num_digits = vli_num_digits(vli, ndigits);
174 if (num_digits == 0)
175 return 0;
176
177 digit = vli[num_digits - 1];
178 for (i = 0; digit; i++)
179 digit >>= 1;
180
181 return ((num_digits - 1) * 64 + i);
182}
183
184/* Set dest from unaligned bit string src. */
185void vli_from_be64(u64 *dest, const void *src, unsigned int ndigits)
186{
187 int i;
188 const u64 *from = src;
189
190 for (i = 0; i < ndigits; i++)
191 dest[i] = get_unaligned_be64(&from[ndigits - 1 - i]);
192}
193EXPORT_SYMBOL(vli_from_be64);
194
195void vli_from_le64(u64 *dest, const void *src, unsigned int ndigits)
196{
197 int i;
198 const u64 *from = src;
199
200 for (i = 0; i < ndigits; i++)
201 dest[i] = get_unaligned_le64(&from[i]);
202}
203EXPORT_SYMBOL(vli_from_le64);
204
205/* Sets dest = src. */
206static void vli_set(u64 *dest, const u64 *src, unsigned int ndigits)
207{
208 int i;
209
210 for (i = 0; i < ndigits; i++)
211 dest[i] = src[i];
212}
213
214/* Returns sign of left - right. */
215int vli_cmp(const u64 *left, const u64 *right, unsigned int ndigits)
216{
217 int i;
218
219 for (i = ndigits - 1; i >= 0; i--) {
220 if (left[i] > right[i])
221 return 1;
222 else if (left[i] < right[i])
223 return -1;
224 }
225
226 return 0;
227}
228EXPORT_SYMBOL(vli_cmp);
229
230/* Computes result = in << c, returning carry. Can modify in place
231 * (if result == in). 0 < shift < 64.
232 */
233static u64 vli_lshift(u64 *result, const u64 *in, unsigned int shift,
234 unsigned int ndigits)
235{
236 u64 carry = 0;
237 int i;
238
239 for (i = 0; i < ndigits; i++) {
240 u64 temp = in[i];
241
242 result[i] = (temp << shift) | carry;
243 carry = temp >> (64 - shift);
244 }
245
246 return carry;
247}
248
249/* Computes vli = vli >> 1. */
250static void vli_rshift1(u64 *vli, unsigned int ndigits)
251{
252 u64 *end = vli;
253 u64 carry = 0;
254
255 vli += ndigits;
256
257 while (vli-- > end) {
258 u64 temp = *vli;
259 *vli = (temp >> 1) | carry;
260 carry = temp << 63;
261 }
262}
263
264/* Computes result = left + right, returning carry. Can modify in place. */
265static u64 vli_add(u64 *result, const u64 *left, const u64 *right,
266 unsigned int ndigits)
267{
268 u64 carry = 0;
269 int i;
270
271 for (i = 0; i < ndigits; i++) {
272 u64 sum;
273
274 sum = left[i] + right[i] + carry;
275 if (sum != left[i])
276 carry = (sum < left[i]);
277
278 result[i] = sum;
279 }
280
281 return carry;
282}
283
284/* Computes result = left + right, returning carry. Can modify in place. */
285static u64 vli_uadd(u64 *result, const u64 *left, u64 right,
286 unsigned int ndigits)
287{
288 u64 carry = right;
289 int i;
290
291 for (i = 0; i < ndigits; i++) {
292 u64 sum;
293
294 sum = left[i] + carry;
295 if (sum != left[i])
296 carry = (sum < left[i]);
297 else
298 carry = !!carry;
299
300 result[i] = sum;
301 }
302
303 return carry;
304}
305
306/* Computes result = left - right, returning borrow. Can modify in place. */
307u64 vli_sub(u64 *result, const u64 *left, const u64 *right,
308 unsigned int ndigits)
309{
310 u64 borrow = 0;
311 int i;
312
313 for (i = 0; i < ndigits; i++) {
314 u64 diff;
315
316 diff = left[i] - right[i] - borrow;
317 if (diff != left[i])
318 borrow = (diff > left[i]);
319
320 result[i] = diff;
321 }
322
323 return borrow;
324}
325EXPORT_SYMBOL(vli_sub);
326
327/* Computes result = left - right, returning borrow. Can modify in place. */
328static u64 vli_usub(u64 *result, const u64 *left, u64 right,
329 unsigned int ndigits)
330{
331 u64 borrow = right;
332 int i;
333
334 for (i = 0; i < ndigits; i++) {
335 u64 diff;
336
337 diff = left[i] - borrow;
338 if (diff != left[i])
339 borrow = (diff > left[i]);
340
341 result[i] = diff;
342 }
343
344 return borrow;
345}
346
347static uint128_t mul_64_64(u64 left, u64 right)
348{
349 uint128_t result;
350#if defined(CONFIG_ARCH_SUPPORTS_INT128)
351 unsigned __int128 m = (unsigned __int128)left * right;
352
353 result.m_low = m;
354 result.m_high = m >> 64;
355#else
356 u64 a0 = left & 0xffffffffull;
357 u64 a1 = left >> 32;
358 u64 b0 = right & 0xffffffffull;
359 u64 b1 = right >> 32;
360 u64 m0 = a0 * b0;
361 u64 m1 = a0 * b1;
362 u64 m2 = a1 * b0;
363 u64 m3 = a1 * b1;
364
365 m2 += (m0 >> 32);
366 m2 += m1;
367
368 /* Overflow */
369 if (m2 < m1)
370 m3 += 0x100000000ull;
371
372 result.m_low = (m0 & 0xffffffffull) | (m2 << 32);
373 result.m_high = m3 + (m2 >> 32);
374#endif
375 return result;
376}
377
378static uint128_t add_128_128(uint128_t a, uint128_t b)
379{
380 uint128_t result;
381
382 result.m_low = a.m_low + b.m_low;
383 result.m_high = a.m_high + b.m_high + (result.m_low < a.m_low);
384
385 return result;
386}
387
388static void vli_mult(u64 *result, const u64 *left, const u64 *right,
389 unsigned int ndigits)
390{
391 uint128_t r01 = { 0, 0 };
392 u64 r2 = 0;
393 unsigned int i, k;
394
395 /* Compute each digit of result in sequence, maintaining the
396 * carries.
397 */
398 for (k = 0; k < ndigits * 2 - 1; k++) {
399 unsigned int min;
400
401 if (k < ndigits)
402 min = 0;
403 else
404 min = (k + 1) - ndigits;
405
406 for (i = min; i <= k && i < ndigits; i++) {
407 uint128_t product;
408
409 product = mul_64_64(left[i], right[k - i]);
410
411 r01 = add_128_128(r01, product);
412 r2 += (r01.m_high < product.m_high);
413 }
414
415 result[k] = r01.m_low;
416 r01.m_low = r01.m_high;
417 r01.m_high = r2;
418 r2 = 0;
419 }
420
421 result[ndigits * 2 - 1] = r01.m_low;
422}
423
424/* Compute product = left * right, for a small right value. */
425static void vli_umult(u64 *result, const u64 *left, u32 right,
426 unsigned int ndigits)
427{
428 uint128_t r01 = { 0 };
429 unsigned int k;
430
431 for (k = 0; k < ndigits; k++) {
432 uint128_t product;
433
434 product = mul_64_64(left[k], right);
435 r01 = add_128_128(r01, product);
436 /* no carry */
437 result[k] = r01.m_low;
438 r01.m_low = r01.m_high;
439 r01.m_high = 0;
440 }
441 result[k] = r01.m_low;
442 for (++k; k < ndigits * 2; k++)
443 result[k] = 0;
444}
445
446static void vli_square(u64 *result, const u64 *left, unsigned int ndigits)
447{
448 uint128_t r01 = { 0, 0 };
449 u64 r2 = 0;
450 int i, k;
451
452 for (k = 0; k < ndigits * 2 - 1; k++) {
453 unsigned int min;
454
455 if (k < ndigits)
456 min = 0;
457 else
458 min = (k + 1) - ndigits;
459
460 for (i = min; i <= k && i <= k - i; i++) {
461 uint128_t product;
462
463 product = mul_64_64(left[i], left[k - i]);
464
465 if (i < k - i) {
466 r2 += product.m_high >> 63;
467 product.m_high = (product.m_high << 1) |
468 (product.m_low >> 63);
469 product.m_low <<= 1;
470 }
471
472 r01 = add_128_128(r01, product);
473 r2 += (r01.m_high < product.m_high);
474 }
475
476 result[k] = r01.m_low;
477 r01.m_low = r01.m_high;
478 r01.m_high = r2;
479 r2 = 0;
480 }
481
482 result[ndigits * 2 - 1] = r01.m_low;
483}
484
485/* Computes result = (left + right) % mod.
486 * Assumes that left < mod and right < mod, result != mod.
487 */
488static void vli_mod_add(u64 *result, const u64 *left, const u64 *right,
489 const u64 *mod, unsigned int ndigits)
490{
491 u64 carry;
492
493 carry = vli_add(result, left, right, ndigits);
494
495 /* result > mod (result = mod + remainder), so subtract mod to
496 * get remainder.
497 */
498 if (carry || vli_cmp(result, mod, ndigits) >= 0)
499 vli_sub(result, result, mod, ndigits);
500}
501
502/* Computes result = (left - right) % mod.
503 * Assumes that left < mod and right < mod, result != mod.
504 */
505static void vli_mod_sub(u64 *result, const u64 *left, const u64 *right,
506 const u64 *mod, unsigned int ndigits)
507{
508 u64 borrow = vli_sub(result, left, right, ndigits);
509
510 /* In this case, p_result == -diff == (max int) - diff.
511 * Since -x % d == d - x, we can get the correct result from
512 * result + mod (with overflow).
513 */
514 if (borrow)
515 vli_add(result, result, mod, ndigits);
516}
517
518/*
519 * Computes result = product % mod
520 * for special form moduli: p = 2^k-c, for small c (note the minus sign)
521 *
522 * References:
523 * R. Crandall, C. Pomerance. Prime Numbers: A Computational Perspective.
524 * 9 Fast Algorithms for Large-Integer Arithmetic. 9.2.3 Moduli of special form
525 * Algorithm 9.2.13 (Fast mod operation for special-form moduli).
526 */
527static void vli_mmod_special(u64 *result, const u64 *product,
528 const u64 *mod, unsigned int ndigits)
529{
530 u64 c = -mod[0];
531 u64 t[ECC_MAX_DIGITS * 2];
532 u64 r[ECC_MAX_DIGITS * 2];
533
534 vli_set(r, product, ndigits * 2);
535 while (!vli_is_zero(r + ndigits, ndigits)) {
536 vli_umult(t, r + ndigits, c, ndigits);
537 vli_clear(r + ndigits, ndigits);
538 vli_add(r, r, t, ndigits * 2);
539 }
540 vli_set(t, mod, ndigits);
541 vli_clear(t + ndigits, ndigits);
542 while (vli_cmp(r, t, ndigits * 2) >= 0)
543 vli_sub(r, r, t, ndigits * 2);
544 vli_set(result, r, ndigits);
545}
546
547/*
548 * Computes result = product % mod
549 * for special form moduli: p = 2^{k-1}+c, for small c (note the plus sign)
550 * where k-1 does not fit into qword boundary by -1 bit (such as 255).
551
552 * References (loosely based on):
553 * A. Menezes, P. van Oorschot, S. Vanstone. Handbook of Applied Cryptography.
554 * 14.3.4 Reduction methods for moduli of special form. Algorithm 14.47.
555 * URL: http://cacr.uwaterloo.ca/hac/about/chap14.pdf
556 *
557 * H. Cohen, G. Frey, R. Avanzi, C. Doche, T. Lange, K. Nguyen, F. Vercauteren.
558 * Handbook of Elliptic and Hyperelliptic Curve Cryptography.
559 * Algorithm 10.25 Fast reduction for special form moduli
560 */
561static void vli_mmod_special2(u64 *result, const u64 *product,
562 const u64 *mod, unsigned int ndigits)
563{
564 u64 c2 = mod[0] * 2;
565 u64 q[ECC_MAX_DIGITS];
566 u64 r[ECC_MAX_DIGITS * 2];
567 u64 m[ECC_MAX_DIGITS * 2]; /* expanded mod */
568 int carry; /* last bit that doesn't fit into q */
569 int i;
570
571 vli_set(m, mod, ndigits);
572 vli_clear(m + ndigits, ndigits);
573
574 vli_set(r, product, ndigits);
575 /* q and carry are top bits */
576 vli_set(q, product + ndigits, ndigits);
577 vli_clear(r + ndigits, ndigits);
578 carry = vli_is_negative(r, ndigits);
579 if (carry)
580 r[ndigits - 1] &= (1ull << 63) - 1;
581 for (i = 1; carry || !vli_is_zero(q, ndigits); i++) {
582 u64 qc[ECC_MAX_DIGITS * 2];
583
584 vli_umult(qc, q, c2, ndigits);
585 if (carry)
586 vli_uadd(qc, qc, mod[0], ndigits * 2);
587 vli_set(q, qc + ndigits, ndigits);
588 vli_clear(qc + ndigits, ndigits);
589 carry = vli_is_negative(qc, ndigits);
590 if (carry)
591 qc[ndigits - 1] &= (1ull << 63) - 1;
592 if (i & 1)
593 vli_sub(r, r, qc, ndigits * 2);
594 else
595 vli_add(r, r, qc, ndigits * 2);
596 }
597 while (vli_is_negative(r, ndigits * 2))
598 vli_add(r, r, m, ndigits * 2);
599 while (vli_cmp(r, m, ndigits * 2) >= 0)
600 vli_sub(r, r, m, ndigits * 2);
601
602 vli_set(result, r, ndigits);
603}
604
605/*
606 * Computes result = product % mod, where product is 2N words long.
607 * Reference: Ken MacKay's micro-ecc.
608 * Currently only designed to work for curve_p or curve_n.
609 */
610static void vli_mmod_slow(u64 *result, u64 *product, const u64 *mod,
611 unsigned int ndigits)
612{
613 u64 mod_m[2 * ECC_MAX_DIGITS];
614 u64 tmp[2 * ECC_MAX_DIGITS];
615 u64 *v[2] = { tmp, product };
616 u64 carry = 0;
617 unsigned int i;
618 /* Shift mod so its highest set bit is at the maximum position. */
619 int shift = (ndigits * 2 * 64) - vli_num_bits(mod, ndigits);
620 int word_shift = shift / 64;
621 int bit_shift = shift % 64;
622
623 vli_clear(mod_m, word_shift);
624 if (bit_shift > 0) {
625 for (i = 0; i < ndigits; ++i) {
626 mod_m[word_shift + i] = (mod[i] << bit_shift) | carry;
627 carry = mod[i] >> (64 - bit_shift);
628 }
629 } else
630 vli_set(mod_m + word_shift, mod, ndigits);
631
632 for (i = 1; shift >= 0; --shift) {
633 u64 borrow = 0;
634 unsigned int j;
635
636 for (j = 0; j < ndigits * 2; ++j) {
637 u64 diff = v[i][j] - mod_m[j] - borrow;
638
639 if (diff != v[i][j])
640 borrow = (diff > v[i][j]);
641 v[1 - i][j] = diff;
642 }
643 i = !(i ^ borrow); /* Swap the index if there was no borrow */
644 vli_rshift1(mod_m, ndigits);
645 mod_m[ndigits - 1] |= mod_m[ndigits] << (64 - 1);
646 vli_rshift1(mod_m + ndigits, ndigits);
647 }
648 vli_set(result, v[i], ndigits);
649}
650
651/* Computes result = product % mod using Barrett's reduction with precomputed
652 * value mu appended to the mod after ndigits, mu = (2^{2w} / mod) and have
653 * length ndigits + 1, where mu * (2^w - 1) should not overflow ndigits
654 * boundary.
655 *
656 * Reference:
657 * R. Brent, P. Zimmermann. Modern Computer Arithmetic. 2010.
658 * 2.4.1 Barrett's algorithm. Algorithm 2.5.
659 */
660static void vli_mmod_barrett(u64 *result, u64 *product, const u64 *mod,
661 unsigned int ndigits)
662{
663 u64 q[ECC_MAX_DIGITS * 2];
664 u64 r[ECC_MAX_DIGITS * 2];
665 const u64 *mu = mod + ndigits;
666
667 vli_mult(q, product + ndigits, mu, ndigits);
668 if (mu[ndigits])
669 vli_add(q + ndigits, q + ndigits, product + ndigits, ndigits);
670 vli_mult(r, mod, q + ndigits, ndigits);
671 vli_sub(r, product, r, ndigits * 2);
672 while (!vli_is_zero(r + ndigits, ndigits) ||
673 vli_cmp(r, mod, ndigits) != -1) {
674 u64 carry;
675
676 carry = vli_sub(r, r, mod, ndigits);
677 vli_usub(r + ndigits, r + ndigits, carry, ndigits);
678 }
679 vli_set(result, r, ndigits);
680}
681
682/* Computes p_result = p_product % curve_p.
683 * See algorithm 5 and 6 from
684 * http://www.isys.uni-klu.ac.at/PDF/2001-0126-MT.pdf
685 */
686static void vli_mmod_fast_192(u64 *result, const u64 *product,
687 const u64 *curve_prime, u64 *tmp)
688{
689 const unsigned int ndigits = 3;
690 int carry;
691
692 vli_set(result, product, ndigits);
693
694 vli_set(tmp, &product[3], ndigits);
695 carry = vli_add(result, result, tmp, ndigits);
696
697 tmp[0] = 0;
698 tmp[1] = product[3];
699 tmp[2] = product[4];
700 carry += vli_add(result, result, tmp, ndigits);
701
702 tmp[0] = tmp[1] = product[5];
703 tmp[2] = 0;
704 carry += vli_add(result, result, tmp, ndigits);
705
706 while (carry || vli_cmp(curve_prime, result, ndigits) != 1)
707 carry -= vli_sub(result, result, curve_prime, ndigits);
708}
709
710/* Computes result = product % curve_prime
711 * from http://www.nsa.gov/ia/_files/nist-routines.pdf
712 */
713static void vli_mmod_fast_256(u64 *result, const u64 *product,
714 const u64 *curve_prime, u64 *tmp)
715{
716 int carry;
717 const unsigned int ndigits = 4;
718
719 /* t */
720 vli_set(result, product, ndigits);
721
722 /* s1 */
723 tmp[0] = 0;
724 tmp[1] = product[5] & 0xffffffff00000000ull;
725 tmp[2] = product[6];
726 tmp[3] = product[7];
727 carry = vli_lshift(tmp, tmp, 1, ndigits);
728 carry += vli_add(result, result, tmp, ndigits);
729
730 /* s2 */
731 tmp[1] = product[6] << 32;
732 tmp[2] = (product[6] >> 32) | (product[7] << 32);
733 tmp[3] = product[7] >> 32;
734 carry += vli_lshift(tmp, tmp, 1, ndigits);
735 carry += vli_add(result, result, tmp, ndigits);
736
737 /* s3 */
738 tmp[0] = product[4];
739 tmp[1] = product[5] & 0xffffffff;
740 tmp[2] = 0;
741 tmp[3] = product[7];
742 carry += vli_add(result, result, tmp, ndigits);
743
744 /* s4 */
745 tmp[0] = (product[4] >> 32) | (product[5] << 32);
746 tmp[1] = (product[5] >> 32) | (product[6] & 0xffffffff00000000ull);
747 tmp[2] = product[7];
748 tmp[3] = (product[6] >> 32) | (product[4] << 32);
749 carry += vli_add(result, result, tmp, ndigits);
750
751 /* d1 */
752 tmp[0] = (product[5] >> 32) | (product[6] << 32);
753 tmp[1] = (product[6] >> 32);
754 tmp[2] = 0;
755 tmp[3] = (product[4] & 0xffffffff) | (product[5] << 32);
756 carry -= vli_sub(result, result, tmp, ndigits);
757
758 /* d2 */
759 tmp[0] = product[6];
760 tmp[1] = product[7];
761 tmp[2] = 0;
762 tmp[3] = (product[4] >> 32) | (product[5] & 0xffffffff00000000ull);
763 carry -= vli_sub(result, result, tmp, ndigits);
764
765 /* d3 */
766 tmp[0] = (product[6] >> 32) | (product[7] << 32);
767 tmp[1] = (product[7] >> 32) | (product[4] << 32);
768 tmp[2] = (product[4] >> 32) | (product[5] << 32);
769 tmp[3] = (product[6] << 32);
770 carry -= vli_sub(result, result, tmp, ndigits);
771
772 /* d4 */
773 tmp[0] = product[7];
774 tmp[1] = product[4] & 0xffffffff00000000ull;
775 tmp[2] = product[5];
776 tmp[3] = product[6] & 0xffffffff00000000ull;
777 carry -= vli_sub(result, result, tmp, ndigits);
778
779 if (carry < 0) {
780 do {
781 carry += vli_add(result, result, curve_prime, ndigits);
782 } while (carry < 0);
783 } else {
784 while (carry || vli_cmp(curve_prime, result, ndigits) != 1)
785 carry -= vli_sub(result, result, curve_prime, ndigits);
786 }
787}
788
789#define SL32OR32(x32, y32) (((u64)x32 << 32) | y32)
790#define AND64H(x64) (x64 & 0xffFFffFF00000000ull)
791#define AND64L(x64) (x64 & 0x00000000ffFFffFFull)
792
793/* Computes result = product % curve_prime
794 * from "Mathematical routines for the NIST prime elliptic curves"
795 */
796static void vli_mmod_fast_384(u64 *result, const u64 *product,
797 const u64 *curve_prime, u64 *tmp)
798{
799 int carry;
800 const unsigned int ndigits = 6;
801
802 /* t */
803 vli_set(result, product, ndigits);
804
805 /* s1 */
806 tmp[0] = 0; // 0 || 0
807 tmp[1] = 0; // 0 || 0
808 tmp[2] = SL32OR32(product[11], (product[10]>>32)); //a22||a21
809 tmp[3] = product[11]>>32; // 0 ||a23
810 tmp[4] = 0; // 0 || 0
811 tmp[5] = 0; // 0 || 0
812 carry = vli_lshift(tmp, tmp, 1, ndigits);
813 carry += vli_add(result, result, tmp, ndigits);
814
815 /* s2 */
816 tmp[0] = product[6]; //a13||a12
817 tmp[1] = product[7]; //a15||a14
818 tmp[2] = product[8]; //a17||a16
819 tmp[3] = product[9]; //a19||a18
820 tmp[4] = product[10]; //a21||a20
821 tmp[5] = product[11]; //a23||a22
822 carry += vli_add(result, result, tmp, ndigits);
823
824 /* s3 */
825 tmp[0] = SL32OR32(product[11], (product[10]>>32)); //a22||a21
826 tmp[1] = SL32OR32(product[6], (product[11]>>32)); //a12||a23
827 tmp[2] = SL32OR32(product[7], (product[6])>>32); //a14||a13
828 tmp[3] = SL32OR32(product[8], (product[7]>>32)); //a16||a15
829 tmp[4] = SL32OR32(product[9], (product[8]>>32)); //a18||a17
830 tmp[5] = SL32OR32(product[10], (product[9]>>32)); //a20||a19
831 carry += vli_add(result, result, tmp, ndigits);
832
833 /* s4 */
834 tmp[0] = AND64H(product[11]); //a23|| 0
835 tmp[1] = (product[10]<<32); //a20|| 0
836 tmp[2] = product[6]; //a13||a12
837 tmp[3] = product[7]; //a15||a14
838 tmp[4] = product[8]; //a17||a16
839 tmp[5] = product[9]; //a19||a18
840 carry += vli_add(result, result, tmp, ndigits);
841
842 /* s5 */
843 tmp[0] = 0; // 0|| 0
844 tmp[1] = 0; // 0|| 0
845 tmp[2] = product[10]; //a21||a20
846 tmp[3] = product[11]; //a23||a22
847 tmp[4] = 0; // 0|| 0
848 tmp[5] = 0; // 0|| 0
849 carry += vli_add(result, result, tmp, ndigits);
850
851 /* s6 */
852 tmp[0] = AND64L(product[10]); // 0 ||a20
853 tmp[1] = AND64H(product[10]); //a21|| 0
854 tmp[2] = product[11]; //a23||a22
855 tmp[3] = 0; // 0 || 0
856 tmp[4] = 0; // 0 || 0
857 tmp[5] = 0; // 0 || 0
858 carry += vli_add(result, result, tmp, ndigits);
859
860 /* d1 */
861 tmp[0] = SL32OR32(product[6], (product[11]>>32)); //a12||a23
862 tmp[1] = SL32OR32(product[7], (product[6]>>32)); //a14||a13
863 tmp[2] = SL32OR32(product[8], (product[7]>>32)); //a16||a15
864 tmp[3] = SL32OR32(product[9], (product[8]>>32)); //a18||a17
865 tmp[4] = SL32OR32(product[10], (product[9]>>32)); //a20||a19
866 tmp[5] = SL32OR32(product[11], (product[10]>>32)); //a22||a21
867 carry -= vli_sub(result, result, tmp, ndigits);
868
869 /* d2 */
870 tmp[0] = (product[10]<<32); //a20|| 0
871 tmp[1] = SL32OR32(product[11], (product[10]>>32)); //a22||a21
872 tmp[2] = (product[11]>>32); // 0 ||a23
873 tmp[3] = 0; // 0 || 0
874 tmp[4] = 0; // 0 || 0
875 tmp[5] = 0; // 0 || 0
876 carry -= vli_sub(result, result, tmp, ndigits);
877
878 /* d3 */
879 tmp[0] = 0; // 0 || 0
880 tmp[1] = AND64H(product[11]); //a23|| 0
881 tmp[2] = product[11]>>32; // 0 ||a23
882 tmp[3] = 0; // 0 || 0
883 tmp[4] = 0; // 0 || 0
884 tmp[5] = 0; // 0 || 0
885 carry -= vli_sub(result, result, tmp, ndigits);
886
887 if (carry < 0) {
888 do {
889 carry += vli_add(result, result, curve_prime, ndigits);
890 } while (carry < 0);
891 } else {
892 while (carry || vli_cmp(curve_prime, result, ndigits) != 1)
893 carry -= vli_sub(result, result, curve_prime, ndigits);
894 }
895
896}
897
898#undef SL32OR32
899#undef AND64H
900#undef AND64L
901
902/* Computes result = product % curve_prime for different curve_primes.
903 *
904 * Note that curve_primes are distinguished just by heuristic check and
905 * not by complete conformance check.
906 */
907static bool vli_mmod_fast(u64 *result, u64 *product,
908 const struct ecc_curve *curve)
909{
910 u64 tmp[2 * ECC_MAX_DIGITS];
911 const u64 *curve_prime = curve->p;
912 const unsigned int ndigits = curve->g.ndigits;
913
914 /* All NIST curves have name prefix 'nist_' */
915 if (strncmp(curve->name, "nist_", 5) != 0) {
916 /* Try to handle Pseudo-Marsenne primes. */
917 if (curve_prime[ndigits - 1] == -1ull) {
918 vli_mmod_special(result, product, curve_prime,
919 ndigits);
920 return true;
921 } else if (curve_prime[ndigits - 1] == 1ull << 63 &&
922 curve_prime[ndigits - 2] == 0) {
923 vli_mmod_special2(result, product, curve_prime,
924 ndigits);
925 return true;
926 }
927 vli_mmod_barrett(result, product, curve_prime, ndigits);
928 return true;
929 }
930
931 switch (ndigits) {
932 case 3:
933 vli_mmod_fast_192(result, product, curve_prime, tmp);
934 break;
935 case 4:
936 vli_mmod_fast_256(result, product, curve_prime, tmp);
937 break;
938 case 6:
939 vli_mmod_fast_384(result, product, curve_prime, tmp);
940 break;
941 default:
942 pr_err_ratelimited("ecc: unsupported digits size!\n");
943 return false;
944 }
945
946 return true;
947}
948
949/* Computes result = (left * right) % mod.
950 * Assumes that mod is big enough curve order.
951 */
952void vli_mod_mult_slow(u64 *result, const u64 *left, const u64 *right,
953 const u64 *mod, unsigned int ndigits)
954{
955 u64 product[ECC_MAX_DIGITS * 2];
956
957 vli_mult(product, left, right, ndigits);
958 vli_mmod_slow(result, product, mod, ndigits);
959}
960EXPORT_SYMBOL(vli_mod_mult_slow);
961
962/* Computes result = (left * right) % curve_prime. */
963static void vli_mod_mult_fast(u64 *result, const u64 *left, const u64 *right,
964 const struct ecc_curve *curve)
965{
966 u64 product[2 * ECC_MAX_DIGITS];
967
968 vli_mult(product, left, right, curve->g.ndigits);
969 vli_mmod_fast(result, product, curve);
970}
971
972/* Computes result = left^2 % curve_prime. */
973static void vli_mod_square_fast(u64 *result, const u64 *left,
974 const struct ecc_curve *curve)
975{
976 u64 product[2 * ECC_MAX_DIGITS];
977
978 vli_square(product, left, curve->g.ndigits);
979 vli_mmod_fast(result, product, curve);
980}
981
982#define EVEN(vli) (!(vli[0] & 1))
983/* Computes result = (1 / p_input) % mod. All VLIs are the same size.
984 * See "From Euclid's GCD to Montgomery Multiplication to the Great Divide"
985 * https://labs.oracle.com/techrep/2001/smli_tr-2001-95.pdf
986 */
987void vli_mod_inv(u64 *result, const u64 *input, const u64 *mod,
988 unsigned int ndigits)
989{
990 u64 a[ECC_MAX_DIGITS], b[ECC_MAX_DIGITS];
991 u64 u[ECC_MAX_DIGITS], v[ECC_MAX_DIGITS];
992 u64 carry;
993 int cmp_result;
994
995 if (vli_is_zero(input, ndigits)) {
996 vli_clear(result, ndigits);
997 return;
998 }
999
1000 vli_set(a, input, ndigits);
1001 vli_set(b, mod, ndigits);
1002 vli_clear(u, ndigits);
1003 u[0] = 1;
1004 vli_clear(v, ndigits);
1005
1006 while ((cmp_result = vli_cmp(a, b, ndigits)) != 0) {
1007 carry = 0;
1008
1009 if (EVEN(a)) {
1010 vli_rshift1(a, ndigits);
1011
1012 if (!EVEN(u))
1013 carry = vli_add(u, u, mod, ndigits);
1014
1015 vli_rshift1(u, ndigits);
1016 if (carry)
1017 u[ndigits - 1] |= 0x8000000000000000ull;
1018 } else if (EVEN(b)) {
1019 vli_rshift1(b, ndigits);
1020
1021 if (!EVEN(v))
1022 carry = vli_add(v, v, mod, ndigits);
1023
1024 vli_rshift1(v, ndigits);
1025 if (carry)
1026 v[ndigits - 1] |= 0x8000000000000000ull;
1027 } else if (cmp_result > 0) {
1028 vli_sub(a, a, b, ndigits);
1029 vli_rshift1(a, ndigits);
1030
1031 if (vli_cmp(u, v, ndigits) < 0)
1032 vli_add(u, u, mod, ndigits);
1033
1034 vli_sub(u, u, v, ndigits);
1035 if (!EVEN(u))
1036 carry = vli_add(u, u, mod, ndigits);
1037
1038 vli_rshift1(u, ndigits);
1039 if (carry)
1040 u[ndigits - 1] |= 0x8000000000000000ull;
1041 } else {
1042 vli_sub(b, b, a, ndigits);
1043 vli_rshift1(b, ndigits);
1044
1045 if (vli_cmp(v, u, ndigits) < 0)
1046 vli_add(v, v, mod, ndigits);
1047
1048 vli_sub(v, v, u, ndigits);
1049 if (!EVEN(v))
1050 carry = vli_add(v, v, mod, ndigits);
1051
1052 vli_rshift1(v, ndigits);
1053 if (carry)
1054 v[ndigits - 1] |= 0x8000000000000000ull;
1055 }
1056 }
1057
1058 vli_set(result, u, ndigits);
1059}
1060EXPORT_SYMBOL(vli_mod_inv);
1061
1062/* ------ Point operations ------ */
1063
1064/* Returns true if p_point is the point at infinity, false otherwise. */
1065static bool ecc_point_is_zero(const struct ecc_point *point)
1066{
1067 return (vli_is_zero(point->x, point->ndigits) &&
1068 vli_is_zero(point->y, point->ndigits));
1069}
1070
1071/* Point multiplication algorithm using Montgomery's ladder with co-Z
1072 * coordinates. From https://eprint.iacr.org/2011/338.pdf
1073 */
1074
1075/* Double in place */
1076static void ecc_point_double_jacobian(u64 *x1, u64 *y1, u64 *z1,
1077 const struct ecc_curve *curve)
1078{
1079 /* t1 = x, t2 = y, t3 = z */
1080 u64 t4[ECC_MAX_DIGITS];
1081 u64 t5[ECC_MAX_DIGITS];
1082 const u64 *curve_prime = curve->p;
1083 const unsigned int ndigits = curve->g.ndigits;
1084
1085 if (vli_is_zero(z1, ndigits))
1086 return;
1087
1088 /* t4 = y1^2 */
1089 vli_mod_square_fast(t4, y1, curve);
1090 /* t5 = x1*y1^2 = A */
1091 vli_mod_mult_fast(t5, x1, t4, curve);
1092 /* t4 = y1^4 */
1093 vli_mod_square_fast(t4, t4, curve);
1094 /* t2 = y1*z1 = z3 */
1095 vli_mod_mult_fast(y1, y1, z1, curve);
1096 /* t3 = z1^2 */
1097 vli_mod_square_fast(z1, z1, curve);
1098
1099 /* t1 = x1 + z1^2 */
1100 vli_mod_add(x1, x1, z1, curve_prime, ndigits);
1101 /* t3 = 2*z1^2 */
1102 vli_mod_add(z1, z1, z1, curve_prime, ndigits);
1103 /* t3 = x1 - z1^2 */
1104 vli_mod_sub(z1, x1, z1, curve_prime, ndigits);
1105 /* t1 = x1^2 - z1^4 */
1106 vli_mod_mult_fast(x1, x1, z1, curve);
1107
1108 /* t3 = 2*(x1^2 - z1^4) */
1109 vli_mod_add(z1, x1, x1, curve_prime, ndigits);
1110 /* t1 = 3*(x1^2 - z1^4) */
1111 vli_mod_add(x1, x1, z1, curve_prime, ndigits);
1112 if (vli_test_bit(x1, 0)) {
1113 u64 carry = vli_add(x1, x1, curve_prime, ndigits);
1114
1115 vli_rshift1(x1, ndigits);
1116 x1[ndigits - 1] |= carry << 63;
1117 } else {
1118 vli_rshift1(x1, ndigits);
1119 }
1120 /* t1 = 3/2*(x1^2 - z1^4) = B */
1121
1122 /* t3 = B^2 */
1123 vli_mod_square_fast(z1, x1, curve);
1124 /* t3 = B^2 - A */
1125 vli_mod_sub(z1, z1, t5, curve_prime, ndigits);
1126 /* t3 = B^2 - 2A = x3 */
1127 vli_mod_sub(z1, z1, t5, curve_prime, ndigits);
1128 /* t5 = A - x3 */
1129 vli_mod_sub(t5, t5, z1, curve_prime, ndigits);
1130 /* t1 = B * (A - x3) */
1131 vli_mod_mult_fast(x1, x1, t5, curve);
1132 /* t4 = B * (A - x3) - y1^4 = y3 */
1133 vli_mod_sub(t4, x1, t4, curve_prime, ndigits);
1134
1135 vli_set(x1, z1, ndigits);
1136 vli_set(z1, y1, ndigits);
1137 vli_set(y1, t4, ndigits);
1138}
1139
1140/* Modify (x1, y1) => (x1 * z^2, y1 * z^3) */
1141static void apply_z(u64 *x1, u64 *y1, u64 *z, const struct ecc_curve *curve)
1142{
1143 u64 t1[ECC_MAX_DIGITS];
1144
1145 vli_mod_square_fast(t1, z, curve); /* z^2 */
1146 vli_mod_mult_fast(x1, x1, t1, curve); /* x1 * z^2 */
1147 vli_mod_mult_fast(t1, t1, z, curve); /* z^3 */
1148 vli_mod_mult_fast(y1, y1, t1, curve); /* y1 * z^3 */
1149}
1150
1151/* P = (x1, y1) => 2P, (x2, y2) => P' */
1152static void xycz_initial_double(u64 *x1, u64 *y1, u64 *x2, u64 *y2,
1153 u64 *p_initial_z, const struct ecc_curve *curve)
1154{
1155 u64 z[ECC_MAX_DIGITS];
1156 const unsigned int ndigits = curve->g.ndigits;
1157
1158 vli_set(x2, x1, ndigits);
1159 vli_set(y2, y1, ndigits);
1160
1161 vli_clear(z, ndigits);
1162 z[0] = 1;
1163
1164 if (p_initial_z)
1165 vli_set(z, p_initial_z, ndigits);
1166
1167 apply_z(x1, y1, z, curve);
1168
1169 ecc_point_double_jacobian(x1, y1, z, curve);
1170
1171 apply_z(x2, y2, z, curve);
1172}
1173
1174/* Input P = (x1, y1, Z), Q = (x2, y2, Z)
1175 * Output P' = (x1', y1', Z3), P + Q = (x3, y3, Z3)
1176 * or P => P', Q => P + Q
1177 */
1178static void xycz_add(u64 *x1, u64 *y1, u64 *x2, u64 *y2,
1179 const struct ecc_curve *curve)
1180{
1181 /* t1 = X1, t2 = Y1, t3 = X2, t4 = Y2 */
1182 u64 t5[ECC_MAX_DIGITS];
1183 const u64 *curve_prime = curve->p;
1184 const unsigned int ndigits = curve->g.ndigits;
1185
1186 /* t5 = x2 - x1 */
1187 vli_mod_sub(t5, x2, x1, curve_prime, ndigits);
1188 /* t5 = (x2 - x1)^2 = A */
1189 vli_mod_square_fast(t5, t5, curve);
1190 /* t1 = x1*A = B */
1191 vli_mod_mult_fast(x1, x1, t5, curve);
1192 /* t3 = x2*A = C */
1193 vli_mod_mult_fast(x2, x2, t5, curve);
1194 /* t4 = y2 - y1 */
1195 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1196 /* t5 = (y2 - y1)^2 = D */
1197 vli_mod_square_fast(t5, y2, curve);
1198
1199 /* t5 = D - B */
1200 vli_mod_sub(t5, t5, x1, curve_prime, ndigits);
1201 /* t5 = D - B - C = x3 */
1202 vli_mod_sub(t5, t5, x2, curve_prime, ndigits);
1203 /* t3 = C - B */
1204 vli_mod_sub(x2, x2, x1, curve_prime, ndigits);
1205 /* t2 = y1*(C - B) */
1206 vli_mod_mult_fast(y1, y1, x2, curve);
1207 /* t3 = B - x3 */
1208 vli_mod_sub(x2, x1, t5, curve_prime, ndigits);
1209 /* t4 = (y2 - y1)*(B - x3) */
1210 vli_mod_mult_fast(y2, y2, x2, curve);
1211 /* t4 = y3 */
1212 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1213
1214 vli_set(x2, t5, ndigits);
1215}
1216
1217/* Input P = (x1, y1, Z), Q = (x2, y2, Z)
1218 * Output P + Q = (x3, y3, Z3), P - Q = (x3', y3', Z3)
1219 * or P => P - Q, Q => P + Q
1220 */
1221static void xycz_add_c(u64 *x1, u64 *y1, u64 *x2, u64 *y2,
1222 const struct ecc_curve *curve)
1223{
1224 /* t1 = X1, t2 = Y1, t3 = X2, t4 = Y2 */
1225 u64 t5[ECC_MAX_DIGITS];
1226 u64 t6[ECC_MAX_DIGITS];
1227 u64 t7[ECC_MAX_DIGITS];
1228 const u64 *curve_prime = curve->p;
1229 const unsigned int ndigits = curve->g.ndigits;
1230
1231 /* t5 = x2 - x1 */
1232 vli_mod_sub(t5, x2, x1, curve_prime, ndigits);
1233 /* t5 = (x2 - x1)^2 = A */
1234 vli_mod_square_fast(t5, t5, curve);
1235 /* t1 = x1*A = B */
1236 vli_mod_mult_fast(x1, x1, t5, curve);
1237 /* t3 = x2*A = C */
1238 vli_mod_mult_fast(x2, x2, t5, curve);
1239 /* t4 = y2 + y1 */
1240 vli_mod_add(t5, y2, y1, curve_prime, ndigits);
1241 /* t4 = y2 - y1 */
1242 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1243
1244 /* t6 = C - B */
1245 vli_mod_sub(t6, x2, x1, curve_prime, ndigits);
1246 /* t2 = y1 * (C - B) */
1247 vli_mod_mult_fast(y1, y1, t6, curve);
1248 /* t6 = B + C */
1249 vli_mod_add(t6, x1, x2, curve_prime, ndigits);
1250 /* t3 = (y2 - y1)^2 */
1251 vli_mod_square_fast(x2, y2, curve);
1252 /* t3 = x3 */
1253 vli_mod_sub(x2, x2, t6, curve_prime, ndigits);
1254
1255 /* t7 = B - x3 */
1256 vli_mod_sub(t7, x1, x2, curve_prime, ndigits);
1257 /* t4 = (y2 - y1)*(B - x3) */
1258 vli_mod_mult_fast(y2, y2, t7, curve);
1259 /* t4 = y3 */
1260 vli_mod_sub(y2, y2, y1, curve_prime, ndigits);
1261
1262 /* t7 = (y2 + y1)^2 = F */
1263 vli_mod_square_fast(t7, t5, curve);
1264 /* t7 = x3' */
1265 vli_mod_sub(t7, t7, t6, curve_prime, ndigits);
1266 /* t6 = x3' - B */
1267 vli_mod_sub(t6, t7, x1, curve_prime, ndigits);
1268 /* t6 = (y2 + y1)*(x3' - B) */
1269 vli_mod_mult_fast(t6, t6, t5, curve);
1270 /* t2 = y3' */
1271 vli_mod_sub(y1, t6, y1, curve_prime, ndigits);
1272
1273 vli_set(x1, t7, ndigits);
1274}
1275
1276static void ecc_point_mult(struct ecc_point *result,
1277 const struct ecc_point *point, const u64 *scalar,
1278 u64 *initial_z, const struct ecc_curve *curve,
1279 unsigned int ndigits)
1280{
1281 /* R0 and R1 */
1282 u64 rx[2][ECC_MAX_DIGITS];
1283 u64 ry[2][ECC_MAX_DIGITS];
1284 u64 z[ECC_MAX_DIGITS];
1285 u64 sk[2][ECC_MAX_DIGITS];
1286 u64 *curve_prime = curve->p;
1287 int i, nb;
1288 int num_bits;
1289 int carry;
1290
1291 carry = vli_add(sk[0], scalar, curve->n, ndigits);
1292 vli_add(sk[1], sk[0], curve->n, ndigits);
1293 scalar = sk[!carry];
1294 num_bits = sizeof(u64) * ndigits * 8 + 1;
1295
1296 vli_set(rx[1], point->x, ndigits);
1297 vli_set(ry[1], point->y, ndigits);
1298
1299 xycz_initial_double(rx[1], ry[1], rx[0], ry[0], initial_z, curve);
1300
1301 for (i = num_bits - 2; i > 0; i--) {
1302 nb = !vli_test_bit(scalar, i);
1303 xycz_add_c(rx[1 - nb], ry[1 - nb], rx[nb], ry[nb], curve);
1304 xycz_add(rx[nb], ry[nb], rx[1 - nb], ry[1 - nb], curve);
1305 }
1306
1307 nb = !vli_test_bit(scalar, 0);
1308 xycz_add_c(rx[1 - nb], ry[1 - nb], rx[nb], ry[nb], curve);
1309
1310 /* Find final 1/Z value. */
1311 /* X1 - X0 */
1312 vli_mod_sub(z, rx[1], rx[0], curve_prime, ndigits);
1313 /* Yb * (X1 - X0) */
1314 vli_mod_mult_fast(z, z, ry[1 - nb], curve);
1315 /* xP * Yb * (X1 - X0) */
1316 vli_mod_mult_fast(z, z, point->x, curve);
1317
1318 /* 1 / (xP * Yb * (X1 - X0)) */
1319 vli_mod_inv(z, z, curve_prime, point->ndigits);
1320
1321 /* yP / (xP * Yb * (X1 - X0)) */
1322 vli_mod_mult_fast(z, z, point->y, curve);
1323 /* Xb * yP / (xP * Yb * (X1 - X0)) */
1324 vli_mod_mult_fast(z, z, rx[1 - nb], curve);
1325 /* End 1/Z calculation */
1326
1327 xycz_add(rx[nb], ry[nb], rx[1 - nb], ry[1 - nb], curve);
1328
1329 apply_z(rx[0], ry[0], z, curve);
1330
1331 vli_set(result->x, rx[0], ndigits);
1332 vli_set(result->y, ry[0], ndigits);
1333}
1334
1335/* Computes R = P + Q mod p */
1336static void ecc_point_add(const struct ecc_point *result,
1337 const struct ecc_point *p, const struct ecc_point *q,
1338 const struct ecc_curve *curve)
1339{
1340 u64 z[ECC_MAX_DIGITS];
1341 u64 px[ECC_MAX_DIGITS];
1342 u64 py[ECC_MAX_DIGITS];
1343 unsigned int ndigits = curve->g.ndigits;
1344
1345 vli_set(result->x, q->x, ndigits);
1346 vli_set(result->y, q->y, ndigits);
1347 vli_mod_sub(z, result->x, p->x, curve->p, ndigits);
1348 vli_set(px, p->x, ndigits);
1349 vli_set(py, p->y, ndigits);
1350 xycz_add(px, py, result->x, result->y, curve);
1351 vli_mod_inv(z, z, curve->p, ndigits);
1352 apply_z(result->x, result->y, z, curve);
1353}
1354
1355/* Computes R = u1P + u2Q mod p using Shamir's trick.
1356 * Based on: Kenneth MacKay's micro-ecc (2014).
1357 */
1358void ecc_point_mult_shamir(const struct ecc_point *result,
1359 const u64 *u1, const struct ecc_point *p,
1360 const u64 *u2, const struct ecc_point *q,
1361 const struct ecc_curve *curve)
1362{
1363 u64 z[ECC_MAX_DIGITS];
1364 u64 sump[2][ECC_MAX_DIGITS];
1365 u64 *rx = result->x;
1366 u64 *ry = result->y;
1367 unsigned int ndigits = curve->g.ndigits;
1368 unsigned int num_bits;
1369 struct ecc_point sum = ECC_POINT_INIT(sump[0], sump[1], ndigits);
1370 const struct ecc_point *points[4];
1371 const struct ecc_point *point;
1372 unsigned int idx;
1373 int i;
1374
1375 ecc_point_add(&sum, p, q, curve);
1376 points[0] = NULL;
1377 points[1] = p;
1378 points[2] = q;
1379 points[3] = ∑
1380
1381 num_bits = max(vli_num_bits(u1, ndigits), vli_num_bits(u2, ndigits));
1382 i = num_bits - 1;
1383 idx = (!!vli_test_bit(u1, i)) | ((!!vli_test_bit(u2, i)) << 1);
1384 point = points[idx];
1385
1386 vli_set(rx, point->x, ndigits);
1387 vli_set(ry, point->y, ndigits);
1388 vli_clear(z + 1, ndigits - 1);
1389 z[0] = 1;
1390
1391 for (--i; i >= 0; i--) {
1392 ecc_point_double_jacobian(rx, ry, z, curve);
1393 idx = (!!vli_test_bit(u1, i)) | ((!!vli_test_bit(u2, i)) << 1);
1394 point = points[idx];
1395 if (point) {
1396 u64 tx[ECC_MAX_DIGITS];
1397 u64 ty[ECC_MAX_DIGITS];
1398 u64 tz[ECC_MAX_DIGITS];
1399
1400 vli_set(tx, point->x, ndigits);
1401 vli_set(ty, point->y, ndigits);
1402 apply_z(tx, ty, z, curve);
1403 vli_mod_sub(tz, rx, tx, curve->p, ndigits);
1404 xycz_add(tx, ty, rx, ry, curve);
1405 vli_mod_mult_fast(z, z, tz, curve);
1406 }
1407 }
1408 vli_mod_inv(z, z, curve->p, ndigits);
1409 apply_z(rx, ry, z, curve);
1410}
1411EXPORT_SYMBOL(ecc_point_mult_shamir);
1412
1413static int __ecc_is_key_valid(const struct ecc_curve *curve,
1414 const u64 *private_key, unsigned int ndigits)
1415{
1416 u64 one[ECC_MAX_DIGITS] = { 1, };
1417 u64 res[ECC_MAX_DIGITS];
1418
1419 if (!private_key)
1420 return -EINVAL;
1421
1422 if (curve->g.ndigits != ndigits)
1423 return -EINVAL;
1424
1425 /* Make sure the private key is in the range [2, n-3]. */
1426 if (vli_cmp(one, private_key, ndigits) != -1)
1427 return -EINVAL;
1428 vli_sub(res, curve->n, one, ndigits);
1429 vli_sub(res, res, one, ndigits);
1430 if (vli_cmp(res, private_key, ndigits) != 1)
1431 return -EINVAL;
1432
1433 return 0;
1434}
1435
1436int ecc_is_key_valid(unsigned int curve_id, unsigned int ndigits,
1437 const u64 *private_key, unsigned int private_key_len)
1438{
1439 int nbytes;
1440 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1441
1442 nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1443
1444 if (private_key_len != nbytes)
1445 return -EINVAL;
1446
1447 return __ecc_is_key_valid(curve, private_key, ndigits);
1448}
1449EXPORT_SYMBOL(ecc_is_key_valid);
1450
1451/*
1452 * ECC private keys are generated using the method of extra random bits,
1453 * equivalent to that described in FIPS 186-4, Appendix B.4.1.
1454 *
1455 * d = (c mod(n–1)) + 1 where c is a string of random bits, 64 bits longer
1456 * than requested
1457 * 0 <= c mod(n-1) <= n-2 and implies that
1458 * 1 <= d <= n-1
1459 *
1460 * This method generates a private key uniformly distributed in the range
1461 * [1, n-1].
1462 */
1463int ecc_gen_privkey(unsigned int curve_id, unsigned int ndigits, u64 *privkey)
1464{
1465 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1466 u64 priv[ECC_MAX_DIGITS];
1467 unsigned int nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1468 unsigned int nbits = vli_num_bits(curve->n, ndigits);
1469 int err;
1470
1471 /* Check that N is included in Table 1 of FIPS 186-4, section 6.1.1 */
1472 if (nbits < 160 || ndigits > ARRAY_SIZE(priv))
1473 return -EINVAL;
1474
1475 /*
1476 * FIPS 186-4 recommends that the private key should be obtained from a
1477 * RBG with a security strength equal to or greater than the security
1478 * strength associated with N.
1479 *
1480 * The maximum security strength identified by NIST SP800-57pt1r4 for
1481 * ECC is 256 (N >= 512).
1482 *
1483 * This condition is met by the default RNG because it selects a favored
1484 * DRBG with a security strength of 256.
1485 */
1486 if (crypto_get_default_rng())
1487 return -EFAULT;
1488
1489 err = crypto_rng_get_bytes(crypto_default_rng, (u8 *)priv, nbytes);
1490 crypto_put_default_rng();
1491 if (err)
1492 return err;
1493
1494 /* Make sure the private key is in the valid range. */
1495 if (__ecc_is_key_valid(curve, priv, ndigits))
1496 return -EINVAL;
1497
1498 ecc_swap_digits(priv, privkey, ndigits);
1499
1500 return 0;
1501}
1502EXPORT_SYMBOL(ecc_gen_privkey);
1503
1504int ecc_make_pub_key(unsigned int curve_id, unsigned int ndigits,
1505 const u64 *private_key, u64 *public_key)
1506{
1507 int ret = 0;
1508 struct ecc_point *pk;
1509 u64 priv[ECC_MAX_DIGITS];
1510 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1511
1512 if (!private_key || !curve || ndigits > ARRAY_SIZE(priv)) {
1513 ret = -EINVAL;
1514 goto out;
1515 }
1516
1517 ecc_swap_digits(private_key, priv, ndigits);
1518
1519 pk = ecc_alloc_point(ndigits);
1520 if (!pk) {
1521 ret = -ENOMEM;
1522 goto out;
1523 }
1524
1525 ecc_point_mult(pk, &curve->g, priv, NULL, curve, ndigits);
1526
1527 /* SP800-56A rev 3 5.6.2.1.3 key check */
1528 if (ecc_is_pubkey_valid_full(curve, pk)) {
1529 ret = -EAGAIN;
1530 goto err_free_point;
1531 }
1532
1533 ecc_swap_digits(pk->x, public_key, ndigits);
1534 ecc_swap_digits(pk->y, &public_key[ndigits], ndigits);
1535
1536err_free_point:
1537 ecc_free_point(pk);
1538out:
1539 return ret;
1540}
1541EXPORT_SYMBOL(ecc_make_pub_key);
1542
1543/* SP800-56A section 5.6.2.3.4 partial verification: ephemeral keys only */
1544int ecc_is_pubkey_valid_partial(const struct ecc_curve *curve,
1545 struct ecc_point *pk)
1546{
1547 u64 yy[ECC_MAX_DIGITS], xxx[ECC_MAX_DIGITS], w[ECC_MAX_DIGITS];
1548
1549 if (WARN_ON(pk->ndigits != curve->g.ndigits))
1550 return -EINVAL;
1551
1552 /* Check 1: Verify key is not the zero point. */
1553 if (ecc_point_is_zero(pk))
1554 return -EINVAL;
1555
1556 /* Check 2: Verify key is in the range [1, p-1]. */
1557 if (vli_cmp(curve->p, pk->x, pk->ndigits) != 1)
1558 return -EINVAL;
1559 if (vli_cmp(curve->p, pk->y, pk->ndigits) != 1)
1560 return -EINVAL;
1561
1562 /* Check 3: Verify that y^2 == (x^3 + a·x + b) mod p */
1563 vli_mod_square_fast(yy, pk->y, curve); /* y^2 */
1564 vli_mod_square_fast(xxx, pk->x, curve); /* x^2 */
1565 vli_mod_mult_fast(xxx, xxx, pk->x, curve); /* x^3 */
1566 vli_mod_mult_fast(w, curve->a, pk->x, curve); /* a·x */
1567 vli_mod_add(w, w, curve->b, curve->p, pk->ndigits); /* a·x + b */
1568 vli_mod_add(w, w, xxx, curve->p, pk->ndigits); /* x^3 + a·x + b */
1569 if (vli_cmp(yy, w, pk->ndigits) != 0) /* Equation */
1570 return -EINVAL;
1571
1572 return 0;
1573}
1574EXPORT_SYMBOL(ecc_is_pubkey_valid_partial);
1575
1576/* SP800-56A section 5.6.2.3.3 full verification */
1577int ecc_is_pubkey_valid_full(const struct ecc_curve *curve,
1578 struct ecc_point *pk)
1579{
1580 struct ecc_point *nQ;
1581
1582 /* Checks 1 through 3 */
1583 int ret = ecc_is_pubkey_valid_partial(curve, pk);
1584
1585 if (ret)
1586 return ret;
1587
1588 /* Check 4: Verify that nQ is the zero point. */
1589 nQ = ecc_alloc_point(pk->ndigits);
1590 if (!nQ)
1591 return -ENOMEM;
1592
1593 ecc_point_mult(nQ, pk, curve->n, NULL, curve, pk->ndigits);
1594 if (!ecc_point_is_zero(nQ))
1595 ret = -EINVAL;
1596
1597 ecc_free_point(nQ);
1598
1599 return ret;
1600}
1601EXPORT_SYMBOL(ecc_is_pubkey_valid_full);
1602
1603int crypto_ecdh_shared_secret(unsigned int curve_id, unsigned int ndigits,
1604 const u64 *private_key, const u64 *public_key,
1605 u64 *secret)
1606{
1607 int ret = 0;
1608 struct ecc_point *product, *pk;
1609 u64 priv[ECC_MAX_DIGITS];
1610 u64 rand_z[ECC_MAX_DIGITS];
1611 unsigned int nbytes;
1612 const struct ecc_curve *curve = ecc_get_curve(curve_id);
1613
1614 if (!private_key || !public_key || !curve ||
1615 ndigits > ARRAY_SIZE(priv) || ndigits > ARRAY_SIZE(rand_z)) {
1616 ret = -EINVAL;
1617 goto out;
1618 }
1619
1620 nbytes = ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
1621
1622 get_random_bytes(rand_z, nbytes);
1623
1624 pk = ecc_alloc_point(ndigits);
1625 if (!pk) {
1626 ret = -ENOMEM;
1627 goto out;
1628 }
1629
1630 ecc_swap_digits(public_key, pk->x, ndigits);
1631 ecc_swap_digits(&public_key[ndigits], pk->y, ndigits);
1632 ret = ecc_is_pubkey_valid_partial(curve, pk);
1633 if (ret)
1634 goto err_alloc_product;
1635
1636 ecc_swap_digits(private_key, priv, ndigits);
1637
1638 product = ecc_alloc_point(ndigits);
1639 if (!product) {
1640 ret = -ENOMEM;
1641 goto err_alloc_product;
1642 }
1643
1644 ecc_point_mult(product, pk, priv, rand_z, curve, ndigits);
1645
1646 if (ecc_point_is_zero(product)) {
1647 ret = -EFAULT;
1648 goto err_validity;
1649 }
1650
1651 ecc_swap_digits(product->x, secret, ndigits);
1652
1653err_validity:
1654 memzero_explicit(priv, sizeof(priv));
1655 memzero_explicit(rand_z, sizeof(rand_z));
1656 ecc_free_point(product);
1657err_alloc_product:
1658 ecc_free_point(pk);
1659out:
1660 return ret;
1661}
1662EXPORT_SYMBOL(crypto_ecdh_shared_secret);
1663
1664MODULE_LICENSE("Dual BSD/GPL");