Tor 0.4.9.0-alpha-dev
curve25519-donna-c64.c
1/* Copyright 2008, Google Inc.
2 * All rights reserved.
3 *
4 * Code released into the public domain.
5 *
6 * curve25519-donna: Curve25519 elliptic curve, public key function
7 *
8 * http://code.google.com/p/curve25519-donna/
9 *
10 * Adam Langley <agl@imperialviolet.org>
11 *
12 * Derived from public domain C code by Daniel J. Bernstein <djb@cr.yp.to>
13 *
14 * More information about curve25519 can be found here
15 * http://cr.yp.to/ecdh.html
16 *
17 * djb's sample implementation of curve25519 is written in a special assembly
18 * language called qhasm and uses the floating point registers.
19 *
20 * This is, almost, a clean room reimplementation from the curve25519 paper. It
21 * uses many of the tricks described therein. Only the crecip function is taken
22 * from the sample implementation.
23 */
24
25#include "orconfig.h"
26
27#include <string.h>
28#include "lib/cc/torint.h"
29
30typedef uint8_t u8;
31typedef uint64_t limb;
32typedef limb felem[5];
33// This is a special gcc mode for 128-bit integers. It's implemented on 64-bit
34// platforms only as far as I know.
35typedef unsigned uint128_t __attribute__((mode(TI)));
36
37#undef force_inline
38#define force_inline __attribute__((always_inline))
39
40/* Sum two numbers: output += in */
41static inline void force_inline
42fsum(limb *output, const limb *in) {
43 output[0] += in[0];
44 output[1] += in[1];
45 output[2] += in[2];
46 output[3] += in[3];
47 output[4] += in[4];
48}
49
50/* Find the difference of two numbers: output = in - output
51 * (note the order of the arguments!)
52 *
53 * Assumes that out[i] < 2**52
54 * On return, out[i] < 2**55
55 */
56static inline void force_inline
57fdifference_backwards(felem out, const felem in) {
58 /* 152 is 19 << 3 */
59 static const limb two54m152 = (((limb)1) << 54) - 152;
60 static const limb two54m8 = (((limb)1) << 54) - 8;
61
62 out[0] = in[0] + two54m152 - out[0];
63 out[1] = in[1] + two54m8 - out[1];
64 out[2] = in[2] + two54m8 - out[2];
65 out[3] = in[3] + two54m8 - out[3];
66 out[4] = in[4] + two54m8 - out[4];
67}
68
69/* Multiply a number by a scalar: output = in * scalar */
70static inline void force_inline
71fscalar_product(felem output, const felem in, const limb scalar) {
72 uint128_t a;
73
74 a = ((uint128_t) in[0]) * scalar;
75 output[0] = ((limb)a) & 0x7ffffffffffff;
76
77 a = ((uint128_t) in[1]) * scalar + ((limb) (a >> 51));
78 output[1] = ((limb)a) & 0x7ffffffffffff;
79
80 a = ((uint128_t) in[2]) * scalar + ((limb) (a >> 51));
81 output[2] = ((limb)a) & 0x7ffffffffffff;
82
83 a = ((uint128_t) in[3]) * scalar + ((limb) (a >> 51));
84 output[3] = ((limb)a) & 0x7ffffffffffff;
85
86 a = ((uint128_t) in[4]) * scalar + ((limb) (a >> 51));
87 output[4] = ((limb)a) & 0x7ffffffffffff;
88
89 output[0] += (a >> 51) * 19;
90}
91
92/* Multiply two numbers: output = in2 * in
93 *
94 * output must be distinct to both inputs. The inputs are reduced coefficient
95 * form, the output is not.
96 *
97 * Assumes that in[i] < 2**55 and likewise for in2.
98 * On return, output[i] < 2**52
99 */
100static inline void force_inline
101fmul(felem output, const felem in2, const felem in) {
102 uint128_t t[5];
103 limb r0,r1,r2,r3,r4,s0,s1,s2,s3,s4,c;
104
105 r0 = in[0];
106 r1 = in[1];
107 r2 = in[2];
108 r3 = in[3];
109 r4 = in[4];
110
111 s0 = in2[0];
112 s1 = in2[1];
113 s2 = in2[2];
114 s3 = in2[3];
115 s4 = in2[4];
116
117 t[0] = ((uint128_t) r0) * s0;
118 t[1] = ((uint128_t) r0) * s1 + ((uint128_t) r1) * s0;
119 t[2] = ((uint128_t) r0) * s2 + ((uint128_t) r2) * s0 + ((uint128_t) r1) * s1;
120 t[3] = ((uint128_t) r0) * s3 + ((uint128_t) r3) * s0 + ((uint128_t) r1) * s2 + ((uint128_t) r2) * s1;
121 t[4] = ((uint128_t) r0) * s4 + ((uint128_t) r4) * s0 + ((uint128_t) r3) * s1 + ((uint128_t) r1) * s3 + ((uint128_t) r2) * s2;
122
123 r4 *= 19;
124 r1 *= 19;
125 r2 *= 19;
126 r3 *= 19;
127
128 t[0] += ((uint128_t) r4) * s1 + ((uint128_t) r1) * s4 + ((uint128_t) r2) * s3 + ((uint128_t) r3) * s2;
129 t[1] += ((uint128_t) r4) * s2 + ((uint128_t) r2) * s4 + ((uint128_t) r3) * s3;
130 t[2] += ((uint128_t) r4) * s3 + ((uint128_t) r3) * s4;
131 t[3] += ((uint128_t) r4) * s4;
132
133 r0 = (limb)t[0] & 0x7ffffffffffff; c = (limb)(t[0] >> 51);
134 t[1] += c; r1 = (limb)t[1] & 0x7ffffffffffff; c = (limb)(t[1] >> 51);
135 t[2] += c; r2 = (limb)t[2] & 0x7ffffffffffff; c = (limb)(t[2] >> 51);
136 t[3] += c; r3 = (limb)t[3] & 0x7ffffffffffff; c = (limb)(t[3] >> 51);
137 t[4] += c; r4 = (limb)t[4] & 0x7ffffffffffff; c = (limb)(t[4] >> 51);
138 r0 += c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffff;
139 r1 += c; c = r1 >> 51; r1 = r1 & 0x7ffffffffffff;
140 r2 += c;
141
142 output[0] = r0;
143 output[1] = r1;
144 output[2] = r2;
145 output[3] = r3;
146 output[4] = r4;
147}
148
149static inline void force_inline
150fsquare_times(felem output, const felem in, limb count) {
151 uint128_t t[5];
152 limb r0,r1,r2,r3,r4,c;
153 limb d0,d1,d2,d4,d419;
154
155 r0 = in[0];
156 r1 = in[1];
157 r2 = in[2];
158 r3 = in[3];
159 r4 = in[4];
160
161 do {
162 d0 = r0 * 2;
163 d1 = r1 * 2;
164 d2 = r2 * 2 * 19;
165 d419 = r4 * 19;
166 d4 = d419 * 2;
167
168 t[0] = ((uint128_t) r0) * r0 + ((uint128_t) d4) * r1 + (((uint128_t) d2) * (r3 ));
169 t[1] = ((uint128_t) d0) * r1 + ((uint128_t) d4) * r2 + (((uint128_t) r3) * (r3 * 19));
170 t[2] = ((uint128_t) d0) * r2 + ((uint128_t) r1) * r1 + (((uint128_t) d4) * (r3 ));
171 t[3] = ((uint128_t) d0) * r3 + ((uint128_t) d1) * r2 + (((uint128_t) r4) * (d419 ));
172 t[4] = ((uint128_t) d0) * r4 + ((uint128_t) d1) * r3 + (((uint128_t) r2) * (r2 ));
173
174 r0 = (limb)t[0] & 0x7ffffffffffff; c = (limb)(t[0] >> 51);
175 t[1] += c; r1 = (limb)t[1] & 0x7ffffffffffff; c = (limb)(t[1] >> 51);
176 t[2] += c; r2 = (limb)t[2] & 0x7ffffffffffff; c = (limb)(t[2] >> 51);
177 t[3] += c; r3 = (limb)t[3] & 0x7ffffffffffff; c = (limb)(t[3] >> 51);
178 t[4] += c; r4 = (limb)t[4] & 0x7ffffffffffff; c = (limb)(t[4] >> 51);
179 r0 += c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffff;
180 r1 += c; c = r1 >> 51; r1 = r1 & 0x7ffffffffffff;
181 r2 += c;
182 } while(--count);
183
184 output[0] = r0;
185 output[1] = r1;
186 output[2] = r2;
187 output[3] = r3;
188 output[4] = r4;
189}
190
191/* Load a little-endian 64-bit number */
192static limb
193load_limb(const u8 *in) {
194 return
195 ((limb)in[0]) |
196 (((limb)in[1]) << 8) |
197 (((limb)in[2]) << 16) |
198 (((limb)in[3]) << 24) |
199 (((limb)in[4]) << 32) |
200 (((limb)in[5]) << 40) |
201 (((limb)in[6]) << 48) |
202 (((limb)in[7]) << 56);
203}
204
205static void
206store_limb(u8 *out, limb in) {
207 out[0] = in & 0xff;
208 out[1] = (in >> 8) & 0xff;
209 out[2] = (in >> 16) & 0xff;
210 out[3] = (in >> 24) & 0xff;
211 out[4] = (in >> 32) & 0xff;
212 out[5] = (in >> 40) & 0xff;
213 out[6] = (in >> 48) & 0xff;
214 out[7] = (in >> 56) & 0xff;
215}
216
217/* Take a little-endian, 32-byte number and expand it into polynomial form */
218static void
219fexpand(limb *output, const u8 *in) {
220 output[0] = load_limb(in) & 0x7ffffffffffff;
221 output[1] = (load_limb(in+6) >> 3) & 0x7ffffffffffff;
222 output[2] = (load_limb(in+12) >> 6) & 0x7ffffffffffff;
223 output[3] = (load_limb(in+19) >> 1) & 0x7ffffffffffff;
224 output[4] = (load_limb(in+24) >> 12) & 0x7ffffffffffff;
225}
226
227/* Take a fully reduced polynomial form number and contract it into a
228 * little-endian, 32-byte array
229 */
230static void
231fcontract(u8 *output, const felem input) {
232 uint128_t t[5];
233
234 t[0] = input[0];
235 t[1] = input[1];
236 t[2] = input[2];
237 t[3] = input[3];
238 t[4] = input[4];
239
240 t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
241 t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
242 t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
243 t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
244 t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
245
246 t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
247 t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
248 t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
249 t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
250 t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
251
252 /* now t is between 0 and 2^255-1, properly carried. */
253 /* case 1: between 0 and 2^255-20. case 2: between 2^255-19 and 2^255-1. */
254
255 t[0] += 19;
256
257 t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
258 t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
259 t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
260 t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
261 t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffff;
262
263 /* now between 19 and 2^255-1 in both cases, and offset by 19. */
264
265 t[0] += 0x8000000000000 - 19;
266 t[1] += 0x8000000000000 - 1;
267 t[2] += 0x8000000000000 - 1;
268 t[3] += 0x8000000000000 - 1;
269 t[4] += 0x8000000000000 - 1;
270
271 /* now between 2^255 and 2^256-20, and offset by 2^255. */
272
273 t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffff;
274 t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffff;
275 t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffff;
276 t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffff;
277 t[4] &= 0x7ffffffffffff;
278
279 store_limb(output, t[0] | (t[1] << 51));
280 store_limb(output+8, (t[1] >> 13) | (t[2] << 38));
281 store_limb(output+16, (t[2] >> 26) | (t[3] << 25));
282 store_limb(output+24, (t[3] >> 39) | (t[4] << 12));
283}
284
285/* Input: Q, Q', Q-Q'
286 * Output: 2Q, Q+Q'
287 *
288 * x2 z3: long form
289 * x3 z3: long form
290 * x z: short form, destroyed
291 * xprime zprime: short form, destroyed
292 * qmqp: short form, preserved
293 */
294static void
295fmonty(limb *x2, limb *z2, /* output 2Q */
296 limb *x3, limb *z3, /* output Q + Q' */
297 limb *x, limb *z, /* input Q */
298 limb *xprime, limb *zprime, /* input Q' */
299 const limb *qmqp /* input Q - Q' */) {
300 limb origx[5], origxprime[5], zzz[5], xx[5], zz[5], xxprime[5],
301 zzprime[5], zzzprime[5];
302
303 memcpy(origx, x, 5 * sizeof(limb));
304 fsum(x, z);
305 fdifference_backwards(z, origx); // does x - z
306
307 memcpy(origxprime, xprime, sizeof(limb) * 5);
308 fsum(xprime, zprime);
309 fdifference_backwards(zprime, origxprime);
310 fmul(xxprime, xprime, z);
311 fmul(zzprime, x, zprime);
312 memcpy(origxprime, xxprime, sizeof(limb) * 5);
313 fsum(xxprime, zzprime);
314 fdifference_backwards(zzprime, origxprime);
315 fsquare_times(x3, xxprime, 1);
316 fsquare_times(zzzprime, zzprime, 1);
317 fmul(z3, zzzprime, qmqp);
318
319 fsquare_times(xx, x, 1);
320 fsquare_times(zz, z, 1);
321 fmul(x2, xx, zz);
322 fdifference_backwards(zz, xx); // does zz = xx - zz
323 fscalar_product(zzz, zz, 121665);
324 fsum(zzz, xx);
325 fmul(z2, zz, zzz);
326}
327
328// -----------------------------------------------------------------------------
329// Maybe swap the contents of two limb arrays (@a and @b), each @len elements
330// long. Perform the swap iff @swap is non-zero.
331//
332// This function performs the swap without leaking any side-channel
333// information.
334// -----------------------------------------------------------------------------
335static void
336swap_conditional(limb a[5], limb b[5], limb iswap) {
337 unsigned i;
338 const limb swap = -iswap;
339
340 for (i = 0; i < 5; ++i) {
341 const limb x = swap & (a[i] ^ b[i]);
342 a[i] ^= x;
343 b[i] ^= x;
344 }
345}
346
347/* Calculates nQ where Q is the x-coordinate of a point on the curve
348 *
349 * resultx/resultz: the x coordinate of the resulting curve point (short form)
350 * n: a little endian, 32-byte number
351 * q: a point of the curve (short form)
352 */
353static void
354cmult(limb *resultx, limb *resultz, const u8 *n, const limb *q) {
355 limb a[5] = {0}, b[5] = {1}, c[5] = {1}, d[5] = {0};
356 limb *nqpqx = a, *nqpqz = b, *nqx = c, *nqz = d, *t;
357 limb e[5] = {0}, f[5] = {1}, g[5] = {0}, h[5] = {1};
358 limb *nqpqx2 = e, *nqpqz2 = f, *nqx2 = g, *nqz2 = h;
359
360 unsigned i, j;
361
362 memcpy(nqpqx, q, sizeof(limb) * 5);
363
364 for (i = 0; i < 32; ++i) {
365 u8 byte = n[31 - i];
366 for (j = 0; j < 8; ++j) {
367 const limb bit = byte >> 7;
368
369 swap_conditional(nqx, nqpqx, bit);
370 swap_conditional(nqz, nqpqz, bit);
371 fmonty(nqx2, nqz2,
372 nqpqx2, nqpqz2,
373 nqx, nqz,
374 nqpqx, nqpqz,
375 q);
376 swap_conditional(nqx2, nqpqx2, bit);
377 swap_conditional(nqz2, nqpqz2, bit);
378
379 t = nqx;
380 nqx = nqx2;
381 nqx2 = t;
382 t = nqz;
383 nqz = nqz2;
384 nqz2 = t;
385 t = nqpqx;
386 nqpqx = nqpqx2;
387 nqpqx2 = t;
388 t = nqpqz;
389 nqpqz = nqpqz2;
390 nqpqz2 = t;
391
392 byte <<= 1;
393 }
394 }
395
396 memcpy(resultx, nqx, sizeof(limb) * 5);
397 memcpy(resultz, nqz, sizeof(limb) * 5);
398}
399
400
401// -----------------------------------------------------------------------------
402// Shamelessly copied from djb's code, tightened a little
403// -----------------------------------------------------------------------------
404static void
405crecip(felem out, const felem z) {
406 felem a,t0,b,c;
407
408 /* 2 */ fsquare_times(a, z, 1); // a = 2
409 /* 8 */ fsquare_times(t0, a, 2);
410 /* 9 */ fmul(b, t0, z); // b = 9
411 /* 11 */ fmul(a, b, a); // a = 11
412 /* 22 */ fsquare_times(t0, a, 1);
413 /* 2^5 - 2^0 = 31 */ fmul(b, t0, b);
414 /* 2^10 - 2^5 */ fsquare_times(t0, b, 5);
415 /* 2^10 - 2^0 */ fmul(b, t0, b);
416 /* 2^20 - 2^10 */ fsquare_times(t0, b, 10);
417 /* 2^20 - 2^0 */ fmul(c, t0, b);
418 /* 2^40 - 2^20 */ fsquare_times(t0, c, 20);
419 /* 2^40 - 2^0 */ fmul(t0, t0, c);
420 /* 2^50 - 2^10 */ fsquare_times(t0, t0, 10);
421 /* 2^50 - 2^0 */ fmul(b, t0, b);
422 /* 2^100 - 2^50 */ fsquare_times(t0, b, 50);
423 /* 2^100 - 2^0 */ fmul(c, t0, b);
424 /* 2^200 - 2^100 */ fsquare_times(t0, c, 100);
425 /* 2^200 - 2^0 */ fmul(t0, t0, c);
426 /* 2^250 - 2^50 */ fsquare_times(t0, t0, 50);
427 /* 2^250 - 2^0 */ fmul(t0, t0, b);
428 /* 2^255 - 2^5 */ fsquare_times(t0, t0, 5);
429 /* 2^255 - 21 */ fmul(out, t0, a);
430}
431
432int curve25519_donna(u8 *, const u8 *, const u8 *);
433
434int
435curve25519_donna(u8 *mypublic, const u8 *secret, const u8 *basepoint) {
436 limb bp[5], x[5], z[5], zmone[5];
437 uint8_t e[32];
438 int i;
439
440 for (i = 0;i < 32;++i) e[i] = secret[i];
441 e[0] &= 248;
442 e[31] &= 127;
443 e[31] |= 64;
444
445 fexpand(bp, basepoint);
446 cmult(x, z, e, bp);
447 crecip(zmone, z);
448 fmul(z, x, zmone);
449 fcontract(mypublic, z);
450 return 0;
451}
Integer definitions used throughout Tor.