2 * Copyright 2015-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the OpenSSL license (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
12 #include <openssl/crypto.h>
14 #include "internal/poly1305.h"
16 typedef void (*poly1305_blocks_f) (void *ctx, const unsigned char *inp,
17 size_t len, unsigned int padbit);
18 typedef void (*poly1305_emit_f) (void *ctx, unsigned char mac[16],
19 const unsigned int nonce[4]);
21 struct poly1305_context {
22 double opaque[24]; /* large enough to hold internal state, declared
23 * 'double' to ensure at least 64-bit invariant
24 * alignment across all platforms and
26 unsigned int nonce[4];
27 unsigned char data[POLY1305_BLOCK_SIZE];
30 poly1305_blocks_f blocks;
35 size_t Poly1305_ctx_size ()
37 return sizeof(struct poly1305_context);
40 /* pick 32-bit unsigned integer in little endian order */
41 static unsigned int U8TOU32(const unsigned char *p)
43 return (((unsigned int)(p[0] & 0xff)) |
44 ((unsigned int)(p[1] & 0xff) << 8) |
45 ((unsigned int)(p[2] & 0xff) << 16) |
46 ((unsigned int)(p[3] & 0xff) << 24));
50 * Implementations can be classified by amount of significant bits in
51 * words making up the multi-precision value, or in other words radix
52 * or base of numerical representation, e.g. base 2^64, base 2^32,
53 * base 2^26. Complementary characteristic is how wide is the result of
54 * multiplication of pair of digits, e.g. it would take 128 bits to
55 * accommodate multiplication result in base 2^64 case. These are used
56 * interchangeably. To describe implementation that is. But interface
57 * is designed to isolate this so that low-level primitives implemented
58 * in assembly can be self-contained/self-coherent.
62 * Even though there is __int128 reference implementation targeting
63 * 64-bit platforms provided below, it's not obvious that it's optimal
64 * choice for every one of them. Depending on instruction set overall
65 * amount of instructions can be comparable to one in __int64
66 * implementation. Amount of multiplication instructions would be lower,
67 * but not necessarily overall. And in out-of-order execution context,
68 * it is the latter that can be crucial...
70 * On related note. Poly1305 author, D. J. Bernstein, discusses and
71 * provides floating-point implementations of the algorithm in question.
72 * It made a lot of sense by the time of introduction, because most
73 * then-modern processors didn't have pipelined integer multiplier.
74 * [Not to mention that some had non-constant timing for integer
75 * multiplications.] Floating-point instructions on the other hand could
76 * be issued every cycle, which allowed to achieve better performance.
77 * Nowadays, with SIMD and/or out-or-order execution, shared or
78 * even emulated FPU, it's more complicated, and floating-point
79 * implementation is not necessarily optimal choice in every situation,
85 typedef unsigned int u32;
88 * poly1305_blocks processes a multiple of POLY1305_BLOCK_SIZE blocks
89 * of |inp| no longer than |len|. Behaviour for |len| not divisible by
90 * block size is unspecified in general case, even though in reference
91 * implementation the trailing chunk is simply ignored. Per algorithm
92 * specification, every input block, complete or last partial, is to be
93 * padded with a bit past most significant byte. The latter kind is then
94 * padded with zeros till block size. This last partial block padding
95 * is caller(*)'s responsibility, and because of this the last partial
96 * block is always processed with separate call with |len| set to
97 * POLY1305_BLOCK_SIZE and |padbit| to 0. In all other cases |padbit|
98 * should be set to 1 to perform implicit padding with 128th bit.
99 * poly1305_blocks does not actually check for this constraint though,
100 * it's caller(*)'s responsibility to comply.
102 * (*) In the context "caller" is not application code, but higher
103 * level Poly1305_* from this very module, so that quirks are
107 poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit);
110 * Type-agnostic "rip-off" from constant_time_locl.h
112 # define CONSTANT_TIME_CARRY(a,b) ( \
113 (a ^ ((a ^ b) | ((a - b) ^ b))) >> (sizeof(a) * 8 - 1) \
116 # if !defined(PEDANTIC) && \
117 (defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16) && \
118 (defined(__SIZEOF_LONG__) && __SIZEOF_LONG__==8)
120 typedef unsigned long u64;
121 typedef unsigned __int128 u128;
128 /* pick 32-bit unsigned integer in little endian order */
129 static u64 U8TOU64(const unsigned char *p)
131 return (((u64)(p[0] & 0xff)) |
132 ((u64)(p[1] & 0xff) << 8) |
133 ((u64)(p[2] & 0xff) << 16) |
134 ((u64)(p[3] & 0xff) << 24) |
135 ((u64)(p[4] & 0xff) << 32) |
136 ((u64)(p[5] & 0xff) << 40) |
137 ((u64)(p[6] & 0xff) << 48) |
138 ((u64)(p[7] & 0xff) << 56));
141 /* store a 32-bit unsigned integer in little endian */
142 static void U64TO8(unsigned char *p, u64 v)
144 p[0] = (unsigned char)((v) & 0xff);
145 p[1] = (unsigned char)((v >> 8) & 0xff);
146 p[2] = (unsigned char)((v >> 16) & 0xff);
147 p[3] = (unsigned char)((v >> 24) & 0xff);
148 p[4] = (unsigned char)((v >> 32) & 0xff);
149 p[5] = (unsigned char)((v >> 40) & 0xff);
150 p[6] = (unsigned char)((v >> 48) & 0xff);
151 p[7] = (unsigned char)((v >> 56) & 0xff);
154 static void poly1305_init(void *ctx, const unsigned char key[16])
156 poly1305_internal *st = (poly1305_internal *) ctx;
163 /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
164 st->r[0] = U8TOU64(&key[0]) & 0x0ffffffc0fffffff;
165 st->r[1] = U8TOU64(&key[8]) & 0x0ffffffc0ffffffc;
169 poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit)
171 poly1305_internal *st = (poly1305_internal *)ctx;
186 while (len >= POLY1305_BLOCK_SIZE) {
188 h0 = (u64)(d0 = (u128)h0 + U8TOU64(inp + 0));
189 h1 = (u64)(d1 = (u128)h1 + (d0 >> 64) + U8TOU64(inp + 8));
191 * padbit can be zero only when original len was
192 * POLY1306_BLOCK_SIZE, but we don't check
194 h2 += (u64)(d1 >> 64) + padbit;
196 /* h *= r "%" p, where "%" stands for "partial remainder" */
197 d0 = ((u128)h0 * r0) +
199 d1 = ((u128)h0 * r1) +
204 /* last reduction step: */
205 /* a) h2:h0 = h2<<128 + d1<<64 + d0 */
207 h1 = (u64)(d1 += d0 >> 64);
208 h2 += (u64)(d1 >> 64);
209 /* b) (h2:h0 += (h2:h0>>130) * 5) %= 2^130 */
210 c = (h2 >> 2) + (h2 & ~3UL);
213 h1 += (c = CONSTANT_TIME_CARRY(h0,c));
214 h2 += CONSTANT_TIME_CARRY(h1,c);
216 * Occasional overflows to 3rd bit of h2 are taken care of
217 * "naturally". If after this point we end up at the top of
218 * this loop, then the overflow bit will be accounted for
219 * in next iteration. If we end up in poly1305_emit, then
220 * comparison to modulus below will still count as "carry
221 * into 131st bit", so that properly reduced value will be
222 * picked in conditional move.
225 inp += POLY1305_BLOCK_SIZE;
226 len -= POLY1305_BLOCK_SIZE;
234 static void poly1305_emit(void *ctx, unsigned char mac[16],
237 poly1305_internal *st = (poly1305_internal *) ctx;
247 /* compare to modulus by computing h + -p */
248 g0 = (u64)(t = (u128)h0 + 5);
249 g1 = (u64)(t = (u128)h1 + (t >> 64));
250 g2 = h2 + (u64)(t >> 64);
252 /* if there was carry into 131st bit, h1:h0 = g1:g0 */
253 mask = 0 - (g2 >> 2);
257 h0 = (h0 & mask) | g0;
258 h1 = (h1 & mask) | g1;
260 /* mac = (h + nonce) % (2^128) */
261 h0 = (u64)(t = (u128)h0 + nonce[0] + ((u64)nonce[1]<<32));
262 h1 = (u64)(t = (u128)h1 + nonce[2] + ((u64)nonce[3]<<32) + (t >> 64));
270 # if defined(_WIN32) && !defined(__MINGW32__)
271 typedef unsigned __int64 u64;
272 # elif defined(__arch64__)
273 typedef unsigned long u64;
275 typedef unsigned long long u64;
283 /* store a 32-bit unsigned integer in little endian */
284 static void U32TO8(unsigned char *p, unsigned int v)
286 p[0] = (unsigned char)((v) & 0xff);
287 p[1] = (unsigned char)((v >> 8) & 0xff);
288 p[2] = (unsigned char)((v >> 16) & 0xff);
289 p[3] = (unsigned char)((v >> 24) & 0xff);
292 static void poly1305_init(void *ctx, const unsigned char key[16])
294 poly1305_internal *st = (poly1305_internal *) ctx;
303 /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
304 st->r[0] = U8TOU32(&key[0]) & 0x0fffffff;
305 st->r[1] = U8TOU32(&key[4]) & 0x0ffffffc;
306 st->r[2] = U8TOU32(&key[8]) & 0x0ffffffc;
307 st->r[3] = U8TOU32(&key[12]) & 0x0ffffffc;
311 poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit)
313 poly1305_internal *st = (poly1305_internal *)ctx;
316 u32 h0, h1, h2, h3, h4, c;
334 while (len >= POLY1305_BLOCK_SIZE) {
336 h0 = (u32)(d0 = (u64)h0 + U8TOU32(inp + 0));
337 h1 = (u32)(d1 = (u64)h1 + (d0 >> 32) + U8TOU32(inp + 4));
338 h2 = (u32)(d2 = (u64)h2 + (d1 >> 32) + U8TOU32(inp + 8));
339 h3 = (u32)(d3 = (u64)h3 + (d2 >> 32) + U8TOU32(inp + 12));
340 h4 += (u32)(d3 >> 32) + padbit;
342 /* h *= r "%" p, where "%" stands for "partial remainder" */
343 d0 = ((u64)h0 * r0) +
347 d1 = ((u64)h0 * r1) +
352 d2 = ((u64)h0 * r2) +
357 d3 = ((u64)h0 * r3) +
364 /* last reduction step: */
365 /* a) h4:h0 = h4<<128 + d3<<96 + d2<<64 + d1<<32 + d0 */
367 h1 = (u32)(d1 += d0 >> 32);
368 h2 = (u32)(d2 += d1 >> 32);
369 h3 = (u32)(d3 += d2 >> 32);
370 h4 += (u32)(d3 >> 32);
371 /* b) (h4:h0 += (h4:h0>>130) * 5) %= 2^130 */
372 c = (h4 >> 2) + (h4 & ~3U);
375 h1 += (c = CONSTANT_TIME_CARRY(h0,c));
376 h2 += (c = CONSTANT_TIME_CARRY(h1,c));
377 h3 += (c = CONSTANT_TIME_CARRY(h2,c));
378 h4 += CONSTANT_TIME_CARRY(h3,c);
380 * Occasional overflows to 3rd bit of h4 are taken care of
381 * "naturally". If after this point we end up at the top of
382 * this loop, then the overflow bit will be accounted for
383 * in next iteration. If we end up in poly1305_emit, then
384 * comparison to modulus below will still count as "carry
385 * into 131st bit", so that properly reduced value will be
386 * picked in conditional move.
389 inp += POLY1305_BLOCK_SIZE;
390 len -= POLY1305_BLOCK_SIZE;
400 static void poly1305_emit(void *ctx, unsigned char mac[16],
403 poly1305_internal *st = (poly1305_internal *) ctx;
404 u32 h0, h1, h2, h3, h4;
405 u32 g0, g1, g2, g3, g4;
415 /* compare to modulus by computing h + -p */
416 g0 = (u32)(t = (u64)h0 + 5);
417 g1 = (u32)(t = (u64)h1 + (t >> 32));
418 g2 = (u32)(t = (u64)h2 + (t >> 32));
419 g3 = (u32)(t = (u64)h3 + (t >> 32));
420 g4 = h4 + (u32)(t >> 32);
422 /* if there was carry into 131st bit, h3:h0 = g3:g0 */
423 mask = 0 - (g4 >> 2);
429 h0 = (h0 & mask) | g0;
430 h1 = (h1 & mask) | g1;
431 h2 = (h2 & mask) | g2;
432 h3 = (h3 & mask) | g3;
434 /* mac = (h + nonce) % (2^128) */
435 h0 = (u32)(t = (u64)h0 + nonce[0]);
436 h1 = (u32)(t = (u64)h1 + (t >> 32) + nonce[1]);
437 h2 = (u32)(t = (u64)h2 + (t >> 32) + nonce[2]);
438 h3 = (u32)(t = (u64)h3 + (t >> 32) + nonce[3]);
443 U32TO8(mac + 12, h3);
447 int poly1305_init(void *ctx, const unsigned char key[16], void *func);
448 void poly1305_blocks(void *ctx, const unsigned char *inp, size_t len,
449 unsigned int padbit);
450 void poly1305_emit(void *ctx, unsigned char mac[16],
451 const unsigned int nonce[4]);
454 void Poly1305_Init(POLY1305 *ctx, const unsigned char key[32])
456 ctx->nonce[0] = U8TOU32(&key[16]);
457 ctx->nonce[1] = U8TOU32(&key[20]);
458 ctx->nonce[2] = U8TOU32(&key[24]);
459 ctx->nonce[3] = U8TOU32(&key[28]);
462 poly1305_init(ctx->opaque, key);
465 * Unlike reference poly1305_init assembly counterpart is expected
466 * to return a value: non-zero if it initializes ctx->func, and zero
467 * otherwise. Latter is to simplify assembly in cases when there no
468 * multiple code paths to switch between.
470 if (!poly1305_init(ctx->opaque, key, &ctx->func)) {
471 ctx->func.blocks = poly1305_blocks;
472 ctx->func.emit = poly1305_emit;
482 * This "eclipses" poly1305_blocks and poly1305_emit, but it's
483 * conscious choice imposed by -Wshadow compiler warnings.
485 # define poly1305_blocks (*poly1305_blocks_p)
486 # define poly1305_emit (*poly1305_emit_p)
489 void Poly1305_Update(POLY1305 *ctx, const unsigned char *inp, size_t len)
493 * As documented, poly1305_blocks is never called with input
494 * longer than single block and padbit argument set to 0. This
495 * property is fluently used in assembly modules to optimize
496 * padbit handling on loop boundary.
498 poly1305_blocks_f poly1305_blocks_p = ctx->func.blocks;
502 if ((num = ctx->num)) {
503 rem = POLY1305_BLOCK_SIZE - num;
505 memcpy(ctx->data + num, inp, rem);
506 poly1305_blocks(ctx->opaque, ctx->data, POLY1305_BLOCK_SIZE, 1);
510 /* Still not enough data to process a block. */
511 memcpy(ctx->data + num, inp, len);
512 ctx->num = num + len;
517 rem = len % POLY1305_BLOCK_SIZE;
520 if (len >= POLY1305_BLOCK_SIZE) {
521 poly1305_blocks(ctx->opaque, inp, len, 1);
526 memcpy(ctx->data, inp, rem);
531 void Poly1305_Final(POLY1305 *ctx, unsigned char mac[16])
534 poly1305_blocks_f poly1305_blocks_p = ctx->func.blocks;
535 poly1305_emit_f poly1305_emit_p = ctx->func.emit;
539 if ((num = ctx->num)) {
540 ctx->data[num++] = 1; /* pad bit */
541 while (num < POLY1305_BLOCK_SIZE)
542 ctx->data[num++] = 0;
543 poly1305_blocks(ctx->opaque, ctx->data, POLY1305_BLOCK_SIZE, 0);
546 poly1305_emit(ctx->opaque, mac, ctx->nonce);
548 /* zero out the state */
549 OPENSSL_cleanse(ctx, sizeof(*ctx));
555 struct poly1305_test {
556 const char *inputhex;
561 static const struct poly1305_test poly1305_tests[] = {
566 "43727970746f6772617068696320466f72756d2052657365617263682047726f"
568 "85d6be7857556d337f4452fe42d506a8""0103808afb0db2fd4abff6af4149f51b",
569 "a8061dc1305136c6c22b8baf0c0127a9"
572 * test vectors from "The Poly1305-AES message-authentication code"
576 "851fc40c3467ac0be05cc20404f3f700""580b3b0f9447bb1e69d095b5928b6dbc",
577 "f4c633c3044fc145f84f335cb81953de"
581 "a0f3080000f46400d0c7e9076c834403""dd3fab2251f11ac759f0887129cc2ee7",
582 "dd3fab2251f11ac759f0887129cc2ee7"
585 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136",
586 "48443d0bb0d21109c89a100b5ce2c208""83149c69b561dd88298a1798b10716ef",
587 "0ee1c16bb73f0f4fd19881753c01cdbe"
590 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
591 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9",
592 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
593 "5154ad0d2cb26e01274fc51148491f1b"
596 * self-generated vectors exercise "significant" lengths, such that
597 * are handled by different code paths
600 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
601 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af",
602 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
603 "812059a5da198637cac7c4a631bee466"
606 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
607 "990c62e48b8018b2c3e4a0fa3134cb67",
608 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
609 "5b88d7f6228b11e2e28579a5c0c1f761"
612 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
613 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
614 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136",
615 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
616 "bbb613b2b6d753ba07395b916aaece15"
619 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
620 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
621 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
622 "663cea190ffb83d89593f3f476b6bc24",
623 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
624 "c794d7057d1778c4bbee0a39b3d97342"
627 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
628 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
629 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
630 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136",
631 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
632 "ffbcb9b371423152d7fca5ad042fbaa9"
635 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
636 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
637 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
638 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136"
639 "812059a5da198637cac7c4a631bee466",
640 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
641 "069ed6b8ef0f207b3e243bb1019fe632"
644 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
645 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
646 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
647 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136"
648 "812059a5da198637cac7c4a631bee4665b88d7f6228b11e2e28579a5c0c1f761",
649 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
650 "cca339d9a45fa2368c2c68b3a4179133"
653 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
654 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
655 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
656 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136"
657 "812059a5da198637cac7c4a631bee4665b88d7f6228b11e2e28579a5c0c1f761"
658 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
659 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
660 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
661 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136",
662 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
663 "53f6e828a2f0fe0ee815bf0bd5841a34"
666 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
667 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
668 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
669 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136"
670 "812059a5da198637cac7c4a631bee4665b88d7f6228b11e2e28579a5c0c1f761"
671 "ab0812724a7f1e342742cbed374d94d136c6b8795d45b3819830f2c04491faf0"
672 "990c62e48b8018b2c3e4a0fa3134cb67fa83e158c994d961c4cb21095c1bf9af"
673 "48443d0bb0d21109c89a100b5ce2c20883149c69b561dd88298a1798b10716ef"
674 "663cea190ffb83d89593f3f476b6bc24d7e679107ea26adb8caf6652d0656136"
675 "812059a5da198637cac7c4a631bee4665b88d7f6228b11e2e28579a5c0c1f761",
676 "12976a08c4426d0ce8a82407c4f48207""80f8c20aa71202d1e29179cbcb555a57",
677 "b846d44e9bbd53cedffbfbb6b7fa4933"
680 * 4th power of the key spills to 131th bit in SIMD key setup
683 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
684 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
685 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
686 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
687 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
688 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
689 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
690 "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff",
691 "ad628107e8351d0f2c231a05dc4a4106""00000000000000000000000000000000",
692 "07145a4c02fe5fa32036de68fabe9066"
696 * poly1305_ieee754.c failed this in final stage
698 "842364e156336c0998b933a6237726180d9e3fdcbde4cd5d17080fc3beb49614"
699 "d7122c037463ff104d73f19c12704628d417c4c54a3fe30d3c3d7714382d43b0"
700 "382a50a5dee54be844b076e8df88201a1cd43b90eb21643fa96f39b518aa8340"
701 "c942ff3c31baf7c9bdbf0f31ae3fa096bf8c63030609829fe72e179824890bc8"
702 "e08c315c1cce2a83144dbbff09f74e3efc770b54d0984a8f19b14719e6363564"
703 "1d6b1eedf63efbf080e1783d32445412114c20de0b837a0dfa33d6b82825fff4"
704 "4c9a70ea54ce47f07df698e6b03323b53079364a5fc3e9dd034392bdde86dccd"
705 "da94321c5e44060489336cb65bf3989c36f7282c2f5d2b882c171e74",
706 "95d5c005503e510d8cd0aa072c4a4d06""6eabc52d11653df47fbf63ab198bcc26",
707 "f248312e578d9d58f8b7bb4d19105431"
710 * AVX2 in poly1305-x86.pl failed this with 176+32 split
713 "248ac31085b6c2adaaa38259a0d7192c5c35d1bb4ef39ad94c38d1c82479e2dd"
714 "2159a077024b0589bc8a20101b506f0a1ad0bbab76e83a83f1b94be6beae74e8"
715 "74cab692c5963a75436b776121ec9f62399a3e66b2d22707dae81933b6277f3c"
716 "8516bcbe26dbbd86f373103d7cf4cad1888c952118fbfbd0d7b4bedc4ae4936a"
717 "ff91157e7aa47c54442ea78d6ac251d324a0fbe49d89cc3521b66d16e9c66a37"
718 "09894e4eb0a4eedc4ae19468e66b81f2"
719 "71351b1d921ea551047abcc6b87a901fde7db79fa1818c11336dbc07244a40eb",
720 "000102030405060708090a0b0c0d0e0f""00000000000000000000000000000000",
721 "bc939bc5281480fa99c6d68c258ec42f"
724 * test vectors from Google
728 "c8afaac331ee372cd6082de134943b17""4710130e9f6fea8d72293850a667d86c",
729 "4710130e9f6fea8d72293850a667d86c",
732 "48656c6c6f20776f726c6421",
733 "746869732069732033322d6279746520""6b657920666f7220506f6c7931333035",
734 "a6f745008f81c916a20dcc74eef2b2f0"
737 "0000000000000000000000000000000000000000000000000000000000000000",
738 "746869732069732033322d6279746520""6b657920666f7220506f6c7931333035",
739 "49ec78090e481ec6c26b33b91ccc0307"
742 "89dab80b7717c1db5db437860a3f70218e93e1b8f461fb677f16f35f6f87e2a9"
743 "1c99bc3a47ace47640cc95c345be5ecca5a3523c35cc01893af0b64a62033427"
744 "0372ec12482d1b1e363561698a578b359803495bb4e2ef1930b17a5190b580f1"
745 "41300df30adbeca28f6427a8bc1a999fd51c554a017d095d8c3e3127daf9f595",
746 "2d773be37adb1e4d683bf0075e79c4ee""037918535a7f99ccb7040fb5f5f43aea",
747 "c85d15ed44c378d6b00e23064c7bcd51"
750 "000000000000000b1703030200000000"
751 "06db1f1f368d696a810a349c0c714c9a5e7850c2407d721acded95e018d7a852"
752 "66a6e1289cdb4aeb18da5ac8a2b0026d24a59ad485227f3eaedbb2e7e35e1c66"
753 "cd60f9abf716dcc9ac42682dd7dab287a7024c4eefc321cc0574e16793e37cec"
754 "03c5bda42b54c114a80b57af26416c7be742005e20855c73e21dc8e2edc9d435"
755 "cb6f6059280011c270b71570051c1c9b3052126620bc1e2730fa066c7a509d53"
756 "c60e5ae1b40aa6e39e49669228c90eecb4a50db32a50bc49e90b4f4b359a1dfd"
757 "11749cd3867fcf2fb7bb6cd4738f6a4ad6f7ca5058f7618845af9f020f6c3b96"
758 "7b8f4cd4a91e2813b507ae66f2d35c18284f7292186062e10fd5510d18775351"
759 "ef334e7634ab4743f5b68f49adcab384d3fd75f7390f4006ef2a295c8c7a076a"
760 "d54546cd25d2107fbe1436c840924aaebe5b370893cd63d1325b8616fc481088"
761 "6bc152c53221b6df373119393255ee72bcaa880174f1717f9184fa91646f17a2"
762 "4ac55d16bfddca9581a92eda479201f0edbf633600d6066d1ab36d5d2415d713"
763 "51bbcd608a25108d25641992c1f26c531cf9f90203bc4cc19f5927d834b0a471"
764 "16d3884bbb164b8ec883d1ac832e56b3918a98601a08d171881541d594db399c"
765 "6ae6151221745aec814c45b0b05b565436fd6f137aa10a0c0b643761dbd6f9a9"
766 "dcb99b1a6e690854ce0769cde39761d82fcdec15f0d92d7d8e94ade8eb83fbe0",
767 "99e5822dd4173c995e3dae0ddefb9774""3fde3b080134b39f76e9bf8d0e88d546",
768 "2637408fe13086ea73f971e3425e2820"
771 * test vectors from Hanno Böck
774 "cccccccccccccccccccccccccccccccccccccccccccccccccc80cccccccccccc"
775 "cccccccccccccccccccccccccccccccccccccccccccccccccccccccccecccccc"
776 "ccccccccccccccccccccccccccccccc5cccccccccccccccccccccccccccccccc"
777 "cccccccccce3cccccccccccccccccccccccccccccccccccccccccccccccccccc"
778 "ccccccccaccccccccccccccccccccce6cccccccccc000000afcccccccccccccc"
779 "ccccfffffff50000000000000000000000000000000000000000000000000000"
780 "00ffffffe7000000000000000000000000000000000000000000000000000000"
781 "0000000000000000000000000000000000000000000000000000719205a8521d"
783 "7f1b0264000000000000000000000000""0000000000000000cccccccccccccccc",
784 "8559b876eceed66eb37798c0457baff9"
787 "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa0000000000"
789 "e0001600000000000000000000000000""0000aaaaaaaaaaaaaaaaaaaaaaaaaaaa",
790 "00bd1258978e205444c9aaaa82006fed"
794 "0c0c0c0c0c0c0c0c0c0c0c0c0c0c0c0c""0c0c0c0c0c0c0c0c0c0c0c0c0c0c0c0c",
795 "06120c0c0c0c0c0c0c0c0c0c0c0c0c0c"
798 "7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b"
799 "7b7b7b7b7b7b7a7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b"
800 "7b7b5c7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b"
801 "7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b6e7b007b7b7b7b7b7b7b7b7b"
802 "7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7a7b7b7b7b7b7b7b7b7b7b7b7b"
803 "7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b5c7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b"
804 "7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b7b"
805 "7b6e7b001300000000b300000000000000000000000000000000000000000000"
806 "f20000000000000000000000000000000000002000efff000900000000000000"
807 "0000000000100000000009000000640000000000000000000000001300000000"
808 "b300000000000000000000000000000000000000000000f20000000000000000"
809 "000000000000000000002000efff00090000000000000000007a000010000000"
810 "000900000064000000000000000000000000000000000000000000000000fc",
811 "00ff0000000000000000000000000000""00000000001e00000000000000007b7b",
812 "33205bbf9e9f8f7212ab9e2ab9b7e4a5"
815 "7777777777777777777777777777777777777777777777777777777777777777"
816 "7777777777777777777777777777777777777777777777777777777777777777"
817 "777777777777777777777777ffffffe9e9acacacacacacacacacacac0000acac"
818 "ec0100acacac2caca2acacacacacacacacacacac64f2",
819 "0000007f0000007f0100002000000000""0000cf77777777777777777777777777",
820 "02ee7c8c546ddeb1a467e4c3981158b9"
823 * test vectors from Andrew Moon
826 "8e993b9f48681273c29650ba32fc76ce48332ea7164d96a4476fb8c531a1186a"
827 "c0dfc17c98dce87b4da7f011ec48c97271d2c20f9b928fe2270d6fb863d51738"
828 "b48eeee314a7cc8ab932164548e526ae90224368517acfeabd6bb3732bc0e9da"
829 "99832b61ca01b6de56244a9e88d5f9b37973f622a43d14a6599b1f654cb45a74"
831 "eea6a7251c1e72916d11c2cb214d3c25""2539121d8e234e652d651fa4c8cff880",
832 "f3ffc7703f9400e52a7dfb4b3d3305d9"
835 "ffffffffffffffffffffffffffffffff",
836 "02000000000000000000000000000000""00000000000000000000000000000000",
837 "03000000000000000000000000000000"
840 "02000000000000000000000000000000",
841 "02000000000000000000000000000000""ffffffffffffffffffffffffffffffff",
842 "03000000000000000000000000000000"
845 "fffffffffffffffffffffffffffffffff0ffffffffffffffffffffffffffffff"
846 "11000000000000000000000000000000",
847 "01000000000000000000000000000000""00000000000000000000000000000000",
848 "05000000000000000000000000000000"
851 "fffffffffffffffffffffffffffffffffbfefefefefefefefefefefefefefefe"
852 "01010101010101010101010101010101",
853 "01000000000000000000000000000000""00000000000000000000000000000000",
854 "00000000000000000000000000000000"
857 "fdffffffffffffffffffffffffffffff",
858 "02000000000000000000000000000000""00000000000000000000000000000000",
859 "faffffffffffffffffffffffffffffff"
861 { /* 5*H+L reduction intermediate */
862 "e33594d7505e43b900000000000000003394d7505e4379cd0100000000000000"
863 "0000000000000000000000000000000001000000000000000000000000000000",
864 "01000000000000000400000000000000""00000000000000000000000000000000",
865 "14000000000000005500000000000000"
867 { /* 5*H+L reduction final */
868 "e33594d7505e43b900000000000000003394d7505e4379cd0100000000000000"
869 "00000000000000000000000000000000",
870 "01000000000000000400000000000000""00000000000000000000000000000000",
871 "13000000000000000000000000000000"
875 static unsigned char hex_digit(char h)
877 int i = OPENSSL_hexchar2int(h);
884 static void hex_decode(unsigned char *out, const char *hex)
889 unsigned char v = hex_digit(*hex++);
891 v |= hex_digit(*hex++);
896 static void hexdump(unsigned char *a, size_t len)
900 for (i = 0; i < len; i++)
901 printf("%02x", a[i]);
906 static const unsigned num_tests =
907 sizeof(poly1305_tests) / sizeof(struct poly1305_test);
909 unsigned char key[32], out[16], expected[16];
912 for (i = 0; i < num_tests; i++) {
913 const struct poly1305_test *test = &poly1305_tests[i];
915 size_t inlen = strlen(test->inputhex);
917 if (strlen(test->keyhex) != sizeof(key) * 2 ||
918 strlen(test->outhex) != sizeof(out) * 2 || (inlen & 1) == 1)
923 hex_decode(key, test->keyhex);
924 hex_decode(expected, test->outhex);
928 hex_decode(in, test->inputhex);
930 Poly1305_Init(&poly1305, key);
931 Poly1305_Update(&poly1305, in, inlen);
932 Poly1305_Final(&poly1305, out);
934 if (memcmp(out, expected, sizeof(expected)) != 0) {
935 printf("Poly1305 test #%d failed.\n", i);
937 hexdump(out, sizeof(out));
938 printf("\nexpected: ");
939 hexdump(expected, sizeof(expected));
945 Poly1305_Init(&poly1305, key);
946 Poly1305_Update(&poly1305, in, 1);
947 Poly1305_Update(&poly1305, in+1, inlen-1);
948 Poly1305_Final(&poly1305, out);
950 if (memcmp(out, expected, sizeof(expected)) != 0) {
951 printf("Poly1305 test #%d/1+(N-1) failed.\n", i);
953 hexdump(out, sizeof(out));
954 printf("\nexpected: ");
955 hexdump(expected, sizeof(expected));
962 size_t half = inlen / 2;
964 Poly1305_Init(&poly1305, key);
965 Poly1305_Update(&poly1305, in, half);
966 Poly1305_Update(&poly1305, in+half, inlen-half);
967 Poly1305_Final(&poly1305, out);
969 if (memcmp(out, expected, sizeof(expected)) != 0) {
970 printf("Poly1305 test #%d/2 failed.\n", i);
972 hexdump(out, sizeof(out));
973 printf("\nexpected: ");
974 hexdump(expected, sizeof(expected));
979 for (half = 16; half < inlen; half += 16) {
980 Poly1305_Init(&poly1305, key);
981 Poly1305_Update(&poly1305, in, half);
982 Poly1305_Update(&poly1305, in+half, inlen-half);
983 Poly1305_Final(&poly1305, out);
985 if (memcmp(out, expected, sizeof(expected)) != 0) {
986 printf("Poly1305 test #%d/%d+%d failed.\n",
987 i, half, inlen-half);
989 hexdump(out, sizeof(out));
990 printf("\nexpected: ");
991 hexdump(expected, sizeof(expected));
1003 # ifdef OPENSSL_CPUID_OBJ
1005 unsigned char buf[8192];
1006 unsigned long long stopwatch;
1007 unsigned long long OPENSSL_rdtsc();
1009 memset (buf,0x55,sizeof(buf));
1010 memset (key,0xAA,sizeof(key));
1012 Poly1305_Init(&poly1305, key);
1014 for (i=0;i<100000;i++)
1015 Poly1305_Update(&poly1305,buf,sizeof(buf));
1017 stopwatch = OPENSSL_rdtsc();
1018 for (i=0;i<10000;i++)
1019 Poly1305_Update(&poly1305,buf,sizeof(buf));
1020 stopwatch = OPENSSL_rdtsc() - stopwatch;
1022 printf("%g\n",stopwatch/(double)(i*sizeof(buf)));
1024 stopwatch = OPENSSL_rdtsc();
1025 for (i=0;i<10000;i++) {
1026 Poly1305_Init(&poly1305, key);
1027 Poly1305_Update(&poly1305,buf,16);
1028 Poly1305_Final(&poly1305,buf);
1030 stopwatch = OPENSSL_rdtsc() - stopwatch;
1032 printf("%g\n",stopwatch/(double)(i));