poly1305.c

Jenkins

poly1305.c

Absolute File Name:

/home/opencoverage/opencoverage/guest-scripts/openssl/src/crypto/poly1305/poly1305.c

Line

Source

Count

* Licensed under the OpenSSL license (the "License"). You may not use

* this file except in compliance with the License. You can obtain a copy

* in the file LICENSE in the source distribution or at

* https://www.openssl.org/source/license.html

#include <stdlib.h>

#include <string.h>

#include <openssl/crypto.h>

#include "internal/poly1305.h"

#include "poly1305_local.h"

size_t Poly1305_ctx_size(void)

{

return sizeof(struct poly1305_context);

executed 13290 times by 1 test: return sizeof(struct poly1305_context);

Executed by:

libcrypto.so.1.1

13290

}

/* pick 32-bit unsigned integer in little endian order */

static unsigned int U8TOU32(const unsigned char *p)

{

return (((unsigned int)(p[0] & 0xff)) |

executed 60844 times by 2 tests:

return (((unsigned int)(p[0] & 0xff)) | ((unsigned int)(p[1] & 0xff) << 8) | ((unsigned int)(p[2] & 0xff) << 16) | ((unsigned int)(p[3] & 0xff) << 24));

Executed by:

libcrypto.so.1.1
poly1305_internal_test

60844

((unsigned int)(p[1] & 0xff) << 8) |

executed 60844 times by 2 tests:

return (((unsigned int)(p[0] & 0xff)) | ((unsigned int)(p[1] & 0xff) << 8) | ((unsigned int)(p[2] & 0xff) << 16) | ((unsigned int)(p[3] & 0xff) << 24));

Executed by:

libcrypto.so.1.1
poly1305_internal_test

60844

((unsigned int)(p[2] & 0xff) << 16) |

executed 60844 times by 2 tests:

return (((unsigned int)(p[0] & 0xff)) | ((unsigned int)(p[1] & 0xff) << 8) | ((unsigned int)(p[2] & 0xff) << 16) | ((unsigned int)(p[3] & 0xff) << 24));

Executed by:

libcrypto.so.1.1
poly1305_internal_test

60844

((unsigned int)(p[3] & 0xff) << 24));

executed 60844 times by 2 tests:

return (((unsigned int)(p[0] & 0xff)) | ((unsigned int)(p[1] & 0xff) << 8) | ((unsigned int)(p[2] & 0xff) << 16) | ((unsigned int)(p[3] & 0xff) << 24));

Executed by:

libcrypto.so.1.1
poly1305_internal_test

60844

}

* Implementations can be classified by amount of significant bits in

* words making up the multi-precision value, or in other words radix

* or base of numerical representation, e.g. base 2^64, base 2^32,

* base 2^26. Complementary characteristic is how wide is the result of

* multiplication of pair of digits, e.g. it would take 128 bits to

* accommodate multiplication result in base 2^64 case. These are used

* interchangeably. To describe implementation that is. But interface

* is designed to isolate this so that low-level primitives implemented

* in assembly can be self-contained/self-coherent.

#ifndef POLY1305_ASM

* Even though there is __int128 reference implementation targeting

* 64-bit platforms provided below, it's not obvious that it's optimal

* choice for every one of them. Depending on instruction set overall

* amount of instructions can be comparable to one in __int64

* implementation. Amount of multiplication instructions would be lower,

* but not necessarily overall. And in out-of-order execution context,

* it is the latter that can be crucial...

* On related note. Poly1305 author, D. J. Bernstein, discusses and

* provides floating-point implementations of the algorithm in question.

* It made a lot of sense by the time of introduction, because most

* then-modern processors didn't have pipelined integer multiplier.

* [Not to mention that some had non-constant timing for integer

* multiplications.] Floating-point instructions on the other hand could

* be issued every cycle, which allowed to achieve better performance.

* Nowadays, with SIMD and/or out-or-order execution, shared or

* even emulated FPU, it's more complicated, and floating-point

* implementation is not necessarily optimal choice in every situation,

* rather contrary...

* <appro@openssl.org>

typedef unsigned int u32;

* poly1305_blocks processes a multiple of POLY1305_BLOCK_SIZE blocks

* of |inp| no longer than |len|. Behaviour for |len| not divisible by

* block size is unspecified in general case, even though in reference

* implementation the trailing chunk is simply ignored. Per algorithm

* specification, every input block, complete or last partial, is to be

* padded with a bit past most significant byte. The latter kind is then

* padded with zeros till block size. This last partial block padding

* is caller(*)'s responsibility, and because of this the last partial

* block is always processed with separate call with |len| set to

* POLY1305_BLOCK_SIZE and |padbit| to 0. In all other cases |padbit|

* should be set to 1 to perform implicit padding with 128th bit.

* poly1305_blocks does not actually check for this constraint though,

* it's caller(*)'s responsibility to comply.

* (*) In the context "caller" is not application code, but higher

* level Poly1305_* from this very module, so that quirks are

* handled locally.

static void

poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit);

* Type-agnostic "rip-off" from constant_time_locl.h

# define CONSTANT_TIME_CARRY(a,b) ( \

(a ^ ((a ^ b) | ((a - b) ^ b))) >> (sizeof(a) * 8 - 1) \

)

# if (defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16) && \

(defined(__SIZEOF_LONG__) && __SIZEOF_LONG__==8)

100

101

typedef unsigned long u64;

102

typedef __uint128_t u128;

typedef struct {

u64 h[3];

u64 r[2];

} poly1305_internal;

/* pick 32-bit unsigned integer in little endian order */

110

static u64 U8TOU64(const unsigned char *p)

111

{

112

return (((u64)(p[0] & 0xff)) |

113

((u64)(p[1] & 0xff) << 8) |

114

((u64)(p[2] & 0xff) << 16) |

115

((u64)(p[3] & 0xff) << 24) |

116

((u64)(p[4] & 0xff) << 32) |

117

((u64)(p[5] & 0xff) << 40) |

118

((u64)(p[6] & 0xff) << 48) |

119

((u64)(p[7] & 0xff) << 56));

120

}

121

122

/* store a 32-bit unsigned integer in little endian */

123

static void U64TO8(unsigned char *p, u64 v)

124

{

125

p[0] = (unsigned char)((v) & 0xff);

126

p[1] = (unsigned char)((v >> 8) & 0xff);

127

p[2] = (unsigned char)((v >> 16) & 0xff);

128

p[3] = (unsigned char)((v >> 24) & 0xff);

129

p[4] = (unsigned char)((v >> 32) & 0xff);

130

p[5] = (unsigned char)((v >> 40) & 0xff);

131

p[6] = (unsigned char)((v >> 48) & 0xff);

132

p[7] = (unsigned char)((v >> 56) & 0xff);

133

}

134

135

static void poly1305_init(void *ctx, const unsigned char key[16])

136

{

137

poly1305_internal *st = (poly1305_internal *) ctx;

/* h = 0 */

st->h[0] = 0;

st->h[1] = 0;

st->h[2] = 0;

/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */

145

st->r[0] = U8TOU64(&key[0]) & 0x0ffffffc0fffffff;

146

st->r[1] = U8TOU64(&key[8]) & 0x0ffffffc0ffffffc;

}

static void

poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit)

151

{

152

poly1305_internal *st = (poly1305_internal *)ctx;

u64 r0, r1;

u64 s1;

u64 h0, h1, h2, c;

u128 d0, d1;

r0 = st->r[0];

r1 = st->r[1];

s1 = r1 + (r1 >> 2);

h0 = st->h[0];

h1 = st->h[1];

h2 = st->h[2];

while (len >= POLY1305_BLOCK_SIZE) {

168

/* h += m[i] */

169

h0 = (u64)(d0 = (u128)h0 + U8TOU64(inp + 0));

170

h1 = (u64)(d1 = (u128)h1 + (d0 >> 64) + U8TOU64(inp + 8));

171

172

* padbit can be zero only when original len was

173

* POLY1306_BLOCK_SIZE, but we don't check

174

175

h2 += (u64)(d1 >> 64) + padbit;

176

177

/* h *= r "%" p, where "%" stands for "partial remainder" */

178

d0 = ((u128)h0 * r0) +

179

((u128)h1 * s1);

180

d1 = ((u128)h0 * r1) +

((u128)h1 * r0) +

(h2 * s1);

h2 = (h2 * r0);

/* last reduction step: */

186

/* a) h2:h0 = h2<<128 + d1<<64 + d0 */

187

h0 = (u64)d0;

188

h1 = (u64)(d1 += d0 >> 64);

189

h2 += (u64)(d1 >> 64);

190

/* b) (h2:h0 += (h2:h0>>130) * 5) %= 2^130 */

191

c = (h2 >> 2) + (h2 & ~3UL);

192

h2 &= 3;

193

h0 += c;

194

h1 += (c = CONSTANT_TIME_CARRY(h0,c));

195

h2 += CONSTANT_TIME_CARRY(h1,c);

196

197

* Occasional overflows to 3rd bit of h2 are taken care of

198

* "naturally". If after this point we end up at the top of

199

* this loop, then the overflow bit will be accounted for

200

* in next iteration. If we end up in poly1305_emit, then

201

* comparison to modulus below will still count as "carry

202

* into 131st bit", so that properly reduced value will be

203

* picked in conditional move.

204

205

206

inp += POLY1305_BLOCK_SIZE;

207

len -= POLY1305_BLOCK_SIZE;

}

st->h[0] = h0;

st->h[1] = h1;

st->h[2] = h2;

}

static void poly1305_emit(void *ctx, unsigned char mac[16],

216

const u32 nonce[4])

217

{

218

poly1305_internal *st = (poly1305_internal *) ctx;

u64 h0, h1, h2;

u64 g0, g1, g2;

u128 t;

u64 mask;

h0 = st->h[0];

h1 = st->h[1];

h2 = st->h[2];

/* compare to modulus by computing h + -p */

229

g0 = (u64)(t = (u128)h0 + 5);

230

g1 = (u64)(t = (u128)h1 + (t >> 64));

231

g2 = h2 + (u64)(t >> 64);

232

233

/* if there was carry into 131st bit, h1:h0 = g1:g0 */

234

mask = 0 - (g2 >> 2);

g0 &= mask;

g1 &= mask;

mask = ~mask;

h0 = (h0 & mask) | g0;

239

h1 = (h1 & mask) | g1;

240

241

/* mac = (h + nonce) % (2^128) */

242

h0 = (u64)(t = (u128)h0 + nonce[0] + ((u64)nonce[1]<<32));

243

h1 = (u64)(t = (u128)h1 + nonce[2] + ((u64)nonce[3]<<32) + (t >> 64));

U64TO8(mac + 0, h0);

U64TO8(mac + 8, h1);

}

# else

# if defined(_WIN32) && !defined(__MINGW32__)

252

typedef unsigned __int64 u64;

253

# elif defined(__arch64__)

254

typedef unsigned long u64;

255

# else

256

typedef unsigned long long u64;

# endif

typedef struct {

u32 h[5];

u32 r[4];

} poly1305_internal;

/* store a 32-bit unsigned integer in little endian */

265

static void U32TO8(unsigned char *p, unsigned int v)

266

{

267

p[0] = (unsigned char)((v) & 0xff);

268

p[1] = (unsigned char)((v >> 8) & 0xff);

269

p[2] = (unsigned char)((v >> 16) & 0xff);

270

p[3] = (unsigned char)((v >> 24) & 0xff);

271

}

272

273

static void poly1305_init(void *ctx, const unsigned char key[16])

274

{

275

poly1305_internal *st = (poly1305_internal *) ctx;

/* h = 0 */

st->h[0] = 0;

st->h[1] = 0;

st->h[2] = 0;

st->h[3] = 0;

st->h[4] = 0;

/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */

285

st->r[0] = U8TOU32(&key[0]) & 0x0fffffff;

286

st->r[1] = U8TOU32(&key[4]) & 0x0ffffffc;

287

st->r[2] = U8TOU32(&key[8]) & 0x0ffffffc;

288

st->r[3] = U8TOU32(&key[12]) & 0x0ffffffc;

}

static void

poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, u32 padbit)

293

{

294

poly1305_internal *st = (poly1305_internal *)ctx;

295

u32 r0, r1, r2, r3;

296

u32 s1, s2, s3;

297

u32 h0, h1, h2, h3, h4, c;

u64 d0, d1, d2, d3;

r0 = st->r[0];

r1 = st->r[1];

r2 = st->r[2];

r3 = st->r[3];

s1 = r1 + (r1 >> 2);

s2 = r2 + (r2 >> 2);

s3 = r3 + (r3 >> 2);

h0 = st->h[0];

h1 = st->h[1];

h2 = st->h[2];

h3 = st->h[3];

h4 = st->h[4];

while (len >= POLY1305_BLOCK_SIZE) {

316

/* h += m[i] */

317

h0 = (u32)(d0 = (u64)h0 + U8TOU32(inp + 0));

318

h1 = (u32)(d1 = (u64)h1 + (d0 >> 32) + U8TOU32(inp + 4));

319

h2 = (u32)(d2 = (u64)h2 + (d1 >> 32) + U8TOU32(inp + 8));

320

h3 = (u32)(d3 = (u64)h3 + (d2 >> 32) + U8TOU32(inp + 12));

321

h4 += (u32)(d3 >> 32) + padbit;

322

323

/* h *= r "%" p, where "%" stands for "partial remainder" */

324

d0 = ((u64)h0 * r0) +

((u64)h1 * s3) +

((u64)h2 * s2) +

((u64)h3 * s1);

d1 = ((u64)h0 * r1) +

((u64)h1 * r0) +

((u64)h2 * s3) +

((u64)h3 * s2) +

(h4 * s1);

d2 = ((u64)h0 * r2) +

((u64)h1 * r1) +

((u64)h2 * r0) +

((u64)h3 * s3) +

(h4 * s2);

d3 = ((u64)h0 * r3) +

((u64)h1 * r2) +

((u64)h2 * r1) +

((u64)h3 * r0) +

(h4 * s3);

h4 = (h4 * r0);

/* last reduction step: */

346

/* a) h4:h0 = h4<<128 + d3<<96 + d2<<64 + d1<<32 + d0 */

347

h0 = (u32)d0;

348

h1 = (u32)(d1 += d0 >> 32);

349

h2 = (u32)(d2 += d1 >> 32);

350

h3 = (u32)(d3 += d2 >> 32);

351

h4 += (u32)(d3 >> 32);

352

/* b) (h4:h0 += (h4:h0>>130) * 5) %= 2^130 */

353

c = (h4 >> 2) + (h4 & ~3U);

354

h4 &= 3;

355

h0 += c;

356

h1 += (c = CONSTANT_TIME_CARRY(h0,c));

357

h2 += (c = CONSTANT_TIME_CARRY(h1,c));

358

h3 += (c = CONSTANT_TIME_CARRY(h2,c));

359

h4 += CONSTANT_TIME_CARRY(h3,c);

360

361

* Occasional overflows to 3rd bit of h4 are taken care of

362

* "naturally". If after this point we end up at the top of

363

* this loop, then the overflow bit will be accounted for

364

* in next iteration. If we end up in poly1305_emit, then

365

* comparison to modulus below will still count as "carry

366

* into 131st bit", so that properly reduced value will be

367

* picked in conditional move.

368

369

370

inp += POLY1305_BLOCK_SIZE;

371

len -= POLY1305_BLOCK_SIZE;

}

st->h[0] = h0;

st->h[1] = h1;

st->h[2] = h2;

st->h[3] = h3;

st->h[4] = h4;

}

static void poly1305_emit(void *ctx, unsigned char mac[16],

382

const u32 nonce[4])

383

{

384

poly1305_internal *st = (poly1305_internal *) ctx;

385

u32 h0, h1, h2, h3, h4;

386

u32 g0, g1, g2, g3, g4;

u64 t;

u32 mask;

h0 = st->h[0];

h1 = st->h[1];

h2 = st->h[2];

h3 = st->h[3];

h4 = st->h[4];

/* compare to modulus by computing h + -p */

397

g0 = (u32)(t = (u64)h0 + 5);

398

g1 = (u32)(t = (u64)h1 + (t >> 32));

399

g2 = (u32)(t = (u64)h2 + (t >> 32));

400

g3 = (u32)(t = (u64)h3 + (t >> 32));

401

g4 = h4 + (u32)(t >> 32);

402

403

/* if there was carry into 131st bit, h3:h0 = g3:g0 */

404

mask = 0 - (g4 >> 2);

g0 &= mask;

g1 &= mask;

g2 &= mask;

g3 &= mask;

mask = ~mask;

h0 = (h0 & mask) | g0;

411

h1 = (h1 & mask) | g1;

412

h2 = (h2 & mask) | g2;

413

h3 = (h3 & mask) | g3;

414

415

/* mac = (h + nonce) % (2^128) */

416

h0 = (u32)(t = (u64)h0 + nonce[0]);

417

h1 = (u32)(t = (u64)h1 + (t >> 32) + nonce[1]);

418

h2 = (u32)(t = (u64)h2 + (t >> 32) + nonce[2]);

419

h3 = (u32)(t = (u64)h3 + (t >> 32) + nonce[3]);

U32TO8(mac + 0, h0);

U32TO8(mac + 4, h1);

U32TO8(mac + 8, h2);

U32TO8(mac + 12, h3);

}

# endif

#else

int poly1305_init(void *ctx, const unsigned char key[16], void *func);

429

void poly1305_blocks(void *ctx, const unsigned char *inp, size_t len,

430

unsigned int padbit);

431

void poly1305_emit(void *ctx, unsigned char mac[16],

432

const unsigned int nonce[4]);

433

#endif

434

435

void Poly1305_Init(POLY1305 *ctx, const unsigned char key[32])

436

{

437

ctx->nonce[0] = U8TOU32(&key[16]);

438

ctx->nonce[1] = U8TOU32(&key[20]);

439

ctx->nonce[2] = U8TOU32(&key[24]);

440

ctx->nonce[3] = U8TOU32(&key[28]);

441

442

#ifndef POLY1305_ASM

443

poly1305_init(ctx->opaque, key);

444

#else

445

446

* Unlike reference poly1305_init assembly counterpart is expected

447

* to return a value: non-zero if it initializes ctx->func, and zero

448

* otherwise. Latter is to simplify assembly in cases when there no

449

* multiple code paths to switch between.

450

451

if (!poly1305_init(ctx->opaque, key, &ctx->func)) {

!poly1305_init...y, &ctx->func)	Description
TRUE	never evaluated
FALSE	evaluated 15211 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

0-15211

452

ctx->func.blocks = poly1305_blocks;

453

ctx->func.emit = poly1305_emit;

454

}

never executed: end of block

#endif

ctx->num = 0;

}

executed 15211 times by 2 tests: end of block

Executed by:

libcrypto.so.1.1
poly1305_internal_test

15211

#ifdef POLY1305_ASM

* This "eclipses" poly1305_blocks and poly1305_emit, but it's

464

* conscious choice imposed by -Wshadow compiler warnings.

465

466

# define poly1305_blocks (*poly1305_blocks_p)

467

# define poly1305_emit (*poly1305_emit_p)

468

#endif

469

470

void Poly1305_Update(POLY1305 *ctx, const unsigned char *inp, size_t len)

{

#ifdef POLY1305_ASM

* As documented, poly1305_blocks is never called with input

475

* longer than single block and padbit argument set to 0. This

476

* property is fluently used in assembly modules to optimize

477

* padbit handling on loop boundary.

478

479

poly1305_blocks_f poly1305_blocks_p = ctx->func.blocks;

#endif

size_t rem, num;

if ((num = ctx->num)) {

(num = ctx->num)	Description
TRUE	evaluated 24990 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 43508 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

24990-43508

484

rem = POLY1305_BLOCK_SIZE - num;

485

if (len >= rem) {

len >= rem	Description
TRUE	evaluated 24846 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 144 times by 1 test Evaluated by: libcrypto.so.1.1

144-24846

486

memcpy(ctx->data + num, inp, rem);

487

poly1305_blocks(ctx->opaque, ctx->data, POLY1305_BLOCK_SIZE, 1);

488

inp += rem;

489

len -= rem;

490

} else {

executed 24846 times by 2 tests: end of block

Executed by:

libcrypto.so.1.1
poly1305_internal_test

24846

491

/* Still not enough data to process a block. */

492

memcpy(ctx->data + num, inp, len);

493

ctx->num = num + len;

494

return;

executed 144 times by 1 test: return;

Executed by:

libcrypto.so.1.1

144

}

}

rem = len % POLY1305_BLOCK_SIZE;

499

len -= rem;

500

501

if (len >= POLY1305_BLOCK_SIZE) {

len >= 16	Description
TRUE	evaluated 34363 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 33991 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

33991-34363

502

poly1305_blocks(ctx->opaque, inp, len, 1);

503

inp += len;

504

}

executed 34363 times by 2 tests: end of block

Executed by:

libcrypto.so.1.1
poly1305_internal_test

34363

505

506

if (rem)

rem	Description
TRUE	evaluated 24965 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 43389 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

24965-43389

507

memcpy(ctx->data, inp, rem);

executed 24965 times by 2 tests: memcpy(ctx->data, inp, rem);

Executed by:

libcrypto.so.1.1
poly1305_internal_test

24965

508

509

ctx->num = rem;

510

}

executed 68354 times by 2 tests: end of block

Executed by:

libcrypto.so.1.1
poly1305_internal_test

68354

511

512

void Poly1305_Final(POLY1305 *ctx, unsigned char mac[16])

513

{

514

#ifdef POLY1305_ASM

515

poly1305_blocks_f poly1305_blocks_p = ctx->func.blocks;

516

poly1305_emit_f poly1305_emit_p = ctx->func.emit;

#endif

size_t num;

if ((num = ctx->num)) {

(num = ctx->num)	Description
TRUE	evaluated 119 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 15092 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

119-15092

521

ctx->data[num++] = 1; /* pad bit */

522

while (num < POLY1305_BLOCK_SIZE)

num < 16	Description
TRUE	evaluated 780 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test
FALSE	evaluated 119 times by 2 tests Evaluated by: libcrypto.so.1.1 poly1305_internal_test

119-780

523

ctx->data[num++] = 0;

executed 780 times by 2 tests: ctx->data[num++] = 0;

Executed by: