This commit is contained in:
Frank Denis
2017-02-23 09:47:12 +01:00
parent cad6561799
commit a1c4cf5b3c
6 changed files with 406 additions and 307 deletions
@@ -1,11 +1,12 @@
#include "poly1305_donna.h"
#include "crypto_verify_16.h"
#include "utils.h"
#include "poly1305_donna.h"
#ifdef HAVE_TI_MODE
# include "poly1305_donna64.h"
#include "poly1305_donna64.h"
#else
# include "poly1305_donna32.h"
#include "poly1305_donna32.h"
#endif
#include "../onetimeauth_poly1305.h"
@@ -19,15 +20,18 @@ poly1305_update(poly1305_state_internal_t *st, const unsigned char *m,
if (st->leftover) {
unsigned long long want = (poly1305_block_size - st->leftover);
if (want > bytes)
if (want > bytes) {
want = bytes;
for (i = 0; i < want; i++)
}
for (i = 0; i < want; i++) {
st->buffer[st->leftover + i] = m[i];
}
bytes -= want;
m += want;
st->leftover += want;
if (st->leftover < poly1305_block_size)
if (st->leftover < poly1305_block_size) {
return;
}
poly1305_blocks(st, st->buffer, poly1305_block_size);
st->leftover = 0;
}
@@ -52,7 +56,7 @@ poly1305_update(poly1305_state_internal_t *st, const unsigned char *m,
static int
crypto_onetimeauth_poly1305_donna(unsigned char *out, const unsigned char *m,
unsigned long long inlen,
unsigned long long inlen,
const unsigned char *key)
{
CRYPTO_ALIGN(64) poly1305_state_internal_t state;
@@ -66,30 +70,32 @@ crypto_onetimeauth_poly1305_donna(unsigned char *out, const unsigned char *m,
static int
crypto_onetimeauth_poly1305_donna_init(crypto_onetimeauth_poly1305_state *state,
const unsigned char *key)
const unsigned char * key)
{
(void) sizeof(int[sizeof (crypto_onetimeauth_poly1305_state) >=
sizeof (poly1305_state_internal_t) ? 1 : -1]);
poly1305_init((poly1305_state_internal_t *)(void *) state, key);
(void) sizeof(int[sizeof(crypto_onetimeauth_poly1305_state) >=
sizeof(poly1305_state_internal_t)
? 1
: -1]);
poly1305_init((poly1305_state_internal_t *) (void *) state, key);
return 0;
}
static int
crypto_onetimeauth_poly1305_donna_update(crypto_onetimeauth_poly1305_state *state,
const unsigned char *in,
unsigned long long inlen)
crypto_onetimeauth_poly1305_donna_update(
crypto_onetimeauth_poly1305_state *state, const unsigned char *in,
unsigned long long inlen)
{
poly1305_update((poly1305_state_internal_t *)(void *) state, in, inlen);
poly1305_update((poly1305_state_internal_t *) (void *) state, in, inlen);
return 0;
}
static int
crypto_onetimeauth_poly1305_donna_final(crypto_onetimeauth_poly1305_state *state,
unsigned char *out)
crypto_onetimeauth_poly1305_donna_final(
crypto_onetimeauth_poly1305_state *state, unsigned char *out)
{
poly1305_finish((poly1305_state_internal_t *)(void *) state, out);
poly1305_finish((poly1305_state_internal_t *) (void *) state, out);
return 0;
}
@@ -97,21 +103,23 @@ crypto_onetimeauth_poly1305_donna_final(crypto_onetimeauth_poly1305_state *state
static int
crypto_onetimeauth_poly1305_donna_verify(const unsigned char *h,
const unsigned char *in,
unsigned long long inlen,
unsigned long long inlen,
const unsigned char *k)
{
unsigned char correct[16];
crypto_onetimeauth_poly1305_donna(correct,in,inlen,k);
crypto_onetimeauth_poly1305_donna(correct, in, inlen, k);
return crypto_verify_16(h,correct);
return crypto_verify_16(h, correct);
}
struct crypto_onetimeauth_poly1305_implementation
crypto_onetimeauth_poly1305_donna_implementation = {
SODIUM_C99(.onetimeauth =) crypto_onetimeauth_poly1305_donna,
SODIUM_C99(.onetimeauth_verify =) crypto_onetimeauth_poly1305_donna_verify,
SODIUM_C99(.onetimeauth_init =) crypto_onetimeauth_poly1305_donna_init,
SODIUM_C99(.onetimeauth_update =) crypto_onetimeauth_poly1305_donna_update,
SODIUM_C99(.onetimeauth_final =) crypto_onetimeauth_poly1305_donna_final
};
crypto_onetimeauth_poly1305_donna_implementation = {
SODIUM_C99(.onetimeauth =) crypto_onetimeauth_poly1305_donna,
SODIUM_C99(.onetimeauth_verify =)
crypto_onetimeauth_poly1305_donna_verify,
SODIUM_C99(.onetimeauth_init =) crypto_onetimeauth_poly1305_donna_init,
SODIUM_C99(.onetimeauth_update =)
crypto_onetimeauth_poly1305_donna_update,
SODIUM_C99(.onetimeauth_final =) crypto_onetimeauth_poly1305_donna_final
};
@@ -3,8 +3,8 @@
#include <stddef.h>
#include "crypto_onetimeauth_poly1305.h"
#include "../onetimeauth_poly1305.h"
#include "crypto_onetimeauth_poly1305.h"
extern struct crypto_onetimeauth_poly1305_implementation
crypto_onetimeauth_poly1305_donna_implementation;
@@ -1,13 +1,14 @@
/*
poly1305 implementation using 32 bit * 32 bit = 64 bit multiplication and 64 bit addition
poly1305 implementation using 32 bit * 32 bit = 64 bit multiplication
and 64 bit addition
*/
#if defined(_MSC_VER)
# define POLY1305_NOINLINE __declspec(noinline)
#define POLY1305_NOINLINE __declspec(noinline)
#elif defined(__GNUC__)
# define POLY1305_NOINLINE __attribute__ ((noinline))
#define POLY1305_NOINLINE __attribute__((noinline))
#else
# define POLY1305_NOINLINE
#define POLY1305_NOINLINE
#endif
#include "private/common.h"
@@ -16,173 +17,219 @@
/* 17 + sizeof(unsigned long long) + 14*sizeof(unsigned long) */
typedef struct poly1305_state_internal_t {
unsigned long r[5];
unsigned long h[5];
unsigned long pad[4];
unsigned long long leftover;
unsigned char buffer[poly1305_block_size];
unsigned char final;
unsigned long r[5];
unsigned long h[5];
unsigned long pad[4];
unsigned long long leftover;
unsigned char buffer[poly1305_block_size];
unsigned char final;
} poly1305_state_internal_t;
static void
poly1305_init(poly1305_state_internal_t *st, const unsigned char key[32])
{
/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
st->r[0] = (LOAD32_LE(&key[ 0]) ) & 0x3ffffff;
st->r[1] = (LOAD32_LE(&key[ 3]) >> 2) & 0x3ffff03;
st->r[2] = (LOAD32_LE(&key[ 6]) >> 4) & 0x3ffc0ff;
st->r[3] = (LOAD32_LE(&key[ 9]) >> 6) & 0x3f03fff;
st->r[4] = (LOAD32_LE(&key[12]) >> 8) & 0x00fffff;
/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
st->r[0] = (LOAD32_LE(&key[0])) & 0x3ffffff;
st->r[1] = (LOAD32_LE(&key[3]) >> 2) & 0x3ffff03;
st->r[2] = (LOAD32_LE(&key[6]) >> 4) & 0x3ffc0ff;
st->r[3] = (LOAD32_LE(&key[9]) >> 6) & 0x3f03fff;
st->r[4] = (LOAD32_LE(&key[12]) >> 8) & 0x00fffff;
/* h = 0 */
st->h[0] = 0;
st->h[1] = 0;
st->h[2] = 0;
st->h[3] = 0;
st->h[4] = 0;
/* h = 0 */
st->h[0] = 0;
st->h[1] = 0;
st->h[2] = 0;
st->h[3] = 0;
st->h[4] = 0;
/* save pad for later */
st->pad[0] = LOAD32_LE(&key[16]);
st->pad[1] = LOAD32_LE(&key[20]);
st->pad[2] = LOAD32_LE(&key[24]);
st->pad[3] = LOAD32_LE(&key[28]);
/* save pad for later */
st->pad[0] = LOAD32_LE(&key[16]);
st->pad[1] = LOAD32_LE(&key[20]);
st->pad[2] = LOAD32_LE(&key[24]);
st->pad[3] = LOAD32_LE(&key[28]);
st->leftover = 0;
st->final = 0;
st->leftover = 0;
st->final = 0;
}
static void
poly1305_blocks(poly1305_state_internal_t *st, const unsigned char *m, unsigned long long bytes)
poly1305_blocks(poly1305_state_internal_t *st, const unsigned char *m,
unsigned long long bytes)
{
const unsigned long hibit = (st->final) ? 0UL : (1UL << 24); /* 1 << 128 */
unsigned long r0,r1,r2,r3,r4;
unsigned long s1,s2,s3,s4;
unsigned long h0,h1,h2,h3,h4;
unsigned long long d0,d1,d2,d3,d4;
unsigned long c;
const unsigned long hibit = (st->final) ? 0UL : (1UL << 24); /* 1 << 128 */
unsigned long r0, r1, r2, r3, r4;
unsigned long s1, s2, s3, s4;
unsigned long h0, h1, h2, h3, h4;
unsigned long long d0, d1, d2, d3, d4;
unsigned long c;
r0 = st->r[0];
r1 = st->r[1];
r2 = st->r[2];
r3 = st->r[3];
r4 = st->r[4];
r0 = st->r[0];
r1 = st->r[1];
r2 = st->r[2];
r3 = st->r[3];
r4 = st->r[4];
s1 = r1 * 5;
s2 = r2 * 5;
s3 = r3 * 5;
s4 = r4 * 5;
s1 = r1 * 5;
s2 = r2 * 5;
s3 = r3 * 5;
s4 = r4 * 5;
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
h3 = st->h[3];
h4 = st->h[4];
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
h3 = st->h[3];
h4 = st->h[4];
while (bytes >= poly1305_block_size) {
/* h += m[i] */
h0 += (LOAD32_LE(m+ 0) ) & 0x3ffffff;
h1 += (LOAD32_LE(m+ 3) >> 2) & 0x3ffffff;
h2 += (LOAD32_LE(m+ 6) >> 4) & 0x3ffffff;
h3 += (LOAD32_LE(m+ 9) >> 6) & 0x3ffffff;
h4 += (LOAD32_LE(m+12) >> 8) | hibit;
while (bytes >= poly1305_block_size) {
/* h += m[i] */
h0 += (LOAD32_LE(m + 0)) & 0x3ffffff;
h1 += (LOAD32_LE(m + 3) >> 2) & 0x3ffffff;
h2 += (LOAD32_LE(m + 6) >> 4) & 0x3ffffff;
h3 += (LOAD32_LE(m + 9) >> 6) & 0x3ffffff;
h4 += (LOAD32_LE(m + 12) >> 8) | hibit;
/* h *= r */
d0 = ((unsigned long long)h0 * r0) + ((unsigned long long)h1 * s4) + ((unsigned long long)h2 * s3) + ((unsigned long long)h3 * s2) + ((unsigned long long)h4 * s1);
d1 = ((unsigned long long)h0 * r1) + ((unsigned long long)h1 * r0) + ((unsigned long long)h2 * s4) + ((unsigned long long)h3 * s3) + ((unsigned long long)h4 * s2);
d2 = ((unsigned long long)h0 * r2) + ((unsigned long long)h1 * r1) + ((unsigned long long)h2 * r0) + ((unsigned long long)h3 * s4) + ((unsigned long long)h4 * s3);
d3 = ((unsigned long long)h0 * r3) + ((unsigned long long)h1 * r2) + ((unsigned long long)h2 * r1) + ((unsigned long long)h3 * r0) + ((unsigned long long)h4 * s4);
d4 = ((unsigned long long)h0 * r4) + ((unsigned long long)h1 * r3) + ((unsigned long long)h2 * r2) + ((unsigned long long)h3 * r1) + ((unsigned long long)h4 * r0);
/* h *= r */
d0 = ((unsigned long long) h0 * r0) + ((unsigned long long) h1 * s4) +
((unsigned long long) h2 * s3) + ((unsigned long long) h3 * s2) +
((unsigned long long) h4 * s1);
d1 = ((unsigned long long) h0 * r1) + ((unsigned long long) h1 * r0) +
((unsigned long long) h2 * s4) + ((unsigned long long) h3 * s3) +
((unsigned long long) h4 * s2);
d2 = ((unsigned long long) h0 * r2) + ((unsigned long long) h1 * r1) +
((unsigned long long) h2 * r0) + ((unsigned long long) h3 * s4) +
((unsigned long long) h4 * s3);
d3 = ((unsigned long long) h0 * r3) + ((unsigned long long) h1 * r2) +
((unsigned long long) h2 * r1) + ((unsigned long long) h3 * r0) +
((unsigned long long) h4 * s4);
d4 = ((unsigned long long) h0 * r4) + ((unsigned long long) h1 * r3) +
((unsigned long long) h2 * r2) + ((unsigned long long) h3 * r1) +
((unsigned long long) h4 * r0);
/* (partial) h %= p */
c = (unsigned long)(d0 >> 26); h0 = (unsigned long)d0 & 0x3ffffff;
d1 += c; c = (unsigned long)(d1 >> 26); h1 = (unsigned long)d1 & 0x3ffffff;
d2 += c; c = (unsigned long)(d2 >> 26); h2 = (unsigned long)d2 & 0x3ffffff;
d3 += c; c = (unsigned long)(d3 >> 26); h3 = (unsigned long)d3 & 0x3ffffff;
d4 += c; c = (unsigned long)(d4 >> 26); h4 = (unsigned long)d4 & 0x3ffffff;
h0 += c * 5; c = (h0 >> 26); h0 = h0 & 0x3ffffff;
h1 += c;
/* (partial) h %= p */
c = (unsigned long) (d0 >> 26);
h0 = (unsigned long) d0 & 0x3ffffff;
d1 += c;
c = (unsigned long) (d1 >> 26);
h1 = (unsigned long) d1 & 0x3ffffff;
d2 += c;
c = (unsigned long) (d2 >> 26);
h2 = (unsigned long) d2 & 0x3ffffff;
d3 += c;
c = (unsigned long) (d3 >> 26);
h3 = (unsigned long) d3 & 0x3ffffff;
d4 += c;
c = (unsigned long) (d4 >> 26);
h4 = (unsigned long) d4 & 0x3ffffff;
h0 += c * 5;
c = (h0 >> 26);
h0 = h0 & 0x3ffffff;
h1 += c;
m += poly1305_block_size;
bytes -= poly1305_block_size;
}
m += poly1305_block_size;
bytes -= poly1305_block_size;
}
st->h[0] = h0;
st->h[1] = h1;
st->h[2] = h2;
st->h[3] = h3;
st->h[4] = h4;
st->h[0] = h0;
st->h[1] = h1;
st->h[2] = h2;
st->h[3] = h3;
st->h[4] = h4;
}
static POLY1305_NOINLINE void
poly1305_finish(poly1305_state_internal_t *st, unsigned char mac[16])
{
unsigned long h0,h1,h2,h3,h4,c;
unsigned long g0,g1,g2,g3,g4;
unsigned long long f;
unsigned long mask;
unsigned long h0, h1, h2, h3, h4, c;
unsigned long g0, g1, g2, g3, g4;
unsigned long long f;
unsigned long mask;
/* process the remaining block */
if (st->leftover) {
unsigned long long i = st->leftover;
st->buffer[i++] = 1;
for (; i < poly1305_block_size; i++)
st->buffer[i] = 0;
st->final = 1;
poly1305_blocks(st, st->buffer, poly1305_block_size);
/* process the remaining block */
if (st->leftover) {
unsigned long long i = st->leftover;
st->buffer[i++] = 1;
for (; i < poly1305_block_size; i++) {
st->buffer[i] = 0;
}
st->final = 1;
poly1305_blocks(st, st->buffer, poly1305_block_size);
}
/* fully carry h */
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
h3 = st->h[3];
h4 = st->h[4];
/* fully carry h */
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
h3 = st->h[3];
h4 = st->h[4];
c = h1 >> 26; h1 = h1 & 0x3ffffff;
h2 += c; c = h2 >> 26; h2 = h2 & 0x3ffffff;
h3 += c; c = h3 >> 26; h3 = h3 & 0x3ffffff;
h4 += c; c = h4 >> 26; h4 = h4 & 0x3ffffff;
h0 += c * 5; c = h0 >> 26; h0 = h0 & 0x3ffffff;
h1 += c;
c = h1 >> 26;
h1 = h1 & 0x3ffffff;
h2 += c;
c = h2 >> 26;
h2 = h2 & 0x3ffffff;
h3 += c;
c = h3 >> 26;
h3 = h3 & 0x3ffffff;
h4 += c;
c = h4 >> 26;
h4 = h4 & 0x3ffffff;
h0 += c * 5;
c = h0 >> 26;
h0 = h0 & 0x3ffffff;
h1 += c;
/* compute h + -p */
g0 = h0 + 5; c = g0 >> 26; g0 &= 0x3ffffff;
g1 = h1 + c; c = g1 >> 26; g1 &= 0x3ffffff;
g2 = h2 + c; c = g2 >> 26; g2 &= 0x3ffffff;
g3 = h3 + c; c = g3 >> 26; g3 &= 0x3ffffff;
g4 = h4 + c - (1UL << 26);
/* compute h + -p */
g0 = h0 + 5;
c = g0 >> 26;
g0 &= 0x3ffffff;
g1 = h1 + c;
c = g1 >> 26;
g1 &= 0x3ffffff;
g2 = h2 + c;
c = g2 >> 26;
g2 &= 0x3ffffff;
g3 = h3 + c;
c = g3 >> 26;
g3 &= 0x3ffffff;
g4 = h4 + c - (1UL << 26);
/* select h if h < p, or h + -p if h >= p */
mask = (g4 >> ((sizeof(unsigned long) * 8) - 1)) - 1;
g0 &= mask;
g1 &= mask;
g2 &= mask;
g3 &= mask;
g4 &= mask;
mask = ~mask;
h0 = (h0 & mask) | g0;
h1 = (h1 & mask) | g1;
h2 = (h2 & mask) | g2;
h3 = (h3 & mask) | g3;
h4 = (h4 & mask) | g4;
/* select h if h < p, or h + -p if h >= p */
mask = (g4 >> ((sizeof(unsigned long) * 8) - 1)) - 1;
g0 &= mask;
g1 &= mask;
g2 &= mask;
g3 &= mask;
g4 &= mask;
mask = ~mask;
/* h = h % (2^128) */
h0 = ((h0 ) | (h1 << 26)) & 0xffffffff;
h1 = ((h1 >> 6) | (h2 << 20)) & 0xffffffff;
h2 = ((h2 >> 12) | (h3 << 14)) & 0xffffffff;
h3 = ((h3 >> 18) | (h4 << 8)) & 0xffffffff;
h0 = (h0 & mask) | g0;
h1 = (h1 & mask) | g1;
h2 = (h2 & mask) | g2;
h3 = (h3 & mask) | g3;
h4 = (h4 & mask) | g4;
/* mac = (h + pad) % (2^128) */
f = (unsigned long long)h0 + st->pad[0] ; h0 = (unsigned long)f;
f = (unsigned long long)h1 + st->pad[1] + (f >> 32); h1 = (unsigned long)f;
f = (unsigned long long)h2 + st->pad[2] + (f >> 32); h2 = (unsigned long)f;
f = (unsigned long long)h3 + st->pad[3] + (f >> 32); h3 = (unsigned long)f;
/* h = h % (2^128) */
h0 = ((h0) | (h1 << 26)) & 0xffffffff;
h1 = ((h1 >> 6) | (h2 << 20)) & 0xffffffff;
h2 = ((h2 >> 12) | (h3 << 14)) & 0xffffffff;
h3 = ((h3 >> 18) | (h4 << 8)) & 0xffffffff;
STORE32_LE(mac + 0, h0);
STORE32_LE(mac + 4, h1);
STORE32_LE(mac + 8, h2);
STORE32_LE(mac + 12, h3);
/* mac = (h + pad) % (2^128) */
f = (unsigned long long) h0 + st->pad[0];
h0 = (unsigned long) f;
f = (unsigned long long) h1 + st->pad[1] + (f >> 32);
h1 = (unsigned long) f;
f = (unsigned long long) h2 + st->pad[2] + (f >> 32);
h2 = (unsigned long) f;
f = (unsigned long long) h3 + st->pad[3] + (f >> 32);
h3 = (unsigned long) f;
/* zero out the state */
sodium_memzero((void *)st, sizeof *st);
STORE32_LE(mac + 0, h0);
STORE32_LE(mac + 4, h1);
STORE32_LE(mac + 8, h2);
STORE32_LE(mac + 12, h3);
/* zero out the state */
sodium_memzero((void *) st, sizeof *st);
}
@@ -1,25 +1,26 @@
/*
poly1305 implementation using 64 bit * 64 bit = 128 bit multiplication and 128 bit addition
poly1305 implementation using 64 bit * 64 bit = 128 bit multiplication
and 128 bit addition
*/
#if defined(__SIZEOF_INT128__)
typedef unsigned __int128 uint128_t;
#else
typedef unsigned uint128_t __attribute__ ((mode(TI)));
typedef unsigned uint128_t __attribute__((mode(TI)));
#endif
#define MUL(out, x, y) out = ((uint128_t)x * y)
#define MUL(out, x, y) out = ((uint128_t) x * y)
#define ADD(out, in) out += in
#define ADDLO(out, in) out += in
#define SHR(in, shift) (unsigned long long)(in >> (shift))
#define LO(in) (unsigned long long)(in)
#define SHR(in, shift) (unsigned long long) (in >> (shift))
#define LO(in) (unsigned long long) (in)
#if defined(_MSC_VER)
# define POLY1305_NOINLINE __declspec(noinline)
#define POLY1305_NOINLINE __declspec(noinline)
#elif defined(__GNUC__)
# define POLY1305_NOINLINE __attribute__ ((noinline))
#define POLY1305_NOINLINE __attribute__((noinline))
#else
# define POLY1305_NOINLINE
#define POLY1305_NOINLINE
#endif
#include "private/common.h"
@@ -28,154 +29,197 @@ typedef unsigned uint128_t __attribute__ ((mode(TI)));
/* 17 + sizeof(unsigned long long) + 8*sizeof(unsigned long long) */
typedef struct poly1305_state_internal_t {
unsigned long long r[3];
unsigned long long h[3];
unsigned long long pad[2];
unsigned long long leftover;
unsigned char buffer[poly1305_block_size];
unsigned char final;
unsigned long long r[3];
unsigned long long h[3];
unsigned long long pad[2];
unsigned long long leftover;
unsigned char buffer[poly1305_block_size];
unsigned char final;
} poly1305_state_internal_t;
static void
poly1305_init(poly1305_state_internal_t *st, const unsigned char key[32])
{
unsigned long long t0,t1;
unsigned long long t0, t1;
/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
t0 = LOAD64_LE(&key[0]);
t1 = LOAD64_LE(&key[8]);
/* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
t0 = LOAD64_LE(&key[0]);
t1 = LOAD64_LE(&key[8]);
st->r[0] = ( t0 ) & 0xffc0fffffff;
st->r[1] = ((t0 >> 44) | (t1 << 20)) & 0xfffffc0ffff;
st->r[2] = ((t1 >> 24) ) & 0x00ffffffc0f;
st->r[0] = (t0) &0xffc0fffffff;
st->r[1] = ((t0 >> 44) | (t1 << 20)) & 0xfffffc0ffff;
st->r[2] = ((t1 >> 24)) & 0x00ffffffc0f;
/* h = 0 */
st->h[0] = 0;
st->h[1] = 0;
st->h[2] = 0;
/* h = 0 */
st->h[0] = 0;
st->h[1] = 0;
st->h[2] = 0;
/* save pad for later */
st->pad[0] = LOAD64_LE(&key[16]);
st->pad[1] = LOAD64_LE(&key[24]);
/* save pad for later */
st->pad[0] = LOAD64_LE(&key[16]);
st->pad[1] = LOAD64_LE(&key[24]);
st->leftover = 0;
st->final = 0;
st->leftover = 0;
st->final = 0;
}
static void
poly1305_blocks(poly1305_state_internal_t *st, const unsigned char *m, unsigned long long bytes)
poly1305_blocks(poly1305_state_internal_t *st, const unsigned char *m,
unsigned long long bytes)
{
const unsigned long long hibit = (st->final) ? 0ULL : (1ULL << 40); /* 1 << 128 */
unsigned long long r0,r1,r2;
unsigned long long s1,s2;
unsigned long long h0,h1,h2;
unsigned long long c;
uint128_t d0,d1,d2,d;
const unsigned long long hibit =
(st->final) ? 0ULL : (1ULL << 40); /* 1 << 128 */
unsigned long long r0, r1, r2;
unsigned long long s1, s2;
unsigned long long h0, h1, h2;
unsigned long long c;
uint128_t d0, d1, d2, d;
r0 = st->r[0];
r1 = st->r[1];
r2 = st->r[2];
r0 = st->r[0];
r1 = st->r[1];
r2 = st->r[2];
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
s1 = r1 * (5 << 2);
s2 = r2 * (5 << 2);
s1 = r1 * (5 << 2);
s2 = r2 * (5 << 2);
while (bytes >= poly1305_block_size) {
unsigned long long t0,t1;
while (bytes >= poly1305_block_size) {
unsigned long long t0, t1;
/* h += m[i] */
t0 = LOAD64_LE(&m[0]);
t1 = LOAD64_LE(&m[8]);
/* h += m[i] */
t0 = LOAD64_LE(&m[0]);
t1 = LOAD64_LE(&m[8]);
h0 += (( t0 ) & 0xfffffffffff);
h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff);
h2 += (((t1 >> 24) ) & 0x3ffffffffff) | hibit;
h0 += ((t0) &0xfffffffffff);
h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff);
h2 += (((t1 >> 24)) & 0x3ffffffffff) | hibit;
/* h *= r */
MUL(d0, h0, r0); MUL(d, h1, s2); ADD(d0, d); MUL(d, h2, s1); ADD(d0, d);
MUL(d1, h0, r1); MUL(d, h1, r0); ADD(d1, d); MUL(d, h2, s2); ADD(d1, d);
MUL(d2, h0, r2); MUL(d, h1, r1); ADD(d2, d); MUL(d, h2, r0); ADD(d2, d);
/* h *= r */
MUL(d0, h0, r0);
MUL(d, h1, s2);
ADD(d0, d);
MUL(d, h2, s1);
ADD(d0, d);
MUL(d1, h0, r1);
MUL(d, h1, r0);
ADD(d1, d);
MUL(d, h2, s2);
ADD(d1, d);
MUL(d2, h0, r2);
MUL(d, h1, r1);
ADD(d2, d);
MUL(d, h2, r0);
ADD(d2, d);
/* (partial) h %= p */
c = SHR(d0, 44); h0 = LO(d0) & 0xfffffffffff;
ADDLO(d1, c); c = SHR(d1, 44); h1 = LO(d1) & 0xfffffffffff;
ADDLO(d2, c); c = SHR(d2, 42); h2 = LO(d2) & 0x3ffffffffff;
h0 += c * 5; c = (h0 >> 44); h0 = h0 & 0xfffffffffff;
h1 += c;
/* (partial) h %= p */
c = SHR(d0, 44);
h0 = LO(d0) & 0xfffffffffff;
ADDLO(d1, c);
c = SHR(d1, 44);
h1 = LO(d1) & 0xfffffffffff;
ADDLO(d2, c);
c = SHR(d2, 42);
h2 = LO(d2) & 0x3ffffffffff;
h0 += c * 5;
c = (h0 >> 44);
h0 = h0 & 0xfffffffffff;
h1 += c;
m += poly1305_block_size;
bytes -= poly1305_block_size;
}
m += poly1305_block_size;
bytes -= poly1305_block_size;
}
st->h[0] = h0;
st->h[1] = h1;
st->h[2] = h2;
st->h[0] = h0;
st->h[1] = h1;
st->h[2] = h2;
}
static POLY1305_NOINLINE void
poly1305_finish(poly1305_state_internal_t *st, unsigned char mac[16])
{
unsigned long long h0,h1,h2,c;
unsigned long long g0,g1,g2;
unsigned long long t0,t1;
unsigned long long h0, h1, h2, c;
unsigned long long g0, g1, g2;
unsigned long long t0, t1;
/* process the remaining block */
if (st->leftover) {
unsigned long long i = st->leftover;
st->buffer[i] = 1;
for (i = i + 1; i < poly1305_block_size; i++)
st->buffer[i] = 0;
st->final = 1;
poly1305_blocks(st, st->buffer, poly1305_block_size);
/* process the remaining block */
if (st->leftover) {
unsigned long long i = st->leftover;
st->buffer[i] = 1;
for (i = i + 1; i < poly1305_block_size; i++) {
st->buffer[i] = 0;
}
st->final = 1;
poly1305_blocks(st, st->buffer, poly1305_block_size);
}
/* fully carry h */
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
/* fully carry h */
h0 = st->h[0];
h1 = st->h[1];
h2 = st->h[2];
c = (h1 >> 44); h1 &= 0xfffffffffff;
h2 += c; c = (h2 >> 42); h2 &= 0x3ffffffffff;
h0 += c * 5; c = (h0 >> 44); h0 &= 0xfffffffffff;
h1 += c; c = (h1 >> 44); h1 &= 0xfffffffffff;
h2 += c; c = (h2 >> 42); h2 &= 0x3ffffffffff;
h0 += c * 5; c = (h0 >> 44); h0 &= 0xfffffffffff;
h1 += c;
c = (h1 >> 44);
h1 &= 0xfffffffffff;
h2 += c;
c = (h2 >> 42);
h2 &= 0x3ffffffffff;
h0 += c * 5;
c = (h0 >> 44);
h0 &= 0xfffffffffff;
h1 += c;
c = (h1 >> 44);
h1 &= 0xfffffffffff;
h2 += c;
c = (h2 >> 42);
h2 &= 0x3ffffffffff;
h0 += c * 5;
c = (h0 >> 44);
h0 &= 0xfffffffffff;
h1 += c;
/* compute h + -p */
g0 = h0 + 5; c = (g0 >> 44); g0 &= 0xfffffffffff;
g1 = h1 + c; c = (g1 >> 44); g1 &= 0xfffffffffff;
g2 = h2 + c - (1ULL << 42);
/* compute h + -p */
g0 = h0 + 5;
c = (g0 >> 44);
g0 &= 0xfffffffffff;
g1 = h1 + c;
c = (g1 >> 44);
g1 &= 0xfffffffffff;
g2 = h2 + c - (1ULL << 42);
/* select h if h < p, or h + -p if h >= p */
c = (g2 >> ((sizeof(unsigned long long) * 8) - 1)) - 1;
g0 &= c;
g1 &= c;
g2 &= c;
c = ~c;
h0 = (h0 & c) | g0;
h1 = (h1 & c) | g1;
h2 = (h2 & c) | g2;
/* select h if h < p, or h + -p if h >= p */
c = (g2 >> ((sizeof(unsigned long long) * 8) - 1)) - 1;
g0 &= c;
g1 &= c;
g2 &= c;
c = ~c;
h0 = (h0 & c) | g0;
h1 = (h1 & c) | g1;
h2 = (h2 & c) | g2;
/* h = (h + pad) */
t0 = st->pad[0];
t1 = st->pad[1];
/* h = (h + pad) */
t0 = st->pad[0];
t1 = st->pad[1];
h0 += (( t0 ) & 0xfffffffffff) ; c = (h0 >> 44); h0 &= 0xfffffffffff;
h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff) + c; c = (h1 >> 44); h1 &= 0xfffffffffff;
h2 += (((t1 >> 24) ) & 0x3ffffffffff) + c; h2 &= 0x3ffffffffff;
h0 += ((t0) &0xfffffffffff);
c = (h0 >> 44);
h0 &= 0xfffffffffff;
h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff) + c;
c = (h1 >> 44);
h1 &= 0xfffffffffff;
h2 += (((t1 >> 24)) & 0x3ffffffffff) + c;
h2 &= 0x3ffffffffff;
/* mac = h % (2^128) */
h0 = ((h0 ) | (h1 << 44));
h1 = ((h1 >> 20) | (h2 << 24));
/* mac = h % (2^128) */
h0 = ((h0) | (h1 << 44));
h1 = ((h1 >> 20) | (h2 << 24));
STORE64_LE(&mac[0], h0);
STORE64_LE(&mac[8], h1);
STORE64_LE(&mac[0], h0);
STORE64_LE(&mac[8], h1);
/* zero out the state */
sodium_memzero((void *)st, sizeof *st);
/* zero out the state */
sodium_memzero((void *) st, sizeof *st);
}
@@ -1,11 +1,12 @@
#include "crypto_onetimeauth_poly1305.h"
#include "onetimeauth_poly1305.h"
#include "crypto_onetimeauth_poly1305.h"
#include "randombytes.h"
#include "runtime.h"
#include "donna/poly1305_donna.h"
#if defined(HAVE_TI_MODE) && defined(HAVE_EMMINTRIN_H)
# include "sse2/poly1305_sse2.h"
#include "sse2/poly1305_sse2.h"
#endif
static const crypto_onetimeauth_poly1305_implementation *implementation =
@@ -21,7 +22,7 @@ crypto_onetimeauth_poly1305(unsigned char *out, const unsigned char *in,
int
crypto_onetimeauth_poly1305_verify(const unsigned char *h,
const unsigned char *in,
unsigned long long inlen,
unsigned long long inlen,
const unsigned char *k)
{
return implementation->onetimeauth_verify(h, in, inlen, k);
@@ -29,38 +30,41 @@ crypto_onetimeauth_poly1305_verify(const unsigned char *h,
int
crypto_onetimeauth_poly1305_init(crypto_onetimeauth_poly1305_state *state,
const unsigned char *key)
const unsigned char * key)
{
return implementation->onetimeauth_init(state, key);
}
int
crypto_onetimeauth_poly1305_update(crypto_onetimeauth_poly1305_state *state,
const unsigned char *in,
unsigned long long inlen)
const unsigned char * in,
unsigned long long inlen)
{
return implementation->onetimeauth_update(state, in, inlen);
}
int
crypto_onetimeauth_poly1305_final(crypto_onetimeauth_poly1305_state *state,
unsigned char *out)
unsigned char * out)
{
return implementation->onetimeauth_final(state, out);
}
size_t
crypto_onetimeauth_poly1305_bytes(void) {
crypto_onetimeauth_poly1305_bytes(void)
{
return crypto_onetimeauth_poly1305_BYTES;
}
size_t
crypto_onetimeauth_poly1305_keybytes(void) {
crypto_onetimeauth_poly1305_keybytes(void)
{
return crypto_onetimeauth_poly1305_KEYBYTES;
}
void
crypto_onetimeauth_poly1305_keygen(unsigned char k[crypto_onetimeauth_poly1305_KEYBYTES])
crypto_onetimeauth_poly1305_keygen(
unsigned char k[crypto_onetimeauth_poly1305_KEYBYTES])
{
randombytes_buf(k, crypto_onetimeauth_poly1305_KEYBYTES);
}
@@ -3,21 +3,17 @@
#define onetimeauth_poly1305_H
typedef struct crypto_onetimeauth_poly1305_implementation {
int (*onetimeauth)(unsigned char *out,
const unsigned char *in,
unsigned long long inlen,
const unsigned char *k);
int (*onetimeauth_verify)(const unsigned char *h,
const unsigned char *in,
unsigned long long inlen,
const unsigned char *k);
int (*onetimeauth)(unsigned char *out, const unsigned char *in,
unsigned long long inlen, const unsigned char *k);
int (*onetimeauth_verify)(const unsigned char *h, const unsigned char *in,
unsigned long long inlen, const unsigned char *k);
int (*onetimeauth_init)(crypto_onetimeauth_poly1305_state *state,
const unsigned char *key);
const unsigned char * key);
int (*onetimeauth_update)(crypto_onetimeauth_poly1305_state *state,
const unsigned char *in,
unsigned long long inlen);
const unsigned char * in,
unsigned long long inlen);
int (*onetimeauth_final)(crypto_onetimeauth_poly1305_state *state,
unsigned char *out);
unsigned char * out);
} crypto_onetimeauth_poly1305_implementation;
#endif