1
1

external: Add ChaCha and Poly1305 implementations from OpenSSH

Signed-off-by: Aris Adamantiadis <aris@0xbadc0de.be>
Reviewed-by: Andreas Schneider <asn@cryptomilk.org>
Этот коммит содержится в:
Aris Adamantiadis 2018-02-28 10:24:51 -06:00 коммит произвёл Andreas Schneider
родитель 72e4500da9
Коммит acc1ade74a
7 изменённых файлов: 430 добавлений и 0 удалений

Просмотреть файл

@ -56,6 +56,7 @@ check_include_file(pty.h HAVE_PTY_H)
check_include_file(utmp.h HAVE_UTMP_H)
check_include_file(termios.h HAVE_TERMIOS_H)
check_include_file(unistd.h HAVE_UNISTD_H)
check_include_file(stdint.h HAVE_STDINT_H)
check_include_file(util.h HAVE_UTIL_H)
check_include_file(libutil.h HAVE_LIBUTIL_H)
check_include_file(sys/time.h HAVE_SYS_TIME_H)

Просмотреть файл

@ -50,6 +50,9 @@
/* Define to 1 if you have the <unistd.h> header file. */
#cmakedefine HAVE_UNISTD_H 1
/* Define to 1 if you have the <stdint.h> header file. */
#cmakedefine HAVE_STDINT_H 1
/* Define to 1 if you have the <openssl/aes.h> header file. */
#cmakedefine HAVE_OPENSSL_AES_H 1

32
include/libssh/chacha.h Обычный файл
Просмотреть файл

@ -0,0 +1,32 @@
/* $OpenBSD: chacha.h,v 1.3 2014/05/02 03:27:54 djm Exp $ */
/*
chacha-merged.c version 20080118
D. J. Bernstein
Public domain.
*/
#ifndef CHACHA_H
#define CHACHA_H
struct chacha_ctx {
uint32_t input[16];
};
#define CHACHA_MINKEYLEN 16
#define CHACHA_NONCELEN 8
#define CHACHA_CTRLEN 8
#define CHACHA_STATELEN (CHACHA_NONCELEN+CHACHA_CTRLEN)
#define CHACHA_BLOCKLEN 64
void chacha_keysetup(struct chacha_ctx *x, const uint8_t *k, uint32_t kbits)
__attribute__((__bounded__(__minbytes__, 2, CHACHA_MINKEYLEN)));
void chacha_ivsetup(struct chacha_ctx *x, const uint8_t *iv, const uint8_t *ctr)
__attribute__((__bounded__(__minbytes__, 2, CHACHA_NONCELEN)))
__attribute__((__bounded__(__minbytes__, 3, CHACHA_CTRLEN)));
void chacha_encrypt_bytes(struct chacha_ctx *x, const uint8_t *m,
uint8_t *c, uint32_t bytes)
__attribute__((__bounded__(__buffer__, 2, 4)))
__attribute__((__bounded__(__buffer__, 3, 4)));
#endif /* CHACHA_H */

18
include/libssh/poly1305.h Обычный файл
Просмотреть файл

@ -0,0 +1,18 @@
/*
* Public Domain poly1305 from Andrew Moon
* poly1305-donna-unrolled.c from https://github.com/floodyberry/poly1305-donna
*/
#ifndef POLY1305_H
#define POLY1305_H
#define POLY1305_KEYLEN 32
#define POLY1305_TAGLEN 16
void poly1305_auth(uint8_t out[POLY1305_TAGLEN], const uint8_t *m, size_t inlen,
const uint8_t key[POLY1305_KEYLEN])
__attribute__((__bounded__(__minbytes__, 1, POLY1305_TAGLEN)))
__attribute__((__bounded__(__buffer__, 2, 3)))
__attribute__((__bounded__(__minbytes__, 4, POLY1305_KEYLEN)));
#endif /* POLY1305_H */

Просмотреть файл

@ -158,9 +158,11 @@ set(libssh_SRCS
wrapper.c
external/bcrypt_pbkdf.c
external/blowfish.c
external/chacha.c
external/ed25519.c
external/fe25519.c
external/ge25519.c
external/poly1305.c
external/sc25519.c
)

218
src/external/chacha.c поставляемый Обычный файл
Просмотреть файл

@ -0,0 +1,218 @@
/*
chacha-merged.c version 20080118
D. J. Bernstein
Public domain.
*/
#include <stdint.h>
#include <stddef.h>
#include <sys/types.h>
#include "libssh/chacha.h"
typedef unsigned int uint32_t;
typedef struct chacha_ctx chacha_ctx;
#define U8C(v) (v##U)
#define U32C(v) (v##U)
#define U8V(v) ((uint8_t)(v) & U8C(0xFF))
#define U32V(v) ((uint32_t)(v) & U32C(0xFFFFFFFF))
#define ROTL32(v, n) \
(U32V((v) << (n)) | ((v) >> (32 - (n))))
#define U8TO32_LITTLE(p) \
(((uint32_t)((p)[0]) ) | \
((uint32_t)((p)[1]) << 8) | \
((uint32_t)((p)[2]) << 16) | \
((uint32_t)((p)[3]) << 24))
#define U32TO8_LITTLE(p, v) \
do { \
(p)[0] = U8V((v) ); \
(p)[1] = U8V((v) >> 8); \
(p)[2] = U8V((v) >> 16); \
(p)[3] = U8V((v) >> 24); \
} while (0)
#define ROTATE(v,c) (ROTL32(v,c))
#define XOR(v,w) ((v) ^ (w))
#define PLUS(v,w) (U32V((v) + (w)))
#define PLUSONE(v) (PLUS((v),1))
#define QUARTERROUND(a,b,c,d) \
a = PLUS(a,b); d = ROTATE(XOR(d,a),16); \
c = PLUS(c,d); b = ROTATE(XOR(b,c),12); \
a = PLUS(a,b); d = ROTATE(XOR(d,a), 8); \
c = PLUS(c,d); b = ROTATE(XOR(b,c), 7);
static const char sigma[16] = "expand 32-byte k";
static const char tau[16] = "expand 16-byte k";
void
chacha_keysetup(chacha_ctx *x,const uint8_t *k,uint32_t kbits)
{
const char *constants;
x->input[4] = U8TO32_LITTLE(k + 0);
x->input[5] = U8TO32_LITTLE(k + 4);
x->input[6] = U8TO32_LITTLE(k + 8);
x->input[7] = U8TO32_LITTLE(k + 12);
if (kbits == 256) { /* recommended */
k += 16;
constants = sigma;
} else { /* kbits == 128 */
constants = tau;
}
x->input[8] = U8TO32_LITTLE(k + 0);
x->input[9] = U8TO32_LITTLE(k + 4);
x->input[10] = U8TO32_LITTLE(k + 8);
x->input[11] = U8TO32_LITTLE(k + 12);
x->input[0] = U8TO32_LITTLE(constants + 0);
x->input[1] = U8TO32_LITTLE(constants + 4);
x->input[2] = U8TO32_LITTLE(constants + 8);
x->input[3] = U8TO32_LITTLE(constants + 12);
}
void
chacha_ivsetup(chacha_ctx *x, const uint8_t *iv, const uint8_t *counter)
{
x->input[12] = counter == NULL ? 0 : U8TO32_LITTLE(counter + 0);
x->input[13] = counter == NULL ? 0 : U8TO32_LITTLE(counter + 4);
x->input[14] = U8TO32_LITTLE(iv + 0);
x->input[15] = U8TO32_LITTLE(iv + 4);
}
void
chacha_encrypt_bytes(chacha_ctx *x,const uint8_t *m,uint8_t *c,uint32_t bytes)
{
uint32_t x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15;
uint32_t j0, j1, j2, j3, j4, j5, j6, j7, j8, j9, j10, j11, j12, j13, j14, j15;
uint8_t *ctarget = NULL;
uint8_t tmp[64];
uint32_t i;
if (!bytes) return;
j0 = x->input[0];
j1 = x->input[1];
j2 = x->input[2];
j3 = x->input[3];
j4 = x->input[4];
j5 = x->input[5];
j6 = x->input[6];
j7 = x->input[7];
j8 = x->input[8];
j9 = x->input[9];
j10 = x->input[10];
j11 = x->input[11];
j12 = x->input[12];
j13 = x->input[13];
j14 = x->input[14];
j15 = x->input[15];
for (;;) {
if (bytes < 64) {
for (i = 0;i < bytes;++i) tmp[i] = m[i];
m = tmp;
ctarget = c;
c = tmp;
}
x0 = j0;
x1 = j1;
x2 = j2;
x3 = j3;
x4 = j4;
x5 = j5;
x6 = j6;
x7 = j7;
x8 = j8;
x9 = j9;
x10 = j10;
x11 = j11;
x12 = j12;
x13 = j13;
x14 = j14;
x15 = j15;
for (i = 20;i > 0;i -= 2) {
QUARTERROUND( x0, x4, x8,x12)
QUARTERROUND( x1, x5, x9,x13)
QUARTERROUND( x2, x6,x10,x14)
QUARTERROUND( x3, x7,x11,x15)
QUARTERROUND( x0, x5,x10,x15)
QUARTERROUND( x1, x6,x11,x12)
QUARTERROUND( x2, x7, x8,x13)
QUARTERROUND( x3, x4, x9,x14)
}
x0 = PLUS(x0,j0);
x1 = PLUS(x1,j1);
x2 = PLUS(x2,j2);
x3 = PLUS(x3,j3);
x4 = PLUS(x4,j4);
x5 = PLUS(x5,j5);
x6 = PLUS(x6,j6);
x7 = PLUS(x7,j7);
x8 = PLUS(x8,j8);
x9 = PLUS(x9,j9);
x10 = PLUS(x10,j10);
x11 = PLUS(x11,j11);
x12 = PLUS(x12,j12);
x13 = PLUS(x13,j13);
x14 = PLUS(x14,j14);
x15 = PLUS(x15,j15);
x0 = XOR(x0,U8TO32_LITTLE(m + 0));
x1 = XOR(x1,U8TO32_LITTLE(m + 4));
x2 = XOR(x2,U8TO32_LITTLE(m + 8));
x3 = XOR(x3,U8TO32_LITTLE(m + 12));
x4 = XOR(x4,U8TO32_LITTLE(m + 16));
x5 = XOR(x5,U8TO32_LITTLE(m + 20));
x6 = XOR(x6,U8TO32_LITTLE(m + 24));
x7 = XOR(x7,U8TO32_LITTLE(m + 28));
x8 = XOR(x8,U8TO32_LITTLE(m + 32));
x9 = XOR(x9,U8TO32_LITTLE(m + 36));
x10 = XOR(x10,U8TO32_LITTLE(m + 40));
x11 = XOR(x11,U8TO32_LITTLE(m + 44));
x12 = XOR(x12,U8TO32_LITTLE(m + 48));
x13 = XOR(x13,U8TO32_LITTLE(m + 52));
x14 = XOR(x14,U8TO32_LITTLE(m + 56));
x15 = XOR(x15,U8TO32_LITTLE(m + 60));
j12 = PLUSONE(j12);
if (!j12) {
j13 = PLUSONE(j13);
/* stopping at 2^70 bytes per nonce is user's responsibility */
}
U32TO8_LITTLE(c + 0,x0);
U32TO8_LITTLE(c + 4,x1);
U32TO8_LITTLE(c + 8,x2);
U32TO8_LITTLE(c + 12,x3);
U32TO8_LITTLE(c + 16,x4);
U32TO8_LITTLE(c + 20,x5);
U32TO8_LITTLE(c + 24,x6);
U32TO8_LITTLE(c + 28,x7);
U32TO8_LITTLE(c + 32,x8);
U32TO8_LITTLE(c + 36,x9);
U32TO8_LITTLE(c + 40,x10);
U32TO8_LITTLE(c + 44,x11);
U32TO8_LITTLE(c + 48,x12);
U32TO8_LITTLE(c + 52,x13);
U32TO8_LITTLE(c + 56,x14);
U32TO8_LITTLE(c + 60,x15);
if (bytes <= 64) {
if (bytes < 64) {
for (i = 0;i < bytes;++i) ctarget[i] = c[i];
}
x->input[12] = j12;
x->input[13] = j13;
return;
}
bytes -= 64;
c += 64;
m += 64;
}
}

156
src/external/poly1305.c поставляемый Обычный файл
Просмотреть файл

@ -0,0 +1,156 @@
/*
* Public Domain poly1305 from Andrew Moon
* poly1305-donna-unrolled.c from https://github.com/floodyberry/poly1305-donna
*/
#include "config.h"
#include <stdint.h>
#include <sys/types.h>
#include "libssh/poly1305.h"
#define mul32x32_64(a,b) ((uint64_t)(a) * (b))
#define U8TO32_LE(p) \
(((uint32_t)((p)[0])) | \
((uint32_t)((p)[1]) << 8) | \
((uint32_t)((p)[2]) << 16) | \
((uint32_t)((p)[3]) << 24))
#define U32TO8_LE(p, v) \
do { \
(p)[0] = (uint8_t)((v)); \
(p)[1] = (uint8_t)((v) >> 8); \
(p)[2] = (uint8_t)((v) >> 16); \
(p)[3] = (uint8_t)((v) >> 24); \
} while (0)
void
poly1305_auth(unsigned char out[POLY1305_TAGLEN], const unsigned char *m, size_t inlen, const unsigned char key[POLY1305_KEYLEN]) {
uint32_t t0,t1,t2,t3;
uint32_t h0,h1,h2,h3,h4;
uint32_t r0,r1,r2,r3,r4;
uint32_t s1,s2,s3,s4;
uint32_t b, nb;
size_t j;
uint64_t t[5];
uint64_t f0,f1,f2,f3;
uint32_t g0,g1,g2,g3,g4;
uint64_t c;
unsigned char mp[16];
/* clamp key */
t0 = U8TO32_LE(key+0);
t1 = U8TO32_LE(key+4);
t2 = U8TO32_LE(key+8);
t3 = U8TO32_LE(key+12);
/* precompute multipliers */
r0 = t0 & 0x3ffffff; t0 >>= 26; t0 |= t1 << 6;
r1 = t0 & 0x3ffff03; t1 >>= 20; t1 |= t2 << 12;
r2 = t1 & 0x3ffc0ff; t2 >>= 14; t2 |= t3 << 18;
r3 = t2 & 0x3f03fff; t3 >>= 8;
r4 = t3 & 0x00fffff;
s1 = r1 * 5;
s2 = r2 * 5;
s3 = r3 * 5;
s4 = r4 * 5;
/* init state */
h0 = 0;
h1 = 0;
h2 = 0;
h3 = 0;
h4 = 0;
/* full blocks */
if (inlen < 16) goto poly1305_donna_atmost15bytes;
poly1305_donna_16bytes:
m += 16;
inlen -= 16;
t0 = U8TO32_LE(m-16);
t1 = U8TO32_LE(m-12);
t2 = U8TO32_LE(m-8);
t3 = U8TO32_LE(m-4);
h0 += t0 & 0x3ffffff;
h1 += ((((uint64_t)t1 << 32) | t0) >> 26) & 0x3ffffff;
h2 += ((((uint64_t)t2 << 32) | t1) >> 20) & 0x3ffffff;
h3 += ((((uint64_t)t3 << 32) | t2) >> 14) & 0x3ffffff;
h4 += (t3 >> 8) | (1 << 24);
poly1305_donna_mul:
t[0] = mul32x32_64(h0,r0) + mul32x32_64(h1,s4) + mul32x32_64(h2,s3) + mul32x32_64(h3,s2) + mul32x32_64(h4,s1);
t[1] = mul32x32_64(h0,r1) + mul32x32_64(h1,r0) + mul32x32_64(h2,s4) + mul32x32_64(h3,s3) + mul32x32_64(h4,s2);
t[2] = mul32x32_64(h0,r2) + mul32x32_64(h1,r1) + mul32x32_64(h2,r0) + mul32x32_64(h3,s4) + mul32x32_64(h4,s3);
t[3] = mul32x32_64(h0,r3) + mul32x32_64(h1,r2) + mul32x32_64(h2,r1) + mul32x32_64(h3,r0) + mul32x32_64(h4,s4);
t[4] = mul32x32_64(h0,r4) + mul32x32_64(h1,r3) + mul32x32_64(h2,r2) + mul32x32_64(h3,r1) + mul32x32_64(h4,r0);
h0 = (uint32_t)t[0] & 0x3ffffff; c = (t[0] >> 26);
t[1] += c; h1 = (uint32_t)t[1] & 0x3ffffff; b = (uint32_t)(t[1] >> 26);
t[2] += b; h2 = (uint32_t)t[2] & 0x3ffffff; b = (uint32_t)(t[2] >> 26);
t[3] += b; h3 = (uint32_t)t[3] & 0x3ffffff; b = (uint32_t)(t[3] >> 26);
t[4] += b; h4 = (uint32_t)t[4] & 0x3ffffff; b = (uint32_t)(t[4] >> 26);
h0 += b * 5;
if (inlen >= 16) goto poly1305_donna_16bytes;
/* final bytes */
poly1305_donna_atmost15bytes:
if (!inlen) goto poly1305_donna_finish;
for (j = 0; j < inlen; j++) mp[j] = m[j];
mp[j++] = 1;
for (; j < 16; j++) mp[j] = 0;
inlen = 0;
t0 = U8TO32_LE(mp+0);
t1 = U8TO32_LE(mp+4);
t2 = U8TO32_LE(mp+8);
t3 = U8TO32_LE(mp+12);
h0 += t0 & 0x3ffffff;
h1 += ((((uint64_t)t1 << 32) | t0) >> 26) & 0x3ffffff;
h2 += ((((uint64_t)t2 << 32) | t1) >> 20) & 0x3ffffff;
h3 += ((((uint64_t)t3 << 32) | t2) >> 14) & 0x3ffffff;
h4 += (t3 >> 8);
goto poly1305_donna_mul;
poly1305_donna_finish:
b = h0 >> 26; h0 = h0 & 0x3ffffff;
h1 += b; b = h1 >> 26; h1 = h1 & 0x3ffffff;
h2 += b; b = h2 >> 26; h2 = h2 & 0x3ffffff;
h3 += b; b = h3 >> 26; h3 = h3 & 0x3ffffff;
h4 += b; b = h4 >> 26; h4 = h4 & 0x3ffffff;
h0 += b * 5; b = h0 >> 26; h0 = h0 & 0x3ffffff;
h1 += b;
g0 = h0 + 5; b = g0 >> 26; g0 &= 0x3ffffff;
g1 = h1 + b; b = g1 >> 26; g1 &= 0x3ffffff;
g2 = h2 + b; b = g2 >> 26; g2 &= 0x3ffffff;
g3 = h3 + b; b = g3 >> 26; g3 &= 0x3ffffff;
g4 = h4 + b - (1 << 26);
b = (g4 >> 31) - 1;
nb = ~b;
h0 = (h0 & nb) | (g0 & b);
h1 = (h1 & nb) | (g1 & b);
h2 = (h2 & nb) | (g2 & b);
h3 = (h3 & nb) | (g3 & b);
h4 = (h4 & nb) | (g4 & b);
f0 = ((h0 ) | (h1 << 26)) + (uint64_t)U8TO32_LE(&key[16]);
f1 = ((h1 >> 6) | (h2 << 20)) + (uint64_t)U8TO32_LE(&key[20]);
f2 = ((h2 >> 12) | (h3 << 14)) + (uint64_t)U8TO32_LE(&key[24]);
f3 = ((h3 >> 18) | (h4 << 8)) + (uint64_t)U8TO32_LE(&key[28]);
U32TO8_LE(&out[ 0], f0); f1 += (f0 >> 32);
U32TO8_LE(&out[ 4], f1); f2 += (f1 >> 32);
U32TO8_LE(&out[ 8], f2); f3 += (f2 >> 32);
U32TO8_LE(&out[12], f3);
}