askiiart/tunsafe-clang15
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
tunsafe-clang15/crypto/blake2s_sse.cpp
Ludvig Strigeus cf92ac7a0c Updates for TunSafe 1.4-rc1 
1.Subfolders in the Config/ directory now show up as submenus.
2.Added a way to run TunSafe as a Windows Service.
  Foreground Mode: The service will disconnect when TunSafe closes.
  Background Mode: The service will stay connected in the background.
  No longer required to run the TunSafe client as Admin as long as
  the service is running.
3.New config setting [Interface].ExcludedIPs to configure IPs that
  should not be routed through TunSafe.
4.Can now automatically start TunSafe when Windows starts
5.New UI with tabs and graphs
6.Cache DNS queries to ensure DNS will succeed if connection fails
7.Recreate tray icon when explorer.exe restarts
8.Renamed window title to TunSafe instead of TunSafe VPN Client
9.Main window is now resizable
10.Disallow roaming endpoint when using AllowedIPs=0.0.0.0/0
   Only the original endpoint is added in the routing table so
   this would result in an endless loop of packets.
11.Display approximate Wireguard framing overhead in stats
12.Preparations for protocol handling with multiple threads
13.Delete the routes we made when disconnecting
14.Fix error message about unable to delete a route when connecting
2018-08-12 03:30:06 +02:00

399 lines
11 KiB
C++
Executable file
Raw Blame History

/*
BLAKE2 reference source code package - optimized C implementations
Copyright 2012, Samuel Neves <sneves@dei.uc.pt>. You may use this under the
terms of the CC0, the OpenSSL Licence, or the Apache Public License 2.0, at
your option. The terms of these licenses can be found at:
- CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
- OpenSSL license : https://www.openssl.org/source/license.html
- Apache 2.0 : http://www.apache.org/licenses/LICENSE-2.0
More information about the BLAKE2 hash function can be found at
https://blake2.net.
*/
#include "stdafx.h"
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "blake2s.h"
#include "crypto_ops.h"
#include <emmintrin.h>
#if defined(HAVE_SSSE3)
#include <tmmintrin.h>
#endif
#if defined(HAVE_SSE41)
#include <smmintrin.h>
#endif
#if defined(HAVE_AVX)
#include <immintrin.h>
#endif
#if defined(HAVE_XOP)
#include <x86intrin.h>
#endif
#include "blake2s-round.h"
#if !defined(__cplusplus) && (!defined(__STDC_VERSION__) || __STDC_VERSION__ < 199901L)
#if defined(_MSC_VER)
#define BLAKE2_INLINE __inline
#elif defined(__GNUC__)
#define BLAKE2_INLINE __inline__
#else
#define BLAKE2_INLINE
#endif
#else
#define BLAKE2_INLINE inline
#endif
static BLAKE2_INLINE uint32_t load32(const void *src) {
#if defined(ARCH_CPU_LITTLE_ENDIAN)
uint32_t w;
memcpy(&w, src, sizeof w);
return w;
#else
const uint8_t *p = (const uint8_t *)src;
return ((uint32_t)(p[0]) << 0) |
((uint32_t)(p[1]) << 8) |
((uint32_t)(p[2]) << 16) |
((uint32_t)(p[3]) << 24);
#endif
}
static BLAKE2_INLINE void store32(void *dst, uint32_t w) {
#if defined(ARCH_CPU_LITTLE_ENDIAN)
memcpy(dst, &w, sizeof w);
#else
uint8_t *p = (uint8_t *)dst;
p[0] = (uint8_t)(w >> 0);
p[1] = (uint8_t)(w >> 8);
p[2] = (uint8_t)(w >> 16);
p[3] = (uint8_t)(w >> 24);
#endif
}
static const uint32_t blake2s_IV[8] =
{
0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL,
0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL
};
/* Some helper functions */
static void blake2s_set_lastnode( blake2s_state *S )
{
S->f[1] = (uint32_t)-1;
}
static int blake2s_is_lastblock( const blake2s_state *S )
{
return S->f[0] != 0;
}
static void blake2s_set_lastblock( blake2s_state *S )
{
if( S->last_node ) blake2s_set_lastnode( S );
S->f[0] = (uint32_t)-1;
}
static void blake2s_increment_counter( blake2s_state *S, const uint32_t inc )
{
uint64_t t = ( ( uint64_t )S->t[1] << 32 ) | S->t[0];
t += inc;
S->t[0] = ( uint32_t )( t >> 0 );
S->t[1] = ( uint32_t )( t >> 32 );
}
/* init2 xors IV with input parameter block */
#if 0
void blake2s_init_param( blake2s_state *S, const blake2s_param *P )
{
size_t i;
/*blake2s_init0( S ); */
const uint8_t * v = ( const uint8_t * )( blake2s_IV );
const uint8_t * p = ( const uint8_t * )( P );
uint8_t * h = ( uint8_t * )( S->h );
/* IV XOR ParamBlock */
memset( S, 0, sizeof( blake2s_state ) );
for( i = 0; i < BLAKE2S_OUTBYTES; ++i ) h[i] = v[i] ^ p[i];
S->outlen = P->digest_length;
}
/* Some sort of default parameter block initialization, for sequential blake2s */
void blake2s_init( blake2s_state *S, size_t outlen )
{
blake2s_param P[1];
assert(outlen && outlen <= BLAKE2S_OUTBYTES);
P->digest_length = (uint8_t)outlen;
P->key_length = 0;
P->fanout = 1;
P->depth = 1;
store32( &P->leaf_length, 0 );
store32( &P->node_offset, 0 );
store16( &P->xof_length, 0 );
P->node_depth = 0;
P->inner_length = 0;
/* memset(P->reserved, 0, sizeof(P->reserved) ); */
memset( P->salt, 0, sizeof( P->salt ) );
memset( P->personal, 0, sizeof( P->personal ) );
blake2s_init_param( S, P );
}
int blake2s_init_key( blake2s_state *S, size_t outlen, const void *key, size_t keylen )
{
blake2s_param P[1];
/* Move interval verification here? */
if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1;
if ( ( !key ) || ( !keylen ) || keylen > BLAKE2S_KEYBYTES ) return -1;
P->digest_length = (uint8_t)outlen;
P->key_length = (uint8_t)keylen;
P->fanout = 1;
P->depth = 1;
store32( &P->leaf_length, 0 );
store32( &P->node_offset, 0 );
store16( &P->xof_length, 0 );
P->node_depth = 0;
P->inner_length = 0;
/* memset(P->reserved, 0, sizeof(P->reserved) ); */
memset( P->salt, 0, sizeof( P->salt ) );
memset( P->personal, 0, sizeof( P->personal ) );
if( blake2s_init_param( S, P ) < 0 )
return -1;
{
uint8_t block[BLAKE2S_BLOCKBYTES];
memset( block, 0, BLAKE2S_BLOCKBYTES );
memcpy( block, key, keylen );
blake2s_update( S, block, BLAKE2S_BLOCKBYTES );
memzero_crypto( block, BLAKE2S_BLOCKBYTES ); /* Burn the key from stack */
}
return 0;
}
#endif
void blake2s_compress_sse( blake2s_state *S, const uint8_t block[BLAKE2S_BLOCKBYTES] )
{
__m128i row1, row2, row3, row4;
__m128i buf1, buf2, buf3, buf4;
#if defined(HAVE_SSE41)
__m128i t0, t1;
#if !defined(HAVE_XOP)
__m128i t2;
#endif
#endif
__m128i ff0, ff1;
#if defined(HAVE_SSSE3) && !defined(HAVE_XOP)
const __m128i r8 = _mm_set_epi8( 12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1 );
const __m128i r16 = _mm_set_epi8( 13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2 );
#endif
#if defined(HAVE_SSE41)
const __m128i m0 = LOADU( block + 00 );
const __m128i m1 = LOADU( block + 16 );
const __m128i m2 = LOADU( block + 32 );
const __m128i m3 = LOADU( block + 48 );
#else
const uint32_t m0 = load32(block + 0 * sizeof(uint32_t));
const uint32_t m1 = load32(block + 1 * sizeof(uint32_t));
const uint32_t m2 = load32(block + 2 * sizeof(uint32_t));
const uint32_t m3 = load32(block + 3 * sizeof(uint32_t));
const uint32_t m4 = load32(block + 4 * sizeof(uint32_t));
const uint32_t m5 = load32(block + 5 * sizeof(uint32_t));
const uint32_t m6 = load32(block + 6 * sizeof(uint32_t));
const uint32_t m7 = load32(block + 7 * sizeof(uint32_t));
const uint32_t m8 = load32(block + 8 * sizeof(uint32_t));
const uint32_t m9 = load32(block + 9 * sizeof(uint32_t));
const uint32_t m10 = load32(block + 10 * sizeof(uint32_t));
const uint32_t m11 = load32(block + 11 * sizeof(uint32_t));
const uint32_t m12 = load32(block + 12 * sizeof(uint32_t));
const uint32_t m13 = load32(block + 13 * sizeof(uint32_t));
const uint32_t m14 = load32(block + 14 * sizeof(uint32_t));
const uint32_t m15 = load32(block + 15 * sizeof(uint32_t));
#endif
row1 = ff0 = LOADU( &S->h[0] );
row2 = ff1 = LOADU( &S->h[4] );
row3 = _mm_loadu_si128( (__m128i const *)&blake2s_IV[0] );
row4 = _mm_xor_si128( _mm_loadu_si128( (__m128i const *)&blake2s_IV[4] ), LOADU( &S->t[0] ) );
ROUND( 0 );
ROUND( 1 );
ROUND( 2 );
ROUND( 3 );
ROUND( 4 );
ROUND( 5 );
ROUND( 6 );
ROUND( 7 );
ROUND( 8 );
ROUND( 9 );
STOREU( &S->h[0], _mm_xor_si128( ff0, _mm_xor_si128( row1, row3 ) ) );
STOREU( &S->h[4], _mm_xor_si128( ff1, _mm_xor_si128( row2, row4 ) ) );
}
#if 0
int blake2s_update( blake2s_state *S, const void *pin, size_t inlen )
{
const unsigned char * in = (const unsigned char *)pin;
if( inlen > 0 )
{
size_t left = S->buflen;
size_t fill = BLAKE2S_BLOCKBYTES - left;
if( inlen > fill )
{
S->buflen = 0;
memcpy( S->buf + left, in, fill ); /* Fill buffer */
blake2s_increment_counter( S, BLAKE2S_BLOCKBYTES );
blake2s_compress( S, S->buf ); /* Compress */
in += fill; inlen -= fill;
while(inlen > BLAKE2S_BLOCKBYTES) {
blake2s_increment_counter(S, BLAKE2S_BLOCKBYTES);
blake2s_compress( S, in );
in += BLAKE2S_BLOCKBYTES;
inlen -= BLAKE2S_BLOCKBYTES;
}
}
memcpy( S->buf + S->buflen, in, inlen );
S->buflen += inlen;
}
return 0;
}
int blake2s_final( blake2s_state *S, void *out, size_t outlen )
{
uint8_t buffer[BLAKE2S_OUTBYTES] = {0};
size_t i;
if( out == NULL || outlen < S->outlen )
return -1;
if( blake2s_is_lastblock( S ) )
return -1;
blake2s_increment_counter( S, (uint32_t)S->buflen );
blake2s_set_lastblock( S );
memset( S->buf + S->buflen, 0, BLAKE2S_BLOCKBYTES - S->buflen ); /* Padding */
blake2s_compress( S, S->buf );
for( i = 0; i < 8; ++i ) /* Output full hash to temp buffer */
store32( buffer + sizeof( S->h[i] ) * i, S->h[i] );
memcpy( out, buffer, S->outlen );
memzero_crypto( buffer, sizeof(buffer) );
return 0;
}
/* inlen, at least, should be uint64_t. Others can be size_t. */
int blake2s( void *out, size_t outlen, const void *in, size_t inlen, const void *key, size_t keylen )
{
blake2s_state S[1];
/* Verify parameters */
if ( NULL == in && inlen > 0 ) return -1;
if ( NULL == out ) return -1;
if ( NULL == key && keylen > 0) return -1;
if( !outlen || outlen > BLAKE2S_OUTBYTES ) return -1;
if( keylen > BLAKE2S_KEYBYTES ) return -1;
if( keylen > 0 )
{
if( blake2s_init_key( S, outlen, key, keylen ) < 0 ) return -1;
}
else
{
if( blake2s_init( S, outlen ) < 0 ) return -1;
}
blake2s_update( S, ( const uint8_t * )in, inlen );
blake2s_final( S, out, outlen );
return 0;
}
#endif
#if defined(SUPERCOP)
int crypto_hash( unsigned char *out, unsigned char *in, unsigned long long inlen )
{
return blake2s( out, BLAKE2S_OUTBYTES, in, inlen, NULL, 0 );
}
#endif
#if defined(BLAKE2S_SELFTEST)
#include <string.h>
#include "blake2-kat.h"
int main( void )
{
uint8_t key[BLAKE2S_KEYBYTES];
uint8_t buf[BLAKE2_KAT_LENGTH];
size_t i, step;
for( i = 0; i < BLAKE2S_KEYBYTES; ++i )
key[i] = ( uint8_t )i;
for( i = 0; i < BLAKE2_KAT_LENGTH; ++i )
buf[i] = ( uint8_t )i;
/* Test simple API */
for( i = 0; i < BLAKE2_KAT_LENGTH; ++i )
{
uint8_t hash[BLAKE2S_OUTBYTES];
blake2s( hash, BLAKE2S_OUTBYTES, buf, i, key, BLAKE2S_KEYBYTES );
if( 0 != memcmp( hash, blake2s_keyed_kat[i], BLAKE2S_OUTBYTES ) )
{
goto fail;
}
}
/* Test streaming API */
for(step = 1; step < BLAKE2S_BLOCKBYTES; ++step) {
for (i = 0; i < BLAKE2_KAT_LENGTH; ++i) {
uint8_t hash[BLAKE2S_OUTBYTES];
blake2s_state S;
uint8_t * p = buf;
size_t mlen = i;
int err = 0;
if( (err = blake2s_init_key(&S, BLAKE2S_OUTBYTES, key, BLAKE2S_KEYBYTES)) < 0 ) {
goto fail;
}
while (mlen >= step) {
if ( (err = blake2s_update(&S, p, step)) < 0 ) {
goto fail;
}
mlen -= step;
p += step;
}
if ( (err = blake2s_update(&S, p, mlen)) < 0) {
goto fail;
}
if ( (err = blake2s_final(&S, hash, BLAKE2S_OUTBYTES)) < 0) {
goto fail;
}
if (0 != memcmp(hash, blake2s_keyed_kat[i], BLAKE2S_OUTBYTES)) {
goto fail;
}
}
}
puts( "ok" );
return 0;
fail:
puts("error");
return -1;
}
#endif
Reference in a new issue View git blame Copy permalink