nihilist/VeraCrypt
1
0
Fork 0
mirror of https://github.com/veracrypt/VeraCrypt synced 2024-11-30 23:13:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

SIMD speed optimization for Kuznyechik cipher implementation (up to 2x speedup). Based on https://github.com/aprelev/libgost15.

Browse Source
This commit is contained in:
Mounir IDRASSI 2017-11-27 09:10:17 +02:00
parent 685fad2d5d
commit f53eb8e260
No known key found for this signature in database
GPG Key ID: DD0C382D5FCFB8FC
10 changed files with 9835 additions and 168 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
src/Common/Crypto.c
View File

@ -254,6 +254,20 @@ void EncipherBlocks (int cipher, void *dataPtr, void *ks, size_t blockCount)
else if (cipher == CAMELLIA) { else if (cipher == CAMELLIA) {
camellia_encrypt_blocks(ks, data, data, (uint32) blockCount); camellia_encrypt_blocks(ks, data, data, (uint32) blockCount);
} }
#endif
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined (_UEFI)
else if (cipher == KUZNYECHIK
&& HasSSE2()
#if defined (TC_WINDOWS_DRIVER) && !defined (_WIN64)
&& (blockCount >= 4) && NT_SUCCESS (KeSaveFloatingPointState (&floatingPointState))
#endif
)
{
kuznyechik_encrypt_blocks (data, data, blockCount, ks);
#if defined (TC_WINDOWS_DRIVER) && !defined (_WIN64)
KeRestoreFloatingPointState (&floatingPointState);
#endif
}
#endif #endif
else if (cipher == GOST89) { else if (cipher == GOST89) {
gost_encrypt(data, data, ks, (int)blockCount); gost_encrypt(data, data, ks, (int)blockCount);
@ -357,6 +371,20 @@ void DecipherBlocks (int cipher, void *dataPtr, void *ks, size_t blockCount)
else if (cipher == CAMELLIA) { else if (cipher == CAMELLIA) {
camellia_decrypt_blocks(ks, data, data, (uint32) blockCount); camellia_decrypt_blocks(ks, data, data, (uint32) blockCount);
} }
#endif
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined (_UEFI)
else if (cipher == KUZNYECHIK
&& HasSSE2()
#if defined (TC_WINDOWS_DRIVER) && !defined (_WIN64)
&& (blockCount >= 4) && NT_SUCCESS (KeSaveFloatingPointState (&floatingPointState))
#endif
)
{
kuznyechik_decrypt_blocks (data, data, blockCount, ks);
#if defined (TC_WINDOWS_DRIVER) && !defined (_WIN64)
KeRestoreFloatingPointState (&floatingPointState);
#endif
}
#endif #endif
else if (cipher == GOST89) { else if (cipher == GOST89) {
gost_decrypt(data, data, ks, (int)blockCount); gost_decrypt(data, data, ks, (int)blockCount);
@ -429,6 +457,7 @@ BOOL CipherSupportsIntraDataUnitParallelization (int cipher)
|| (cipher == GOST89) || (cipher == GOST89)
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined (_UEFI) #if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined (_UEFI)
|| (cipher == SERPENT && HasSSE2()) || (cipher == SERPENT && HasSSE2())
|| (cipher == KUZNYECHIK && HasSSE2())
#endif #endif
#if CRYPTOPP_BOOL_X64 #if CRYPTOPP_BOOL_X64
|| (cipher == TWOFISH) || (cipher == TWOFISH)

1
src/Crypto/Crypto.vcxproj
View File

@ -218,6 +218,7 @@
<ClCompile Include="cpu.c" /> <ClCompile Include="cpu.c" />
<ClCompile Include="GostCipher.c" /> <ClCompile Include="GostCipher.c" />
<ClCompile Include="kuznyechik.c" /> <ClCompile Include="kuznyechik.c" />
<ClCompile Include="kuznyechik_simd.c" />
<ClCompile Include="Rmd160.c" /> <ClCompile Include="Rmd160.c" />
<ClCompile Include="SerpentFast.c" /> <ClCompile Include="SerpentFast.c" />
<ClCompile Include="SerpentFast_simd.cpp" /> <ClCompile Include="SerpentFast_simd.cpp" />

3
src/Crypto/Crypto.vcxproj.filters
View File

@ -54,6 +54,9 @@
<ClCompile Include="Camellia.c"> <ClCompile Include="Camellia.c">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
<ClCompile Include="kuznyechik_simd.c">
<Filter>Source Files</Filter>
</ClCompile>
</ItemGroup> </ItemGroup>
<ItemGroup> <ItemGroup>
<ClInclude Include="Aes.h"> <ClInclude Include="Aes.h">

1
src/Crypto/Sources
View File

@ -35,6 +35,7 @@ SOURCES = \
GostCipher.c \ GostCipher.c \
Streebog.c \ Streebog.c \
kuznyechik.c \ kuznyechik.c \
kuznyechik_simd.c \
Whirlpool.c \ Whirlpool.c \
Camellia.c \ Camellia.c \
Camellia_$(TC_ARCH).S \ Camellia_$(TC_ARCH).S \

393
src/Crypto/kuznyechik.c
View File

@ -4,14 +4,21 @@ and released into public domain.
*/ */
#include "kuznyechik.h" #include "kuznyechik.h"
// #include <memory.h> #include "cpu.h"
// #include <algorithm> #include "misc.h"
// #include "portability.h"
#ifdef _MSC_VER #ifdef _MSC_VER
#define inline __forceinline #define inline __forceinline
#endif #endif
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
void kuznyechik_set_key_simd(const byte* key, kuznyechik_kds *kds);
void kuznyechik_encrypt_block_simd(byte* out, const byte* in, kuznyechik_kds* kds);
void kuznyechik_encrypt_blocks_simd(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds);
void kuznyechik_decrypt_block_simd(byte* out, const byte* in, kuznyechik_kds* kds);
void kuznyechik_decrypt_blocks_simd(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds);
#endif
//#define CPPCRYPTO_DEBUG //#define CPPCRYPTO_DEBUG
static const byte S[256] = { static const byte S[256] = {
@ -2136,199 +2143,257 @@ and released into public domain.
{LL(0x45aba4f6433784cc), LL(0x1dffec46132c75de)}, {LL(0x4e257c42d5ada17e), LL(0x1e80a3e223281c39)}, {LL(0xf65f342ea7db0310), LL(0x1f14273f33953b64)}, {LL(0x619b141e58d8a75e), LL(0x20a8ed9c45c16af1)} {LL(0x45aba4f6433784cc), LL(0x1dffec46132c75de)}, {LL(0x4e257c42d5ada17e), LL(0x1e80a3e223281c39)}, {LL(0xf65f342ea7db0310), LL(0x1f14273f33953b64)}, {LL(0x619b141e58d8a75e), LL(0x20a8ed9c45c16af1)}
}; };
static inline void LS(uint64 x1, uint64 x2, uint64* t1, uint64* t2) #define LS(x1,x2,t1,t2) { \
{ t1 = T[0][(byte)(x1)][0] ^ T[1][(byte)(x1 >> 8)][0] ^ T[2][(byte)(x1 >> 16)][0] ^ T[3][(byte)(x1 >> 24)][0] ^ T[4][(byte)(x1 >> 32)][0] ^ T[5][(byte)(x1 >> 40)][0] ^ \
*t1 = T[0][(byte)(x1)][0] ^ T[1][(byte)(x1 >> 8)][0] ^ T[2][(byte)(x1 >> 16)][0] ^ T[3][(byte)(x1 >> 24)][0] ^ T[4][(byte)(x1 >> 32)][0] ^ T[5][(byte)(x1 >> 40)][0] ^ T[6][(byte)(x1 >> 48)][0] ^ T[7][(byte)(x1 >> 56)][0] ^ T[8][(byte)(x2)][0] ^ T[9][(byte)(x2 >> 8)][0] ^ T[10][(byte)(x2 >> 16)][0] ^ T[11][(byte)(x2 >> 24)][0] ^ \
T[6][(byte)(x1 >> 48)][0] ^ T[7][(byte)(x1 >> 56)][0] ^ T[8][(byte)(x2)][0] ^ T[9][(byte)(x2 >> 8)][0] ^ T[10][(byte)(x2 >> 16)][0] ^ T[11][(byte)(x2 >> 24)][0] ^ T[12][(byte)(x2 >> 32)][0] ^ T[13][(byte)(x2 >> 40)][0] ^ T[14][(byte)(x2 >> 48)][0] ^ T[15][(byte)(x2 >> 56)][0]; \
T[12][(byte)(x2 >> 32)][0] ^ T[13][(byte)(x2 >> 40)][0] ^ T[14][(byte)(x2 >> 48)][0] ^ T[15][(byte)(x2 >> 56)][0]; t2 = T[0][(byte)(x1)][1] ^ T[1][(byte)(x1 >> 8)][1] ^ T[2][(byte)(x1 >> 16)][1] ^ T[3][(byte)(x1 >> 24)][1] ^ T[4][(byte)(x1 >> 32)][1] ^ T[5][(byte)(x1 >> 40)][1] ^ \
*t2 = T[0][(byte)(x1)][1] ^ T[1][(byte)(x1 >> 8)][1] ^ T[2][(byte)(x1 >> 16)][1] ^ T[3][(byte)(x1 >> 24)][1] ^ T[4][(byte)(x1 >> 32)][1] ^ T[5][(byte)(x1 >> 40)][1] ^ T[6][(byte)(x1 >> 48)][1] ^ T[7][(byte)(x1 >> 56)][1] ^ T[8][(byte)(x2)][1] ^ T[9][(byte)(x2 >> 8)][1] ^ T[10][(byte)(x2 >> 16)][1] ^ T[11][(byte)(x2 >> 24)][1] ^ \
T[6][(byte)(x1 >> 48)][1] ^ T[7][(byte)(x1 >> 56)][1] ^ T[8][(byte)(x2)][1] ^ T[9][(byte)(x2 >> 8)][1] ^ T[10][(byte)(x2 >> 16)][1] ^ T[11][(byte)(x2 >> 24)][1] ^ T[12][(byte)(x2 >> 32)][1] ^ T[13][(byte)(x2 >> 40)][1] ^ T[14][(byte)(x2 >> 48)][1] ^ T[15][(byte)(x2 >> 56)][1]; \
T[12][(byte)(x2 >> 32)][1] ^ T[13][(byte)(x2 >> 40)][1] ^ T[14][(byte)(x2 >> 48)][1] ^ T[15][(byte)(x2 >> 56)][1];
} }
static inline void ILS(uint64 x1, uint64 x2, uint64* t1, uint64* t2) #define ILS(x1,x2,t1,t2) { \
{ t1 = IT[0][(byte)(x1)][0] ^ IT[1][(byte)(x1 >> 8)][0] ^ IT[2][(byte)(x1 >> 16)][0] ^ IT[3][(byte)(x1 >> 24)][0] ^ IT[4][(byte)(x1 >> 32)][0] ^ IT[5][(byte)(x1 >> 40)][0] ^ \
*t1 = IT[0][(byte)(x1)][0] ^ IT[1][(byte)(x1 >> 8)][0] ^ IT[2][(byte)(x1 >> 16)][0] ^ IT[3][(byte)(x1 >> 24)][0] ^ IT[4][(byte)(x1 >> 32)][0] ^ IT[5][(byte)(x1 >> 40)][0] ^ IT[6][(byte)(x1 >> 48)][0] ^ IT[7][(byte)(x1 >> 56)][0] ^ IT[8][(byte)(x2)][0] ^ IT[9][(byte)(x2 >> 8)][0] ^ IT[10][(byte)(x2 >> 16)][0] ^ IT[11][(byte)(x2 >> 24)][0] ^ \
IT[6][(byte)(x1 >> 48)][0] ^ IT[7][(byte)(x1 >> 56)][0] ^ IT[8][(byte)(x2)][0] ^ IT[9][(byte)(x2 >> 8)][0] ^ IT[10][(byte)(x2 >> 16)][0] ^ IT[11][(byte)(x2 >> 24)][0] ^ IT[12][(byte)(x2 >> 32)][0] ^ IT[13][(byte)(x2 >> 40)][0] ^ IT[14][(byte)(x2 >> 48)][0] ^ IT[15][(byte)(x2 >> 56)][0]; \
IT[12][(byte)(x2 >> 32)][0] ^ IT[13][(byte)(x2 >> 40)][0] ^ IT[14][(byte)(x2 >> 48)][0] ^ IT[15][(byte)(x2 >> 56)][0]; t2 = IT[0][(byte)(x1)][1] ^ IT[1][(byte)(x1 >> 8)][1] ^ IT[2][(byte)(x1 >> 16)][1] ^ IT[3][(byte)(x1 >> 24)][1] ^ IT[4][(byte)(x1 >> 32)][1] ^ IT[5][(byte)(x1 >> 40)][1] ^ \
*t2 = IT[0][(byte)(x1)][1] ^ IT[1][(byte)(x1 >> 8)][1] ^ IT[2][(byte)(x1 >> 16)][1] ^ IT[3][(byte)(x1 >> 24)][1] ^ IT[4][(byte)(x1 >> 32)][1] ^ IT[5][(byte)(x1 >> 40)][1] ^ IT[6][(byte)(x1 >> 48)][1] ^ IT[7][(byte)(x1 >> 56)][1] ^ IT[8][(byte)(x2)][1] ^ IT[9][(byte)(x2 >> 8)][1] ^ IT[10][(byte)(x2 >> 16)][1] ^ IT[11][(byte)(x2 >> 24)][1] ^ \
IT[6][(byte)(x1 >> 48)][1] ^ IT[7][(byte)(x1 >> 56)][1] ^ IT[8][(byte)(x2)][1] ^ IT[9][(byte)(x2 >> 8)][1] ^ IT[10][(byte)(x2 >> 16)][1] ^ IT[11][(byte)(x2 >> 24)][1] ^ IT[12][(byte)(x2 >> 32)][1] ^ IT[13][(byte)(x2 >> 40)][1] ^ IT[14][(byte)(x2 >> 48)][1] ^ IT[15][(byte)(x2 >> 56)][1]; \
IT[12][(byte)(x2 >> 32)][1] ^ IT[13][(byte)(x2 >> 40)][1] ^ IT[14][(byte)(x2 >> 48)][1] ^ IT[15][(byte)(x2 >> 56)][1];
} }
static inline void ILSS(uint64 x1, uint64 x2, uint64* t1, uint64* t2) #define ILSS(x1,x2,t1,t2) { \
{ t1 = IT[0][S[(byte)(x1)]][0] ^ IT[1][S[(byte)(x1 >> 8)]][0] ^ IT[2][S[(byte)(x1 >> 16)]][0] ^ IT[3][S[(byte)(x1 >> 24)]][0] ^ IT[4][S[(byte)(x1 >> 32)]][0] ^ IT[5][S[(byte)(x1 >> 40)]][0] ^ \
*t1 = IT[0][S[(byte)(x1)]][0] ^ IT[1][S[(byte)(x1 >> 8)]][0] ^ IT[2][S[(byte)(x1 >> 16)]][0] ^ IT[3][S[(byte)(x1 >> 24)]][0] ^ IT[4][S[(byte)(x1 >> 32)]][0] ^ IT[5][S[(byte)(x1 >> 40)]][0] ^ IT[6][S[(byte)(x1 >> 48)]][0] ^ IT[7][S[(byte)(x1 >> 56)]][0] ^ IT[8][S[(byte)(x2)]][0] ^ IT[9][S[(byte)(x2 >> 8)]][0] ^ IT[10][S[(byte)(x2 >> 16)]][0] ^ IT[11][S[(byte)(x2 >> 24)]][0] ^ \
IT[6][S[(byte)(x1 >> 48)]][0] ^ IT[7][S[(byte)(x1 >> 56)]][0] ^ IT[8][S[(byte)(x2)]][0] ^ IT[9][S[(byte)(x2 >> 8)]][0] ^ IT[10][S[(byte)(x2 >> 16)]][0] ^ IT[11][S[(byte)(x2 >> 24)]][0] ^ IT[12][S[(byte)(x2 >> 32)]][0] ^ IT[13][S[(byte)(x2 >> 40)]][0] ^ IT[14][S[(byte)(x2 >> 48)]][0] ^ IT[15][S[(byte)(x2 >> 56)]][0]; \
IT[12][S[(byte)(x2 >> 32)]][0] ^ IT[13][S[(byte)(x2 >> 40)]][0] ^ IT[14][S[(byte)(x2 >> 48)]][0] ^ IT[15][S[(byte)(x2 >> 56)]][0]; t2 = IT[0][S[(byte)(x1)]][1] ^ IT[1][S[(byte)(x1 >> 8)]][1] ^ IT[2][S[(byte)(x1 >> 16)]][1] ^ IT[3][S[(byte)(x1 >> 24)]][1] ^ IT[4][S[(byte)(x1 >> 32)]][1] ^ IT[5][S[(byte)(x1 >> 40)]][1] ^ \
*t2 = IT[0][S[(byte)(x1)]][1] ^ IT[1][S[(byte)(x1 >> 8)]][1] ^ IT[2][S[(byte)(x1 >> 16)]][1] ^ IT[3][S[(byte)(x1 >> 24)]][1] ^ IT[4][S[(byte)(x1 >> 32)]][1] ^ IT[5][S[(byte)(x1 >> 40)]][1] ^ IT[6][S[(byte)(x1 >> 48)]][1] ^ IT[7][S[(byte)(x1 >> 56)]][1] ^ IT[8][S[(byte)(x2)]][1] ^ IT[9][S[(byte)(x2 >> 8)]][1] ^ IT[10][S[(byte)(x2 >> 16)]][1] ^ IT[11][S[(byte)(x2 >> 24)]][1] ^ \
IT[6][S[(byte)(x1 >> 48)]][1] ^ IT[7][S[(byte)(x1 >> 56)]][1] ^ IT[8][S[(byte)(x2)]][1] ^ IT[9][S[(byte)(x2 >> 8)]][1] ^ IT[10][S[(byte)(x2 >> 16)]][1] ^ IT[11][S[(byte)(x2 >> 24)]][1] ^ IT[12][S[(byte)(x2 >> 32)]][1] ^ IT[13][S[(byte)(x2 >> 40)]][1] ^ IT[14][S[(byte)(x2 >> 48)]][1] ^ IT[15][S[(byte)(x2 >> 56)]][1]; \
IT[12][S[(byte)(x2 >> 32)]][1] ^ IT[13][S[(byte)(x2 >> 40)]][1] ^ IT[14][S[(byte)(x2 >> 48)]][1] ^ IT[15][S[(byte)(x2 >> 56)]][1];
} }
static inline void ISI(byte* val) #define ISI(val) { \
{ (val)[0] = IS[(val)[0]]; \
val[0] = IS[val[0]]; (val)[1] = IS[(val)[1]]; \
val[1] = IS[val[1]]; (val)[2] = IS[(val)[2]]; \
val[2] = IS[val[2]]; (val)[3] = IS[(val)[3]]; \
val[3] = IS[val[3]]; (val)[4] = IS[(val)[4]]; \
val[4] = IS[val[4]]; (val)[5] = IS[(val)[5]]; \
val[5] = IS[val[5]]; (val)[6] = IS[(val)[6]]; \
val[6] = IS[val[6]]; (val)[7] = IS[(val)[7]]; \
val[7] = IS[val[7]];
} }
static inline void F(uint64 k00, uint64 k01, uint64 k10, uint64 k11, int i, uint64* o00, uint64* o01, uint64* o10, uint64* o11) #define F(k00,k01,k10,k11,i,o00,o01,o10,o11) { \
{ o10 = k00; \
*o10 = k00; o11 = k01; \
*o11 = k01; k00 ^= C[i][0]; \
k00 ^= C[i][0]; k01 ^= C[i][1]; \
k01 ^= C[i][1]; LS(k00, k01, o00, o01); \
LS(k00, k01, o00, o01); o00 ^= k10; \
*o00 ^= k10; o01 ^= k11; \
*o01 ^= k11;
} }
static inline void FK(uint64* k00, uint64* k01, uint64* k10, uint64* k11, int ist) #define FK(k00,k01,k10,k11,ist) { \
{ for (i = 0; i < 8; i += 2) \
uint64 t00, t01, t10, t11; { \
int i; F(k00, k01, k10, k11, i + ist, t00, t01, t10, t11); \
for (i = 0; i < 8; i += 2) F(t00, t01, t10, t11, i + 1 + ist, k00, k01, k10, k11); \
{ } \
F(*k00, *k01, *k10, *k11, i + ist, &t00, &t01, &t10, &t11);
F(t00, t01, t10, t11, i + 1 + ist, k00, k01, k10, k11);
}
} }
void kuznyechik_set_key(const byte* key, kuznyechik_kds* kds) void kuznyechik_set_key(const byte* key, kuznyechik_kds* kds)
{ {
int i; #if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined(_UEFI) && (!defined (TC_WINDOWS_DRIVER) || (!defined (DEBUG) && defined (_WIN64)))
uint64 k00 = *(const uint64*)key; if(HasSSE2())
uint64 k01 = *(((const uint64*)key) + 1);
uint64 k10 = *(((const uint64*)key) + 2);
uint64 k11 = *(((const uint64*)key) + 3);
kds->rke[0][0] = k00;
kds->rke[0][1] = k01;
kds->rke[1][0] = k10;
kds->rke[1][1] = k11;
FK(&k00, &k01, &k10, &k11, 0);
kds->rke[2][0] = k00;
kds->rke[2][1] = k01;
kds->rke[3][0] = k10;
kds->rke[3][1] = k11;
FK(&k00, &k01, &k10, &k11, 8);
kds->rke[4][0] = k00;
kds->rke[4][1] = k01;
kds->rke[5][0] = k10;
kds->rke[5][1] = k11;
FK(&k00, &k01, &k10, &k11, 16);
kds->rke[6][0] = k00;
kds->rke[6][1] = k01;
kds->rke[7][0] = k10;
kds->rke[7][1] = k11;
FK(&k00, &k01, &k10, &k11, 24);
kds->rke[8][0] = k00;
kds->rke[8][1] = k01;
kds->rke[9][0] = k10;
kds->rke[9][1] = k11;
kds->rkd[0][0] = kds->rke[0][0];
kds->rkd[0][1] = kds->rke[0][1];
for (i = 1; i < 10; i++)
{ {
uint64 t1 = kds->rke[i][0], t2 = kds->rke[i][1]; kuznyechik_set_key_simd (key, kds);
kds->rkd[i][0] = t1; kds->rkd[i][1] = t2;
ILSS(t1, t2, &kds->rkd[i][0], &kds->rkd[i][1]);
} }
else
#endif
{
int i;
uint64 k00 = *(const uint64*)key;
uint64 k01 = *(((const uint64*)key) + 1);
uint64 k10 = *(((const uint64*)key) + 2);
uint64 k11 = *(((const uint64*)key) + 3);
uint64 t00, t01, t10, t11;
kds->rke[0] = k00;
kds->rke[1] = k01;
kds->rke[2] = k10;
kds->rke[3] = k11;
FK(k00, k01, k10, k11, 0);
kds->rke[4] = k00;
kds->rke[5] = k01;
kds->rke[6] = k10;
kds->rke[7] = k11;
FK(k00, k01, k10, k11, 8);
kds->rke[8] = k00;
kds->rke[9] = k01;
kds->rke[10] = k10;
kds->rke[11] = k11;
FK(k00, k01, k10, k11, 16);
kds->rke[12] = k00;
kds->rke[13] = k01;
kds->rke[14] = k10;
kds->rke[15] = k11;
FK(k00, k01, k10, k11, 24);
kds->rke[16] = k00;
kds->rke[17] = k01;
kds->rke[18] = k10;
kds->rke[19] = k11;
kds->rkd[0] = kds->rke[0];
kds->rkd[1] = kds->rke[1];
for (i = 1; i < 10; i++)
{
uint64 t1 = kds->rke[2*i], t2 = kds->rke[2*i+1];
kds->rkd[2*i] = t1; kds->rkd[2*i + 1] = t2;
ILSS(t1, t2, kds->rkd[2*i], kds->rkd[2*i+1]);
}
}
#ifdef CPPCRYPTO_DEBUG #ifdef CPPCRYPTO_DEBUG
for(int i = 0; i < 10; i++) for(int i = 0; i < 10; i++)
printf("key[%d]: { 0x%016I64X, 0x%016I64X }\n", i, kds->rke[i][0], kds->rke[i][1]); printf("key[%d]: { 0x%016I64X, 0x%016I64X }\n", i, kds->rke[2*i], kds->rke[2*i+1]);
#endif #endif
} }
void kuznyechik_encrypt_block(byte* out, const byte* in, kuznyechik_kds* kds) void kuznyechik_encrypt_block(byte* out, const byte* in, kuznyechik_kds* kds)
{ {
uint64 x1 = *(const uint64*)in; #if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined(_UEFI) && (!defined (TC_WINDOWS_DRIVER) || (!defined (DEBUG) && defined (_WIN64)))
uint64 x2 = *(((const uint64*)in)+1); if(HasSSE2())
uint64 t1, t2; {
x1 ^= kds->rke[0][0]; kuznyechik_encrypt_block_simd (out, in, kds);
x2 ^= kds->rke[0][1]; }
LS(x1, x2, &t1, &t2); else
t1 ^= kds->rke[1][0]; #endif
t2 ^= kds->rke[1][1]; {
LS(t1, t2, &x1, &x2); uint64 x1 = *(const uint64*)in;
x1 ^= kds->rke[2][0]; uint64 x2 = *(((const uint64*)in)+1);
x2 ^= kds->rke[2][1]; uint64 t1, t2;
LS(x1, x2, &t1, &t2); x1 ^= kds->rke[0];
t1 ^= kds->rke[3][0]; x2 ^= kds->rke[1];
t2 ^= kds->rke[3][1]; LS(x1, x2, t1, t2);
LS(t1, t2, &x1, &x2); t1 ^= kds->rke[2];
x1 ^= kds->rke[4][0]; t2 ^= kds->rke[3];
x2 ^= kds->rke[4][1]; LS(t1, t2, x1, x2);
LS(x1, x2, &t1, &t2); x1 ^= kds->rke[4];
t1 ^= kds->rke[5][0]; x2 ^= kds->rke[5];
t2 ^= kds->rke[5][1]; LS(x1, x2, t1, t2);
LS(t1, t2, &x1, &x2); t1 ^= kds->rke[6];
x1 ^= kds->rke[6][0]; t2 ^= kds->rke[7];
x2 ^= kds->rke[6][1]; LS(t1, t2, x1, x2);
LS(x1, x2, &t1, &t2); x1 ^= kds->rke[8];
t1 ^= kds->rke[7][0]; x2 ^= kds->rke[9];
t2 ^= kds->rke[7][1]; LS(x1, x2, t1, t2);
LS(t1, t2, &x1, &x2); t1 ^= kds->rke[10];
x1 ^= kds->rke[8][0]; t2 ^= kds->rke[11];
x2 ^= kds->rke[8][1]; LS(t1, t2, x1, x2);
LS(x1, x2, &t1, &t2); x1 ^= kds->rke[12];
t1 ^= kds->rke[9][0]; x2 ^= kds->rke[13];
t2 ^= kds->rke[9][1]; LS(x1, x2, t1, t2);
*(uint64*)out = t1; t1 ^= kds->rke[14];
*(((uint64*)out) + 1) = t2; t2 ^= kds->rke[15];
LS(t1, t2, x1, x2);
x1 ^= kds->rke[16];
x2 ^= kds->rke[17];
LS(x1, x2, t1, t2);
t1 ^= kds->rke[18];
t2 ^= kds->rke[19];
*(uint64*)out = t1;
*(((uint64*)out) + 1) = t2;
}
}
void kuznyechik_encrypt_blocks(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds)
{
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined(_UEFI) && (!defined (DEBUG) || !defined (TC_WINDOWS_DRIVER))
if(HasSSE2())
{
kuznyechik_encrypt_blocks_simd (out, in, blocks, kds);
}
else
#endif
{
while (blocks)
{
kuznyechik_encrypt_block (out, in, kds);
in += 16;
out += 16;
blocks--;
}
}
} }
void kuznyechik_decrypt_block(byte* out, const byte* in, kuznyechik_kds* kds) void kuznyechik_decrypt_block(byte* out, const byte* in, kuznyechik_kds* kds)
{ {
uint64 x1 = *(const uint64*)in; #if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined(_UEFI) && (!defined (TC_WINDOWS_DRIVER) || (!defined (DEBUG) && defined (_WIN64)))
uint64 x2 = *(((const uint64*)in) + 1); if(HasSSE2())
uint64 t1, t2; {
kuznyechik_decrypt_block_simd (out, in, kds);
}
else
#endif
{
uint64 x1 = *(const uint64*)in;
uint64 x2 = *(((const uint64*)in) + 1);
uint64 t1, t2;
ILSS(x1, x2, &t1, &t2); ILSS(x1, x2, t1, t2);
t1 ^= kds->rkd[9][0]; t1 ^= kds->rkd[18];
t2 ^= kds->rkd[9][1]; t2 ^= kds->rkd[19];
ILS(t1, t2, &x1, &x2); ILS(t1, t2, x1, x2);
x1 ^= kds->rkd[8][0]; x1 ^= kds->rkd[16];
x2 ^= kds->rkd[8][1]; x2 ^= kds->rkd[17];
ILS(x1, x2, &t1, &t2); ILS(x1, x2, t1, t2);
t1 ^= kds->rkd[7][0]; t1 ^= kds->rkd[14];
t2 ^= kds->rkd[7][1]; t2 ^= kds->rkd[15];
ILS(t1, t2, &x1, &x2); ILS(t1, t2, x1, x2);
x1 ^= kds->rkd[6][0]; x1 ^= kds->rkd[12];
x2 ^= kds->rkd[6][1]; x2 ^= kds->rkd[13];
ILS(x1, x2, &t1, &t2); ILS(x1, x2, t1, t2);
t1 ^= kds->rkd[5][0]; t1 ^= kds->rkd[10];
t2 ^= kds->rkd[5][1]; t2 ^= kds->rkd[11];
ILS(t1, t2, &x1, &x2); ILS(t1, t2, x1, x2);
x1 ^= kds->rkd[4][0]; x1 ^= kds->rkd[8];
x2 ^= kds->rkd[4][1]; x2 ^= kds->rkd[9];
ILS(x1, x2, &t1, &t2); ILS(x1, x2, t1, t2);
t1 ^= kds->rkd[3][0]; t1 ^= kds->rkd[6];
t2 ^= kds->rkd[3][1]; t2 ^= kds->rkd[7];
ILS(t1, t2, &x1, &x2); ILS(t1, t2, x1, x2);
x1 ^= kds->rkd[2][0]; x1 ^= kds->rkd[4];
x2 ^= kds->rkd[2][1]; x2 ^= kds->rkd[5];
ILS(x1, x2, &t1, &t2); ILS(x1, x2, t1, t2);
t1 ^= kds->rkd[1][0]; t1 ^= kds->rkd[2];
t2 ^= kds->rkd[1][1]; t2 ^= kds->rkd[3];
ISI((byte*)&t1); ISI((byte*)&t1);
ISI((byte*)&t2); ISI((byte*)&t2);
t1 ^= kds->rkd[0][0]; t1 ^= kds->rkd[0];
t2 ^= kds->rkd[0][1]; t2 ^= kds->rkd[1];
*(uint64*)out = t1; *(uint64*)out = t1;
*(((uint64*)out) + 1) = t2; *(((uint64*)out) + 1) = t2;
}
} }
void kuznyechik_decrypt_blocks(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds)
{
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && !defined(_UEFI) && (!defined (DEBUG) || !defined (TC_WINDOWS_DRIVER))
if(HasSSE2())
{
kuznyechik_decrypt_blocks_simd (out, in, blocks, kds);
}
else
#endif
{
while (blocks)
{
kuznyechik_decrypt_block (out, in, kds);
in += 16;
out += 16;
blocks--;
}
}
}
#if 0 #if 0
static inline uint8_t mul_gf(uint8_t x, uint8_t y, uint16_t p) { static inline uint8_t mul_gf(uint8_t x, uint8_t y, uint16_t p) {

6
src/Crypto/kuznyechik.h
View File

@ -16,14 +16,16 @@ extern "C" {
typedef struct _kuznyechik_kds typedef struct _kuznyechik_kds
{ {
uint64 rke[10][2]; uint64 rke[20];
uint64 rkd[10][2]; uint64 rkd[20];
} kuznyechik_kds; } kuznyechik_kds;
#define KUZNYECHIK_KS (sizeof(kuznyechik_kds)) #define KUZNYECHIK_KS (sizeof(kuznyechik_kds))
void kuznyechik_encrypt_block(byte* out, const byte* in, kuznyechik_kds* kds); void kuznyechik_encrypt_block(byte* out, const byte* in, kuznyechik_kds* kds);
void kuznyechik_encrypt_blocks(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds);
void kuznyechik_decrypt_block(byte* out, const byte* in, kuznyechik_kds* kds); void kuznyechik_decrypt_block(byte* out, const byte* in, kuznyechik_kds* kds);
void kuznyechik_decrypt_blocks(byte* out, const byte* in, size_t blocks, kuznyechik_kds* kds);
void kuznyechik_set_key(const byte* key, kuznyechik_kds *kds); void kuznyechik_set_key(const byte* key, kuznyechik_kds *kds);
#ifdef __cplusplus #ifdef __cplusplus

9517
src/Crypto/kuznyechik_simd.c Normal file
View File

File diff suppressed because it is too large Load Diff

48
src/Volume/Cipher.cpp
View File

@ -462,5 +462,53 @@ namespace VeraCrypt
{ {
kuznyechik_set_key (key, (kuznyechik_kds *) ScheduledKey.Ptr()); kuznyechik_set_key (key, (kuznyechik_kds *) ScheduledKey.Ptr());
} }
void CipherKuznyechik::EncryptBlocks (byte *data, size_t blockCount) const
{
if (!Initialized)
throw NotInitialized (SRC_POS);
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
if ((blockCount >= 4)
&& IsHwSupportAvailable())
{
kuznyechik_encrypt_blocks (data, data, blockCount, (kuznyechik_kds *) ScheduledKey.Ptr());
}
else
#endif
Cipher::EncryptBlocks (data, blockCount);
}
void CipherKuznyechik::DecryptBlocks (byte *data, size_t blockCount) const
{
if (!Initialized)
throw NotInitialized (SRC_POS);
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
if ((blockCount >= 4)
&& IsHwSupportAvailable())
{
kuznyechik_decrypt_blocks (data, data, blockCount, (kuznyechik_kds *) ScheduledKey.Ptr());
}
else
#endif
Cipher::DecryptBlocks (data, blockCount);
}
bool CipherKuznyechik::IsHwSupportAvailable () const
{
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
static bool state = false;
static bool stateValid = false;
if (!stateValid)
{
state = HasSSE2() ? true : false;
stateValid = true;
}
return state;
#else
return false;
#endif
}
bool Cipher::HwSupportEnabled = true; bool Cipher::HwSupportEnabled = true;
} }

2
src/Volume/Cipher.h
View File

@ -104,13 +104,13 @@ namespace VeraCrypt
TC_CIPHER (Serpent, 16, 32); TC_CIPHER (Serpent, 16, 32);
TC_CIPHER (Twofish, 16, 32); TC_CIPHER (Twofish, 16, 32);
TC_CIPHER (Camellia, 16, 32); TC_CIPHER (Camellia, 16, 32);
TC_CIPHER (Kuznyechik, 16, 32);
#undef TC_CIPHER_ADD_METHODS #undef TC_CIPHER_ADD_METHODS
#define TC_CIPHER_ADD_METHODS #define TC_CIPHER_ADD_METHODS
TC_CIPHER (Gost89, 16, 32); TC_CIPHER (Gost89, 16, 32);
TC_CIPHER (Gost89StaticSBOX, 16, 32); TC_CIPHER (Gost89StaticSBOX, 16, 32);
TC_CIPHER (Kuznyechik, 16, 32);
#undef TC_CIPHER #undef TC_CIPHER

1
src/Volume/Volume.make
View File

@ -79,6 +79,7 @@ OBJS += ../Crypto/Camellia.o
OBJS += ../Crypto/GostCipher.o OBJS += ../Crypto/GostCipher.o
OBJS += ../Crypto/Streebog.o OBJS += ../Crypto/Streebog.o
OBJS += ../Crypto/kuznyechik.o OBJS += ../Crypto/kuznyechik.o
OBJS += ../Crypto/kuznyechik_simd.o
OBJS += ../Common/Crc.o OBJS += ../Common/Crc.o
OBJS += ../Common/Endian.o OBJS += ../Common/Endian.o