1
0
mirror of https://github.com/hashcat/hashcat.git synced 2024-11-30 03:48:17 +00:00
hashcat/OpenCL/m30901_a3-pure.cl

366 lines
9.7 KiB
Common Lisp
Raw Normal View History

/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
//#define NEW_SIMD_CODE
// #define SECP256K1_TMPS_TYPE CONSTANT_AS
#define SECP256K1_TMPS_TYPE PRIVATE_AS
#ifdef KERNEL_STATIC
#include M2S(INCLUDE_PATH/inc_vendor.h)
#include M2S(INCLUDE_PATH/inc_types.h)
#include M2S(INCLUDE_PATH/inc_platform.cl)
#include M2S(INCLUDE_PATH/inc_common.cl)
#include M2S(INCLUDE_PATH/inc_scalar.cl)
#include M2S(INCLUDE_PATH/inc_hash_sha256.cl)
#include M2S(INCLUDE_PATH/inc_hash_ripemd160.cl)
#include M2S(INCLUDE_PATH/inc_ecc_secp256k1.cl)
#endif
// or use set_precomputed_basepoint_g () instead:
// (set SECP256K1_TMPS_TYPE to CONSTANT_AS above:)
// CONSTANT_AS const secp256k1_t preG =
// {
// {
// SECP256K1_G_PRE_COMPUTED_00, SECP256K1_G_PRE_COMPUTED_01, SECP256K1_G_PRE_COMPUTED_02,
// SECP256K1_G_PRE_COMPUTED_03, SECP256K1_G_PRE_COMPUTED_04, SECP256K1_G_PRE_COMPUTED_05,
// SECP256K1_G_PRE_COMPUTED_06, SECP256K1_G_PRE_COMPUTED_07, SECP256K1_G_PRE_COMPUTED_08,
// SECP256K1_G_PRE_COMPUTED_09, SECP256K1_G_PRE_COMPUTED_10, SECP256K1_G_PRE_COMPUTED_11,
// SECP256K1_G_PRE_COMPUTED_12, SECP256K1_G_PRE_COMPUTED_13, SECP256K1_G_PRE_COMPUTED_14,
// SECP256K1_G_PRE_COMPUTED_15, SECP256K1_G_PRE_COMPUTED_16, SECP256K1_G_PRE_COMPUTED_17,
// SECP256K1_G_PRE_COMPUTED_18, SECP256K1_G_PRE_COMPUTED_19, SECP256K1_G_PRE_COMPUTED_20,
// SECP256K1_G_PRE_COMPUTED_21, SECP256K1_G_PRE_COMPUTED_22, SECP256K1_G_PRE_COMPUTED_23,
// SECP256K1_G_PRE_COMPUTED_24, SECP256K1_G_PRE_COMPUTED_25, SECP256K1_G_PRE_COMPUTED_26,
// SECP256K1_G_PRE_COMPUTED_27, SECP256K1_G_PRE_COMPUTED_28, SECP256K1_G_PRE_COMPUTED_29,
// SECP256K1_G_PRE_COMPUTED_30, SECP256K1_G_PRE_COMPUTED_31, SECP256K1_G_PRE_COMPUTED_32,
// SECP256K1_G_PRE_COMPUTED_33, SECP256K1_G_PRE_COMPUTED_34, SECP256K1_G_PRE_COMPUTED_35,
// SECP256K1_G_PRE_COMPUTED_36, SECP256K1_G_PRE_COMPUTED_37, SECP256K1_G_PRE_COMPUTED_38,
// SECP256K1_G_PRE_COMPUTED_39, SECP256K1_G_PRE_COMPUTED_40, SECP256K1_G_PRE_COMPUTED_41,
// SECP256K1_G_PRE_COMPUTED_42, SECP256K1_G_PRE_COMPUTED_43, SECP256K1_G_PRE_COMPUTED_44,
// SECP256K1_G_PRE_COMPUTED_45, SECP256K1_G_PRE_COMPUTED_46, SECP256K1_G_PRE_COMPUTED_47,
// SECP256K1_G_PRE_COMPUTED_48, SECP256K1_G_PRE_COMPUTED_49, SECP256K1_G_PRE_COMPUTED_50,
// SECP256K1_G_PRE_COMPUTED_51, SECP256K1_G_PRE_COMPUTED_52, SECP256K1_G_PRE_COMPUTED_53,
// SECP256K1_G_PRE_COMPUTED_54, SECP256K1_G_PRE_COMPUTED_55, SECP256K1_G_PRE_COMPUTED_56,
// SECP256K1_G_PRE_COMPUTED_57, SECP256K1_G_PRE_COMPUTED_58, SECP256K1_G_PRE_COMPUTED_59,
// SECP256K1_G_PRE_COMPUTED_60, SECP256K1_G_PRE_COMPUTED_61, SECP256K1_G_PRE_COMPUTED_62,
// SECP256K1_G_PRE_COMPUTED_63, SECP256K1_G_PRE_COMPUTED_64, SECP256K1_G_PRE_COMPUTED_65,
// SECP256K1_G_PRE_COMPUTED_66, SECP256K1_G_PRE_COMPUTED_67, SECP256K1_G_PRE_COMPUTED_68,
// SECP256K1_G_PRE_COMPUTED_69, SECP256K1_G_PRE_COMPUTED_70, SECP256K1_G_PRE_COMPUTED_71,
// SECP256K1_G_PRE_COMPUTED_72, SECP256K1_G_PRE_COMPUTED_73, SECP256K1_G_PRE_COMPUTED_74,
// SECP256K1_G_PRE_COMPUTED_75, SECP256K1_G_PRE_COMPUTED_76, SECP256K1_G_PRE_COMPUTED_77,
// SECP256K1_G_PRE_COMPUTED_78, SECP256K1_G_PRE_COMPUTED_79, SECP256K1_G_PRE_COMPUTED_80,
// SECP256K1_G_PRE_COMPUTED_81, SECP256K1_G_PRE_COMPUTED_82, SECP256K1_G_PRE_COMPUTED_83,
// SECP256K1_G_PRE_COMPUTED_84, SECP256K1_G_PRE_COMPUTED_85, SECP256K1_G_PRE_COMPUTED_86,
// SECP256K1_G_PRE_COMPUTED_87, SECP256K1_G_PRE_COMPUTED_88, SECP256K1_G_PRE_COMPUTED_89,
// SECP256K1_G_PRE_COMPUTED_90, SECP256K1_G_PRE_COMPUTED_91, SECP256K1_G_PRE_COMPUTED_92,
// SECP256K1_G_PRE_COMPUTED_93, SECP256K1_G_PRE_COMPUTED_94, SECP256K1_G_PRE_COMPUTED_95,
// }
// };
DECLSPEC u32 hex_convert_u32 (PRIVATE_AS const u32 c)
{
return (c & 15) + (c >> 6) * 9;
}
DECLSPEC u32 hex_u32_to_u32 (PRIVATE_AS const u32 hex0, PRIVATE_AS const u32 hex1)
{
u32 v = 0;
v |= hex_convert_u32 ((hex0 >> 0) & 0xff) << 28;
v |= hex_convert_u32 ((hex0 >> 8) & 0xff) << 24;
v |= hex_convert_u32 ((hex0 >> 16) & 0xff) << 20;
v |= hex_convert_u32 ((hex0 >> 24) & 0xff) << 16;
v |= hex_convert_u32 ((hex1 >> 0) & 0xff) << 12;
v |= hex_convert_u32 ((hex1 >> 8) & 0xff) << 8;
v |= hex_convert_u32 ((hex1 >> 16) & 0xff) << 4;
v |= hex_convert_u32 ((hex1 >> 24) & 0xff) << 0;
return (v);
}
KERNEL_FQ void m30901_mxx (KERN_ATTR_VECTOR ())
{
/**
* modifier
*/
const u64 gid = get_global_id (0);
if (gid >= GID_CNT) return;
/**
* base
*/
const u32 pw_len = pws[gid].pw_len;
if (pw_len != 64) return;
// copy password to w
u32 w[16];
for (u32 i = 0; i < 16; i++) // pw_len / 4
{
w[i] = pws[gid].i[i];
}
for (u32 i = 1; i < 16; i++)
{
if (is_valid_hex_32 (w[i]) == 0) return;
}
secp256k1_t preG; // need to change SECP256K1_TMPS_TYPE above to: PRIVATE_AS
set_precomputed_basepoint_g (&preG);
/**
* loop
*/
u32 w0l = w[0];
for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
{
u32x w0r = words_buf_r[il_pos / VECT_SIZE];
const u32 w0 = w0l | w0r;
w[0] = w0;
if (is_valid_hex_32 (w[0]) == 0) continue;
// convert password from hex to binary
u32 tmp[16] = { 0 };
for (u32 i = 0, j = 0; i < 8; i += 1, j += 2)
{
tmp[i] = hex_u32_to_u32 (w[j + 0], w[j + 1]);
}
u32 prv_key[9];
prv_key[0] = tmp[7];
prv_key[1] = tmp[6];
prv_key[2] = tmp[5];
prv_key[3] = tmp[4];
prv_key[4] = tmp[3];
prv_key[5] = tmp[2];
prv_key[6] = tmp[1];
prv_key[7] = tmp[0];
// convert: pub_key = G * prv_key
u32 x[8];
u32 y[8];
point_mul_xy (x, y, prv_key, &preG);
// to public key:
u32 pub_key[16] = { 0 }; // why is re-using the "tmp" variable here slower ?
const u32 type = 0x02 | (y[0] & 1);
pub_key[8] = (x[0] << 24);
pub_key[7] = (x[0] >> 8) | (x[1] << 24);
pub_key[6] = (x[1] >> 8) | (x[2] << 24);
pub_key[5] = (x[2] >> 8) | (x[3] << 24);
pub_key[4] = (x[3] >> 8) | (x[4] << 24);
pub_key[3] = (x[4] >> 8) | (x[5] << 24);
pub_key[2] = (x[5] >> 8) | (x[6] << 24);
pub_key[1] = (x[6] >> 8) | (x[7] << 24);
pub_key[0] = (x[7] >> 8) | (type << 24);
// calculate HASH160 for pub key
sha256_ctx_t ctx;
sha256_init (&ctx);
sha256_update (&ctx, pub_key, 33); // length of public key: 33
sha256_final (&ctx);
for (u32 i = 0; i < 8; i++) tmp[i] = ctx.h[i];
// tmp[ 8] = 0; tmp[ 9] = 0; tmp[10] = 0; tmp[11] = 0;
// tmp[12] = 0; tmp[13] = 0; tmp[14] = 0; tmp[15] = 0;
for (u32 i = 8; i < 16; i++) tmp[i] = 0;
// now let's do RIPEMD-160 on the sha256sum
ripemd160_ctx_t rctx;
ripemd160_init (&rctx);
ripemd160_update_swap (&rctx, tmp, 32);
ripemd160_final (&rctx);
const u32 r0 = rctx.h[0];
const u32 r1 = rctx.h[1];
const u32 r2 = rctx.h[2];
const u32 r3 = rctx.h[3];
COMPARE_M_SCALAR (r0, r1, r2, r3);
}
}
KERNEL_FQ void m30901_sxx (KERN_ATTR_VECTOR ())
{
/**
* modifier
*/
const u64 gid = get_global_id (0);
if (gid >= GID_CNT) return;
/**
* digest
*/
const u32 search[4] =
{
digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R2],
digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R3]
};
/**
* base
*/
const u32 pw_len = pws[gid].pw_len;
if (pw_len != 64) return;
// copy password to w
u32 w[16];
for (u32 i = 0; i < 16; i++) // pw_len / 4
{
w[i] = pws[gid].i[i];
}
for (u32 i = 1; i < 16; i++)
{
if (is_valid_hex_32 (w[i]) == 0) return;
}
secp256k1_t preG; // need to change SECP256K1_TMPS_TYPE above to: PRIVATE_AS
set_precomputed_basepoint_g (&preG);
/**
* loop
*/
u32 w0l = w[0];
for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
{
u32x w0r = words_buf_r[il_pos / VECT_SIZE];
const u32 w0 = w0l | w0r;
w[0] = w0;
if (is_valid_hex_32 (w[0]) == 0) continue;
// convert password from hex to binary
u32 tmp[16] = { 0 };
for (u32 i = 0, j = 0; i < 8; i += 1, j += 2)
{
tmp[i] = hex_u32_to_u32 (w[j + 0], w[j + 1]);
}
u32 prv_key[9];
prv_key[0] = tmp[7];
prv_key[1] = tmp[6];
prv_key[2] = tmp[5];
prv_key[3] = tmp[4];
prv_key[4] = tmp[3];
prv_key[5] = tmp[2];
prv_key[6] = tmp[1];
prv_key[7] = tmp[0];
// convert: pub_key = G * prv_key
u32 x[8];
u32 y[8];
point_mul_xy (x, y, prv_key, &preG);
// to public key:
u32 pub_key[16] = { 0 }; // why is re-using the "tmp" variable here slower ?
const u32 type = 0x02 | (y[0] & 1);
pub_key[8] = (x[0] << 24);
pub_key[7] = (x[0] >> 8) | (x[1] << 24);
pub_key[6] = (x[1] >> 8) | (x[2] << 24);
pub_key[5] = (x[2] >> 8) | (x[3] << 24);
pub_key[4] = (x[3] >> 8) | (x[4] << 24);
pub_key[3] = (x[4] >> 8) | (x[5] << 24);
pub_key[2] = (x[5] >> 8) | (x[6] << 24);
pub_key[1] = (x[6] >> 8) | (x[7] << 24);
pub_key[0] = (x[7] >> 8) | (type << 24);
// calculate HASH160 for pub key
sha256_ctx_t ctx;
sha256_init (&ctx);
sha256_update (&ctx, pub_key, 33); // length of public key: 33
sha256_final (&ctx);
for (u32 i = 0; i < 8; i++) tmp[i] = ctx.h[i];
// tmp[ 8] = 0; tmp[ 9] = 0; tmp[10] = 0; tmp[11] = 0;
// tmp[12] = 0; tmp[13] = 0; tmp[14] = 0; tmp[15] = 0;
for (u32 i = 8; i < 16; i++) tmp[i] = 0;
// now let's do RIPEMD-160 on the sha256sum
ripemd160_ctx_t rctx;
ripemd160_init (&rctx);
ripemd160_update_swap (&rctx, tmp, 32);
ripemd160_final (&rctx);
const u32 r0 = rctx.h[0];
const u32 r1 = rctx.h[1];
const u32 r2 = rctx.h[2];
const u32 r3 = rctx.h[3];
COMPARE_S_SCALAR (r0, r1, r2, r3);
}
}