1
mirror of https://github.com/hashcat/hashcat synced 2024-11-24 14:27:14 +01:00
hashcat/OpenCL/m01800-pure.cl
2022-02-07 13:31:22 +01:00

423 lines
10 KiB
Common Lisp

/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
#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_hash_sha512.cl)
#endif
#define COMPARE_S M2S(INCLUDE_PATH/inc_comp_single.cl)
#define COMPARE_M M2S(INCLUDE_PATH/inc_comp_multi.cl)
typedef struct sha512crypt_tmp
{
u64 l_alt_result[8];
u64 l_p_bytes[2];
u64 l_s_bytes[2];
// pure version
u32 alt_result[16];
u32 p_bytes[64];
u32 s_bytes[64];
} sha512crypt_tmp_t;
KERNEL_FQ void m01800_init (KERN_ATTR_TMPS (sha512crypt_tmp_t))
{
/**
* base
*/
const u64 gid = get_global_id (0);
if (gid >= GID_CNT) return;
/**
* init
*/
const u32 pw_len = pws[gid].pw_len;
u32 w[64] = { 0 };
for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
{
w[idx] = pws[gid].i[idx];
}
for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
{
w[idx] = hc_swap32_S (w[idx]);
}
const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
u32 s[64] = { 0 };
for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
{
s[idx] = salt_bufs[SALT_POS_HOST].salt_buf[idx];
}
for (u32 i = 0, idx = 0; i < salt_len; i += 4, idx += 1)
{
s[idx] = hc_swap32_S (s[idx]);
}
/**
* prepare
*/
sha512_ctx_t ctx;
sha512_init (&ctx);
sha512_update (&ctx, w, pw_len);
sha512_update (&ctx, s, salt_len);
sha512_update (&ctx, w, pw_len);
sha512_final (&ctx);
u32 final[32] = { 0 };
final[ 0] = h32_from_64_S (ctx.h[0]);
final[ 1] = l32_from_64_S (ctx.h[0]);
final[ 2] = h32_from_64_S (ctx.h[1]);
final[ 3] = l32_from_64_S (ctx.h[1]);
final[ 4] = h32_from_64_S (ctx.h[2]);
final[ 5] = l32_from_64_S (ctx.h[2]);
final[ 6] = h32_from_64_S (ctx.h[3]);
final[ 7] = l32_from_64_S (ctx.h[3]);
final[ 8] = h32_from_64_S (ctx.h[4]);
final[ 9] = l32_from_64_S (ctx.h[4]);
final[10] = h32_from_64_S (ctx.h[5]);
final[11] = l32_from_64_S (ctx.h[5]);
final[12] = h32_from_64_S (ctx.h[6]);
final[13] = l32_from_64_S (ctx.h[6]);
final[14] = h32_from_64_S (ctx.h[7]);
final[15] = l32_from_64_S (ctx.h[7]);
// alt_result
sha512_init (&ctx);
sha512_update (&ctx, w, pw_len);
sha512_update (&ctx, s, salt_len);
int pl;
for (pl = pw_len; pl > 64; pl -= 64)
{
sha512_update (&ctx, final, 64);
}
u32 t_final[32] = { 0 };
#ifdef _unroll
#pragma unroll
#endif
for (int i = 0; i < 16; i++) t_final[i] = final[i];
truncate_block_16x4_be_S (t_final + 0, t_final + 4, t_final + 8, t_final + 12, pl);
sha512_update (&ctx, t_final, pl);
for (int cnt = pw_len; cnt > 0; cnt >>= 1)
{
if ((cnt & 1) != 0)
{
sha512_update (&ctx, final, 64);
}
else
{
sha512_update (&ctx, w, pw_len);
}
}
sha512_final (&ctx);
tmps[gid].alt_result[ 0] = h32_from_64_S (ctx.h[0]);
tmps[gid].alt_result[ 1] = l32_from_64_S (ctx.h[0]);
tmps[gid].alt_result[ 2] = h32_from_64_S (ctx.h[1]);
tmps[gid].alt_result[ 3] = l32_from_64_S (ctx.h[1]);
tmps[gid].alt_result[ 4] = h32_from_64_S (ctx.h[2]);
tmps[gid].alt_result[ 5] = l32_from_64_S (ctx.h[2]);
tmps[gid].alt_result[ 6] = h32_from_64_S (ctx.h[3]);
tmps[gid].alt_result[ 7] = l32_from_64_S (ctx.h[3]);
tmps[gid].alt_result[ 8] = h32_from_64_S (ctx.h[4]);
tmps[gid].alt_result[ 9] = l32_from_64_S (ctx.h[4]);
tmps[gid].alt_result[10] = h32_from_64_S (ctx.h[5]);
tmps[gid].alt_result[11] = l32_from_64_S (ctx.h[5]);
tmps[gid].alt_result[12] = h32_from_64_S (ctx.h[6]);
tmps[gid].alt_result[13] = l32_from_64_S (ctx.h[6]);
tmps[gid].alt_result[14] = h32_from_64_S (ctx.h[7]);
tmps[gid].alt_result[15] = l32_from_64_S (ctx.h[7]);
// p_bytes
sha512_init (&ctx);
for (u32 j = 0; j < pw_len; j++)
{
sha512_update (&ctx, w, pw_len);
}
sha512_final (&ctx);
final[ 0] = h32_from_64_S (ctx.h[0]);
final[ 1] = l32_from_64_S (ctx.h[0]);
final[ 2] = h32_from_64_S (ctx.h[1]);
final[ 3] = l32_from_64_S (ctx.h[1]);
final[ 4] = h32_from_64_S (ctx.h[2]);
final[ 5] = l32_from_64_S (ctx.h[2]);
final[ 6] = h32_from_64_S (ctx.h[3]);
final[ 7] = l32_from_64_S (ctx.h[3]);
final[ 8] = h32_from_64_S (ctx.h[4]);
final[ 9] = l32_from_64_S (ctx.h[4]);
final[10] = h32_from_64_S (ctx.h[5]);
final[11] = l32_from_64_S (ctx.h[5]);
final[12] = h32_from_64_S (ctx.h[6]);
final[13] = l32_from_64_S (ctx.h[6]);
final[14] = h32_from_64_S (ctx.h[7]);
final[15] = l32_from_64_S (ctx.h[7]);
u32 p_final[64] = { 0 };
int idx;
for (pl = pw_len, idx = 0; pl > 64; pl -= 64, idx += 16)
{
p_final[idx + 0] = final[ 0];
p_final[idx + 1] = final[ 1];
p_final[idx + 2] = final[ 2];
p_final[idx + 3] = final[ 3];
p_final[idx + 4] = final[ 4];
p_final[idx + 5] = final[ 5];
p_final[idx + 6] = final[ 6];
p_final[idx + 7] = final[ 7];
p_final[idx + 8] = final[ 8];
p_final[idx + 9] = final[ 9];
p_final[idx + 10] = final[10];
p_final[idx + 11] = final[11];
p_final[idx + 12] = final[12];
p_final[idx + 13] = final[13];
p_final[idx + 14] = final[14];
p_final[idx + 15] = final[15];
}
truncate_block_16x4_be_S (final + 0, final + 4, final + 8, final + 12, pl);
p_final[idx + 0] = final[ 0];
p_final[idx + 1] = final[ 1];
p_final[idx + 2] = final[ 2];
p_final[idx + 3] = final[ 3];
p_final[idx + 4] = final[ 4];
p_final[idx + 5] = final[ 5];
p_final[idx + 6] = final[ 6];
p_final[idx + 7] = final[ 7];
p_final[idx + 8] = final[ 8];
p_final[idx + 9] = final[ 9];
p_final[idx + 10] = final[10];
p_final[idx + 11] = final[11];
p_final[idx + 12] = final[12];
p_final[idx + 13] = final[13];
p_final[idx + 14] = final[14];
p_final[idx + 15] = final[15];
#ifdef _unroll
#pragma unroll
#endif
for (int i = 0; i < 64; i++) tmps[gid].p_bytes[i] = p_final[i];
// s_bytes
sha512_init (&ctx);
for (u32 j = 0; j < 16 + (tmps[gid].alt_result[0] >> 24); j++)
{
sha512_update (&ctx, s, salt_len);
}
sha512_final (&ctx);
final[ 0] = h32_from_64_S (ctx.h[0]);
final[ 1] = l32_from_64_S (ctx.h[0]);
final[ 2] = h32_from_64_S (ctx.h[1]);
final[ 3] = l32_from_64_S (ctx.h[1]);
final[ 4] = h32_from_64_S (ctx.h[2]);
final[ 5] = l32_from_64_S (ctx.h[2]);
final[ 6] = h32_from_64_S (ctx.h[3]);
final[ 7] = l32_from_64_S (ctx.h[3]);
final[ 8] = h32_from_64_S (ctx.h[4]);
final[ 9] = l32_from_64_S (ctx.h[4]);
final[10] = h32_from_64_S (ctx.h[5]);
final[11] = l32_from_64_S (ctx.h[5]);
final[12] = h32_from_64_S (ctx.h[6]);
final[13] = l32_from_64_S (ctx.h[6]);
final[14] = h32_from_64_S (ctx.h[7]);
final[15] = l32_from_64_S (ctx.h[7]);
u32 s_final[64] = { 0 };
for (pl = salt_len, idx = 0; pl > 64; pl -= 64, idx += 16)
{
s_final[idx + 0] = final[ 0];
s_final[idx + 1] = final[ 1];
s_final[idx + 2] = final[ 2];
s_final[idx + 3] = final[ 3];
s_final[idx + 4] = final[ 4];
s_final[idx + 5] = final[ 5];
s_final[idx + 6] = final[ 6];
s_final[idx + 7] = final[ 7];
s_final[idx + 8] = final[ 8];
s_final[idx + 9] = final[ 9];
s_final[idx + 10] = final[10];
s_final[idx + 11] = final[11];
s_final[idx + 12] = final[12];
s_final[idx + 13] = final[13];
s_final[idx + 14] = final[14];
s_final[idx + 15] = final[15];
}
truncate_block_16x4_be_S (final + 0, final + 4, final + 8, final + 12, pl);
s_final[idx + 0] = final[ 0];
s_final[idx + 1] = final[ 1];
s_final[idx + 2] = final[ 2];
s_final[idx + 3] = final[ 3];
s_final[idx + 4] = final[ 4];
s_final[idx + 5] = final[ 5];
s_final[idx + 6] = final[ 6];
s_final[idx + 7] = final[ 7];
s_final[idx + 8] = final[ 8];
s_final[idx + 9] = final[ 9];
s_final[idx + 10] = final[10];
s_final[idx + 11] = final[11];
s_final[idx + 12] = final[12];
s_final[idx + 13] = final[13];
s_final[idx + 14] = final[14];
s_final[idx + 15] = final[15];
#ifdef _unroll
#pragma unroll
#endif
for (int i = 0; i < 64; i++) tmps[gid].s_bytes[i] = s_final[i];
}
KERNEL_FQ void m01800_loop (KERN_ATTR_TMPS (sha512crypt_tmp_t))
{
/**
* base
*/
const u64 gid = get_global_id (0);
if (gid >= GID_CNT) return;
const u32 pw_len = pws[gid].pw_len;
const u32 salt_len = salt_bufs[SALT_POS_HOST].salt_len;
u32 alt_result[32] = { 0 };
#ifdef _unroll
#pragma unroll
#endif
for (int i = 0; i < 16; i++) alt_result[i] = tmps[gid].alt_result[i];
/* Repeatedly run the collected hash value through SHA512 to burn
CPU cycles. */
for (u32 i = 0, j = LOOP_POS; i < LOOP_CNT; i++, j++)
{
sha512_ctx_t ctx;
sha512_init (&ctx);
if (j & 1)
{
sha512_update_global (&ctx, tmps[gid].p_bytes, pw_len);
}
else
{
sha512_update (&ctx, alt_result, 64);
}
if (j % 3)
{
sha512_update_global (&ctx, tmps[gid].s_bytes, salt_len);
}
if (j % 7)
{
sha512_update_global (&ctx, tmps[gid].p_bytes, pw_len);
}
if (j & 1)
{
sha512_update (&ctx, alt_result, 64);
}
else
{
sha512_update_global (&ctx, tmps[gid].p_bytes, pw_len);
}
sha512_final (&ctx);
alt_result[ 0] = h32_from_64_S (ctx.h[0]);
alt_result[ 1] = l32_from_64_S (ctx.h[0]);
alt_result[ 2] = h32_from_64_S (ctx.h[1]);
alt_result[ 3] = l32_from_64_S (ctx.h[1]);
alt_result[ 4] = h32_from_64_S (ctx.h[2]);
alt_result[ 5] = l32_from_64_S (ctx.h[2]);
alt_result[ 6] = h32_from_64_S (ctx.h[3]);
alt_result[ 7] = l32_from_64_S (ctx.h[3]);
alt_result[ 8] = h32_from_64_S (ctx.h[4]);
alt_result[ 9] = l32_from_64_S (ctx.h[4]);
alt_result[10] = h32_from_64_S (ctx.h[5]);
alt_result[11] = l32_from_64_S (ctx.h[5]);
alt_result[12] = h32_from_64_S (ctx.h[6]);
alt_result[13] = l32_from_64_S (ctx.h[6]);
alt_result[14] = h32_from_64_S (ctx.h[7]);
alt_result[15] = l32_from_64_S (ctx.h[7]);
}
#ifdef _unroll
#pragma unroll
#endif
for (int i = 0; i < 16; i++) tmps[gid].alt_result[i] = alt_result[i];
}
KERNEL_FQ void m01800_comp (KERN_ATTR_TMPS (sha512crypt_tmp_t))
{
/**
* base
*/
const u64 gid = get_global_id (0);
if (gid >= GID_CNT) return;
const u64 lid = get_local_id (0);
const u32 r0 = hc_swap32_S (tmps[gid].alt_result[0]);
const u32 r1 = hc_swap32_S (tmps[gid].alt_result[1]);
const u32 r2 = hc_swap32_S (tmps[gid].alt_result[2]);
const u32 r3 = hc_swap32_S (tmps[gid].alt_result[3]);
#define il_pos 0
#ifdef KERNEL_STATIC
#include COMPARE_M
#endif
}