2 * Authors.....: Jens Steube <jens.steube@gmail.com>
3 * Gabriele Gristina <matrix@hashcat.net>
10 #include "include/constants.h"
11 #include "include/kernel_vendor.h"
18 #include "include/kernel_functions.c"
19 #include "OpenCL/types_ocl.c"
20 #include "OpenCL/common.c"
22 #define COMPARE_S "OpenCL/check_single_comp4.c"
23 #define COMPARE_M "OpenCL/check_multi_comp4.c"
27 #define SBOG_LPSti64 \
28 s_sbob_sl64[0][(t[0] >> (i * 8)) & 0xff] ^ \
29 s_sbob_sl64[1][(t[1] >> (i * 8)) & 0xff] ^ \
30 s_sbob_sl64[2][(t[2] >> (i * 8)) & 0xff] ^ \
31 s_sbob_sl64[3][(t[3] >> (i * 8)) & 0xff] ^ \
32 s_sbob_sl64[4][(t[4] >> (i * 8)) & 0xff] ^ \
33 s_sbob_sl64[5][(t[5] >> (i * 8)) & 0xff] ^ \
34 s_sbob_sl64[6][(t[6] >> (i * 8)) & 0xff] ^ \
35 s_sbob_sl64[7][(t[7] >> (i * 8)) & 0xff]
39 __constant u64 sbob_sl64[8][256] =
2107 __constant u64 sbob_rc64[12][8] =
2231 static void streebog_g (u64 h[8], const u64 m[8], __local u64 (*s_sbob_sl64)[256])
2238 for (int i = 0; i < 8; i++)
2243 for (int i = 0; i < 8; i++)
2245 k[i] = SBOG_LPSti64;
2249 for (int i = 0; i < 8; i++)
2254 for (int r = 0; r < 12; r++)
2257 for (int i = 0; i < 8; i++)
2263 for (int i = 0; i < 8; i++)
2265 s[i] = SBOG_LPSti64;
2268 for (int i = 0; i < 8; i++)
2270 t[i] = k[i] ^ sbob_rc64[r][i];
2274 for (int i = 0; i < 8; i++)
2276 k[i] = SBOG_LPSti64;
2281 for (int i = 0; i < 8; i++)
2283 h[i] ^= s[i] ^ k[i] ^ m[i];
2287 __kernel void m11800_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
2293 const u32 gid = get_global_id (0);
2294 const u32 lid = get_local_id (0);
2295 const u32 lsz = get_local_size (0);
2298 * shared lookup table
2301 __local u64 s_sbob_sl64[8][256];
2303 for (u32 i = lid; i < 256; i += lsz)
2305 s_sbob_sl64[0][i] = sbob_sl64[0][i];
2306 s_sbob_sl64[1][i] = sbob_sl64[1][i];
2307 s_sbob_sl64[2][i] = sbob_sl64[2][i];
2308 s_sbob_sl64[3][i] = sbob_sl64[3][i];
2309 s_sbob_sl64[4][i] = sbob_sl64[4][i];
2310 s_sbob_sl64[5][i] = sbob_sl64[5][i];
2311 s_sbob_sl64[6][i] = sbob_sl64[6][i];
2312 s_sbob_sl64[7][i] = sbob_sl64[7][i];
2315 barrier (CLK_LOCAL_MEM_FENCE);
2317 if (gid >= gid_max) return;
2325 wordl0[0] = pws[gid].i[ 0];
2326 wordl0[1] = pws[gid].i[ 1];
2327 wordl0[2] = pws[gid].i[ 2];
2328 wordl0[3] = pws[gid].i[ 3];
2332 wordl1[0] = pws[gid].i[ 4];
2333 wordl1[1] = pws[gid].i[ 5];
2334 wordl1[2] = pws[gid].i[ 6];
2335 wordl1[3] = pws[gid].i[ 7];
2351 const u32 pw_l_len = pws[gid].pw_len;
2353 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
2355 append_0x80_2x4 (wordl0, wordl1, pw_l_len);
2357 switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
2364 for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
2366 const u32 pw_r_len = combs_buf[il_pos].pw_len;
2368 const u32 pw_len = pw_l_len + pw_r_len;
2372 wordr0[0] = combs_buf[il_pos].i[0];
2373 wordr0[1] = combs_buf[il_pos].i[1];
2374 wordr0[2] = combs_buf[il_pos].i[2];
2375 wordr0[3] = combs_buf[il_pos].i[3];
2379 wordr1[0] = combs_buf[il_pos].i[4];
2380 wordr1[1] = combs_buf[il_pos].i[5];
2381 wordr1[2] = combs_buf[il_pos].i[6];
2382 wordr1[3] = combs_buf[il_pos].i[7];
2398 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
2400 switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
2405 w[ 0] = wordl0[0] | wordr0[0];
2406 w[ 1] = wordl0[1] | wordr0[1];
2407 w[ 2] = wordl0[2] | wordr0[2];
2408 w[ 3] = wordl0[3] | wordr0[3];
2409 w[ 4] = wordl1[0] | wordr1[0];
2410 w[ 5] = wordl1[1] | wordr1[1];
2411 w[ 6] = wordl1[2] | wordr1[2];
2412 w[ 7] = wordl1[3] | wordr1[3];
2413 w[ 8] = wordl2[0] | wordr2[0];
2414 w[ 9] = wordl2[1] | wordr2[1];
2415 w[10] = wordl2[2] | wordr2[2];
2416 w[11] = wordl2[3] | wordr2[3];
2417 w[12] = wordl3[0] | wordr3[0];
2418 w[13] = wordl3[1] | wordr3[1];
2419 w[14] = wordl3[1] | wordr3[1];
2420 w[15] = wordl3[1] | wordr3[1];
2422 append_0x01_4x4 (&w[0], &w[1], &w[2], &w[3], pw_len);
2425 * reverse message block
2430 m[0] = hl32_to_64 (w[15], w[14]);
2431 m[1] = hl32_to_64 (w[13], w[12]);
2432 m[2] = hl32_to_64 (w[11], w[10]);
2433 m[3] = hl32_to_64 (w[ 9], w[ 8]);
2434 m[4] = hl32_to_64 (w[ 7], w[ 6]);
2435 m[5] = hl32_to_64 (w[ 5], w[ 4]);
2436 m[6] = hl32_to_64 (w[ 3], w[ 2]);
2437 m[7] = hl32_to_64 (w[ 1], w[ 0]);
2439 m[0] = swap64 (m[0]);
2440 m[1] = swap64 (m[1]);
2441 m[2] = swap64 (m[2]);
2442 m[3] = swap64 (m[3]);
2443 m[4] = swap64 (m[4]);
2444 m[5] = swap64 (m[5]);
2445 m[6] = swap64 (m[6]);
2446 m[7] = swap64 (m[7]);
2448 // state buffer (hash)
2461 streebog_g (h, m, s_sbob_sl64);
2472 z[7] = swap64 ((u64) (pw_len * 8));
2474 streebog_g (h, z, s_sbob_sl64);
2475 streebog_g (h, m, s_sbob_sl64);
2477 const u32 r0 = l32_from_64 (h[0]);
2478 const u32 r1 = h32_from_64 (h[0]);
2479 const u32 r2 = l32_from_64 (h[1]);
2480 const u32 r3 = h32_from_64 (h[1]);
2486 __kernel void m11800_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
2490 __kernel void m11800_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
2494 __kernel void m11800_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
2500 const u32 gid = get_global_id (0);
2501 const u32 lid = get_local_id (0);
2502 const u32 lsz = get_local_size (0);
2505 * shared lookup table
2508 __local u64 s_sbob_sl64[8][256];
2510 for (u32 i = lid; i < 256; i += lsz)
2512 s_sbob_sl64[0][i] = sbob_sl64[0][i];
2513 s_sbob_sl64[1][i] = sbob_sl64[1][i];
2514 s_sbob_sl64[2][i] = sbob_sl64[2][i];
2515 s_sbob_sl64[3][i] = sbob_sl64[3][i];
2516 s_sbob_sl64[4][i] = sbob_sl64[4][i];
2517 s_sbob_sl64[5][i] = sbob_sl64[5][i];
2518 s_sbob_sl64[6][i] = sbob_sl64[6][i];
2519 s_sbob_sl64[7][i] = sbob_sl64[7][i];
2522 barrier (CLK_LOCAL_MEM_FENCE);
2524 if (gid >= gid_max) return;
2530 const u32 search[4] =
2532 digests_buf[digests_offset].digest_buf[DGST_R0],
2533 digests_buf[digests_offset].digest_buf[DGST_R1],
2534 digests_buf[digests_offset].digest_buf[DGST_R2],
2535 digests_buf[digests_offset].digest_buf[DGST_R3]
2544 wordl0[0] = pws[gid].i[ 0];
2545 wordl0[1] = pws[gid].i[ 1];
2546 wordl0[2] = pws[gid].i[ 2];
2547 wordl0[3] = pws[gid].i[ 3];
2551 wordl1[0] = pws[gid].i[ 4];
2552 wordl1[1] = pws[gid].i[ 5];
2553 wordl1[2] = pws[gid].i[ 6];
2554 wordl1[3] = pws[gid].i[ 7];
2570 const u32 pw_l_len = pws[gid].pw_len;
2572 if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
2574 append_0x80_2x4 (wordl0, wordl1, pw_l_len);
2576 switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
2583 for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
2585 const u32 pw_r_len = combs_buf[il_pos].pw_len;
2587 const u32 pw_len = pw_l_len + pw_r_len;
2591 wordr0[0] = combs_buf[il_pos].i[0];
2592 wordr0[1] = combs_buf[il_pos].i[1];
2593 wordr0[2] = combs_buf[il_pos].i[2];
2594 wordr0[3] = combs_buf[il_pos].i[3];
2598 wordr1[0] = combs_buf[il_pos].i[4];
2599 wordr1[1] = combs_buf[il_pos].i[5];
2600 wordr1[2] = combs_buf[il_pos].i[6];
2601 wordr1[3] = combs_buf[il_pos].i[7];
2617 if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
2619 switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
2624 w[ 0] = wordl0[0] | wordr0[0];
2625 w[ 1] = wordl0[1] | wordr0[1];
2626 w[ 2] = wordl0[2] | wordr0[2];
2627 w[ 3] = wordl0[3] | wordr0[3];
2628 w[ 4] = wordl1[0] | wordr1[0];
2629 w[ 5] = wordl1[1] | wordr1[1];
2630 w[ 6] = wordl1[2] | wordr1[2];
2631 w[ 7] = wordl1[3] | wordr1[3];
2632 w[ 8] = wordl2[0] | wordr2[0];
2633 w[ 9] = wordl2[1] | wordr2[1];
2634 w[10] = wordl2[2] | wordr2[2];
2635 w[11] = wordl2[3] | wordr2[3];
2636 w[12] = wordl3[0] | wordr3[0];
2637 w[13] = wordl3[1] | wordr3[1];
2638 w[14] = wordl3[1] | wordr3[1];
2639 w[15] = wordl3[1] | wordr3[1];
2641 append_0x01_4x4 (&w[0], &w[1], &w[2], &w[3], pw_len);
2644 * reverse message block
2649 m[0] = hl32_to_64 (w[15], w[14]);
2650 m[1] = hl32_to_64 (w[13], w[12]);
2651 m[2] = hl32_to_64 (w[11], w[10]);
2652 m[3] = hl32_to_64 (w[ 9], w[ 8]);
2653 m[4] = hl32_to_64 (w[ 7], w[ 6]);
2654 m[5] = hl32_to_64 (w[ 5], w[ 4]);
2655 m[6] = hl32_to_64 (w[ 3], w[ 2]);
2656 m[7] = hl32_to_64 (w[ 1], w[ 0]);
2658 m[0] = swap64 (m[0]);
2659 m[1] = swap64 (m[1]);
2660 m[2] = swap64 (m[2]);
2661 m[3] = swap64 (m[3]);
2662 m[4] = swap64 (m[4]);
2663 m[5] = swap64 (m[5]);
2664 m[6] = swap64 (m[6]);
2665 m[7] = swap64 (m[7]);
2667 // state buffer (hash)
2680 streebog_g (h, m, s_sbob_sl64);
2691 z[7] = swap64 ((u64) (pw_len * 8));
2693 streebog_g (h, z, s_sbob_sl64);
2694 streebog_g (h, m, s_sbob_sl64);
2696 const u32 r0 = l32_from_64 (h[0]);
2697 const u32 r1 = h32_from_64 (h[0]);
2698 const u32 r2 = l32_from_64 (h[1]);
2699 const u32 r3 = h32_from_64 (h[1]);
2705 __kernel void m11800_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
2709 __kernel void m11800_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)