/**
 * Author......: See docs/credits.txt
 * License.....: MIT
 */

//shared mem too small
//#define NEW_SIMD_CODE

#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_simd.cl)
#include M2S(INCLUDE_PATH/inc_hash_md4.cl)
#include M2S(INCLUDE_PATH/inc_hash_md5.cl)
#include M2S(INCLUDE_PATH/inc_cipher_rc4.cl)
#endif

typedef struct krb5pa
{
  u32 user[16];
  u32 realm[16];
  u32 salt[32];
  u32 timestamp[16];
  u32 checksum[4];

} krb5pa_t;

DECLSPEC int decrypt_and_check (LOCAL_AS u32 *S, PRIVATE_AS u32 *data, PRIVATE_AS u32 *timestamp_ct, const u64 lid)
{
  rc4_init_128 (S, data, lid);

  u32 out[4];

  u8 j = 0;

  j = rc4_next_16 (S,  0, j, timestamp_ct + 0, out, lid);

  if ((out[3] & 0xffff0000) != 0x30320000) return 0;

  j = rc4_next_16 (S, 16, j, timestamp_ct + 4, out, lid);

  if (((out[0] & 0xff) < '0') || ((out[0] & 0xff) > '9')) return 0; out[0] >>= 8;
  if (((out[0] & 0xff) < '0') || ((out[0] & 0xff) > '9')) return 0; out[0] >>= 8;
  if (((out[0] & 0xff) < '0') || ((out[0] & 0xff) > '9')) return 0; out[0] >>= 8;
  if (((out[0] & 0xff) < '0') || ((out[0] & 0xff) > '9')) return 0;
  if (((out[1] & 0xff) < '0') || ((out[1] & 0xff) > '9')) return 0; out[1] >>= 8;
  if (((out[1] & 0xff) < '0') || ((out[1] & 0xff) > '9')) return 0; out[1] >>= 8;
  if (((out[1] & 0xff) < '0') || ((out[1] & 0xff) > '9')) return 0; out[1] >>= 8;
  if (((out[1] & 0xff) < '0') || ((out[1] & 0xff) > '9')) return 0;
  if (((out[2] & 0xff) < '0') || ((out[2] & 0xff) > '9')) return 0; out[2] >>= 8;
  if (((out[2] & 0xff) < '0') || ((out[2] & 0xff) > '9')) return 0; out[2] >>= 8;
  if (((out[2] & 0xff) < '0') || ((out[2] & 0xff) > '9')) return 0; out[2] >>= 8;
  if (((out[2] & 0xff) < '0') || ((out[2] & 0xff) > '9')) return 0;

  return 1;
}

DECLSPEC void hmac_md5_pad (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, PRIVATE_AS u32 *ipad, PRIVATE_AS u32 *opad)
{
  w0[0] = w0[0] ^ 0x36363636;
  w0[1] = w0[1] ^ 0x36363636;
  w0[2] = w0[2] ^ 0x36363636;
  w0[3] = w0[3] ^ 0x36363636;
  w1[0] = w1[0] ^ 0x36363636;
  w1[1] = w1[1] ^ 0x36363636;
  w1[2] = w1[2] ^ 0x36363636;
  w1[3] = w1[3] ^ 0x36363636;
  w2[0] = w2[0] ^ 0x36363636;
  w2[1] = w2[1] ^ 0x36363636;
  w2[2] = w2[2] ^ 0x36363636;
  w2[3] = w2[3] ^ 0x36363636;
  w3[0] = w3[0] ^ 0x36363636;
  w3[1] = w3[1] ^ 0x36363636;
  w3[2] = w3[2] ^ 0x36363636;
  w3[3] = w3[3] ^ 0x36363636;

  ipad[0] = MD5M_A;
  ipad[1] = MD5M_B;
  ipad[2] = MD5M_C;
  ipad[3] = MD5M_D;

  md5_transform (w0, w1, w2, w3, ipad);

  w0[0] = w0[0] ^ 0x6a6a6a6a;
  w0[1] = w0[1] ^ 0x6a6a6a6a;
  w0[2] = w0[2] ^ 0x6a6a6a6a;
  w0[3] = w0[3] ^ 0x6a6a6a6a;
  w1[0] = w1[0] ^ 0x6a6a6a6a;
  w1[1] = w1[1] ^ 0x6a6a6a6a;
  w1[2] = w1[2] ^ 0x6a6a6a6a;
  w1[3] = w1[3] ^ 0x6a6a6a6a;
  w2[0] = w2[0] ^ 0x6a6a6a6a;
  w2[1] = w2[1] ^ 0x6a6a6a6a;
  w2[2] = w2[2] ^ 0x6a6a6a6a;
  w2[3] = w2[3] ^ 0x6a6a6a6a;
  w3[0] = w3[0] ^ 0x6a6a6a6a;
  w3[1] = w3[1] ^ 0x6a6a6a6a;
  w3[2] = w3[2] ^ 0x6a6a6a6a;
  w3[3] = w3[3] ^ 0x6a6a6a6a;

  opad[0] = MD5M_A;
  opad[1] = MD5M_B;
  opad[2] = MD5M_C;
  opad[3] = MD5M_D;

  md5_transform (w0, w1, w2, w3, opad);
}

DECLSPEC void hmac_md5_run (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, PRIVATE_AS u32 *ipad, PRIVATE_AS u32 *opad, PRIVATE_AS u32 *digest)
{
  digest[0] = ipad[0];
  digest[1] = ipad[1];
  digest[2] = ipad[2];
  digest[3] = ipad[3];

  md5_transform (w0, w1, w2, w3, digest);

  w0[0] = digest[0];
  w0[1] = digest[1];
  w0[2] = digest[2];
  w0[3] = digest[3];
  w1[0] = 0x80;
  w1[1] = 0;
  w1[2] = 0;
  w1[3] = 0;
  w2[0] = 0;
  w2[1] = 0;
  w2[2] = 0;
  w2[3] = 0;
  w3[0] = 0;
  w3[1] = 0;
  w3[2] = (64 + 16) * 8;
  w3[3] = 0;

  digest[0] = opad[0];
  digest[1] = opad[1];
  digest[2] = opad[2];
  digest[3] = opad[3];

  md5_transform (w0, w1, w2, w3, digest);
}

DECLSPEC void kerb_prepare (PRIVATE_AS const u32 *w0, PRIVATE_AS const u32 *w1, const u32 pw_len, PRIVATE_AS const u32 *checksum, PRIVATE_AS u32 *digest)
{
  /**
   * pads
   */

  u32 w0_t[4];
  u32 w1_t[4];
  u32 w2_t[4];
  u32 w3_t[4];

  w0_t[0] = w0[0];
  w0_t[1] = w0[1];
  w0_t[2] = w0[2];
  w0_t[3] = w0[3];
  w1_t[0] = w1[0];
  w1_t[1] = w1[1];
  w1_t[2] = w1[2];
  w1_t[3] = w1[3];
  w2_t[0] = 0;
  w2_t[1] = 0;
  w2_t[2] = 0;
  w2_t[3] = 0;
  w3_t[0] = 0;
  w3_t[1] = 0;
  w3_t[2] = 0;
  w3_t[3] = 0;

  // K=MD4(Little_indian(UNICODE(pwd))

  append_0x80_2x4 (w0_t, w1_t, pw_len);

  make_utf16le (w1_t, w2_t, w3_t);
  make_utf16le (w0_t, w0_t, w1_t);

  w3_t[2] = pw_len * 8 * 2;
  w3_t[3] = 0;

  digest[0] = MD4M_A;
  digest[1] = MD4M_B;
  digest[2] = MD4M_C;
  digest[3] = MD4M_D;

  md4_transform (w0_t, w1_t, w2_t, w3_t, digest);

  // K1=MD5_HMAC(K,1); with 1 encoded as little indian on 4 bytes (01000000 in hexa);

  w0_t[0] = digest[0];
  w0_t[1] = digest[1];
  w0_t[2] = digest[2];
  w0_t[3] = digest[3];
  w1_t[0] = 0;
  w1_t[1] = 0;
  w1_t[2] = 0;
  w1_t[3] = 0;
  w2_t[0] = 0;
  w2_t[1] = 0;
  w2_t[2] = 0;
  w2_t[3] = 0;
  w3_t[0] = 0;
  w3_t[1] = 0;
  w3_t[2] = 0;
  w3_t[3] = 0;

  u32 ipad[4];
  u32 opad[4];

  hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad);

  w0_t[0] = 1;
  w0_t[1] = 0x80;
  w0_t[2] = 0;
  w0_t[3] = 0;
  w1_t[0] = 0;
  w1_t[1] = 0;
  w1_t[2] = 0;
  w1_t[3] = 0;
  w2_t[0] = 0;
  w2_t[1] = 0;
  w2_t[2] = 0;
  w2_t[3] = 0;
  w3_t[0] = 0;
  w3_t[1] = 0;
  w3_t[2] = (64 + 4) * 8;
  w3_t[3] = 0;

  hmac_md5_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest);

  // K3=MD5_HMAC(K1,checksum);

  w0_t[0] = digest[0];
  w0_t[1] = digest[1];
  w0_t[2] = digest[2];
  w0_t[3] = digest[3];
  w1_t[0] = 0;
  w1_t[1] = 0;
  w1_t[2] = 0;
  w1_t[3] = 0;
  w2_t[0] = 0;
  w2_t[1] = 0;
  w2_t[2] = 0;
  w2_t[3] = 0;
  w3_t[0] = 0;
  w3_t[1] = 0;
  w3_t[2] = 0;
  w3_t[3] = 0;

  hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad);

  w0_t[0] = checksum[0];
  w0_t[1] = checksum[1];
  w0_t[2] = checksum[2];
  w0_t[3] = checksum[3];
  w1_t[0] = 0x80;
  w1_t[1] = 0;
  w1_t[2] = 0;
  w1_t[3] = 0;
  w2_t[0] = 0;
  w2_t[1] = 0;
  w2_t[2] = 0;
  w2_t[3] = 0;
  w3_t[0] = 0;
  w3_t[1] = 0;
  w3_t[2] = (64 + 16) * 8;
  w3_t[3] = 0;

  hmac_md5_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest);
}

DECLSPEC void m07500 (LOCAL_AS u32 *S, PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 pw_len, KERN_ATTR_FUNC_ESALT (krb5pa_t))
{
  /**
   * modifiers are taken from args
   */

  /**
   * salt
   */

  u32 checksum[4];

  checksum[0] = esalt_bufs[DIGESTS_OFFSET_HOST].checksum[0];
  checksum[1] = esalt_bufs[DIGESTS_OFFSET_HOST].checksum[1];
  checksum[2] = esalt_bufs[DIGESTS_OFFSET_HOST].checksum[2];
  checksum[3] = esalt_bufs[DIGESTS_OFFSET_HOST].checksum[3];

  u32 timestamp_ct[8];

  timestamp_ct[0] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[0];
  timestamp_ct[1] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[1];
  timestamp_ct[2] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[2];
  timestamp_ct[3] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[3];
  timestamp_ct[4] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[4];
  timestamp_ct[5] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[5];
  timestamp_ct[6] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[6];
  timestamp_ct[7] = esalt_bufs[DIGESTS_OFFSET_HOST].timestamp[7];

  /**
   * loop
   */

  u32 w0l = w0[0];

  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
  {
    const u32x w0r = ix_create_bft (bfs_buf, il_pos);

    const u32x w0lr = w0l | w0r;

    u32x w0_t[4];
    u32x w1_t[4];
    u32x w2_t[4];
    u32x w3_t[4];

    w0_t[0] = w0lr;
    w0_t[1] = w0[1];
    w0_t[2] = w0[2];
    w0_t[3] = w0[3];
    w1_t[0] = w1[0];
    w1_t[1] = w1[1];
    w1_t[2] = w1[2];
    w1_t[3] = w1[3];
    w2_t[0] = w2[0];
    w2_t[1] = w2[1];
    w2_t[2] = w2[2];
    w2_t[3] = w2[3];
    w3_t[0] = w3[0];
    w3_t[1] = w3[1];
    w3_t[2] = w3[2];
    w3_t[3] = w3[3];

    /**
     * kerberos
     */

    u32 digest[4];

    kerb_prepare (w0_t, w1_t, pw_len, checksum, digest);

    u32 tmp[4];

    tmp[0] = digest[0];
    tmp[1] = digest[1];
    tmp[2] = digest[2];
    tmp[3] = digest[3];

    if (decrypt_and_check (S, tmp, timestamp_ct, lid) == 1)
    {
      if (hc_atomic_inc (&hashes_shown[DIGESTS_OFFSET_HOST]) == 0)
      {
        mark_hash (plains_buf, d_return_buf, SALT_POS_HOST, DIGESTS_CNT, 0, DIGESTS_OFFSET_HOST + 0, gid, il_pos, 0, 0);
      }
    }
  }
}

KERNEL_FQ void m07500_m04 (KERN_ATTR_ESALT (krb5pa_t))
{
  /**
   * base
   */

  const u64 lid = get_local_id (0);
  const u64 gid = get_global_id (0);
  const u64 lsz = get_local_size (0);

  if (gid >= GID_CNT) return;

  u32 w0[4];

  w0[0] = pws[gid].i[ 0];
  w0[1] = pws[gid].i[ 1];
  w0[2] = pws[gid].i[ 2];
  w0[3] = pws[gid].i[ 3];

  u32 w1[4];

  w1[0] = 0;
  w1[1] = 0;
  w1[2] = 0;
  w1[3] = 0;

  u32 w2[4];

  w2[0] = 0;
  w2[1] = 0;
  w2[2] = 0;
  w2[3] = 0;

  u32 w3[4];

  w3[0] = 0;
  w3[1] = 0;
  w3[2] = 0;
  w3[3] = 0;

  const u32 pw_len = pws[gid].pw_len & 63;

  /**
   * main
   */

  LOCAL_VK u32 S[64 * FIXED_LOCAL_SIZE];

  m07500 (S, w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, kernel_param, gid, lid, lsz);
}

KERNEL_FQ void m07500_m08 (KERN_ATTR_ESALT (krb5pa_t))
{
  /**
   * base
   */

  const u64 lid = get_local_id (0);
  const u64 gid = get_global_id (0);
  const u64 lsz = get_local_size (0);

  if (gid >= GID_CNT) return;

  u32 w0[4];

  w0[0] = pws[gid].i[ 0];
  w0[1] = pws[gid].i[ 1];
  w0[2] = pws[gid].i[ 2];
  w0[3] = pws[gid].i[ 3];

  u32 w1[4];

  w1[0] = pws[gid].i[ 4];
  w1[1] = pws[gid].i[ 5];
  w1[2] = pws[gid].i[ 6];
  w1[3] = pws[gid].i[ 7];

  u32 w2[4];

  w2[0] = 0;
  w2[1] = 0;
  w2[2] = 0;
  w2[3] = 0;

  u32 w3[4];

  w3[0] = 0;
  w3[1] = 0;
  w3[2] = 0;
  w3[3] = 0;

  const u32 pw_len = pws[gid].pw_len & 63;

  /**
   * main
   */

  LOCAL_VK u32 S[64 * FIXED_LOCAL_SIZE];

  m07500 (S, w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, kernel_param, gid, lid, lsz);
}

KERNEL_FQ void m07500_m16 (KERN_ATTR_ESALT (krb5pa_t))
{
}

KERNEL_FQ void m07500_s04 (KERN_ATTR_ESALT (krb5pa_t))
{
  /**
   * base
   */

  const u64 lid = get_local_id (0);
  const u64 gid = get_global_id (0);
  const u64 lsz = get_local_size (0);

  if (gid >= GID_CNT) return;

  u32 w0[4];

  w0[0] = pws[gid].i[ 0];
  w0[1] = pws[gid].i[ 1];
  w0[2] = pws[gid].i[ 2];
  w0[3] = pws[gid].i[ 3];

  u32 w1[4];

  w1[0] = 0;
  w1[1] = 0;
  w1[2] = 0;
  w1[3] = 0;

  u32 w2[4];

  w2[0] = 0;
  w2[1] = 0;
  w2[2] = 0;
  w2[3] = 0;

  u32 w3[4];

  w3[0] = 0;
  w3[1] = 0;
  w3[2] = 0;
  w3[3] = 0;

  const u32 pw_len = pws[gid].pw_len & 63;

  /**
   * main
   */

  LOCAL_VK u32 S[64 * FIXED_LOCAL_SIZE];

  m07500 (S, w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, kernel_param, gid, lid, lsz);
}

KERNEL_FQ void m07500_s08 (KERN_ATTR_ESALT (krb5pa_t))
{
  /**
   * base
   */

  const u64 lid = get_local_id (0);
  const u64 gid = get_global_id (0);
  const u64 lsz = get_local_size (0);

  if (gid >= GID_CNT) return;

  u32 w0[4];

  w0[0] = pws[gid].i[ 0];
  w0[1] = pws[gid].i[ 1];
  w0[2] = pws[gid].i[ 2];
  w0[3] = pws[gid].i[ 3];

  u32 w1[4];

  w1[0] = pws[gid].i[ 4];
  w1[1] = pws[gid].i[ 5];
  w1[2] = pws[gid].i[ 6];
  w1[3] = pws[gid].i[ 7];

  u32 w2[4];

  w2[0] = 0;
  w2[1] = 0;
  w2[2] = 0;
  w2[3] = 0;

  u32 w3[4];

  w3[0] = 0;
  w3[1] = 0;
  w3[2] = 0;
  w3[3] = 0;

  const u32 pw_len = pws[gid].pw_len & 63;

  /**
   * main
   */

  LOCAL_VK u32 S[64 * FIXED_LOCAL_SIZE];

  m07500 (S, w0, w1, w2, w3, pw_len, pws, rules_buf, combs_buf, bfs_buf, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, kernel_param, gid, lid, lsz);
}

KERNEL_FQ void m07500_s16 (KERN_ATTR_ESALT (krb5pa_t))
{
}