/**
 * Author......: Netherlands Forensic Institute
 *                based upon 17010
 * License.....: MIT
 */

//#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_hash_sha1.cl)
#include M2S(INCLUDE_PATH/inc_cipher_cast.h)
#endif

typedef struct gpg
{
  u32 cipher_algo;
  u32 iv[4]; // make this dynamic based on the input hash.. iv_size can be 8 bytes or 16 bytes
  u32 modulus_size;
  u32 encrypted_data[384];
  u32 encrypted_data_size;

} gpg_t;

typedef struct gpg_tmp
{
  // buffer for a maximum of 256 + 8 characters, we extend it to 320 characters so it's always 64 byte aligned
  u32 salted_pw_block[80];
  // actual number of bytes in 'salted_pwd' that are used since salt and password are copied multiple times into the buffer
  u32 salted_pw_block_len;

  u32 h[10];

  u32 w0[4];
  u32 w1[4];
  u32 w2[4];
  u32 w3[4];

  u32 len;

} gpg_tmp_t;


DECLSPEC u32 hc_bytealign_le_S (const u32 a, const u32 b, const int c)
{
  const int c_mod_4 = c & 3;

  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 0) || defined IS_GENERIC
  const u32 r = l32_from_64_S ((v64_from_v32ab_S (b, a) >> (c_mod_4 * 8)));
  #endif

  #if ((defined IS_AMD || defined IS_HIP) && HAS_VPERM == 1) || defined IS_NV

  #if defined IS_NV
  const int selector = (0x76543210 >> (c_mod_4 * 4)) & 0xffff;
  #endif

  #if (defined IS_AMD || defined IS_HIP)
  const int selector = l32_from_64_S (0x0706050403020100UL >> (c_mod_4 * 8));
  #endif

  const u32 r = hc_byte_perm (b, a, selector);
  #endif

  return r;
}

DECLSPEC void memcat_le_S (PRIVATE_AS u32 *block, const u32 offset, PRIVATE_AS const u32 *append, u32 len)
{
  const u32 start_index = (offset - 1) >> 2;
  const u32 count = ((offset + len + 3) >> 2) - start_index;
  const int off_mod_4 = offset & 3;
  const int off_minus_4 = 4 - off_mod_4;

  block[start_index] |= hc_bytealign_le_S (append[0], 0, off_minus_4);

  for (u32 idx = 1; idx < count; idx++)
  {
    block[start_index + idx] = hc_bytealign_le_S (append[idx], append[idx - 1], off_minus_4);
  }
}

DECLSPEC void memzero_le_S (PRIVATE_AS u32 *block, const u32 start_offset, const u32 end_offset)
{
  const u32 start_idx = start_offset / 4;

  // zero out bytes in the first u32 starting from 'start_offset'
  // math is a bit complex to avoid shifting by 32 bits, which is not possible on some architectures
  block[start_idx] &= ~(0xffffffff << ((start_offset & 3) * 8));

  const u32 end_idx = (end_offset + 3) / 4;

  // zero out bytes in u32 units -- note that the last u32 is completely zeroed!
  for (u32 i = start_idx + 1; i < end_idx; i++)
  {
    block[i] = 0;
  }
}

DECLSPEC void memzero_be_S (PRIVATE_AS u32 *block, const u32 start_offset, const u32 end_offset)
{
  const u32 start_idx = start_offset / 4;

  // zero out bytes in the first u32 starting from 'start_offset'
  // math is a bit complex to avoid shifting by 32 bits, which is not possible on some architectures
  block[start_idx] &= ~(0xffffffff >> ((start_offset & 3) * 8));

  const u32 end_idx = (end_offset + 3) / 4;

  // zero out bytes in u32 units -- note that the last u32 is completely zeroed!
  for (u32 i = start_idx + 1; i < end_idx; i++)
  {
    block[i] = 0;
  }
}

DECLSPEC void cast128_decrypt_cfb (GLOBAL_AS const u32 *encrypted_data, int data_len, PRIVATE_AS const u32 *iv, PRIVATE_AS const u32 *key, PRIVATE_AS u32 *decrypted_data)
{
  u8 essiv[8];
  for (int j=0; j<8; j++) { essiv[j] = 0; }

  // TODO remove this casting, would speedup the attack
  // We need to do this casting to get values in local memory and have them not be constant.
  u32 lencrypted_data[384]; // I'd prefer not to hardcode to 384,  but rest of kernel uses the same value
  for (u32 i = 0; i < (data_len + 3) / 4; i += 4)
  {
    lencrypted_data[i + 0] = encrypted_data[i + 0];
    lencrypted_data[i + 1] = encrypted_data[i + 1];
    lencrypted_data[i + 2] = encrypted_data[i + 2];
    lencrypted_data[i + 3] = encrypted_data[i + 3];
  }
  u8 *lencrypted_data8 =  (u8*)lencrypted_data;
  u8 *decrypted_data8 = (u8*)decrypted_data;
  u8 *key8 = (u8*)key;


  // Copy the IV, since this will be modified
  // essiv[0] = iv[0];  // IV is zero for our example, but we load it dynamically..
  // essiv[1] = iv[1];  // IV is zero for our example, but we load it dynamically..
  // essiv[2] = 0;
  // essiv[3] = 0; //TODO load IV dynamically, code doesn't make any sense currently as essiv is now a u8

  CAST_KEY ck;
  Cast5SetKey(&ck, 16, key8);

  // Decrypt an CAST5 encrypted block
  for (u32 i = 0; i < (data_len + 3) ; i += 8)
  {
    Cast5Encrypt(essiv, &decrypted_data8[i], &ck);

    for (int j=0; j<8; j++) { decrypted_data8[i+j] ^= lencrypted_data8[i + j]; }

    // Note: Not necessary if you are only decrypting a single block!
    for (int j=0; j<8; j++) {
      essiv[j] = lencrypted_data8[i + j];
    }
  }
}

DECLSPEC int check_decoded_data (PRIVATE_AS u32 *decoded_data, const u32 decoded_data_size)
{
  // Check the SHA-1 of the decrypted data which is stored at the end of the decrypted data
  const u32 sha1_byte_off = (decoded_data_size - 20);
  const u32 sha1_u32_off = sha1_byte_off / 4;

  u32 expected_sha1[5];

  expected_sha1[0] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 1], decoded_data[sha1_u32_off + 0], sha1_byte_off);
  expected_sha1[1] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 2], decoded_data[sha1_u32_off + 1], sha1_byte_off);
  expected_sha1[2] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 3], decoded_data[sha1_u32_off + 2], sha1_byte_off);
  expected_sha1[3] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 4], decoded_data[sha1_u32_off + 3], sha1_byte_off);
  expected_sha1[4] = hc_bytealign_le_S (decoded_data[sha1_u32_off + 5], decoded_data[sha1_u32_off + 4], sha1_byte_off);



  memzero_le_S (decoded_data, sha1_byte_off, 384 * sizeof(u32));

  sha1_ctx_t ctx;

  sha1_init (&ctx);

  sha1_update_swap (&ctx, decoded_data, sha1_byte_off);

  sha1_final (&ctx);

  return (expected_sha1[0] == hc_swap32_S (ctx.h[0]))
      && (expected_sha1[1] == hc_swap32_S (ctx.h[1]))
      && (expected_sha1[2] == hc_swap32_S (ctx.h[2]))
      && (expected_sha1[3] == hc_swap32_S (ctx.h[3]))
      && (expected_sha1[4] == hc_swap32_S (ctx.h[4]));
}

KERNEL_FQ void m17040_init (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
{
  const u64 gid = get_global_id (0);

  if (gid >= GID_CNT) return;

  const u32 pw_len = pws[gid].pw_len;
  const u32 salted_pw_len = (salt_bufs[SALT_POS_HOST].salt_len + pw_len);

  u32 salted_pw_block[80];

  // concatenate salt and password -- the salt is always 8 bytes
  salted_pw_block[0] = salt_bufs[SALT_POS_HOST].salt_buf[0];
  salted_pw_block[1] = salt_bufs[SALT_POS_HOST].salt_buf[1];

  for (u32 idx = 0; idx < 64; idx++) salted_pw_block[idx + 2] = pws[gid].i[idx];

  // zero remainder of buffer
  for (u32 idx = 66; idx < 80; idx++) salted_pw_block[idx] = 0;

  // create a number of copies for efficiency
  const u32 copies = 80 * sizeof(u32) / salted_pw_len;

  for (u32 idx = 1; idx < copies; idx++)
  {
    memcat_le_S (salted_pw_block, idx * salted_pw_len, salted_pw_block, salted_pw_len);
  }

  for (u32 idx = 0; idx < 80; idx++)
  {
    tmps[gid].salted_pw_block[idx] = hc_swap32_S (salted_pw_block[idx]);
  }

  tmps[gid].salted_pw_block_len = (copies * salted_pw_len);

  tmps[gid].h[0] = SHA1M_A;
  tmps[gid].h[1] = SHA1M_B;
  tmps[gid].h[2] = SHA1M_C;
  tmps[gid].h[3] = SHA1M_D;
  tmps[gid].h[4] = SHA1M_E;
  tmps[gid].h[5] = SHA1M_A;
  tmps[gid].h[6] = SHA1M_B;
  tmps[gid].h[7] = SHA1M_C;
  tmps[gid].h[8] = SHA1M_D;
  tmps[gid].h[9] = SHA1M_E;

  tmps[gid].len = 0;
}

KERNEL_FQ void m17040_loop_prepare (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
{
  const u64 gid = get_global_id (0);

  if (gid >= GID_CNT) return;

  tmps[gid].h[0] = SHA1M_A;
  tmps[gid].h[1] = SHA1M_B;
  tmps[gid].h[2] = SHA1M_C;
  tmps[gid].h[3] = SHA1M_D;
  tmps[gid].h[4] = SHA1M_E;
  tmps[gid].h[5] = SHA1M_A;
  tmps[gid].h[6] = SHA1M_B;
  tmps[gid].h[7] = SHA1M_C;
  tmps[gid].h[8] = SHA1M_D;
  tmps[gid].h[9] = SHA1M_E;

  tmps[gid].w0[0] = 0;
  tmps[gid].w0[1] = 0;
  tmps[gid].w0[2] = 0;
  tmps[gid].w0[3] = 0;
  tmps[gid].w1[0] = 0;
  tmps[gid].w1[1] = 0;
  tmps[gid].w1[2] = 0;
  tmps[gid].w1[3] = 0;
  tmps[gid].w2[0] = 0;
  tmps[gid].w2[1] = 0;
  tmps[gid].w2[2] = 0;
  tmps[gid].w2[3] = 0;
  tmps[gid].w3[0] = 0;
  tmps[gid].w3[1] = 0;
  tmps[gid].w3[2] = 0;
  tmps[gid].w3[3] = 0;

  tmps[gid].len = 0;
}

KERNEL_FQ void m17040_loop (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
{
  const u64 gid = get_global_id (0);
  const u64 lid = get_local_id (0);

  if (gid >= GID_CNT) return;

  // get the prepared buffer from the gpg_tmp_t struct into a local buffer
  u32 salted_pw_block[80];
  for (int i = 0; i < 80; i++) salted_pw_block[i] = tmps[gid].salted_pw_block[i];


  const u32 salted_pw_block_len = tmps[gid].salted_pw_block_len;

  // do we really need this, since the salt is always length 8?
  if (salted_pw_block_len == 0) return;

  /**
   * context load
   */

  sha1_ctx_t ctx;

  for (int i = 0; i < 5; i++) ctx.h[i] = tmps[gid].h[i];

  for (int i = 0; i < 4; i++) ctx.w0[i] = tmps[gid].w0[i];
  for (int i = 0; i < 4; i++) ctx.w1[i] = tmps[gid].w1[i];
  for (int i = 0; i < 4; i++) ctx.w2[i] = tmps[gid].w2[i];
  for (int i = 0; i < 4; i++) ctx.w3[i] = tmps[gid].w3[i];

  const u32 pw_len = pws[gid].pw_len;
  const u32 salted_pw_len = (salt_bufs[SALT_POS_HOST].salt_len + pw_len);
  const u32 remaining_bytes = salted_pw_len % 4;

  ctx.len = tmps[gid].len;

  memzero_be_S (salted_pw_block, salted_pw_len, salted_pw_block_len);
  // zero out last bytes of password if not a multiple of 4
  // TODO do we need this wo don't feed the remainder to the hashing algorithm anyway..??
  sha1_update (&ctx, salted_pw_block, salted_pw_len);
  sha1_final (&ctx);

  /**
   * context save
   */

  for (int i = 0; i < 5; i++) tmps[gid].h[i] = ctx.h[i];
  // this is the sha1 hash of the salt+password:

  for (int i = 0; i < 4; i++) tmps[gid].w0[i] = ctx.w0[i];
  for (int i = 0; i < 4; i++) tmps[gid].w1[i] = ctx.w1[i];
  for (int i = 0; i < 4; i++) tmps[gid].w2[i] = ctx.w2[i];
  for (int i = 0; i < 4; i++) tmps[gid].w3[i] = ctx.w3[i];

  tmps[gid].len = ctx.len;
}

KERNEL_FQ void m17040_comp (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
{
  // not in use here, special case...
}

KERNEL_FQ void m17040_aux1 (KERN_ATTR_TMPS_ESALT (gpg_tmp_t, gpg_t))
{
  /**
   * modifier
   */

  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;

  // retrieve and use the SHA-1 as the key for CAST5
  u32 cast_key[5];
  for (int i = 0; i < 5; i++) cast_key[i] = hc_swap32_S (tmps[gid].h[i]);

  u32 iv[4] = {0};
  for (int idx = 0; idx < 4; idx++) iv[idx] = esalt_bufs[DIGESTS_OFFSET_HOST].iv[idx];

  u32 decoded_data[384];

  const u32 enc_data_size = esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data_size;

  cast128_decrypt_cfb (esalt_bufs[DIGESTS_OFFSET_HOST].encrypted_data, enc_data_size, iv, cast_key, decoded_data);

  if (check_decoded_data (decoded_data, enc_data_size))
  {
    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, 0, 0, 0);
    }
  }
}