path: root/aes_keywrap.c

/*
 * aes_keywrap.c
 * -------------
 * Implementation of RFC 5649 over Cryptech AES core.
 *
 * Authors: Rob Austein
 * Copyright (c) 2015-2018, NORDUnet A/S
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met:
 * - Redistributions of source code must retain the above copyright notice,
 *   this list of conditions and the following disclaimer.
 *
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 *
 * - Neither the name of the NORDUnet nor the names of its contributors may
 *   be used to endorse or promote products derived from this software
 *   without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
 * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * Note that there are two different block sizes involved here: the
 * key wrap algorithm deals entirely with 64-bit blocks, while AES
 * itself deals with 128-bit blocks.  In practice, this is not as
 * confusing as it sounds, because we combine two 64-bit blocks to
 * create one 128-bit block just prior to performing an AES operation,
 * then split the result back to 64-bit blocks immediately afterwards.
 */

#include <stdint.h>
#include <string.h>

#include "hal.h"
#include "hal_internal.h"

/*
 * Enable use of the experimental keywrap core, if present.
 */

static int use_keywrap_core = 1;

int hal_aes_use_keywrap_core(int onoff)
{
  use_keywrap_core = (onoff && hal_core_find(KEYWRAP_NAME, NULL) != NULL);
  return use_keywrap_core;
}


/*
 * How long the ciphertext will be for a given plaintext length.
 * This rounds up the length to a multiple of 8, and adds 8 for the IV.
 */

size_t hal_aes_keywrap_ciphertext_length(const size_t plaintext_length)
{
  return (plaintext_length + 15) & ~7;
}


/*
 * Check the KEK, then load it into the AES core.
 * Note that our AES core only supports 128 and 256 bit keys.
 *
 * This should work without modification for the experimental keywrap core,
 * because ADDR_KEY0 and ADDR_CONFIG are at the same offsets.
 */

typedef enum { key_encrypting, key_decrypting } key_action_t;

static hal_error_t load_kek(const hal_core_t *core, const uint8_t *K, const size_t K_len, const key_action_t action)
{
  uint8_t config[4];
  hal_error_t err;

  if (K == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  memset(config, 0, sizeof(config));

  switch (K_len) {
  case bitsToBytes(128):
    config[3] &= ~AES_CONFIG_KEYLEN;
    break;
  case bitsToBytes(256):
    config[3] |=  AES_CONFIG_KEYLEN;
    break;
  case bitsToBytes(192):
    return HAL_ERROR_UNSUPPORTED_KEY;
  default:
    return HAL_ERROR_BAD_ARGUMENTS;
  }

  switch (action) {
  case key_encrypting:
    config[3] |=  AES_CONFIG_ENCDEC;
    break;
  case key_decrypting:
    config[3] &= ~AES_CONFIG_ENCDEC;
    break;
  default:
    return HAL_ERROR_BAD_ARGUMENTS;
  }

  /*
   * Load the KEK and tell the core to expand it.
   */

  if ((err = hal_io_write(core, AES_ADDR_KEY0, K, K_len))                 != HAL_OK ||
      (err = hal_io_write(core, AES_ADDR_CONFIG, config, sizeof(config))) != HAL_OK ||
      (err = hal_io_init(core))                                           != HAL_OK)
    return err;

  return HAL_OK;
}


/*
 * Use the experimental keywrap core to wrap/unwrap len bytes of plaintext.
 * The wrapped/unwrapped key is returned in the same buffer.
 */

static hal_error_t do_keywrap_core(const hal_core_t *core, uint8_t * const C, const size_t len, const key_action_t action)
{
#ifndef min
#define min(a,b) ((a) < (b) ? (a) : (b))
#endif

  hal_error_t err;

  hal_assert(core != NULL && C != NULL && len > 0);

  const size_t calculated_C_len = hal_aes_keywrap_ciphertext_length(len);
  /* KEYWRAP_LEN_R_DATA is the number of 4-byte data registers in the core. */
  if (calculated_C_len - 8 > KEYWRAP_LEN_R_DATA * 4)
    return HAL_ERROR_BAD_ARGUMENTS;

  if (action == key_decrypting) {
    /* write the first 8 bytes to A */
    if ((err = hal_io_write(core, KEYWRAP_ADDR_A0, C, 8)) != HAL_OK)
      return err;
  }

  /* write the unpadded length to LENGTH */
  uint32_t nel = htonl(len);
  if ((err = hal_io_write(core, KEYWRAP_ADDR_LENGTH, (const uint8_t *)&nel, 4)) != HAL_OK)
    return err;

  /* write the data to R_DATA */
  if ((err = hal_io_write(core, KEYWRAP_ADDR_R_DATA, C + 8, calculated_C_len - 8)) != HAL_OK)
      return err;

  /* start the wrap/unwrap operation, and wait for it to complete */
  if ((err = hal_io_next(core)) != HAL_OK ||
      (err = hal_io_wait_ready(core)) != HAL_OK)
    return err;

  /* read the A registers */
  if ((err = hal_io_read(core, KEYWRAP_ADDR_A0, C, 8)) != HAL_OK)
    return err;

  /* read the data from R_DATA */
  if ((err = hal_io_read(core, KEYWRAP_ADDR_R_DATA, C + 8, calculated_C_len - 8)) != HAL_OK)
      return err;

  return HAL_OK;
}


/*
 * Process one block.  Since AES Key Wrap always deals with 64-bit
 * half blocks and since the bus is going to break this up into 32-bit
 * words no matter what we do, we can eliminate a few gratuitous
 * memcpy() operations by receiving our arguments as two half blocks.
 *
 * Since the length of these half blocks is constant, there's no real
 * point in passing the length as an argument, we'd just be checking a
 * constant against a constant and a smart compiler will optimize
 * the whole check out.
 *
 * Just be VERY careful if you change anything here.
 */

static hal_error_t do_block(const hal_core_t *core, uint8_t *b1, uint8_t *b2)
{
  hal_error_t err;

  hal_assert(b1 != NULL && b2 != NULL);

  if ((err = hal_io_write(core, AES_ADDR_BLOCK0, b1, 8)) != HAL_OK ||
      (err = hal_io_write(core, AES_ADDR_BLOCK2, b2, 8)) != HAL_OK ||
      (err = hal_io_next(core))                          != HAL_OK ||
      (err = hal_io_wait_ready(core))                    != HAL_OK ||
      (err = hal_io_read(core, AES_ADDR_RESULT0, b1, 8)) != HAL_OK ||
      (err = hal_io_read(core, AES_ADDR_RESULT2, b2, 8)) != HAL_OK)
    return err;

  return HAL_OK;
}


/*
 * Wrap plaintext Q using KEK K, placing result in C.
 *
 * Q and C can overlap.  For encrypt-in-place, use Q = C + 8 (that is,
 * leave 8 empty bytes before the plaintext).
 *
 * Use hal_aes_keywrap_ciphertext_length() to calculate the correct
 * buffer size.
 */

hal_error_t hal_aes_keywrap(hal_core_t *core,
                            const uint8_t *K, const size_t K_len,
                            const uint8_t * const Q,
                            const size_t m,
                            uint8_t *C,
                            size_t *C_len)
{
  const size_t calculated_C_len = hal_aes_keywrap_ciphertext_length(m);
  const int free_core = (core == NULL);
  hal_error_t err;
  size_t n;

  hal_assert(calculated_C_len % 8 == 0);

  if (Q == NULL || C == NULL || C_len == NULL || *C_len < calculated_C_len)
    return HAL_ERROR_BAD_ARGUMENTS;

  /* If we're passed a core, we should figure out which one it is.
   * In practice, core is always NULL, so this is UNTESTED CODE.
   */
  if (core) {
    const hal_core_info_t *info = hal_core_info(core);
    if (memcmp(info->name, KEYWRAP_NAME, 8) == 0)
      use_keywrap_core = 1;
    else if (memcmp(info->name, AES_CORE_NAME, 8) != 0)
      /* I have no idea what this is */
      return HAL_ERROR_BAD_ARGUMENTS;
  }
  else {
    const char *core_name = (use_keywrap_core ? KEYWRAP_NAME : AES_CORE_NAME);
    if ((err = hal_core_alloc(core_name, &core, NULL)) != HAL_OK)
      return err;
  }

  if ((err = load_kek(core, K, K_len, key_encrypting)) != HAL_OK)
      goto out;

  *C_len = calculated_C_len;

  if (C + 8 != Q)
    memmove(C + 8, Q, m);
  if (m % 8 != 0)
    memset(C + 8 + m, 0, 8 -  (m % 8));

  if (use_keywrap_core) {
    if ((err = do_keywrap_core(core, C, m, key_encrypting)) != HAL_OK)
      goto out;
  }
  else {
    C[0] = 0xA6;
    C[1] = 0x59;
    C[2] = 0x59;
    C[3] = 0xA6;
    C[4] = (m >> 24) & 0xFF;
    C[5] = (m >> 16) & 0xFF;
    C[6] = (m >>  8) & 0xFF;
    C[7] = (m >>  0) & 0xFF;

    n = calculated_C_len/8 - 1;
    if (n == 1) {
      if ((err = do_block(core, C, C + 8)) != HAL_OK)
        goto out;
    }

    else {
      for (size_t j = 0; j <= 5; j++) {
        for (size_t i = 1; i <= n; i++) {
          uint32_t t = n * j + i;
          if ((err = do_block(core, C, C + i * 8)) != HAL_OK)
            goto out;
          C[7] ^= t & 0xFF; t >>= 8;
          C[6] ^= t & 0xFF; t >>= 8;
          C[5] ^= t & 0xFF; t >>= 8;
          C[4] ^= t & 0xFF;
        }
      }
    }
  }

out:
  if (free_core)
    hal_core_free(core);
  return err;
}


/*
 * Unwrap ciphertext C using KEK K, placing result in Q.
 *
 * Q should be the same size as C.  Q and C can overlap.
 */

hal_error_t hal_aes_keyunwrap(hal_core_t *core,
                              const uint8_t *K, const size_t K_len,
                              const uint8_t * const C,
                              const size_t C_len,
                              uint8_t *Q,
                              size_t *Q_len)
{
  const int free_core = core == NULL;
  hal_error_t err;
  size_t n;
  size_t m;

  if (C == NULL || Q == NULL || C_len % 8 != 0 || C_len < 16 || Q_len == NULL || *Q_len < C_len)
    return HAL_ERROR_BAD_ARGUMENTS;

  /* If we're passed a core, we should figure out which one it is.
   * In practice, core is always NULL, so this is UNTESTED CODE.
   */
  if (core) {
    const hal_core_info_t *info = hal_core_info(core);
    if (memcmp(info->name, KEYWRAP_NAME, 8) == 0)
      use_keywrap_core = 1;
    else if (memcmp(info->name, AES_CORE_NAME, 8) != 0)
      /* I have no idea what this is */
      return HAL_ERROR_BAD_ARGUMENTS;
  }
  else {
    const char *core_name = (use_keywrap_core ? KEYWRAP_NAME : AES_CORE_NAME);
    if ((err = hal_core_alloc(core_name, &core, NULL)) != HAL_OK)
      return err;
  }

  if ((err = load_kek(core, K, K_len, key_decrypting)) != HAL_OK)
    goto out;

  n = (C_len / 8) - 1;

  if (Q != C)
    memmove(Q, C, C_len);

  if (use_keywrap_core) {
    if ((err = do_keywrap_core(core, Q, C_len - 8, key_decrypting)) != HAL_OK)
      goto out;
  }
  else {
    if (n == 1) {
      if ((err = do_block(core, Q, Q + 8)) != HAL_OK)
        goto out;
    }

    else {
      for (long j = 5; j >= 0; j--) {
        for (size_t i = n; i >= 1; i--) {
          uint32_t t = n * j + i;
          Q[7] ^= t & 0xFF; t >>= 8;
          Q[6] ^= t & 0xFF; t >>= 8;
          Q[5] ^= t & 0xFF; t >>= 8;
          Q[4] ^= t & 0xFF;
          if ((err = do_block(core, Q, Q + i * 8)) != HAL_OK)
            goto out;
        }
      }
    }
  }

  if (Q[0] != 0xA6 || Q[1] != 0x59 || Q[2] != 0x59 || Q[3] != 0xA6) {
    err = HAL_ERROR_KEYWRAP_BAD_MAGIC;
    goto out;
  }

  m = (((((Q[4] << 8) + Q[5]) << 8) + Q[6]) << 8) + Q[7];

  if (m <= 8 * (n - 1) || m > 8 * n) {
    err = HAL_ERROR_KEYWRAP_BAD_LENGTH;
    goto out;
  }

  if (m % 8 != 0)
    for (size_t i = m + 8; i < 8 * (n + 1); i++)
      if (Q[i] != 0x00) {
        err = HAL_ERROR_KEYWRAP_BAD_PADDING;
        goto out;
      }

  *Q_len = m;

  memmove(Q, Q + 8, m);

out:
  if (free_core)
    hal_core_free(core);
  return err;
}

/*
 * "Any programmer who fails to comply with the standard naming, formatting,
 *  or commenting conventions should be shot.  If it so happens that it is
 *  inconvenient to shoot him, then he is to be politely requested to recode
 *  his program in adherence to the above standard."
 *                      -- Michael Spier, Digital Equipment Corporation
 *
 * Local variables:
 * indent-tabs-mode: nil
 * End:
 */