linux-crypto.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* potential underfow in crypto/lrw.c setkey() setkey
@ 2019-05-09 11:06 Dan Carpenter
  2019-05-10  4:03 ` Eric Biggers
  0 siblings, 1 reply; 2+ messages in thread
From: Dan Carpenter @ 2019-05-09 11:06 UTC (permalink / raw)
  Cc: linux-crypto

crypto/lrw.c
    72  static int setkey(struct crypto_skcipher *parent, const u8 *key,
    73                    unsigned int keylen)
    74  {
    75          struct priv *ctx = crypto_skcipher_ctx(parent);
    76          struct crypto_skcipher *child = ctx->child;
    77          int err, bsize = LRW_BLOCK_SIZE;
    78          const u8 *tweak = key + keylen - bsize;
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Smatch thinks that keylen is user controlled from zero to some upper
bound.  How do we know it's >= LRW_BLOCK_SIZE (16)?

I find the crypto code sort of hard to follow...  There are a bunch of
setkey pointers and they're sometimes called recursively.  Is there
some trick or hints?

    79          be128 tmp = { 0 };
    80          int i;
    81  
    82          crypto_skcipher_clear_flags(child, CRYPTO_TFM_REQ_MASK);
    83          crypto_skcipher_set_flags(child, crypto_skcipher_get_flags(parent) &
    84                                           CRYPTO_TFM_REQ_MASK);
    85          err = crypto_skcipher_setkey(child, key, keylen - bsize);
    86          crypto_skcipher_set_flags(parent, crypto_skcipher_get_flags(child) &
    87                                            CRYPTO_TFM_RES_MASK);
    88          if (err)
    89                  return err;
    90  
    91          if (ctx->table)
    92                  gf128mul_free_64k(ctx->table);
    93  
    94          /* initialize multiplication table for Key2 */
    95          ctx->table = gf128mul_init_64k_bbe((be128 *)tweak);
    96          if (!ctx->table)
    97                  return -ENOMEM;
    98  
    99          /* initialize optimization table */
   100          for (i = 0; i < 128; i++) {
   101                  setbit128_bbe(&tmp, i);
   102                  ctx->mulinc[i] = tmp;
   103                  gf128mul_64k_bbe(&ctx->mulinc[i], ctx->table);
   104          }
   105  
   106          return 0;
   107  }

regards,
dan carpenter

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: potential underfow in crypto/lrw.c setkey() setkey
  2019-05-09 11:06 potential underfow in crypto/lrw.c setkey() setkey Dan Carpenter
@ 2019-05-10  4:03 ` Eric Biggers
  0 siblings, 0 replies; 2+ messages in thread
From: Eric Biggers @ 2019-05-10  4:03 UTC (permalink / raw)
  To: Dan Carpenter; +Cc: linux-crypto

On Thu, May 09, 2019 at 02:06:22PM +0300, Dan Carpenter wrote:
> crypto/lrw.c
>     72  static int setkey(struct crypto_skcipher *parent, const u8 *key,
>     73                    unsigned int keylen)
>     74  {
>     75          struct priv *ctx = crypto_skcipher_ctx(parent);
>     76          struct crypto_skcipher *child = ctx->child;
>     77          int err, bsize = LRW_BLOCK_SIZE;
>     78          const u8 *tweak = key + keylen - bsize;
>                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Smatch thinks that keylen is user controlled from zero to some upper
> bound.  How do we know it's >= LRW_BLOCK_SIZE (16)?

See create() in crypto/lrw.c, which is the function that creates an LRW
instance.  It sets .min_keysize and .max_keysize:

        inst->alg.min_keysize = crypto_skcipher_alg_min_keysize(alg) +
                                LRW_BLOCK_SIZE;
        inst->alg.max_keysize = crypto_skcipher_alg_max_keysize(alg) +
                                LRW_BLOCK_SIZE;

Then when sometime calls crypto_skcipher_setkey(), it calls skcipher_setkey() in
crypto/skcipher.c which verifies the key length is in within the bounds the
algorithm declares:

        if (keylen < cipher->min_keysize || keylen > cipher->max_keysize) {
                crypto_skcipher_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
                return -EINVAL;
        }

> 
> I find the crypto code sort of hard to follow...  There are a bunch of
> setkey pointers and they're sometimes called recursively.  Is there
> some trick or hints?
> 
>     79          be128 tmp = { 0 };
>     80          int i;
>     81  
>     82          crypto_skcipher_clear_flags(child, CRYPTO_TFM_REQ_MASK);
>     83          crypto_skcipher_set_flags(child, crypto_skcipher_get_flags(parent) &
>     84                                           CRYPTO_TFM_REQ_MASK);
>     85          err = crypto_skcipher_setkey(child, key, keylen - bsize);
>     86          crypto_skcipher_set_flags(parent, crypto_skcipher_get_flags(child) &
>     87                                            CRYPTO_TFM_RES_MASK);
>     88          if (err)
>     89                  return err;

LRW is a template for a block cipher mode of operation which is implemented on
top of ECB.  So, LRW's setkey() method calls setkey() on the underlying ECB
transform.  Similarly, ECB then may call setkey() of the underlying block cipher
such as AES, or alternatively it may be implemented directly.

- Eric

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2019-05-10  4:03 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2019-05-09 11:06 potential underfow in crypto/lrw.c setkey() setkey Dan Carpenter
2019-05-10  4:03 ` Eric Biggers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).