When decrypting with CBC mode, the in and out buffers should be the same size. Even though the IV and ciphertext are contiguous, the in buffer points to the first byte of the ciphertext. The sanity check has been updated accordingly.

Because of this error, the CBC mode tests are reporting the wrong size for the decrypted buffer (80 - 16 - 16 == 48 != 64). This has been corrected.

Also, since the loop in the decryption is writing to the out buffer, the loop conditional has been changed to `n < outlen`.  This should avoid any future errors if `inlen` changes so that it’s no longer equal to `outlen`.

Fix incorrect buffer size

In one of the ECC auxiliary test methods, a buffer 'd' was being
erased with memset using a wrong buffer size (NUM_ECC_WORDS instead
of 4*NUM_ECC_WORDS). This bug was not affecting the correctness of
ECC nor any other crypto primtive. In fact, it was not affecting even
the ECC test itself since the content of such buffer was being
overwritten rigth after the erasure procedure.

This fix removes a warning triggered when compiling the code.

Remove the name of Constanza Heath from AUTHORS as she is not
maintaining the project anymore.

Fixes Issue #30.

portabillity: Make unnecessary VLA into statically sized array

Use tests based on NIST SP 800-90A HMAC_DRBG testvectors.

Signed-off-by: Thomas Ebert Hansen <[email protected]>
Signed-off-by: Rafael Misoczki <[email protected]>

In ecc_dh.c it is widely used to zero out the memory used for holding
sensitive information, such as private keys. This memory is cleared by
a memset, but as this is done right at function return, this memset
will most likely be optimized out by the compiler, as the memory will
no longer be used, hence no need to set it. To prevent the memset from
being optimized out, it is necessary to place a memory barrier at the
memset, but such a barrier is compiler depended.

To solve the above, a _set_secure function has been introduced. This
defaults to an inlined function calling memset and, in cases where the
compiler defines __GNUC__, adds a memory barrier which ensures that
the memset is not optimized out. For compilers, not defining the
__GNUC__, it may be necessary to define TINYCRYPT_ARCH_HAS_SET_SECURE
which then declares the _set_secure function extern, thus meaning that
it will require target implementation, which in turn allows for
architecture/compiler dependent implementation.

This also fixes the memset's in uECC_make_key and uECC_make_key_with_d
which actually was missing the memory barrier.

Fixes #32

Signed-off-by: Danny Oerndrup <[email protected]>

Abstraction for architecture dependent secure memset.

Patch for #39 Sidechannel resistence of uECC_sign disabled

archiving repo