On Fri, Jul 10, 2015 at 07:57:11PM +0200, Stephan Mueller wrote:

Am Freitag, 10. Juli 2015, 21:39:22 schrieb Herbert Xu:

Hi Herbert,

quoted

On Thu, Jul 09, 2015 at 01:38:07PM +0200, Stephan Mueller wrote:

quoted

Actually, I found a problem that I have overlooked initally:
rfc4106-gcm-aesni causes a problem. For encryption/decryption with the
same tests for other rfc4106 implementations, I get an EINVAL.

Did you update your test vectors? The AD must now include the IV.
This is the reason I disabled the AEAD interface in 4.2 by requiring
any exported AEAD algorithm to carry the CRYPTO_ALG_AEAD_NEW flag.

I actually did not. But it works with the gcm equivalents as well as the C 
version of RFC4106. Only rfc4106-gcm-aesni is affected.

Weird.  The C version does the very same check:

static int crypto_rfc4106_decrypt(struct aead_request *req)
{
        if (req->assoclen != 16 && req->assoclen != 20)
                return -EINVAL;

Cheers,
-- 
Email: Herbert Xu [off-list ref]
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt