Am Mittwoch, 22. April 2015, 13:48:46 schrieb Herbert Xu:
Hi Herbert,
> On Wed, Apr 22, 2015 at 06:36:59AM +0200, Stephan Mueller wrote:
> > +static int crypto_kw_decrypt(struct aead_request *req)
> > +{
> > + struct crypto_aead *aead = crypto_aead_reqtfm(req);
> > + struct crypto_kw_ctx *ctx = crypto_aead_ctx(aead);
> > + struct crypto_cipher *tfm = ctx->child;
> > + unsigned long alignmask = crypto_cipher_alignmask(tfm);
> > + unsigned int src_nbytes, dst_nbytes, i;
> > + struct scatter_walk src_walk, dst_walk;
> > + struct crypto_kw_block block;
>
> Why isn't this aligned like tbe_buffer?
>
> > + u8 tmpblock[SEMIBSIZE];
> > + u64 t = 6 * ((req->cryptlen - SEMIBSIZE) >> 3);
> > + int ret = -EAGAIN;
> > + struct scatterlist src, dst;
> > + /* IV of KW defined by section 6.2 */
> > + u8 *default_iv = "\xA6\xA6\xA6\xA6\xA6\xA6\xA6\xA6";
> > + unsigned int outcryptlen = req->cryptlen - SEMIBSIZE;
> > +
> > + /*
> > + * Require at least 3 semiblocks as defined in SP800-38F and ensure
> > + * that the given data is aligned to semiblock.
> > + */
> > + if (req->cryptlen < (3 * SEMIBSIZE) || req->cryptlen % 8)
> > + return -EINVAL;
> > +
> > + /*
> > + * src scatterlist is read only. dst scatterlist is r/w. During the
> > + * first loop, src points to req->src and dst to req->dst. For any
> > + * subsequent round, the code operates on req->dst only.
> > + */
> > + crypto_kw_copy_scatterlist(req->src, &src);
> > + crypto_kw_copy_scatterlist(req->dst, &dst);
> > +
> > + for (i = 0; i < 6; i++) {
> > + u8 tbe_buffer[SEMIBSIZE + alignmask];
> > + /* alignment for the crypto_xor operation */
>
> You're setting alignmask to that of the child transform, which
> may have no requirements on alignment at all. So you need to
> ensure that it's at least 4-byte aligned for crypto_xor.
Will do in next installment.
>
> > + inst->alg.cra_alignmask = alg->cra_alignmask | (__alignof__(u64) - 1);
>
> Where does this 8-byte alignment requirement come from?
Well, I am accessing the data in 8-byte chunks. Moreover, in the scatterwalk
copy functions, I search through the scatterlists in 8 byte increments. If,
say, a scatterwalk is not a multiple of 8 bytes, the scatterwalk logic will
not process the last chunk of memory.
>
> You also never actually pass any input data directly to the child,
> except for the key so you don't need to specify the child's alignment
> here at all.
Will change that.
>
> Cheers,
--
Ciao
Stephan
--
To unsubscribe from this list: send the line "unsubscribe linux-crypto" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
