crypto: ahash - remove support for nonzero alignmask

Currently, the ahash API checks the alignment of all key and result
buffers against the algorithm's declared alignmask, and for any
unaligned buffers it falls back to manually aligned temporary buffers.

This is virtually useless, however. First, since it does not apply to
the message, its effect is much more limited than e.g. is the case for
the alignmask for "skcipher". Second, the key and result buffers are
given as virtual addresses and cannot (in general) be DMA'ed into, so
drivers end up having to copy to/from them in software anyway. As a
result it's easy to use memcpy() or the unaligned access helpers.

The crypto_hash_walk_*() helper functions do use the alignmask to align
the message. But with one exception those are only used for shash
algorithms being exposed via the ahash API, not for native ahashes, and
aligning the message is not required in this case, especially now that
alignmask support has been removed from shash. The exception is the
n2_core driver, which doesn't set an alignmask.

In any case, no ahash algorithms actually set a nonzero alignmask
anymore. Therefore, remove support for it from ahash. The benefit is
that all the code to handle "misaligned" buffers in the ahash API goes
away, reducing the overhead of the ahash API.

This follows the same change that was made to shash.

Signed-off-by: Eric Biggers <[e-mail]>
Signed-off-by: Herbert Xu <[e-mail]>

Documentation/crypto/devel-algos.rst

···
 
 Some of the drivers will want to use the Generic ScatterWalk in case the
 implementation needs to be fed separate chunks of the scatterlist which
-contains the input data. The buffer containing the resulting hash will
-always be properly aligned to .cra_alignmask so there is no need to
-worry about this.
+contains the input data.

crypto/ahash.c

···
 
 static int hash_walk_next(struct crypto_hash_walk *walk)
 {
-	unsigned int alignmask = walk->alignmask;
 	unsigned int offset = walk->offset;
 	unsigned int nbytes = min(walk->entrylen,
 				  ((unsigned int)(PAGE_SIZE)) - offset);
 
 	walk->data = kmap_local_page(walk->pg);
 	walk->data += offset;
-
-	if (offset & alignmask) {
-		unsigned int unaligned = alignmask + 1 - (offset & alignmask);
-
-		if (nbytes > unaligned)
-			nbytes = unaligned;
-	}
-
 	walk->entrylen -= nbytes;
 	return nbytes;
 }
···
 
 int crypto_hash_walk_done(struct crypto_hash_walk *walk, int err)
 {
-	unsigned int alignmask = walk->alignmask;
-
 	walk->data -= walk->offset;
-
-	if (walk->entrylen && (walk->offset & alignmask) && !err) {
-		unsigned int nbytes;
-
-		walk->offset = ALIGN(walk->offset, alignmask + 1);
-		nbytes = min(walk->entrylen,
-			     (unsigned int)(PAGE_SIZE - walk->offset));
-		if (nbytes) {
-			walk->entrylen -= nbytes;
-			walk->data += walk->offset;
-			return nbytes;
-		}
-	}
 
 	kunmap_local(walk->data);
 	crypto_yield(walk->flags);
···
 		return 0;
 	}
 
-	walk->alignmask = crypto_ahash_alignmask(crypto_ahash_reqtfm(req));
 	walk->sg = req->src;
 	walk->flags = req->base.flags;
 
 	return hash_walk_new_entry(walk);
 }
 EXPORT_SYMBOL_GPL(crypto_hash_walk_first);
-
-static int ahash_setkey_unaligned(struct crypto_ahash *tfm, const u8 *key,
-				unsigned int keylen)
-{
-	unsigned long alignmask = crypto_ahash_alignmask(tfm);
-	int ret;
-	u8 *buffer, *alignbuffer;
-	unsigned long absize;
-
-	absize = keylen + alignmask;
-	buffer = kmalloc(absize, GFP_KERNEL);
-	if (!buffer)
-		return -ENOMEM;
-
-	alignbuffer = (u8 *)ALIGN((unsigned long)buffer, alignmask + 1);
-	memcpy(alignbuffer, key, keylen);
-	ret = tfm->setkey(tfm, alignbuffer, keylen);
-	kfree_sensitive(buffer);
-	return ret;
-}
 
 static int ahash_nosetkey(struct crypto_ahash *tfm, const u8 *key,
 			  unsigned int keylen)
···
 int crypto_ahash_setkey(struct crypto_ahash *tfm, const u8 *key,
 			unsigned int keylen)
 {
-	unsigned long alignmask = crypto_ahash_alignmask(tfm);
-	int err;
-
-	if ((unsigned long)key & alignmask)
-		err = ahash_setkey_unaligned(tfm, key, keylen);
-	else
-		err = tfm->setkey(tfm, key, keylen);
+	int err = tfm->setkey(tfm, key, keylen);
 
 	if (unlikely(err)) {
 		ahash_set_needkey(tfm);
···
 			  bool has_state)
 {
 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
-	unsigned long alignmask = crypto_ahash_alignmask(tfm);
 	unsigned int ds = crypto_ahash_digestsize(tfm);
 	struct ahash_request *subreq;
 	unsigned int subreq_size;
···
 	reqsize = ALIGN(reqsize, crypto_tfm_ctx_alignment());
 	subreq_size += reqsize;
 	subreq_size += ds;
-	subreq_size += alignmask & ~(crypto_tfm_ctx_alignment() - 1);
 
 	flags = ahash_request_flags(req);
 	gfp = (flags & CRYPTO_TFM_REQ_MAY_SLEEP) ?  GFP_KERNEL : GFP_ATOMIC;
···
 	ahash_request_set_callback(subreq, flags, cplt, req);
 
 	result = (u8 *)(subreq + 1) + reqsize;
-	result = PTR_ALIGN(result, alignmask + 1);
 
 	ahash_request_set_crypt(subreq, req->src, result, req->nbytes);
 
···
 	kfree_sensitive(subreq);
 }
 
-static void ahash_op_unaligned_done(void *data, int err)
-{
-	struct ahash_request *areq = data;
-
-	if (err == -EINPROGRESS)
-		goto out;
-
-	/* First copy req->result into req->priv.result */
-	ahash_restore_req(areq, err);
-
-out:
-	/* Complete the ORIGINAL request. */
-	ahash_request_complete(areq, err);
-}
-
-static int ahash_op_unaligned(struct ahash_request *req,
-			      int (*op)(struct ahash_request *),
-			      bool has_state)
-{
-	int err;
-
-	err = ahash_save_req(req, ahash_op_unaligned_done, has_state);
-	if (err)
-		return err;
-
-	err = op(req->priv);
-	if (err == -EINPROGRESS || err == -EBUSY)
-		return err;
-
-	ahash_restore_req(req, err);
-
-	return err;
-}
-
-static int crypto_ahash_op(struct ahash_request *req,
-			   int (*op)(struct ahash_request *),
-			   bool has_state)
-{
-	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
-	unsigned long alignmask = crypto_ahash_alignmask(tfm);
-	int err;
-
-	if ((unsigned long)req->result & alignmask)
-		err = ahash_op_unaligned(req, op, has_state);
-	else
-		err = op(req);
-
-	return crypto_hash_errstat(crypto_hash_alg_common(tfm), err);
-}
-
 int crypto_ahash_final(struct ahash_request *req)
 {
 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
···
 	if (IS_ENABLED(CONFIG_CRYPTO_STATS))
 		atomic64_inc(&hash_get_stat(alg)->hash_cnt);
 
-	return crypto_ahash_op(req, tfm->final, true);
+	return crypto_hash_errstat(alg, tfm->final(req));
 }
 EXPORT_SYMBOL_GPL(crypto_ahash_final);
 
···
 		atomic64_add(req->nbytes, &istat->hash_tlen);
 	}
 
-	return crypto_ahash_op(req, tfm->finup, true);
+	return crypto_hash_errstat(alg, tfm->finup(req));
 }
 EXPORT_SYMBOL_GPL(crypto_ahash_finup);
 
···
 {
 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
 	struct hash_alg_common *alg = crypto_hash_alg_common(tfm);
+	int err;
 
 	if (IS_ENABLED(CONFIG_CRYPTO_STATS)) {
 		struct crypto_istat_hash *istat = hash_get_stat(alg);
···
 	}
 
 	if (crypto_ahash_get_flags(tfm) & CRYPTO_TFM_NEED_KEY)
-		return crypto_hash_errstat(alg, -ENOKEY);
+		err = -ENOKEY;
+	else
+		err = tfm->digest(req);
 
-	return crypto_ahash_op(req, tfm->digest, false);
+	return crypto_hash_errstat(alg, err);
 }
 EXPORT_SYMBOL_GPL(crypto_ahash_digest);
 

crypto/shash.c

···
 	if (alg->digestsize > HASH_MAX_DIGESTSIZE)
 		return -EINVAL;
 
+	/* alignmask is not useful for hashes, so it is not supported. */
+	if (base->cra_alignmask)
+		return -EINVAL;
+
 	base->cra_flags &= ~CRYPTO_ALG_TYPE_MASK;
 
 	if (IS_ENABLED(CONFIG_CRYPTO_STATS))
···
 	int err;
 
 	if (alg->descsize > HASH_MAX_DESCSIZE)
-		return -EINVAL;
-
-	/* alignmask is not useful for shash, so it is not supported. */
-	if (base->cra_alignmask)
 		return -EINVAL;
 
 	if ((alg->export && !alg->import) || (alg->import && !alg->export))

include/crypto/internal/hash.h

···
 	char *data;
 
 	unsigned int offset;
-	unsigned int alignmask;
+	unsigned int flags;
 
 	struct page *pg;
 	unsigned int entrylen;
 
 	unsigned int total;
 	struct scatterlist *sg;
-
-	unsigned int flags;
 };
 
 struct ahash_instance {

include/linux/crypto.h

···
  *	  crypto_aead_walksize() (with the remainder going at the end), no chunk
  *	  can cross a page boundary or a scatterlist element boundary.
  *    ahash:
- *	- The result buffer must be aligned to the algorithm's alignmask.
  *	- crypto_ahash_finup() must not be used unless the algorithm implements
  *	  ->finup() natively.
  */
···
  * @cra_ctxsize: Size of the operational context of the transformation. This
  *		 value informs the kernel crypto API about the memory size
  *		 needed to be allocated for the transformation context.
- * @cra_alignmask: Alignment mask for the input and output data buffer. The data
- *		   buffer containing the input data for the algorithm must be
- *		   aligned to this alignment mask. The data buffer for the
- *		   output data must be aligned to this alignment mask. Note that
- *		   the Crypto API will do the re-alignment in software, but
- *		   only under special conditions and there is a performance hit.
- *		   The re-alignment happens at these occasions for different
- *		   @cra_u types: cipher -- For both input data and output data
- *		   buffer; ahash -- For output hash destination buf; shash --
- *		   For output hash destination buf.
- *		   This is needed on hardware which is flawed by design and
- *		   cannot pick data from arbitrary addresses.
+ * @cra_alignmask: For cipher, skcipher, lskcipher, and aead algorithms this is
+ *		   1 less than the alignment, in bytes, that the algorithm
+ *		   implementation requires for input and output buffers.  When
+ *		   the crypto API is invoked with buffers that are not aligned
+ *		   to this alignment, the crypto API automatically utilizes
+ *		   appropriately aligned temporary buffers to comply with what
+ *		   the algorithm needs.  (For scatterlists this happens only if
+ *		   the algorithm uses the skcipher_walk helper functions.)  This
+ *		   misalignment handling carries a performance penalty, so it is
+ *		   preferred that algorithms do not set a nonzero alignmask.
+ *		   Also, crypto API users may wish to allocate buffers aligned
+ *		   to the alignmask of the algorithm being used, in order to
+ *		   avoid the API having to realign them.  Note: the alignmask is
+ *		   not supported for hash algorithms and is always 0 for them.
  * @cra_priority: Priority of this transformation implementation. In case
  *		  multiple transformations with same @cra_name are available to
  *		  the Crypto API, the kernel will use the one with highest