Re: [PATCH v8 10/12] crypto: x86/aesni - Use the proper data type in struct aesni_xts_ctx

[Eric Biggers]

On Sun, Jun 04, 2023 at 03:02:32PM -0700, Chang S. Bae wrote:
> On 6/4/2023 8:34 AM, Eric Biggers wrote:
> > 
> > To re-iterate what I said on v6, the runtime alignment to a 16-byte boundary
> > should happen when translating the raw crypto_skcipher_ctx() into the pointer to
> > the aes_xts_ctx.  It should not happen when accessing each individual field in
> > the aes_xts_ctx.
> > 
> > Yet, this code is still doing runtime alignment when accessing each individual
> > field, as the second argument to aes_set_key_common() is 'void *raw_ctx' which
> > aes_set_key_common() runtime-aligns to crypto_aes_ctx.
> > 
> > We should keep everything consistent, which means making aes_set_key_common()
> > take a pointer to crypto_aes_ctx and not do the runtime alignment.
> 
> Let me clarify what is the problem this patch tried to solve here. The
> current struct aesni_xts_ctx is ugly. So, the main story is let's fix it
> before using the code for AES-KL.
> 
> Then, the rework part may be applicable for code re-usability. That seems to
> be okay to do here.
> 
> Fixing the runtime alignment entirely seems to be touching other code than
> AES-XTS. Yes, that's ideal cleanup for consistency. But, it seems to be less
> relevant in this series.  I'd be happy to follow up on that improvement
> though.

IMO the issue is that your patch makes the code (including the XTS code)
inconsistent because it makes it use a mix of both approaches: it aligns each
field individually, *and* it aligns the ctx up-front.  I was hoping to switch
fully from the former approach to the latter approach, instead of switching from
the former approach to a mix of the two approaches as you are proposing.

The following on top of this patch is what I am asking for.  I think it would be
appropriate to fold into this patch.

diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index 589648142c173..ad1ae7a88b59d 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -228,10 +228,10 @@ static inline struct aesni_xts_ctx *aes_xts_ctx(struct crypto_skcipher *tfm)
 	return (struct aesni_xts_ctx *)aes_align_addr(crypto_skcipher_ctx(tfm));
 }
 
-static int aes_set_key_common(struct crypto_tfm *tfm, void *raw_ctx,
+static int aes_set_key_common(struct crypto_tfm *tfm,
+			      struct crypto_aes_ctx *ctx,
 			      const u8 *in_key, unsigned int key_len)
 {
-	struct crypto_aes_ctx *ctx = aes_ctx(raw_ctx);
 	int err;
 
 	if (key_len != AES_KEYSIZE_128 && key_len != AES_KEYSIZE_192 &&
@@ -252,7 +252,8 @@ static int aes_set_key_common(struct crypto_tfm *tfm, void *raw_ctx,
 static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
 		       unsigned int key_len)
 {
-	return aes_set_key_common(tfm, crypto_tfm_ctx(tfm), in_key, key_len);
+	return aes_set_key_common(tfm, aes_ctx(crypto_tfm_ctx(tfm)),
+				  in_key, key_len);
 }
 
 static void aesni_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
@@ -285,7 +286,7 @@ static int aesni_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
 			         unsigned int len)
 {
 	return aes_set_key_common(crypto_skcipher_tfm(tfm),
-				  crypto_skcipher_ctx(tfm), key, len);
+				  aes_ctx(crypto_skcipher_ctx(tfm)), key, len);
 }
 
 static int ecb_encrypt(struct skcipher_request *req)