Re: [PATCH v5 8/8] crypto, x86/sha: Eliminate casts on asm implementations

[Kees Cook]

On Wed, Nov 13, 2019 at 11:55:30AM -0800, Eric Biggers wrote:
> On Wed, Nov 13, 2019 at 10:25:16AM -0800, Kees Cook wrote:
> > In order to avoid CFI function prototype mismatches, this removes the
> > casts on assembly implementations of sha1/256/512 accelerators. The
> > safety checks from BUILD_BUG_ON() remain.
> > 
> > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > ---
> >  arch/x86/crypto/sha1_ssse3_glue.c   | 61 ++++++++++++-----------------
> >  arch/x86/crypto/sha256_ssse3_glue.c | 31 +++++++--------
> >  arch/x86/crypto/sha512_ssse3_glue.c | 28 ++++++-------
> >  3 files changed, 50 insertions(+), 70 deletions(-)
> > 
> > diff --git a/arch/x86/crypto/sha1_ssse3_glue.c b/arch/x86/crypto/sha1_ssse3_glue.c
> > index 639d4c2fd6a8..a151d899f37a 100644
> > --- a/arch/x86/crypto/sha1_ssse3_glue.c
> > +++ b/arch/x86/crypto/sha1_ssse3_glue.c
> > @@ -27,11 +27,8 @@
> >  #include <crypto/sha1_base.h>
> >  #include <asm/simd.h>
> >  
> > -typedef void (sha1_transform_fn)(u32 *digest, const char *data,
> > -				unsigned int rounds);
> > -
> >  static int sha1_update(struct shash_desc *desc, const u8 *data,
> > -			     unsigned int len, sha1_transform_fn *sha1_xform)
> > +			     unsigned int len, sha1_block_fn *sha1_xform)
> >  {
> >  	struct sha1_state *sctx = shash_desc_ctx(desc);
> >  
> > @@ -39,48 +36,44 @@ static int sha1_update(struct shash_desc *desc, const u8 *data,
> >  	    (sctx->count % SHA1_BLOCK_SIZE) + len < SHA1_BLOCK_SIZE)
> >  		return crypto_sha1_update(desc, data, len);
> >  
> > -	/* make sure casting to sha1_block_fn() is safe */
> > +	/* make sure sha1_block_fn() use in generic routines is safe */
> >  	BUILD_BUG_ON(offsetof(struct sha1_state, state) != 0);
> 
> This update to the comment makes no sense, since sha1_block_fn() is obviously
> safe in the helpers, and this says nothing about the assembly functions.
> Instead this should say something like:
> 
> 	/*
> 	 * Make sure that struct sha1_state begins directly with the 160-bit
> 	 * SHA1 internal state, as this is what the assembly functions expect.
> 	 */
> 
> Likewise for SHA-256 and SHA-512, except for those it would be a 256-bit and
> 512-bit internal state respectively.

Thanks! Agreed, that is much clearer.

> > -asmlinkage void sha1_transform_ssse3(u32 *digest, const char *data,
> > -				     unsigned int rounds);
> > +asmlinkage void sha1_transform_ssse3(struct sha1_state *digest,
> > +				     u8 const *data, int rounds);
> 
> 'u8 const' is unconventional.  Please use 'const u8' instead.

Yeah, I noticed that but decided to disrupt less. Fixed now.

> Also, this function prototype is also given in a comment in the corresponding
> assembly file.  Can you please update that too, and also leave a comment in the
> assembly file like "struct sha1_state is assumed to begin with u32 state[5]."?
> 
> Likewise for all the other SHA-1, and SHA-256, and SHA-512 assembly functions,
> except it would be u32 state[8] for SHA-256 and u64 state[8] for SHA-512.

I love how each uses an entirely different comment style. :) Updated!

-- 
Kees Cook