On Wed, 13 Nov 2019 at 22:28, Sami Tolvanen <samitolvanen@xxxxxxxxxx> wrote:
>
> On Wed, Nov 13, 2019 at 12:04 PM Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
> >
> > On Tue, Nov 12, 2019 at 02:30:46PM -0800, Sami Tolvanen wrote:
> > > Declare assembly functions with the expected function type
> > > instead of casting pointers in C to avoid type mismatch failures
> > > with Control-Flow Integrity (CFI) checking.
> > >
> > > Signed-off-by: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
> > > ---
> > > arch/arm64/crypto/sha1-ce-glue.c | 12 +++++-------
> > > arch/arm64/crypto/sha2-ce-glue.c | 26 +++++++++++---------------
> > > arch/arm64/crypto/sha256-glue.c | 30 ++++++++++++------------------
> > > arch/arm64/crypto/sha512-ce-glue.c | 23 ++++++++++-------------
> > > arch/arm64/crypto/sha512-glue.c | 13 +++++--------
> > > 5 files changed, 43 insertions(+), 61 deletions(-)
> > >
> > > diff --git a/arch/arm64/crypto/sha1-ce-glue.c b/arch/arm64/crypto/sha1-ce-glue.c
> > > index bdc1b6d7aff7..3153a9bbb683 100644
> > > --- a/arch/arm64/crypto/sha1-ce-glue.c
> > > +++ b/arch/arm64/crypto/sha1-ce-glue.c
> > > @@ -25,7 +25,7 @@ struct sha1_ce_state {
> > > u32 finalize;
> > > };
> > >
> > > -asmlinkage void sha1_ce_transform(struct sha1_ce_state *sst, u8 const *src,
> > > +asmlinkage void sha1_ce_transform(struct sha1_state *sst, u8 const *src,
> > > int blocks);
> >
> > Please update the comments in the corresponding assembly files too.
> >
> > Also, this change doesn't really make sense because the assembly functions still
> > expect struct sha1_ce_state, and they access sha1_ce_state::finalize which is
> > not present in struct sha1_state. There should either be wrapper functions that
> > explicitly do the cast from sha1_state to sha1_ce_state, or there should be
> > comments in the assembly files that very clearly explain that although the
> > function prototype takes sha1_state, it's really assumed to be a sha1_ce_state.
>
> Agreed, this needs a comment explaining the type mismatch. I'm also
> fine with using wrapper functions and explicitly casting the
> parameters instead of changing function declarations. Herbert, Ard,
> any preferences?
>
I guess the former would be cleaner, using container_of() rather than
a blind cast to make the code more self-documenting. The extra branch
shouldn't really matter.
