Re: [PATCH] crypto: add missing kernel_fpu_end() call

[Ard Biesheuvel]

On Fri, 6 Aug 2021 at 11:05, Ard Biesheuvel <ardb@xxxxxxxxxx> wrote:
>
> On Fri, 6 Aug 2021 at 10:23, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Jul 30, 2021 at 07:11:55PM +0530, Shreyansh Chouhan wrote:
> > > xts_crypt() code doesn't call kernel_fpu_end() after calling
> > > kernel_fpu_begin() if walk.nbytes is 0. Add a call to kernel_fpu_end()
> > > for this case.
> > >
> > > Reported-by: syzbot+20191dc583eff8602d2d@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > Signed-off-by: Shreyansh Chouhan <chouhan.shreyansh630@xxxxxxxxx>
> > > ---
> > >  arch/x86/crypto/aesni-intel_glue.c | 3 +++
> > >  1 file changed, 3 insertions(+)
> >
> > Ard?
> >
> > > diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
> > > index 2144e54a6c89..bd55a0cd7bde 100644
> > > --- a/arch/x86/crypto/aesni-intel_glue.c
> > > +++ b/arch/x86/crypto/aesni-intel_glue.c
> > > @@ -894,6 +894,9 @@ static int xts_crypt(struct skcipher_request *req, bool encrypt)
> > >                       kernel_fpu_begin();
> > >       }
> > >
> > > +     if (walk.nbytes == 0)
> > > +             kernel_fpu_end();
> > > +
>
> Don't we end up calling kernel_fpu_end() twice this way if we do enter
> the while() loop at least once?
>

How about the below instead, does that work?

--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -849,7 +849,7 @@ static int xts_crypt(struct skcipher_request *req,
bool encrypt)
                return -EINVAL;

        err = skcipher_walk_virt(&walk, req, false);
-       if (err)
+       if (err || !walk.nbytes)
                return err;

        if (unlikely(tail > 0 && walk.nbytes < walk.total)) {