Re: [PATCH] crypto: mpi: add NULL checks to mpi_normalize().

[Roman Smirnov]

On Fri, 2024-08-02 at 20:46 +0800, Herbert Xu wrote:
> On Tue, Jul 16, 2024 at 11:28:25AM +0300, Roman Smirnov wrote:
> > If a->d is NULL, the NULL pointer will be dereferenced. It
> > is necessary to prevent this case. There is at least one call
> > stack that can lead to it:
> > 
> >     mpi_ec_curve_point()
> >       ec_pow2()
> >         ec_mulm()
> >           ec_mod()
> >             mpi_mod()
> >               mpi_fdiv_r()
> >                 mpi_tdiv_r()
> >                   mpi_tdiv_qr()
> >                     mpi_resize()
> >                       kcalloc()
> > 
> > mpi_resize can return -ENOMEM, but this case is not handled in any way.
> > 
> > Next, dereferencing takes place:
> > 
> >     mpi_ec_curve_point()
> >       mpi_cmp()
> >         do_mpi_cmp()
> >           mpi_normalize()
> > 
> > Found by Linux Verification Center (linuxtesting.org) with Svace.
> > 
> > Signed-off-by: Roman Smirnov <r.smirnov@xxxxxx>
> > ---
> >  lib/crypto/mpi/mpi-bit.c | 3 +++
> >  1 file changed, 3 insertions(+)
> 
> I've just posted a patch to remove mpi_ec_curve_point and mpi_tdiv_qr.
> Are there any other code paths with the same problem?

Svace found a similar case but it is no longer relevant:

NULL constant:
    mpi_ec_mul_point()
      ec_mulm(z3, point->z, z2, ctx)
        ec_mod()
          mpi_mod()
            mpi_fdiv_r()
              mpi_tdiv_r()
                mpi_tdiv_qr()
                  mpi_resize()
                    kcalloc()

Dereference:
    mpi_ec_mul_point()
      ec_invm(z3, z3, ctx)
        mpi_invm()
          mpi_cmp_ui()
            mpi_normalize()
> 
> Thanks,