Re: [PATCH v2 56/63] RDMA/mlx5: Use struct_group() to zero struct mlx5_ib_mr

[Kees Cook <keescook@chromium.org>]

On Thu, Aug 19, 2021 at 01:47:57PM -0300, Jason Gunthorpe wrote:
> On Thu, Aug 19, 2021 at 09:19:08AM -0700, Kees Cook wrote:
> > On Thu, Aug 19, 2021 at 09:27:16AM -0300, Jason Gunthorpe wrote:
> > > On Tue, Aug 17, 2021 at 11:05:26PM -0700, Kees Cook wrote:
> > > > In preparation for FORTIFY_SOURCE performing compile-time and run-time
> > > > field bounds checking for memset(), avoid intentionally writing across
> > > > neighboring fields.
> > > > 
> > > > Add struct_group() to mark region of struct mlx5_ib_mr that should be
> > > > initialized to zero.
> > > > 
> > > > Cc: Leon Romanovsky <leon@kernel.org>
> > > > Cc: Doug Ledford <dledford@redhat.com>
> > > > Cc: Jason Gunthorpe <jgg@ziepe.ca>
> > > > Cc: linux-rdma@vger.kernel.org
> > > > Signed-off-by: Kees Cook <keescook@chromium.org>
> > > >  drivers/infiniband/hw/mlx5/mlx5_ib.h | 4 +++-
> > > >  1 file changed, 3 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
> > > > index bf20a388eabe..f63bf204a7a1 100644
> > > > +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
> > > > @@ -644,6 +644,7 @@ struct mlx5_ib_mr {
> > > >  	struct ib_umem *umem;
> > > >  
> > > >  	/* This is zero'd when the MR is allocated */
> > > > +	struct_group(cleared,
> > > >  	union {
> > > >  		/* Used only while the MR is in the cache */
> > > >  		struct {
> > > > @@ -691,12 +692,13 @@ struct mlx5_ib_mr {
> > > >  			bool is_odp_implicit;
> > > >  		};
> > > >  	};
> > > > +	);
> > > >  };
> > > >  
> > > >  /* Zero the fields in the mr that are variant depending on usage */
> > > >  static inline void mlx5_clear_mr(struct mlx5_ib_mr *mr)
> > > >  {
> > > > -	memset(mr->out, 0, sizeof(*mr) - offsetof(struct mlx5_ib_mr, out));
> > > > +	memset(&mr->cleared, 0, sizeof(mr->cleared));
> > > >  }
> > > 
> > > Why not use the memset_after(mr->umem) here?
> > 
> > I can certainly do that instead. In this series I've tended to opt
> > for groupings so the position of future struct member additions are
> > explicitly chosen. (i.e. reducing the chance that a zeroing of the new
> > member be a surprise.)
> 
> I saw the earlier RDMA patches where using other memset techniques
> though? Were there flex arrays or something that made groups infeasible?

Which do you mean? When doing the conversions I tended to opt for
struct_group() since it provides more robust "intentionality". Strictly
speaking, the new memset helpers are doing field-spanning writes, but the
"clear to the end" pattern was so common it made sense to add the helpers,
as they're a bit less disruptive. It's totally up to you! :)

-- 
Kees Cook