Re: [PATCH rdma-rc] IB/mlx5: Fix use-after-free error while accessing ev_file pointer

[Leon Romanovsky <leon@kernel.org>]

On Mon, Aug 12, 2019 at 10:59:07AM -0400, Doug Ledford wrote:
> On Thu, 2019-08-08 at 15:34 +0000, Jason Gunthorpe wrote:
> > On Thu, Aug 08, 2019 at 06:32:00PM +0300, Yishai Hadas wrote:
> > > > > diff --git a/drivers/infiniband/hw/mlx5/devx.c
> > > > > b/drivers/infiniband/hw/mlx5/devx.c
> > > > > index 2d1b3d9609d9..af5bbb35c058 100644
> > > > > +++ b/drivers/infiniband/hw/mlx5/devx.c
> > > > > @@ -2644,12 +2644,13 @@ static int devx_async_event_close(struct
> > > > > inode *inode, struct file *filp)
> > > > >   	struct devx_async_event_file *ev_file = filp-
> > > > > >private_data;
> > > >
> > > > This line is wrong, it should be
> > > >
> > > >    	struct devx_async_event_file *ev_file =
> > > > container_of(struct
> > > >   	                   devx_async_event_file, filp-
> > > > >private_data, uobj);
> > >
> > > You suggested the below 2 lines instead of the above one line,
> > > correct ? as
> > > struct devx_async_event_file wraps uobj as its first field this is
> > > logically
> > > equal, agree ?
> >
> > Yes, it is wrong only in the use of the type system, the private_data
> > void * should only ever be cast to a ib_uobject.
> >
> > > struct ib_uobject *uobj = filp->private_data;
> >
> > This could be done with a cast
> >
> > > > It is also a bit redundant to store the mlx5_ib_dev in the
> > > > devx_async_event_file as uobj->ucontext->dev is the same pointer.
> > > >
> > > Post hot unplug uobj->ucontext might not be accessible, isn't it ?
> > > Current code should be fine for that.
> >
> > That is the other kind of weird thing, all this destruction could be
> > done on unplug..
> >
> > > Are we fine to take this patch ?
> >
> > Yes
>
> Thanks, applied to for-rc.

Doug,

I don't see it in WIP branch, did you push it?

Thanks

>
> --
> Doug Ledford <dledford@redhat.com>
>     GPG KeyID: B826A3330E572FDD
>     Fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD