Re: [PATCH] net: sched: Fix memory exposure from short TCA_U32_SEL

[Joe Perches <joe@perches.com>]

On Sun, 2018-08-26 at 18:32 +0100, Al Viro wrote:
> On Sat, Aug 25, 2018 at 11:19:30PM -0700, Kees Cook wrote:
> > > > -     n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
> > > > +     n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
> > > 
> > > ITYM
> > >         n = kzalloc(offsetof(struct tc_u_common, sel.keys[s->nkeys]), GFP_KERNEL);
> > 
> > I prefer to reuse sel_size and keep typeof() to keep things tied to
> > "n" more directly. *shrug*
> 
> This is rather search-hostile, though.  Fresh example from the same
> area: where are struct tcf_proto instances created?  Is it true that
> each is followed by ->ops->init()?  Is it true that ->ops->init()
> is never called twice for the same instance?  Is it true that
> ->ops->destroy() is called exactly once between successful ->ops->init()
> and freeing the object?
> 
> That's precisely the kind of questions you end up asking when learning
> a new area.  Your variant makes those harder to answer; it does make
> it easier to catch local problems on casual grep, but it's hell both
> on the newbies trying to make sense of an area and on the old hands
> from different areas.
> 
> That, BTW, is why I hate the use of sizeof(*p) in kmalloc, etc.
> arguments.  typeof is even worse in that respect.

True.  Semantic searches via tools like coccinelle could help here
but those searches are quite a bit slower than straightforward greps.