Re: [Cocci] [PATCH] coccinelle: misc: add array_size_dup script to detect missed overlow checks

From: Kees Cook
Date: Thu Jun 18 2020 - 15:52:56 EST


On Wed, Jun 17, 2020 at 08:54:03PM +0200, Julia Lawall wrote:
>
>
> On Wed, 17 Jun 2020, Kees Cook wrote:
>
> > On Mon, Jun 15, 2020 at 01:20:45PM +0300, Denis Efremov wrote:
> > > +@as@
> > > +expression E1, E2;
> > > +@@
> > > +
> > > +array_size(E1, E2)
> >
> > BTW, is there a way yet in Coccinelle to match a fully qualified (?)
> > identifier? For example, if I have two lines in C:
> >
> > A)
> > array_size(variable, 5);
> > B)
> > array_size(instance->member.size, 5);
> > C)
> > array_size(instance->member.size + 1, 5);
> > D)
> > array_size(function_call(variable), 5);
> >
> >
> > This matches A, B, C, and D:
> >
> > @@
> > expression ARG1;
> > expression ARG2;
> > @@
> >
> > array_size(ARG1, ARG2);
> >
> >
> > This matches only A:
> >
> > @@
> > identifier ARG1;
> > expression ARG2;
> > @@
> >
> > array_size(ARG1, ARG2);
> >
> >
> > How do I get something to match A and B but not C and D (i.e. I do not
> > want to match any operations, function calls, etc, only a variable,
> > which may be identified through dereference, array index, or struct
> > member access.)
>
> \(i\|e.fld\|e->fld\)
>
> would probably do what you want. It will also match cases where e is a
> function/macr call, but that is unlikely.
>
> If you want a single metavariable that contains the whole thing, you can
> have an expression metavariable E and then write:
>
> \(\(i\|e.fld\|e->fld\) \& E\)

Can you give an example of how that would look for an @@ section?

The problem I have is that I don't know the depth or combination of such
metavariables. There are a lot of combinations:

a
a.b
a.b.c
a.b.c.d
a.b.c->d
a.b->c
a.b->c.d
a.b->c->d
a->b
a->b.c
a->b.c.d
a->b.c->d
a->b->c
a->b->c.d
a->b->c->d
...


--
Kees Cook