The issue, as I understand it, is that you need to know whether an identifier in certain contexts is a type name or not. C99 increases the number of contexts. Consider:

    x * y;

If x is a type, this declares y as a pointer to elements of that type, if you're at the beginning of a block or in C99. If x is a variable, this is a multiplication in void context, unless you're at the top level of the translation unit. A somewhat more interesting example is

    x *f();

which may declare f or call it, depending on what x is. Of course, lacking the ability to overload the multiplication operator, there's no reason you'd ever do a multiplication in void context, but it is legal.

If your lexer can distinguish type names from other identifiers, the problem is solved, and you can use BNF from there.