Re: [Patch][BZ 14594] Demangle function pointers before testing them

[Jeff Law <law at redhat dot com>]

On 09/21/2012 01:30 AM, Andreas Schwab wrote:
> Jeff Law <law@redhat.com> writes:
>
> > The problem is the code is testing the mangled pointer against null which
> > is blatently wrong
>
> No, it is perfectly correct: the NULL pointer is never mangled in the
> first place.  Demangling the NULL pointer will never result in a NULL
> pointer, and all you get is a crash.
You're still missing my point.  If the clear pointer is _NOT_ NULL and 
the mangling function produces a NULL result, then the existing tests 
will do the wrong thing.

I see your point, namely that because a NULL clear pointer is not 
mangled an additional test is necessary.  ie the code really needs to do 
this:

if (ptr != NULL)
  DEMANGLE (ptr);

if (ptr != NULL)
  *(ptr) (args);


> That is trivial.  Just make sure the canary is never a valid pointer
> (you lose a bit of randomness, but that should not be a problem).
And how do you propose to do that, keeping in mind architectures where 
functions can start on a byte boundary and can live virtually anywhere 
in the address space?

jeff