https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104964

Siddhesh Poyarekar <siddhesh at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED

--- Comment #10 from Siddhesh Poyarekar <siddhesh at gcc dot gnu.org> ---
OK, I have a representative reproducer, which TBH is not too different from the
one you posted, just that it succeeds with __builtin_object_size and fails with
__builtin_dynamic_object_size:


struct __string_ext
{
  char s_str[0];
};

typedef struct
{
  int o_prefix;
  struct __string_ext i;
} string_obj;

#define SUFFIX ".suffix"

string_obj *
__acl_to_any_text (unsigned long n)
{
  unsigned long off = 0;
  unsigned long size = sizeof SUFFIX;

  string_obj *obj = __builtin_malloc (sizeof (string_obj) + size);

  if (n == 0)
          __builtin_unreachable ();

  while (n-- != 0)
    {
      if (off + 1 > size - sizeof SUFFIX)
        {
          size <<= 1;
          string_obj *tmp = __builtin_realloc (obj, sizeof (string_obj) +
                                               size);
          if (!tmp)
            __builtin_unreachable ();
          obj = tmp;
        }
      obj->i.s_str[off++] = 'A';
    }

  char *t = obj->i.s_str + off;
  __strcpy_chk (t, SUFFIX, __builtin_dynamic_object_size (t, 1));

  return obj;
}

int
main ()
{
  string_obj *s = __acl_to_any_text (32);

  __builtin_printf ("%zu: %s\n", __builtin_strlen (s->i.s_str), s->i.s_str);
  return 0;
}


$ gcc/cc1 -g -o test.s -quiet -Wall -O3 fs3.c
fs3.c: In function ‘__acl_to_any_text’:
fs3.c:40:3: warning: ‘__builtin___memcpy_chk’ writing 8 bytes into a region of
size 0 overflows the destination [-Wstringop-overflow=]
   40 |   __strcpy_chk (t, SUFFIX, __builtin_dynamic_object_size (t, 1));
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


The only reason why __builtin_object_size fails is because of the non-constant
OFF.  If that is removed, __builtin_object_size also returns the declared size
of s_str, i.e. 0.  The check for a traditionally declared trailing array ()i.e.
a[0] or  a[1]) seems to be broken for nested structs like the above.  Change
that to s_str[] (the struct then needs another member above) and it works fine.

Reply via email to