Re: thoughts on reallocarray, explicit_bzero?

["Daniel Cegiełka" <daniel.cegielka@gmail.com>]

2015-01-29 11:04 GMT+01:00 Szabolcs Nagy <nsz@port70.net>:
> * Daniel Cegie??ka <daniel.cegielka@gmail.com> [2015-01-29 10:30:40 +0100]:
>> yet another secure_memzero(). A better solution would be to promote a
>> single standard (eg. memset_s()) and the expectation that the compiler
>> will respect it.
>>
>
> i think you don't know the semantics of memset_s
> (it uses nonsense types, has superflous arguments, handles
> constraint violations through global state etc)


btw. memset_s() is an attempt to solve the same problem. However, this
version will not work with LTO:

ftp://ftp.netbsd.org/pub/NetBSD/misc/apb/memset_s.20120224.diff

#include <sys/cdefs.h>

__RCSID("$NetBSD$");

#define __STDC_WANT_LIB_EXT1__ 1
#include <errno.h>
#include <stdint.h>
#include <string.h>

/*
 * __memset_vp is a volatile pointer to a function.
 * It is initialised to point to memset, and should never be changed.
 */
static void * (* const volatile __memset_vp)(void *, int, size_t)
= (memset);

#undef memset_s /* in case it was defined as a macro */

errno_t
memset_s(void *s, rsize_t smax, int c, rsize_t n)
{
errno_t err = 0;

if (s == NULL) {
err = EINVAL;
goto out;
}
if (smax > RSIZE_MAX) {
err = E2BIG;
goto out;
}
if (n > RSIZE_MAX) {
err = E2BIG;
n = smax;
}
if (n > smax) {
err = EOVERFLOW;
n = smax;
}

/* Calling through a volatile pointer should never be optimised away. */
(*__memset_vp)(s, c, n);

    out:
if (err == 0)
return 0;
else {
errno = err;
/* XXX call runtime-constraint handler */
return err;
}
}