rand_win.c may use BCryptGenRandom with BCRYPT_USE_SYSTEM_PREFERRED_RNG

[noloader]

I believe there's a few issues with rand_win.c.

First, BCryptGenRandom is called with BCRYPT_USE_SYSTEM_PREFERRED_RNG. On some version of Windows, I believe that's the infamous Dual EC generator. Apparently Microsoft has been infiltrated by the NSA, too (as if NSA_KEY was not evidence enough).

Second, BCryptGenRandom is used without a call to BCryptOpenAlgorithmProvider. I'm fairly certain using BCryptOpenAlgorithmProvider like below avoids Dual EC. Its not as convenient as using BCRYPT_USE_SYSTEM_PREFERRED_RNG (gee, what a coincidence):

#ifdef RAND_WINDOWS_USE_BCRYPT
# include <bcrypt.h>
# ifndef BCRYPT_SUCCESS
#  define BCRYPT_SUCCESS(Status) (((NTSTATUS)(Status)) >= 0)
# endif
#else
# include <wincrypt.h>
#endif
...

#if RAND_WINDOWS_USE_BCRYPT
NTSTATUS ret = ::BCryptOpenAlgorithmProvider(&m_handle,
                                             BCRYPT_RNG_ALGORITHM,
                                             MS_PRIMITIVE_PROVIDER, 0);
ASSERT(BCRYPT_SUCCESS(ret));
...
#endif

Third, the Bcrypt library defines success as (NTSTATUS)ret >= 0, and not (NTSTATUS)ret == STATUS_SUCCESS.

[Mari2411]

(NTSTATUS)ret >= 0, (NTSTATUS)ret

== STATUS_SUCCESS

[richsalz]

Fixed with the recent RAND system updates.