mbedtls_platform_gmtime_r windows bug

[miroslav-mastny-regosys]

Description

• Type: Bug
• Priority: Major

---

OS
windows (C++ Builder)

mbed TLS build:
Version: 2.16.9

WIN32 function gmtime_s has parameters and return value wrong order.

Should be:
#if defined(_WIN32) && !defined(EFIX64) && !defined(EFI32)
return( ( gmtime_s( tt, tm_buf ) == 0 ) ? NULL: tm_buf );
#elif !defined(PLATFORM_UTIL_USE_GMTIME)

Currently implemented:
#if defined(_WIN32) && !defined(EFIX64) && !defined(EFI32)
return( ( gmtime_s( tm_buf, tt ) == 0 ) ? tm_buf : NULL );
#elif !defined(PLATFORM_UTIL_USE_GMTIME)

[gabor-mezei-arm]

It looks like the C++ Builder using the standard C function instead of the windows' one.

Can you help me about which macro is defined by C++ Builder and what are their value?
These are the interested macros:
_POSIX_VERSION
_POSIX_THREAD_SAFE_FUNCTIONS
__STDC__

[miroslav-mastny-regosys]

Any POSIX macro is not set.
Macro STDC is set to 1 when building as 32bit C project and ANSI conformance compiler flag is set (it seems is not set by default and there is not direct switch in project setting - only user compiler switch field is useable for that)

C++Builder definition of gmtime_s in file <time.h>:
struct tm * _RTLENTRY _EXPFUNC gmtime_s(const time_t * _RESTRICT clock,
struct tm * _RESTRICT result);

[gabor-mezei-arm]

The gmtime handling is not prepared for standard-C-like gmtime_s function in windows.

[gstrauss]

Is this still an issue? The usage of gmtime_s() is centralized to a single use in library/platform_util.c:mbedtls_platform_gmtime_r() and the argument order looks correct to me.

[gabor-mezei-arm]

Yes, the gmtime_s function should be called the same way as gmtime_r if the __STDC__ and _WIN32 macros are defined.

The standard (C11):
struct tm *gmtime_s( const time_t *restrict timer, struct tm *restrict buf );

The MSVC provides:
errno_t gmtime_s( struct tm* tmDest, const __time_t* sourceTime );

[gstrauss]

ick. Now I see that the signature is incompatible depending on the "standards" view in the headers.
https://en.cppreference.com/w/c/chrono/gmtime
Since gmtime_s() was added in C11 Annex K, is the definition properly hidden by checking for the following?

#define __STDC_WANT_LIB_EXT1__ 1
#include <time.h>
...
#ifdef __STDC_LIB_EXT1__
/* gmtime_s() C11 prototype view */
#else
/* gmtime_s() MSVC prototype view */
#endif

[daverodgman]

Fixed by #4211