crt: Check for overflows in POSIX 32-bit ftime() function

POSIX ftime() function can fail and return -1. So if the time value in
32-bit ftime call does not into struct __timeb32 then signal EOVERFLOW.

Signed-off-by: LIU Hao <lh_mouse@126.com>

mingw-w64-crt/misc/ftime32.c

diff --git a/mingw-w64-crt/misc/ftime32.c b/mingw-w64-crt/misc/ftime32.c
index 41cc31a..2ff79ba 100644
--- a/mingw-w64-crt/misc/ftime32.c
+++ b/mingw-w64-crt/misc/ftime32.c
@@ -4,12 +4,31 @@
  * No warranty is given; refer to the file DISCLAIMER.PD within this package.
  */
 
+#include <stdint.h>
+#include <errno.h>
 #include <sys/timeb.h>
 
 int __cdecl ftime32(struct __timeb32 *tb32);
 int __cdecl ftime32(struct __timeb32 *tb32)
 {
-    _ftime32(tb32);
+    /*
+     * Both 32-bit and 64-bit MS _ftime functions have void return value and do not signal overflow.
+     * So if 32-bit POSIX ftime function wants to detect overflow it has to call 64-bit _ftime function.
+     * msvc defines ftime as alias to _ftime, which always fills all members even if they overflow.
+     * So for compatibility with application code written for msvc ftime function, always fill
+     * in our mingw-w64 POSIX ftime function all __timeb32 members, even if they are overflowed.
+     * And if overflow happens, correctly sets errno to EOVERFLOW and returns negative value.
+     */
+    struct __timeb64 tb64;
+    _ftime64(&tb64);
+    tb32->time = (__time32_t)tb64.time; /* truncate */
+    tb32->millitm = tb64.millitm;
+    tb32->timezone = tb64.timezone;
+    tb32->dstflag = tb64.dstflag;
+    if (tb64.time < INT32_MIN || tb64.time > INT32_MAX) {
+        errno = EOVERFLOW;
+        return -1;
+    }
     return 0;
 }