Fixed ctor/dtor issue allowing use of static qualifiers for non primitives.

 * split CRT in a C and C++ part
 * use "fake" entry point to init CRT and set a DriverUnload routine for de-init
 * added -Wl,--exclude-all-symbols to DRIVER_LDFLAGS

Signed-off-by: Toni Uhlig <matzeton@googlemail.com>

12 files changed, 235 insertions, 86 deletions

diff --git a/DriverThread.cpp b/CRT/DriverThread.cpp
index efc4023..efc4023 100644
--- a/DriverThread.cpp
+++ b/CRT/DriverThread.cpp
diff --git a/DriverThread.hpp b/CRT/DriverThread.hpp
index d00db1b..d00db1b 100644
--- a/DriverThread.hpp
+++ b/CRT/DriverThread.hpp
diff --git a/EASTL-compat/kcrt.cpp b/CRT/kcrt.c
index 885dadb..ccff181 100644
--- a/EASTL-compat/kcrt.cpp
+++ b/CRT/kcrt.c
@@ -1,14 +1,30 @@
 /*
  * Shameless copy pasta from: https://github.com/sidyhe/dxx
+ * and: https://github.com/liupengs/Mini-CRT
+ * and some minor modifications.
  */
 
-#include <cstdio>
-#include <cstdlib>
-
 #include <ntddk.h>
 
 #define KCRT_POOL_DEFAULT_TAG 0xDEADBEEF
 
+extern void (*__CTOR_LIST__)();
+extern void (*__DTOR_LIST__)();
+extern NTSTATUS __cdecl DriverEntry(_In_ struct _DRIVER_OBJECT * DriverObject, _In_ PUNICODE_STRING RegistryPath);
+extern void __cdecl DriverUnload(_In_ struct _DRIVER_OBJECT * DriverObject);
+
+DRIVER_INITIALIZE __cdecl _CRT_DriverEntry;
+DRIVER_UNLOAD __cdecl _CRT_DriverUnload;
+
+typedef void (*__cdecl init_and_deinit_fn)(void);
+typedef void (*__cdecl atexit_func_t)(void);
+
+typedef struct _func_node
+{
+    atexit_func_t func;
+    struct _func_node * next;
+} func_node;
+
 typedef struct _MALLOC_HEADER
 {
     ULONG32 Tags;
@@ -17,6 +33,8 @@ typedef struct _MALLOC_HEADER
 } MALLOC_HEADER, *PMALLOC_HEADER;
 C_ASSERT(sizeof(MALLOC_HEADER) % sizeof(void *) == 0);
 
+static func_node * atexit_list = NULL;
+
 // dynamic memory mgmt
 
 PMALLOC_HEADER GET_MALLOC_HEADER(PVOID ptr)
@@ -41,6 +59,28 @@ ULONG_PTR GET_MALLOC_SIZE(PVOID ptr)
 
 // c runtime
 
+static int register_atexit(atexit_func_t func)
+{
+    func_node * node;
+    if (!func)
+        return -1;
+
+    node = (func_node *)malloc(sizeof(func_node));
+
+    if (node == 0)
+        return -1;
+
+    node->func = func;
+    node->next = atexit_list;
+    atexit_list = node;
+    return 0;
+}
+
+int __cdecl atexit(atexit_func_t func)
+{
+    return register_atexit(func);
+}
+
 void __cdecl free(void * ptr)
 {
     if (ptr)
@@ -107,56 +147,13 @@ void * __cdecl realloc(void * ptr, size_t new_size)
     return NULL;
 }
 
-// new & delete
-
-void * __cdecl operator new(std::size_t size)
-{
-    return malloc(size);
-}
-
-void * __cdecl operator new[](size_t size)
-{
-    return malloc(size);
-}
-
-void __cdecl operator delete(void * ptr)
-{
-    free(ptr);
-}
-
-void __cdecl operator delete(void * ptr, size_t)
-{
-    free(ptr);
-}
-
-void __cdecl operator delete[](void * ptr, long long unsigned int)
-{
-    free(ptr);
-}
-
-void __cdecl operator delete[](void * ptr)
-{
-    free(ptr);
-}
-
-// EASTL
-
-void * operator new[](size_t size, const char *, int, unsigned, const char *, int)
-{
-    return malloc(size);
-}
-void * operator new[](size_t size, size_t, size_t, const char *, int, unsigned, const char *, int)
-{
-    return malloc(size);
-}
-
-extern "C" void __cxa_pure_virtual(void)
+void __cdecl __cxa_pure_virtual(void)
 {
     // definitly not perfect, but we get at least a notification
     while (1)
     {
         DbgPrint("Pure virtual function call..\n");
-        LARGE_INTEGER li = { .QuadPart = -10000000 };
+        LARGE_INTEGER li = {.QuadPart = -10000000};
         KeDelayExecutionThread(KernelMode, TRUE, &li);
     }
 }
@@ -182,7 +179,7 @@ extern "C" void __cxa_pure_virtual(void)
         }                                                                                                              \
     } while (0)
 
-extern "C" float ceilf(float x)
+float __cdecl ceilf(float x)
 {
     union {
         float f;
@@ -213,3 +210,64 @@ extern "C" float ceilf(float x)
     }
     return u.f;
 }
+
+// functions called in DRIVER_INITIALIZE and DRIVER_UNLOAD
+
+static void __cdecl __ctors(void)
+{
+    unsigned long long int const * const * const l = (unsigned long long int const * const * const)&__CTOR_LIST__;
+    unsigned long long int i = (unsigned long long int)*l;
+    init_and_deinit_fn const * p;
+
+    if (i == (unsigned long long int)-1)
+    {
+        for (i = 1; l[i] != NULL; i++)
+            ;
+        i--;
+    }
+
+    p = (init_and_deinit_fn *)&l[i];
+
+    while (i--)
+    {
+        (**p--)();
+    }
+}
+
+static void __cdecl __dtors(void)
+{
+    func_node * p = atexit_list;
+    for (; p != NULL; p = p->next)
+    {
+        p->func();
+        free(p);
+    }
+    atexit_list = NULL;
+}
+
+void __cdecl KCRT_OnDriverEntry(void)
+{
+    __ctors();
+}
+
+void __cdecl KCRT_OnDriverUnload(void)
+{
+    __dtors();
+}
+
+void __cdecl _CRT_DriverUnload(_In_ struct _DRIVER_OBJECT * DriverObject)
+{
+    DriverUnload(DriverObject);
+
+    KCRT_OnDriverUnload();
+}
+
+NTSTATUS __cdecl _CRT_DriverEntry(_In_ struct _DRIVER_OBJECT * DriverObject, _In_ PUNICODE_STRING RegistryPath)
+{
+    KCRT_OnDriverEntry();
+
+    /* support for service stopping and CRT de-init */
+    DriverObject->DriverUnload = _CRT_DriverUnload;
+
+    return DriverEntry(DriverObject, RegistryPath);
+}
diff --git a/CRT/kcrt.cpp b/CRT/kcrt.cpp
new file mode 100644
index 0000000..32681ab
--- /dev/null
+++ b/CRT/kcrt.cpp
@@ -0,0 +1,51 @@
+/*
+ * Shameless copy pasta from: https://github.com/sidyhe/dxx
+ * and: https://github.com/liupengs/Mini-CRT
+ * and some minor modifications.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+
+// new & delete
+
+void * __cdecl operator new(std::size_t size)
+{
+    return malloc(size);
+}
+
+void * __cdecl operator new[](size_t size)
+{
+    return malloc(size);
+}
+
+void __cdecl operator delete(void * ptr)
+{
+    free(ptr);
+}
+
+void __cdecl operator delete(void * ptr, size_t)
+{
+    free(ptr);
+}
+
+void __cdecl operator delete[](void * ptr, long long unsigned int)
+{
+    free(ptr);
+}
+
+void __cdecl operator delete[](void * ptr)
+{
+    free(ptr);
+}
+
+// EASTL
+
+void * operator new[](size_t size, const char *, int, unsigned, const char *, int)
+{
+    return malloc(size);
+}
+void * operator new[](size_t size, size_t, size_t, const char *, int, unsigned, const char *, int)
+{
+    return malloc(size);
+}
diff --git a/EASTL-compat/ucrt.cpp b/CRT/ucrt.cpp
index 19ea583..19ea583 100644
--- a/EASTL-compat/ucrt.cpp
+++ b/CRT/ucrt.cpp
diff --git a/Makefile b/Makefile
index ec29919..de851e0 100644
--- a/Makefile
+++ b/Makefile
@@ -51,9 +51,9 @@ distclean: clean
 
 clean:
 	$(MAKE) -C $(DPP_ROOT) -f Makefile.deps clean
-	rm -f $(1_OBJECTS) $(1_TARGET)
-	rm -f $(2_OBJECTS) $(2_TARGET)
-	rm -f $(3_OBJECTS) $(3_TARGET)
+	rm -f $(1_OBJECTS) $(1_TARGET) $(1_TARGET).map
+	rm -f $(2_OBJECTS) $(2_TARGET) $(2_TARGET).map
+	rm -f $(3_OBJECTS) $(3_TARGET) $(3_TARGET).map
 
 .PHONY: all install distclean clean
 .DEFAULT_GOAL := all
diff --git a/Makefile.deps b/Makefile.deps
index 9f883f8..ca3d22d 100644
--- a/Makefile.deps
+++ b/Makefile.deps
@@ -46,8 +46,9 @@ deps-build: \
 	$(LOCAL_MINGW64_CC) \
 	$(EASTL_STATIC_LIB) \
 	$(DRIVER_ADDITIONAL_OBJS) \
-	$(DRIVER_EASTL_COMPAT) \
-	$(USER_EASTL_COMPAT) \
+	$(DRIVER_CRT) \
+	$(DRIVER_CRTPLUSPLUS) \
+	$(USER_CRT) \
 	$(SIGNTOOL_PREFIX)
 
 deps: deps-print-local-notice deps-build
@@ -73,7 +74,7 @@ distclean: clean
 	git submodule deinit --all
 
 clean:
-	rm -f $(DRIVER_EASTL_COMPAT) $(USER_EASTL_COMPAT) $(EASTL_STATIC_LIB)
+	rm -f $(DRIVER_CRT) $(DRIVER_CRTPLUSPLUS) $(USER_CRT) $(EASTL_STATIC_LIB)
 	rm -f $(DRIVER_ADDITIONAL_OBJS)
 	rm -rf EASTL-build
 
diff --git a/Makefile.inc b/Makefile.inc
index 153e84c..3535464 100644
--- a/Makefile.inc
+++ b/Makefile.inc
@@ -18,12 +18,16 @@ CMAKE = cmake
 CC = $(LOCAL_MINGW64_CC)
 CXX = $(dir $(CC))x86_64-w64-mingw32-g++
 DDK_INCLUDE_DIR = $(dir $(CC))../x86_64-w64-mingw32/include/ddk
-CFLAGS := -Wall -Wextra -m64 -fPIC -fvisibility=hidden \
+CFLAGS := -Wall -Wextra -Wno-sign-compare \
+	-m64 -fPIC -fvisibility=hidden \
 	-ffunction-sections -fdata-sections -fno-builtin -ffreestanding \
-	-I$(DPP_ROOT) -I$(DDK_INCLUDE_DIR) \
+	-I$(DPP_ROOT)/CRT -I$(DDK_INCLUDE_DIR) \
 	-D__INTRINSIC_DEFINED_InterlockedBitTestAndSet \
 	-D__INTRINSIC_DEFINED_InterlockedBitTestAndReset
-CXXFLAGS := -fno-exceptions -fno-rtti
+ifneq ($(WERROR),)
+CFLAGS += -Werror
+endif
+CXXFLAGS := -fno-exceptions -fno-rtti -fuse-cxa-atexit
 EASTL_CXXFLAGS := -I$(DPP_ROOT)/EASTL/include -I$(DPP_ROOT)/EASTL/test/packages/EABase/include/Common \
 	-DEASTL_THREAD_SUPPORT_AVAILABLE=0 \
 	-DEASTL_EXCEPTIONS_ENABLED=0 \
@@ -39,24 +43,27 @@ DRIVER_LDFLAGS := -shared \
 	-Wl,--file-alignment,0x200 \
 	-Wl,--section-alignment,0x1000 \
 	-Wl,--stack,0x100000 \
-	-Wl,--entry,DriverEntry \
 	-Wl,--gc-sections \
+	-Wl,--exclude-all-symbols \
+	-Wl,--entry,_CRT_DriverEntry \
 	-nostartfiles -nodefaultlibs -nostdlib
 DRIVER_LIBS := -lntoskrnl -lhal
-USER_LDFLAGS := -Wl,--dynamicbase -Wl,--nxcompat \
-	-Wl,--gc-sections
+USER_LDFLAGS := -Wl,--dynamicbase -Wl,--nxcompat -Wl,--gc-sections
 
-DRIVER_ADDITIONAL_DEPS := $(DPP_ROOT)/DriverThread.cpp $(DPP_ROOT)/DriverThread.hpp
-DRIVER_ADDITIONAL_OBJS := $(DPP_ROOT)/DriverThread.opp
+DRIVER_ADDITIONAL_DEPS := $(DPP_ROOT)/CRT/DriverThread.cpp $(DPP_ROOT)/CRT/DriverThread.hpp
+DRIVER_ADDITIONAL_OBJS := $(DPP_ROOT)/CRT/DriverThread.opp
 
 EASTL_DEPS := $(wildcard $(DPP_ROOT)/EASTL/source/*.cpp) $(wildcard $(DPP_ROOT)/EASTL/include/EASTL/*.h)
 EASTL_STATIC_LIB := $(DPP_ROOT)/EASTL-build/libEASTL.a
 
-DRIVER_EASTL_COMPAT_DEPS := $(DPP_ROOT)/EASTL-compat/kcrt.cpp
-DRIVER_EASTL_COMPAT := $(DPP_ROOT)/EASTL-compat/kcrt.opp
+DRIVER_CRT_DEPS := $(DPP_ROOT)/CRT/kcrt.c
+DRIVER_CRT := $(DPP_ROOT)/CRT/kcrt.o
+
+DRIVER_CRTPLUSPLUS_DEPS := $(DPP_ROOT)/CRT/kcrt.cpp $(DPP_ROOT)/CRT/kcrt.c
+DRIVER_CRTPLUSPLUS := $(DPP_ROOT)/CRT/kcrt.opp $(DPP_ROOT)/CRT/kcrt.o
 
-USER_EASTL_COMPAT_DEPS := $(DPP_ROOT)/EASTL-compat/ucrt.cpp
-USER_EASTL_COMPAT := $(DPP_ROOT)/EASTL-compat/ucrt.opp
+USER_CRT_DEPS := $(DPP_ROOT)/CRT/ucrt.cpp
+USER_CRT := $(DPP_ROOT)/CRT/ucrt.opp
 
 is_set = \
 	$(if $1,, \
@@ -72,8 +79,9 @@ define CHECK_REQUIRED_PATHS
 	$(call path_exists,$(DDK_INCLUDE_DIR))
 	$(call path_exists,$(DRIVER_ADDITIONAL_OBJS))
 	$(call path_exists,$(EASTL_STATIC_LIB))
-	$(call path_exists,$(DRIVER_EASTL_COMPAT))
-	$(call path_exists,$(USER_EASTL_COMPAT))
+	$(call path_exists,$(DRIVER_CRT))
+	$(call path_exists,$(DRIVER_CRTPLUSPLUS))
+	$(call path_exists,$(USER_CRT))
 endef
 
 define BUILD_C_OBJECT
@@ -99,8 +107,10 @@ define LINK_C_KERNEL_TARGET
 	$(Q)$(CC) \
 		$(CFLAGS) \
 		$(DRIVER_LDFLAGS) \
+		-Wl,-Map='$(2).map' \
 		-o '$(2)' \
 		$(1) \
+		$(DRIVER_CRT) \
 		$(DRIVER_LIBS)
 	@echo 'LD  $(2)'
 endef
@@ -126,10 +136,11 @@ define LINK_CPP_KERNEL_TARGET
 		$(CXXFLAGS) \
 		$(EASTL_CXXFLAGS) \
 		$(DRIVER_LDFLAGS) \
+		-Wl,-Map='$(2).map' \
 		-o '$(2)' \
 		$(1) \
 		$(DRIVER_ADDITIONAL_OBJS) \
-		$(DRIVER_EASTL_COMPAT) \
+		$(DRIVER_CRTPLUSPLUS) \
 		$(EASTL_STATIC_LIB) \
 		$(DRIVER_LIBS)
 	@echo 'LD  $(2)'
@@ -146,7 +157,7 @@ define LINK_CPP_USER_TARGET
 		$(USER_LDFLAGS) \
 		-o '$(2)' \
 		$(1) \
-		$(USER_EASTL_COMPAT) \
+		$(USER_CRT) \
 		$(EASTL_STATIC_LIB)
 	@echo 'LD  $(2)'
 endef
diff --git a/README.md b/README.md
index d8c8173..c808e5d 100644
--- a/README.md
+++ b/README.md
@@ -67,19 +67,31 @@ install: $(DRIVER_TARGET)
     $(call INSTALL_EXEC_SIGN,$(DRIVER_TARGET))
 ```
 
-## Known issues
+## The CRT and CRT++
 
-Defining classes (by value, not pointer) with a static qualifier won't work because something seems broken with the static initialization of (non-primitive) globals.
-Meaning code `static MyClass test;` fails whereas `static MyClass * test = new MyClass();` will sill work.
+This project uses a very very rudimentary CRT for C and C++ projects.
+Please keep in mind that depending on what you want to do the CRT may lack features you are familiar with.
+Usually copy&pasting them from various online sources should be sufficient.
 
-The issue is related to `__static_initialization_and_destruction` as it does not call the appropriate ctor's and dtor's (pushing dtor's via `atexit()`).
-The latter one may be related to `-Wl,--subsystem,native` as in kernel space there is usually no need for calling `atexit()`.
-I've tried adding C++ initializers manually, but they did not survive linker optimizations. There may be custom linker script required to fix it.
+Remember: The CRT and CRT++ **sets a driver unload function** meaning that code .e.g.:
+
+```C
+NTSTATUS MyDriverEntry(_In_ struct _DRIVER_OBJECT * DriverObject, _In_ PUNICODE_STRING RegistryPath)
+{
+    DriverObject->DriverUnload = MyDriverUnload;
+}
+```
+
+shouldn't be used. Instead the function `DriverUnload` will be called.
+So make sure that the symbol `DriverUnload` exists and has the usual ddk function signature:
+`void DriverUnload(_In_ struct _DRIVER_OBJECT * DriverObject)`.
+This is required to make ctors/dtors work without calling additional functions in `DriverEntry` / `DriverUnload`.
 
 ## Thanks!
 
 - [Zeranoe](https://github.com/Zeranoe/mingw-w64-build) for the Mingw64 build script
 - [sidyhe](https://github.com/sidyhe/dxx) for some copy paste ready CRT code ;)
+- [liupengs](https://github.com/liupengs/Mini-CRT) helped me to fix the ctor/dtor issue
 
 and last but not least:
 
diff --git a/ddk-template-cplusplus-EASTL.cpp b/ddk-template-cplusplus-EASTL.cpp
index e099488..d3cb0e3 100644
--- a/ddk-template-cplusplus-EASTL.cpp
+++ b/ddk-template-cplusplus-EASTL.cpp
@@ -107,9 +107,6 @@ extern "C"
 
         DbgPrint("%s\n", "Hello ring0!");
 
-        /* support for service stopping */
-        DriverObject->DriverUnload = DriverUnload;
-
         stl_test();
         more_stl_test();
 
diff --git a/ddk-template-cplusplus.cpp b/ddk-template-cplusplus.cpp
index 380603a..f9167ec 100644
--- a/ddk-template-cplusplus.cpp
+++ b/ddk-template-cplusplus.cpp
@@ -35,6 +35,29 @@ public:
     }
 };
 
+class DerivedWithCDtor : public Derived
+{
+public:
+    explicit DerivedWithCDtor(unsigned int value)
+    {
+        some_value = value;
+        DbgPrint("%s\n", "DerivedWithCDtor-Ctor.");
+    }
+    ~DerivedWithCDtor()
+    {
+        DbgPrint("%s\n", "DerivedWithCDtor-Dtor.");
+    }
+    void doSmth(void)
+    {
+        DbgPrint("SomeValue: %X\n", some_value);
+    }
+
+private:
+    unsigned int some_value = 0;
+};
+
+static DerivedWithCDtor some_static(0xDEADC0DE);
+
 struct threadContext
 {
     DriverThread::Semaphore sem;
@@ -70,6 +93,8 @@ static void test_cplusplus(void)
     ctx.dth.WaitForTermination();
     ctx.dth.WaitForTermination();
     DbgPrint("MainThread EOF\n");
+
+    some_static.doSmth();
 }
 
 extern "C"
@@ -86,9 +111,6 @@ extern "C"
         DbgPrint("%s\n", "Hello ring0!");
         cdtor_test = new TestSmth();
 
-        /* support for service stopping */
-        DriverObject->DriverUnload = DriverUnload;
-
         test_cplusplus();
 
         return STATUS_SUCCESS;
diff --git a/ddk-template.c b/ddk-template.c
index 9667b18..2d7becd 100644
--- a/ddk-template.c
+++ b/ddk-template.c
@@ -13,9 +13,6 @@ NTSTATUS DriverEntry(
 
     DbgPrint("%s\n", "Hello ring0!");
 
-    /* support for service stopping */
-    DriverObject->DriverUnload = DriverUnload;
-
     return STATUS_SUCCESS;
 }