Deleting an unmanaged object from Finalize throws an AccessViolationException - c#

I'm writing a wrapper library using C++/CLI for an unmanaged C++ project. The goal is to expose this library to C# applications. Here's what I have so far.
#pragma once
#include "Settings.h"
#include "Settings/SettingsPrivate.h"
public ref class Settings
{
public:
Settings();
virtual ~Settings();
protected:
!Settings();
public:
unsigned char GetModel(int iNumber);
private:
CSettings* m_pSettings;
};
#include "stdafx.h"
#include "Managed/Settings.h"
Settings::Settings()
{
// Pointer to unmanaged object
m_pSettings = new CSettings();
}
Settings::~Settings()
{
this->!Settings();
}
Settings::!Settings()
{
if (m_pSettings)
{
delete m_pSettings;
m_pSettings = NULL;
}
}
unsigned char Settings::GetModel(int iNumber)
{
return m_pSettingss->GetModel(iNumber);
}
The code executes fine in the test application I wrote. The function call succeeds. The problem is that when the GC Finalizes this object, it throws an exception.
An unhandled exception of type 'System.AccessViolationException' occurred in Wrapper.dll
Additional information: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
I can't see any obvious reason why this exception is being thrown. I tried disposing of the object explicitly by calling Dispose from my C# application. It still throws the same exception.
Here's the test application:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
namespace WrapperTest
{
class Program
{
static void Main(string[] args)
{
Settings settings = new Settings();
byte b = settings.GetModel(0);
settings.Dispose();
return;
}
}
}
Would someone point out what I'm doing wrong??

It was a project configuration error. The code actually ran fine in Release mode.
In debug mode I linked in some release DLL's when I should have been statically linking against debug libraries. Why this caused memory corruption I haven't yet investigated but it has solved the problem.
Otherwise the code posted above is correct.

You should enable more verbose heap debugging functionality. I tested your code and it did not fail for me. I had to use malloc/free instead of new/delete as you didnt define CSettings, however the effect should be the same.
I added this to make sure I had enough heap churn to trigger a failure if their was any corruption;
unsigned char Settings::GetModel(int iNumber)
{
for(int i=0; i < iNumber; i++)
free(malloc(1024));
return iNumber;
}
For me your code worked with no failure. You should look into some of your compilation settings, are you linking to an external library for CSettings? If so you need to make sure the CRT is the same version, etc. Also you know you have to use one of the DLL run time's not /MT or /MTd.

Related

Why does calling libsass functions with .NET P/Invoke fail?

I’ve compiled libsass 3.3.6 into a DLL with VS 2015 using the included solution files. Running the code below causes the program to crash immediately with no output.
using System;
using System.Runtime.InteropServices;
namespace Sass.Cli {
static class Sass {
[DllImport(#"C:\...\libsass\win\bin\libsass.dll")]
public static extern String libsass_version();
}
class Program {
static void Main(string[] args) {
Console.WriteLine(Sass.libsass_version());
}
}
}
The source for the invoked function is at sass.cpp:61.
const char* ADDCALL libsass_version(void)
Both the DLL and the above code were compiled for x86. Running the VS debugger on crash gives me the following exception: Unhandled exception at 0x771A9841 (ntdll.dll) in Sass.Cli.exe: 0xC0000374: A heap has been corrupted (parameters: 0x771DC8D0).
Is this something that can be fixed or that I’ve overlooked? Or is it just the case that libsass DLLs aren’t currently working?
When a string is used as a return type then the framework assumes it was allocated by calling CoTaskMemAlloc. It then copies the content of the string and deallocation by calling CoTaskMemFree. That is the explanation for your error.
Solve it by changing the return value to IntPtr and getting the string content with Marshal.PtrToStringAnsi. You won't need to deal located anything since this version string will be static.
I can't tell what the calling convention is. You should check the code and docs to find out. Still, this function is so simple that it behaves the same way for both cdecl and stdcall so that can't explain the error. Still, you will need to get it right for the rest of the library.
Note that I am guessing a little here. You've not provided full details. You should consult the code and docs to double check my guesswork.

C# code not catching SEHExceptions

I have a C# application that invokes a managed C++ DLL that deliberately accesses to an invalid address; I enabled SEH Exceptions in my C++ project, I added a _se_translator_function into my C++ code and I also added a signal handler when a SIGSEGV occurs. Using my C++ code from a purely native test, everything works perfectly, but when I invoke my C++ code from a .net app, the app crashes with a:
Unhandled Exception: System.AccessViolationException: Attempted to
read or write protected memory. This is often an indication that other
memory is corrupt.
at K.killnative() in c:\users\ebascon\documents\visual studio 2013\projects\consoleapplication3\consoleapplication4\source.cpp:line 32
This is my C# console app:
namespace ConsoleApplication3 {
class Program {
static void Main(string[] args) {
try {
var k = new NativeKiller();
k.kill();
}
catch (Exception ex) {
Console.WriteLine("Catching " + ex);
}
}
}
}
and this is the C++/CLI code invoked:
void MXXExceptionTranslator(unsigned int, struct _EXCEPTION_POINTERS*)
{
throw std::exception("Crash happens");
}
void killnative() {
try {
_set_se_translator(MXXExceptionTranslator);
signal(SIGSEGV, [](int) {
puts("Exception");
exit(-1);
});
int* p = reinterpret_cast<int*>(0xDEADBEEF);
printf("%d\n", *p);
}
catch (...) { //Removing this catch does not change anything
puts("Doing nothing");
}
}
public ref class NativeKiller {
public:
void kill() {
killnative();
}
};
What do you think I am doing wrong? In my real world problem, I need this C++/CLI process (that is a bridge with a legacy app) to log an error message and die gracefully instead of popping the "The program stopped working" window.
Thanks in advance,
Ernesto
This is the good kind of problem to have, it helps you discover that you are not building your code correctly. The C++/CLI compiler is pretty powerful, almost too powerful, and can translate almost any native code into IL. The exact same kind of IL that a C# compiler generates. And it is treated the same at runtime, the jitter translates it to machine code at runtime.
This is not usually something you actually want. Native C or C++ code ought to be translated directly to machine code by the compiler. The MSDN article for _set_se_translator() does a decent job of warning about this:
When using _set_se_translator from managed code (code compiled with /clr) or mixed native and managed code, be aware that the translator affects exceptions generated in native code only. Any managed exceptions generated in managed code (such as when raising System::Exception) are not routed through the translator function.
The usual way to fall in the pit of success is by compiling the native code separately in its own source file or library project. But even easier is to take advantage of the C++/CLI compiler's ability to dynamically switch back-and-forth between IL and machine code generation in a single source file. Fix:
#pragma managed(push, off)
void MXXExceptionTranslator(unsigned int, struct _EXCEPTION_POINTERS*) { ... }
void killnative() { ... }
#pragma managed(pop)
public ref class NativeKiller { ... }
And you'll now see that the exception translator works fine.

Shared memory between C++ DLL and C# code

I am currently working on a project with really short deadline, so I don't have much time to understand everything. Also, I am not an expert in C++ development and memory management.
So, what I am trying to do is to create a DLL in with both C and C++ code. Then, I would like to call this DLL in a C# code. Currently, the communication between C++ and C# is OK. The problem comes up when I try to transfer a string from the DLL to the C# code. The error is this one :
System.AccessViolationException: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
at Microsoft.Win32.Win32Native.CoTaskMemFree(IntPtr ptr)
at System.StubHelpers.CSTRMarshaler.ClearNative(IntPtr pNative)
at NMSPRecognitionWrapper.Program.GetResultsExt()
at NMSPRecognitionWrapper.Program.<Main>b__0() in <my dir>\Program.cs:line 54
at NMSPRecognitionWrapper.Program.StartRecognitionExt()
at NMSPRecognitionWrapper.Program.Main(String[] args) in <my dir>\Program.cs:line 60
Also, I can give you some piece of code below (really simplified !). Actually, the C++ expose two methods : StartRecognition() launch operations to get some data from microphone, then process them and store the results. GetResults() return an instance of the results previously stored. The WrapperCallback() allows the C# part to be called when a Result is able for processing. The C# part, when the Callback is called, will ask to get the results using the GetResults() method.
I know the architecture may seem really inappropriate in this presentation, but I don't want to explain the whole project to validate the model, please be sure everything is correct.
To finish, the problem is when the C# callback call the GetResults() method. Trying to access to the resultsForCS seems to be impossible from the C#.
C++ part - header
// NMSPRecognitionLib.h
#pragma once
#include <iostream>
using namespace std;
extern "C" __declspec(dllexport) char* GetResults();
extern "C" static void DoWork();
extern "C" __declspec(dllexport) void StartRecognition();
C++ part - sources
#include "stdafx.h"
#include "NMSPRecognitionLib.h"
static char * resultsForCS;
static SUCCESS ProcessResult(NMSPCONNECTION_OBJECTS *pNmspConnectionObjects, LH_OBJECT hResult)
{
[...]
char* szResult;
[...]
resultsForCS = szResult;
DoWork();
[...]
return Success;
error:
return Failure;
} /* End of ProcessResult */
extern "C" __declspec(dllexport) char* GetResults()
{
return resultsForCS;
}
extern "C"
{
typedef void (*callback_function)();
callback_function gCBF;
__declspec(dllexport) void WrapperCallback(callback_function callback) {
gCBF = callback;
}
static void DoWork() {
gCBF();
}
}
extern "C" __declspec(dllexport) void StartRecognition()
{
char* argv[] = { "path", "params" };
entryPoint(2, argv);
}
C# part
class Program
{
[DllImport("NMSPRecognitionLib.dll", EntryPoint = "GetResults")]
[return: MarshalAs(UnmanagedType.LPStr)]
public static extern string GetResultsExt();
public delegate void message_callback_delegate();
[DllImport("NMSPRecognitionLib.dll", EntryPoint = "WrapperCallback")]
public static extern void WrapperCallbackExt(message_callback_delegate callback);
[DllImport("NMSPRecognitionLib.dll", EntryPoint = "StartRecognition")]
public static extern void StartRecognitionExt();
static void Main(string[] args)
{
WrapperCallbackExt(
delegate()
{
Console.WriteLine(GetResultsExt());
}
);
StartRecognitionExt();
Console.WriteLine("\nPress any key to finish... ");
var nothing = Console.ReadLine();
}
}
I understand that the problem comes because I am using a pointer to store the results (char *), but I actually don't know how to do this in another way. The szResults type is char * too and I can't change this !
Yes, the return type is the problem. The pinvoke marshaller must do something to release the memory that was allocated for the string. The contract is that memory allocations that need to be released by the caller must be allocated from the COM heap. CoTaskMemAlloc() in native code, also exposed in .NET as Marshal.AllocCoTaskMem().
This rarely comes to a good end, most native code allocates with malloc() or ::operator new, allocating from a heap that's created by the C runtime library. The wrong heap. So inevitably the CoTaskMemFree() call will fail. Ignored silently in Windows XP and earlier, a kaboom on Vista and up.
You must stop the pinvoke marshaller from trying to release the memory. Do so by declaring the return value as IntPtr. And use Marshal.PtrToStringAnsi() to recover the string.
You still have a Big Problem, the kind of problem that bedevils any native code that tries to use this function as well. You still have a string buffer that needs to be released. You cannot do that from C#, you can't pinvoke the correct version of free() or ::operator delete. A memory leak is inevitable. The only thing you can hope for is that the native code takes care of it, somehow. If it doesn't then you must use C++/CLI to interop with it. With the additional requirement that the native code needs to be rebuilt with the same compiler so that it uses the same shared CRT. Code that's difficult to use correctly from native code is also hard to pinvoke. That's a design flaw, always allow the caller to pass a buffer to be filled in so there's never a question who owns the memory.
Looking at:
at Microsoft.Win32.Win32Native.CoTaskMemFree(IntPtr ptr)
at System.StubHelpers.CSTRMarshaler.ClearNative(IntPtr pNative)
at NMSPRecognitionWrapper.Program.GetResultsExt()
I can see that your callback is called, but the runtime tries to free some memory. I think it assumes your pointer would be to com memory. Try converting the string yourself, it is easy!
[DllImport("NMSPRecognitionLib.dll", EntryPoint = "GetResults")]
public static extern IntPtr GetResultsExt();
[...]
string result = Marshal.PtrToStringAnsi(GetResultsExt())
No 100% guarantee, but worth a try.
I have found that it is usually easier to write a wrapper in C++/CLI around the C++ native code. A C++/CLI class can directly call and use native C++, but is accessible from C# (and any .Net language). In my experience, using DLLImport as you do leads to hard to debug and find errors.

Handling errors while using unmanaged code in a managed one ( C++, C, C++/CLI, C#)

I am using a badly written 3rd party (C/C++) Api. I used it from managed code( C++/ CLI). Get sometimes "access violation errors". And this crash whole application. I know i can not handle those errors[ what can i do if a pointer acess to illegal memory location etc].
But I do not want my application crash as a whole. At least if there is a real problem, my application gracefully should say "OK.I can not do my job.BYE. " :-) then it least execute some alternative scenarious and finally close itself.
But there seems to be no way, to catch( may be wrong term, the rigth word may be to be informed about) access violation and similiar errors.Is there a way to be informed about those errors. So i can execute my alternative scenarious.
PS: Standard Exception handling does not solve this.
#include "stdafx.h"
#include <iostream>
using namespace System;
using namespace std;
static void ThrowingManagedException()
{
throw gcnew ArgumentException("For no good reason");
}
static void ThrowingNativeException()
{
throw std::exception("For no good reason");
}
static void SomeBadThingsHappen()
{
short a[1];
a[0]=1;
a[2]= 2; // SomeOne make stupid mistake
}
int main(array<System::String ^> ^args)
{
Console::WriteLine(L"Test Exceptions");
try
{
SomeBadThingsHappen();
//ThrowingNativeException();
//ThrowingManagedException();
}
catch(Exception^ e)
{
Console::WriteLine("Something awful happened: "+ e);
}
Console::WriteLine("Press enter to exit");
Console::Read();
return 0;
}
If you're sure that the problems are bugs in the library, and not a result of you passing in bad arguments, then your most robust option is inter-process communication with a hosting process that loads the library. That way your OS process separation keeps the library from bringing down your application.
You can try to catch access violations in-process, using SEH, but if the library writes to wild pointers and not simple null pointers, then the process won't survive even with an exception handler.
Your example will not cause an access violation, it's a buffer overrun of a buffer on the stack, so the adjacent memory location contains some other valid data which gets stomped on.

What is a canonical way to produce a segmentation fault in C#?

I am interested in the shortest, neatest piece of C# code around that will reliably produce a segfault - ideally without directly calling any unmanaged code.
What you're after is somewhat unclear but I suppose this is as good as any answer so far, and it is about as minimal as you can get.
System.Runtime.InteropServices.Marshal.ReadInt32(IntPtr.Zero);
Michael's answer wasn't working for me, perhaps that case is caught now. Marshal.ReadInt32() just gives me a "SystemError: Attempted to read or write protected memory." with .NET 4.5 on Windows for various passed values. I used the following however which segfaults for me both on Windows and under mono 4.0.4.1:
using System.Runtime.InteropServices;
[UnmanagedFunctionPointer(CallingConvention.Cdecl)]
public delegate void UNMANAGED_CALLBACK();
public void Crash()
{
var crash = (UNMANAGED_CALLBACK)Marshal.GetDelegateForFunctionPointer((IntPtr) 123, typeof(UNMANAGED_CALLBACK));
crash();
}
As noted in the comments above, there's no such thing as a segfault in Windows, and you didn't say anything about mono on Linux. So I'm going to assume you actually meant an access violation.
Here's a way to get one:
unsafe {
int* a = (int*) -4;
*a = 0;
}
(Must be compiled with the /unsafe option.)
My first try used 0 as the address, but that turned out to throw a plain old NullReferenceException, which you can get without unsafe code. But the negative address gets an AccessViolationException on my Vista x64 box.
Compile with csc with the /unsafe option:
class Program
{
static void Main(string[] args)
{
unsafe
{
int *p = null;
*p = 5;
}
}
}

Categories

Resources