crash in release build, but it works on debug build

[kevinding]

Hi All,

We meet an evil condition for our project. Our project has 3 layers.
A C# layer to do some business logic, and Managed C++ layer translate
managed values to native ones or vice verse.(vs2005, /clr:oldSyntax) .
UI layer are written in native C++.

Currently, we meet a random crash bug in release build. It is ok in
debug build. If we replace the managed c++ dll with the debug one, it
works also. So we think there must be something wrong in Managed C++
layer.

We intensively convert System::String to TCHAR * in Managed C++
layer. The code snippet is as belows:

TCHAR * strName = NULL;
System::String * Name = dv->Name; //get the value from C# layer
if(Name)
{
nLength = Name->Length;
strName = new TCHAR[nLength+1];
ptr = Marshal::String ToHGlobalUni(Na me);
if(ptr != IntPtr::Zero)
{
lstrcpy(strName ,
reinterpret_cas t<__wchar_t*>(s tatic_cast<void *>(ptr)));
}
}

The crash are random. But all crashes are occur after some code try
to alloc some native memory.

Call stack in Crash Scenario 1
Unhandled exception at 0x77fcca14 (NTDLL.DLL) : 0xC0000005: Access
violation writing location 0x00000000.

NTDLL.DLL!_Rtlp CoalesceFreeBlo cks@16
NTDLL.DLL!_Rtlp ExtendHeap@8
NTDLL.DLL!_RtlA llocateHeap@12
msvcr80.dll!mal loc
msvcr80.dll!ope rator new
05f1d90c()
mscorwks.dll!Se curityPolicy::I sDefaultThreadS ecurityInfo()(
...

Call stack in Crash Scenario 2
Unhandled exception at 0x77fcc024 (NTDLL.DLL): 0xC0000005: Access
violation writing location 0x00000052.

NTDLL.DLL!_RtlA llocateHeap@12
KERNEL32.DLL!_L ocalAlloc@8
MFC71u.dll!CNoT rackObject::ope rator new
MFC71u.dll!CThr eadSlotData::Se tValue
MFC71u.dll!CThr eadLocalObject: :GetData
MFC71u.dll!AFX_ MAINTAIN_STATE2 ::AFX_MAINTAIN_ STATE2
xxxxxx.dll!02c1 0101() (it is our UI dll)
...

But when I use Performance Monitor, the private heap is used
little. Also the usage of CLR memory is very low also. What is happen
to the memory heap?

Do you know what is wrong to Managed C++ module? Is there a good
tool to monitor the memory to the Interop program? Or there are some
special compiler setting for release build.

Any help are appreciated.

Thanks.

ding

[Oleg Starodumov]

Such call stacks usually mean that the native heap is corrupted.
You can try to enable PageHeap to detect the place where
the corruption occurs.
http://www.debuginfo.com/tips/userbpntdll.html

Regards,
Oleg
[VC++ MVP http://www.debuginfo.com/]

[kevinding]

Hi Oleg,

We had solved the bug. The developer had wrote some code that cause
buffer overflow. We are very new with this kind of bug in Managed C++,
it always crashed when any code want to allocate the memory. We found
Heap Manager is corrupt already, but we don't think it more.

Thanks very much for your kindly help.

Regards,

ding
Oleg Starodumov 写�：

Such call stacks usually mean that the native heap is corrupted.
You can try to enable PageHeap to detect the place where
the corruption occurs.
http://www.debuginfo.com/tips/userbpntdll.html

Regards,
Oleg
[VC++ MVP http://www.debuginfo.com/]