[原]调试实战——使用windbg调试TerminateThread导致的死锁

原调试 debug windbg 死锁 deadlock

前言

项目里的一个升级程序偶尔会死锁，查看dump后发现是死在了ShellExecuteExW里。经验少，不知道为什么，于是在高端调试论坛里发帖求助，链接如下http://advdbg.org/forums/6520/ShowPost.aspx

根据张银奎老师的描述可知，应该是拥有关键段的线程意外结束了。仔细检查项目中的代码，发现程序中有使用TerminateThread()来强制杀线程的代码。很可疑，于是写了一个测试程序，还原了这个问题。

{% note info %}

这也是几年前在项目中遇到的一个问题，我对之前的笔记进行了整理重新发布于此。

{% endnote %}

问题重现

重现方法

主程序会加载一个DLL，并调用该DLL的导出函数创建一个线程，然后调用TerminateThread()强制杀死这个线程，然后调用RunProcess()（内部封装了对ShellExecuteEx()的调用）执行一个新进程，会卡死在ShellExecuteEx()。为了让问题更容易重现，特地在DllMain()的参数ul_reason_for_call为DLL_THREAD_DETACH时，强制睡眠了5秒。

代码摘录

主工程 testTerminateThread

//testTerminateThread.cpp
#include "stdafx.h"
#include "windows.h"
#include "process.h"

typedef HANDLE (*pfnGenerateThread)();

HANDLE RunProcess(const TCHAR* app_name, const TCHAR* cmd) {
    SHELLEXECUTEINFO shex = {sizeof(SHELLEXECUTEINFO)};
    shex.fMask = SEE_MASK_NOCLOSEPROCESS;
    shex.lpVerb = _T("open");
    shex.lpFile = app_name; 
    shex.lpParameters = cmd; 
    shex.lpDirectory = NULL; 
    shex.nShow = SW_NORMAL;

    if (!::ShellExecuteEx(&shex))
    {
        return INVALID_HANDLE_VALUE;
    }

    return shex.hProcess;
}

int _tmain(int argc, _TCHAR* argv[])
{
    while ( 1 )
    {
        HMODULE hModule = LoadLibrary(_T("testDll.dll"));
        if ( NULL == hModule )
            return 0;

        pfnGenerateThread pfn = (pfnGenerateThread)GetProcAddress(hModule, "GenerateThread");
        if ( NULL == pfn )
            return 0;

        HANDLE hThread = pfn();

        // give thread time to start up
        Sleep(1000);
        
        // terminate thread.
        BOOL bOk = TerminateThread(hThread, 0);

        // dead lock in this function...
        RunProcess(argv[0], NULL);

        FreeLibrary(hModule);
    }

    return 0;
}

DLL工程 testDll

// DllMain.cpp
#include "stdafx.h"
#include "windows.h"

BOOL APIENTRY DllMain( HMODULE hModule, DWORD ul_reason_for_call, LPVOID lpReserved ) {

    switch (ul_reason_for_call)
    {
    case DLL_PROCESS_ATTACH:
        OutputDebugString(L"====> DLL_PROCESS_ATTACH called.n");
        break;
    case DLL_THREAD_ATTACH:
        OutputDebugString(L"----> DLL_THREAD_ATTACH called.n");
        break;
    case DLL_THREAD_DETACH:
        OutputDebugString(L"<---- DLL_THREAD_DETACH called.n");
        // with LdrpLoaderLock held! sleep 5 seconds. 
        Sleep(5000);
        break;
    case DLL_PROCESS_DETACH:
        OutputDebugString(L"<==== DLL_PROCESS_DETACH called.n");
        break;
    }
    return TRUE;
}

// testDll.cpp
#include "stdafx.h"
#include "stdio.h"
#include "process.h"
#include "windows.h"

void OutputCurrentThreadId() {
    TCHAR szBuffer[1024];
    swprintf_s(szBuffer, L"thread [0x%x], running & exiting...n", GetCurrentThreadId());
    OutputDebugString(szBuffer);
    return;
}

unsigned __stdcall testProc(void *) {
    OutputCurrentThreadId();
    return 0;
}

HANDLE GenerateThread() {
    HANDLE hThread = (HANDLE)_beginthreadex(NULL, 0, &testProc, NULL, 0, NULL);
    return hThread;
}

debug-deadlock-caused-by-TerminateThread-Demo.zip 10.63 KB

问题分析

运行测试程序前先打开DbgView监视调试信息，然后运行测试程序。

DebugView

从日志可知，我们启动的测试线程的线程id为0x1400。

当程序hang住后，使用windbg附加。附加成功后，先运行~*kvn查看线程及每个线程的的调用栈信息。发现只有一个0号线程（1号线程是windbg附加到进程时产生的）。

0:001> ~*kvn

   0  Id: 18c0.1008 Suspend: 1 Teb: 7ffdf000 Unfrozen
 # ChildEBP RetAddr  Args to Child              
00 002bf614 775a6a64 77592278 00000064 00000000 ntdll!KiFastSystemCallRet (FPO: [0,0,0])
01 002bf618 77592278 00000064 00000000 00000000 ntdll!NtWaitForSingleObject 0xc (FPO: [3,0,0])
02 002bf67c 7759215c 00000000 00000000 00000001 ntdll!RtlpWaitOnCriticalSection 0x13e (FPO: [Non-Fpo])
03 002bf6a4 775c00e1 77637340 77bf1b77 00000000 ntdll!RtlEnterCriticalSection 0x150 (FPO: [Non-Fpo])
04 002bf6dc 75587bc3 00000001 00000000 002bf704 ntdll!LdrLockLoaderLock 0xe4 (FPO: [Non-Fpo])
05 002bf728 7679215d 00000000 002bf73c 00000104 KERNELBASE!GetModuleFileNameW 0x75 (FPO: [Non-Fpo])
06 002bf948 76792112 002bfbb0 002bf968 7ffdb000 SHELL32!InRunDllProcess 0x39 (FPO: [Non-Fpo])
*** WARNING: Unable to verify checksum for C:UsersBianChengNanDocumentsVisual Studio 2012ProjectstestTerminateThreadDebugtestTerminateThread.exe
07 002bf95c 013714db 002bfa44 002bfcbc 002bfbc0 SHELL32!ShellExecuteExW 0x51 (FPO: [Non-Fpo])
08 002bfbb0 01371685 000ac518 00000000 00000000 testTerminateThread!RunProcess 0xdb (FPO: [Non-Fpo]) (CONV: cdecl) [c:usersbianchengnandocumentsvisual studio 2012projectstestterminatethreadtestterminatethreadtestterminatethread.cpp @ 28]
09 002bfcbc 01371c69 00000001 000ac510 000ae660 testTerminateThread!wmain 0xc5 (FPO: [Non-Fpo]) (CONV: cdecl) [c:usersbianchengnandocumentsvisual studio 2012projectstestterminatethreadtestterminatethreadtestterminatethread.cpp @ 59]
0a 002bfd0c 01371e5d 002bfd20 758ced6c 7ffdb000 testTerminateThread!__tmainCRTStartup 0x199 (FPO: [Non-Fpo]) (CONV: cdecl) [f:ddvctoolscrt_bldself_x86crtsrccrtexe.c @ 533]
0b 002bfd14 758ced6c 7ffdb000 002bfd60 775c37eb testTerminateThread!wmainCRTStartup 0xd (FPO: [Non-Fpo]) (CONV: cdecl) [f:ddvctoolscrt_bldself_x86crtsrccrtexe.c @ 377]
0c 002bfd20 775c37eb 7ffdb000 77bf10cb 00000000 kernel32!BaseThreadInitThunk 0xe (FPO: [Non-Fpo])
0d 002bfd60 775c37be 01371082 7ffdb000 00000000 ntdll!__RtlUserThreadStart 0x70 (FPO: [Non-Fpo])
0e 002bfd78 00000000 01371082 7ffdb000 00000000 ntdll!_RtlUserThreadStart 0x1b (FPO: [Non-Fpo])

#  1  Id: 18c0.193c Suspend: 1 Teb: 7ffde000 Unfrozen
 # ChildEBP RetAddr  Args to Child              
00 0133fbac 775ff20f 76a71677 00000000 00000000 ntdll!DbgBreakPoint (FPO: [0,0,0])
01 0133fbdc 758ced6c 00000000 0133fc28 775c37eb ntdll!DbgUiRemoteBreakin 0x3c (FPO: [Non-Fpo])
02 0133fbe8 775c37eb 00000000 76a71183 00000000 kernel32!BaseThreadInitThunk 0xe (FPO: [Non-Fpo])
03 0133fc28 775c37be 775ff1d3 00000000 00000000 ntdll!__RtlUserThreadStart 0x70 (FPO: [Non-Fpo])
04 0133fc40 00000000 775ff1d3 00000000 00000000 ntdll!_RtlUserThreadStart 0x1b (FPO: [Non-Fpo])

通过调用栈，我们发现程序卡在了ShellExecuteExW里。

运行!cs -l看下输出结果：

0:001> !cs -l
-----------------------------------------
DebugInfo          = 0x77637540
Critical section   = 0x77637340 (ntdll!LdrpLoaderLock 0x0)
LOCKED
LockCount          = 0x1
WaiterWoken        = No
OwningThread       = 0x00001400
RecursionCount     = 0x1
LockSemaphore      = 0x64
SpinCount          = 0x00000000

注意OwningThread的值0x00001400 正是我们生成的测试线程，与我们在DbgView里看到的线程id一致。但是该线程已经被我们杀死了，它在被杀死前获得了进程加载锁0x77637340 (ntdll!LdrpLoaderLock 0x0)。

至此，真相大白。

总结

不要随便用TerminateThread来强行杀死线程！
windbg真是windows下的调试神器。
!cs -l可以帮助我们快速的查找到死锁的关键段。

参考资料

《软件调试》
《格蠹汇编》
《windows核心编程（第 5 版）》尤其是第20章
Dynamic-Link Library Best Practices

秒客网