记录程序崩溃时的调用堆栈

在程序release之后，不可避免的会存在一些bug，测试人员和最终用户如何在发现bug之后指导开发人员进行更正呢？在MS的网站上，有一篇名为"Under the hook"的文章，讲述了如何把程序崩溃时的函数调用情况记录为日志的方法，对此感兴趣的读者可以去看一看原文,那里提供源代码和原理的说明。

文章的作者提供了一个MSJExceptionHandler类来实现这一功能，这个类的使用方法很简单，只要把这个类加入到你的工程中并和你的程序一起编译就可以了，由于在这个类的实现文件中把自己定义为一个全局的类对象，所以，不用加入任何代码，#include都不需要。

当程序崩溃时，MSJExceptionHandler就会把崩溃时的堆栈调用情况记录在一个.rpt文件中，软件的测试人员或最终用户只要把这个文件发给你，而你使用记事本打开这个文件就可以查看崩溃原因了。你需要在发行软件的时候，为你的程序生成一个或几个map文件，用于定位出错的文件和函数。(我的另一篇blog中有关于生成map文件和定位错误的详细说明)为了方便使用，这里附上该类的完整代码：

// msjexhnd.h

#ifndef __MSJEXHND_H__
#define __MSJEXHND_H__

class MSJExceptionHandler
{
    public:
    
    MSJExceptionHandler( );
    ~MSJExceptionHandler( );
    
    void SetLogFileName( PTSTR pszLogFileName );

    private:

    // entry point where control comes on an unhandled exception
    static LONG WINAPI MSJUnhandledExceptionFilter(
                                PEXCEPTION_POINTERS pExceptionInfo );

    // where report info is extracted and generated 
    static void GenerateExceptionReport( PEXCEPTION_POINTERS pExceptionInfo );

    // Helper functions
    static LPTSTR GetExceptionString( DWORD dwCode );
    static BOOL GetLogicalAddress(  PVOID addr, PTSTR szModule, DWORD len,
                                    DWORD& section, DWORD& offset );
    static void IntelStackWalk( PCONTEXT pContext );
    static int __cdecl _tprintf(const TCHAR * format, ...);

    // Variables used by the class
    static TCHAR m_szLogFileName[MAX_PATH];
    static LPTOP_LEVEL_EXCEPTION_FILTER m_previousFilter;
    static HANDLE m_hReportFile;
};

extern MSJExceptionHandler g_MSJExceptionHandler;   //  global instance of class

#endif

// msjexhnd.cpp

//==========================================
// Matt Pietrek
// Microsoft Systems Journal, April 1997
// FILE: MSJEXHND.CPP
//==========================================
#include <windows.h>
#include <tchar.h>
#include "msjexhnd.h"

//============================== Global Variables =============================

//
// Declare the static variables of the MSJExceptionHandler class
//
TCHAR MSJExceptionHandler::m_szLogFileName[MAX_PATH];
LPTOP_LEVEL_EXCEPTION_FILTER MSJExceptionHandler::m_previousFilter;
HANDLE MSJExceptionHandler::m_hReportFile;

MSJExceptionHandler g_MSJExceptionHandler;  // Declare global instance of class

//============================== Class Methods =============================

//=============
// Constructor 
//=============
MSJExceptionHandler::MSJExceptionHandler( )
{
    // Install the unhandled exception filter function
    m_previousFilter = SetUnhandledExceptionFilter(MSJUnhandledExceptionFilter);

    // Figure out what the report file will be named, and store it away
    GetModuleFileName( 0, m_szLogFileName, MAX_PATH );

    // Look for the '.' before the "EXE" extension.  Replace the extension
    // with "RPT"
    PTSTR pszDot = _tcsrchr( m_szLogFileName, _T('.') );
    if ( pszDot )
    {
        pszDot++;   // Advance past the '.'
        if ( _tcslen(pszDot) >= 3 )
            _tcscpy( pszDot, _T("RPT") );   // "RPT" -> "Report"
    }
}

//============
// Destructor 
//============
MSJExceptionHandler::~MSJExceptionHandler( )
{
    SetUnhandledExceptionFilter( m_previousFilter );
}

//==============================================================
// Lets user change the name of the report file to be generated 
//==============================================================
void MSJExceptionHandler::SetLogFileName( PTSTR pszLogFileName )
{
    _tcscpy( m_szLogFileName, pszLogFileName );
}

//===========================================================
// Entry point where control comes on an unhandled exception 
//===========================================================
LONG WINAPI MSJExceptionHandler::MSJUnhandledExceptionFilter(
                                    PEXCEPTION_POINTERS pExceptionInfo )
{
    m_hReportFile = CreateFile( m_szLogFileName,
                                GENERIC_WRITE,
                                0,
                                0,
                                OPEN_ALWAYS,
                                FILE_FLAG_WRITE_THROUGH,
                                0 );

    if ( m_hReportFile )
    {
        SetFilePointer( m_hReportFile, 0, 0, FILE_END );

        GenerateExceptionReport( pExceptionInfo );

        CloseHandle( m_hReportFile );
        m_hReportFile = 0;
    }

    if ( m_previousFilter )
        return m_previousFilter( pExceptionInfo );
    else
        return EXCEPTION_CONTINUE_SEARCH;
}

//===========================================================================
// Open the report file, and write the desired information to it.  Called by 
// MSJUnhandledExceptionFilter                                               
//===========================================================================
void MSJExceptionHandler::GenerateExceptionReport(
    PEXCEPTION_POINTERS pExceptionInfo )
{
    // Start out with a banner
    _tprintf( _T("//=====================================================\n") );

    PEXCEPTION_RECORD pExceptionRecord = pExceptionInfo->ExceptionRecord;

    // First print information about the type of fault
    _tprintf(   _T("Exception code: %08X %s\n"),
                pExceptionRecord->ExceptionCode,
                GetExceptionString(pExceptionRecord->ExceptionCode) );

    // Now print information about where the fault occured
    TCHAR szFaultingModule[MAX_PATH];
    DWORD section, offset;
    GetLogicalAddress(  pExceptionRecord->ExceptionAddress,
                        szFaultingModule,
                        sizeof( szFaultingModule ),
                        section, offset );

    _tprintf( _T("Fault address:  %08X %02X:%08X %s\n"),
                pExceptionRecord->ExceptionAddress,
                section, offset, szFaultingModule );

    PCONTEXT pCtx = pExceptionInfo->ContextRecord;

    // Show the registers
    #ifdef _M_IX86  // Intel Only!
    _tprintf( _T("\nRegisters:\n") );

    _tprintf(_T("EAX:%08X\nEBX:%08X\nECX:%08X\nEDX:%08X\nESI:%08X\nEDI:%08X\n"),
            pCtx->Eax, pCtx->Ebx, pCtx->Ecx, pCtx->Edx, pCtx->Esi, pCtx->Edi );

    _tprintf( _T("CS:EIP:%04X:%08X\n"), pCtx->SegCs, pCtx->Eip );
    _tprintf( _T("SS:ESP:%04X:%08X  EBP:%08X\n"),
                pCtx->SegSs, pCtx->Esp, pCtx->Ebp );
    _tprintf( _T("DS:%04X  ES:%04X  FS:%04X  GS:%04X\n"),
                pCtx->SegDs, pCtx->SegEs, pCtx->SegFs, pCtx->SegGs );
    _tprintf( _T("Flags:%08X\n"), pCtx->EFlags );

    // Walk the stack using x86 specific code
    IntelStackWalk( pCtx );

    #endif

    _tprintf( _T("\n") );
}

//======================================================================
// Given an exception code, returns a pointer to a static string with a 
// description of the exception                                         
//======================================================================
LPTSTR MSJExceptionHandler::GetExceptionString( DWORD dwCode )
{
    #define EXCEPTION( x ) case EXCEPTION_##x: return _T(#x);

    switch ( dwCode )
    {
        EXCEPTION( ACCESS_VIOLATION )
        EXCEPTION( DATATYPE_MISALIGNMENT )
        EXCEPTION( BREAKPOINT )
        EXCEPTION( SINGLE_STEP )
        EXCEPTION( ARRAY_BOUNDS_EXCEEDED )
        EXCEPTION( FLT_DENORMAL_OPERAND )
        EXCEPTION( FLT_DIVIDE_BY_ZERO )
        EXCEPTION( FLT_INEXACT_RESULT )
        EXCEPTION( FLT_INVALID_OPERATION )
        EXCEPTION( FLT_OVERFLOW )
        EXCEPTION( FLT_STACK_CHECK )
        EXCEPTION( FLT_UNDERFLOW )
        EXCEPTION( INT_DIVIDE_BY_ZERO )
        EXCEPTION( INT_OVERFLOW )
        EXCEPTION( PRIV_INSTRUCTION )
        EXCEPTION( IN_PAGE_ERROR )
        EXCEPTION( ILLEGAL_INSTRUCTION )
        EXCEPTION( NONCONTINUABLE_EXCEPTION )
        EXCEPTION( STACK_OVERFLOW )
        EXCEPTION( INVALID_DISPOSITION )
        EXCEPTION( GUARD_PAGE )
        EXCEPTION( INVALID_HANDLE )
    }

    // If not one of the "known" exceptions, try to get the string
    // from NTDLL.DLL's message table.

    static TCHAR szBuffer[512] = { 0 };

    FormatMessage(  FORMAT_MESSAGE_IGNORE_INSERTS | FORMAT_MESSAGE_FROM_HMODULE,
                    GetModuleHandle( _T("NTDLL.DLL") ),
                    dwCode, 0, szBuffer, sizeof( szBuffer ), 0 );

    return szBuffer;
}

//==============================================================================
// Given a linear address, locates the module, section, and offset containing  
// that address.                                                               
//                                                                             
// Note: the szModule paramater buffer is an output buffer of length specified 
// by the len parameter (in characters!)                                       
//==============================================================================
BOOL MSJExceptionHandler::GetLogicalAddress(
        PVOID addr, PTSTR szModule, DWORD len, DWORD& section, DWORD& offset )
{
    MEMORY_BASIC_INFORMATION mbi;

    if ( !VirtualQuery( addr, &mbi, sizeof(mbi) ) )
        return FALSE;

    DWORD hMod = (DWORD)mbi.AllocationBase;

    if ( !GetModuleFileName( (HMODULE)hMod, szModule, len ) )
        return FALSE;

    // Point to the DOS header in memory
    PIMAGE_DOS_HEADER pDosHdr = (PIMAGE_DOS_HEADER)hMod;

    // From the DOS header, find the NT (PE) header
    PIMAGE_NT_HEADERS pNtHdr = (PIMAGE_NT_HEADERS)(hMod + pDosHdr->e_lfanew);

    PIMAGE_SECTION_HEADER pSection = IMAGE_FIRST_SECTION( pNtHdr );

    DWORD rva = (DWORD)addr - hMod; // RVA is offset from module load address

    // Iterate through the section table, looking for the one that encompasses
    // the linear address.
    for (   unsigned i = 0;
            i < pNtHdr->FileHeader.NumberOfSections;
            i++, pSection++ )
    {
        DWORD sectionStart = pSection->VirtualAddress;
        DWORD sectionEnd = sectionStart
                    + max(pSection->SizeOfRawData, pSection->Misc.VirtualSize);

        // Is the address in this section???
        if ( (rva >= sectionStart) && (rva <= sectionEnd) )
        {
            // Yes, address is in the section.  Calculate section and offset,
            // and store in the "section" & "offset" params, which were
            // passed by reference.
            section = i+1;
            offset = rva - sectionStart;
            return TRUE;
        }
    }

    return FALSE;   // Should never get here!
}

//============================================================
// Walks the stack, and writes the results to the report file 
//============================================================
void MSJExceptionHandler::IntelStackWalk( PCONTEXT pContext )
{
    _tprintf( _T("\nCall stack:\n") );

    _tprintf( _T("Address   Frame     Logical addr  Module\n") );

    DWORD pc = pContext->Eip;
    PDWORD pFrame, pPrevFrame;
    
    pFrame = (PDWORD)pContext->Ebp;

    do
    {
        TCHAR szModule[MAX_PATH] = _T("");
        DWORD section = 0, offset = 0;

        GetLogicalAddress((PVOID)pc, szModule,sizeof(szModule),section,offset );

        _tprintf( _T("%08X  %08X  %04X:%08X %s\n"),
                    pc, pFrame, section, offset, szModule );

        pc = pFrame[1];

        pPrevFrame = pFrame;

        pFrame = (PDWORD)pFrame[0]; // precede to next higher frame on stack

        if ( (DWORD)pFrame & 3 )    // Frame pointer must be aligned on a
            break;                  // DWORD boundary.  Bail if not so.

        if ( pFrame <= pPrevFrame )
            break;

        // Can two DWORDs be read from the supposed frame address?          
        if ( IsBadWritePtr(pFrame, sizeof(PVOID)*2) )
            break;

    } while ( 1 );
}

//============================================================================
// Helper function that writes to the report file, and allows the user to use 
// printf style formating                                                     
//============================================================================
int __cdecl MSJExceptionHandler::_tprintf(const TCHAR * format, ...)
{
    TCHAR szBuff[1024];
    int retValue;
    DWORD cbWritten;
    va_list argptr;
          
    va_start( argptr, format );
    retValue = wvsprintf( szBuff, format, argptr );
    va_end( argptr );

    WriteFile( m_hReportFile, szBuff, retValue * sizeof(TCHAR), &cbWritten, 0 );

    return retValue;
}