如何用 WinDbg 调试Linux上的 .NET程序

news2024/11/24 18:32:40

一：背景

1. 讲故事

最新版本 1.2402.24001.0 的WinDbg真的让人很兴奋，可以将自己伪装成 GDB 来和远程的 GDBServer 打通来实现对 Linux 上 .NET程序进行调试，这样就可以继续使用熟悉的WinDbg 命令，在这个版本中我觉得 WinDbg 不再是 WinDbg，而是 XDbg 了，画个简图如下：

简图有了，接下来就要付出实践了。

二：实操 Linux 上 .NET调试

1. 测试程序

本想在 CentOS7 上安装 .NET8，不大好装，这里就用一个现存的 .NETCore 3.1 吧，测试代码如下：


    internal class Program
    {
        static void Main(string[] args)
        {
            while (true)
            {
                Console.WriteLine($"{DateTime.Now},tid={Thread.CurrentThread.ManagedThreadId}");
                Thread.Sleep(1000);
            }
        }
    }

代码非常简单，就是1s输出一条记录，接下来编译成x64部署到 Centos7 上。


[root@localhost data]# ls
ConsoleApp7  ConsoleApp7.deps.json  ConsoleApp7.dll  ConsoleApp7.pdb  ConsoleApp7.runtimeconfig.json

2. 安装GDBServer

在 linux 上安装 gdbserver 比较简单，使用 yum 安装即可 yum install gdb-gdbserver ，输出如下：


[root@localhost data]# yum install gdb-gdbserver
Loaded plugins: fastestmirror, langpacks
Loading mirror speeds from cached hostfile
 * base: mirror-hk.koddos.net
 * centos-sclo-rh: ftp.sjtu.edu.cn
 * centos-sclo-sclo: ftp.sjtu.edu.cn
 * epel: mirror.hoster.kz
 * extras: ftp.sjtu.edu.cn
 * updates: mirror-hk.koddos.net
Package gdb-gdbserver-7.6.1-120.el7.x86_64 already installed and latest version
Nothing to do
[root@localhost data]# gdbserver –version
Usage:	gdbserver [OPTIONS] COMM PROG [ARGS ...]
	gdbserver [OPTIONS] --attach COMM PID
	gdbserver [OPTIONS] --multi COMM

COMM may either be a tty device (for serial debugging), or 
HOST:PORT to listen for a TCP connection.

Options:
  --debug               Enable general debugging output.
  --remote-debug        Enable remote protocol debugging output.
  --version             Display version information and exit.
  --wrapper WRAPPER --  Run WRAPPER to start new programs.
  --once                Exit after the first connection has closed.

安装好之后，接下来用 gdbserver 来启动我们的程序，并启动调试端口为 1234，参考如下：


[root@localhost data]# gdbserver 192.168.128.130:1234 dotnet ConsoleApp7.dll
Process dotnet created; pid = 3643
Listening on port 1234

3. 使用 windbg 连接

打开Windbg后，选择 Connect to remote debugger 选项, 在连接字符串中填入 gdb:server=192.168.128.130,port=1234 即可，截图如下：

连接好之后，会有一个初始中断，直接输入g就好了，输出如下：


64-bit machine not using 64-bit API

************* Path validation summary **************
Response                         Time (ms)     Location
Deferred                                       SRV*C:\mysymbols*https://msdl.microsoft.com/download/symbols
Symbol search path is: SRV*C:\mysymbols*https://msdl.microsoft.com/download/symbols
Executable search path is: 
Unknown System Version 0 UP Free x64
System Uptime: not available
Process Uptime: not available
Reloading current modules
ModLoad: 00005555`55554000 00005555`555770cd   /usr/share/dotnet/dotnet
ModLoad: 00007fff`f7bbf000 00007fff`f7dda488   /lib64/libpthread.so.0
ModLoad: 00007fff`f79bb000 00007fff`f7bbe130   /lib64/libdl.so.2
ModLoad: 00007fff`f76b3000 00007fff`f79ba420   /lib64/libstdc++.so.6
ModLoad: 00007fff`f73b1000 00007fff`f76b2138   /lib64/libm.so.6
ModLoad: 00007fff`f719b000 00007fff`f73b0400   /lib64/libgcc_s.so.1
ModLoad: 00007fff`f6dcd000 00007fff`f719a200   /lib64/libc.so.6
ModLoad: 00007fff`f7ddb000 00007fff`f7ffe150   /lib64/ld-linux-x86-64.so.2
ModLoad: 00007fff`f7f72000 00007fff`f7fda288   /usr/share/dotnet/host/fxr/6.0.26/libhostfxr.so
ModLoad: 00007fff`f6b7c000 00007fff`f6dcc3b0   /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/libhostpolicy.so
ModLoad: 00007fff`f63e7000 00007fff`f6b7bac8   /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/libcoreclr.so
ModLoad: 00007fff`f61df000 00007fff`f63e6c38   /lib64/librt.so.1
ModLoad: 00007fff`f57d2000 00007fff`f59dd8c0   /lib64/libnuma.so.1
ModLoad: 00007fff`f3142000 00007fff`f3413dac   /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/libclrjit.so
ModLoad: 00007fff`f2f31000 00007fff`f3141468   /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/System.Native.so
ModLoad: 00007fff`f2d26000 00007fff`f2f30488   /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/System.Globalization.Native.so
ModLoad: 00007fff`f29ad000 00007fff`f2d25fe0   /lib64/libicuuc.so.50
ModLoad: 00007fff`f13da000 00007fff`f29ac030   /lib64/libicudata.so.50
ModLoad: 00007fff`f0fdb000 00007fff`f13d9340   /lib64/libicui18n.so.50
...................
ReadVirtual() failed in GetXStateConfiguration() first read attempt (error == 0.)
Unable to load image /lib64/libpthread.so.0, Win32 error 0n2
*** WARNING: Unable to verify timestamp for /lib64/libpthread.so.0
Unable to load image /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/libcoreclr.so, Win32 error 0n2
*** WARNING: Unable to verify timestamp for /usr/share/dotnet/shared/Microsoft.NETCore.App/3.1.32/libcoreclr.so
libpthread_so!_pthread_cond_timedwait+0x132:
00007fff`f7bcade2 4989c6          mov     r14,rax
0:000> g

有些人可能会好奇，为什么 WinDbg 能伪装成 GDB 来和 GDBServer 来通讯，这其实得益于 WinDbg 是一个宿主，它可以被很多外来的插件无线扩容自己的功能，这和 Linux 的分而治之恰恰相反。。。

接下来可以用 .chain 命令观察插件列表，其中的 GDBServerComposition 和 ELFBinComposition 让这项功能得到实现。


0:000> .chain
Extension DLL chain:
    GDBServerComposition: image 10.0.27553.1004, API 0.0.0, 
        [path: C:\Program Files\WindowsApps\Microsoft.WinDbg_1.2402.24001.0_x64__8wekyb3d8bbwe\amd64\winext\GDBServerComposition.dll]
    ELFBinComposition: image 10.0.27553.1004, API 0.0.0, 
        [path: C:\Program Files\WindowsApps\Microsoft.WinDbg_1.2402.24001.0_x64__8wekyb3d8bbwe\amd64\winext\ELFBinComposition.dll]
    dbghelp: image 10.0.27553.1004, API 10.0.6, 
        [path: C:\Program Files\WindowsApps\Microsoft.WinDbg_1.2402.24001.0_x64__8wekyb3d8bbwe\amd64\dbghelp.dll]
    uext: image 10.0.27553.1004, API 1.0.0, 
        [path: C:\Program Files\WindowsApps\Microsoft.WinDbg_1.2402.24001.0_x64__8wekyb3d8bbwe\amd64\winext\uext.dll]

接下来就可以做验证了，研究 coreclr 源码，你会发现在 Linux 上 .NET 的 Sleep 函数是借助于底层的 pthread_cond_timedwait 函数，Linux并没有提供类似Windows 的SleepEx这样的系统调用，这就比较坑了，参考如下：


PAL_ERROR CPalSynchronizationManager::ThreadNativeWait(
    ThreadNativeWaitData* ptnwdNativeWaitData,
    DWORD dwTimeout,
    ThreadWakeupReason* ptwrWakeupReason,
    DWORD* pdwSignaledObject)
{
	//...
    while (FALSE == ptnwdNativeWaitData->iPred)
    {
        if (INFINITE == dwTimeout)
        {
            iWaitRet = pthread_cond_wait(&ptnwdNativeWaitData->cond,
                &ptnwdNativeWaitData->mutex);
        }
        else
        {
            iWaitRet = pthread_cond_timedwait(&ptnwdNativeWaitData->cond,
                &ptnwdNativeWaitData->mutex,
                &tsAbsTmo);
        }
    }
	//...
}