0:000> k 25# Child-SP RetAddr Call Site00 00000000`007fc8d8 00007ffd`87439b13 ntdll!NtWaitForAlertByThreadId+0x1401 00000000`007fc8e0 00007ffd`87439a06 ntdll!RtlpWaitOnAddressWithTimeout+0x4302 00000000`007fc910 00007ffd`8743987d ntdll!RtlpWaitOnAddress+0xae03 00000000`007fc980 00007ffd`87435fdc ntdll!RtlpWaitOnCriticalSection+0xd904 00000000`007fc9f0 00007ffd`87435ef0 ntdll!RtlpEnterCriticalSectionContended+0xdc05 00000000`007fca20 00007ffd`536839ea ntdll!RtlEnterCriticalSection+0x4006 00000000`007fca50 00007ffd`5368470a AcLayers!NS_VirtualRegistry::CRegLock::CRegLock+0x1a07 00000000`007fca90 00007ffd`536726d2 AcLayers!NS_VirtualRegistry::APIHook_RegOpenKeyExW+0x2a08 00000000`007fcb10 00007ffd`778e550b AcLayers!NS_WRPMitigation::APIHook_RegOpenKeyExW+0x4209 00000000`007fcb60 00007ffd`778e5437 xxx!GetCodePageForFont+0xa70a 00000000`007fcc90 00007ffd`778e5296 xxx!CToolTipsMgr::NewFont+0x1130b 00000000`007fcda0 00007ffd`778e18f9 xxx!CToolTipsMgr::LoadTheme+0xb20c 00000000`007fcdd0 00007ffd`84b9ca66 xxx!CToolTipsMgr::s_ToolTipsWndProc+0x1b90d 00000000`007fce10 00007ffd`84b9c34b user32!UserCallWinProcCheckWow+0x2660e 00000000`007fcf90 00007ffd`4f36b1cc user32!CallWindowProcW+0x8b0f 00000000`007fcfe0 00007ffd`4f39ccac System_Windows_Forms_ni!System.Windows.Forms.NativeWindow.DefWndProc+0x9c10 00000000`007fd090 00007ffd`4f39cc05 System_Windows_Forms_ni!System.Windows.Forms.ToolTip.WndProc+0x9c11 00000000`007fd260 00007ffd`4f36a3a3 System_Windows_Forms_ni!System.Windows.Forms.ToolTip.ToolTipNativeWindow.WndProc+0x1512 00000000`007fd290 00007ffd`4f9e1161 System_Windows_Forms_ni!System.Windows.Forms.NativeWindow.Callback+0xc313 00000000`007fd330 00007ffd`52c8222e System_Windows_Forms_ni+0x8d116114 00000000`007fd3a0 00007ffd`84b9ca66 clr!UMThunkStub+0x6e15 00000000`007fd430 00007ffd`84b9c78c user32!UserCallWinProcCheckWow+0x26616 00000000`007fd5b0 00007ffd`84bb3b32 user32!DispatchClientMessage+0x9c17 00000000`007fd610 00007ffd`874c22c4 user32!__fnINLPCREATESTRUCT+0xa218 00000000`007fd670 00007ffd`836a1f24 ntdll!KiUserCallbackDispatcherContinue19 00000000`007fd7e8 00007ffd`84ba15df win32u!NtUserCreateWindowEx+0x141a 00000000`007fd7f0 00007ffd`84ba11d4 user32!VerNtUserCreateWindowEx+0x20f1b 00000000`007fdb80 00007ffd`84ba1012 user32!CreateWindowInternal+0x1b41c 00000000`007fdce0 00007ffd`4f3e8098 user32!CreateWindowExW+0x821d 00000000`007fdd70 00007ffd`4f3696f0 System_Windows_Forms_ni+0x2d8098...
从卦象看,很明显主线程卡在 NtWaitForAlertByThreadId 上,这是有问题的,接下来我们仔细解读下线程栈垫片
,详情可以看一下《软件调试》,它主要用来处理一些系统级兼容性的问题,然后可以看到它在查询注册表时有一个lock操作在非托管代码中,lock 一般都用 临界区(CriticalSection) 实现,那到底它等待的临界区被谁持有着呢?2. 谁持有着临界区锁要想获取锁的持有信息,可以使用 !cs -l
或者 !locks
,但这里要提醒一下,在真实的dump分析过程中,有时候不准,所以更好的办法就是从线程栈上提取,那怎么提取呢? 其实就是寻找 ntdll!RtlEnterCriticalSection
方法的第一个参数即可,方法签名如下:VOID RtlEnterCriticalSection( PRTL_CRITICAL_SECTION CriticalSection);
接下来反汇编下 00007ffd536839ea
处的代码,看看 rcx 寄存器是怎么传下来的0:000> ub 00007ffd`536839eaAcLayers!NS_VirtualRegistry::OPENKEY::AddEnumEntries<NS_VirtualRegistry::VIRTUALVAL>+0x11a:00007ffd`536839ce cc int 300007ffd`536839cf cc int 3AcLayers!NS_VirtualRegistry::CRegLock::CRegLock:00007ffd`536839d0 48895c2408 mov qword ptr [rsp+8],rbx00007ffd`536839d5 57 push rdi00007ffd`536839d6 4883ec30 sub rsp,30h00007ffd`536839da 488bf9 mov rdi,rcx00007ffd`536839dd 488d0d4c7f0300 lea rcx,[AcLayers!NS_VirtualRegistry::csRegCriticalSection (00007ffd`536bb930)]00007ffd`536839e4 ff15ae660100 call qword ptr [AcLayers!_imp_EnterCriticalSection (00007ffd`5369a098)]
从卦象上看,很吉利,这个 rcx 原来是一个全局变量 AcLayers!NS_VirtualRegistry::csRegCriticalSection
, 接下来用 !cs 观察下到底被谁持有着0:000> !cs AcLayers!NS_VirtualRegistry::csRegCriticalSection-----------------------------------------Critical section = 0x00007ffd536bb930 (AcLayers!NS_VirtualRegistry::csRegCriticalSection+0x0)DebugInfo = 0x000000001c4e58e0LOCKEDLockCount = 0x2WaiterWoken = NoOwningThread = 0x0000000000001d20RecursionCount = 0x1LockSemaphore = 0xFFFFFFFFSpinCount = 0x00000000020007ce
这又是一副吉卦,可以看到当前持有线程是 1d20
,那这个线程正在做什么呢?3. 1d20 线程为什么持锁不释放案情往前推进了一步,我们切过去观察下这个线程栈0:000> ~~[1d20]sntdll!NtDelayExecution+0x14:00007ffd`874bec14 c3 ret0:028> kL# Child-SP RetAddr Call Site00 00000000`33ccd948 00007ffd`83955381 ntdll!NtDelayExecution+0x1401 00000000`33ccd950 00007ffd`6d4a2361 KERNELBASE!SleepEx+0xa102 00000000`33ccd9f0 00007ffd`8520a75c perfts!CloseLagPerfData+0x2103 00000000`33ccda30 00007ffd`85209ccd advapi32!CloseExtObjectLibrary+0xec04 00000000`33ccda90 00007ffd`8396dc6a advapi32!PerfRegCloseKey+0x15d05 00000000`33ccdae0 00007ffd`839715e6 KERNELBASE!BaseRegCloseKeyInternal+0x7206 00000000`33ccdb10 00007ffd`83935209 KERNELBASE!ClosePredefinedHandle+0x9607 00000000`33ccdb40 00007ffd`53685d71 KERNELBASE!RegCloseKey+0x14908 00000000`33ccdba0 00007ffd`53683ae5 AcLayers!NS_VirtualRegistry::CVirtualRegistry::CloseKey+0xbd09 00000000`33ccdbf0 00007ffd`51c7737e AcLayers!NS_VirtualRegistry::APIHook_RegCloseKey+0x250a 00000000`33ccdc30 00007ffd`51bf4be2 mscorlib_ni+0x58737e0b 00000000`33ccdce0 00007ffd`513c356a mscorlib_ni!Microsoft.Win32.RegistryKey.Dispose+0x720c 00000000`33ccdd20 00007ffd`513c34b9 System_ni!System.Diagnostics.PerformanceCounterLib.GetStringTable+0x41a...13 00000000`33cce050 00007ffd`513bfe3c System_ni!System.Diagnostics.PerformanceCounter..ctor+0xd714 00000000`33cce0a0 00007ffc`f45cb2ce System_ni!System.Diagnostics.PerformanceCounter..ctor+0x1c15 00000000`33cce0d0 00007ffc`f45cb14c 0x00007ffc`f45cb2ce16 00000000`33cce120 00007ffc`f45cb023 0x00007ffc`f45cb14c...
从卦中看,这个线程貌似在用 CloseLagPerfData
方法关闭一些东西时一直在Sleep等待,可以反汇编 00007ffd6d4a2361
处代码看看等待多久0:028> ub 00007ffd`6d4a2361perfts!CloseLagPerfData+0x5:00007ffd`6d4a2345 55 push rbp00007ffd`6d4a2346 488bec mov rbp,rsp00007ffd`6d4a2349 4883ec30 sub rsp,30h00007ffd`6d4a234d e8720e0000 call perfts!LagCounterManager::Cleanup (00007ffd`6d4a31c4)00007ffd`6d4a2352 33db xor ebx,ebx00007ffd`6d4a2354 eb0b jmp perfts!CloseLagPerfData+0x21 (00007ffd`6d4a2361)00007ffd`6d4a2356 b964000000 mov ecx,64h00007ffd`6d4a235b ff15c74e0000 call qword ptr [perfts!_imp_Sleep (00007ffd`6d4a7228)]...
从卦中的 mov ecx,64h
可以看到是 Sleep(100) 毫秒,更多细节也没空继续追究了,但不管怎么样,它是由上层的计数器类 PerformanceCounter
引发的,这里学一下 4S 店的做法,让朋友能不能不要调用 PerformanceCounter
这个类,咱躲开他就可以了,截图如下:去掉之后,朋友反馈问题消失三:总结 说来也奇怪,最近发现了二起由 PerformanceCounter
引发的程序卡死,把经验留在这里,希望后来人少踩坑吧(图片来源网络,侵删)
0 评论