AuroraRuntime/Source/Debug/ExceptionWatcher.Win32.cpp

/***
    Copyright (C) 2021 J Reece Wilson (a/k/a "Reece"). All rights reserved.

    File: ExceptionWatcher.Win32.cpp
    Date: 2021-6-12
    Author: Reece
***/
#include <Source/RuntimeInternal.hpp>
#include "Debug.hpp"
#include "ExceptionWatcher.NT.hpp"
#include "ExceptionWatcher.Win32.hpp"
#include "Stack.Win32.hpp"

#include <Source/Process/AuProcessMap.Win32.hpp>
#include <Source/Telemetry/Telemetry.hpp>

#include <Windows.h>
#include <Dbghelp.h>
#include <codecvt>

#include <vcruntime_exception.h>
#include <ehdata.h>

#include <Source/Process/AuProcessMap.hpp>
#include <Source/IO/FS/FS.hpp>

#include <Source/Grug/AuGrug.hpp>
#include <Source/Exit/AuExit.hpp>

#if 0
    #include <WerApi.h>
#endif

static thread_local int gDebugLocked = 0;

namespace Aurora::Debug
{
    #define EXCEPTION_ENTRY(n) {n, #n}
    static const AuHashMap<DWORD, AuString> kExceptionTable
    {
        EXCEPTION_ENTRY(STILL_ACTIVE),
        EXCEPTION_ENTRY(EXCEPTION_ACCESS_VIOLATION),
        EXCEPTION_ENTRY(EXCEPTION_DATATYPE_MISALIGNMENT),
        EXCEPTION_ENTRY(EXCEPTION_BREAKPOINT),
        EXCEPTION_ENTRY(EXCEPTION_SINGLE_STEP),
        EXCEPTION_ENTRY(EXCEPTION_ARRAY_BOUNDS_EXCEEDED),
        EXCEPTION_ENTRY(EXCEPTION_FLT_DENORMAL_OPERAND),
        EXCEPTION_ENTRY(EXCEPTION_FLT_DIVIDE_BY_ZERO),
        EXCEPTION_ENTRY(EXCEPTION_FLT_INEXACT_RESULT),
        EXCEPTION_ENTRY(EXCEPTION_FLT_INVALID_OPERATION),
        EXCEPTION_ENTRY(EXCEPTION_FLT_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_FLT_STACK_CHECK),
        EXCEPTION_ENTRY(EXCEPTION_FLT_UNDERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_INT_DIVIDE_BY_ZERO),
        EXCEPTION_ENTRY(EXCEPTION_INT_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_PRIV_INSTRUCTION),
        EXCEPTION_ENTRY(EXCEPTION_IN_PAGE_ERROR),
        EXCEPTION_ENTRY(EXCEPTION_ILLEGAL_INSTRUCTION),
        EXCEPTION_ENTRY(EXCEPTION_NONCONTINUABLE_EXCEPTION),
        EXCEPTION_ENTRY(EXCEPTION_STACK_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_INVALID_DISPOSITION),
        EXCEPTION_ENTRY(EXCEPTION_GUARD_PAGE),
        EXCEPTION_ENTRY(EXCEPTION_INVALID_HANDLE),
        //EXCEPTION_ENTRY(EXCEPTION_POSSIBLE_DEADLOCK),
        EXCEPTION_ENTRY(CONTROL_C_EXIT),
        EXCEPTION_ENTRY(DBG_CONTROL_C)
    };
#undef EXCEPTION_ENTRY
#define EXCEPTION_ENTRY(n) {n, true}
    static const AuHashMap<DWORD, bool> kExceptionFatalTable
    {
        EXCEPTION_ENTRY(EXCEPTION_ACCESS_VIOLATION),
        EXCEPTION_ENTRY(EXCEPTION_DATATYPE_MISALIGNMENT),
        EXCEPTION_ENTRY(EXCEPTION_ARRAY_BOUNDS_EXCEEDED),
        EXCEPTION_ENTRY(EXCEPTION_FLT_DENORMAL_OPERAND),
        EXCEPTION_ENTRY(EXCEPTION_FLT_DIVIDE_BY_ZERO),
        EXCEPTION_ENTRY(EXCEPTION_FLT_INEXACT_RESULT),
        EXCEPTION_ENTRY(EXCEPTION_FLT_INVALID_OPERATION),
        EXCEPTION_ENTRY(EXCEPTION_FLT_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_FLT_STACK_CHECK),
        EXCEPTION_ENTRY(EXCEPTION_FLT_UNDERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_INT_DIVIDE_BY_ZERO),
        EXCEPTION_ENTRY(EXCEPTION_INT_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_PRIV_INSTRUCTION),
        EXCEPTION_ENTRY(EXCEPTION_IN_PAGE_ERROR),
        EXCEPTION_ENTRY(EXCEPTION_ILLEGAL_INSTRUCTION),
        EXCEPTION_ENTRY(EXCEPTION_NONCONTINUABLE_EXCEPTION),
        EXCEPTION_ENTRY(EXCEPTION_STACK_OVERFLOW),
        EXCEPTION_ENTRY(EXCEPTION_INVALID_DISPOSITION),
        EXCEPTION_ENTRY(EXCEPTION_GUARD_PAGE)
    };
#undef EXCEPTION_ENTRY
   
    void PlatformHandleFatalEx2(bool fatal, CONTEXT &ctx, bool bNoExit);

    static bool IsReadable(const void *address)
    {
        MEMORY_BASIC_INFORMATION info;

        if (!VirtualQuery(address, &info, sizeof(info)))
        {
            return false;
        }

        if (!info.BaseAddress)
        {
            return false;
        }

        return (info.Protect & (PAGE_READONLY | PAGE_READWRITE)) != 0;
    }

    bool InPanic();

    static LONG CALLBACK HandleVectorException(_EXCEPTION_POINTERS *ExceptionInfo)
    {
        if (ExceptionInfo->ExceptionRecord->ExceptionCode == DBG_CONTROL_C)
        {
            Exit::PostLevel(AuThreads::GetThread(), Exit::ETriggerLevel::eSigTerminate);
            return AuExchange(Exit::gHasCanceled, false) ? EXCEPTION_CONTINUE_EXECUTION : EXCEPTION_CONTINUE_SEARCH;
        }

        // https://www.youtube.com/embed/w6P03sTzSqM?start=2&autoplay=1
        if (ExceptionInfo->ExceptionRecord->ExceptionCode < STATUS_GUARD_PAGE_VIOLATION)
        {
            return EXCEPTION_CONTINUE_SEARCH;
        }

    #if defined(AU_CFG_ID_SHIP)

        // you what?
        if (ExceptionInfo->ExceptionRecord->ExceptionCode == EXCEPTION_BREAKPOINT)
        {
            SysPanic("");
            return EXCEPTION_CONTINUE_EXECUTION;
        }

    #else

        // debugger go brrr
        if (ExceptionInfo->ExceptionRecord->ExceptionCode == EXCEPTION_BREAKPOINT)
        {
            return EXCEPTION_CONTINUE_SEARCH;
        }

    #endif

        // debug builds can go do something stupid
        // QA builds, staging, rel in any form should just give up trying if we're under a panic
    #if !defined(AU_CFG_ID_DEBUG)
        if (InPanic())
        {
            return EXCEPTION_CONTINUE_SEARCH;
        }
    #endif

        // something dumb like this mess w hat i had before.
        // gave up trying 
        auto minimal = gDebugLocked++;
        if (minimal >= 5)
        {
            SysPanic("Nested Exception");
        }

        bool cxxThrow = ExceptionInfo->ExceptionRecord->ExceptionInformation[0] == EH_MAGIC_NUMBER1;
        bool cxxThrowPure = ExceptionInfo->ExceptionRecord->ExceptionInformation[0] == EH_PURE_MAGIC_NUMBER1;

        void *exception {};
        HMODULE handle {};
        ThrowInfo *pThrowInfo {};

        if ((ExceptionInfo->ExceptionRecord->ExceptionCode == EH_EXCEPTION_NUMBER) &&
            (ExceptionInfo->ExceptionRecord->NumberParameters >= 3) &&
            ((cxxThrow) ||
            (cxxThrowPure))
            )
        {
            pThrowInfo = reinterpret_cast<ThrowInfo *>(ExceptionInfo->ExceptionRecord->ExceptionInformation[2]);

            if (pThrowInfo)
            {
                auto attribs = pThrowInfo->attributes;

                if (_EH_RELATIVE_TYPEINFO)
                {
                    handle = reinterpret_cast<HMODULE>(ExceptionInfo->ExceptionRecord->ExceptionInformation[3]);
                }

                exception = reinterpret_cast<void *>(ExceptionInfo->ExceptionRecord->ExceptionInformation[1]);
            }
        }

        bool isCritical = AuExists(kExceptionFatalTable, ExceptionInfo->ExceptionRecord->ExceptionCode);

        auto handleNoCppObject = [&]() -> AuString
        {
            const AuString *msg;
            if (AuTryFind(kExceptionTable, ExceptionInfo->ExceptionRecord->ExceptionCode, msg))
            {
                return *msg;
            }
            else
            {
                return AuToString(ExceptionInfo->ExceptionRecord->ExceptionCode);
            }
        };

        StackTrace backtrace;
        {
            static bool bRunOnce {};
            struct ArrowEx : Grug::Arrow
            {
                AuFunction<void()> callback;
            };
            ArrowEx empty;

            empty.callback = [&]()
            {
                try
                {
                    if (AuExchange(bRunOnce, true))
                    {
                        Telemetry::Mayday();
                    }

                    ParseStack(ExceptionInfo->ContextRecord, backtrace);
                }
                catch (...)
                {

                }
            };

            empty.pCallback =[](Grug::Arrow *pBase) -> void
            {
                auto pEx = (ArrowEx *)pBase;
                pEx->callback();
            };

            Grug::HurlArrow(&empty, empty.pCallback, {});
            Grug::ArrowWait(&empty);
        }


        CONTEXT ctx {};
        ctx.ContextFlags = CONTEXT_ALL;

        if (!GetThreadContext(GetCurrentThread(), &ctx))
        {
            Debug::Panic();
        }

        AuVoidFunc doReportLocal;
        {

        #if defined(AU_CFG_ID_INTERNAL) || defined(AU_CFG_ID_DEBUG)
            const bool kShouldPrintErrors = true;
        #else
            const bool kShouldPrintErrors = false;
        #endif

            doReportLocal = [&]()
            {
                ReportSEH(handle, exception, pThrowInfo, handleNoCppObject, backtrace, [&](const AuString &str)
                {
                    // Pre-submit callback -> its showtime
                    if ((isCritical || kShouldPrintErrors) && (minimal == 0))
                    {
                        if (gRuntimeConfig.debug.bPrintExceptionStackTracesOut)
                        {
                            AuLogWarn("NT Exception: 0x{:x}, {}", ExceptionInfo->ExceptionRecord->ExceptionCode, str);
                            AuLogWarn("{}", StringifyStackTrace(backtrace));
                        }
                    }
                });
            };
        }

        auto pThread = AuThreads::GetThread();

        {
            static bool bRunOnce {};
            struct ArrowEx : Grug::Arrow
            {
                AuFunction<void()> callback;
            };
            ArrowEx empty;

            empty.callback = [&]()
            {
                doReportLocal();

                ReportStackTrace(backtrace, "");

                try
                {
                    if (isCritical ||
                        gRuntimeConfig.debug.bIsExceptionThrowFatal) // exception = literally anything
                    {
                        Exit::PostLevel(pThread, Aurora::Exit::ETriggerLevel::eFatalException);
                        PlatformHandleFatalEx2(true, ctx, true);
                    }
                    else
                    {
                        Exit::PostLevel(pThread, Aurora::Exit::ETriggerLevel::eProblematicEvent);
                        PlatformHandleFatalEx2(false, ctx, true);
                    }
                }
                catch (...)
                {

                }
            };

            empty.pCallback =[](Grug::Arrow *pBase) -> void
            {
                auto pEx = (ArrowEx *)pBase;
                pEx->callback();
            };

            Grug::HurlArrow(&empty, empty.pCallback, {});
            Grug::ArrowWait(&empty);

            ReportStackTrace(backtrace, "");
        }

        gDebugLocked = 0;
        return EXCEPTION_CONTINUE_SEARCH;
    }

    static AuString GetDumpName()
    {
        try
        {
            auto pProcessName = Process::GetProcessName();
            auto tm = Time::ToCivilTime(Time::CurrentClockMS());
            return (pProcessName ? *pProcessName : AuString()) +
                " " +
                AuLocale::TimeDateToFileNameISO8601(tm) + ".dmp";
        }
        catch (...)
        {
            return "errordate.dmp";
        }
    }

#if defined(AU_ENABLE_NATIVE_MINIDUMP)
    void SaveMinidump(_EXCEPTION_POINTERS *ExceptionInfo, bool isFatal)
    {
        bool ok {};
        MINIDUMP_EXCEPTION_INFORMATION info {};
        info.ThreadId = GetCurrentThreadId();

        info.ClientPointers = false;
        info.ExceptionPointers = ExceptionInfo;

        static const DWORD flags = MiniDumpWithFullMemory |
            MiniDumpWithFullMemoryInfo |
            MiniDumpWithHandleData |
            MiniDumpWithUnloadedModules |
            MiniDumpWithThreadInfo;

        std::wstring path;
        AuString utf8Path;
        while (path.empty())
        {
            try
            {
                utf8Path = AuIOFS::NormalizePathRet("./Logs/Crashes/" + GetDumpName());
                path = Locale::ConvertFromUTF8(utf8Path);
            }
            catch (...)
            {

            }
        }

        AuIOFS::CreateDirectories(utf8Path, true);

        auto hFile = Win32Open(path.c_str(), GENERIC_READ | GENERIC_WRITE, 0, false, CREATE_ALWAYS, 0, FILE_ATTRIBUTE_NORMAL);
        if (hFile == INVALID_HANDLE_VALUE)
        {
            AuLogWarn("Couldn't open minidump file. Has a debugger locked the .dmp file?");
            goto miniDumpOut;
        }

        if (!pMiniDumpWriteDump)
        {
            goto miniDumpOut;
        }

        ok = pMiniDumpWriteDump(GetCurrentProcess(), GetCurrentProcessId(), hFile, (MINIDUMP_TYPE)flags, &info, nullptr, nullptr);
        if (!ok)
        {
            AuLogWarn("Couldn't write minidump: {:x}", GetLastError());
            goto miniDumpOut;
        }

        CloseHandle(hFile);

    miniDumpOut:

        try
        {
            AuDebug::PrintError();
        }
        catch (...)
        {
        }

        if (isFatal)
        {
            try
            {
                Grug::GrugFlushFlushs();
                Grug::GrugFlushWrites();
            }
            catch (...)
            {

            }

            Win32Terminate();
        }
    }
#endif 

    void BlackboxReport(_EXCEPTION_POINTERS *ExceptionInfo, bool fatal)
    {
        HANDLE hFile;
        std::wstring path;
        AuString utf8Path;
        bool ok { true };
        MINIDUMP_EXCEPTION_INFORMATION info;

        if (fatal)
        {
            AuDebug::AddMemoryCrunch();
        }

        auto dumpName = GetDumpName();

        info.ClientPointers = false;
        info.ThreadId = GetCurrentThreadId();
        info.ExceptionPointers = ExceptionInfo;

        static const DWORD flags = MiniDumpWithDataSegs |
            MiniDumpWithFullMemoryInfo |
            MiniDumpWithHandleData |
            MiniDumpWithUnloadedModules |
            MiniDumpWithThreadInfo;

        while (path.empty())
        {
            try
            {
                utf8Path = AuIOFS::NormalizePathRet("./Logs/Crashes/" + dumpName);
                path = Locale::ConvertFromUTF8(utf8Path);
            }
            catch (...)
            {

            }
        }

        AuIOFS::CreateDirectories(utf8Path, true); // potentially unsafe / could throw inside

        hFile = Win32Open(path.c_str(), GENERIC_READ | GENERIC_WRITE, 0, false, CREATE_ALWAYS, 0, FILE_ATTRIBUTE_NORMAL);
        if (hFile != INVALID_HANDLE_VALUE)
        {
            AuLogWarn("[1] Couldn't open minidump file. Has a debugger locked the .dmp file?");
            goto miniMiniDumpOut;
        }

        if (!pMiniDumpWriteDump)
        {
            goto miniMiniDumpOut;
        }

        ok = pMiniDumpWriteDump(GetCurrentProcess(), GetCurrentProcessId(), hFile, (MINIDUMP_TYPE)flags, &info, nullptr, nullptr);
        if (!ok)
        {
            AuLogWarn("Couldn't write minidump: {:x}", GetLastError());
            goto miniMiniDumpOut;
        }

        CloseHandle(hFile);

    miniMiniDumpOut:

        if (fatal)
        {
            Telemetry::NewBlackBoxEntryMinidump report {};
            report.includesRx = false;
            report.resource.path = dumpName;
            report.resource.type = Telemetry::ENewBlackBoxResourceType::eLocal;
            Telemetry::ReportDyingBreath(report);

            Win32Terminate();
        }
    }

    static void CacheInternalBuildSymbols()
    {
    #if defined(AU_CFG_ID_INTERNAL) || defined(AU_CFG_ID_DEBUG)
        if (pSymInitialize)
        {
            pSymInitialize(GetCurrentProcess(), NULL, TRUE);
        }
    #endif 
    }

    static void DisableWindowsErrorReporting()
    {
    #if 0
        // Windows has this annoying watchdog that triggers when your main loop doesnt respond after a while
        // It's aggressive in its approach, giving the users a choice to forcefully terminate as soon as they spam click a busy app,
        //  or never at all. latterly, its not uncommon for the app to not come back up, bc win32.
        // It's too easy to trigger the watchdog and impossible to stop it from deciding the windowed application must die
        auto pProcName = Process::GetProcessName();
        if (!pProcName)
        {
            AuLogWarn("Couldn't disable Microsoft Glowware Reporting!");
            return;
        }
     
        // Yes, this is prone to module-name conflict issues because it's a dumb registry utility function
        // This isn't what we want, but it's probably what the user wants
        // MSFT is such a loving company, they'll do almost nothing to ensure they wont steal all your data
        WerAddExcludedApplication(AuLocale::ConvertFromUTF8(pProcName->data()).c_str(), false);
    #else

        // This implementation doesn't require a single IAT entry to a pointless dll
        auto pProcName = Process::GetProcessName();
        if (!pProcName)
        {
            AuLogWarn("Couldn't disable Microsoft Glowware Reporting!");
            return;
        }

        auto procNameWide = AuLocale::ConvertFromUTF8(pProcName->c_str());

        HKEY hKey;
        if (pRegOpenKeyExW &&
            pRegOpenKeyExW(HKEY_CURRENT_USER, L"SOFTWARE\\Microsoft\\Windows\\Windows Error Reporting\\ExcludedApplications", 0, KEY_WRITE, &hKey) == ERROR_SUCCESS)
        {
            DWORD bioluminescenceReductionFactor { 1 };
            if (pRegSetValueExW)
            {
                (void)pRegSetValueExW(hKey, procNameWide.c_str(), 0, REG_DWORD, (const BYTE *)&bioluminescenceReductionFactor, sizeof(DWORD));
            }
            if (pRegCloseKey)
            {
                pRegCloseKey(hKey);
            }
        }

    #endif
    }

    void InitWin32()
    {
        // ...
        if (gRuntimeConfig.debug.bNonshipPrecachesSymbols)
        {
            CacheInternalBuildSymbols();
        }

        // ...
        DisableWindowsErrorReporting();

        // ..
        if (gRuntimeConfig.debug.bEnableWin32RootExceptionHandler)
        {
            AddVectoredExceptionHandler(1, HandleVectorException);
        }
    }
}