Merge pull request #2490 from goaaats/feat/catch_clr_errors

Catch CLR exceptions

Dalamud.Boot/crashhandler_shared.h

@@ -6,6 +6,8 @@
 #define WIN32_LEAN_AND_MEAN
 #include <Windows.h>
 
+#define CUSTOM_EXCEPTION_EXTERNAL_EVENT 0x12345679
+
 struct exception_info
 {
     LPEXCEPTION_POINTERS pExceptionPointers;

Dalamud.Boot/dllmain.cpp

@@ -331,6 +331,51 @@ HRESULT WINAPI InitializeImpl(LPVOID lpParam, HANDLE hMainThreadContinue) {
         logging::I("VEH was disabled manually");
     }
 
+    // ============================== CLR Reporting =================================== //
+
+    // This is pretty horrible - CLR just doesn't provide a way for us to handle these events, and the API for it
+    // was pushed back to .NET 11, so we have to hook ReportEventW and catch them ourselves for now.
+    // Ideally all of this will go away once they get to it.
+    static std::shared_ptr<hooks::global_import_hook<decltype(ReportEventW)>> s_report_event_hook;
+    s_report_event_hook = std::make_shared<hooks::global_import_hook<decltype(ReportEventW)>>(
+        "advapi32.dll!ReportEventW (global import, hook_clr_report_event)", L"advapi32.dll", "ReportEventW");
+    s_report_event_hook->set_detour([hook = s_report_event_hook.get()](
+        HANDLE hEventLog,
+        WORD wType,
+        WORD wCategory,
+        DWORD dwEventID,
+        PSID lpUserSid,
+        WORD wNumStrings,
+        DWORD dwDataSize,
+        LPCWSTR* lpStrings,
+        LPVOID lpRawData)-> BOOL {
+
+        // Check for CLR Error Event IDs
+        // https://github.com/dotnet/runtime/blob/v10.0.0/src/coreclr/vm/eventreporter.cpp#L370
+        if (dwEventID != 1026 && // ERT_UnhandledException: The process was terminated due to an unhandled exception
+            dwEventID != 1025 && // ERT_ManagedFailFast: The application requested process termination through System.Environment.FailFast
+            dwEventID != 1023 && // ERT_UnmanagedFailFast: The process was terminated due to an internal error in the .NET Runtime
+            dwEventID != 1027 && // ERT_StackOverflow: The process was terminated due to a stack overflow
+            dwEventID != 1028)   // ERT_CodeContractFailed: The application encountered a bug.  A managed code contract (precondition, postcondition, object invariant, or assert) failed
+        {
+            return hook->call_original(hEventLog, wType, wCategory, dwEventID, lpUserSid, wNumStrings, dwDataSize, lpStrings, lpRawData);
+        }
+
+        if (wNumStrings == 0 || lpStrings == nullptr) {
+            logging::W("ReportEventW called with no strings.");
+            return hook->call_original(hEventLog, wType, wCategory, dwEventID, lpUserSid, wNumStrings, dwDataSize, lpStrings, lpRawData);
+        }
+
+        // In most cases, DalamudCrashHandler will kill us now, so call original here to make sure we still write to the event log.
+        const BOOL original_ret = hook->call_original(hEventLog, wType, wCategory, dwEventID, lpUserSid, wNumStrings, dwDataSize, lpStrings, lpRawData);
+
+        const std::wstring error_details(lpStrings[0]);
+        veh::raise_external_event(error_details);
+
+        return original_ret;
+    });
+    logging::I("ReportEventW hook installed.");
+
     // ============================== Dalamud ==================================== //
 
     if (static_cast<int>(g_startInfo.BootWaitMessageBox) & static_cast<int>(DalamudStartInfo::WaitMessageboxFlags::BeforeDalamudEntrypoint))

Dalamud.Boot/veh.cpp

@@ -31,6 +31,8 @@ HANDLE g_crashhandler_process = nullptr;
 HANDLE g_crashhandler_event = nullptr;
 HANDLE g_crashhandler_pipe_write = nullptr;
 
+wchar_t g_external_event_info[16384] = L"";
+
 std::recursive_mutex g_exception_handler_mutex;
 
 std::chrono::time_point<std::chrono::system_clock> g_time_start;
@@ -190,7 +192,11 @@ LONG exception_handler(EXCEPTION_POINTERS* ex)
     DuplicateHandle(GetCurrentProcess(), g_crashhandler_event, g_crashhandler_process, &exinfo.hEventHandle, 0, TRUE, DUPLICATE_SAME_ACCESS);
 
     std::wstring stackTrace;
-    if (!g_clr)
+    if (ex->ExceptionRecord->ExceptionCode == CUSTOM_EXCEPTION_EXTERNAL_EVENT)
+    {
+        stackTrace = std::wstring(g_external_event_info);
+    }
+    else if (!g_clr)
     {
         stackTrace = L"(no CLR stack trace available)";
     }
@@ -251,6 +257,12 @@ LONG WINAPI structured_exception_handler(EXCEPTION_POINTERS* ex)
 
 LONG WINAPI vectored_exception_handler(EXCEPTION_POINTERS* ex)
 {
+    // special case for CLR exceptions, always trigger crash handler
+    if (ex->ExceptionRecord->ExceptionCode == CUSTOM_EXCEPTION_EXTERNAL_EVENT)
+    {
+        return exception_handler(ex);
+    }
+
     if (ex->ExceptionRecord->ExceptionCode == 0x12345678)
     {
         // pass
@@ -434,3 +446,10 @@ bool veh::remove_handler()
     }
     return false;
 }
+
+void veh::raise_external_event(const std::wstring& info)
+{
+    const auto info_size = std::min(info.size(), std::size(g_external_event_info) - 1);
+    wcsncpy_s(g_external_event_info, info.c_str(), info_size);
+    RaiseException(CUSTOM_EXCEPTION_EXTERNAL_EVENT, 0, 0, nullptr);
+}

Dalamud.Boot/veh.h

@@ -4,4 +4,5 @@ namespace veh
 {
     bool add_handler(bool doFullDump, const std::string& workingDirectory);
     bool remove_handler();
+    void raise_external_event(const std::wstring& info);
 }

DalamudCrashHandler/DalamudCrashHandler.cpp

@@ -292,35 +292,43 @@ std::wstring to_address_string(const DWORD64 address, const bool try_ptrderef =
 
 void print_exception_info(HANDLE hThread, const EXCEPTION_POINTERS& ex, const CONTEXT& ctx, std::wostringstream& log) {
     std::vector<EXCEPTION_RECORD> exRecs;
-    if (ex.ExceptionRecord) {
+    if (ex.ExceptionRecord)
+    {
         size_t rec_index = 0;
         size_t read;
-        exRecs.emplace_back();
+
         for (auto pRemoteExRec = ex.ExceptionRecord;
-             pRemoteExRec
-             && rec_index < 64
-             && ReadProcessMemory(g_hProcess, pRemoteExRec, &exRecs.back(), sizeof exRecs.back(), &read)
-             && read >= offsetof(EXCEPTION_RECORD, ExceptionInformation)
-             && read >= static_cast<size_t>(reinterpret_cast<const char*>(&exRecs.back().ExceptionInformation[exRecs.back().NumberParameters]) - reinterpret_cast<const char*>(&exRecs.back()));
-             rec_index++) {
+             pRemoteExRec && rec_index < 64;
+             rec_index++)
+        {
+            exRecs.emplace_back();
+
+            if (!ReadProcessMemory(g_hProcess, pRemoteExRec, &exRecs.back(), sizeof exRecs.back(), &read)
+                || read < offsetof(EXCEPTION_RECORD, ExceptionInformation)
+                || read < static_cast<size_t>(reinterpret_cast<const char*>(&exRecs.back().ExceptionInformation[exRecs.
+                    back().NumberParameters]) - reinterpret_cast<const char*>(&exRecs.back())))
+            {
+                exRecs.pop_back();
+                break;
+            }
 
             log << std::format(L"\nException Info #{}\n", rec_index);
             log << std::format(L"Address: {:X}\n", exRecs.back().ExceptionCode);
             log << std::format(L"Flags: {:X}\n", exRecs.back().ExceptionFlags);
             log << std::format(L"Address: {:X}\n", reinterpret_cast<size_t>(exRecs.back().ExceptionAddress));
-            if (!exRecs.back().NumberParameters)
-                continue;
+            if (exRecs.back().NumberParameters)
+            {
                 log << L"Parameters: ";
-            for (DWORD i = 0; i < exRecs.back().NumberParameters; ++i) {
+                for (DWORD i = 0; i < exRecs.back().NumberParameters; ++i)
+                {
                     if (i != 0)
                         log << L", ";
                     log << std::format(L"{:X}", exRecs.back().ExceptionInformation[i]);
                 }
+            }
 
             pRemoteExRec = exRecs.back().ExceptionRecord;
-            exRecs.emplace_back();
         }
-        exRecs.pop_back();
     }
 
     log << L"\nCall Stack\n{";
@@ -930,9 +938,19 @@ int main() {
             } while (false);
         }
 
+        const bool is_external_event = exinfo.ExceptionRecord.ExceptionCode == CUSTOM_EXCEPTION_EXTERNAL_EVENT;
+
         std::wostringstream log;
+
+        if (!is_external_event)
+        {
             log << std::format(L"Unhandled native exception occurred at {}", to_address_string(exinfo.ContextRecord.Rip, false)) << std::endl;
             log << std::format(L"Code: {:X}", exinfo.ExceptionRecord.ExceptionCode) << std::endl;
+        }
+        else
+        {
+            log << L"CLR error occurred" << std::endl;
+        }
 
         if (shutup)
             log << L"======= Crash handler was globally muted(shutdown?) =======" << std::endl;
@@ -949,9 +967,19 @@ int main() {
         if (pProgressDialog)
             pProgressDialog->SetLine(3, L"Refreshing Module List", FALSE, NULL);
 
+        std::wstring window_log_str;
+
+        // Cut the log here for external events, the rest is unreadable and doesn't matter since we can't get
+        // symbols for mixed-mode stacks yet.
+        if (is_external_event)
+            window_log_str = log.str();
+
         SymRefreshModuleList(GetCurrentProcess());
         print_exception_info(exinfo.hThreadHandle, exinfo.ExceptionPointers, exinfo.ContextRecord, log);
-        const auto window_log_str = log.str();
+
+        if (!is_external_event)
+            window_log_str = log.str();
+
         print_exception_info_extended(exinfo.ExceptionPointers, exinfo.ContextRecord, log);
         std::wofstream(logPath) << log.str();