NativeAOT createdump fork/exec for crash dump generation

src/coreclr/CMakeLists.txt

@@ -12,6 +12,7 @@ include(../../eng/native/configurepaths.cmake)
 include(${CLR_ENG_NATIVE_DIR}/configurecompiler.cmake)
 
 include_directories("${CLR_SRC_NATIVE_DIR}")
+include_directories("${CLR_SRC_NATIVE_DIR}/inc")
 
 if(MSVC)
   set(CMAKE_CXX_STANDARD_LIBRARIES "") # do not link against standard win32 libs i.e. kernel32, uuid, user32, etc.

src/coreclr/debug/createdump/crashinfo.cpp

@@ -22,6 +22,7 @@ CrashInfo::CrashInfo(const CreateDumpOptions& options) :
     m_gatherFrames(options.CrashReport),
     m_crashThread(options.CrashThread),
     m_signal(options.Signal),
+    m_exceptionRecord(options.ExceptionRecord),
     m_moduleInfos(&ModuleInfoCompare),
     m_mainModule(nullptr),
     m_cbModuleMappings(0),
@@ -39,7 +40,7 @@ CrashInfo::CrashInfo(const CreateDumpOptions& options) :
     m_siginfo.si_signo = options.Signal;
     m_siginfo.si_code = options.SignalCode;
     m_siginfo.si_errno = options.SignalErrno;
-    m_siginfo.si_addr = options.SignalAddress;
+    m_siginfo.si_addr = (void*)options.SignalAddress;
 }
 
 CrashInfo::~CrashInfo()
@@ -193,6 +194,9 @@ CrashInfo::GatherCrashInfo(DumpType dumpType)
     {
         return false;
     }
+    // Add the special (fake) memory region for the special diagnostics info
+    MemoryRegion special(PF_R, SpecialDiagInfoAddress, SpecialDiagInfoAddress + PAGE_SIZE);
+    m_memoryRegions.insert(special);
 #ifdef __APPLE__
     InitializeOtherMappings();
 #endif

src/coreclr/debug/createdump/crashinfo.h

@@ -55,6 +55,7 @@ class CrashInfo : public ICLRDataEnumMemoryRegionsCallback, public ICLRDataLoggi
     bool m_gatherFrames;                            // if true, add the native and managed stack frames to the thread info
     pid_t m_crashThread;                            // crashing thread id or 0 if none
     uint32_t m_signal;                              // crash signal code or 0 if none
+    uint64_t m_exceptionRecord;                     // exception record address or 0 if none
     std::string m_name;                             // exe name
     siginfo_t m_siginfo;                            // signal info (if any)
     std::string m_coreclrPath;                      // the path of the coreclr module or empty if none
@@ -115,6 +116,7 @@ class CrashInfo : public ICLRDataEnumMemoryRegionsCallback, public ICLRDataLoggi
     inline const bool GatherFrames() const { return m_gatherFrames; }
     inline const pid_t CrashThread() const { return m_crashThread; }
     inline const uint32_t Signal() const { return m_signal; }
+    inline const uint64_t ExceptionRecord () const { return m_exceptionRecord; }
     inline const std::string& Name() const { return m_name; }
     inline const ModuleInfo* MainModule() const { return m_mainModule; }
     inline const uint64_t RuntimeBaseAddress() const { return m_runtimeBaseAddress; }

src/coreclr/debug/createdump/createdump.h

@@ -119,7 +119,8 @@ typedef struct
     int Signal;
     int SignalCode;
     int SignalErrno;
-    void* SignalAddress;
+    uint64_t SignalAddress;
+    uint64_t ExceptionRecord;
 } CreateDumpOptions;
 
 #ifdef HOST_UNIX
@@ -136,6 +137,7 @@ typedef struct
 #include "crashreportwriter.h"
 #include "dumpwriter.h"
 #include "runtimeinfo.h"
+#include "specialdiaginfo.h"
 #endif
 
 #ifndef MAX_LONGPATH

src/coreclr/debug/createdump/createdumpmain.cpp

@@ -71,11 +71,10 @@ int createdump_main(const int argc, const char* argv[])
     options.Signal = 0;
     options.CrashThread = 0;
     options.Pid = 0;
-#if defined(HOST_UNIX) && !defined(HOST_OSX)
     options.SignalCode = 0;
     options.SignalErrno = 0;
-    options.SignalAddress = nullptr;
-#endif
+    options.SignalAddress = 0;
+    options.ExceptionRecord = 0;
     bool help = false;
     int exitCode = 0;
 
@@ -141,7 +140,11 @@ int createdump_main(const int argc, const char* argv[])
             }
             else if (strcmp(*argv, "--address") == 0)
             {
-                options.SignalAddress = (void*)atoll(*++argv);
+                options.SignalAddress = atoll(*++argv);
+            }
+            else if (strcmp(*argv, "--exception-record") == 0)
+            {
+                options.ExceptionRecord = atoll(*++argv);
             }
 #endif
             else if ((strcmp(*argv, "-d") == 0) || (strcmp(*argv, "--diag") == 0))

src/coreclr/debug/createdump/dumpwriter.cpp

@@ -32,6 +32,27 @@ DumpWriter::OpenDump(const char* dumpFileName)
     return true;
 }
 
+bool
+DumpWriter::WriteDiagInfo(size_t size)
+{
+    // Write the diagnostics info header
+    SpecialDiagInfoHeader header = {
+        {SPECIAL_DIAGINFO_SIGNATURE},
+        SPECIAL_DIAGINFO_VERSION,
+        m_crashInfo.ExceptionRecord()
+    };
+    if (!WriteData(&header, sizeof(header))) {
+        return false;
+    }
+    size_t alignment = size - sizeof(header);
+    assert(alignment < sizeof(m_tempBuffer));
+    memset(m_tempBuffer, 0, alignment);
+    if (!WriteData(m_tempBuffer, alignment)) {
+        return false;
+    }
+    return true;
+}
+
 // Write all of the given buffer, handling short writes and EINTR. Return true iff successful.
 bool
 DumpWriter::WriteData(int fd, const void* buffer, size_t length)

src/coreclr/debug/createdump/dumpwriterelf.cpp

@@ -176,28 +176,37 @@ DumpWriter::WriteDump()
         size_t size = memoryRegion.Size();
         total += size;
 
-        while (size > 0)
+        if (address == SpecialDiagInfoAddress)
         {
-            size_t bytesToRead = std::min(size, sizeof(m_tempBuffer));
-            size_t read = 0;
-
-            if (!m_crashInfo.ReadProcessMemory((void*)address, m_tempBuffer, bytesToRead, &read)) {
-                printf_error("Error reading memory at %" PRIA PRIx64 " size %08zx FAILED %s (%d)\n", address, bytesToRead, strerror(g_readProcessMemoryErrno), g_readProcessMemoryErrno);
+            if (!WriteDiagInfo(size)) {
                 return false;
             }
-
-            // This can happen if the target process dies before createdump is finished
-            if (read == 0) {
-                printf_error("Error reading memory at %" PRIA PRIx64 " size %08zx returned 0 bytes read: %s (%d)\n", address, bytesToRead, strerror(g_readProcessMemoryErrno), g_readProcessMemoryErrno);
-                return false;
-            }
-
-            if (!WriteData(m_tempBuffer, read)) {
-                return false;
+        }
+        else
+        {
+            while (size > 0)
+            {
+                size_t bytesToRead = std::min(size, sizeof(m_tempBuffer));
+                size_t read = 0;
+
+                if (!m_crashInfo.ReadProcessMemory((void*)address, m_tempBuffer, bytesToRead, &read)) {
+                    printf_error("Error reading memory at %" PRIA PRIx64 " size %08zx FAILED %s (%d)\n", address, bytesToRead, strerror(g_readProcessMemoryErrno), g_readProcessMemoryErrno);
+                    return false;
+                }
+
+                // This can happen if the target process dies before createdump is finished
+                if (read == 0) {
+                    printf_error("Error reading memory at %" PRIA PRIx64 " size %08zx returned 0 bytes read: %s (%d)\n", address, bytesToRead, strerror(g_readProcessMemoryErrno), g_readProcessMemoryErrno);
+                    return false;
+                }
+
+                if (!WriteData(m_tempBuffer, read)) {
+                    return false;
+                }
+
+                address += read;
+                size -= read;
             }
-
-            address += read;
-            size -= read;
         }
     }
 

src/coreclr/debug/createdump/dumpwriterelf.h

@@ -56,6 +56,7 @@ class DumpWriter
     static bool WriteData(int fd, const void* buffer, size_t length);
 
 private:
+    bool WriteDiagInfo(size_t size);
     bool WriteProcessInfo();
     bool WriteAuxv();
     size_t GetNTFileInfoSize(size_t* alignmentBytes = nullptr);

src/coreclr/debug/createdump/dumpwritermacho.cpp

@@ -229,7 +229,13 @@ DumpWriter::WriteSegments()
             (segment.initprot & VM_PROT_EXECUTE) ? 'x' : '-',
             segment.initprot);
 
-        if (address == SpecialThreadInfoAddress)
+        if (address == SpecialDiagInfoAddress)
+        {
+            if (!WriteDiagInfo(size)) {
+                return false;
+            }
+        }
+        else if (address == SpecialThreadInfoAddress)
         {
             // Write the header
             SpecialThreadInfoHeader header = {

src/coreclr/debug/createdump/dumpwritermacho.h

@@ -42,6 +42,7 @@ class DumpWriter
     static bool WriteData(int fd, const void* buffer, size_t length);
 
 private:
+    bool WriteDiagInfo(size_t size);
     void BuildSegmentLoadCommands();
     void BuildThreadLoadCommands();
     bool WriteHeader(uint64_t* pFileOffset);

src/coreclr/debug/createdump/specialdiaginfo.h

@@ -0,0 +1,32 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+// ******************************************************************************
+// WARNING!!!: This code is also used by SOS in the diagnostics repo. Should be
+// updated in a backwards and forwards compatible way.
+// See: https://github.com/dotnet/diagnostics/blob/main/src/SOS/inc/specialdiaginfo.h
+// ******************************************************************************
+
+// This is a special memory region added to ELF and MachO dumps that contains extra diagnostics
+// information like the exception record for a crash for a NativeAOT app. The exception record
+// contains the pointer to the JSON formatted crash info.
+
+#define SPECIAL_DIAGINFO_SIGNATURE "DIAGINFOHEADER"
+#define SPECIAL_DIAGINFO_VERSION 1
+
+#ifdef __APPLE__
+const uint64_t SpecialDiagInfoAddress = 0x7fffffff10000000;
+#else
+#if TARGET_64BIT
+const uint64_t SpecialDiagInfoAddress = 0x00007ffffff10000;
+#else
+const uint64_t SpecialDiagInfoAddress = 0x7fff1000;
+#endif
+#endif
+
+struct SpecialDiagInfoHeader
+{
+    char Signature[16];
+    int32_t Version;
+    uint64_t ExceptionRecordAddress;
+};

src/coreclr/nativeaot/Runtime/CMakeLists.txt

@@ -142,6 +142,7 @@ else()
 
   list(APPEND COMMON_RUNTIME_SOURCES
     unix/PalRedhawkUnix.cpp
+    unix/PalCreateDump.cpp
     ${GC_DIR}/unix/gcenv.unix.cpp
     ${GC_DIR}/unix/numasupport.cpp
     ${GC_DIR}/unix/events.cpp

src/coreclr/nativeaot/Runtime/DebugHeader.cpp

@@ -17,6 +17,7 @@
 #include "thread.h"
 #include "threadstore.h"
 
+extern uint8_t g_CrashInfoBuffer[];
 GPTR_DECL(MethodTable, g_pFreeObjectEEType);
 
 struct DebugTypeEntry
@@ -195,6 +196,11 @@ extern "C" void PopulateDebugHeaders()
     MAKE_DEBUG_FIELD_ENTRY(ThreadBuffer, m_rgbAllocContextBuffer);
     MAKE_DEBUG_FIELD_ENTRY(ThreadBuffer, m_threadId);
     MAKE_DEBUG_FIELD_ENTRY(ThreadBuffer, m_pThreadStressLog);
+    MAKE_DEBUG_FIELD_ENTRY(ThreadBuffer, m_pExInfoStackHead);
+
+    MAKE_SIZE_ENTRY(ExInfo);
+    MAKE_DEBUG_FIELD_ENTRY(ExInfo, m_pPrevExInfo);
+    MAKE_DEBUG_FIELD_ENTRY(ExInfo, m_exception);
 
     MAKE_SIZE_ENTRY(MethodTable);
     MAKE_DEBUG_FIELD_ENTRY(MethodTable, m_uBaseSize);
@@ -244,6 +250,8 @@ extern "C" void PopulateDebugHeaders()
     MAKE_SIZE_ENTRY(RuntimeInstance);
     MAKE_DEBUG_FIELD_ENTRY(RuntimeInstance, m_pThreadStore);
 
+    MAKE_GLOBAL_ENTRY(g_CrashInfoBuffer);
+
     RuntimeInstance *g_pTheRuntimeInstance = GetRuntimeInstance();
     MAKE_GLOBAL_ENTRY(g_pTheRuntimeInstance);
 
@@ -269,4 +277,9 @@ extern "C" void PopulateDebugHeaders()
     static_assert(MethodTable::Flags::IsGenericFlag          == 0x02000000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Flags. If you change this value you must bump major_version_number.");
     static_assert(MethodTable::Flags::ElementTypeMask        == 0x7C000000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Flags. If you change this value you must bump major_version_number.");
     static_assert(MethodTable::Flags::ElementTypeShift       == 26,         "The debugging data contract has a hard coded dependency on this value of MethodTable::Flags. If you change this value you must bump major_version_number.");
+    static_assert(MethodTable::Flags::HasComponentSizeFlag   == 0x80000000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Flags. If you change this value you must bump major_version_number.");
+
+    static_assert(MethodTable::Kinds::CanonicalEEType        == 0x00000000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Kinds. If you change this value you must bump major_version_number.");
+    static_assert(MethodTable::Kinds::ParameterizedEEType    == 0x00020000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Kinds. If you change this value you must bump major_version_number.");
+    static_assert(MethodTable::Kinds::GenericTypeDefEEType   == 0x00030000, "The debugging data contract has a hard coded dependency on this value of MethodTable::Kinds. If you change this value you must bump major_version_number.");
 }

src/coreclr/nativeaot/Runtime/RuntimeInstance.cpp

@@ -37,7 +37,7 @@ bool ShouldHijackForGcStress(uintptr_t CallsiteIP, HijackType ht);
 #include "shash.inl"
 
 #define MAX_CRASHINFOBUFFER_SIZE 8192
-uint8_t g_CrashInfoBuffer[MAX_CRASHINFOBUFFER_SIZE];
+uint8_t g_CrashInfoBuffer[MAX_CRASHINFOBUFFER_SIZE] = { 0 };
 
 ThreadStore *   RuntimeInstance::GetThreadStore()
 {
@@ -50,6 +50,14 @@ COOP_PINVOKE_HELPER(uint8_t *, RhGetCrashInfoBuffer, (int32_t* pcbMaxSize))
     return g_CrashInfoBuffer;
 }
 
+#if TARGET_UNIX
+#include "PalCreateDump.h"
+COOP_PINVOKE_HELPER(void, RhCreateCrashDumpIfEnabled, (PEXCEPTION_RECORD pExceptionRecord, PCONTEXT pExContext))
+{
+    PalCreateCrashDumpIfEnabled(pExceptionRecord, pExContext);
+}
+#endif
+
 COOP_PINVOKE_HELPER(uint8_t *, RhGetRuntimeVersion, (int32_t* pcbLength))
 {
     *pcbLength = sizeof(CLR_PRODUCT_VERSION) - 1;           // don't include the terminating null

src/coreclr/nativeaot/Runtime/eventpipe/ds-rt-aot.cpp

@@ -248,4 +248,5 @@ ds_rt_aot_set_environment_variable (const ep_char16_t *name, const ep_char16_t *
     return SetEnvironmentVariableW(reinterpret_cast<LPCWSTR>(name), reinterpret_cast<LPCWSTR>(value)) ? S_OK : HRESULT_FROM_WIN32(GetLastError());
 #endif
 }
+
 #endif /* ENABLE_PERFTRACING */

src/coreclr/nativeaot/Runtime/eventpipe/ds-rt-aot.h

@@ -6,6 +6,7 @@
 #define DIAGNOSTICS_RT_AOT_H
 
 #include <eventpipe/ds-rt-config.h>
+#include <generatedumpflags.h>
 
 #ifdef ENABLE_PERFTRACING
 #include "ep-rt-aot.h"
@@ -181,8 +182,24 @@ ds_rt_generate_core_dump (
 {
     STATIC_CONTRACT_NOTHROW;
 
-    // Eventpipe driven core_dump is not currently supported in NativeAOT
-    return DS_IPC_E_NOTSUPPORTED;
+    ds_ipc_result_t result = DS_IPC_E_FAIL;
+#ifdef TARGET_UNIX
+    uint32_t flags = ds_generate_core_dump_command_payload_get_flags(payload);
+    if (commandId == DS_DUMP_COMMANDID_GENERATE_CORE_DUMP)
+    {
+        // For the old commmand, this payload field is a bool of whether to enable logging
+        flags = flags != 0 ? GenerateDumpFlagsLoggingEnabled : 0;
+    }
+    const ep_char16_t *dumpName = ds_generate_core_dump_command_payload_get_dump_name (payload);
+    int32_t dumpType = static_cast<int32_t>(ds_generate_core_dump_command_payload_get_dump_type (payload));
+    ep_char8_t *dumpNameUtf8 = ep_rt_utf16le_to_utf8_string (dumpName, ep_rt_utf16_string_len (dumpName));
+    extern bool PalGenerateCoreDump(const char* dumpName, int dumpType, uint32_t flags, char* errorMessageBuffer, int cbErrorMessageBuffer);
+    if (PalGenerateCoreDump(dumpNameUtf8, dumpType, flags, errorMessageBuffer, cbErrorMessageBuffer))
+    {
+        result = DS_IPC_S_OK;
+    }
+#endif
+    return result;
 }
 
 /*

src/coreclr/nativeaot/Runtime/unix/HardwareExceptions.cpp

@@ -10,6 +10,7 @@
 #include "UnixContext.h"
 #include "HardwareExceptions.h"
 #include "UnixSignals.h"
+#include "PalCreateDump.h"
 
 #if defined(HOST_APPLE)
 #include <mach/mach.h>
@@ -559,6 +560,8 @@ void SIGSEGVHandler(int code, siginfo_t *siginfo, void *context)
         // Restore the original or default handler and restart h/w exception
         RestoreSignalHandler(code, &g_previousSIGSEGV);
     }
+
+    PalCreateCrashDumpIfEnabled(code, siginfo);
 }
 
 // Handler for the SIGFPE signal
@@ -579,6 +582,8 @@ void SIGFPEHandler(int code, siginfo_t *siginfo, void *context)
         // Restore the original or default handler and restart h/w exception
         RestoreSignalHandler(code, &g_previousSIGFPE);
     }
+
+    PalCreateCrashDumpIfEnabled(code, siginfo);
 }
 
 // Initialize hardware exception handling