Support loading standalone GC from an absolute path using a new config named DOTNET_GCPath (#101874)

Author: cshung

src/coreclr/gc/gcconfig.h

@@ -137,7 +137,8 @@ class GCConfigStringHolder
     INT_CONFIG   (GCEnabledInstructionSets,  "GCEnabledInstructionSets",  NULL,                                -1,                 "Specifies whether GC can use AVX2 or AVX512F - 0 for neither, 1 for AVX2, 3 for AVX512F")\
     INT_CONFIG   (GCConserveMem,             "GCConserveMemory",          "System.GC.ConserveMemory",          0,                  "Specifies how hard GC should try to conserve memory - values 0-9")                       \
     INT_CONFIG   (GCWriteBarrier,            "GCWriteBarrier",            NULL,                                0,                  "Specifies whether GC should use more precise but slower write barrier")                  \
-    STRING_CONFIG(GCName,                    "GCName",                    "System.GC.Name",                                        "Specifies the path of the standalone GC implementation.")                                \
+    STRING_CONFIG(GCName,                    "GCName",                    "System.GC.Name",                                        "Specifies the name of the standalone GC implementation.")                                \
+    STRING_CONFIG(GCPath,                    "GCPath",                    "System.GC.Path",                                        "Specifies the path of the standalone GC implementation.")                                \
     INT_CONFIG   (GCSpinCountUnit,           "GCSpinCountUnit",           NULL,                                0,                  "Specifies the spin count unit used by the GC.")                                          \
     INT_CONFIG   (GCDynamicAdaptationMode,   "GCDynamicAdaptationMode",   "System.GC.DynamicAdaptationMode",   0,                  "Enable the GC to dynamically adapt to application sizes.")
 // This class is responsible for retreiving configuration information

src/coreclr/inc/clrconfigvalues.h

@@ -284,6 +284,7 @@ CONFIG_DWORD_INFO(INTERNAL_GcStressOnDirectCalls, W("GcStressOnDirectCalls"), 0,
 RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_HeapVerify, W("HeapVerify"), 0, "When set verifies the integrity of the managed heap on entry and exit of each GC")
 RETAIL_CONFIG_DWORD_INFO(EXTERNAL_GCCpuGroup, W("GCCpuGroup"), 0, "Specifies if to enable GC to support CPU groups")
 RETAIL_CONFIG_STRING_INFO(EXTERNAL_GCName, W("GCName"), "")
+RETAIL_CONFIG_STRING_INFO(EXTERNAL_GCPath, W("GCPath"), "")
 /**
  * This flag allows us to force the runtime to use global allocation context on Windows x86/amd64 instead of thread allocation context just for testing purpose.
  * The flag is unsafe for a subtle reason. Although the access to the g_global_alloc_context is protected under a lock. The implementation of

src/coreclr/inc/utilcode.h

@@ -55,6 +55,8 @@
 #define CoreLibSatelliteName_A "System.Private.CoreLib.resources"
 #define CoreLibSatelliteNameLen 32
 
+bool ValidateModuleName(LPCWSTR pwzModuleName);
+
 class StringArrayList;
 
 #if !defined(_DEBUG_IMPL) && defined(_DEBUG) && !defined(DACCESS_COMPILE)

src/coreclr/nativeaot/Runtime/clrgc.enabled.cpp

@@ -58,52 +58,111 @@ HRESULT InitializeStandaloneGC()
     return GCHeapUtilities::InitializeStandaloneGC();
 }
 
+//  Validate that the name used to load the GC is just a simple file name
+//  and does not contain something that could be used in a non-qualified path.
+//  For example, using the string "..\..\..\clrgc.dll" we might attempt to
+//  load a GC from the root of the drive.
+//
+//  The minimal set of characters that we must check for and exclude are:
+//  On all platforms:
+//     '/'  - (forward slash)
+//  On Windows:
+//     '\\' - (backslash)
+//     ':'  - (colon)
+//
+//  Returns false if we find any of these characters in 'pwzModuleName'
+//  Returns true if we reach the null terminator without encountering
+//  any of these characters.
+//
+bool ValidateModuleName(const char* pwzModuleName)
+{
+    const char* pCurChar = pwzModuleName;
+    char curChar;
+    do {
+        curChar = *pCurChar;
+        if (curChar == '/'
+#ifdef TARGET_WINDOWS
+            || (curChar == '\\') || (curChar == ':')
+#endif
+        )
+        {
+            //  Return false if we find any of these character in 'pwzJitName'
+            return false;
+        }
+        pCurChar++;
+    } while (curChar != 0);
+
+    //  Return true; we have reached the null terminator
+    //
+    return true;
+}
+
 HRESULT GCHeapUtilities::InitializeStandaloneGC()
 {
     char* moduleName;
+    char* modulePath;
+
+    if (!RhConfig::Environment::TryGetStringValue("GCPath", &modulePath))
+    {
+        modulePath = nullptr;
+    }
 
     if (!RhConfig::Environment::TryGetStringValue("GCName", &moduleName))
     {
-        return GCHeapUtilities::InitializeDefaultGC();
+        moduleName = nullptr;
     }
 
-    NewArrayHolder<char> moduleNameHolder(moduleName);
-    HANDLE executableModule = PalGetModuleHandleFromPointer((void*)&PalGetModuleHandleFromPointer);
-    const TCHAR * executableModulePath = NULL;
-    PalGetModuleFileName(&executableModulePath, executableModule);
-    char* convertedExecutableModulePath = PalCopyTCharAsChar(executableModulePath);
-    if (!convertedExecutableModulePath)
+    if (!(moduleName || modulePath))
     {
-        return E_OUTOFMEMORY;
+        return GCHeapUtilities::InitializeDefaultGC();
     }
-    NewArrayHolder<char> convertedExecutableModulePathHolder(convertedExecutableModulePath);
+
+    NewArrayHolder<char> moduleNameHolder(moduleName);
+    if (!modulePath)
     {
-        char* p = convertedExecutableModulePath;
-        char* q = nullptr;
-        while (*p != '\0')
+        if (!ValidateModuleName(moduleName))
+        {
+            LOG((LF_GC, LL_FATALERROR, "GC initialization failed to load the Standalone GC library.\n"));
+            return E_FAIL;
+        }
+
+        HANDLE executableModule = PalGetModuleHandleFromPointer((void*)&PalGetModuleHandleFromPointer);
+        const TCHAR * executableModulePath = NULL;
+        PalGetModuleFileName(&executableModulePath, executableModule);
+        char* convertedExecutableModulePath = PalCopyTCharAsChar(executableModulePath);
+        if (!convertedExecutableModulePath)
+        {
+            return E_OUTOFMEMORY;
+        }
+        NewArrayHolder<char> convertedExecutableModulePathHolder(convertedExecutableModulePath);
         {
-            if (*p == DIRECTORY_SEPARATOR_CHAR)
+            char* p = convertedExecutableModulePath;
+            char* q = nullptr;
+            while (*p != '\0')
             {
-                q = p;
+                if (*p == DIRECTORY_SEPARATOR_CHAR)
+                {
+                    q = p;
+                }
+                p++;
             }
-            p++;
+            assert(q != nullptr);
+            q++;
+            *q = '\0';
         }
-        assert(q != nullptr);
-        q++;
-        *q = '\0';
-    }
-    size_t folderLength = strlen(convertedExecutableModulePath);
-    size_t nameLength = strlen(moduleName);
-    char* moduleFullPath = new (nothrow) char[folderLength + nameLength + 1];
-    if (!moduleFullPath)
-    {
-        return E_OUTOFMEMORY;
+        size_t folderLength = strlen(convertedExecutableModulePath);
+        size_t nameLength = strlen(moduleName);
+        modulePath = new (nothrow) char[folderLength + nameLength + 1];
+        if (!modulePath)
+        {
+            return E_OUTOFMEMORY;
+        }
+        strcpy(modulePath, convertedExecutableModulePath);
+        strcpy(modulePath + folderLength, moduleName);
     }
-    NewArrayHolder<char> moduleFullPathHolder(moduleFullPath);
-    strcpy(moduleFullPath, convertedExecutableModulePath);
-    strcpy(moduleFullPath + folderLength, moduleName);
 
-    HANDLE hMod = PalLoadLibrary(moduleFullPath);
+    NewArrayHolder<char> modulePathHolder(modulePath);
+    HANDLE hMod = PalLoadLibrary(modulePath);
 
     if (!hMod)
     {

src/coreclr/utilcode/util.cpp

@@ -32,6 +32,45 @@ bool g_arm64_atomics_present = false;
 
 #endif //!DACCESS_COMPILE
 
+//  Validate that the name used to load the JIT/GC is just a simple file name
+//  and does not contain something that could be used in a non-qualified path.
+//  For example, using the string "..\..\..\myjit.dll" we might attempt to
+//  load a JIT from the root of the drive.
+//
+//  The minimal set of characters that we must check for and exclude are:
+//  On all platforms:
+//     '/'  - (forward slash)
+//  On Windows:
+//     '\\' - (backslash)
+//     ':'  - (colon)
+//
+//  Returns false if we find any of these characters in 'pwzModuleName'
+//  Returns true if we reach the null terminator without encountering
+//  any of these characters.
+//
+bool ValidateModuleName(LPCWSTR pwzModuleName)
+{
+    LPCWSTR pCurChar = pwzModuleName;
+    wchar_t curChar;
+    do {
+        curChar = *pCurChar;
+        if (curChar == '/'
+#ifdef TARGET_WINDOWS
+            || (curChar == '\\') || (curChar == ':')
+#endif
+        )
+        {
+            //  Return false if we find any of these character in 'pwzJitName'
+            return false;
+        }
+        pCurChar++;
+    } while (curChar != 0);
+
+    //  Return true; we have reached the null terminator
+    //
+    return true;
+}
+
 //*****************************************************************************
 // Convert a string of hex digits into a hex value of the specified # of bytes.
 //*****************************************************************************

src/coreclr/vm/codeman.cpp

@@ -1669,45 +1669,6 @@ struct JIT_LOAD_DATA
 // Here's the global data for JIT load and initialization state.
 JIT_LOAD_DATA g_JitLoadData;
 
-//  Validate that the name used to load the JIT is just a simple file name
-//  and does not contain something that could be used in a non-qualified path.
-//  For example, using the string "..\..\..\myjit.dll" we might attempt to
-//  load a JIT from the root of the drive.
-//
-//  The minimal set of characters that we must check for and exclude are:
-//  On all platforms:
-//     '/'  - (forward slash)
-//  On Windows:
-//     '\\' - (backslash)
-//     ':'  - (colon)
-//
-//  Returns false if we find any of these characters in 'pwzJitName'
-//  Returns true if we reach the null terminator without encountering
-//  any of these characters.
-//
-static bool ValidateJitName(LPCWSTR pwzJitName)
-{
-    LPCWSTR pCurChar = pwzJitName;
-    wchar_t curChar;
-    do {
-        curChar = *pCurChar;
-        if (curChar == '/'
-#ifdef TARGET_WINDOWS
-            || (curChar == '\\') || (curChar == ':')
-#endif
-        )
-        {
-            //  Return false if we find any of these character in 'pwzJitName'
-            return false;
-        }
-        pCurChar++;
-    } while (curChar != 0);
-
-    //  Return true; we have reached the null terminator
-    //
-    return true;
-}
-
 CORINFO_OS getClrVmOs();
 
 #define LogJITInitializationError(...) \
@@ -1770,7 +1731,7 @@ static void LoadAndInitializeJIT(LPCWSTR pwzJitName DEBUGARG(LPCWSTR pwzJitPath)
             return;
         }
 
-        if (ValidateJitName(pwzJitName))
+        if (ValidateModuleName(pwzJitName))
         {
             // Load JIT from next to CoreCLR binary
             PathString CoreClrFolderHolder;

src/coreclr/vm/gcheaputilities.cpp

@@ -5,6 +5,7 @@
 #include "configuration.h"
 #include "gcheaputilities.h"
 #include "appdomain.hpp"
+#include "hostinformation.h"
 
 #include "../gc/env/gcenv.ee.h"
 #include "../gc/env/gctoeeinterface.standalone.inl"
@@ -85,7 +86,6 @@ PTR_VOID GCHeapUtilities::GetGCModuleBase()
 
 namespace
 {
-
 // This block of code contains all of the state necessary to handle incoming
 // EtwCallbacks before the GC has been initialized. This is a tricky problem
 // because EtwCallbacks can appear at any time, even when we are just about
@@ -161,18 +161,47 @@ void StashKeywordAndLevel(bool isPublicProvider, GCEventKeyword keywords, GCEven
 }
 
 #ifdef FEATURE_STANDALONE_GC
-HMODULE LoadStandaloneGc(LPCWSTR libFileName)
+HMODULE LoadStandaloneGc(LPCWSTR libFileName, LPCWSTR libFilePath)
 {
     LIMITED_METHOD_CONTRACT;
+    HMODULE result = nullptr;
+
+    if (libFilePath)
+    {
+        return CLRLoadLibrary(libFilePath);
+    }
 
-    // Look for the standalone GC module next to the clr binary
-    PathString libPath = GetInternalSystemDirectory();
-    libPath.Append(libFileName);
+    if (!ValidateModuleName(libFileName))
+    {
+        LOG((LF_GC, LL_INFO100, "Invalid GC name found %s\n", libFileName));
+        return nullptr;
+    }
+
+    SString appBase;
+    if (HostInformation::GetProperty("APP_CONTEXT_BASE_DIRECTORY", appBase))
+    {
+        PathString libPath = appBase.GetUnicode();
+        libPath.Append(libFileName);
 
-    LOG((LF_GC, LL_INFO100, "Loading standalone GC from path %s\n", libPath.GetUTF8()));
+        LOG((LF_GC, LL_INFO100, "Loading standalone GC from appBase %s\n", libPath.GetUTF8()));
+
+        LPCWSTR libraryName = libPath.GetUnicode();
+        result = CLRLoadLibrary(libraryName);
+    }
+
+    if (result == nullptr)
+    {
+        // Look for the standalone GC module next to the clr binary
+        PathString libPath = GetInternalSystemDirectory();
+        libPath.Append(libFileName);
+
+        LOG((LF_GC, LL_INFO100, "Loading standalone GC by coreclr %s\n", libPath.GetUTF8()));
+
+        LPCWSTR libraryName = libPath.GetUnicode();
+        result = CLRLoadLibrary(libraryName);
+    }
 
-    LPCWSTR libraryName = libPath.GetUnicode();
-    return CLRLoadLibrary(libraryName);
+    return result;
 }
 #endif // FEATURE_STANDALONE_GC
 
@@ -182,21 +211,24 @@ HMODULE LoadStandaloneGc(LPCWSTR libFileName)
 //
 // See Documentation/design-docs/standalone-gc-loading.md for details
 // on the loading protocol in use here.
-HRESULT LoadAndInitializeGC(LPCWSTR standaloneGcLocation)
+HRESULT LoadAndInitializeGC(LPCWSTR standaloneGCName, LPCWSTR standaloneGCPath)
 {
     LIMITED_METHOD_CONTRACT;
 
 #ifndef FEATURE_STANDALONE_GC
     LOG((LF_GC, LL_FATALERROR, "EE not built with the ability to load standalone GCs"));
     return E_FAIL;
 #else
-    HMODULE hMod = LoadStandaloneGc(standaloneGcLocation);
+    HMODULE hMod = LoadStandaloneGc(standaloneGCName, standaloneGCPath);
     if (!hMod)
     {
         HRESULT err = GetLastError();
 #ifdef LOGGING
-        MAKE_UTF8PTR_FROMWIDE(standaloneGcLocationUtf8, standaloneGcLocation);
-        LOG((LF_GC, LL_FATALERROR, "Load of %s failed\n", standaloneGcLocationUtf8));
+        LPCWSTR standaloneGCNameLogging = standaloneGCName ? standaloneGCName : W("");
+        LPCWSTR standaloneGCPathLogging = standaloneGCPath ? standaloneGCPath : W("");
+        MAKE_UTF8PTR_FROMWIDE(standaloneGCNameUtf8, standaloneGCNameLogging);
+        MAKE_UTF8PTR_FROMWIDE(standaloneGCPathUtf8, standaloneGCPathLogging);
+        LOG((LF_GC, LL_FATALERROR, "Load of %s or %s failed\n", standaloneGCNameUtf8, standaloneGCPathUtf8));
 #endif // LOGGING
         return __HRESULT_FROM_WIN32(err);
     }
@@ -344,16 +376,17 @@ HRESULT GCHeapUtilities::LoadAndInitialize()
     assert(g_gc_load_status == GC_LOAD_STATUS_BEFORE_START);
     g_gc_load_status = GC_LOAD_STATUS_START;
 
-    LPCWSTR standaloneGcLocation = Configuration::GetKnobStringValue(W("System.GC.Name"), CLRConfig::EXTERNAL_GCName);
+    LPCWSTR standaloneGCName = Configuration::GetKnobStringValue(W("System.GC.Name"), CLRConfig::EXTERNAL_GCName);
+    LPCWSTR standaloneGCPath = Configuration::GetKnobStringValue(W("System.GC.Path"), CLRConfig::EXTERNAL_GCPath);
     g_gc_dac_vars.major_version_number = GC_INTERFACE_MAJOR_VERSION;
     g_gc_dac_vars.minor_version_number = GC_INTERFACE_MINOR_VERSION;
-    if (!standaloneGcLocation)
+    if (!standaloneGCName && !standaloneGCPath)
     {
         return InitializeDefaultGC();
     }
     else
     {
-        return LoadAndInitializeGC(standaloneGcLocation);
+        return LoadAndInitializeGC(standaloneGCName, standaloneGCPath);
     }
 }