src: support fs and v8 binding data in snapshot

Author: joyeecheung

lib/fs.js

@@ -67,6 +67,10 @@ const {
 
 const pathModule = require('path');
 const { isArrayBufferView } = require('internal/util/types');
+
+// We need to get the statValues from the binding at the callsite since
+// it's re-initialized after deserialization.
+
 const binding = internalBinding('fs');
 const { Buffer } = require('buffer');
 const {
@@ -82,7 +86,7 @@ const {
   uvException
 } = require('internal/errors');
 
-const { FSReqCallback, statValues } = binding;
+const { FSReqCallback } = binding;
 const { toPathIfFileURL } = require('internal/url');
 const internalUtil = require('internal/util');
 const {
@@ -1781,8 +1785,8 @@ function realpathSync(p, options) {
 
     // Continue if not a symlink, break if a pipe/socket
     if (knownHard[base] || cache?.get(base) === base) {
-      if (isFileType(statValues, S_IFIFO) ||
-          isFileType(statValues, S_IFSOCK)) {
+      if (isFileType(binding.statValues, S_IFIFO) ||
+          isFileType(binding.statValues, S_IFSOCK)) {
         break;
       }
       continue;
@@ -1924,8 +1928,8 @@ function realpath(p, options, callback) {
 
     // Continue if not a symlink, break if a pipe/socket
     if (knownHard[base]) {
-      if (isFileType(statValues, S_IFIFO) ||
-          isFileType(statValues, S_IFSOCK)) {
+      if (isFileType(binding.statValues, S_IFIFO) ||
+          isFileType(binding.statValues, S_IFSOCK)) {
         return callback(null, encodeRealpathResult(p, options));
       }
       return process.nextTick(LOOP);

lib/internal/bootstrap/node.js

@@ -347,6 +347,9 @@ process.emitWarning = emitWarning;
   // Note: only after this point are the timers effective
 }
 
+require('fs');
+internalBinding('v8');
+
 function setupPrepareStackTrace() {
   const {
     setEnhanceStackForFatalException,

lib/v8.js

@@ -72,12 +72,13 @@ function getHeapSnapshot() {
   return new HeapSnapshotStream(handle);
 }
 
+// We need to get the buffer from the binding at the callsite since
+// it's re-initialized after deserialization.
+const binding = internalBinding('v8');
+
 const {
   cachedDataVersionTag,
   setFlagsFromString: _setFlagsFromString,
-  heapStatisticsBuffer,
-  heapSpaceStatisticsBuffer,
-  heapCodeStatisticsBuffer,
   updateHeapStatisticsBuffer,
   updateHeapSpaceStatisticsBuffer,
   updateHeapCodeStatisticsBuffer,
@@ -106,7 +107,7 @@ const {
   kCodeAndMetadataSizeIndex,
   kBytecodeAndMetadataSizeIndex,
   kExternalScriptSourceSizeIndex
-} = internalBinding('v8');
+} = binding;
 
 const kNumberOfHeapSpaces = kHeapSpaces.length;
 
@@ -116,7 +117,7 @@ function setFlagsFromString(flags) {
 }
 
 function getHeapStatistics() {
-  const buffer = heapStatisticsBuffer;
+  const buffer = binding.heapStatisticsBuffer;
 
   updateHeapStatisticsBuffer();
 
@@ -137,7 +138,7 @@ function getHeapStatistics() {
 
 function getHeapSpaceStatistics() {
   const heapSpaceStatistics = new Array(kNumberOfHeapSpaces);
-  const buffer = heapSpaceStatisticsBuffer;
+  const buffer = binding.heapSpaceStatisticsBuffer;
 
   for (let i = 0; i < kNumberOfHeapSpaces; i++) {
     updateHeapSpaceStatisticsBuffer(i);
@@ -154,7 +155,7 @@ function getHeapSpaceStatistics() {
 }
 
 function getHeapCodeStatistics() {
-  const buffer = heapCodeStatisticsBuffer;
+  const buffer = binding.heapCodeStatisticsBuffer;
 
   updateHeapCodeStatisticsBuffer();
   return {

node.gyp

@@ -641,6 +641,7 @@
         'src/node_report_module.cc',
         'src/node_report_utils.cc',
         'src/node_serdes.cc',
+        'src/node_serializable.cc',
         'src/node_sockaddr.cc',
         'src/node_stat_watcher.cc',
         'src/node_symbols.cc',
@@ -743,12 +744,14 @@
         'src/node_report.h',
         'src/node_revert.h',
         'src/node_root_certs.h',
+        'src/node_serializable.h',
         'src/node_sockaddr.h',
         'src/node_sockaddr-inl.h',
         'src/node_stat_watcher.h',
         'src/node_union_bytes.h',
         'src/node_url.h',
         'src/node_version.h',
+        'src/node_v8.h',
         'src/node_v8_platform-inl.h',
         'src/node_wasi.h',
         'src/node_watchdog.h',

src/README.md

@@ -422,7 +422,7 @@ that state is through the use of `Environment::AddBindingData`, which gives
 binding functions access to an object for storing such state.
 That object is always a [`BaseObject`][].
 
-Its class needs to have a static `binding_data_name` field based on a
+Its class needs to have a static `type_name` field based on a
 constant string, in order to disambiguate it from other classes of this type,
 and which could e.g. match the binding’s name (in the example above, that would
 be `cares_wrap`).
@@ -433,7 +433,7 @@ class BindingData : public BaseObject {
  public:
   BindingData(Environment* env, Local<Object> obj) : BaseObject(env, obj) {}
 
-  static constexpr FastStringKey binding_data_name { "http_parser" };
+  static constexpr FastStringKey type_name { "http_parser" };
 
   std::vector<char> parser_buffer;
   bool parser_buffer_in_use = false;

src/aliased_buffer.h

@@ -198,6 +198,11 @@ class AliasedBufferBase {
     return js_array_.Get(isolate_);
   }
 
+  void Release() {
+    DCHECK_NULL(index_);
+    js_array_.Reset();
+  }
+
   /**
   *  Get the underlying v8::ArrayBuffer underlying the TypedArray and
   *  overlaying the native buffer

src/base_object.h

@@ -158,6 +158,9 @@ class BaseObject : public MemoryRetainer {
 
   virtual inline void OnGCCollect();
 
+  bool is_serializable() const { return is_serializable_; }
+  void set_is_serializable(bool val) { is_serializable_ = val; }
+
  private:
   v8::Local<v8::Object> WrappedObject() const override;
   bool IsRootNode() const override;
@@ -206,6 +209,7 @@ class BaseObject : public MemoryRetainer {
 
   Environment* env_;
   PointerData* pointer_data_ = nullptr;
+  bool is_serializable_ = false;
 };
 
 // Global alias for FromJSObject() to avoid churn.

src/env-inl.h

@@ -358,7 +358,7 @@ inline T* Environment::GetBindingData(v8::Local<v8::Context> context) {
       context->GetAlignedPointerFromEmbedderData(
           ContextEmbedderIndex::kBindingListIndex));
   DCHECK_NOT_NULL(map);
-  auto it = map->find(T::binding_data_name);
+  auto it = map->find(T::type_name);
   if (UNLIKELY(it == map->end())) return nullptr;
   T* result = static_cast<T*>(it->second.get());
   DCHECK_NOT_NULL(result);
@@ -377,7 +377,7 @@ inline T* Environment::AddBindingData(
       context->GetAlignedPointerFromEmbedderData(
           ContextEmbedderIndex::kBindingListIndex));
   DCHECK_NOT_NULL(map);
-  auto result = map->emplace(T::binding_data_name, item);
+  auto result = map->emplace(T::type_name, item);
   CHECK(result.second);
   DCHECK_EQ(GetBindingData<T>(context), item.get());
   return item.get();
@@ -1081,6 +1081,17 @@ void Environment::ForEachBaseObject(T&& iterator) {
   }
 }
 
+template <typename T>
+void Environment::ForEachBindingData(T&& iterator) {
+  BindingDataStore* map = static_cast<BindingDataStore*>(
+      context()->GetAlignedPointerFromEmbedderData(
+          ContextEmbedderIndex::kBindingListIndex));
+  DCHECK_NOT_NULL(map);
+  for (auto& it : *map) {
+    iterator(it.first, it.second);
+  }
+}
+
 void Environment::modify_base_object_count(int64_t delta) {
   base_object_count_ += delta;
 }

src/env.cc

@@ -8,9 +8,11 @@
 #include "node_buffer.h"
 #include "node_context_data.h"
 #include "node_errors.h"
+#include "node_file.h"
 #include "node_internals.h"
 #include "node_options-inl.h"
 #include "node_process.h"
+#include "node_v8.h"
 #include "node_v8_platform-inl.h"
 #include "node_worker.h"
 #include "req_wrap-inl.h"
@@ -1258,6 +1260,7 @@ EnvSerializeInfo Environment::Serialize(SnapshotCreator* creator) {
   EnvSerializeInfo info;
   Local<Context> ctx = context();
 
+  SerializeBindingData(this, creator, &info);
   // Currently all modules are compiled without cache in builtin snapshot
   // builder.
   info.native_modules = std::vector<std::string>(
@@ -1324,6 +1327,9 @@ std::ostream& operator<<(std::ostream& output,
 
 std::ostream& operator<<(std::ostream& output, const EnvSerializeInfo& i) {
   output << "{\n"
+         << "// -- bindings begins --\n"
+         << i.bindings << ",\n"
+         << "// -- bindings ends --\n"
          << "// -- native_modules begins --\n"
          << i.native_modules << ",\n"
          << "// -- native_modules ends --\n"
@@ -1349,9 +1355,29 @@ std::ostream& operator<<(std::ostream& output, const EnvSerializeInfo& i) {
   return output;
 }
 
+void Environment::EnqueueDeserializeRequest(DeserializeRequest request) {
+  deserialize_requests_.push_back(std::move(request));
+}
+
+void Environment::RunDeserializeRequests() {
+  HandleScope scope(isolate());
+  Local<Context> ctx = context();
+  Isolate* is = isolate();
+  while (!deserialize_requests_.empty()) {
+    DeserializeRequest request(std::move(deserialize_requests_.front()));
+    deserialize_requests_.pop_front();
+    Local<Object> holder = request.holder.Get(is);
+    request.cb(ctx, holder, request.info);
+    request.holder.Reset();  // unnecessary?
+    request.info->Delete();
+  }
+}
+
 void Environment::DeserializeProperties(const EnvSerializeInfo* info) {
   Local<Context> ctx = context();
 
+  RunDeserializeRequests();
+
   native_modules_in_snapshot = info->native_modules;
   async_hooks_.Deserialize(ctx);
   immediate_info_.Deserialize(ctx);

src/env.h

@@ -37,6 +37,7 @@
 #include "node_main_instance.h"
 #include "node_options.h"
 #include "node_perf_common.h"
+#include "node_serializable.h"
 #include "req_wrap.h"
 #include "util.h"
 #include "uv.h"
@@ -936,8 +937,22 @@ struct PropInfo {
   SnapshotIndex index;  // In the snapshot
 };
 
+typedef void (*DeserializeRequestCallback)(v8::Local<v8::Context>,
+                                           v8::Local<v8::Object> holder,
+                                           InternalFieldInfo* info);
+struct DeserializeRequest {
+  DeserializeRequestCallback cb;
+  v8::Global<v8::Object> holder;
+  InternalFieldInfo* info;  // Owned by the request
+
+  // Move constructor
+  DeserializeRequest(DeserializeRequest&& other) = default;
+};
+
 struct EnvSerializeInfo {
+  std::vector<PropInfo> bindings;
   std::vector<std::string> native_modules;
+
   AsyncHooks::SerializeInfo async_hooks;
   TickInfo::SerializeInfo tick_info;
   ImmediateInfo::SerializeInfo immediate_info;
@@ -971,6 +986,8 @@ class Environment : public MemoryRetainer {
 
   void PrintAllBaseObjects();
   void VerifyNoStrongBaseObjects();
+  void EnqueueDeserializeRequest(DeserializeRequest request);
+  void RunDeserializeRequests();
   // Should be called before InitializeInspector()
   void InitializeDiagnostics();
 
@@ -1407,6 +1424,9 @@ class Environment : public MemoryRetainer {
   void AddUnmanagedFd(int fd);
   void RemoveUnmanagedFd(int fd);
 
+  template <typename T>
+  void ForEachBindingData(T&& iterator);
+
  private:
   inline void ThrowError(v8::Local<v8::Value> (*fun)(v8::Local<v8::String>),
                          const char* errmsg);
@@ -1495,6 +1515,8 @@ class Environment : public MemoryRetainer {
   bool is_in_inspector_console_call_ = false;
 #endif
 
+  std::list<DeserializeRequest> deserialize_requests_;
+
   // handle_wrap_queue_ and req_wrap_queue_ needs to be at a fixed offset from
   // the start of the class because it is used by
   // src/node_postmortem_metadata.cc to calculate offsets and generate debug