author Tom Ritter <tom@mozilla.com>
Mon, 24 Jun 2019 13:20:49 +0000
changeset 537136 828a590988ad2726fcf884fd3af84d1d20ffc141
parent 528177 9e33009abc2a84be589ebc0d301fabc865cb92cb
permissions -rw-r--r--
Bug 1560651 - Correctly set time clamping/jittering for Workers. r=baku, a=RyanVM Differential Revision: https://phabricator.services.mozilla.com/D35629

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
 * vim: set ts=8 sts=2 et sw=2 tw=80:
 * This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

// Interfaces by which the embedding can interact with the Debugger API.

#ifndef js_Debug_h
#define js_Debug_h

#include "mozilla/Assertions.h"
#include "mozilla/Attributes.h"
#include "mozilla/MemoryReporting.h"

#include "jsapi.h"
#include "jspubtd.h"

#include "js/GCAPI.h"
#include "js/RootingAPI.h"
#include "js/TypeDecls.h"

namespace js {
class Debugger;
}  // namespace js

namespace JS {
namespace dbg {

// [SMDOC] Debugger builder API
// Helping embedding code build objects for Debugger
// -------------------------------------------------
// Some Debugger API features lean on the embedding application to construct
// their result values. For example, Debugger.Frame.prototype.scriptEntryReason
// calls hooks provided by the embedding to construct values explaining why it
// invoked JavaScript; if F is a frame called from a mouse click event handler,
// F.scriptEntryReason would return an object of the form:
//   { eventType: "mousedown", event: <object> }
// where <object> is a Debugger.Object whose referent is the event being
// dispatched.
// However, Debugger implements a trust boundary. Debuggee code may be
// considered untrusted; debugger code needs to be protected from debuggee
// getters, setters, proxies, Object.watch watchpoints, and any other feature
// that might accidentally cause debugger code to set the debuggee running. The
// Debugger API tries to make it easy to write safe debugger code by only
// offering access to debuggee objects via Debugger.Object instances, which
// ensure that only those operations whose explicit purpose is to invoke
// debuggee code do so. But this protective membrane is only helpful if we
// interpose Debugger.Object instances in all the necessary spots.
// SpiderMonkey's compartment system also implements a trust boundary. The
// debuggee and debugger are always in different compartments. Inter-compartment
// work requires carefully tracking which compartment each JSObject or JS::Value
// belongs to, and ensuring that is is correctly wrapped for each operation.
// It seems precarious to expect the embedding's hooks to implement these trust
// boundaries. Instead, the JS::dbg::Builder API segregates the code which
// constructs trusted objects from that which deals with untrusted objects.
// Trusted objects have an entirely different C++ type, so code that improperly
// mixes trusted and untrusted objects is caught at compile time.
// In the structure shown above, there are two trusted objects, and one
// untrusted object:
// - The overall object, with the 'eventType' and 'event' properties, is a
//   trusted object. We're going to return it to D.F.p.scriptEntryReason's
//   caller, which will handle it directly.
// - The Debugger.Object instance appearing as the value of the 'event' property
//   is a trusted object. It belongs to the same Debugger instance as the
//   Debugger.Frame instance whose scriptEntryReason accessor was called, and
//   presents a safe reflection-oriented API for inspecting its referent, which
//   is:
// - The actual event object, an untrusted object, and the referent of the
//   Debugger.Object above. (Content can do things like replacing accessors on
//   Event.prototype.)
// Using JS::dbg::Builder, all objects and values the embedding deals with
// directly are considered untrusted, and are assumed to be debuggee values. The
// only way to construct trusted objects is to use Builder's own methods, which
// return a separate Object type. The only way to set a property on a trusted
// object is through that Object type. The actual trusted object is never
// exposed to the embedding.
// So, for example, the embedding might use code like the following to construct
// the object shown above, given a Builder passed to it by Debugger:
//    bool
//    MyScriptEntryReason::explain(JSContext* cx,
//                                 Builder& builder,
//                                 Builder::Object& result)
//    {
//        JSObject* eventObject = ... obtain debuggee event object somehow ...;
//        if (!eventObject) {
//            return false;
//        }
//        result = builder.newObject(cx);
//        return result &&
//               result.defineProperty(cx, "eventType",
//                                     SafelyFetchType(eventObject)) &&
//               result.defineProperty(cx, "event", eventObject);
//    }
// Object::defineProperty also accepts an Object as the value to store on the
// property. By its type, we know that the value is trusted, so we set it
// directly as the property's value, without interposing a Debugger.Object
// wrapper. This allows the embedding to builted nested structures of trusted
// objects.
// The Builder and Builder::Object methods take care of doing whatever
// compartment switching and wrapping are necessary to construct the trusted
// values in the Debugger's compartment.
// The Object type is self-rooting. Construction, assignment, and destruction
// all properly root the referent object.

class BuilderOrigin;

class Builder {
  // The Debugger instance whose client we are building a value for. We build
  // objects in this object's compartment.
  PersistentRootedObject debuggerObject;

  // debuggerObject's Debugger structure, for convenience.
  js::Debugger* debugger;

  // Check that |thing| is in the same compartment as our debuggerObject. Used
  // for assertions when constructing BuiltThings. We can overload this as we
  // add more instantiations of BuiltThing.
  void assertBuilt(JSObject* obj);
  void assertBuilt(JSObject* obj) {}

  // A reference to a trusted object or value. At the moment, we only use it
  // with JSObject*.
  template <typename T>
  class BuiltThing {
    friend class BuilderOrigin;

    // The Builder to which this trusted thing belongs.
    Builder& owner;

    // A rooted reference to our value.
    PersistentRooted<T> value;

    BuiltThing(JSContext* cx, Builder& owner_,
               T value_ = SafelyInitialized<T>())
        : owner(owner_), value(cx, value_) {

    // Forward some things from our owner, for convenience.
    js::Debugger* debugger() const { return owner.debugger; }
    JSObject* debuggerObject() const { return owner.debuggerObject; }

    BuiltThing(const BuiltThing& rhs) : owner(rhs.owner), value(rhs.value) {}
    BuiltThing& operator=(const BuiltThing& rhs) {
      MOZ_ASSERT(&owner == &rhs.owner);
      value = rhs.value;
      return *this;

    explicit operator bool() const {
      // If we ever instantiate BuiltThing<Value>, this might not suffice.
      return value;

    BuiltThing() = delete;

  // A reference to a trusted object, possibly null. Instances of Object are
  // always properly rooted. They can be copied and assigned, as if they were
  // pointers.
  class Object : private BuiltThing<JSObject*> {
    friend class Builder;        // for construction
    friend class BuilderOrigin;  // for unwrapping

    typedef BuiltThing<JSObject*> Base;

    // This is private, because only Builders can create Objects that
    // actually point to something (hence the 'friend' declaration).
    Object(JSContext* cx, Builder& owner_, HandleObject obj)
        : Base(cx, owner_, obj.get()) {}

    bool definePropertyToTrusted(JSContext* cx, const char* name,
                                 JS::MutableHandleValue value);

    Object(JSContext* cx, Builder& owner_) : Base(cx, owner_, nullptr) {}
    Object(const Object& rhs) : Base(rhs) {}

    // Our automatically-generated assignment operator can see our base
    // class's assignment operator, so we don't need to write one out here.

    // Set the property named |name| on this object to |value|.
    // If |value| is a string or primitive, re-wrap it for the debugger's
    // compartment.
    // If |value| is an object, assume it is a debuggee object and make a
    // Debugger.Object instance referring to it. Set that as the propery's
    // value.
    // If |value| is another trusted object, store it directly as the
    // property's value.
    // On error, report the problem on cx and return false.
    bool defineProperty(JSContext* cx, const char* name, JS::HandleValue value);
    bool defineProperty(JSContext* cx, const char* name,
                        JS::HandleObject value);
    bool defineProperty(JSContext* cx, const char* name, Object& value);

    using Base::operator bool;

  // Build an empty object for direct use by debugger code, owned by this
  // Builder. If an error occurs, report it on cx and return a false Object.
  Object newObject(JSContext* cx);

  Builder(JSContext* cx, js::Debugger* debugger);

// Debugger itself instantiates this subclass of Builder, which can unwrap
// BuiltThings that belong to it.
class BuilderOrigin : public Builder {
  template <typename T>
  T unwrapAny(const BuiltThing<T>& thing) {
    MOZ_ASSERT(&thing.owner == this);
    return thing.value.get();

  BuilderOrigin(JSContext* cx, js::Debugger* debugger_)
      : Builder(cx, debugger_) {}

  JSObject* unwrap(Object& object) { return unwrapAny(object); }

// Finding the size of blocks allocated with malloc
// ------------------------------------------------
// Debugger.Memory wants to be able to report how many bytes items in memory are
// consuming. To do this, it needs a function that accepts a pointer to a block,
// and returns the number of bytes allocated to that block. SpiderMonkey itself
// doesn't know which function is appropriate to use, but the embedding does.

// Tell Debuggers in |cx| to use |mallocSizeOf| to find the size of
// malloc'd blocks.
JS_PUBLIC_API void SetDebuggerMallocSizeOf(JSContext* cx,
                                           mozilla::MallocSizeOf mallocSizeOf);

// Get the MallocSizeOf function that the given context is using to find the
// size of malloc'd blocks.
JS_PUBLIC_API mozilla::MallocSizeOf GetDebuggerMallocSizeOf(JSContext* cx);

// Debugger and Garbage Collection Events
// --------------------------------------
// The Debugger wants to report about its debuggees' GC cycles, however entering
// JS after a GC is troublesome since SpiderMonkey will often do something like
// force a GC and then rely on the nursery being empty. If we call into some
// Debugger's hook after the GC, then JS runs and the nursery won't be
// empty. Instead, we rely on embedders to call back into SpiderMonkey after a
// GC and notify Debuggers to call their onGarbageCollection hook.

// Determine whether it's necessary to call FireOnGarbageCollectionHook() after
// a GC. This is only required if there are debuggers with an
// onGarbageCollection hook observing a global in the set of collected zones.
JS_PUBLIC_API bool FireOnGarbageCollectionHookRequired(JSContext* cx);

// For each Debugger that observed a debuggee involved in the given GC event,
// call its `onGarbageCollection` hook.
JS_PUBLIC_API bool FireOnGarbageCollectionHook(
    JSContext* cx, GarbageCollectionEvent::Ptr&& data);

// Return true if the given value is a Debugger object, false otherwise.
JS_PUBLIC_API bool IsDebugger(JSObject& obj);

// Append each of the debuggee global objects observed by the Debugger object
// |dbgObj| to |vector|. Returns true on success, false on failure.
JS_PUBLIC_API bool GetDebuggeeGlobals(JSContext* cx, JSObject& dbgObj,
                                      MutableHandleObjectVector vector);

// Hooks for reporting where JavaScript execution began.
// Our performance tools would like to be able to label blocks of JavaScript
// execution with the function name and source location where execution began:
// the event handler, the callback, etc.
// Construct an instance of this class on the stack, providing a JSContext
// belonging to the runtime in which execution will occur. Each time we enter
// JavaScript --- specifically, each time we push a JavaScript stack frame that
// has no older JS frames younger than this AutoEntryMonitor --- we will
// call the appropriate |Entry| member function to indicate where we've begun
// execution.

class MOZ_STACK_CLASS JS_PUBLIC_API AutoEntryMonitor {
  JSContext* cx_;
  AutoEntryMonitor* savedMonitor_;

  explicit AutoEntryMonitor(JSContext* cx);

  // SpiderMonkey reports the JavaScript entry points occuring within this
  // AutoEntryMonitor's scope to the following member functions, which the
  // embedding is expected to override.
  // It is important to note that |asyncCause| is owned by the caller and its
  // lifetime must outlive the lifetime of the AutoEntryMonitor object. It is
  // strongly encouraged that |asyncCause| be a string constant or similar
  // statically allocated string.

  // We have begun executing |function|. Note that |function| may not be the
  // actual closure we are running, but only the canonical function object to
  // which the script refers.
  virtual void Entry(JSContext* cx, JSFunction* function,
                     HandleValue asyncStack, const char* asyncCause) = 0;

  // Execution has begun at the entry point of |script|, which is not a
  // function body. (This is probably being executed by 'eval' or some
  // JSAPI equivalent.)
  virtual void Entry(JSContext* cx, JSScript* script, HandleValue asyncStack,
                     const char* asyncCause) = 0;

  // Execution of the function or script has ended.
  virtual void Exit(JSContext* cx) {}

}  // namespace dbg
}  // namespace JS

#endif /* js_Debug_h */