Avoiding sun.misc.Unsafe and embracing modules in Java libraries: missing links

[Rafael Winterhalter]

As for final methods, this is a problem with delegating and subclassing
proxies. For delegation, I would even claim that this is a bigger problem
as the original code is invoked on an unprepared instance. With
subclassing, the final method is simply invoked in its non-proxied form
what should retain consistent state. Also, subclassing proxies can emulate
self-invocation without proxying by setting a flag on the instance that is
reset after the proxied instance completes its super method call. Finally,
subclassing proxies avoid confusion when using instances as monitors etc.

The problem with creating instances without invoking a constructor that I
see when using the lookup facility, for instance, is that this lookup would
need access to the entire class hierarchy. If I defined a class Foo extends
Bar extends Object and Foo and Bar resided in different packages and I
wanted to invoke an instance of Foo without a constructor call, my lookup
might have the right access for Foo but not for Bar. Should this then
require another lookup to skip Bar's constructor as well. And if not, could
I not just define a subclass of Foo, get a handle for it and now
instantiate a subclass of Foo that can be used as an instance of Foo? This
would break both Foo's and Bar's encapsulation without enforcing any
restriction.

I still find instantiating non-serializable classes without constructor
calls not meaningful for production environments, at least I could not
think of a good use case. For a stubbing server, I claim that the mocking
must work on a higher level then an object instance. Since such a server
needs deployment, I would also call this server a production application
even if it might never run in a production environment. As for application
tests, it can make sense for mocking and other unit testing frameworks and
I think a Java agent would be a good faciliator as owners of an
Instrumentation instance already can emulate this behavior as I had
described above.

Best regards, Rafael



2018-04-06 5:37 GMT+02:00 Henri Tremblay <henri.tremblay at gmail.com>:

> Answers inline
>
> On 4 April 2018 at 08:05, Rafael Winterhalter <rafael.wth at gmail.com>
> wrote:
>
>> Hei Henri,
>>
>> for Maven, Gradle and IDEs I think it is sensible to expect that these
>> runners would resolve such a test module unless it is explicitly specified
>> otherwise. Already today, those runners patch the module under test in
>> order to include the test code. I would expect of these runners to adopt
>> similarly to a test module as they adopted to the module system in general.
>>
>> -> Yes. I agree.
>
>
>> As for Spring's use case, I believe that the approach was chosen such
>> that class proxies resemble interface proxies as much as possible. In
>> general however, I think that Spring could just generate a subclass proxy
>> and delegate to the overridden methods instead of delegating to an instance
>> of the proxied class. If the proxy class mimics the proxied class's
>> constructors' signatures, this proxy can be instantiated exactly like the
>> proxied class would have been. This way, using Objenesis and the like is no
>> longer necessary and I argue that one should avoid exposing this
>> functionality if there is a reasonable alternative that does not require
>> any derivation from Javas programming model. As a matter of fact, I often
>> find developers confused about how fields are not set on such Spring
>> proxies and I even found possible exploits in code because of it. For
>> example, imagine a class:
>>
>> -> I think extending and delegating to the super class was the case a
> long time ago but was causing all sorts of problems. I agree that proxying
> and delegating to the real class also causes other problems. We should ask
> them what caused then to pick the later. Doing that, we also need to be
> cautious about final methods.
>
> class Foo {
>>   public final production;
>>   Foo(boolean production) { this.production = production; } // Only call
>> from unit test
>>   public Foo() { this(true); }
>>   public void doSomethingSensible() {
>>     if (production) { doSecurityCheck(); }
>>     // do something ...
>>   }
>> }
>>
>> With Objenesis, you can create an instance of the above class and set it
>> to test mode with the production field being false as the default value. In
>> contrast, for the case of serialization, I already argued that support for
>> serializable types should probably go into MethodHandles.Lookup similar to
>> how defineClass was added. This does not render an additional security
>> threat, if Foo was serializable I could just synthetically create a
>> serialized vale of an instance with production being false and deserialize
>> this value in a production environment. On a side note, Spring would create
>> a class with the above field value automatically if Foo was a bean, making
>> this exploit very easy even when using a security manager as the proxy
>> instance is normally available to all sorts of code.
>>
>> Finally, I disagree with the argument that this sort of API should be
>> accessible to "standard code" only shielded by a security manager because
>> this feature is useful today. The module system becomes useless the moment
>> you punch even a little hole into it. Therefore, I find it important to
>> remove sun.misc.Unsafe as quickly as possible as the module system can be
>> easily circumvented until it is removed and since the JPMS only show its
>> full benefit once this is no longer possible. But in order to keep testing
>> frameworks that have a good reason to break module boundaries up and
>> running, I think adding a test modules is a reasonable compromise. As an
>> additional benefit, it would allow libraries like Mockito that should not
>> be used in production to self-identify.
>>
>
> ->  I don't see the breach in JPMS in particular. Reflection is
> restricted. Creating a class without calling a constructor is reflection.
> -> Using mockito in production is weird. Using Mockito in a stubbing
> server can make sense. But you can advocate it requires to add some
> parameters to the JVM to make it work.
>
>>
>> One could even argue that this ability to instantiate instance is nothing
>> that needs to be exposed to a Java agent so maybe this functionality could
>> also be part of the mentioned test module instead of being a part of
>> instrumentation. On the other hand, if there is a use case not being
>> considered for a Java agent, this agent might be tempted to open
>> jdk.internal.misc to gain this access, one might therefore just decide to
>> add the method to Instrumentation to offer an easy migration path to
>> applications that aim to migrate their code with a minimal delta.
>>
>> Best regards, Rafael
>>
>> 2018-04-03 21:26 GMT+02:00 Henri Tremblay <henri.tremblay at gmail.com>:
>>
>>> Hi Rafael,
>>>
>>> I don't like the idea to have to explicitly load a module to create
>>> mocks.
>>>
>>> Because Unsafe.allocateInstance is of course used to create mock but
>>> also proxies. For instance, Spring uses it extensively through Objenesis.
>>> And different frameworks doing serialization.
>>>
>>> Also, It means Gradle and Maven would need to be updated to load the
>>> module automatically in test context. And all IDE. A lot of annoyance for
>>> not much benefit since everybody is currently happy.
>>>
>>> I've been advocating that if creating a class without calling a
>>> constructor is useless in multiple contexts, it should be easily available
>>> through the API. I don't know exactly where to put it but it should be
>>>
>>>    - Available to all (java.base)
>>>    - Protected by the security manager
>>>
>>> For completeness, there are 4 ways to create a class without calling a
>>> constructor right now that I'm aware of:
>>>
>>>    - Unsafe.allocateInstance
>>>    - sun.reflect.ReflectionFactory.newConstructorForSerialization (my
>>>    favorite because it's the fastest one)
>>>    - Generate an extending class forgetting to call the super
>>>    constructor (so it's not exactly that same class that is instantiated). It
>>>    requires -Xverify:none
>>>    - Generate a class extending MagicAccessorImpl that will then
>>>    instantiates the wanted class but calling the wrong constructor
>>>
>>> Regards,
>>> Henri
>>>
>>>
>>>
>>> On 1 April 2018 at 17:02, Rafael Winterhalter <rafael.wth at gmail.com>
>>> wrote:
>>>
>>>> Hello,
>>>>
>>>> I am the/a maintainer of the libraries Byte Buddy, cglib and Mockito
>>>> with
>>>> countless dependencies upstream and I wanted to give a summary of
>>>> adopting
>>>> the JPMS and migrating away from sun.misc.Unsafe.
>>>>
>>>> 1. Java agents cannot define auxiliary classes.
>>>>
>>>> Byte Buddy does support the JPMS fully, however, it still relies on
>>>> sun.misc.Unsafe::defineClass for its Java agent API and currently
>>>> breaks on
>>>> Java 11 as this method was removed in a recent EA build. The reason for
>>>> using Unsafe is that many instrumentations need to define auxiliary
>>>> classes
>>>> to aid an instrumentation similar to javac which sometimes needs to
>>>> define
>>>> anonymous classes or even synthetic classes. For example, if a Java
>>>> agent
>>>> wants to register an event listener to some framework, such listeners
>>>> often
>>>> declare multiple methods what makes it impossible to fullfil the
>>>> listener
>>>> contract using a lambda expression. Instead, one typically injects an
>>>> additional class into the same package as the instrumented class. In
>>>> this
>>>> case, it is not possible to use MethodHandles.Lookup::defineClass as
>>>> the
>>>> class file transformer does not necessarily have private access to the
>>>> lookup of the instrumented class.
>>>>
>>>> The current workarounds are:
>>>>
>>>> a) Open the package jdk.internal.misc to gain access to this package's
>>>> Unsafe class. This can be done via Instrumentation::redefineModule.
>>>> b) Open the java.lang package to access ClassLoader via reflection.
>>>> c) Open the java.lang package to access the internal lookup with global
>>>> access rights.
>>>>
>>>> Of these solutions only (b) relies on standard API and is guaranteed to
>>>> function in the future but the solution still feels hacky and does not
>>>> work
>>>> for instrumentations of classes on the bootstrap loader. Opening
>>>> packages
>>>> also implies a risk of being applied carelessly since opening the
>>>> package
>>>> to the agent's module most likely opens the package to the unnamed
>>>> module
>>>> of the system class loader what invites to breaches of the JPMS
>>>> encapsulation by code that does not ship with the agent.
>>>>
>>>> To offer a better solution, I would like to suggest one of the
>>>> following:
>>>>
>>>> a) Add a method defineClass(ClassLoader, byte[], ProtectionDomain) to
>>>> the
>>>> Instrumentation interface that works similar to Unsafe::defineClass.
>>>> This
>>>> would provide a very simple migration path. Since agents have access to
>>>> jdk.internal.misc, adding this method does not add any capabilities to
>>>> the
>>>> agent, it merley avoids using internal API that might change.
>>>> b) Supply a MethodHandles.Lookup instance to the
>>>> ClassFileTransformer::transform API where the instance represents the
>>>> instrumented class's access rights. This does however carry the risk of
>>>> invoking the lookupClass method which would either load the instrumented
>>>> class prematurely causing a circularity error or return an unexpected
>>>> value
>>>> such as null. Since the lookup API generally relies on loaded types,
>>>> there
>>>> are probably other problems such as invoking Lookup::privateLookupIn
>>>> before
>>>> all involved types are loaded.
>>>>
>>>> For the sake of simlicity and since easy migration paths make a quick
>>>> adoption easier, I would suggestion solution (a), also in the light that
>>>> quick and dirty migrations might still choose option (b) to save time
>>>> and
>>>> also since (b) might cause problems when types are not yet loaded.
>>>>
>>>> 2. Java proxies cannot invoke default methods of proxied interfaces
>>>>
>>>> The Java proxy API does not currently allow the invocation of an
>>>> overridden
>>>> default method since
>>>> the InvocationHandler API only supplies an instance of
>>>> java.lang.reflection.Method. In Java 8, it was always possible to get
>>>> hold
>>>> of method handle of any method of the proxied interface and to
>>>> specialize
>>>> the handle on the interface type to invoke the default implementation.
>>>> With
>>>> the JPMS, even if an interface type is exported, this same type might
>>>> not
>>>> be opened to another module. This implies that if an InvocationHandler
>>>> is
>>>> defined by this module to which the interface is exported, this module's
>>>> InvocationHandler cannot resolve a specialized method handle to a
>>>> default
>>>> method of the proxied interface. As a matter of fact, such a call is
>>>> impossible in this scenario whereas the same call is possible if the
>>>> proxy
>>>> is implemented manually at compile time.
>>>>
>>>> As a solution, I suggest to provide an argument to the InvocationHandler
>>>> that represents a lookup instance with the access rights of the proxy
>>>> class. Using this lookup, the specialized method handles could be
>>>> resolved.
>>>>
>>>> 3. Mocking and serialization libraries still require
>>>> Unsafe::allocateInstance.
>>>>
>>>> For Mockito, it is required to instantiate any class without invoking a
>>>> constructor with potential side-effects. This is of course a grose
>>>> violation of Java's general contract for class instantiation but this
>>>> feature is very useful.
>>>>
>>>> Using a Java agent, it is already possible to emulate this feature
>>>> without
>>>> internal API by instrumenting all constructors of all classes in the
>>>> hierarchy of a mocked class by transforming all constructors into the
>>>> following pseudo-code:
>>>>
>>>> SomeConstructor(Void arg) {
>>>>   if (MockitoBootHelper.THREAD_LOCAL.isMockInstantiatingMode()) {
>>>>     super(null); // any constructor of the super class with default
>>>> values
>>>> for all arguments
>>>>   } else {
>>>>     // original constructor code...
>>>>   }
>>>> }
>>>>
>>>> Before instantiating a mock, the thread local value that is held by a
>>>> bootstrap-loader injected class is set to true such that a side-effect
>>>> free
>>>> construction is achieved.
>>>>
>>>> This is of course too expensive and has negative effects on performance
>>>> due
>>>> to additional branching and JIT-byte code limits such that one would
>>>> rather
>>>> open jdk.internal.misc to access the Unsafe instantiation mechanism if a
>>>> Java agent is already available.
>>>>
>>>> However, mocking and serialization libraries are not typically loaded
>>>> as a
>>>> Java agent. Also, I think that the actual requirements are different. My
>>>> suggestion here is:
>>>>
>>>> a) For serialization libraries, I think that adding
>>>> MethodHandles.Lookup::newInstance(Class<? extends Serializable>) with
>>>> standard deserialization mechanics would be sufficient.
>>>> b) For mocking libraries, this does not suffice as mocks can be of any
>>>> class. I understand that this breaks encapsulation but for unit tests, I
>>>> argue that the benefit of using these libraries outweights the benefit
>>>> of
>>>> full encapsulation within a unit test.
>>>>
>>>> As Mockito is typically run within a build which uses a JDK, we could
>>>> attach to the current VM using the attachment API. Since Java 9, this
>>>> is no
>>>> longer possible to attach to the same VM but instead we start a helper
>>>> JVM
>>>> process that applies the attachment indirectly. Unfortunately, this is a
>>>> rather costly operation what is especially problematic when running a
>>>> single unit test. (The difference to this approach over Unsafe is about
>>>> half a second on average.)
>>>>
>>>> To overcome this, I would like to suggest to:
>>>>
>>>> a) Add a method Instrumentation::allocateInstance(Class). Java agents
>>>> can
>>>> already emulate this privilege as described above, this is therefore
>>>> merely
>>>> a convenience.
>>>> b) Add a module jdk.test to JDK 11 with a class
>>>> JavaTest::getInstrumentation that returns an instrumentation instance
>>>> for
>>>> the current JVM. This module should not be resolved by default but only
>>>> when requiring it explicitly similar to the EE modules after Java 9.
>>>>
>>>> I think this solution carries two benefits:
>>>>
>>>> a) Test libraries like Mockito can only be used in a testing scope. We
>>>> experience regularly that Mockito is used in production environments.
>>>> The
>>>> library is not meant for that and we warn and document that this is not
>>>> intended use but people regularly ignore this directive. By requiring
>>>> this
>>>> module, this form of abuse would no longer be possible and the JVM could
>>>> even give a meaningful error message if this use was intended.
>>>> b) The Instrumentation instance can be used for other meaningful testing
>>>> approaches. For example, Mockito offers an inline mock maker where the
>>>> mocking logic is inlined into a method body rather then creating a
>>>> subclass. This approach mainly targets final classes which have become
>>>> more
>>>> common especially since the Kotlin language became popular. To supply
>>>> this
>>>> alternative mock maker, Mockito attempts attaching an agent to the
>>>> current
>>>> VM (directly or indirectly, depending on the VM's version) which suffers
>>>> the additional costs for attaching an agent that I mentioned before.
>>>>
>>>> Thanks for reading this and I hope that you can consider these, as of
>>>> today, very practiced use cases. The JPMS migration has gone quite
>>>> well, I
>>>> find. With these outstanding problems, JDK 11 could be the first release
>>>> where a majority of Java programs would no longer need to rely on
>>>> internal
>>>> API.
>>>>
>>>> Best regards, Rafael
>>>>
>>>
>>>
>>
>