Nicolas Fränkel
Posted on April 4, 2021
A couple of years ago, I attended a talk of former colleague (but still friend) Volker Simonis. It gave me the idea to dig a bit into the subject of how to secure the JVM. From the material, I created a series of blog posts as well as a talk.
From that point on, I submitted the talk at meetups and conferences, where it was well-received. Because I like to explore different areas, I stopped to submit other proposals. Still, the talk is in my portfolio, and it was requested again in 2021. I have already presented it twice since the beginning of the year at the time of this writing.
It allowed me to update the demo with version 16 of the JDK. In this blog post, I want to share some findings regarding the security changes regarding changing a field's type across JDK versions.
Fun with JDK 8
Let's start with the JDK. Here's a quiz I show early in my talk:
Foo foo = new Foo();
Class<Foo> clazz = foo.getClass();
Field field = clazz.getDeclaredField("hidden");
Field type = Field.class.getDeclaredField("type");
AccessibleObject.setAccessible(
new AccessibleObject[]{field, type}, true);
type.set(field, String.class);
field.set(foo, "This should print 5!");
Object hidden = field.get(foo);
System.out.println(hidden);
class Foo {
private int hidden = 5;
}
Take some time to guess the result of executing this program when running it with a JDK 8.
Here's the relevant class diagram to help you:
As can be seen, Field
has a type
attribute that contains... its type. With the above code, one can change the type of hidden
from int
to String
so that the above code executes and prints "This should print 5!"
.
With JDK 16, the snippet doesn't work anymore. It throws a runtime exception instead:
Exception in thread "main" java.lang.NoSuchFieldException: type
at java.base/java.lang.Class.getDeclaredField(Class.java:2549)
at ch.frankel.blog.FirstAttempt.main(FirstAttempt.java:12)
The exception explicitly mentions line 12: Field.class.getDeclaredField("type")
. It seems as if the implementation of the Field
class changed.
Looking at the source code of JDK 16
Let's look at the source code in JDK 16:
public final class Field extends AccessibleObject implements Member {
private Class<?> clazz;
private int slot;
// This is guaranteed to be interned by the VM in the 1.4
// reflection implementation
private String name;
private Class<?> type; // 1
// ...
}
- Interestingly, the
field
type is there.
If the field is present, why do we get the exception? We need to dive a bit into the code to understand the reason.
Here's the sequence diagram of Class.getDeclaredField()
:
The diagram reveals two interesting bits:
- The
Reflection
class manages a cache to improve performance. - A field named
fieldFilterMap
filters out the fields that reflective access return.
Let's investigate the Reflection
class to understand the runtime doesn't find the type
attribute:
static {
fieldFilterMap = Map.of(
Reflection.class, ALL_MEMBERS,
AccessibleObject.class, ALL_MEMBERS,
Class.class, Set.of("classLoader", "classData"),
ClassLoader.class, ALL_MEMBERS,
Constructor.class, ALL_MEMBERS,
Field.class, ALL_MEMBERS, // 1
Method.class, ALL_MEMBERS,
Module.class, ALL_MEMBERS,
System.class, Set.of("security")
);
methodFilterMap = Map.of();
}
- All of the
Field
attributes are filtered out!
For this reason, none of the attributes of Field
are accessible via reflection!
An alternative way to change the type
Since version 9, the JDK offers a new API to access fields as part of the java.lang.invoke
package.
Here's a quite simplified class diagram focusing on our usage:
One can use the API to access the type
attribute as above. The code looks like the following:
var foo = new Foo();
var clazz = foo.getClass();
var lookup = MethodHandles.privateLookupIn(Field.class, MethodHandles.lookup());
var type = lookup.findVarHandle(Field.class, "type", Class.class);
var field = clazz.getDeclaredField("hidden");
type.set(field, String.class);
field.setAccessible(true);
field.set(foo, "This should print 5!");
var hidden = field.get(foo);
System.out.println(hidden);
But running the code yields the following:
Exception in thread "main" java.lang.IllegalArgumentException: Can not set int field ch.frankel.blog.Foo.hidden to java.lang.String
at java.base/jdk.internal.reflect.UnsafeFieldAccessorImpl.throwSetIllegalArgumentException(UnsafeFieldAccessorImpl.java:167)
at java.base/jdk.internal.reflect.UnsafeFieldAccessorImpl.throwSetIllegalArgumentException(UnsafeFieldAccessorImpl.java:171)
at java.base/jdk.internal.reflect.UnsafeIntegerFieldAccessorImpl.set(UnsafeIntegerFieldAccessorImpl.java:98)
at java.base/java.lang.reflect.Field.set(Field.java:793)
at ch.frankel.blog.FinalAttempt.main(FinalAttempt.java:16)
Though the code compiles and runs, it throws at field.set(foo, "This should print 5!")
. We reference the type
field and can change it without any issue, but it still complains.
The reason lies in the last line of the getDeclaredField()
method:
public Field getDeclaredField(String name)
throws NoSuchFieldException, SecurityException {
Objects.requireNonNull(name);
SecurityManager sm = System.getSecurityManager();
if (sm != null) {
checkMemberAccess(sm, Member.DECLARED, Reflection.getCallerClass(), true);
}
Field field = searchFields(privateGetDeclaredFields(false), name);
if (field ## null) {
throw new NoSuchFieldException(name);
}
return getReflectionFactory().copyField(field); // 1
}
- Return a copy of the
Field
object, not theField
itself.
Since the JDK code returns a copy of the field, the change happens on this copy, and we cannot change the original field's type.
Conclusion
Though Java touts itself as a statically-typed language, version 8 of the JVM allows us to change the type at runtime dynamically. One of my favorite jokes during the talk mentioned above is that though we have learned that Java is statically-typed, it is dynamically-typed in reality.
We can track the change precisely in Java 12: the version 11 of the Reflection
class shows a basic fieldFilterMap
; the version 12 shows a fully-configured one. Hence, if you want to avoid nasty surprises, you should upgrade to the latter, if not the latest.
To go further:
- Focus on JVM Security
- What does the sun.reflect.CallerSensitive annotation mean?
- Secure Coding Guidelines for Java SE
Orginally published at A Java Geek on April 4th, 2021
Posted on April 4, 2021
Join Our Newsletter. No Spam, Only the good stuff.
Sign up to receive the latest update from our blog.