RFR: 8284877: Check type compatibility before looking up method from receiver's vtable [v2]

[David Holmes]

On Mon, 19 Jun 2023 01:53:47 GMT, Yi Yang <yyang at openjdk.org> wrote:

>>> Is it possible to add this sanity check as part of `-Xcheck:jni`? I agree that this check should not be added by default.
>> 
>> @dcubed-ojdk , @y1yang0 such a check is already part of `-Xcheck:jni`. See `src/hotspot/share/prims/jniCheck.cpp` and the `jniCheck::validate_call` method - it does a receiver typecheck:
>> 
>> if (obj != nullptr) {
>>     oop recv = jniCheck::validate_object(thr, obj);
>>     assert(recv != nullptr, "validate_object checks that");
>>     Klass* rk = recv->klass();
>> 
>>     // Check that the object is a subtype of method holder too.
>>     if (!rk->is_subtype_of(holder)) {
>>       ReportJNIFatalError(thr, fatal_wrong_class_or_method);
>>     }
>>   }
>
>> validate_call
> 
> This check is exercised when jni_CallStaticVoidMethod is called, while aforementioned case is that JNI wrongly constructs an object and does a normal virtual call and gets correct result.

@y1yang0  as far as I can see it is already being called via the following macro/wrapper:

#define WRAPPER_CallMethod(ResultType, Result) \
JNI_ENTRY_CHECKED(ResultType,  \
  checked_jni_Call##Result##Method(JNIEnv *env, \
                                   jobject obj, \
                                   jmethodID methodID, \
                                   ...)) \
    functionEnter(thr); \
    va_list args; \
    IN_VM( \
      jniCheck::validate_call(thr, nullptr, methodID, obj); \
    ) \
    va_start(args,methodID); \
    ResultType result =UNCHECKED()->Call##Result##MethodV(env, obj, methodID, \
                                                          args); \
    va_end(args); \
    thr->set_pending_jni_exception_check("Call"#Result"Method"); \
    functionExit(thr); \
    return result; \
JNI_END \

-------------

PR Comment: https://git.openjdk.org/jdk/pull/8241#issuecomment-1596406550