RFR: JDK-8068721 - RMI-IIOP communication fails when ConcurrentHashMap is passed to remote method

[Mark Sheppard]

Thanks Chris
I can include the proposed changes if you wish.
my original fix had similar PutField change to writeObject, but I left 
it from this review so
that it could be addressed separately.

regards
Mark

On 24/03/2015 09:38, Chris Hegarty wrote:
> Martin,
>
> On 23 Mar 2015, at 22:15, Martin Buchholz <martinrb at google.com> wrote:
>
>> Let us know if the serialization code of the collection classes can be
>> improved.
> I think there are a few minor cleanups that would be beneficial in CHM:
>
>   1) Add @serialField so that the serializable stream fields show up in the
>       javadoc ( serial form ), since they are still part of the serial form, even
>       though not used in the implementation. This is just documenting
>       existing behaviour/form.
>
>   2) Mark correctly identified a small hole in the putFields() spec, which
>       should be fixed. A minor, benign, change in CHM writeObject
>       can avoid this ( spec ambiguity ).
>
> Please consider the following change:
>
> diff --git a/src/java.base/share/classes/java/util/concurrent/ConcurrentHashMap.java b/src/java.base/share/classes/java/util/concurrent/ConcurrentHashMap.java
> --- a/src/java.base/share/classes/java/util/concurrent/ConcurrentHashMap.java
> +++ b/src/java.base/share/classes/java/util/concurrent/ConcurrentHashMap.java
> @@ -35,6 +35,7 @@
>   
>   package java.util.concurrent;
>   
> +import java.io.ObjectOutputStream;
>   import java.io.ObjectStreamField;
>   import java.io.Serializable;
>   import java.lang.reflect.ParameterizedType;
> @@ -599,7 +600,12 @@
>       /** Number of CPUS, to place bounds on some sizings */
>       static final int NCPU = Runtime.getRuntime().availableProcessors();
>   
> -    /** For serialization compatibility. */
> +    /**
> +     * For serialization compatibility.
> +     * @serialField segments Segment[]
> +     * @serialField segmentMask int
> +     * @serialField segmentShift int
> +     */
>       private static final ObjectStreamField[] serialPersistentFields = {
>           new ObjectStreamField("segments", Segment[].class),
>           new ObjectStreamField("segmentMask", Integer.TYPE),
> @@ -1400,9 +1406,10 @@
>               new Segment<?,?>[DEFAULT_CONCURRENCY_LEVEL];
>           for (int i = 0; i < segments.length; ++i)
>               segments[i] = new Segment<K,V>(LOAD_FACTOR);
> -        s.putFields().put("segments", segments);
> -        s.putFields().put("segmentShift", segmentShift);
> -        s.putFields().put("segmentMask", segmentMask);
> +        ObjectOutputStream.PutField streamFields = s.putFields();
> +        streamFields.put("segments", segments);
> +        streamFields.put("segmentShift", segmentShift);
> +        streamFields.put("segmentMask", segmentMask);
>           s.writeFields();
>   
>           Node<K,V>[] t;
>
> -Chris.
>
>> On Mon, Mar 23, 2015 at 2:25 PM, Mark Sheppard <mark.sheppard at oracle.com>
>> wrote:
>>
>>> Hi
>>> please oblige and review the following fix
>>>
>>> http://cr.openjdk.java.net/~msheppar/8068721/jdk9/corba/webrev/
>>> http://cr.openjdk.java.net/~msheppar/8068721/jdk9/jdk/webrev/
>>>
>>>
>>> which addresses the issue in
>>> https://bugs.openjdk.java.net/browse/JDK-8068721
>>>
>>> This relates to RMI-IIOP and the interplay between custom marshalling of
>>> ValueTypes and
>>> the corresponding serialization of a Java class, in this case
>>> ConcurrentHashMap.
>>>
>>> ConcurrentHashMap changed its structure in jdk8 from that used in jdk7.
>>> This resulted in modification to the readObject and writeObject methods,
>>> and in particular, former serial fields were removed, resulting in
>>> writeObject using PutField and readObject using defaultReadObject.
>>> The writeObject invokes the putFields method of an ObjectOutputStream
>>> multiple times, and assumes
>>> that it will receive the same PutField object instance for each
>>> invocation. The spec
>>> doesn't endorse this behaviour - but that's what the implementation of
>>> ObjectOutputStream
>>> provides. However in the case of RMI-IIOP, the OutputStreamHook, a
>>> subclass of ObjectOutputStream, returns a new instance for each
>>> putFields invocation. Thus, the ConcurrentHashMap writeObject results in
>>> improper serialization in the context
>>> of RMI-IIOP.
>>> In the unmarshalling flow of ConcurrentHashMap, the readObject now uses
>>> defaultReadObject rather than GetField
>>> In this call flow the IIOPInputStream attempts to ignore any primitive
>>> field, in the serialized data, that doesn't
>>> correspond with a reflective field of the object. However, it leaves the
>>> primitive in the stream.
>>> Thus, in the case of ConcurrentHashMap, which has serialized two integers
>>> and
>>> an array of Segments (subclass of ReentrantLock), this results in erroneous
>>> deserialization, with a misinterpretation of a value tag in the stream as
>>> an array length
>>> and an attempt to allocate a very large array ensues, with an exception
>>> being thrown.
>>>
>>> The OutputStreamHook now returns the same instance of PutField allocated
>>> for each separate call of putFields method.
>>> This highlights a need to tighten up and disambiguate the
>>> OutputObjectStream spec for putFields.
>>>
>>> IIOPInputStream now removes the primitive values from the stream.
>>>
>>> regards
>>> Mark
>>>