[jboss-jira] [JBoss JIRA] (JASSIST-259) Concurrency issue in ProxyFactory

[Shigeru Chiba (JIRA)]

    [ https://issues.jboss.org/browse/JASSIST-259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150344#comment-13150344 ] 

Shigeru Chiba commented on JASSIST-259:
---------------------------------------

You're right.  I found [this commit|https://github.com/jboss-javassist/javassist/commit/bd1c47e0f555ada2d250a52f4a3b6697107486b0#diff-784aef550e16e266f5ecb6a3ce357fd1] introduced this bug.

A fix will be:

{{    private Class createClass1() {
        Class result = thisClass;
        if (result == null) {
            ClassLoader cl = getClassLoader();
            synchronized (proxyCache) {
                if (factoryUseCache)
                    createClass2(cl);
                else 
                    createClass3(cl);
                result = thisClass;
                // don't retain any unwanted references
                thisClass = null;
            }
        }

        return result;
    }}}

> Concurrency issue in ProxyFactory
> ---------------------------------
>
>                 Key: JASSIST-259
>                 URL: https://issues.jboss.org/browse/JASSIST-259
>             Project: Javassist
>          Issue Type: Bug
>    Affects Versions: 3.12.0.GA
>         Environment: AIX, IBM J9 VM (build 2.8, JRE 1.8.0 AIX ppc64-64 Compressed References 20150722_258693 (JIT enabled, AOT enabled)
>            Reporter: Christian Beikov
>            Assignee: Shigeru Chiba
>            Priority: Critical
>
> Look at the following line to understand the issue
> https://github.com/jboss-javassist/javassist/blob/master/src/main/javassist/util/proxy/ProxyFactory.java#L435
> If there are 2 threads T1 and T2 at this line, T1 pauses and T2 runs through, on some systems T1 may encounter thisClass to be null after it wakes up again.
> This leads to Hibernate getting a null class in JavassistProxyFactory and trying to create an instance of that which fails with a NPE.
> Can anyone please confirm this issue? Maybe there is something in the JMM that I am not aware of that makes this code safe, but as far as I can tell, this looks like a real bug that just happens to uncover itself on an environment like J9 on AIX.



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)