[ 
https://issues.apache.org/jira/browse/FELIX-2791?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Sahoo updated FELIX-2791:
-------------------------

    Attachment: patch.txt

PFA a patch that solves the issue. It introduces a barrier and a flag to make 
sure watcher threads don't run until fileinstall has finished initialisation. 
Unlike the previous patch, this one resets the boolean initialised field in 
stop(). Please review if you want to, as I plan to commit and make a new 
release of fileinstall ASAP. Thanks. 

> NPE in getStartLevel due to race condition
> ------------------------------------------
>
>                 Key: FELIX-2791
>                 URL: https://issues.apache.org/jira/browse/FELIX-2791
>             Project: Felix
>          Issue Type: Bug
>          Components: File Install
>    Affects Versions: fileinstall-3.1.4
>            Reporter: Sahoo
>            Assignee: Sahoo
>            Priority: Blocker
>             Fix For: fileinstall-3.1.6
>
>         Attachments: patch.txt
>
>
> We are using FileInstall 3.1.4 and one of our user is seeing the following 
> exception repeated in the directory watcher main loop continuously:
> java.lang.NullPointerException
>         at 
> org.apache.felix.fileinstall.internal.FileInstall.getStartLevel(FileInstall.java:283)
>         at 
> org.apache.felix.fileinstall.internal.FileInstall.getStartLevel(FileInstall.java:276)
>         at 
> org.apache.felix.fileinstall.internal.DirectoryWatcher.run(DirectoryWatcher.java:237)
> In main loop, we have serious trouble: java.lang.NullPointerException
> NPE comes from FileInstall.java:283, which looks like this:
>             return (StartLevel) startLevel.waitForService(timeout);
> It looks like a concurrency bug to me in FileInstall. The only explanation 
> that I have so far is that in the target environment, code is getting 
> reordered and hence directory watcher thread is spwaned before startLevel is 
> initialzed.. Since startLevel is not declared volatile, nor is it accessed 
> from a synchronized block, DirectoryWatcher thread continues to see the stale 
> value.
> So, I suggest we change startLevel to a volatile field. Any comments?

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to