felix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sahoo (JIRA)" <j...@apache.org>
Subject [jira] Created: (FELIX-2748) Possible synchronization issue with BundleContext.getBundle()
Date Mon, 27 Dec 2010 16:30:47 GMT
Possible synchronization issue with BundleContext.getBundle()

                 Key: FELIX-2748
                 URL: https://issues.apache.org/jira/browse/FELIX-2748
             Project: Felix
          Issue Type: Bug
          Components: Framework
    Affects Versions: framework-3.0.6
            Reporter: Sahoo

On 12/26/10 23:24, Sahoo wrote:
> Hi,
> Looking at the code, it seems to me that BundleContext.getBundle() can return a bundle
even though the BundleContext has been invalidated. I say this based on the following reasoning:
> public BundleContext.getBundle()
> {
>         checkValidity();
>         return m_bundle;
> }
> private void checkValidity() // non-synchronized method
> {
>         if (m_valid) // m_valid is a non-volatile field
>         {
>             switch (m_bundle.getState())
>             {
>                 case Bundle.ACTIVE:
>                 case Bundle.STARTING:
>                 case Bundle.STOPPING:
>                     return;
>             }
>         }
>         throw new IllegalStateException("Invalid BundleContext.");
> }
> protected void invalidate()    {        m_valid = false;    }
> class Bundle  {
>   public int getState() // non-synchronized method
>     {
>         return m_state;
>     }
> }
> time t1: thread1 has access to bundleContext.
> time t2: thread2 calls b.update()
> time t3: bundle has been stopped and is now in STARTING state. thread1 calls bundleContext.getBundle().
> Since m_valid is not a volatile field, it can see m_valid as true, although thread2 would
have set it to false. Since m_state is volatile, it sees its value as STARTING. So, it fails
to detect invalidation of the bundle and is returned with the bundle object instead of IllegalStateException.
> Am I missing anything here?

No, I think you are correct.

I'll have to think about it a little more, but I think the real issue here is the fact that
the validity of the bundle context has to be more directly tied to the bundle. They can't
be treated as independent values, because even if they were both volatile or synchronized,
you could still end up with a change in state between checking each value.

I think the simpler BundleContext.checkValidity() method should be something like:

    m_bundle._getBundleContext() == this

Which says as long as "this" bundle context is the bundle's bundle context, then it is valid.
I think. I'll still need to think about it. :-)

Want to open a bug on this issue?

Is this a theoretical exercise or are you suffering from this?

-> richard

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

View raw message