Re: RFR(xs): 8181419: Race in jdwp invoker handling may lead to crashes or invalid results

Thu, 15 Jun 2017 11:01:38 -0700

Hi Thomas,

Sur.
I'll also run some internal jdi tests.

Thanks,
Serguei


On 6/13/17 21:17, Thomas Stüfe wrote:
Hi Sergey,

thank you for the review! 

I ran jdi jtreg tests on both AIX and Linux x64, I had some issues but they seemed preexisting (same issues without my patch). Could you please also run tests you think are necessary on your platforms? Thanks!

best Regards, Thomas

On Wed, Jun 14, 2017 at 1:36 AM, serguei.spit...@oracle.com <serguei.spit...@oracle.com> wrote:
On 6/13/17 14:16, serguei.spit...@oracle.com wrote:
Hi Thomas,

The fix looks great to me.

 Forgot to say that the copyright comment need an update.

Thanks,
Serguei


Thank you for identifying and fixing the problem!

Thanks,
Serguei


On 6/13/17 06:55, Thomas Stüfe wrote:
Ping... Anyone?

On Thu, Jun 1, 2017 at 2:18 PM, Thomas Stüfe <thomas.stu...@gmail.com> wrote:
Hi all,

please take a look at this proposed fix for a theoretical race in the jdwp library.


In short, this is an addition to Severin's fix to the jdwp invoke handling (https://bugs.openjdk.java.net/browse/JDK-8153711).

We have a potential race condition where the delayed cleanup of the saved returnvalue object reference and the exception reference (released in deletePotentiallySavedGlobalRefs() ) may be overtaken by a new request which populates the thread request structure anew. If this happens, deletePotentiallySavedGlobalRefs() may actually release the return value / exception references of the follow up request, if that one was already processed.

The solution I choose is safe and conservative. We still release both references, but use the locally saved JNI references. We just avoid accessing the thread local request structure after it has been cleared for reuse. This keeps timing and locking behaviour unchanged.

I am currently running jtreg tests for com/sun/jdi on AIX and Linux.

Kind Regards, Thomas





Reply via email to