[concurrency-interest] Here's why Atomic*FieldReference access checking is broken

David M. Lloyd david.lloyd at redhat.com
Sat Oct 4 11:14:16 EDT 2014


On 10/03/2014 10:15 PM, Alan Bateman wrote:
> On 03/10/2014 08:13, David M. Lloyd wrote:
>> :
>>
>> Why the problem occurs
>> ----------------------
>> The root of the problem traces back to
>> SecurityManager.checkMemberAccess().  This method is the one remaining
>> method in all of SecurityManager which uses the calling class context
>> (stack) in order to determine the nature of the access check that is
>> needed.
> Are you sure you see this in JDK 8 too? I ask because I remember David
> Holmes changed the Atomic*Updater methods to call getDeclaredField in a
> privileged block (JDK-7103570). Also there work in JDK 8 on caller
> sensitive methods (JEP 176). As part of this then SM.checkMemberAccess
> was deprecated and usages in the JDK dropped (Class.getDeclaredField and
> the others no longer use it).

Ah, I will double check.  I admit my JDK 8 checkout is (still) very old, 
and it appeared like the code was similar, and given the prior 
discussion it did not occur to me to check out more recent code.  But 
I'll check with a newer JDK 8 - if I can target the workaround, all the 
better.

Either way though I do feel that a JDK 7 backport is warranted, 
especially considering that Java EE 7 is the state of the art for EE 
right now.

-- 
- DML


More information about the Concurrency-interest mailing list