Hi Chris, I agree that resetting the TCCL makes sense, and looks like the minimal safer thing to do here. InnocuousThread should probably be considered for the future, but that's a first step in this direction :-)
best regards, -- daniel On 18/08/16 08:58, Chris Hegarty wrote:
The SunPKCS11 poller thread has no need of any user defined class loader, so should set its context class loader to null before starting, so as to not inadvertently retain a reference to the creating thread’s context class loader. In other areas that suffered from a similar issue we changed to use an InnocuousThread, but I cannot fully satisfy myself that this is a safe substation here, so I opted for the safest minimal fix. A future refactoring exercise should consider using InnocuousThread. diff -r 92c31ec731eb src/jdk.crypto.pkcs11/share/classes/sun/security/pkcs11/SunPKCS11.java --- a/src/jdk.crypto.pkcs11/share/classes/sun/security/pkcs11/SunPKCS11.java Wed Aug 10 11:54:12 2016 +0100 +++ b/src/jdk.crypto.pkcs11/share/classes/sun/security/pkcs11/SunPKCS11.java Wed Aug 10 16:32:41 2016 +0100 @@ -809,20 +809,21 @@ } } // create the poller thread, if not already active private void createPoller() { if (poller != null) { return; } final TokenPoller poller = new TokenPoller(this); Thread t = new Thread(null, poller, "Poller " + getName(), 0, false); + t.setContextClassLoader(null); t.setDaemon(true); t.setPriority(Thread.MIN_PRIORITY); t.start(); this.poller = poller; } // destroy the poller thread, if active private void destroyPoller() { if (poller != null) { poller.disable(); https://bugs.openjdk.java.net/browse/JDK-8156841 -Chris.
