I asked a number of times for a security audit to be made of the security implications of JPA and it was never taken up. Most of the vendors make extensive use of privileged operations including getting system properties, reflection, and class loader operations and IMHO not enough attention was paid to getting it right.

That includes us. To be perfectly frank, in all the years of Kodo and OpenJPA, it has simply never been an issue.

Which doesn't mean that we shouldn't endeavor to fix the problems and do it all the right way, security-wise. I'm just pointing out that the demand for well-behaved security-sensitive applications is not as high as one might expect.



On May 23, 2007, at 9:43 AM, Craig L Russell wrote:

On May 22, 2007, at 7:22 PM, Kevin Sutter wrote:

Here's my take (just to generate some discussion)...

Right now, it doesn't seem like OpenJPA is ready for Java 2 Security.

It's a bug that OpenJPA doesn't use doPrivileged blocks around security-protected APIs.

Can you file a JIRA issue with as many of these cases as you can find? They need to be fixed.

As
Albert has pointed out, there only seems to be two places in the code where doPriv blocks exist. It would seem that any application-managed path that would attempt to access secure operations would require a doPriv block.

Right.

This may also apply to the container-managed paths, but more than likely, these paths have some type of application-server wrapper around the OpenJPA
objects and they could do the proper doPrivs.

Doubtful. It would be a bug for trusted code to wrap a call to untrusted code in a doPrivileged block. That would defeat the architecture of the security model. Rather, each component that needs to access secure resources needs to wrap the call in doPrivileged blocks.

It would also seem that we
would need to provide instructions for proper updating of the policy files
(for both the application-managed and container-managed scenarios).

Correct.

I know we're hitting these type of problems in the WebSphere environment. I
would be surprised if other app servers won't be experiencing similar
problems if Java 2 security is turned on. We're just trying to figure out
the who's responsible for what processing.

I've been through this exercise with JDO and put doPrivileged blocks around everything that needed it. It turned out to be about 40 places in the code. Not a big deal.

Patrick, were there any discussions on the expert group concerning the
relationship between JPA and the Java 2 Security?

I asked a number of times for a security audit to be made of the security implications of JPA and it was never taken up. Most of the vendors make extensive use of privileged operations including getting system properties, reflection, and class loader operations and IMHO not enough attention was paid to getting it right.

Craig

Thanks,
Kevin

On 5/17/07, Albert Lee <[EMAIL PROTECTED]> wrote:

I ran into the following exception when I enabled Java 2 security in the
Java EE environment using openjpa in the WebSphere environment:

java.security.AccessControlException: Access denied (
java.lang.RuntimePermission getClassLoader)
    at java.security.AccessController.checkPermission(
AccessController.java
:104)
at java.lang.SecurityManager.checkPermission (SecurityManager.java:547)
    at com.ibm.ws.security.core.SecurityManager.checkPermission(
SecurityManager.java:189)
    at java.lang.Thread.getContextClassLoader(Thread.java:490)
    at org.apache.openjpa.lib.conf.Configurations.findDerivedLoader(
Configurations.java:232)
    at org.apache.openjpa.lib.conf.Configurations.newInstance(
Configurations.java:194)
    at org.apache.openjpa.lib.conf.ObjectValue.newInstance(
ObjectValue.java
:103)
    at org.apache.openjpa.lib.conf.PluginValue.instantiate(
PluginValue.java
:101)
    at org.apache.openjpa.lib.conf.ObjectValue.instantiate(
ObjectValue.java
:79)
    at

org.apache.openjpa.conf.OpenJPAConfigurationImpl.getDataCacheManager Instance
(OpenJPAConfigurationImpl.java:583)
    at org.apache.openjpa.kernel.AbstractBrokerFactory.newBroker(
AbstractBrokerFactory.java:169)
    at org.apache.openjpa.kernel.DelegatingBrokerFactory.newBroker(
DelegatingBrokerFactory.java:142)
    at

org.apache.openjpa.persistence.EntityManagerFactoryImpl.createEntity Manager
(
EntityManagerFactoryImpl.java:190)
    at
com.ibm.websphere.ejb3sample.counter.StatelessCounterBean.getTheValu e
(StatelessCounterBean.java:63)

The scenario is a openjpa entity manager factory is injected to a
stateless
session bean and it is trying to create an EntityManager from the factory. Since the factory is directly injected in the application, the container
has
no involvment in handling the AccessController.doPrivileged(). Another similiar scenario is Persistence.createEntityManagerFactory() is called
from
within a stateless session bean, in which a similiar but different
security
related symptom is surfaced. These tests run successfully when Java 2 security is disabled. A security policy has put in place in the app server
to give all permissions to the openjpa jar files in the app server.

For experimentation, I add a doPrivilege block in the
Configurations.findDerivedLoader where the above exception took place and
I
was able to by-pass the failure and the doPriv seems to work. However I
went
into the same exception in different places when getSystemClassLoader()
and
other privileged operations are used.

Questions:
1) How is security being handled in openjpa or JPA in general?
2) What is the philosphy of putting doPrivilege construct around security sensitive code in openjpa? I only find 2 instances of doPrivilege usage in
openJPA.
3) Who is responsible to define and enable security in a app server
environment?
4) Is injecting a provider entity manager factory to user code an valid
procedure? I understand EntityManager proxy/wrapper is needed for
persistence context injection but I see no reason why provider's entity
manager factory can not be injected to user code.

Am I way off base regarding security in OpenJPA and/or JPA in general?

Any insights into this topics is greatly appreciated.
Thanks.
Albert Lee.


Craig Russell
Architect, Sun Java Enterprise System http://java.sun.com/products/jdo
408 276-5638 mailto:[EMAIL PROTECTED]
P.S. A good JDO? O, Gasp!


Reply via email to