Security Requirements on Activation Group Descriptors used with phoenix or rmid

Some Jini network technology-based services may be activatable and as such will require the use of an activation daemon. The standard Java(TM) 2 Standard Edition (J2SE(TM)) Remote Method Invocation (Java RMI) activation daemon is rmid. The JGDMS release provides phoenix, an alternative configurable implementation of a Java RMI activation daemon.

This note describes the security requirements to specify an activation group descriptor (of type java.rmi.activation.ActivationGroupDesc) for use with activatable objects handled by either phoenix or rmid.

Note: We encourage developers to transition from rmid and to use its configurable alternative phoenix instead. The phoenix activation daemon supports configurable exporters for the remote objects that are part of the activation system and also supports flexible access control policies for the activation system. See the org.apache.river.phoenix package documentation for details on configuring and starting phoenix.

Introduction and Rationale

An activation daemon, such as phoenix or rmid, executes subprocesses which run virtual machines for the Java platform (VMs). It does this to keep activation groups separate from one another, and to allow the use of special VM engines and options for different activation groups.

An activation group is specified by an activation group descriptor that describes the information necessary for an activation daemon to start the activation group. An application can register an activation group descriptor (of type ActivationGroupDesc) directly with the ActivationSystem of either phoenix or rmid by calling the ActivationSystem's registerGroup method. Alternatively, an activation group can be registered via the service starter API, by invoking the create method on a org.apache.river.start.SharedActivationGroupDescriptor instance. As a side effect of this invocation, an ActivationGroupDesc is created and registered with the ActivationSystem as above.

An activation group descriptor used to start an activation group affects the subprocess environment, command, command options, and arguments for starting the activation group, and is flexible enough that with an insecure security policy, any command, even a non-Java-platform command or program, can be run under the guise of starting the group. There is clearly a need for a security control to prevent unauthorized specification of an activation group descriptor and hence unauthorized execution of a command. Note that whether an activation group is registered directly with the ActivationSystem or registered indirectly as a result of the SharedActivationGroupDescriptor.create method, the security requirements outlined below are the same for each situation.

The activation daemons phoenix and rmid have different schemes for preventing an unauthorized activation group descriptor from being used. The phoenix activation daemon may be configured with an access control policy on its ActivationSystem that prevents an unauthorized ActivationGroupDesc from either being registered via the ActivationSystem.registerGroup method or being modified via the ActivationSystem.setActivationGroupDesc method. The rmid activation daemon has a slightly different control point: it verifies that an ActivationGroupDesc is authorized (according to rmid's security policy) just before using it to create an activation group. The details of each access control model are discussed in turn below. Although phoenix and rmid have different points in which they check permissions for such access, the basic permission requirements are the same. Also, the permission types that phoenix uses are modeled after the permission types used by rmid (ExecPermission and ExecOptionPermission); each permission type has the same class name, but is in a different package.

---

phoenix

Since applications need to configure the access control policy for registering and updating an activation group descriptor with the ActivationSystem, the phoenix implementation includes a special net.jini.export.Exporter implementation for the ActivationSystem, org.apache.river.phoenix.SystemAccessExporter, that allows the ActivationSystem's access control policy to be configured. The phoenix implementation also includes a net.jini.jeri.InvocationLayerFactory implementation, org.apache.river.phoenix.SystemAccessILFactory, to use with net.jini.jeri.BasicJeriExporter. This InvocationLayerFactory implementation supports the same configurable access control policy for the ActivationSystem.

By default, phoenix's ActivationSystem is exported via BasicJeriExporter with a SystemAccessILFactory that produces a net.jini.jeri.Dispatcher that enforces an access control policy on calls to the registerGroup and setActivationGroupDesc methods. The default access control policy is specified by the org.apache.river.phoenix.DefaultGroupPolicy.checkGroup method.

This default group policy requires the following when an ActivationGroupDesc is being recorded via the registerGroup or setActivationGroupDesc method:

• The group class name in the descriptor is either null (indicating the default) or specifically org.apache.river.phoenix.ActivationGroupImpl.
• For each property in the descriptor's property overrides, the calling context is granted the permission org.apache.river.phoenix.ExecOptionPermission with a target of the form "-Dname=value" (where name is the name of the property and value is the value of the property).
• If the descriptor contains a non-null command environment, the calling context is granted the permission org.apache.river.phoenix.ExecPermission with the command path as the target. Also, for each command option, the calling context is granted ExecOptionPermission with the option as a target.

These permissions must be specified in phoenix's security policy file, indicated on the command line when starting up phoenix:

java -J-Djava.security.policy=configDir/phoenix.policy  
     -J-Djava.rmi.server.codebase=codebasePath [otherOptions]
     -jar installDir/lib/phoenix.jar configOptions

where configDir is the directory containing phoenix's security policy file, codebasePath is a codebase path (space-separated list of URLs that serves up phoenix's download JAR file (installDir/lib/phoenix-dl.jar), otherOptions are any other standard options for the java command, and configOptions are the options (typically a filename or URL) of a phoenix configuration.

Some things to keep in mind when defining permissions for phoenix are:

• Although in many cases all or part of the command line will be case insensitive (e.g. host names, files, and command paths on some operating systems), the checks ExecPermission and ExecOptionPermission perform are case sensitive, so it is important to make sure the case of the targets of the ExecPermissions and ExecOptionPermissions match the case of the command line elements in the ActivationGroupDesc.CommandEnvironment.
• Because phoenix can be configured to support authentication and access control, it may be reasonable to grant more broad ExecPermissions and ExecOptionPermissions to a narrow set of trusted subjects.

Note that the permission class names ExecPermission and ExecOptionPermission are the same as those used for rmid's access control policy except for the package name, which, for phoenix, is org.apache.river.phoenix. The examples given for rmid below are the same as for phoenix except that the package name for the permissions needs to be changed to org.apache.river.phoenix.

---

rmid

In releases of the Java 2 SDK prior to 1.2.2_006, the need to control the security of an ActivationGroupDesc was satisfied by the same stopgap measure used for rmiregistry access control: if the caller was running on the same host as the remote object, then all access was allowed; otherwise, no access to sensitive functions was allowed. Registering or changing an ActivationGroupDesc is an example of a sensitive function; activating an object is an example of a non-sensitive function that anyone can perform. This security mechanism was predicated on the assumption that unprivileged code is not allowed to connect to port 1098 on the executing host.

However, that assumption is no longer as certain, and a stronger security model has been developed. In this new model, an extra security control has been added: before being executed, the command and options requested by an ActivationGroupDesc are checked by the security manager of rmid, which then references the security policy file that was set for rmid from the command line (for authorization).

An "exec policy" is a Java class employed by the Java RMI activation daemon to check commands and command-line options used to launch a VM when deciding whether or not an ActivationGroupDesc is authorized. The exec policy is adopted when rmid is started. The value of the sun.rmi.activation.execPolicy property dictates the policy that rmid uses to determine whether or not the information in an ActivationGroupDesc may be used to launch a VM for an activation group. Some exec policies are built into rmid, and rmid can be instructed to use one of them by defining a property that specifies a built-in policy instead of the exec policy's class name.

In the exec policy represented by the value none (discussed later, in the section, How to achieve the behavior of rmid released in previous versions of the Java 2 SDK), the new security control is disabled, and execution is compatible with the rmid found in previous releases of the Java 2 SDK.

In the default exec policy (discussed in the section, How to assign and maintain security controls), the authorized commands and options are encoded in the security policy file, which gets set using the "-J-Djava.security.policy=..." option on the rmid command line. Each command and option string in each ActivationGroupDesc is separately authorized or unauthorized. While there is no authentication of a caller (an authorized option is authorized for everyone's use), there is still a check to ensure that the registration of the ActivationGroupDesc originates from the local host.

The exec policy can be set to the name of another class that rmid will instantiate; the exec policy object will then be expected to perform the security check on ActivationGroupDescs at group activation time. This option is the most flexible, but discussion of its use is beyond the scope of this document. See the rmid tool documentation (manpage) for details, at:

     http://java.sun.com/j2se/1.4/docs/tooldocs/solaris/rmid.html

and

     http://java.sun.com/j2se/1.4/docs/tooldocs/windows/rmid.html

How to achieve the behavior of rmid released in previous versions of the Java 2 SDK

The new security control can be effectively bypassed by assigning an exec policy that permits all commands and all options. There is a short way to specify this behavior by setting the exec policy value to none:

   rmid -J-Dsun.rmi.activation.execPolicy=none

which will run rmid in the normal way, but will automatically authorize all ActivationGroupDescs registered from the local host.

For the user who can control the rmid host to guarantee the assumptions of Java 2 SDK, v1.2.x, or who is simply impatient with the new security control, the none keyword is a convenient way to get rmid up and running quickly. Remember that this is a way to sacrifice security for temporary convenience, and therefore is not recommended.

How to assign and maintain security controls

The default exec policy is used if the sun.rmi.activation.execPolicy system property is not set for the rmid tool, or if the value of the execPolicy property is set to default. In this case, a security policy file must be specified on rmid's command line.

In the default case, the java.security.policy system property must be set for rmid. For example on a Microsoft Windows platform:

   rmid -J-Djava.security.policy=configDir\rmid.policy

The file named as the value of this property should contain a Java 2 platform security policy file. An example which has appropriate syntax for the Microsoft Windows platform is:

   grant {
     // permissions granted to everyone
     permission com.sun.rmi.rmid.ExecPermission "C:\\jdk\\bin\\java_g";
     permission com.sun.rmi.rmid.ExecOptionPermission "-Djava.security.policy=configDir\\lookup.policy";
   };

The two permission types shown above, ExecPermission and ExecOptionPermission, are the only two security permissions relevant to the new security control. Notice that they are granted to all codebases (grant {): you should always be cautious when granting permissions to all codebases, because these permissions are also granted to unknown downloaded code. See the rmid reference pages for a complete description of the syntax for these permissions.

The straightforward way to build up an effective rmid.policy is:

1. Start with an empty rmid.policy.
2. Run rmid -J-Djava.security.policy=rmid.policy.
3. Run a client program to activate an object (you might need to register the object if you haven't already).
4. Catch and printStackTrace the AccessControlException that gets thrown by the activation attempt (usually the client program will do this anyway).
   

   Note on restartable Activatable objects: Some activatable objects are activated automatically by rmid (if ActivationDesc.getRestartMode() is true); in this case, rmid will printStackTrace the exception to its System.err, preceded by the English-language message "rmid: unable to restart service".

5. Look in the AccessControlException to see the failed exception. For example, the exception:

   java.security.AccessControlException: access denied (com.sun.rmi.rmid.ExecOptionPermission -cp)

   means that the permission you need to add to the security policy file is:

      grant {
        permission com.sun.rmi.rmid.ExecOptionPermission "-cp";
      };
   

6. Kill the rmid process (rmid -stop) and return to step 2.

Specific examples for some River technology-enabled service implementations

Here's an example of an rmid.policy which will run Reggie (the contributed River lookup service implementation from Sun Microsystems) in the Solaris(TM) Operating System (Solaris OS):

  grant {
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.security.policy=configDir/lookup.policy";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.rmi.server.codebase=*";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-cp";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "installDir/lib/reggie.jar";
  };

Here's an example to run Mahalo (the contributed River transaction manager service implementation from Sun Microsystems) in the Solaris OS:

  grant {
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.security.policy=configDir/txn.policy";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.rmi.server.codebase=*";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-cp";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "installDir/lib/mahalo.jar";
  };

Here's one for FrontEndSpace (a contributed implementation of JavaSpaces technology from Sun Microsystems) in the Solaris OS:

  grant {
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.security.policy=configDir/books.policy";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Djava.rmi.server.codebase=*";
    permission com.sun.rmi.rmid.ExecOptionPermission
	       "-Xbootclasspath/p::installDir/lib/outrigger.jar:installDir/lib/transient-outrigger.jar:installDir/lib/pro.zip";
  };

http://www.apache.org/licenses/LICENSE-2.0