Valentina Casola » 9.Java security architecture

Components of Java

The development environment:

• development lifecycle;
• Java language features;
• class files and bytecode.

The execution environment:

• the Java Virtual Machine (JVM).

Interfaces and architectures:

• e.g., Java beans, RMI, JDBC, etc.

The Java Virtual Machine (JVM)

JVM components

Class loaders:

• locate and load classes into the JVM;
• primordial class loader:
  • loads trusted classes (system classes found on the boot class path);
  • integral part of the JVM.

Class loader instances:

• instances of java.net.URLClassLoader (which extends SecureClassLoader);
• loads untrusted classes from the local file system or from the network and passes them to the class file verifier;
• application developers can implement their own class loaders.

Class file verifier

• checks untrusted class files:
  • size and structure of the class file;
  • bytecode integrity (references, illegal operations, …);
  • some run-time characteristics (e.g., stack overflow).
• a class is accepted only if it passes the test.

JVM components

Native method loader:

• native methods are needed to access some of the underlying operating system functions (e.g., graphics and networking features);
• once loaded, native code is stored in the native method area for easy access.

The heap:

• memory used to store objects during execution;
• how objects are stored is implementation specific
  execution engine;
• a virtual processor that executes bytecode;
• has virtual registers, stack, etc.;
• performs memory management, thread management, calls to native methods, etc.

JVM components

Security Manager:

• enforces access control at run-time (e.g., prevents applets from reading or writing to the file system, accessing the network, printing, …);
• application developers can implement their own Security Manager.

or use the policy based SM implementation provided by the JDK

JIT – Just In Time compiler:

• performance overhead due to interpreting bytecode;
• translates bytecode into native code on-the-fly.

• works on a method-by-method basis;
• the first time a method is called, it is translated into native code and stored in the class area;
• future calls to the same method run the native code.

all this happens after the class has been loaded and verified.

Java security models

The need for Java security.
The sandbox (Java 1.0).
The concept of trusted code (Java 1.1).
Fine grained access control (Java 2).

The need for Java security

Code mobility can be useful (though not indispensable):

• may reduce bandwidth requirements;
• improve functionality of web services.

but downloaded executable content is dangerous:

• the source may be unknown hence untrusted;
• hostile applets may modify or destroy data in your file system;
• hostile applets may read private data from your file system;
• hostile applets may install other hostile code on your system (e.g., virus, back-door, keyboard sniffer, …);
• hostile applets may try to attack someone else from your system (making you appear responsible for the attack);
• hostile applets may use (up) the resources of your system (DoS).

all this may happen without you knowing about it.

J2SE

Java 2: Each Class Has A ProtectionDomain

The three pillars of Java security

The Security Manager;
Class loaders;
The bytecode verifier.

……. Further readings…..

The Security Manager

Ensures that the permissions specified in the policy file are not overridden.
Implements a checkPermission() method, which:

• takes a permission object as parameter, and;
• returns a yes or a no (based on the code source and the permissions granted for that code source in the policy file).

CheckPermission() is called from trusted system classes:

• e.g., if you want to open a socket you need to create a Socket object;
• the Socket class is a trusted system class that always invokes the checkPermission() method.

This requires that:

• all system resources are accessible only via trusted system classes;
• trusted system classes cannot be overwritten (ensured by the class loading mechanism).

The Security Manager (cont’d)

the JVM allows only one SM to be active at a time
there is a default SM provided by the JDK
Java programs (applications, applets, beans, …) can replace the default SM by their own SM only if they have permission to do so

• two permissions are needed:
  • create an instance of SecurityManager;
  • set an SM instance as active.

example:

grant CodeBase “…”, SignedBy “…” {

permission java.lang.RuntimePermission “createSecurityManager”;

permission java.lang.RuntimePermission “setSecurityManager”;};

invoking the SecurityManager constructor or the setSecurityManager() method will call the checkPermissions() method of the current SM and verify if the caller has the necessary permissions.

Class loaders

Separate name spaces

• classes loaded by a class loader instance belong to the same name space;
• since classes with the same name may exist on different Web sites, different Web sites are handled by different instances of the applet class loader;
• a class in one name space cannot access a class in another name space.

→ classes from different Web sites cannot access each other.

establish the protection domain (set of permissions) for a loaded class;
enforce a search order that prevents trusted system classes from being replaced by classes from less trusted sources
… see next two slide …

Class loading process

When a class is referenced

• JVM: invokes the class loader associated with the requesting program;
• class loader: has the class already been loaded?

yes:

• does the program have permission to access the class?
  • yes: return object reference
  • no: security exception

no:  does the program have permission to create the requested class?

yes:

• first delegate loading task to parent;
• if parent returns success, then return (class is loaded);
• if parent returned failure, then load class and return.

no: security exception.

Class loading task delegation

Byte code verifier

Performs static analysis of the bytecode
syntactic analysis:

• all arguments to flow control instructions must cause branches to the start of a valid instruction;
• all references to local variables must be legal;
• all references to the constant pool must be to an entry of appropriate type;
• all opcodes must have the correct number of arguments;
• exception handlers must start at the beginning of a valid instruction;
• …

Data flow analysis:

• attempts to reconstruct the behavior of the code at run time without actually running the code;
• keeps track only of types not the actual values in the stack and in local variables.

it is theoretically impossible to identify all problems that may occur at run time with static analysis