Domain Object Security (ACLs)

17.1 Overview

Complex applications often will find the need to define access permissions not simply at a web request or method invocation level. Instead, security decisions need to comprise both who (Authentication), where (MethodInvocation) and what (SomeDomainObject). In other words, authorization decisions also need to consider the actual domain object instance subject of a method invocation.

Imagine you're designing an application for a pet clinic. There will be two main groups of users of your Spring-based application: staff of the pet clinic, as well as the pet clinic's customers. The staff will have access to all of the data, whilst your customers will only be able to see their own customer records. To make it a little more interesting, your customers can allow other users to see their customer records, such as their "puppy preschool" mentor or president of their local "Pony Club". Using Spring Security as the foundation, you have several approaches that can be used:

  1. Write your business methods to enforce the security. You could consult a collection within the Customer domain object instance to determine which users have access. By using the SecurityContextHolder.getContext().getAuthentication(), you'll be able to access the Authentication object.

  2. Write an AccessDecisionVoter to enforce the security from the GrantedAuthority[]s stored in the Authentication object. This would mean your AuthenticationManager would need to populate the Authentication with custom GrantedAuthority[]s representing each of the Customer domain object instances the principal has access to.

  3. Write an AccessDecisionVoter to enforce the security and open the target Customer domain object directly. This would mean your voter needs access to a DAO that allows it to retrieve the Customer object. It would then access the Customer object's collection of approved users and make the appropriate decision.

Each one of these approaches is perfectly legitimate. However, the first couples your authorization checking to your business code. The main problems with this include the enhanced difficulty of unit testing and the fact it would be more difficult to reuse the Customer authorization logic elsewhere. Obtaining the GrantedAuthority[]s from the Authentication object is also fine, but will not scale to large numbers of Customers. If a user might be able to access 5,000 Customers (unlikely in this case, but imagine if it were a popular vet for a large Pony Club!) the amount of memory consumed and time required to construct the Authentication object would be undesirable. The final method, opening the Customer directly from external code, is probably the best of the three. It achieves separation of concerns, and doesn't misuse memory or CPU cycles, but it is still inefficient in that both the AccessDecisionVoter and the eventual business method itself will perform a call to the DAO responsible for retrieving the Customer object. Two accesses per method invocation is clearly undesirable. In addition, with every approach listed you'll need to write your own access control list (ACL) persistence and business logic from scratch.

Fortunately, there is another alternative, which we'll talk about below.

17.2 Key Concepts

Spring Security's ACL services are shipped in the spring-security-acl-xxx.jar. You will need to add this JAR to your classpath to use Spring Security's domain object instance security capabilities.

Spring Security's domain object instance security capabilities centre on the concept of an access control list (ACL). Every domain object instance in your system has its own ACL, and the ACL records details of who can and can't work with that domain object. With this in mind, Spring Security delivers three main ACL-related capabilities to your application:

  • A way of efficiently retrieving ACL entries for all of your domain objects (and modifying those ACLs)

  • A way of ensuring a given principal is permitted to work with your objects, before methods are called

  • A way of ensuring a given principal is permitted to work with your objects (or something they return), after methods are called

As indicated by the first bullet point, one of the main capabilities of the Spring Security ACL module is providing a high-performance way of retrieving ACLs. This ACL repository capability is extremely important, because every domain object instance in your system might have several access control entries, and each ACL might inherit from other ACLs in a tree-like structure (this is supported out-of-the-box by Spring Security, and is very commonly used). Spring Security's ACL capability has been carefully designed to provide high performance retrieval of ACLs, together with pluggable caching, deadlock-minimizing database updates, independence from ORM frameworks (we use JDBC directly), proper encapsulation, and transparent database updating.

Given databases are central to the operation of the ACL module, let's explore the four main tables used by default in the implementation. The tables are presented below in order of size in a typical Spring Security ACL deployment, with the table with the most rows listed last:

  • ACL_SID allows us to uniquely identify any principal or authority in the system ("SID" stands for "security identity"). The only columns are the ID, a textual representation of the SID, and a flag to indicate whether the textual representation refers to a principal name or a GrantedAuthority. Thus, there is a single row for each unique principal or GrantedAuthority. When used in the context of receiving a permission, a SID is generally called a "recipient".

  • ACL_CLASS allows us to uniquely identify any domain object class in the system. The only columns are the ID and the Java class name. Thus, there is a single row for each unique Class we wish to store ACL permissions for.

  • ACL_OBJECT_IDENTITY stores information for each unique domain object instance in the system. Columns include the ID, a foreign key to the ACL_CLASS table, a unique identifier so we know which ACL_CLASS instance we're providing information for, the parent, a foreign key to the ACL_SID table to represent the owner of the domain object instance, and whether we allow ACL entries to inherit from any parent ACL. We have a single row for every domain object instance we're storing ACL permissions for.

  • Finally, ACL_ENTRY stores the individual permissions assigned to each recipient. Columns include a foreign key to the ACL_OBJECT_IDENTITY, the recipient (ie a foreign key to ACL_SID), whether we'll be auditing or not, and the integer bit mask that represents the actual permission being granted or denied. We have a single row for every recipient that receives a permission to work with a domain object.

As mentioned in the last paragraph, the ACL system uses integer bit masking. Don't worry, you need not be aware of the finer points of bit shifting to use the ACL system, but suffice to say that we have 32 bits we can switch on or off. Each of these bits represents a permission, and by default the permissions are read (bit 0), write (bit 1), create (bit 2), delete (bit 3) and administer (bit 4). It's easy to implement your own Permission instance if you wish to use other permissions, and the remainder of the ACL framework will operate without knowledge of your extensions.

It is important to understand that the number of domain objects in your system has absolutely no bearing on the fact we've chosen to use integer bit masking. Whilst you have 32 bits available for permissions, you could have billions of domain object instances (which will mean billions of rows in ACL_OBJECT_IDENTITY and quite probably ACL_ENTRY). We make this point because we've found sometimes people mistakenly believe they need a bit for each potential domain object, which is not the case.

Now that we've provided a basic overview of what the ACL system does, and what it looks like at a table structure, let's explore the key interfaces. The key interfaces are:

  • Acl: Every domain object has one and only one Acl object, which internally holds the AccessControlEntrys as well as knows the owner of the Acl. An Acl does not refer directly to the domain object, but instead to an ObjectIdentity. The Acl is stored in the ACL_OBJECT_IDENTITY table.

  • AccessControlEntry: An Acl holds multiple AccessControlEntrys, which are often abbreviated as ACEs in the framework. Each ACE refers to a specific tuple of Permission, Sid and Acl. An ACE can also be granting or non-granting and contain audit settings. The ACE is stored in the ACL_ENTRY table.

  • Permission: A permission represents a particular immutable bit mask, and offers convenience functions for bit masking and outputting information. The basic permissions presented above (bits 0 through 4) are contained in the BasePermission class.

  • Sid: The ACL module needs to refer to principals and GrantedAuthority[]s. A level of indirection is provided by the Sid interface, which is an abbreviation of "security identity". Common classes include PrincipalSid (to represent the principal inside an Authentication object) and GrantedAuthoritySid. The security identity information is stored in the ACL_SID table.

  • ObjectIdentity: Each domain object is represented internally within the ACL module by an ObjectIdentity. The default implementation is called ObjectIdentityImpl.

  • AclService: Retrieves the Acl applicable for a given ObjectIdentity. In the included implementation (JdbcAclService), retrieval operations are delegated to a LookupStrategy. The LookupStrategy provides a highly optimized strategy for retrieving ACL information, using batched retrievals (BasicLookupStrategy) and supporting custom implementations that leverage materialized views, hierarchical queries and similar performance-centric, non-ANSI SQL capabilities.

  • MutableAclService: Allows a modified Acl to be presented for persistence. It is not essential to use this interface if you do not wish.

Please note that our out-of-the-box AclService and related database classes all use ANSI SQL. This should therefore work with all major databases. At the time of writing, the system had been successfully tested using Hypersonic SQL, PostgreSQL, Microsoft SQL Server and Oracle.

Two samples ship with Spring Security that demonstrate the ACL module. The first is the Contacts Sample, and the other is the Document Management System (DMS) Sample. We suggest taking a look over these for examples.

17.3 Getting Started

To get starting using Spring Security's ACL capability, you will need to store your ACL information somewhere. This necessitates the instantiation of a DataSource using Spring. The DataSource is then injected into a JdbcMutableAclService and BasicLookupStrategy instance. The latter provides high-performance ACL retrieval capabilities, and the former provides mutator capabilities. Refer to one of the samples that ship with Spring Security for an example configuration. You'll also need to populate the database with the four ACL-specific tables listed in the last section (refer to the ACL samples for the appropriate SQL statements).

Once you've created the required schema and instantiated JdbcMutableAclService, you'll next need to ensure your domain model supports interoperability with the Spring Security ACL package. Hopefully ObjectIdentityImpl will prove sufficient, as it provides a large number of ways in which it can be used. Most people will have domain objects that contain a public Serializable getId() method. If the return type is long, or compatible with long (eg an int), you will find you need not give further consideration to ObjectIdentity issues. Many parts of the ACL module rely on long identifiers. If you're not using long (or an int, byte etc), there is a very good chance you'll need to reimplement a number of classes. We do not intend to support non-long identifiers in Spring Security's ACL module, as longs are already compatible with all database sequences, the most common identifier data type, and are of sufficient length to accommodate all common usage scenarios.

The following fragment of code shows how to create an Acl, or modify an existing Acl:

// Prepare the information we'd like in our access control entry (ACE)
ObjectIdentity oi = new ObjectIdentityImpl(Foo.class, new Long(44));
Sid sid = new PrincipalSid("Samantha");
Permission p = BasePermission.ADMINISTRATION;

// Create or update the relevant ACL
MutableAcl acl = null;
try {
  acl = (MutableAcl) aclService.readAclById(oi);
} catch (NotFoundException nfe) {
  acl = aclService.createAcl(oi);

// Now grant some permissions via an access control entry (ACE)
acl.insertAce(acl.getEntries().length, p, sid, true);

In the example above, we're retrieving the ACL associated with the "Foo" domain object with identifier number 44. We're then adding an ACE so that a principal named "Samantha" can "administer" the object. The code fragment is relatively self-explanatory, except the insertAce method. The first argument to the insertAce method is determining at what position in the Acl the new entry will be inserted. In the example above, we're just putting the new ACE at the end of the existing ACEs. The final argument is a boolean indicating whether the ACE is granting or denying. Most of the time it will be granting (true), but if it is denying (false), the permissions are effectively being blocked.

Spring Security does not provide any special integration to automatically create, update or delete ACLs as part of your DAO or repository operations. Instead, you will need to write code like shown above for your individual domain objects. It's worth considering using AOP on your services layer to automatically integrate the ACL information with your services layer operations. We've found this quite an effective approach in the past.

Once you've used the above techniques to store some ACL information in the database, the next step is to actually use the ACL information as part of authorization decision logic. You have a number of choices here. You could write your own AccessDecisionVoter or AfterInvocationProvider that respectively fires before or after a method invocation. Such classes would use AclService to retrieve the relevant ACL and then call Acl.isGranted(Permission[] permission, Sid[] sids, boolean administrativeMode) to decide whether permission is granted or denied. Alternately, you could use our AclEntryVoter, AclEntryAfterInvocationProvider or AclEntryAfterInvocationCollectionFilteringProvider classes. All of these classes provide a declarative-based approach to evaluating ACL information at runtime, freeing you from needing to write any code. Please refer to the sample applications to learn how to use these classes.