Fork me on GitHub

Best Practices for Authorization

Before you get started

Threat Model

Before you start coding, creating content or setting up access control set aside some time to consider what is needed when it comes to securing your application (and what could go wrong). In other words: write a threat model and make sure you keep updating it as you continue developing.

The following references provide a good overview as well as guidance on how to build a threat model:

Content Modelling

As suggested in Jackrabbbit Wiki the content hierarchy in your JCR repository should be designed and not just happen. Access control requirements tend to be a good driver.

Make sure the content design allows for a readable and manageable access control setup later on to secure your data. Excessive complexity is often a strong indicator for problems with your content model, making its security error prone and difficult to reason about (and might ultimately might lead to issues with scaling).

Here is an example of an access control setup (with Sling RepoInit) illustrating why content with different access requirements should be kept in separate trees and how complexity may yield undesired effects (see also section ‘Remember inheritance’ below):

  # Content design that results in complex in ac-setup and a vulnerability
  # ----------------------------------------------------------------------
  
  create path /content
  create path /content/public
  create path /content/content2/also_public   # extra folder with public information
  create path /content/sensitive_info         # sensitive data mixed with regular non-sentive content

  set ACL on /content 
     deny  everyone   jcr:all    # most likely redundant
     allow readers    jcr:read
     allow editors    jcr:read, jcr:write
     deny  readers    jcr:read restriction(rep:subtrees, /sensitive_info)        # what about editors or a subject being both reader and editor?
     allow everyone   jcr:read restriction(rep:subtrees, /public, /also_public) # different public folders??
     
     # ... and what happens with a new node /content/public/abc/sensitive_info?
  end 
  
  # Improved content design
  # ------------------------------------------------------------------------------
  
  create path /content
  create path /content/public
  create path /sensitive_info
  
  set ACL on /content 
      allow readers    jcr:read
      allow editors    jcr:read, jcr:write
  end
  
  set ACL on /content/public     
      allow everyone   jcr:read
  end
  
  # NOTE: no ac setup for sensitive info as neither of the roles must have access

Define Roles and Tasks

Finally, write down basic characteristics and demands of your application without getting into access control details or making any assumptions on how your needs will reflected in the repository:

  • what roles are present
  • what kind of tasks are those roles designed to perform
  • define if you have services accessing the repository and what kind of tasks they need to complete

Note, that this document should be human readable not go into implementation details: Instead of writing principal ‘content-authors’ needs jcr:write on /content, define that you have an asset ‘content’, define what kind of data it contains and how sensitive the data is (similar to the threat model). Then identify what roles are going to interact with this data and how they interact: for example you may identify a role that just reading data, a second role that is expected to read and write, and a third one that will only approve new content and publish it).

General Best Practices

Know how to get what you need

Familiarize yourself with JCR access control management and Oak authorization design and extensions before starting to edit the permission setup of your Oak installation. This will help you avoid common pitfalls. If you find yourself granting your content-writers role full access to just make it work, you probably left your application vulnerable.

Principle of least privilege

Keep in mind that not having any permissions granted is equivalent to denying everything (which is in this case redundant). Start without any access and then keep granting permissions as needed, following the principle of least privilege. In other words: only grant the minimal set of privileges required to perform a particular task.

Verification

Write tests upfront and verify that for each role and task the expected effective permissions (see definition of roles) are granted. Neither less nor more.

Ideally, your tests will fail as soon as someone is attempting to make any change to the permission setup. Granting additional permissions may open up the door for a privilege escalation and revoking permissions will break your application (if it doesn't you did not follow the principle of least privilege).

This may also include assertions that no permissions are granted at resources that are outside the scope of a given role/task.

Example: Pseud-code for a permission validator

  PermissionTestRule ptr = ...
  ptr.newPermissionValidator(EveryonePrincipal.NAME)
        .hasNoPermission("/")
        .hasNoPermission("/content")
        .hasNoPermission("/content/protected", READ)
        .hasOnlyPermission("/content/public", READ)
        .hasOnlyPermission("/content/public/child-item, READ)
        .hasNoPermission("/sensitive_info")
        .validate(); 

Oak Specific Best Practices

Avoid deny

All authorization models present with Apache Jackrabbbit Oak start without any access granted by default i.e. implicit deny everywhere. It is therefore recommended to only grant access where needed and avoid adding explicit deny access control entries. In particular in combination with subsequent allow rules the overall effect will be hard to understand as soon as multiple principals are contained in a given subject.

Be wary if you find yourself adding combinations of denies and allows as it might highlight problematic patterns in your content model that will be hard to understand and secure over time.

Avoid redundancy

Don't specify redundant access control setup just to be on the safe side:

  • If access is granted, avoid repeating the same setup down the hierarchy.
  • Avoid setup for principals with administrative access for which permission evaluation is omitted. It might even create a false sense of security.
  • Avoid redundant membership as it will impact performance of permission evaluation

Principal by principle

Oak authorization is designed to work with java.security.Principal which is an abstract representation of any kind of entity like e.g. individual, a role, a corporation, a login id or even a service.

While JCR specification does not define how the repository knows about principals, Jackrabbit API defines a Principal Management extension.

Not every principal is a user/group

Oak allows plugging custom sources of principals which are all reflected through the principal management API. Therefore, don't assume that every principal is backed by a user or a group. The repository's user management is just one potential source of principals.

Example : everyone
 // everyone always exists even if there is no such group in the user management
 
 PrincipalManager principalMgr = ((JackrabbitSession) session).getPrincipalManager();
 Principal everyone = principalManager.getEveryone();

Membership is no guarantee

Similarly, make sure you always evaluate permissions to verify if a subject has access granted instead of checking if a user is member of a group. How access control defined for a particular group principal affects its members is an implementation detail of the authorization setup.

Example : administrative access

In the default authorization model full access to the repository can be configured for selected user or group principals. (see Configuration Parameters for the default permission evaluation). If you wish to determine if a given subject has full access, don't assume that there is a group ‘administrators’ and that its members have full access.

Stick with group principals

It is preferable to set up access control for group principals instead of individual user principals and then make sure your PrincipalProvider resolves principal membership according to your needs.

Further, note that the default authorization model will give precedence to user principals upon evaluation in other words default access control entries for user principals will overwrite the effect of groups irrespective of the order in the list (see next section).

The above rule is particularly important for the anonymous user marking access with GuestCredentials. If you setup access control for anonymous it will result in the guest account to have effective permissions that do not apply for any authenticated session.

What is usually intended instead is setting up permissions for the everyone group. See also PrincipalManager.getEveryone()

Understand default access control and permission management

Remember inheritance

When designing your access control setup keep in mind that effective permissions are inherited down the node hierarchy: allowing jcr:read for content-readers role on /content will also grant content-readers access to all nodes and properties in the subtree (e.g. /content/project1 or /content/project1/jcr:title).

In addition, effective permissions get inherited through (nested) group principals according to the set of principals resolved and added to the javax.security.auth.Subject upon repository login.

See Permission Evaluation in Detail for additional information as well as the exercises at L3_PrecedenceRulesTest

Built-in privileges

JSR 382 defines a set of built-in privileges and how they apply to repository operations (see https://s.apache.org/jcr-2.0-javadoc/javax/jcr/security/Privilege.html). The default set has been extended by Oak to cover additional features outside of the scope defined by JCR (like e.g. index or user management). The complete list can be found in Privilege Management : The Default Implementation.

The minimal set of privileges required for each operation are outlined in Mapping API Calls to Privileges and Mapping Privileges to Items.

Privileges affecting the parent node

Note in particular for add/removing a node jcr:addChildNodes and jcr:removeChildNodes are required on the parent node respectively i.e. allowing for modification of the child-node collection. In addition jcr:removeNode needs to be granted on the target node of the removal.

Thus, the following subtle difference apply when evaluation effective permissions vs. privileges (see also Permissions vs Privileges) and exercises at L4_PrivilegesAndPermissionsTest.java):

    String parentPath = "/content/parent";
    String toRemove = "/content/parent/child";
    String toAdd = "/content/parent/newchild";
    
    Session session = ...
            
    // Testing Privileges
    // ----------------------------------------------------------------------------------------
    
    AccessControlManager acMgr = session.getAccessControlManager();
    Privilege jcrAddChildNodes = acMgr.privilegeFromName(Privilege.JCR_ADD_CHILD_NODES);
    Privilege jcrRemoveChildNodes = acMgr.privilegeFromName(Privilege.JCR_REMOVE_CHILD_NODES);
    Privilege jcrRemoveNode = acMgr.privilegeFromName(JCR_REMOVE_NODE);
    
    // test if (unspecified) child nodes can be added/removed from the parent
    boolean canModifyChildCollection = acMgr.hasPrivileges(parentPath, new Privilege[]{jcrAddChildNodes, jcrRemoveChildNodes});
    
    # test if existing child node can be removed
    boolean canRemoveNode = acMgr.hasPrivileges(toRemove, new Privilege[]{jcrRemoveNode});
    
    
    // Testing Permissions (on the target node NOT on the parent)
    // ----------------------------------------------------------------------------------------

    // test if not-yet existing node could be added at /content/parent/newchild
    boolean canAddNode = session.hasPermission(toAdd, Session.ACTION_ADD_NODE);
    
    // test if the existing child node can be removed 
    boolean canRemoveItem = session.hasPermission(toRemove, Session.ACTION_REMOVE);
    boolean canRemoveNode = session.hasPermission(toRemove, JackrabbitSession.ACTION_REMOVE_NODE);
    
    // test if a non-existing node could be removed (not possible with privilege evaluation);
    boolean canRemoveNode = session.hasPermission(toAdd, JackrabbitSession.ACTION_REMOVE_NODE);

Leverage PrivilegeCollection

Since Oak 1.42.0 the Jackrabbit API defines a new interface PrivilegeCollection the offers improved support for testing effective privileges (see also OAK-9494). It allows avoiding repeated calls to AccessControlManager.hasPrivileges and manual resolution of aggregated privileges when dealing with the privilege array returned by AccessControlManager.getPrivileges.

    // Using PrivilegeCollection
    // ----------------------------------------------------------------------------------------
    
    JackrabbitAccessControlManager acMgr = ...

    PrivilegeCollection pc = acMgr.getPrivilegeCollection(parentPath);
    
    boolean canRemoveChildNodes = pc.includes(Privilege.JCR_REMOVE_CHILD_NODES);
    boolean canModifyChildCollection = pc.includes(Privilege.JCR_REMOVE_CHILD_NODES, Privilege.JCR_ADD_CHILD_NODES);
    
    boolean hasAllPrivileges = pc.includes(Privilege.JCR_ALL);
    assertFalse(hasAllPrivileges);
    
    Privilege[] privilegesOnParentNode = pc.getPrivileges();  

Use restrictions to limit effect

Apart from picking the minimal set of privileges you can further minimize the risk of privilege escalation by narrowing the effect of a given access control setup on certain items in the subtree. This is achieved by creating access control entries that come with an additional restriction.

Note though, that restrictions can affect readability. So, you may want to find a balance between enhanced security and simplicity. Revisiting your content design early on will likely be the better choice.

See section Restriction Management for additional details as well as lessons L7_RestrictionsTest and L8_GlobRestrictionTest in the Oak exercise module.

Access control setup for system user

If your Oak setup supports principal-based authorization (see Managing Access by Principal) it is recommended to leverage it for system sessions associated with OSGi service. It helps to keep application owned access control setup apart from regular content.

See also Service Authentication in Apache Sling.

Leverage customizations

Oak allows for customization and extensions of all parts of the authorization setup. If you find yourself struggling to reflect your needs with the built-in functionality, consider extending and customizing the authorization configuration of the repository.

Leverage custom privileges

If you identify application specific operations that cannot be reflected using the built-in privileges, Oak allows
registering custom privileges (see section Privilege Management).

However, note that the built-in permission evaluation will not enforce those custom privileges. Instead, you have to enforce it in your application or write a custom authorization model (see section Combining Multiple Authorization Models)

In the example above you might find that publishing content cannot easily be secured using built-in privileges and end up registering a custom myapp:publish privilege.

Leverage custom restrictions

Default authorization in Oak allows to limit the effect of individual JCR access control entries by means of restrictions. See section Restriction Management for the built-in restrictions and instructions on how to plug custom restrictions into the security setup. Be aware though of the potential performance impact of any additional evaluation.

Leverage a custom authorization model

If you find that that built-in authorization model is not suited to reflect your needs and setting up access control becomes cumbersome and overly complex, consider customizing authorization setup (see section Combining Multiple Authorization Models).

The oak-exercise module defines a couple of examples to illustrate alternative approaches. The corresponding training material is located in section Advanced Authorization Topics.