AEM 6.4 has reached the end of extended support and this documentation is no longer updated. For further details, see our technical support periods. Find the supported versions here.
The main way of getting an administrative session or resource resolver in AEM was using the SlingRepository.loginAdministrative()
and ResourceResolverFactory.getAdministrativeResourceResolver()
methods provided by Sling.
However, neither of these methods were designed around the principle of least privilege and make it too easy for a developer not to plan for a proper structure and corresponding Access Control Levels (ACLs) for their content early on. If a vulnerability is present in such a service it often leads to privilege escalations to the admin
user, even if the code itself would not need administrative privileges to work.
There may be cases where the admin session is not used, or the feature is disabled entirely. If this is the case with your implementation, make sure you remove the feature altogether or fit it with NOP code.
Whenever possible refactor your feature so that the given, authenticated request session can be used for reading or writing content. If this is not doable, it can often be achieved by applying the priorities following the ones below.
Many issues can be resolved by restructuring the content. Keep these simple rules in mind when doing the restructure:
Change access control
Refine content structure
Refactor your code to be a proper service
Also, make sure that any new features you develop adhere to these principles:
Security requirements should drive the content structure
Make use of nodetypes
Respect privacy settings
/profile
node.Whether you apply access control while restructuring content or when you do it for a new service user, you must apply the strictest ACLs possible. Use all possible facilities of access control:
For example, instead of applying jcr:read
on /apps
, only apply it to /apps/*/components/*/analytics
Use restrictions
Apply ACLs for node types
Limit permissions
jcr:write
permission; use jcr:modifyProperties
insteadIf the above fails, Sling 7 offers a Service User Mapping service, which allows to configure a bundle-to-user mapping and two corresponding API methods: [SlingRepository.loginService()](https://sling.apache.org/apidocs/sling7/org/apache/sling/jcr/api/SlingRepository.html#loginService-java.lang.String-java.lang.String-)
and [ResourceResolverFactory.getServiceResourceResolver()](https://sling.apache.org/apidocs/sling7/org/apache/sling/api/resource/ResourceResolverFactory.html#getServiceResourceResolver-java.util.Map-)
which return a session/resource resolver with the privileges of a configured user only. These methods have the following characteristics:
They allow mapping services to users
They make it possinle to define sub-service users
The central configuration point is: org.apache.sling.serviceusermapping.impl.ServiceUserMapperImpl
service-id
= service-name
[ “:” subservice-name ]
service-id
is mapped to a resource resolver and/or JCR repository user ID for authentication
service-name
is the symbolic name of the bundle providing the service
A service user is a JCR user with no password set and a minimal set of privileges that are necessary to perform a specific task. Having no password set means that it will not be possible to login with a service user.
A way to deprecate an administrative session is to replace it with service user sessions. It could also be replaced by multiple sub-service users if needed.
To replace the admin session with a service user, you should perform the following steps:
Identify the necessary permissions for your service, keeping the principle of least permission in mind.
Check if there is already a user available with exactly the permission setup you need. Create a new system service user if no existing user matches your needs. RTC is needed to create a new service user. Sometimes, it makes sense to create multiple sub-service users (for example, one for writing and one for reading) to compartmentalize access even more.
Setup and test ACEs for your user.
Add a service-user
mapping for your service and for user/sub-users
Make the service user sling feature available to your bundle: update to the most recent version of org.apache.sling.api
.
Replace the admin-session
in your code with the loginService
or getServiceResourceResolver
APIs.
After you verified that no user in the list of AEM service users is applicable for your use case and the corresponding RTC issues have been approved, you can go ahead and add the new user to the default content.
The recommended approach is to create a service user to use the repository explorer at https://<server>:<port>/crx/explorer/index.jsp
The goal is to get a valid jcr:uuid
property which is mandatory in order to create the user via a content package installation.
You can create service users by:
Going to the repository explorer at https://<server>:<port>/crx/explorer/index.jsp
Logging in as admin by pressing the Log In link in the upper left corner of the screen.
Next, create and name your system user. In order to create the user as a system one, set the intermediate path as system
and add optional subfolders depending on your needs:
Verify that your system user node looks as follows:
Note that there are no mixin types associated with service users. This means that there will be no access control policies for system users.
When adding the corresponding .content.xml to the content of your bundle, make sure you have set the rep:authorizableId
and that the primary type is rep:SystemUser
. It should look like this:
<?xml version="1.0" encoding="UTF-8"?>
<jcr:root xmlns:jcr="https://www.jcp.org/jcr/1.0" xmlns:rep="internal"
jcr:primaryType="rep:SystemUser"
jcr:uuid="4917dd68-a0c1-3021-b5b7-435d0044b0dd"
rep:principalName="authentication-service"
rep:authorizableId="authentication-service"/>
To add a mapping from your service to the corresponding System Users you need to create a factory configuration for the [ServiceUserMapper](https://sling.apache.org/apidocs/sling7/org/apache/sling/serviceusermapping/ServiceUserMapper.html)
service. To keep this modular such configurations can be provided using the Sling amend mechanism. The recommended way to install such configurations with your bundle is by using Sling Initial Content Loading:
Create a subfolder SLING-INF/content below the src/main/resources folder of your bundle
In this folder create a file named org.apache.sling.serviceusermapping.impl.ServiceUserMapperImpl.amended-<some unique name for your factory configuration>.xml with the content of your factory configuration (including all sub-service user mappings). Example:
Create a SLING-INF/content
folder below the src/main/resources
folder of your bundle;
In this folder create a file named org.apache.sling.serviceusermapping.impl.ServiceUserMapperImpl.amended-<a unique name for your factory configuration>.xml
with the content of your factory configuration, including all sub-service user mappings.
For illustration purposes, take a the file called org.apache.sling.serviceusermapping.impl.ServiceUserMapperImpl.amended-com.adobe.granite.auth.saml.xml
:
<?xml version="1.0" encoding="UTF-8"?>
<node>
<primaryNodeType>sling:OsgiConfig</primaryNodeType>
<property>
<name>user.default</name>
<value></value>
</property>
<property>
<name>user.mapping</name>
<values>
<value>com.adobe.granite.auth.saml=authentication-service</value>
</values>
</property>
</node>
Reference the Sling initial content in the configuration of the maven-bundle-plugin
in the pom.xml
of your bundle. Example:
<Sling-Initial-Content>
SLING-INF/content;path:=/libs/system/config;overwrite:=true;
</Sling-Initial-Content>
Install your bundle and make sure the factory configuration has been installed. You can do this by:
Calls to loginAdministrative()
often appear together with shared sessions. These sessions are acquired on service activation and are only logged out after the service is stopped. Although this is common practice, it leads to two problems:
The most obvious solution for the security risk is to simply replace the loginAdministrative()
call with a loginService()
one to a user with restricted privileges. However, this will not have any impact on any potential performance degradation. A possibility to mitigate that is to wrap all requested information in an object that has no association with the session. Then, create (or destroy) the session on demand.
The recommended approach is to refactor the service’s API to give the caller control over the creation/destruction of the session.
JSPs cannot use loginService()
, because there is no associated service. However, administrative sessions in JSPs are usually a sign of a violation of the MVC paradigm.
This can be fixed in two ways:
The first method is the preferred one.
When processing events or jobs, and in some cases workflows, the corresponding session that triggered the event is usually lost. This leads to event handlers and job processors often using administrative sessions to do their work. There are different conceivable approaches to solve this problem, each with their advantages and disatvantages:
Pass the user-id
in the event payload and use impersonation.
Advantages: Easy to use.
Disadvantages: Still uses loginAdministrative()
. It re-authenticates a request that has already been authenticated.
Create or reuse a service user that has access to the data.
Advantages: Consistent with the current design. Needs minimal change.
Disadvantages: Needs very powerful service users to be flexible, which can easily lead to privilege escalations. Circumvents the security model.
Pass a serialization of the Subject
in the event payload, and create a ResourceResolver
based on that subject. One example would be using the JAAS doAsPrivileged
in the ResourceResolverFactory
.
Advantages: Clean implementation from a security standpoint. It avoids re-authentication and it operates with the original privileges. Security relevant code is transparent to the consumer of the event.
Disadvantages: Needs refactoring. The fact that the security relevant code transparent to the consumer of the event might also lead to problems.
The third approach is currently the preferred processing technique.
Within workflow process implementations the corresponding user session that triggered the workflow is usually lost. This leads to workflow processes often using administrative sessions to perform their work.
In order to fix these issues, it is recommended that the same approaches mentioned in Processing Events, Replication Preprocessors and Jobs be used.
There are a couple of administrative sessions used in sling POST processor implementations. Usually, administrative sessions are used to access nodes that are pending deletion within the POST being processed. In consequence, they are no longer available via the request session. A node pending deletion may be accessed to disclose metada that otherwise should not be accessible.