Providing Document-Level Security Using Sentry

For role-based access control of a collection, an administrator modifies a Sentry role so it has query, update, or administrative access, as described above.

Collection-level authorization is useful when the access control requirements for the documents in the collection are the same, but users may want to restrict access to a subset of documents in a collection. This finer-grained restriction could be achieved by defining separate collections for each subset, but this is difficult to manage, requires duplicate documents for each collection, and requires that these documents be kept synchronized.

Document-level access control solves this issue by associating authorization tokens with each document in the collection. This enables granting Sentry roles access to sets of documents in a collection.

Document-Level Security Model

Document-level security depends on a chain of relationships between users, groups, roles, and documents.

  • Users are assigned to groups.
  • Groups are assigned to roles.
  • Roles are stored as "authorization tokens" in a specified field in the documents.

Document-level security supports restricting which documents can be viewed by which users. Access is provided by adding roles as "authorization tokens" to a specified document field. Conversely, access is implicitly denied by omitting roles from the specified field. In other words, in a document-level security enabled environment, a user might submit a query that matches a document; if the user is not part of a group that has a role has been granted access to the document, the result is not returned.

For example, Alice might belong to the administrators group. The administrators group may belong to the doc-mgmt role. A document could be ingested and the doc-mgmt role could be added at ingest time. In such a case, if Alice submitted a query that matched the document, Search would return the document, since Alice is then allowed to see any document with the "doc-mgmt" authorization token.

Similarly, Bob might belong to the guests group. The guests group may belong to the public-browser role. If Bob tried the same query as Alice, but the document did not have the public-browser role, Search would not return the result because Bob does not belong to a group that is associated with a role that has access.

Note that collection-level authorization rules still apply, if enabled. Even if Alice is able to view a document given document-level authorization rules, if she is not allowed to query the collection, the query will fail.

Roles are typically added to documents when those documents are ingested, either via the standard Solr APIs or, if using morphlines, the setValues morphline command.

Enabling Document-Level Security

Cloudera Search supports document-level security in Search for CDH 5.1 and later. Document-level security requires collection-level security. Configuring collection-level security is described earlier in this topic.

Document-level security is disabled by default, so the first step in using document-level security is to enable the feature by modifying the solrconfig.xml.secure file. Remember to replace the solrconfig.xml with this file, as described in Enabling Sentry in Cloudera Search for CDH 5.

To enable document-level security, change solrconfig.xml.secure. The default file contents are as follows:

<searchComponent name="queryDocAuthorization">
    <!-- Set to true to enabled document-level authorization -->

    <bool name="enabled">false</bool>


    <!-- Field where the auth tokens are stored in the document -->
    <str name="sentryAuthField">sentry_auth</str>


    <!-- Auth token defined to allow any role to access the document.
         Uncomment to enable. -->

    <!--<str name="allRolesToken">*</str>-->

</searchComponent>
  • The enabled Boolean determines whether document-level authorization is enabled. To enable document level security, change this setting to true.
  • The sentryAuthField string specifies the name of the field that is used for storing authorization information. You can use the default setting of sentry_auth or you can specify some other string to be used for assigning values during ingest.
  • The allRolesToken string represents a special token defined to allow any role access to the document. By default, this feature is disabled. To enable this feature, uncomment the specification and specify the token. This token should be different from the name of any sentry role to avoid collision. By default it is "*". This feature is useful when first configuring document level security or it can be useful in granting all roles access to a document when the set of roles may change. See Best Practices for additional information.

Best Practices

Using allRolesToken

You may want to grant every user that belongs to a role access to certain documents. One way to accomplish this is to specify all known roles in the document, but this requires updating or re-indexing the document if you add a new role. Alternatively, an allUser role, specified in the Sentry .ini file, could contain all valid groups, but this role would need to be updated every time a new group was added to the system. Instead, specifying allRolesToken allows any user that belongs to a valid role to access the document. This access requires no updating as the system evolves.

In addition, allRolesToken may be useful for transitioning a deployment to use document-level security. Instead of having to define all the roles upfront, all the documents can be specified with allRolesToken and later modified as the roles are defined.

Consequences of Document-Level Authorization Only Affecting Queries

Document-level security does not prevent users from modifying documents or performing other update operations on the collection. Update operations are only governed by collection-level authorization.

Document-level security can be used to prevent documents being returned in query results. If users are not granted access to a document, those documents are not returned even if that user submits a query that matches those documents. This does not have affect attempted updates.

Consequently, it is possible for a user to not have access to a set of documents based on document-level security, but to still be able to modify the documents via their collection-level authorization update rights. This means that a user can delete all documents in the collection. Similarly, a user might modify all documents, adding their authorization token to each one. After such a modification, the user could access any document via querying. Therefore, if you are restricting access using document-level security, consider granting collection-level update rights only to those users you trust and assume they will be able to access every document in the collection.

Limitations on Query Size

By default queries support up to 1024 Boolean clauses. As a result, queries containing more that 1024 clauses may cause errors. Because authorization information is added by Sentry as part of a query, using document-level security can increase the number of clauses. In the case where users belong to many roles, even simple queries can become quite large. If a query is too large, an error of the following form occurs:

org.apache.lucene.search.BooleanQuery$TooManyClauses: maxClauseCount is set to 1024
To change the supported number of clauses, edit the maxBooleanClauses setting in solrconfig.xml. For example, to allow 2048 clauses, you would edit the setting so it appears as follows: 
<maxBooleanClauses>2048</maxBooleanClauses>

For maxBooleanClauses to be applied as expected, make any change to this value to all collections and then restart the service. You must make this change to all collections because this option modifies a global Lucene property, affecting all Solr cores. If different solrconfig.xml files have different values for this property, the effective value is determined per host, based on the first Solr core to be initialized.