Create and manage attribute-based access control (ABAC) policies

Beta

This feature is in Beta.

This page describes how to configure row filter and column mask policies in Unity Catalog. For more information on attribute-based access control (ABAC) and policies, see Unity Catalog attribute-based access control (ABAC). To apply tags to objects, see Governed tags and Apply tags to Unity Catalog securable objects.

Enable ABAC

The ABAC beta is enabled at the workspace level. Databricks cannot enforce ABAC on read operations to shared catalogs unless ABAC is also enabled in each workspace that accesses those catalogs.

To enable the ABAC beta, do the following:

1. As a workspace admin, click your username in the top bar of the Databricks workspace.
2. From the menu, select Previews.
3. Set the Attribute Based Access Control toggle to On.

Compute requirements

To use ABAC policies, you must use one of the following compute configurations:

• Serverless compute
• Standard compute on Databricks Runtime 16.4 or above
• Dedicated compute on Databricks Runtime 16.4 or above with fine-grained access control filtering enabled

Standard and dedicated compute running older runtimes cannot access tables secured by ABAC. As a temporary workaround, you can configure ABAC to apply only to a specific group. Add the users you want to restrict to that group. Users who are not in the group can still access the tables.

Policy quotas

The following lists the number of ABAC policies you can create on different securable objects:

• Catalog: 15 policies per catalog
• Schema: 15 policies per schema
• Table: 5 policies per table

Create a policy on an object

To create a row filter or column mask policy, you must have:

• The ABAC beta enabled in your workspace. See Enable ABAC.
• An existing user-defined function (UDF) that implements the row filter or column mask logic you want to enforce.
  For best practices, limitations, and example UDFs, see UDFs for ABAC policies best practices.
• MANAGE on the object or ownership of the object.

Compute requirements

• You must use compute on Databricks Runtime 16.4 or above or serverless compute.

Compute running older runtimes cannot access tables secured by ABAC. As a temporary workaround, you can configure ABAC to apply only to a specific group. Add the users you want to restrict to that group. Users who are not in the group can still access the tables.

Catalog Explorer

1. In your Databricks workspace, click Catalog.

2. Select the object that determines the policy scope, such as a catalog, schema, or table.

3. Click the Policies tab.

4. Click New policy.

5. In General, enter a name and description for your policy.

6. In Principals:

   • In Applied to…, search for and select the principals that the policy will govern.
   • In Except for…, search for and select any principals to exclude from the policy. For example, you might exclude a user who belongs to a group the policy applies to.

   

7. In Type & target:

   • In Policy type, select Row Filter or Column Mask.
   • In Policy target, select the scope of the policy. This can be a broad scope, such as an entire catalog or schema, or a narrower scope, such as specific tables or columns within it.
   • In Table level condition, specify the condition matching the tables to which this policy applies. For example, hasTag("tag") or hasTagValue("tag", "value").

8. In Function, select a function for this policy to use and enter required parameters.

   

9. Click Create policy.

SQL

The following is the general syntax for creating a policy:

SQL

CREATE POLICY <policy_name>
ON <securable_type> <securable_name>
COMMENT '<policy_description>'
-- One of the following:
  ROW FILTER <udf_name>
  | COLUMN MASK <udf_name> ON COLUMN <target_column>
TO <principal_name>[, <principal_name>, ...]
[EXCEPT <principal_name>[, <principal_name>, ...]]
FOR TABLES
[WHEN hasTag('<key>') OR hasTagValue('<key>', '<value>')]
MATCH COLUMNS hasTag('<key>') OR hasTagValue('<key>', '<value>') AS <alias>
USING COLUMNS <alias>[, <alias>, ...];

When you define table level conditions (MATCH COLUMNS), the policy applies to the table only if at least one column satisfies all column match conditions. If none of the columns satisfy all listed conditions, the policy doesn't apply to the table, and Databricks skips any row filtering or column masking.

For example, if a policy defines MATCH COLUMNS condition1, condition2, condition3, the policy takes effect only if at least one column in the table satisfies all three conditions. A policy can include up to three column conditions in its MATCH COLUMNS clause.

When you define table-level conditions (MATCH COLUMNS), the policy applies to the table only if every condition matches at least one column. If a condition doesn’t match any columns in the table, the policy doesn’t apply to the table, and Databricks skips any row filtering or column masking.

For example, if a policy defines MATCH COLUMNS condition1, condition2, condition3, the policy takes effect only if each of the three conditions matches at least one column in the table.

A policy can include up to three column conditions in its MATCH COLUMNS clause.

This example defines a row filter policy that excludes rows for European customers from queries by US-based analysts:

SQL

CREATE POLICY hide_eu_customers
ON SCHEMA prod.customers
COMMENT 'Hide rows with European customers from sensitive tables'
ROW FILTER non_eu_region
TO us_analysts
FOR TABLES
MATCH COLUMNS
  hasTag('geo_region') AS region
USING COLUMNS (region);

This example defines a column mask policy that hides social security numbers from US analysts, except for those with in the admins group:

SQL

CREATE POLICY mask_SSN
ON SCHEMA prod.customers
COMMENT 'Mask social security numbers'
COLUMN MASK mask_SSN
TO us_analysts
EXCEPT admins
FOR TABLES
MATCH COLUMNS
  hasTagValue('pii', 'ssn') AS ssn
ON COLUMN ssn;

warning

Tag data is stored as plain text and may be replicated globally. Do not use tag names, values, or descriptors that could compromise the security of your resources. For example, do not use tag names, values or descriptors that contain personal or sensitive information.

Edit a policy

Permissions required: MANAGE on the object or the owner of the object.

Catalog Explorer

1. In your Databricks workspace, click Catalog.
2. Select the object that determines the policy scope, such as a catalog, schema, or table.
3. Click the Policies tab.
4. Select the policy and make edits.
5. Click Update policy.

SQL

SQL

CREATE OR REPLACE POLICY <policy_name>
ON <securable_type> <securable_name>
COMMENT '<policy_description>'
-- One of the following:
  ROW FILTER <udf_name>
  | COLUMN MASK <udf_name> ON COLUMN <target_column>
TO <principal_name>[, <principal_name>, ...]
[EXCEPT <principal_name>[, <principal_name>, ...]]
FOR TABLES
[WHEN hasTag('<key>') OR hasTagValue('<key>', '<value>')]
MATCH COLUMNS hasTag('<key>') OR hasTagValue('<key>', '<value>') AS <alias>
USING COLUMNS <alias>[, <alias>, ...];

Delete a policy

Permissions required: MANAGE on the object or the owner of the object.

Catalog Explorer

1. In your Databricks workspace, click Catalog.
2. Select the object that determines the policy scope, such as a catalog, schema, or table.
3. Click the Policies tab.
4. Select the policy.
5. Click Delete policy.

SQL

SQL

DROP POLICY <policy_name> ON <securable_type> <securable_name>

Troubleshooting multiple filters or masks

ABAC enforces a limit of one row filter per table and one column mask per column. This prevents unclear results when multiple filters or masks interact.

When Databricks detects multiple filters or masks during policy evaluation, it throws an INVALID_PARAMETER_VALUE.UC_ABAC_MULTIPLE_ROW_FILTERS or COLUMN_MASKS_FEATURE_NOT_SUPPORTED.MULTIPLE_MASKS error. This behavior is by design and blocks access to the table until the conflict is resolved.

Understanding how multiple row filters or masks occur

Multiple filters or masks can arise in several ways:

• A single policy generates multiple filters or masks.
  This can happen when multiple columns match the policy conditions. For example, the following policy defines a row filter based on columns tagged with region=EMEA:

  SQL

  CREATE OR REPLACE POLICY region_filter_policy
  ON TABLE my_catalog.my_schema.customer_data
  ROW FILTER my_catalog.my_schema.filter_by_region
  TO account users
  FOR TABLES
  MATCH COLUMNS
    hasTagValue('region', 'EMEA') AS region_cols
  USING COLUMNS (region_cols);

  If the customer_data table has multiple columns tagged with region=EMEA, this single policy generates one row filter per matching column. When Databricks evaluates the policy, it detects multiple row filters and throws the error.

• Multiple policies define filters or masks on the same table or column. If more than one ABAC policy applies to the same table or column, Databricks detects multiple effective filters or masks.

• A table or column already has a manually applied filter or mask. Conflicts can also occur when a table or column includes both a manually applied (non-ABAC) row filter or column mask and one or more ABAC-defined filters or masks.

How to resolve the error

You can resolve the multiple row filters error with any of the following approaches:

• Refine your policy's column matches: Update the MATCH COLUMNS clause to be more specific, ensuring that it matches only one column. For example, combine multiple conditions to narrow down the match.

• Adjust your governed tags: Review which columns have the governed tags that trigger the policy. Remove or modify these tags if they shouldn't be included in the row filter.

• Restructure your policies: Instead of relying on conditions that might match multiple columns, consider creating separate policies with explicit column targeting. This gives you more control over which columns trigger row filters.