Row-Level Security enables customers to control access to rows in a database table based on the characteristics of the user executing a query (e.g., group membership or execution context).
Row-Level Security (RLS) simplifies the design and coding of security in your application. RLS enables you to implement restrictions on data row access. For example ensuring that workers can access only those data rows that are pertinent to their department, or restricting a customer's data access to only the data relevant to their company.
The access restriction logic is located in the database tier rather than away from the data in another application tier. The database system applies the access restrictions every time that data access is attempted from any tier. This makes your security system more reliable and robust by reducing the surface area of your security system.
RLS supports two types of security predicates.
Filter predicates silently filter the rows available to read operations (SELECT, UPDATE, and DELETE).
Block predicates explicitly block write operations (AFTER INSERT, AFTER UPDATE, BEFORE UPDATE, BEFORE DELETE) that violate the predicate.
Access to row-level data in a table is restricted by a security predicate defined as an inline table-valued function. The function is then invoked and enforced by a security policy. For filter predicates, there is no indication to the application that rows have been filtered from the result set; if all rows are filtered, then a null set will be returned. For block predicates, any operations that violate the predicate will fail with an error.
Filter predicates are applied while reading data from the base table, and it affects all get operations: SELECT, DELETE (i.e. user cannot delete rows that are filtered), and UPDATE (i.e. user cannot update rows that are filtered, although it is possible to update rows in such way that they will be subsequently filtered). Block predicates affect all write operations.
AFTER INSERT and AFTER UPDATE predicates can prevent users from updating rows to values that violate the predicate.
BEFORE UPDATE predicates can prevent users from updating rows that currently violate the predicate.
BEFORE DELETE predicates can block delete operations.
Both filter and block predicates and security policies have the following behavior:
You may define a predicate function that joins with another table and/or invokes a function. If the security policy is created with
SCHEMABINDING = ON, then the join or function is accessible from the query and works as expected without any additional permission checks. If the security policy is created with
SCHEMABINDING = OFF, then users will need SELECT or EXECUTE permissions on these additional tables and functions in order to query the target table.
You may define a predicate function that joins with another table and/or invokes a function. The join/function is accessible from the query and works as expected without any additional permission checks.
You may issue a query against a table that has a security predicate defined but disabled. Any rows that would have been filtered or blocked are not affected.
If the dbo user, a member of the db_owner role, or the table owner queries against a table that has a security policy defined and enabled, the rows are filtered or blocked as defined by the security policy.
Attempts to alter the schema of a table bound by a schema bound security policy will result in an error. However, columns not referenced by the predicate can be altered.
Attempts to add a predicate on a table that already has one defined for the specified operation (regardless of whether it is enabled or disabled) results in an error.
For schema bound security policies, attempts to modify a function used as a predicate on a table within a security policy results in an error.
Defining multiple active security policies that contain non-overlapping predicates, succeeds.
Filter predicates have the following behavior:
Define a security policy that filters the rows of a table. The application is unaware that any rows have been filtered for SELECT, UPDATE, and DELETE operations, including situations where all the rows have been filtered out. The application can INSERT any rows, regardless of whether or not they will be filtered during any other operation.
Block predicates have the following behavior:
Block predicates for UPDATE are split into separate operations for BEFORE and AFTER. Consequently, you cannot, for example, block users from updating a row to have a value higher than the current one. If this kind of logic is required, you must use triggers with the DELETED and INSERTED intermediate tables to reference the old and new values together.
The optimizer will not check an AFTER UPDATE block predicate if none of the columns used by the predicate function were changed. For example: Alice should not be able to change a salary to be greater than 100,000, but she should be able to change the address of an employee whose salary is already greater than 100,000 (and thus already violates the predicate).
No changes have been made to the bulk APIs, including BULK INSERT. This means that block predicates AFTER INSERT will apply to bulk insert operations just as they would regular insert operations.
Here are design examples of how RLS can be used:
A hospital can create a security policy that allows nurses to view data rows for their own patients only.
A bank can create a policy to restrict access to rows of financial data based on the employee's business division, or based on the employee's role within the company.
A multi-tenant application can create a policy to enforce a logical separation of each tenant's data rows from every other tenant's rows. Efficiencies are achieved by the storage of data for many tenants in a single table. Of course, each tenant can see only its data rows.
RLS filter predicates are functionally equivalent to appending a WHERE clause. The predicate can be as sophisticated as business practices dictate, or the clause can be as simple as
WHERE TenantId = 42.
In more formal terms, RLS introduces predicate based access control. It features a flexible, centralized, predicate-based evaluation that can take into consideration metadata or any other criteria the administrator determines as appropriate. The predicate is used as a criterion to determine whether or not the user has the appropriate access to the data based on user attributes. Label-based access control can be implemented by using predicate-based access control.
Creating, altering, or dropping security policies requires the ALTER ANY SECURITY POLICY permission. Creating or dropping a security policy requires ALTER permission on the schema.
Additionally the following permissions are required for each predicate that is added:
SELECT and REFERENCES permissions on the function being used as a predicate.
REFERENCES permission on the target table being bound to the policy.
REFERENCES permission on every column from the target table used as arguments.
Security policies apply to all users, including dbo users in the database. Dbo users can alter or drop security policies however their changes to security policies can be audited. If high privileged users (such as sysadmin or db_owner) need to see all rows to troubleshoot or validate data, the security policy must be written to allow that.
If a security policy is created with
SCHEMABINDING = OFF, then to query the target table, users must have the SELECT or EXECUTE permission on the predicate function and any additional tables, views, or functions used within the predicate function. If a security policy is created with
SCHEMABINDING = ON(the default), then these permission checks are bypassed when users query the target table.
It is highly recommended to create a separate schema for the RLS objects (predicate function and security policy).
The ALTER ANY SECURITY POLICY permission is intended for highly-privileged users (such as a security policy manager). The security policy manager does not require SELECT permission on the tables they protect.
Avoid type conversions in predicate functions to avoid potential runtime errors.
Avoid recursion in predicate functions wherever possible to avoid performance degradation. The query optimizer will try to detect direct recursions, but is not guaranteed to find indirect recursions (i.e., where a second function calls the predicate function).
Avoid using excessive table joins in predicate functions to maximize performance.
Avoid predicate logic that depends on session-specific SET options: While unlikely to be used in practical applications, predicate functions whose logic depends on certain session-specific SET options can leak information if users are able to execute arbitrary queries. For example, a predicate function that implicitly converts a string to datetime could filter different rows based on the SET DATEFORMAT option for the current session. In general, predicate functions should abide by the following rules:
Predicate functions should not implicitly convert character strings to date, smalldatetime, datetime, datetime2, or datetimeoffset, or vice versa, because these conversions are affected by the SET DATEFORMAT (Transact-SQL) and SET LANGUAGE (Transact-SQL) options. Instead, use the CONVERT function and explicitly specify the style parameter.
Predicate functions should not rely on the value of the first day of the week, because this value is affected by the SET DATEFIRST (Transact-SQL) option.
Predicate functions should not rely on arithmetic or aggregation expressions returning NULL in case of error (such as overflow or divide-by-zero), because this behavior is affected by the SET ANSI_WARNINGS (Transact-SQL), SET NUMERIC_ROUNDABORT (Transact-SQL), and SET ARITHABORT (Transact-SQL) options.
Predicate functions should not compare concatenated strings with NULL, because this behavior is affected by the SET CONCAT_NULL_YIELDS_NULL (Transact-SQL) option.
Security Note: Side-Channel Attacks
Malicious security policy manager: It is important to observe that a malicious security policy manager, with sufficient permissions to create a security policy on top of a sensitive column and having permission to create or alter inline table valued functions, can collude with another user that has select permissions on a table to perform data exfiltration by maliciously creating inline table valued functions designed to use side channel attacks to infer data. Such attacks would require collusion (or excessive permissions granted to a malicious user) and would likely require several iterations of modifying the policy (requiring permission to remove the predicate in order to break the schema binding), modifying the inline table valued functions, and repeatedly running select statements on the target table. It is strongly recommended to limit permissions as it is necessary and to monitor for any suspicious activity such as constantly changing policies and inline table valued functions related to row-level security.
Carefully crafted queries: It is possible to cause information leakage through the use of carefully crafted queries. For example,
SELECT 1/(SALARY-100000) FROM PAYROLL WHERE NAME='John Doe' would let a malicious user know that John Doe's salary is $100,000. Even though there is a security predicate in place to prevent a malicious user from directly querying other people's salary, the user can determine when the query returns a divide-by-zero exception.
In general, row-level security will work as expected across features. However, there are a few exceptions. This section documents several notes and caveats for using row-level security with certain other features of SQL Server.
DBCC SHOW_STATISTICS reports statistics on unfiltered data, and thus can leak information otherwise protected by a security policy. For this reason, in order to view a statistics object for a table with a row-level security policy, the user must own the table or the user must be a member of the sysadmin fixed server role, the db_owner fixed database role, or the db_ddladmin fixed database role.
Filestream RLS is incompatible with Filestream.
Polybase RLS is incompatible with Polybase.
Memory-Optimized TablesThe inline table-valued function used as a security predicate on a memory-optimized table must be defined using the
WITH NATIVE_COMPILATIONoption. With this option, language features not supported by memory-optimized tables will be banned and the appropriate error will be issued at creation time. For more information, see the Row-Level Security in Memory Optimized Tables section in Introduction to Memory-Optimized Tables.
Indexed views In general, security policies can be created on top of views, and views can be created on top of tables that are bound by security policies. However, indexed views cannot be created on top of tables that have a security policy, because row lookups via the index would bypass the policy.
Change Data Capture Change Data Capture can leak entire rows that should be filtered to members of db_owner or users who are members of the "gating" role specified when CDC is enabled for a table (note: you can explicitly set this to NULL to enable all users to access the change data). In effect, db_owner and members of this gating role can see all data changes on a table, even if there is a security policy on the table.
Change Tracking Change Tracking can leak the primary key of rows that should be filtered to users with both SELECT and VIEW CHANGE TRACKING permissions. Actual data values are not leaked; only the fact that column A was updated/inserted/deleted for the row with B primary key. This is problematic if the primary key contains a confidential element, such as a Social Security Number. However, in practice, this CHANGETABLE is almost always joined with the original table in order to get the latest data.
Full-Text Search A performance hit is expected for queries using the following Full-Text Search and Semantic Search functions, because of an extra join introduced to apply row-level security and avoid leaking the primary keys of rows that should be filtered: CONTAINSTABLE, FREETEXTTABLE, semantickeyphrasetable, semanticsimilaritydetailstable, semanticsimilaritytable.
Columnstore Indexes RLS is compatible with both clustered and non-clustered columnstore indexes. However, because row-level security applies a function, it is possible that the optimizer may modify the query plan such that it does not use batch mode.
Partitioned Views Block predicates cannot be defined on partitioned views, and partitioned views cannot be created on top of tables that use block predicates. Filter predicates are compatible with partitioned views.
Temporal tables are compatible with RLS. However, security predicates on the current table are not automatically replicated to the history table. To apply a security policy to both the current and the history tables, you must individually add a security predicate on each table.
A. Scenario for users who authenticate to the database
This short example creates three users, creates and populates a table with 6 rows, then creates an inline table valued function and a security policy for the table. The example shows how select statements are filtered for the various users.
Create three user accounts that will demonstrate different access capabilities.
CREATE USER Manager WITHOUT LOGIN; CREATE USER Sales1 WITHOUT LOGIN; CREATE USER Sales2 WITHOUT LOGIN;
Create a simple table to hold data.
CREATE TABLE Sales ( OrderID int, SalesRep sysname, Product varchar(10), Qty int );
Populate the table with 6 rows of data, showing 3 orders for each sales representative.
INSERT Sales VALUES (1, 'Sales1', 'Valve', 5), (2, 'Sales1', 'Wheel', 2), (3, 'Sales1', 'Valve', 4), (4, 'Sales2', 'Bracket', 2), (5, 'Sales2', 'Wheel', 5), (6, 'Sales2', 'Seat', 5); -- View the 6 rows in the table SELECT * FROM Sales;
Grant read access on the table to each of the users.
GRANT SELECT ON Sales TO Manager; GRANT SELECT ON Sales TO Sales1; GRANT SELECT ON Sales TO Sales2;
Create a new schema, and an inline table valued function. The function returns 1 when a row in the SalesRep column is the same as the user executing the query (
@SalesRep = USER_NAME()) or if the user executing the query is the Manager user (
USER_NAME() = 'Manager').
CREATE SCHEMA Security; GO CREATE FUNCTION Security.fn_securitypredicate(@SalesRep AS sysname) RETURNS TABLE WITH SCHEMABINDING AS RETURN SELECT 1 AS fn_securitypredicate_result WHERE @SalesRep = USER_NAME() OR USER_NAME() = 'Manager';
Create a security policy adding the function as a filter predicate. The state must be set to ON to enable the policy.
CREATE SECURITY POLICY SalesFilter ADD FILTER PREDICATE Security.fn_securitypredicate(SalesRep) ON dbo.Sales WITH (STATE = ON);
Now test the filtering predicate, by selected from the Sales table as each user.
EXECUTE AS USER = 'Sales1'; SELECT * FROM Sales; REVERT; EXECUTE AS USER = 'Sales2'; SELECT * FROM Sales; REVERT; EXECUTE AS USER = 'Manager'; SELECT * FROM Sales; REVERT;
The Manager should see all 6 rows. The Sales1 and Sales2 users should only see their own sales.
Alter the security policy to disable the policy.
ALTER SECURITY POLICY SalesFilter WITH (STATE = OFF);
Now the Sales1 and Sales2 users can see all 6 rows.
B. Scenario for users who connect to the database through a middle-tier application
This example shows how a middle-tier application can implement connection filtering, where application users (or tenants) share the same SQL Server user (the application). The application sets the current application user ID in SESSION_CONTEXT (Transact-SQL) after connecting to the database, and then security policies transparently filter rows that shouldn't be visible to this ID, and also block the user from inserting rows for the wrong user ID. No other app changes are necessary .
Create a simple table to hold data.
CREATE TABLE Sales ( OrderId int, AppUserId int, Product varchar(10), Qty int );
Populate the table with 6 rows of data, showing 3 orders for each application user.
INSERT Sales VALUES (1, 1, 'Valve', 5), (2, 1, 'Wheel', 2), (3, 1, 'Valve', 4), (4, 2, 'Bracket', 2), (5, 2, 'Wheel', 5), (6, 2, 'Seat', 5);
Create a low-privileged user that the application will use to connect.
-- Without login only for demo CREATE USER AppUser WITHOUT LOGIN; GRANT SELECT, INSERT, UPDATE, DELETE ON Sales TO AppUser; -- Never allow updates on this column DENY UPDATE ON Sales(AppUserId) TO AppUser;
Create a new schema and predicate function, which will use the application user ID stored in SESSION_CONTEXT to filter rows.
CREATE SCHEMA Security; GO CREATE FUNCTION Security.fn_securitypredicate(@AppUserId int) RETURNS TABLE WITH SCHEMABINDING AS RETURN SELECT 1 AS fn_securitypredicate_result WHERE DATABASE_PRINCIPAL_ID() = DATABASE_PRINCIPAL_ID('AppUser') AND CAST(SESSION_CONTEXT(N'UserId') AS int) = @AppUserId; GO
Create a security policy that adds this function as a filter predicate and a block predicate on
Sales. The block predicate only needs AFTER INSERT, because BEFORE UPDATE and BEFORE DELETE are already filtered, and AFTER UPDATE is unnecessary because the
AppUserId column cannot be updated to other values, due to the column permission set earlier.
CREATE SECURITY POLICY Security.SalesFilter ADD FILTER PREDICATE Security.fn_securitypredicate(AppUserId) ON dbo.Sales, ADD BLOCK PREDICATE Security.fn_securitypredicate(AppUserId) ON dbo.Sales AFTER INSERT WITH (STATE = ON);
Now we can simulate the connection filtering by selecting from the
Sales table after setting different user IDs in SESSION_CONTEXT. In practice, the application is responsible for setting the current user ID in SESSION_CONTEXT after opening a connection.
EXECUTE AS USER = 'AppUser'; EXEC sp_set_session_context @key=N'UserId', @value=1; SELECT * FROM Sales; GO -- Note: @read_only prevents the value from changing again -- until the connection is closed (returned to the connection pool) EXEC sp_set_session_context @key=N'UserId', @value=2, @read_only=1; SELECT * FROM Sales; GO INSERT INTO Sales VALUES (7, 1, 'Seat', 12); -- error: blocked from inserting row for the wrong user ID GO REVERT; GO
CREATE SECURITY POLICY (Transact-SQL)
ALTER SECURITY POLICY (Transact-SQL)
DROP SECURITY POLICY (Transact-SQL)
CREATE FUNCTION (Transact-SQL)
Create User-defined Functions (Database Engine)