As a running up to the new release of SQL 2016 I would like to share some research I did about the new features that will be available. Last time I talked about Dynamic Data Masking. This time I will discuss another very interesting new feature: Row Level Security. We know Row Level Security on Analysis Service level, but Microsoft now also implements it on Database Service level.
With the security we know in the current SQL versions (everything before SQL 2016) it is possible to grant or deny users access to select tables, schemas or to execute stored procedures, … What was not possible is to hide certain rows for a certain user. At least not without any joins to other tables. With this new feature in SQL it is possible. And through this post I would like to share how you can do this practically. Of course more detailed info can be found on Books Online. Personally I find this new feature very interesting because it can decrease your development time for reports. Because you don’t need to implement all the security each time in all your queries that you are using. You only implement the security once on database level. So my example is all about a DWH that stores the data of a supermarket. We will focus on a table that contains aggregated data.
So we have 2 dimensions (Departments & Sales Regions), 1 fact table (Sales) and 1 table that keeps track of who can see what (UserRights).
Below you can find the script to generate the tables. Also 3 logins are created to test our functions. As a personal best practice I like to keep everything that has to do with security in a separate Schema.
--1. Schema Creation CREATE SCHEMA DWH; GO CREATE SCHEMA [Security]; GO --2. Table Creation IF OBJECT_ID('DWH.DimDepartment') IS NOT NULL DROP TABLE DWH.DimDepartment CREATE TABLE DWH.DimDepartment( DepartmentID INT IDENTITY PRIMARY KEY, DepartmentName NVARCHAR(50) ) IF OBJECT_ID('DWH.DimSalesRegion') IS NOT NULL DROP TABLE DWH.DimSalesRegion CREATE TABLE DWH.DimSalesRegion( SalesRegionID INT IDENTITY PRIMARY KEY, SalesRegionName NVARCHAR(50) ) IF OBJECT_ID('DWH.FactSales') IS NOT NULL DROP TABLE DWH.FactSales CREATE TABLE DWH.FactSales( SalesID INT IDENTITY PRIMARY KEY, DepartmentID INT, SalesRegionID INT, SalesDateID INT, SalesAmount DECIMAL(18,8), CONSTRAINT FK_FactSales_DepartmentID FOREIGN KEY (DepartmentID) REFERENCES DWH.DimDepartment(DepartmentID), CONSTRAINT FK_FactSales_SalesRegionID FOREIGN KEY (SalesRegionID) REFERENCES DWH.DimSalesREgion(SalesRegionID) ) IF OBJECT_ID('Security.UserRights') IS NOT NULL DROP TABLE Security.UserRights CREATE TABLE Security.UserRights( UserRightsID INT IDENTITY PRIMARY KEY, UserLogin NVARCHAR(50), DepartmentID INT, SalesRegionID INT ) --3. Table Data INSERT INTO DWH.DimDepartment VALUES ('Ice Cream'), ('Soft Drinks'), ('Healthy Snacks') INSERT INTO DWH.DimSalesRegion VALUES ('Central Europe'), ('Eastern Europe'), ('Northern Europe'), ('Southern Europe'), ('Western Europe') INSERT INTO DWH.FactSales (DepartmentID, SalesRegionID, SalesDateID, SalesAmount) SELECT DepartmentID, SalesRegionID, DateID, ABS(CAST(NEWID() AS binary(6)) %1000) AS SalesAmount FROM DWH.DimDepartment CROSS JOIN DWH.DimSalesRegion CROSS JOIN (SELECT TOP 100 CONVERT(NVARCHAR, DATEADD(DAY,- ABS(CAST(NEWID() AS binary(6)) %1000) + 1, GETDATE()), 112) AS DateID FROM sys.columns) Dates ORDER BY DateID INSERT INTO Security.UserRights VALUES ('User1', 1, NULL), ('User2', NULL, 1), ('User3', 1, 1) --4. Create Users & Grant permissions CREATE USER User1 WITHOUT LOGIN; CREATE USER User2 WITHOUT LOGIN; CREATE USER User3 WITHOUT LOGIN; GRANT SELECT ON DWH.FactSales TO User1; GRANT SELECT ON DWH.FactSales TO User2; GRANT SELECT ON DWH.FactSales TO User3; GRANT SELECT ON DWH.DimDepartment TO User1; GRANT SELECT ON DWH.DimDepartment TO User2; GRANT SELECT ON DWH.DimDepartment TO User3; GRANT SELECT ON DWH.DimSalesRegion TO User1; GRANT SELECT ON DWH.DimSalesRegion TO User2; GRANT SELECT ON DWH.DimSalesRegion TO User3;
Our first step is to create a function that will tell us if the user can see the data (yes/no).
CREATE FUNCTION [Security].RowLevelSecurityPredicate(@DepartmentId INT, @SalesRegionID INT) RETURNS TABLE WITH SCHEMABINDING AS RETURN SELECT TOP 1 1 RowLevelSecurityPredicateResult FROM Security.UserRights WHERE (DepartmentID = @DepartmentID OR SalesRegionID = @SalesRegionID) AND UserLogin = USER_NAME(); GO
This function has 2 parameters, our department and sales region. By making use of these together with our logged on user, we filter our UserRights table. If this query returns data, it means the user can see the info. So 1 row is more than enough.
Next step is the new feature: implementing the security. Because the function alone, will not do anything.
CREATE SECURITY POLICY [Security].UserFilter ADD FILTER PREDICATE [Security].RowLevelSecurityPredicate(DepartmentID, SalesRegionID) ON DWH.FactSales, ADD FILTER PREDICATE [Security].RowLevelSecurityPredicate(DepartmentID, NULL) ON DWH.DimDepartment, ADD FILTER PREDICATE [Security].RowLevelSecurityPredicate(NULL, SalesRegionID) ON DWH.DimSalesRegion WITH (STATE = ON); GO
Let’s get in the code.
First line is like for any sql object telling what we want to make. For row-level-security this is the SECURITY POLICY object. Lines 2-3, 4-5 and 5-6 is your policy definition. For each table we want to filter we add a FILTER PREDICATE to tell which function can tell us what data can be shown or not. Your function accepts the columns from the table that you want to filter.
You can add as much PREDICATE’s as you want. The last is to enable or disable a security policy. This can be done with a separate statement too:
ALTER SECURITY POLICY [Security].UserFilter WITH (STATE = OFF); ALTER SECURITY POLICY [Security].UserFilter WITH (STATE = ON);
So what are the results?
EXECUTE AS USER = 'User1'; SELECT DepartmentId, SalesRegionId, SUM(SalesAmount) AS SalesAmount FROM DWH.FactSales GROUP BY DepartmentId, SalesRegionId; SELECT * FROM DWH.DimDepartment; SELECT * FROM DWH.DimSalesRegion; REVERT;
EXECUTE AS USER = 'User2'; SELECT DepartmentId, SalesRegionId, SUM(SalesAmount) AS SalesAmount FROM DWH.FactSales GROUP BY DepartmentId, SalesRegionId; SELECT * FROM DWH.DimDepartment; SELECT * FROM DWH.DimSalesRegion; REVERT;
EXECUTE AS USER = 'User3'; SELECT DepartmentId, SalesRegionId, SUM(SalesAmount) AS SalesAmount FROM DWH.FactSales GROUP BY DepartmentId, SalesRegionId; SELECT * FROM DWH.DimDepartment; SELECT * FROM DWH.DimSalesRegion; REVERT;
So as you can see, it works nicely! The user can only see the data he is allowed to see. What is even cooler about all of this, is that it not only works for SELECT statements but also for UPDATE. Personally I’m really looking forward to implement this new feature into the solutions I will develop at client side.