5 minutes: Grant Role to Package

The EOUC (EMEA Oracle User Group Community) hosts a special session on sunday during Oracle Open World. This session contains twelve things about Oracle 12c. Each of these 12 things is hosted by a different speaker, so this means that each section is only five minutes long (Strictly guarded by Debra Lilley).
Thankfully I was chosen to be one of the speaker during this special session, very excited and also very scared. Five minutes is not a whole lot to tell the story of PL/SQL security enhancements in 12c.

This blogpost will cover my five minutes of fame.

The thing that I will cover is Code Based Access Control (or actually only a small example where you grant privileges to a package).

1. Setting the stage

For this example I will use Schema A which will contain two tables (T1 and T2) and a package (PKG).
The structure of the tables is irrelevant, and so is the actual implementation of the package.

2. The Package's Objective

The actual implementation of the package is not really relevant, what is important to know is that it would query table T1 and would manipulate some data of table T2. However it should work on schema A's table T1 and on the calling schema's table T2.
The reason that schema A has a table T2 is that it would be impossible to compile the package if table T2 was missing. Table T2 only acts as a template table in schema A.
The calling schema should have a table named T2 in order for the package to be able to work.

note: the calling schema is the schema which would call the package that schema A has defined.

The package (located in schema A) should therefor be compiled with Invoker Rights (authid current_user) as opposed to Definer Rights (authid definer).
The reference to table T1 should be fully qualified (a.t1) and the reference to table T2 should not be fully qualified (t2).
Because of the Invoker Rights specification on the package PKG the table T2 of the calling schema is used. And because of the full qualification of table T1 (a.T1) the correct table is referenced (the one in schema A).
For a calling schema to be able to use this package (PKG) - let's assume the calling schema is called B - EXECUTE privileges need to be granted to schema B

grant execute on pkg to b;

3. The Old way (< 2013) and the flaw

Before the release of Oracle 12c, only granting execute privileges on package PKG to schema B would not be sufficient.
When invoking the package:

begin
   a.pkg;
end;

there would be an exception raised because of the reference to table T1 which resides in schema A.

ORA-00942: table or view does not exist

What you needed to do was to grant privileges on schema A's table T1 as well.

grant select on t1 to b;

Now all works fine. The package can be executed from schema B, and they lived happily ever after.
So, what is the security issue here?
Not only can the package be executed from schema B, the table T1 (in schema A) can also be queried directly without using the package. What if there are columns in table T1 that contain sensitive information?

4. Fixing the flaw

The solution is to grant privileges to the package and not to the user.
First of all: revoke the select privileges on schema A's T1 from schema B

revoke select on t1 from b

Next create a role and grant the select privileges on schema A's table T1 to that role. What you couldn't do before was grant a role to a package, but now you can.

create role only_this_role;

grant select on a.t1 to only_this_role;

grant only_this_role to package a.pkg;

And that's it.
The execute privileges on the package still exists of course otherwise schema B wouldn't be able to execute the package.
Table T1 (in schema A) can not be called directly by schema B any longer.
Now only the package can make calls to table T1.
This adheres to the concept of least privileges.

Order, Order.. Sorting Happens Last

While preparing for my session at Oracle Open World on "Oracle 12c for Developers" I ran into a little remarkable thing.

When sorting a dataset, the sorting is always done last. That is what I was taught anyway. There is probably some obscure way to detect the exact execution plan, but personally I never bothered to go and investigate.
When I was created some test scripts I found there was a way to see that sorting really happens as the final step.

Oracle 12c made some enhancements in the Subquery Factoring clause. Specifically now it is possible to create Functions and Procedures in the WITH clause.
Something like the following:

SQL> with
  2     function formatname (p_name in varchar2)
  3        return varchar2
  4     is
  5     begin
  6        return initcap (p_name);
  7     end formatname;
  8  select ename
  9        ,formatname(ename)  formatted
 10    from emp;

ENAME      FORMATTED
---------- -------------------------
SMITH      Smith
ALLEN      Allen
WARD       Ward
JONES      Jones
MARTIN     Martin
...

After the WITH keyword (first line) a simple function is defined (lines 2 - 7). This function is used in the SELECT part of the query (line 9).

When you want to include a named query in the WITH clause, it has to be after the Procedures and Functions that you have defined.
In the following query, I included a named query called ordered_emps which is a resultset of the EMP table sorted by ENAME.

SQL> set serveroutput on
SQL> with
  2    procedure show (p_what in varchar2)
  3    is
  4    begin
  5      dbms_output.put_line ('input is: '||p_what);
  6    end show;
  7    function formatname (p_name in varchar2)
  8      return varchar2
  9    is
 10    begin
 11       show (p_name);
 12      return initcap (p_name);
 13    end formatname;
 14  ordered_emps as
 15    (select ename from emp order by ename asc)
 16  select ename
 17        ,formatname(ename) formatted
 18    from ordered_emps
 19  /
 
 ENAME       FORMATTED
---------- -----------
ADAMS      Adams
ALLEN      Allen
BLAKE      Blake
CLARK      Clark
FORD       Ford
JAMES      James
JONES      Jones
KING       King
MARTIN     Martin
MILLER     Miller
SCOTT      Scott
SMITH      Smith
TURNER     Turner
WARD       Ward

14 rows selected.

input is: SMITH
input is: ALLEN
input is: WARD
input is: JONES
input is: MARTIN
input is: BLAKE
input is: CLARK
input is: SCOTT
input is: KING
input is: TURNER
input is: ADAMS
input is: JAMES
input is: FORD
input is: MILLER

After the WITH keyword, a procedure (SHOW) is defined which acts as a wrapper for DBMS_OUTPUT.PUT_LINE (lines 2 - 6). Next a function (FORMATNAME) is defined which calls the SHOW procedure with the given input and formats the name to INITCAP (lines 7-13). Next a named query is defined ORDERED_EMPS (lines 14 - 15), and finally the "real" query is specified calling the FORMATNAME function.

Because SERVEROUTPUT is turned on, you can see the results as produced by the query as well as the calls to the SHOW procedure.
The resultset by the query is sorted based on the ENAME as we expect, the calls to the function (and procedure) happen as the data is fetched from the table in no particular order which can be seen in the output from DBMS_OUTPUT.