Re: Oracle's Virtual Private Database functionality

[ Sorry if this appears twice on the list: sent it via a non-subscribed email the first time around. Of course, while waiting for it to appear I figured out a workaround; see the "Late breaking update" at the end. ]

In the spirit of "tell us what you're trying to do..."

I'd like to mimic a subset of Oracle's Virtual Private Database functionality (see http://dba-oracle.com/oracle_news/2004_1_16_vpd_security.htm for an overview) in Postgres: based on some per-connection setting, I'd like a query to return a different set of rows. In VPD, the Oracle engine actually applies a defined predicate (say, country_code='USA') to every query. The idea is that a given set of users can only see rows in a table that match this predicate, while the other rows are invisible to them.

Now for the "how I tried to do it" part...

I thought I was on my way to doing this in Postgres by making use of schemas and search_path:

CREATE USER user1 UNENCRYPTED PASSWORD 'user1';
CREATE SCHEMA AUTHORIZATION user1;

CREATE SCHEMA canada;
GRANT USAGE ON SCHEMA canada TO user1;
CREATE TABLE canada.row_limiter (country_code VARCHAR(3));
INSERT INTO canada.row_limiter VALUES('CAN');
GRANT SELECT ON canada.row_limiter TO user1;

CREATE SCHEMA usa;
GRANT USAGE ON SCHEMA usa TO user1;
CREATE TABLE usa.row_limiter (country_code VARCHAR(3));
INSERT INTO usa.row_limiter VALUES('USA');
GRANT SELECT ON usa.row_limiter TO user1;

\CONNECT - user1

SET SEARCH_PATH TO '$user',canada;

CREATE TABLE my_data (country_code VARCHAR(3), data NUMERIC);
INSERT INTO my_data VALUES('USA',11);
INSERT INTO my_data VALUES('USA',12);
INSERT INTO my_data VALUES('USA',13);

INSERT INTO my_data VALUES('CAN',21);
INSERT INTO my_data VALUES('CAN',22);
INSERT INTO my_data VALUES('CAN',23);

CREATE VIEW my_data_v AS
SELECT md.*
FROM my_data md,
row_limiter rl
WHERE rl.country_code=md.country_code;

SELECT * FROM my_data_v;

-- Looks great - I only see Canadian data!!
-- country_code | data
----------------+------
-- CAN | 21
-- CAN | 22
-- CAN | 23

SET SEARCH_PATH TO '$user',usa;

SELECT * FROM my_data_v;

-- Darn, I still only see Canadian data :-(
-- country_code | data
----------------+------
-- CAN | 21
-- CAN | 22
-- CAN | 23

\d my_data_v

View definition:
SELECT md.country_code, md.data
FROM my_data md, CANADA.row_limiter rl -- <--ah, and here's the reason...
WHERE rl.country_code::text = md.country_code::text;

It's apparent why: the view determines which table it's going to use at view creation time, not at query time, so this method is no good.

Is there a "right way" to accomplish what I'm trying to do?

Late breaking update:

Looks like a function returning SETOF does the trick, so it must not resolve the table early like the view does:

CREATE OR REPLACE FUNCTION my_data_f() RETURNS SETOF my_data AS '
SELECT md.country_code,
md.data
FROM my_data md, row_limiter rl
WHERE rl.country_code = md.country_code;
' LANGUAGE sql;

SET SEARCH_PATH TO '$user',usa;
SELECT * FROM my_data_f();
country_code | data
--------------+------
USA | 11
USA | 12
USA | 13

SET SEARCH_PATH TO '$user',canada;
SELECT * FROM my_data_f();
country_code | data
--------------+------
CAN | 21
CAN | 22
CAN | 23

Can I rely on this behavior? Is this the best way to do what I'm after?2 to accomplish what I'm trying to do?

Late breaking update:

Looks like a function returning SETOF does the trick, so it must not resolve the table early like the view does:

CREATE OR REPLACE FUNCTION my_data_f() RETURNS SETOF my_data AS '
SELECT md.country_code,
md.data
FROM my_data md, row_limiter rl
WHERE rl.country_code = md.country_code;
' LANGUAGE sql;

SET SEARCH_PATH TO '$user',usa;
SELECT * FROM my_data_f();
country_code | data
--------------+------
USA | 11
USA | 12
USA | 13

SET SEARCH_PATH TO '$user',canada;
SE

Check out CREATE RULE command,

Regards,
Qingqing

""Doug Bloebaum"" <blabes(at)myrealbox(dot)com>
news:1110552518(dot)1f7b963cblabes(at)myrealbox(dot)com(dot)(dot)(dot)
> [ Sorry if this appears twice on the list: sent it via a non-subscribed
email the first time around. Of course, while waiting for it to appear I
figured out a workaround; see the "Late breaking update" at the end. ]
>
> In the spirit of "tell us what you're trying to do..."
>
> I'd like to mimic a subset of Oracle's Virtual Private Database
functionality (see
http://dba-oracle.com/oracle_news/2004_1_16_vpd_security.htm for an
overview) in Postgres: based on some per-connection setting, I'd like a
query to return a different set of rows. In VPD, the Oracle engine actually
applies a defined predicate (say, country_code='USA') to every query. The
idea is that a given set of users can only see rows in a table that match
this predicate, while the other rows are invisible to them.
>
> Now for the "how I tried to do it" part...
>
> I thought I was on my way to doing this in Postgres by making use of
schemas and search_path:
>
> CREATE USER user1 UNENCRYPTED PASSWORD 'user1';
> CREATE SCHEMA AUTHORIZATION user1;
>
> CREATE SCHEMA canada;
> GRANT USAGE ON SCHEMA canada TO user1;
> CREATE TABLE canada.row_limiter (country_code VARCHAR(3));
> INSERT INTO canada.row_limiter VALUES('CAN');
> GRANT SELECT ON canada.row_limiter TO user1;
>
> CREATE SCHEMA usa;
> GRANT USAGE ON SCHEMA usa TO user1;
> CREATE TABLE usa.row_limiter (country_code VARCHAR(3));
> INSERT INTO usa.row_limiter VALUES('USA');
> GRANT SELECT ON usa.row_limiter TO user1;
>
> \CONNECT - user1
>
> SET SEARCH_PATH TO '$user',canada;
>
> CREATE TABLE my_data (country_code VARCHAR(3), data NUMERIC);
> INSERT INTO my_data VALUES('USA',11);
> INSERT INTO my_data VALUES('USA',12);
> INSERT INTO my_data VALUES('USA',13);
>
> INSERT INTO my_data VALUES('CAN',21);
> INSERT INTO my_data VALUES('CAN',22);
> INSERT INTO my_data VALUES('CAN',23);
>
> CREATE VIEW my_data_v AS
> SELECT md.*
> FROM my_data md,
> row_limiter rl
> WHERE rl.country_code=md.country_code;
>
> SELECT * FROM my_data_v;
>
> -- Looks great - I only see Canadian data!!
> -- country_code | data
> ----------------+------
> -- CAN | 21
> -- CAN | 22
> -- CAN | 23
>
> SET SEARCH_PATH TO '$user',usa;
>
> SELECT * FROM my_data_v;
>
> -- Darn, I still only see Canadian data :-(
> -- country_code | data
> ----------------+------
> -- CAN | 21
> -- CAN | 22
> -- CAN | 23
>
> \d my_data_v
>
> View definition:
> SELECT md.country_code, md.data
> FROM my_data md, CANADA.row_limiter rl -- <--ah, and here's the
reason...
> WHERE rl.country_code::text = md.country_code::text;
>
>
> It's apparent why: the view determines which table it's going to use at
view creation time, not at query time, so this method is no good.
>
> Is there a "right way" to accomplish what I'm trying to do?
>
> Late breaking update:
>
> Looks like a function returning SETOF does the trick, so it must not
resolve the table early like the view does:
>
> CREATE OR REPLACE FUNCTION my_data_f() RETURNS SETOF my_data AS '
> SELECT md.country_code,
> md.data
> FROM my_data md, row_limiter rl
> WHERE rl.country_code = md.country_code;
> ' LANGUAGE sql;
>
> SET SEARCH_PATH TO '$user',usa;
> SELECT * FROM my_data_f();
> country_code | data
> --------------+------
> USA | 11
> USA | 12
> USA | 13
>
> SET SEARCH_PATH TO '$user',canada;
> SELECT * FROM my_data_f();
> country_code | data
> --------------+------
> CAN | 21
> CAN | 22
> CAN | 23
>
> Can I rely on this behavior? Is this the best way to do what I'm after?2
to accomplish what I'm trying to do?
>
> Late breaking update:
>
> Looks like a function returning SETOF does the trick, so it must not
resolve the table early like the view does:
>
> CREATE OR REPLACE FUNCTION my_data_f() RETURNS SETOF my_data AS '
> SELECT md.country_code,
> md.data
> FROM my_data md, row_limiter rl
> WHERE rl.country_code = md.country_code;
> ' LANGUAGE sql;
>
> SET SEARCH_PATH TO '$user',usa;
> SELECT * FROM my_data_f();
> country_code | data
> --------------+------
> USA | 11
> USA | 12
> USA | 13
>
> SET SEARCH_PATH TO '$user',canada;
> SE
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 9: the planner will ignore your desire to choose an index scan if your
> joining column's datatypes do not match
>