On 3/10/07, Kenneth Downs <ken@xxxxxxxxxx> wrote:
I tried this, but couldn't achieve what I wanted with it, it's also marked as "temp stuff could change in the future".
Alvaro Herrera wrote:David Legault escribió:
That's basically what I've done with my past questions on the ROLE system in
place. Since roles are global, I wanted it fine grained to the DB level so I
had to append DB_ in front of each role name and by using current_database()
inside my functions, I could hide that from the exterior.
Hmm, there used to be a facility to restrict users to specific
databases, enabled by db_user_namespace (not by default).
I tried this, but couldn't achieve what I wanted with it, it's also marked as "temp stuff could change in the future".
It seems to still work on 8.2 ...
there is also the 'samegroup' facility in pg_hba.conf. We create a group named after each database, and a person cannot get into a database unless they are in that group.
I append the name of the DB on front of the roles and it works flawlessly, you can't have 2 DB with the same name in the cluster. It's also transparent to the application since the functions to interact with the roles + other tables use current_database() which gets the name of the DB to which the groups are assigned.
All databases are REVOKED from public as all functions. When a user is created he is granted connect on the database to which he belongs. So even if someone gets the credentials to db1_user and attempts to connect to db2 with it it will fail.
David
