From the PostgreSQL docs:
The concept of roles subsumes the concepts of "users" and "groups". In
PostgreSQL versions before 8.1, users and groups were distinct kinds
of entities, but now there are only roles. Any role can act as a user,
a group, or both.
Why did they make this change in 8.1?
Perhaps it's easier from the C coders point of view, with a single Role class (struct)?
More details:
CREATE USER is equivalent to CREATE ROLE except that CREATE USER gives the LOGIN permission to the user/role.
(I'm about to design a permission system for my webapp, hence I'm interested in this.)
The merge has many advantages and no disadvantages. For instance, you can now seamlessly convert a "user" to a "group" and vice versa by adding / removing the LOGIN privilege.
ALTER ROLE myrole LOGIN;
ALTER ROLE myrole NOLOGIN;
Or you can GRANT membership in any other login ("user") or non-login role ("group") to a role:
GRANT joe TO sue;
You can still:
CREATE USER james;
That's just a role with login privilege now. Or:
CREATE GROUP workers;
That's effectively the same as CREATE ROLE now.
The manual has it all.
I found this thread in the PostgreSQL-Hackers list, from June 6, 2003, that in the end suggests that users and groups and roles be consolidated. (Thanks Craig Ringer for suggesting that I check the pgsql-hackers list archives.)
Here are some benefits mentioned (those that I found).
allow groups to have groups as members
the ACL code would be simplified
the GRANT/REVOKE syntax and the display format for ACL lists could be
simplified, since there'd be no need for a syntactic marker as to
whether a given name is a user or a group.
In some circumstances I could see it making sense to allow logging in
directly as a group/role/whatchacallit
This would also solve the problem that information_schema views will
show only owned objects
[makes it easier to] representing privileges granted to groups [since
you'd simply reuse the role related code?]
From the manual:
The SQL standard defines the concepts of users and roles, but it
regards them as distinct concepts and leaves all commands defining
users to be specified by each database implementation. In PostgreSQL
we have chosen to unify users and roles into a single kind of entity.
Roles therefore have many more optional attributes than they do in the
standard.
Having a distinction between users and groups doesn't gain you anything.
AFAIK the motivation for changing it was to simplify uses like:
One user masquerading as another, eg a superuser simulating a reduced permissions user. With unified roles this becomes just another change of current role, no different to changing primary group.
Groups that are members of other groups to implement granular access permissions.
If you want the details, though, you're best off checking out the archives of the pgsql-hackers list for the period, and the git history (converted from CVS).
Related
I am new to Postgres, sorry if the question is basic:)
I need to create a personalized role for any customer already existing in the DB.
Role name must be client_{first_name}_{last_name} (without curly brackets).
Also, this customer can only access his own data in "table_1" and "table_2" tables.
I took ALICE STEWART (id 51).
What I did:
I created role Customer with Select privilege. Then tried to create role for ALICE STEWART and I get an infinite recursion error.
But how to create role for one customer so that he could access only his info in 2 tables?
How to do that properly?
There is nothing wrong with personalized database users, but I would only consider that if you don't have very many users. I wouldn't want to deal with a pg_authid that contains 100000 users.
I would not play this with permissions, but with row level security. The exact way in which that would work depends on your data. The easiest way is to have and owner column that lists the user that can see the data.
Watch out for overly complicated policies, as they will make your queries slow.
Snowflake follows the role-based access control (RBAC) paradigm. Best practice for RBAC is, to have functional and access roles managing either user and clients or access privileges. This creates in worst-case a variety of roles that inherits permissions from and to each other. By nature, one can easily lose sight.
In snowflake, grants to roles and users are stored in ACCESS_USAGE.GRANTS_TO_ROLES and ACCESS_USAGE.GRANTS_TO_USERS. What is a proper approach to identify the data stewards/owner of a role automatically (if not labeled explicitly in a 3rd party tooling)?
Options I thought of:
recursive lookup of OWNERSHIP privileges of roles of roles (will generate a lot of false positives)
recursive discovery of a service account that has advanced permission to a role and lookup the service account owner
lookup over usage pattern of executed queries (might be actually more consumers than producers)
A couple of options:
Populate the role’s comment field with the relevant Data Steward information
Use Tags (in public preview)
I'd like to ask Postgres experts about a confusing sentence which I found in the documentation. In particular, I refer to the GRANT command, where the documentation states:
If the role executing GRANT holds the required privileges indirectly via more than one role membership path, it is unspecified which containing role will be recorded as having done the grant. In such cases it is best practice to use SET ROLE to become the specific role you want to do the GRANT as.
If I understand correctly, this is related to role inheritance. In particular, you might have a role C which inherits a permission from both role A and role B. In this case, if a user has role C and grants the permission, then Postgres might non-deterministically stipulate that the permission was granted by either A or B. To avoid this ambiguity, the user can issue SET ROLE A or SET ROLE B to force a lesser role and clarify how the granting should occur.
Some questions about this and a more general one:
Is this reading correct or does the sentence mean something different?
What if, in the prior example, a user has the permission via A and C (as opposed to A and B)? In that case, the recommended practice of using "SET ROLE to become the specific role you want to do the GRANT as" does not seem to help, because setting the role to C still leaves an ambiguity due to inheritance.
More generally, is there any good reference documentation where the role system of Postgres is compared against traditional SQL grant diagrams? Some parts of the Postgres implementation are not straightforward and I'd like to learn more about it, which is complicated to do via experiments alone.
Thanks in advance.
Yes, in my opinion, you are right:
You understand it correctly.
It is no matters if the role that grants a permission has innerited the privilege or not. For each permission that you have postgress registers the role that granted the privilege.
I don't know. If you find it, please post a comment.
What is the differences beween roles and users.
A role can have many users. An example might be an Admin role, which could mean someone who's assigned and their backup. Both would have the same rights, which are embodied in the role. Individuals can come and go, but roles remain.
A role typically defines a business function (or set of functions) performed by one or more users. Examples would be 'customer service agent' or 'business analyst'. A user is an individual person who is included in the role - Bob, Nancy, and Steve might be assigned to the customer service agent role.
This makes is easier to assign permissions to database objects. You can assign permissions to the role, and any user who belongs to that role inherits the same set of permissions.
On a technical level, see other answers. On a practical level, when you have large user set with fluid permissioning needs due to changing roles, the difference is that assigning per-user permissions means any change in user responsibilities necessitates permissions changes on MULTIPLE database objects that user needs to be added to/removed from perms.
Whereas if the perms are assigned to the roles, the only change is in the role membership.
The latter is both significantly less resource taxing on DBAs, and due to less work needed, significantly less likely to suffer from operator error (e.g. less work to do => less chance to screw it up) and thus more secure.
I was trying to create new role that will have all privileges of the PUBLIC role and then remove all of the privileges from PUBLIC role afterwards. This is for security purposes.
This is the problem. I couldn't grant SYS./1005bd30_LnkdConstant, and others with the same format, to my new role.
sample:
SYS./10076b23_OraCustomDatumClosur
SYS./100c1606_StandardMidiFileRead
ORDSYS./1013c29d_PlanarImageServerPro
.
.
.
Do I really need these or my new "public" role can do without those?
Any help will be much appreciated.
Let me take a quick guess here. The problem you are having is that the object names are case sensitive. The quick fix is to enclose the object names in double quotes, like this.
GRANT EXECUTE ON SYS."/1005bd30_LnkdConstant" TO mynewpublicrole;
You indicate that you "couldn't grant [EXECUTE ON] SYS./1005bd30_LnkdConstant" to a role.
I take that to mean that when you ran the GRANT statement, Oracle raised an exception, most likely,
ORA-00903: invalid table name
Enclosing the objectname in double quotes (as shown in the example) should fix that problem.
It's not possible to answer the question whether your new role needs EXECUTE privilege on those objects or not. Well, the role doesn't necessarily need them. The question is whether the user needs them or not (whether granted directly, or granted indirectly through roles.) That can be determined through thorough testing.
Some other comments.
If your intention is to create a new role and grant that role to all users, I don't see that security is changed or improved. So, I'm going to assume that is not the case.
It would appear you are trying to apply the principle of "least privilege". I applaud that effort.
One of the most common patterns I see application developers follow is to have the application connect to the database as the owner of the schema objects. What that means is that the application has all sorts of privileges it probably doesn't need, e.g. DROP TABLE, ALTER PROCEDURE, etc.
The pattern we use is to have an "OWNER" user that owns the schema objects, and a separate "APP" user that has specific privileges it needs on the "OWNER" objects, and synonyms for the "OWNER" objects. (The synonyms allow the OWNER.object to be referenced without being qualified with the OWNER.) It almost goes without saying, we do not grant privileges to PUBLIC, we grant to roles where needed.
I mention this because it's a pattern we use for implementing the principle of "least privilege".
For other security concerns, I recommend you review the "Oracle Security Checklist" white paper:
http://www.oracle.com/technology/deploy/security/database-security/pdf/twp_security_checklist_database.pdf
Some other possible exceptions you may have encountered when executing the GRANT statement:
ORA-01031: insufficient privileges
or
ORA-04042: procedure, function, package, or package body does not exist.
In either of those cases, make sure you connect to the database as SYS (SYSDBA) to grant the privileges. We almost always grant privileges as the owner of the object, rather than have some other user as the GRANTEE. I almost never use the "WITH GRANT OPTION" on object privileges. It's a simpler model, and avoids any potential problem with dependency trees.