Re: [HACKERS] Permissions not working?

>
> Something seems to be wrong here unless I am doing something wrong that
> I can't see.  Here is a dump of one session.
>
> version
> -------------------------------------------------------------------------
> PostgreSQL 6.3.2 on i386-unknown-netbsd1.3B, compiled by gcc 2.7.2.2+myc1
>
> relname|relacl
> -------+--------------
> client |{"=","root=r"}
>
> getpgusername
> -------------
> carol
>
> vex=> select * from client;
> ERROR:  client: Permission denied.
>
> Exactly what I expected.  Now, here is another.

    I  assume user carol does not have usesuper set in pg_shadow.

> version
> --------------------------------------------------------------------------
> PostgreSQL 6.4.0 on i386-unknown-netbsd1.3.2, compiled by gcc 2.7.2.2+myc1
>
> relname|relacl
> -------+---------------
> client |{"=","db=arwR"}
>
> getpgusername
> -------------
> darcy
>
> trends=> select * from client;
> [Rows returned as if I had permissions]

    Here I assume user darcy has usesuper set in pg_shadow. Check
    and correct me if I'm wrong. The superuser flag is set if you
    allow darcy to create users on createuser time.

> Am I missing something or did table permission protections get lost on
> the way to 6.4?

    The only thing changed is  that  relations  accessed  due  to
    rewrite  rules get checked against the owner of the relation,
    the rules are fired for (ev_class attribute  of  pg_rewrite).
    This  was  only  done for read access due to rules in 6.3 and
    now does also check for append/write access since we open all
    rules to regular users in 6.4.

    Someone  can  now  setup a view from tables she has access to
    and then grant access to the view but does not need to  grant
    access to the tables the view is made of. This is how we make
    pg_user (a view) publicly readable but protect pg_shadow (the
    selected table) from public access.

    Or  someone  can  setup rules on insert, update and delete to
    one table (granted) that do logging of these  events  into  a
    log table (not granted).

    All  the  required  permissions are checked during the actual
    query rewriting.  Thus, later ACL changes will  correctly  be
    in effect. Example:

    Table t1 owner user_a granted select to user_b
    View  v1 owner user_b granted select to user_c

    user_c can select from v1

    Now user_a revokes select on t1 from user_b

    user_c gets 't1: permission denied' on select from v1

    But if user_b is a superuser (usesuper set)

    user_c can still select from v1


Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#======================================== jwieck@debis.com (Jan Wieck) #