Thread

1. Fwd: Bug with groups and access permissions (7.0.0RC1): more information

   Guillaume Perréal <perreal@lyon.cemagref.fr> — 2000-05-24T10:22:39Z

   Well, I have more information about theses bugs:
   
   > It seems there is a 'referential integrity' bug with groups and permissions:
   > 
   > db=> CREATE USER user1;
   > CREATE USER
   > db=> CREATE GROUP group1 WITH SYSID 4 USER user1;
   > CREATE GROUP
   > db=> CREATE TABLE table1 (field1 integer PRIMARY KEY);
   > CREATE
   > db=> REVOKE ALL ON table1 FROM PUBLIC;
   > CHANGE
   > db=> GRANT ALL ON table1 TO GROUP group1;
   > CHANGE
   > db=> DROP GROUP group1;
   > DROP GROUP
   > \z
   > NOTICE:  get_groname: group 4 not found
   > The connection to the server was lost. Attempting reset: Failed.
   > !=>
   
   It seems that dropping the group don't delete it from the tables permissions
   (acl).
   Then, when Postgresql try to display them, it can't find the name of the dropped
   group.
   
   > db=> DROP USER userx;
   > ERROR: group groupy does not exist.
   > db=> SELECT * from PG_GROUP;
   > ...
   > groupy | 5 | {1,4}
   > ...
   > db=> DROP GROUP groupy;
   > DROP
   > 
   
   Apparently, this tied with databases:
   
   - Create a group 'foo' in the database 'foodb':
   	\c foodb postgres
   	CREATE GROUP foo;
   
   - Add the user 'userfoo' in this group:
   	ALTER GROUP foo ADD USER 'userfoo;'
   
   - Reconnect to database 'bardb':
   	\c bardb
   
   - Trying to drop the user 'userfoo' leads to "group foo does not exist":
   	DROP USER userfoo;
   
   But either dropping 'userfoo' from 'foodb' or dropping group 'foo' from 'bardb'
   works! 
   	
   Guillaume Perréal - Stagiaire MIAG
   Cemagref (URH), Lyon, France
   Tél: (+33) 4.72.20.87.64
   
   

2. Re: [GENERAL] Fwd: Bug with groups and access permissions (7.0.0RC1): more information

   Tom Lane <tgl@sss.pgh.pa.us> — 2000-05-24T15:01:55Z

   This stuff all seems to be fixed in current sources.  There may be a
   post-release patch or two involved, but the bulk of the fixes are in
   7.0 final.
   
   			regards, tom lane