Hello Right now, slapd ignore attribute ACL when performing an add operation. If you have privilegied users that can add entries, it means that you can prevent them from modifying attributes in existing entries, but you cannot prevent them from adding an entry with a read-only attribute. The problem can be interesting with an attribute such as authzTo, where the whole access control can be circumvented by any user that can create an entry in the tree. IMO this behavior was not intended, but if it was, then it should be clearly documented. Below is a patch that cause attribute ACL to be checked for add operations. It is done in the backend, so if it is acceptable, then I will have to do it for other backends. I wonder if the modrdn operation shoulnd't be subject to the same sanity checks. Any thought? Does it look right? diff -U2 -r1.174 add.c --- servers/slapd/back-bdb/add.c 26 Aug 2008 23:45:35 -0000 1.174 +++ servers/slapd/back-bdb/add.c 27 Sep 2008 15:54:58 -0000 @@ -300,4 +300,22 @@ } + /* + * Check ACL for attribute write access + */ + if (!acl_check_modlist(op, oe, op->ora_modlist)) { + switch( opinfo.boi_err ) { + case DB_LOCK_DEADLOCK: + case DB_LOCK_NOTGRANTED: + goto retry; + } + + Debug( LDAP_DEBUG_TRACE, + LDAP_XSTRING(bdb_add) ": no write access to attribute\n", + 0, 0, 0 ); + rs->sr_err = LDAP_INSUFFICIENT_ACCESS; + rs->sr_text = "no write access to attribute"; + goto return_results;; + } + if ( eid == NOID ) { rs->sr_err = bdb_next_id( op->o_bd, &eid ); -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz manu(a)netbsd.org

On Sep 27, 2008, at 8:59 AM, Emmanuel Dreyfus wrote: > Hello > > Right now, slapd ignore attribute ACL when performing an add > operation. I note that this is the expected behavior, been so for many, many years.

> If you have privilegied users that can add entries, it means that > you can > prevent them from modifying attributes in existing entries, but you > cannot > prevent them from adding an entry with a read-only attribute. > > The problem can be interesting with an attribute such as authzTo, > where the > whole access control can be circumvented by any user that can create > an > entry in the tree. IMO this behavior was not intended, but if it was, It was. Likewise the behavior of rename. > then it should be clearly documented. I recall it being noted somewhere in the documents, but likely not as clear as it should be. I recall discussing this ACL/authzTo issue long ago. > Below is a patch that cause attribute ACL to be checked for add > operations. > It is done in the backend, so if it is acceptable, then I will have > to do it > for other backends. I wonder if the modrdn operation shoulnd't be > subject to > the same sanity checks. > > Any thought? Does it look right? I have no opinion as whether this change should be made or not. I'm merely providing some background information.... -- Kurt > > diff -U2 -r1.174 add.c > --- servers/slapd/back-bdb/add.c 26 Aug 2008 23:45:35 > -0000 1.174 > +++ servers/slapd/back-bdb/add.c 27 Sep 2008 15:54:58 -0000 > @@ -300,4 +300,22 @@ > } > > + /* > + * Check ACL for attribute write access > + */ > + if (!acl_check_modlist(op, oe, op->ora_modlist)) { > + switch( opinfo.boi_err ) { > + case DB_LOCK_DEADLOCK: > + case DB_LOCK_NOTGRANTED: > + goto retry; > + } > + > + Debug( LDAP_DEBUG_TRACE, > + LDAP_XSTRING(bdb_add) ": no write access to > attribute\n", > + 0, 0, 0 ); > + rs->sr_err = LDAP_INSUFFICIENT_ACCESS; > + rs->sr_text = "no write access to attribute"; > + goto return_results;; > + } > + > if ( eid == NOID ) { > rs->sr_err = bdb_next_id( op->o_bd,&eid ); > > -- > Emmanuel Dreyfus > http://hcpnet.free.fr/pubz > manu(a)netbsd.org