[openib-general] RE: osmtest/OpenSM: ServiceGID and busy status

[Hal Rosenstock]

On Mon, 2005-09-12 at 11:53, Eitan Zahavi wrote:
> Hal Rosenstock wrote:
> 
> >>
> >>Proxy would be allowed. There are 2 possibilities:
> >>1. Allow valid looking GIDs
> >>or
> >>2. Only allow GIDs present in the subnet
> > 
> > 
> > I think that C15-0.0.1.13 (IBA 1.2 p.896) results in the second
> > possibility above as it states:
> > 
> > C15-0.1.13: SA shall reject as invalid any attempt to create, modify, or
> > delete a ServiceRecord in which the ServiceP_Key is not present in the
> > P_Key Tables of both the port identified by the ServiceGID and the port
> > from which the request came. 
> > 
> Thanks for finding/pointing it. I am not sure this is covered in the
> implementation as of today. As it would cause the osmtest requests to be dropped.
> 
> Let us add it to the todo list.
> 
> Thanks
> > So that is more stringent and the ServiceGID must be a (valid) GID in
> > the subnet.
> 
> > 
> > I believe the test ais ttempting to create the SR with an invalid
> > ServiceGID as its subnet prefix is 0. It should be rejected by the SA
> > per the rule above.

In terms of the SA code, it does not enforce the PKey sharing as
follows:

  /* if the pkey given is an invalid pkey - return TRUE. */
  if(ib_pkey_is_invalid(pkey))
  {
    osm_log( p_log, OSM_LOG_DEBUG,
             "osm_physp_has_pkey: "
             "Given invalid PKey - we treat it loosely and allow it.\n");
    res = TRUE;
    goto Exit;
  }

Wonder what relies on this (looseness)...

-- Hal