Re: [PATCH 0/5] target: Add se_session->sess_kref + convert iscsi/qla2xxx fabrics

["Nicholas A. Bellinger" <nab@xxxxxxxxxxxxxxx>]

On Mon, 2012-02-27 at 09:47 -0800, Roland Dreier wrote:
> On Mon, Feb 27, 2012 at 4:12 AM, Nicholas A. Bellinger
> <nab@xxxxxxxxxxxxxxx> wrote:
> > This series is a WIP for adding se_session->sess_kref into target core to
> > address a long standing set of issues wrt to active session shutdown and
> > explict NodeACL group configuration.
> >
> > This includes changing iscsi-target to co-exist with internal reference
> > counting, and converting qla2xxx internal qla_tgt_sess->sess_kref to use
> > se_session->sess_kref for all cases.  It also re-factors qla_tgt_sess
> > shutdown -> release code to run outside of qla_hw_data->hardware_lock +
> > interrupt execution using system_wq process context.
> >
> > So far the changes have been tested with normal session creation/deletion
> > for iscsi-target, loopback, tcm_fc(fcoe) and qla2xxx WWPNs, but need
> > more testing with active I/O shutdown.  Please have a look..
> 
> I was thinking about this stuff over the weekend, and I'm starting to wonder
> if the core notion of "session" is a good match for the way qla2xxx uses it.
> 
> Recall that qla2xxx creates a "global" session when firmware tells it about
> an initiator logging in, and a "local" session when it gets a command from
> someone that it doesn't have a global session for yet.
> 
> [The creation of local sessions is obviously racy if multiple commands
> come in before any session is created, but that's not the main point
> here]
> 
> In any case, if we tear down the TCM session structure when we remove
> ACLs for an initiator talking to qla2xxx, that doesn't log the initiator out
> on the FC side, and it doesn't stop the initiator from continuing to send
> commands (and triggering creation of a new local session).
> 

As per current generate_node_acls=1 usage, the session is restarted
after explict NodeACL removal and only has access to read-only LUN=0
RAMDISK_MCP.

> For qla2xxx at least, ACLs are really about which LUNs an initiator sees.
> Even if an initiator has no LUNs mapped to it, it can still do a PLOGI into
> us etc.  In brief I guess I wonder if tcm_qla2xxx sessions should be more
> like tcm_loop sessions, instead of trying to act like iSCSI.  (and maybe
> we need to make the core stuff more generic, and leave the iSCSI
> semantics in the iSCSI fabric module...)
> 

I'm not sure what you mean here wrt to tcm_loop, as sessions are
generated from a configfs attribute and mapped 1:1 with target
endpoints.

Any more thoughts here..?

> I should probably read the tcm_fc code and see what it does too...
> 

tcm_fc currently uses it's own session referencing that can be converted
to se_sess->sess_kref usage without too much effort.

--nab


--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html