Message-Id: <1305835419.32066.20.camel@haakon2.linux-iscsi.org>
Date:	Thu, 19 May 2011 13:03:39 -0700
From:	"Nicholas A. Bellinger" <nab@...ux-iscsi.org>
To:	Bart Van Assche <bvanassche@....org>
Cc:	Roland Dreier <roland@...estorage.com>,
	Jason Gunthorpe <jgunthorpe@...idianresearch.com>,
	linux-kernel <linux-kernel@...r.kernel.org>,
	linux-scsi <linux-scsi@...r.kernel.org>,
	linux-rmda <linux-rdma@...r.kernel.org>,
	Vu Pham <vu@...lanox.com>, David Dillow <dillowda@...l.gov>,
	James Bottomley <James.Bottomley@...senpartnership.com>
Subject: Re: [RFC] ib_srpt: initial .40-rc1 drivers/infiniband/ulp/srpt
	merge

On Thu, 2011-05-19 at 12:40 +0200, Bart Van Assche wrote:
> On Thu, May 19, 2011 at 6:18 AM, Nicholas A. Bellinger
> <nab@...ux-iscsi.org> wrote:
> > The srpt_port->port_wwn patch in question does
> > current ensure that sport->enabled has been set via an configfs
> > attribute at:
> >
> >   /sys/kernel/config/target/srpt/$IB_PORT_GUID/tpgt_1/enable
> >
> > and will reject all SRP login attempts to an individual struct srpt_port
> > until the attribute has been explictly triggered.
> >
> > This allows ib_srpt to follow what is expected by rtsadmin-v2 +
> > lio-utils, and used to generate /etc/target/srpt_start.sh used to save
> > persistent fabric configuration.  Currently other fabrics like
> > iscsi-target and tcm_qla2xxx expect to be able to reject fabric login
> > requests before the full set of WWPN endpoints, LUNs, NodeACLs +
> > MappedLUNs have been recreated during an typical init.d/target start
> > operation.
> >
> > I think it makes sense to do the same for the SRPT control plane on an
> > individual HCA port GUID basis as long as there are no underlying fabric
> > issues, and that Roland is happy.  In terms of supporting more than one
> > type of /sys/kernel/config/target/srpt/$WWPN/$TPGT/ layout, I would
> > really like to avoid this for mainline code unless there is a really
> > good reason..
> 
> I'm still wondering whether it was a good idea to switch from ib<n>
> style names to port GIDs. Not only will that make it hard for users to
> match the ib<n> names already assigned to InfiniBand HCAs with target
> port names but this also deviates from the naming scheme used for any
> other Linux device. Ethernet devices e.g. appear as eth<n> in /dev
> instead of a MAC address like 00:1a:64:10:18:6f.
> 

So from the user perspective in rtsadmin/rtslib, the Port GUIDs are
automatically presented in long-form after being extracted and parsed
from /sys/class/infiniband/*/ports/*/gids/0 input via ib_srpt.spec here:

http://www.linux-iscsi.org/wiki/SCSI_RDMA_Protocol/RTSadmin#Specification_file

The available IB Port GUIs appear as <TAB> completable input
in /ib_srpt/ during a 'create wwn=' operation shown here:

http://www.linux-iscsi.org/wiki/SCSI_RDMA_Protocol/RTSadmin#Configuration

There is no reason why we can't present a more verbose informational
display based on individual available HW fabric WWPNs. Eg: things like
HW port status information, driver, firmware rev, etc..

So aside from the user side of things, I think using a IB Port GUID
makes more sense to ensure that /sys/kernel/config/target/srpt/
configuration does not depend upon the ordering of symbolic HCA names
in /sys/class/infiniband/, which AFAIK are not guarenteed to be
persistent as HCAs are added/removed from the system.

Ensuring that we can reference target fabric WWPNs using an underlying
HW reference instead of a symbolic name is important for this reason,
and is currently how other HW target mode drivers including
tcm_fc(openfcoe) currently function.

Thanks,

--nab

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists