[Stgt-devel] [PATCH] Move device type to LUN

[FUJITA Tomonori]

From: Hannes Reinecke <hare at suse.de>
Subject: Re: [Stgt-devel] [PATCH] Move device type to LUN
Date: Tue, 12 Jun 2007 14:44:03 +0200

> FUJITA Tomonori wrote:
> > From: Hannes Reinecke <hare at suse.de>
> > Subject: Re: [Stgt-devel] [PATCH] Move device type to LUN
> > Date: Mon, 11 Jun 2007 17:15:19 +0200
> > 
> >> Pete Wyckoff wrote:
> >>> hare at suse.de wrote on Tue, 05 Jun 2007 15:38 +0200:
> >>>> This patch moves the device type down to the LUN structure.
> >>>> And in doing so we now also have the proper peripheral device
> >>>> type and peripheral device qualifier attributes for the INQUIRY
> >>>> data.
> >>> Makes sense to me too.
> >>>
> >>>> One thing puzzles me, though: do we support commands with no
> >>>> LUN attached to it? IE is it valid to have 'cmd->dev == NULL'?
> >>>> If so: where is the point here? If that's our handling of a
> >>>> non-existing LUN 0 we should rather add a proper LUN 0 and
> >>>> treat cmd->dev == NULL as an error case ...
> >>> In the SCSI model, every device must have at least one LUN for
> >>> handling REPORT LUNS and a couple other commands.  It is addressed
> >>> as LUN 0 or using the "well-known" LUN for the command.  In the stgt
> >>> abstraction, though, there is no magic LUN like this.  Instead
> >>> things like spc_inqury use cmd->dev == NULL to handle this case.
> >>>
> >> Ah. Hence.
> >>
> >>> Perhaps it is reasonable to create a special device for these
> >>> commands.  Up in target.c, you could assign cmd->dev to target->dev,
> >>> where that is the special device, paralleling the way that
> >>> target->cmd_queue is used.  As a side effect, lots of "if (cmd->dev)
> >>> ... else ILLEGAL_REQUEST" clauses can be removed from device code
> >>> too.
> >>>
> >> Well, I actually thought of creating a proper LUN 0 with type 0xc.
> > 
> > We still need to handle cmd->dev == NULL case though probably we can
> > remove cmd->dev == NULL case in device type code.
> > 
> > I don't have storage systems that work in your way. Is it common?
> > 
> > 
> Yes, quite common. HP and EMC (to name but a few) do it this way.
> Most storage arrays actually refuse to attach any devices to LUN0, as
> this is a pure management LUN.
> 
> I will draft up a patch which creates a proper controller LUN0.

Ok, here's a patch. This depends on the patch that I've just
submitted.

This adds scc support and fixes ibmvio. iSCSI works. ibmvio works for
me, but it detects scsi devices in a unique way, so I'm not sure it
works for everyone.

As described, a scsi controler device is automatically created as lun0
with a new target.

lily:/home/fujita# ./tgt/usr/tgtadm --lld iscsi --op new --mode target --tid 1 -T iqn.2001-04.com.example:storage.disk2.tulip.sys1.xyz
lily:/home/fujita# ./tgt/usr/tgtadm --op show --mode target
Target 1: iqn.2001-04.com.example:storage.disk2.tulip.sys1.xyz
    System information:
        Driver: iscsi
        Status: running
    I_T nexus information:
    LUN information:
        LUN: 0
            Type: raid
            SCSI ID: deadbeaf1:0
            SCSI SN: beaf10
            Size: 0
            Backing store: No backing store
    Account information:
    ACL information:


The patch is still very hacky but shows how it works. Let me know this
works or not.