[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: Mon, 18 Jun 2007 08:29:13 +0200

> FUJITA Tomonori wrote:
> > From: FUJITA Tomonori <fujita.tomonori at lab.ntt.co.jp>
> > Subject: Re: [Stgt-devel] [PATCH] Move device type to LUN
> > Date: Fri, 15 Jun 2007 13:36:12 +0900
> > 
> >> 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.
> > 
> > I did some cleanups.
> > 
> > http://git.kernel.org/?p=linux/kernel/git/tomo/tgt.git;a=shortlog;h=scc
> > 
> Regarding the disk_type_names cleanups:
> - What is the rationale for this? Just truncating the full name to
> something unintelligible doesn't seem like the smartest move to me,
> especially as we're not facing any string length restrictions here ...

They are the names that lsscsi uses. But I don't have strong
preference.


> Regarding device type registration cleanups:
> - We're losing the ability to select sg_bst for passthrough during
>   registration. Is that intentional?

Yeah. I plan to fix it though I'm not hurry since it needs bsg, which
is supposed to be merged into 2.6.23.