Re: [storm] What if a SCSI Response PDU has GOOD status anda DataSegment?

I agree, Option 1 is preferred.  The direction at T10 so far, is that
nothing is done at the SCSI layer to enable this capability.  As you
noted for iSCSI, the other transports basically say the same thing -
they do not prohibit it.

The changes at T10 have so far, indicated no transport layer changes.

	Fred

-----Original Message-----
From: Mallikarjun Chadalapaka [mailto:cbm@chadalapaka.com] 
Sent: Monday, November 30, 2009 9:50 PM
To: 'Julian Satran'; Black_David@emc.com
Cc: storm@ietf.org
Subject: Re: [storm] What if a SCSI Response PDU has GOOD status anda
DataSegment?

IMHO, I think the right forum to make a call on this is T10.  Let's
leave it
to T10 to define the proper behavior.  If T10 decides that "GOOD with
DataSegment" can simply be done without any new SCSI requirements, I
think
we'll have two options.

*Option 1*	
iSCSI should state the following in the main text and elaborate as an
implementer's advice.

Require that "GOOD with DataSegment" status be correctly handled by
*iSCSI*
implementations - i.e. although it doesn't imply for higher layers in
the
stack.  Next revision of consolidated iSCSI draft should simply make it
clear in that a DataSegment is allowed with any non-CHECK CONDITION
status
(RFC 3720 does not prohibit it either).

*Option 2*  
Define a new iSCSI negotiation key to explicitly allow "GOOD with
DataSegment" (or more generally, non-CHECK CONDITION) at the iSCSI
level.

If T10 OTOH defines new SCSI-isms to formally sanction this case (e.g.
mode
pages, commands etc.), iSCSI doesn't need to change anything although
still
an iSCSI implementer's note wouldn't hurt.

My preferred option is #1 - i.e. I like us not to add an iSCSI
negotiation
key for a SCSI feature.  It's something we have avoided doing so far by
design.

Thanks.

Mallikarjun

> -----Original Message-----
> From: storm-bounces@ietf.org [mailto:storm-bounces@ietf.org] On Behalf
> Of Julian Satran
> Sent: Thursday, November 26, 2009 1:22 AM
> To: Black_David@emc.comBlack_David@emc.com
> Cc: storm@ietf.org
> Subject: Re: [storm] What if a SCSI Response PDU has GOOD status and a
> DataSegment?
> 
> David,
> 
> The key is a good idea ... but not a great one :-)
> As you certainly know there are many storage controllers (targets) on
> the market that support
> iSCSI, FCP, iSER and even several at the same time (for different
> sessions usually as we and T10 never seriously considered I_T nexi on
> different transports).
> 
> Their SCSI layer (in the ones I know about) is completely transport
> oblivious so the transport is not really the layer you want to
> introduce status+sense for statuses other than bad (unit check).
> 
> At the SCSI level it may make a lot of sense (no pun intended either).
> iSCSI will support it (without any key negotiation being required) as
> iSCSI does not prohibit it :-). I don't recall exactly what FCP had to
> say on this but I recall vaguely that there is no explicit
> interdiction. However some silicon vendors might be "surprised" if it
> happens.
> 
> Regards,
> Julo
> 
> 
> On 25/11/2009, at 19:57, <Black_David@emc.com> <Black_David@emc.com>
> wrote:
> 
> > [WG chair hat OFF]
> >
> > Here's a little more background.  There's an open issue in T10
> > about whether sending sense data with GOOD status can just be
> > done, vs. whether the target ought to know in advance that
> > the initiator is prepared for this.
> >
> > In keeping with the IETF advice to be "conservative in what
> > is sent and liberal in what is accepted", if T10 decides not
> > to address this at the SCSI level, then it'll be reasonable
> > to define an iSCSI negotiation key to indicate that the
> > initiator is prepared to accept (in the sense of "tolerate",
> > i.e., won't cause an error or worse, rather than any promise
> > to "process" or "understand") sense data for GOOD status.
> >
> > In addition (as Ralph suggests), it may make sense (pun intended)
> > to define this key to cover all status values, not just GOOD,
> > even though only GOOD is currently under discussion in T10.
> >
> > Thanks,
> > --David
> >
> >
> >> -----Original Message-----
> >> From: storm-bounces@ietf.org [mailto:storm-bounces@ietf.org]
> >> On Behalf Of Ralph Weber
> >> Sent: Tuesday, November 24, 2009 7:14 PM
> >> To: storm@ietf.org
> >> Subject: [storm] What if a SCSI Response PDU has GOOD status
> >> and a DataSegment?
> >>
> >> One of the questions that T10 has been wrestling with lately
> >> concerns a case where their equivalent of the iSCSI Response
> >> PDU contains a status other than CHECK CONDITION and sense
> >> data in a Data Segment.
> >>
> >> What will initiators do if handed one of these beasts?
> >>
> >> Note: GOOD is the current favorite doppelganger for CHECK
> >> CONDITION, but BUSY et al. should be considered too when
> >> thinking about what initiators might do.
> >>
> >> The current belief in T10 is that initiators most likely
> >> will ignore any Data Segment found in a Response PDU with
> >> GOOD status. This is based on the notion that initiators
> >> do not squander the extra cycles needed to validate entire
> >> format of a Response PDU.
> >>
> >> What do the implementers of iSCSI initiators think?
> >>
> >> All the best,
> >>
> >> .Ralph
> >>
> >> _______________________________________________
> >> storm mailing list
> >> storm@ietf.org
> >> https://www.ietf.org/mailman/listinfo/storm
> >>
> >>
> > _______________________________________________
> > storm mailing list
> > storm@ietf.org
> > https://www.ietf.org/mailman/listinfo/storm
> 
> _______________________________________________
> storm mailing list
> storm@ietf.org
> https://www.ietf.org/mailman/listinfo/storm

_______________________________________________
storm mailing list
storm@ietf.org
https://www.ietf.org/mailman/listinfo/storm