Re: Target Reset handling

["Matt Wakeley" <matt_wakeley@agilent.com>]

One could make the argument that a "big hammer" reset is desireable,
especially in new implementations that might hang - the initiator could
attempt to "power cycle" the box if it hangs.  The problem is that at this
reset is at a really high level (the iSCSI/TCP/IP stack must be working in the
first place to even receive and process the command). FC loop defined a "link
primitive" that could be used the reset the entire device (the LIP_RESET) that
required only the lowest level link primitive state machine to be running to
process it - and yank on a global reset line.  If iSCSI/TCP/IP is in a state
that it can receive and decode the "big hammer" reset iSCSI PDU, I don't think
it would be in a state that it needs to be reset.

-Matt Wakeley
Agilent Technologies

Robert Snively wrote:

> I do not believe that the added complexity of a cold and hot
> reset is appropriate.  SCSI-2 had such a concept, and it was ultimately
> determined that there was no benefit to it, and the complexity was
> immense.
>
> I strongly feel that the SCSI Target Reset should do exactly
> no more and no less than the corresponding function does in all
> other SCSI devices.
>
> If a session/connection reset is required, it should be performed
> by session/connection control mechanisms already defined by TCP/IP.
> Its effect on the SCSI target should be defined, but may or may
> not be the same as the SCSI Target Reset. FCP-2 provides some examples
> of the kind of things that you may choose to reset in table 4.
>
> No "C" bit should be used.
>
> The session/connection reset should not be in the CDB or in the
> task management functions.
>
> Bob
>
> >  -----Original Message-----
> >  From: Mallikarjun C. [mailto:cbm@rose.hp.com]
> >  Sent: Tuesday, September 05, 2000 11:14 AM
> >  To: ips@ece.cmu.edu
> >  Subject: Re: Target Reset handling (fwd)
> >
> >
> >  Apologies if this is a second copy, I had not seen my copy of
> >  this posting or on the web archive even after more than an hour....
> >  --
> >  Mallikarjun
> >
> >
> >  Forwarded message:
> >
> >  Date: Tue, 05 Sep 2000 10:03:56 PDT
> >  To: ips@ece.cmu.edu
> >  Subject: Re: Target Reset handling
> >  In-Reply-To: <C1256950.001C76C2.00@d12mta02.de.ibm.com>;
> >  from "julian_satran@il.ibm.com" at Sep 4, 100 8:08 am
> >  Status: RO
> >
> >  Julian,
> >
> >  Thanks for the suggestion.
> >
> >  I am in broad agreement with this kind of definition, if the
> >  rest of the folks accept it.  I would however suggest a single
> >  Target Reset task management function (in accordance with SAM-2),
> >  and provide the choice of a "cold-reset" with a "C" bit in the
> >  Task Management Command PDU.  Assuming that we do this, let me
> >  try to summarize our tentative agreement:
> >
> >  o Target Reset task management requires a response, unless the
> >    the C-bit is set in which case the sessions would be terminated
> >    as well and no response can be expected.
> >
> >  o A Unit Attention AER shall be reported to all currently logged-in
> >    initiators, in accordance with the provisions contained in clauses
> >    7.9 and 8.3.6 of SPC-2 (basically, to be reported only if agreed
> >    in advance between the initiator and the LU).
> >
> >  Regards.
> >  --
> >  Mallikarjun
> >  M/S 5601
> >  Networked Storage Architecture
> >  HP Storage Organization
> >  Hewlett-Packard, Roseville.
> >  cbm@rose.hp.com
> >
> >
> >
> >  >Mallikarjun,
> >  >
> >  >How about having two different function:
> >  >
> >  >target warm-reset (connections stay) and target cold-rest
> >  (connections get
> >  >also reset)?
> >  >
> >  >Julo
> >  >
> >  >"Mallikarjun C." <cbm@rose.hp.com> on 03/09/2000 06:07:07
> >  >
> >  >Please respond to cbm@rose.hp.com
> >  >
> >  >To:   ips@ece.cmu.edu
> >  >cc:    (bcc: Julian Satran/Haifa/IBM)
> >  >Subject:  Re: Target Reset handling
> >  >
> >  >
> >  >
> >  >
> >  >Julian,
> >  >
> >  >Let me try again, I was arguing that the protocol stack (I
> >  assume you
> >  >mean the state information in various layers by this) and the TCP
> >  >connections be *not* reset - since I do not see a need for the SCSI
> >  >transport mechanism to be reset/cleared in the context of
> >  "SCSI target
> >  >reset".  I would again request your attention on the FC
> >  precedent, and
> >  >the software expectations of a confirmed target reset.
> >  Logging in after
> >  >an arbitrary "long time" and assuming the target reset to
> >  be complete is
> >  >simply unreliable.
> >  >
> >  >I do not see a need for a special "iSCSI reset".  I believe that the
> >  >SCSI target reset task management request is completely adequate to
> >  >address our requirements.  All I am proposing is a change
> >  in its current
> >  >definition in view of the various reasons I had already
> >  stated, and the
> >  >benefits of making this change (not the least of which is SAM-2
> >  >compliance).  Addition of a new reset mechanism would also
> >  defeat our
> >  >common goal to keep a fairly lean protocol.
> >  >
> >  >The question of security context needs to addressed
> >  regardless of the
> >  >SCSI transport behavior on a target reset.  If security
> >  context is being
> >  >designed as part of the transient operating environment (as
> >  opposed to
> >  >say, creating/modifying mode pages), then my first guess
> >  would be that
> >  >it has to be reset as well to initial values, on a target
> >  reset.  I am
> >  >not sure what you implied, but are you suggesting that the security
> >  >context cannot be reset with TCP connections living across
> >  a SCSI target
> >  >reset?  Or, did I totally miss something?
> >  >
> >  >Regards.
> >  >
> >  >Mallikarjun
> >  >M/S 5601
> >  >Networked Storage Architecture
> >  >HP Storage Organization
> >  >Hewlett-Packard, Roseville.
> >  >cbm@rose.hp.com
> >  >
> >
> >
> >