Third party persistent reservations source I_T nexus function of XCopy
Kevin D Butt
kdbutt at us.ibm.com
Wed May 1 09:07:50 PDT 2013
Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1305012_f.htm">HTML-formatted message</a>
Fred,
I think we are assuming pretty much the same model you described. There
is an application client who is the controlling entity. It does whatever
it wants to do with the CSCD. It then tells the copy manager to do
whatever it is the application client wants the copy manager to do. The
application client is the one with the onus to control how the copy
manager uses reservations (this seems apparent since we have some
descriptors for registration and register and move defined). This seems
like the "cooperative use" you describe. Figure 17 in SPC-4r36e shows
that the copy manager may be in a logical unit, in a SCSI target, or as a
stand alone device.
If you take one of the scenarios where the copy manager is in a different
device than the application client, then everything I discussed below is
an issue. There is no way for the application client to have the
information necessary to "move" the reservation to the copy manager. It
can tell the copy manager to move the reservation back to the application
client, but that is of no worth if there is no way for the reservation to
get moved to the copy manager - or to get established in the copy manager.
The CSCD descriptors do provide a RELATIVE INITIATOR PORT IDENTIFIER
field, but that does not provide the application client the knowledge of
what the TransportID of the initiator port is.
Note that the registration model only allows the exclusive access type of
reservations to be moved with a REGISTER AND MOVE service action. This is
where the exclusive reservation I spoke of comes from.
The REGISTER function allows the copy manager to join a shared reservation
if one already exists (e.g., already created by the application client),
but ignores how this would be cleaned up.
How can the copy manager have a reservation with which to use the REGISTER
AND MOVE required in the Third party persistent reservations source I_T
nexus function (DESCRIPTOR TYPE CODE 15h)?
Thanks,
Kevin D. Butt
SCSI & Fibre Channel Architect, Tape Firmware
Data Protection & Retention
MS 6TYA, 9000 S. Rita Rd., Tucson, AZ 85744
Tel: 520-799-5280
Fax: 520-799-2723 (T/L:321)
Email address: kdbutt at us.ibm.com
http://www-03.ibm.com/servers/storage/
From: "Knight, Frederick" <Frederick.Knight at netapp.com>
To: Kevin D Butt/Tucson/IBM at IBMUS, Ralph Weber <roweber at ieee.org>,
Cc: "owner-t10 at t10.org" <owner-t10 at t10.org>, T10 Reflector
<t10 at t10.org>
Date: 05/01/2013 06:50 AM
Subject: RE: Third party persistent reservations source I_T nexus
function of XCopy
It sounds like you are assuming a model where the application client and
the copy manager are independently accessing the CSCD. I believe the
model assumes that the application client and the copy manager are engaged
in cooperative use of the device.
The application client establishes the reservation (and may write some
data). Then the XCOPY command is used to have the copy manager write a
bunch of data (this involves the ?move? operation). When complete, the
application client again takes control (and again may write some data).
Maybe the application client is writing headers and trailers, but the copy
manager is the one writing the data stream in the middle.
I don?t believe the model was designed for a copy manager to have
exclusive access (e.g. an independent reservation) without cooperation
with the application client.
Fred
From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of Kevin D
Butt
Sent: Tuesday, April 30, 2013 12:19 PM
To: Ralph Weber
Cc: owner-t10 at t10.org; T10 Reflector
Subject: Re: Third party persistent reservations source I_T nexus function
of XCopy
Thanks Ralph!
I have more questions for all.
My engineer came back to me claiming that there are issues with the entire
reservation model of extended copy.
The Register function described in 6.4.6.17 Register persistent
reservation key function (DESCRIPTOR TYPE CODE 14h) is available for the
application client to tell the copy manager to register with the CSCD but
there is no function to tell the copy manager to reserve the CSCD.
Therefore, the only use for the existing register function (14h) is to
join an existing All Registrants or Registrants Only reservation that
already exists on the CSCD. Furthermore, there is no explicitly stated
method for leaving that reservation or unregistering. One might be
tempted to send an additional register function (14h) with a zero key as
the last segment descriptor but that would not work should a failure occur
in a prior segment. Then you would be back to the original issue of
having a registration hanging around with no way to remove it (without
PREEMPT or CLEAR).
The other reservation function is 6.4.6.18 Third party persistent
reservations source I_T nexus function (DESCRIPTOR TYPE CODE 15h) which is
only valid for the exclusive access type reservations. Part of the
description of this function restricts it to tape devices and does not
allow disk devices. This seems to be overly restrictive. The other issue
is how the copy manager gets this reservation in the first place. The
only way would seem to have the application client perform a REGISTER AND
MOVE on the CSCD to the copy manager. However, this requires the
application client to know the relative target port of the CSCD and the
initiator port that the copy manager will be using. The application
client does not know this information and so there seems to be no way for
the copy manager to ever have a reservation with the CSCD.
Are we missing something, or is the reservation model for extended copy
truly incomplete?
It seems that at a minimum there should be a new function to reserve the
CSCD and a way to release the CSCD once the copy manager is done with it.
The release could be a new bit in the existing register function that
indicates that the copy manager shall attempt to release the CSCD after
all segments have been processed - similar to the instructions to do the
Third party persistent reservations source I_T nexus function (15h) after
all segments have been processed.
Thanks,
Kevin D. Butt
SCSI & Fibre Channel Architect, Tape Firmware
Data Protection & Retention
MS 6TYA, 9000 S. Rita Rd., Tucson, AZ 85744
Tel: 520-799-5280
Fax: 520-799-2723 (T/L:321)
Email address: kdbutt at us.ibm.com
http://www-03.ibm.com/servers/storage/
From: Ralph Weber <roweber at ieee.org>
To: T10 Reflector <t10 at t10.org>,
Date: 04/29/2013 07:08 PM
Subject: Re: Third party persistent reservations source I_T nexus
function of XCopy
Sent by: owner-t10 at t10.org
* From the T10 Reflector (t10 at t10.org), posted by:
* Ralph Weber <roweber at ieee.org>
*
I believe the "intent" mentioned in question 1 matches the rather
lengthy description that follows the question (up to, but not including,
question 2).
I believe that any errors encountered during the processing of a Third
party persistent reservations source I_T nexus segment descriptor should
not overwrite an error that has already been encountered prior to the
processing of the descriptor. There are other ways to determine the
state of the persistent reservation, but there are few, if any, ways to
determine the nature of the first error.
This makes the answers to questions 2 and 3 be the following:
Question 2: The COMMAND-SPECIFIC INFORMATION field shall not be altered.
Question 3: The COMMAND-SPECIFIC INFORMATION field shall set to the
relative position of the first segment descriptor that detected an error.
All the best,
.Ralph
On 4/26/2013 5:43 PM, Kevin D Butt wrote:
> One of my engineers asked me about a function in extended copy that is
> confusing him. It confuses me too and I would like to know what the
> expected behavior is.
>
> SPC-4 states:
> <<
> 6.4.6.18 Third party persistent reservations source I_T nexus function
>
> The segment descriptor format shown in table 165 instructs the copy
> manager to send a PERSISTENT RESERVATION OUT command with REGISTER AND
> MOVE service action (see 5.13.8) with the specified I_T nexus after
> all other segment descriptors have been processed. If an error is
> detected any time after receiving a third party persistent source
> reservation I_T nexus segment descriptor, the PERSISTENT RESERVATION
> OUT command REGISTER AND MOVE service action shall be processed before
> the copy operation (see 5.17.4.3) originated by the EXTENDED COPY
> command is completed.
>
> This segment descriptor should be placed at or near the beginning of
> the list of segment descriptors to assure the copy manager processes
> the PERSISTENT RESERVATION OUT command with REGISTER AND MOVE service
> action in the event of an error that terminates the processing of
> segment descriptors. If an error is detected in a segment descriptor
> and third party persistent reservations source I_T nexus segment
> descriptor has not been processed, the copy manager shall not send a
> PERSISTENT RESERVATION OUT command with REGISTER AND MOVE service
action.
>
> Placing more than one source third party persistent reservations
> source I_T nexus segment descriptor in the list of descriptors is not
> an error. All source third party persistent reservations source I_T
> nexus segment descriptors known to the copy manager shall be processed
> after all other segment descriptors have been processed.
> >>
>
> This appears to say that this function requires a REGISTER AND MOVE to
> be process after all other segments have been processed, presumably to
> transfer the reservation after all the copy functions have been
> processed. This segment should be early in the list so it can be
> processed to transfer the reservation even if there is a failure while
> attempting to process any of the segments that occur in the list after
> this function. And, there is a requirement to process this segment
> even if there is a failure in one of the other segments. However, if
> the failure occurs in one of the segments prior to this function, then
> the REGISTER AND MOVE shall not be processed. Further, there are
> allowed to be multiple instances of this function and all instances
> are processed after all the other segments have been processed.
> Presumably, these functions would be processed in the order they
> appear in the list.
>
> This seems to be the behavior described, but it is difficult to be
> sure this is the intent.
> *Question 1: *Is this the intent?
>
> Assuming this is the correct behavior, then assume a parameter list
> with 10 segments. There is a Third party persistent reservations
> source I_T nexus function in the third segment descriptor and another
> that is in the sixth segment descriptor. Further, assume that there
> is an error in processing the eighth segment descriptor.
>
> In 5.17.7.4 EXTENDED COPY command errors detected during processing of
> segment descriptors it states:
> After an exception condition is detected during segment descriptor
> processing:
> b) the copy manager shall indicate the segment that was being
> processed at the time of the exception by writing the segment number
> to the third and fourth bytes of the COMMAND-SPECIFIC INFORMATION
> field. The segment number is based on the relative position of the
> segment descriptor in the parameter list (see 5.17.7.1)
> In this case the segment number put in the third and fourth bytes of
> the COMMAND-SPECIFIC INFORMATION field is 7 (i.e., the eighth
> descriptor).
>
> *Question 2: *What should be put into the COMMAND-SPECIFIC INFORMATION
> field if there is also a failure in processing the REGISTER AND MOVE
> in the third or sixth segment descriptor that is required to be
> processed at the end after all other segments have been processed or
> after an error occurs?
>
> *Question 3:* What should be put into the COMMAND-SPECIFIC
> INFORMATION field if all copy segments complete without error, but
> then during the REGISTER AND MOVE one of the third or sixth segment
> has an error? If we indicate the failing Third party persistent
> reservations source I_T nexus function then we lose the fact that all
> the other segments completed successfully.
>
> Thanks,
>
> Kevin D. Butt
> SCSI & Fibre Channel Architect, Tape Firmware
> Data Protection & Retention
> MS 6TYA, 9000 S. Rita Rd., Tucson, AZ 85744
> Tel: 520-799-5280
> Fax: 520-799-2723 (T/L:321)
> Email address: kdbutt at us.ibm.com
> http://www-03.ibm.com/servers/storage/
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org
More information about the T10
mailing list