Persistent reserve, a off topic question.

[di9...@gmail.com]

Hey

I try to understand the mechanics of persistent reserve in SPC-3.

My problem is off topic and not related to libiscsi but the number of people who know SCSI and

SAM/SPC/SBC inside out is not that many. If you feel there is better spot for this question just kindly

point me in the right direction.

My job is to create a SCSI target and right now I am working with persistent reserve. But I find SPC and

related documents a bit harsh to on the mind so to speak and not so clear as I would wish. Thus I would

very much like to know that I got it.

- A IT Nexus can have at most ONE persistent reservation key at any time.

- A registered IT Nexus is a nexus that have a reservation key.

- Adding a new key to a IT Nexus that have a key will replace the old key with the new key.

- Multiple IT Nexuses may share the same reservation key, no conflict since the IT Nexus together

  with the reservation key form a unique identifier.

- A registration contains of one reservation key and one or more IT Nexuses.

- If a registration is removed, the reservation key and all IT Nexuses in the registration is removed.

  The IT Nexuses is now unregistered.

- A reservation or a release of a reservation is always done on the IT Nexus that issued the reserve

  and release service action.

Is this correct ?

Thanks for your time.

// Jim Gunnarsson

[Lee Duncan]

Hi Jim:

On May 5, 2016, at 1:38 AM, di9...@gmail.com wrote:

Hey

I try to understand the mechanics of persistent reserve in SPC-3.

Many before you have tried and failed. :)

Have you looked at my persistent reservation document:  http://gonzoleeman.net/documents/scsi-3-pgr-tutorial-v10.pdf

My problem is off topic and not related to libiscsi but the number of people who know SCSI and

SAM/SPC/SBC inside out is not that many. If you feel there is better spot for this question just kindly

point me in the right direction.

My job is to create a SCSI target and right now I am working with persistent reserve. But I find SPC and

related documents a bit harsh to on the mind so to speak and not so clear as I would wish. Thus I would

very much like to know that I got it.

This is actually a really large task. I know, I’ve helped do this before. There are endless side-cases that are … not well defined, and which almost take a lawyer if you are going straight from the specification. :)

Just to make a SCSI-2 functional target is not trivial, but then adding group reservations is a *lot* harder. Have you looked at some of the software iSCSI targets, like iet or tgt? These all have the same issues, i.e. how to look like a valid SCSI target, including code to support SCSI-3 PGR (persistent group reservations).

Did you know that the libiscsi tests include a set on persistent reservations (that I added)?

- A IT Nexus can have at most ONE persistent reservation key at any time.

It has been a while since I worked on this stuff, but I’m pretty sure that is incorrect.

- A registered IT Nexus is a nexus that have a reservation key.

Yes, registering is the act of registering a key.

- Adding a new key to a IT Nexus that have a key will replace the old key with the new key.

From a single ITN, yes.

- Multiple IT Nexuses may share the same reservation key, no conflict since the IT Nexus together

  with the reservation key form a unique identifier.

Yes, they could, but that would be dumb. Each initiator should have it’s own key, I believe. But much depends on the reservation type.

- A registration contains of one reservation key and one or more IT Nexuses.

Not really.

- If a registration is removed, the reservation key and all IT Nexuses in the registration is removed.

  The IT Nexuses is now unregistered.

Depends on the registration type.

- A reservation or a release of a reservation is always done on the IT Nexus that issued the reserve

  and release service action.

Depends on the registration type.

Is this correct ?

Thanks for your time.

// Jim Gunnarsson

Please see my document. Then you can email me directly if you have more questions.

— 

Lee Duncan

SUSE Labs