RE: iSCSI: aborting an immediate command with ABORT TASK

[Eddy Quicksall <eddy_quicksall@ivivity.com>]

How about another bit in the abort that says it is aborting an immediate command and hence does not advance the window?

 

Eddy

-----Original Message-----
From: Julian Satran [mailto:Julian_Satran@il.ibm.com]
Sent: Thursday, September 05, 2002 9:00 AM
To: Eddy Quicksall
Cc: 'Black_David@emc.com'; ips@ece.cmu.edu; owner-ips@ece.cmu.edu; tonyb@cybernetics.com
Subject: RE: iSCSI: aborting an immediate command with ABORT TASK

I don't think we want to. The TM made immediate are done so with good reason - to allow expedited consideration.
And there are no known scenarios (to us) in which it does not work.

Julo

Eddy Quicksall <eddy_quicksall@ivivity.com> Sent by: owner-ips@ece.cmu.edu 05/09/02 15:46

To:        "'Black_David@emc.com'" <Black_David@emc.com>, tonyb@cybernetics.com         cc:        ips@ece.cmu.edu         Subject:        RE: iSCSI: aborting an immediate command with ABORT TASK

Can we say that aborts with the immediate bit only abort commands with the
immediate bit? And then say that the window is not advanced under that
situation?

Eddy

-----Original Message-----
From: Black_David@emc.com [mailto:Black_David@emc.com]
Sent: Thursday, September 05, 2002 2:51 AM
To: tonyb@cybernetics.com
Cc: ips@ece.cmu.edu
Subject: RE: iSCSI: aborting an immediate command with ABORT TASK


This thread's taken a while to sort through, but there
does appear to be a problem in here.

> If the missing command was non-immediate, then yes, the plug-in avoids
> having to resend the missing command.  But, if the missing command was
> immediate, then there is no hole to plug.  In this case, the spec requires
> the target to plug a hole which does not exist, which is the problem that
I
> am trying to point out.

I'm working off of -15.  Section 2.2.2.1 contains the following statements:

                CmdSN always contains the number to be
     assigned to the next Command PDU.

  Commands meant for immediate delivery are marked with an immediate
  delivery flag; they MUST also carry the current CmdSN.

So suppose CmdSN is 7, and the initiator sends an immediate command
with CmdSN 7 and then decides to abort it, and sends TASK ABORT as
an immediate command (CmdSN is 7 and not advanced).  Suppose the TASK
ABORT crosses the response to successful completion of the immediate
command on the wire.  At this point the following text in Section
9.6.1 applies:

    b)  if the Referenced Task Tag does not identify an existing task
    but if the CmdSN indicated by the RefCmdSN field in the Task Man-
     agement function request is within the valid CmdSN window (between
     MaxCmdSN and ExpCmdSN), targets must consider the CmdSN received
     and return the "Function complete" response.

7 is still the CmdSN of the next non-immediate command to be sent,
so it is within the window (ExpCmdSN is also 7).  Following these
directions, the Target considers CmdSN 7 to be received, and advances
its window.  Now, when the initiator sends its next non-immediate
command (CmdSN=7), its CmdSN is outside the window, and the target
bit-buckets the command instead of executing it.

This went awry because the task to be aborted was (implicitly)
identified by its CmdSN, and not its Task Tag resulting in an attempt
to abort an immediate command actually clobbering the next non-
immediate one.  I think Tony gets credit for finding another problem.

While I'm in here, it also appears that the text describing the
mapping of iSCSI task management response codes to SCSI service
responses needs to change to map 1 (Task does not exist) to
FUNCTION COMPLETE rather than FUNCTION REJECTED to line up with
Section 6.2 of SAM-2.

Thanks,
--David
---------------------------------------------------
David L. Black, Senior Technologist
EMC Corporation, 42 South St., Hopkinton, MA  01748
+1 (508) 249-6449            FAX: +1 (508) 497-8018
black_david@emc.com       Mobile: +1 (978) 394-7754
---------------------------------------------------