SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    RE: iSCSI: aborting an immediate command with ABORT TASK



    -----Original Message-----
    From:
    owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of Julian Satran (Actcom)
    Sent: Friday, August 30, 2002 7:22 PM
    To:
    tonyb@cybernetics.com
    Cc: ips@ece.cmu.edu
    Subject: Re: iSCSI: aborting an immediate command with ABORT TASK
     
    > Immediate commands do not create holes.
     
    True.
     
    > Immediate commands can be aborted ONLY if referenced by their ITT.
     
    Immediate commands are uniquely identified by their ITT, not their CmdSN.  So yes, an existing immediate command at the target can be aborted ONLY if the ITT matches the Referenced Task Tag.
     
    >Again RefCmdSN is used only if the command is not found by ITT - and then only to avoid having to send a duplicate command.
    > Julo
    It seems that the implicit assumption that you are making and I am not is that the target will always find a command with ITT equal to the Referenced Task Tag when the initiator attempts to abort an immediate command.  It follows that if this were the case, then the target would never use the RefCmdSN improperly.  However, I do not see any mechanism that guarantees this assumption, especially given the following three observations:
     
    1) The possible out-of-order arrival of an abort and the immediate command upon which it is supposed to act
    2) The race between an immediate command finishing normally (causing the target to free the command and forget its associated ITT) and the initiator aborting it
    3) The immediate command to be aborted may have been discarded due to a header digest error
     
    In the examples I gave previously, there are several specific situations in which the assumption does not hold, leading to incorrect behavior of the target.
     
    Tony
    ----- Original Message -----
    Sent: 30 August, 2002 21:29
    Subject: RE: iSCSI: aborting an immediate command with ABORT TASK

    In my example, the command with the Initiator Task Tag matching the Referenced Task Tag is not at the target.  When the target considers the command with RefCmdSN received, it assumes _incorrectly_ that the command with RefCmdSN must have Initiator Task Tag == Referenced Task Tag.  That assumption can be incorrect when immediate commands are used because multiple commands with different Initiator Task Tags can have the same CmdSN.  This is just another way of looking at the problem, but the problem still exists.
     
     
      -----Original Message-----
    From: Julian Satran [mailto:Julian_Satran@il.ibm.com]
    Sent: Friday, August 30, 2002 2:15 PM
    To: tonyb@cybernetics.com
    Cc: ips@ece.cmu.edu; owner-ips@ece.cmu.edu
    Subject: Re: iSCSI: aborting an immediate command with ABORT TASK

    Tony,

    As it was already pointed out CmSN is a SECONDARY identifier - the primary being the Referenced Task Tag.
    I was introduced to simplify handling holes (commands discarded or arriving late).
    In your example A and B have different ITT or one of them gets bumped.

    Julo



    "Tony Battersby" <tonyb@cybernetics.com>
    Sent by: owner-ips@ece.cmu.edu

    30/08/02 20:16
    Please respond to tonyb

           
            To:        <ips@ece.cmu.edu>
            cc:        
            Subject:        iSCSI: aborting an immediate command with ABORT TASK

           


    It seems that an initiator attempting to abort an immediate command may
    inadvertently cause the next non-immediate command to be aborted instead.
    For example,

    Initiator sends an immediate command "A" with CmdSN 10.
    Initiator sends a non-immediate command "B" with CmdSN 10.
    Initiator sends an immediate ABORT TASK command "C" with CmdSN 11 to abort
    command "A" (RefCmdSN == 10, Referenced Task Tag == Initiator Task Tag of
    "A").

    If the target either:
    1) Receives "C" before "A" or "B", OR
    2) Receives "C" before "B" but after executing and freeing "A",

    ...then the target will not have a matching task for the abort in its queue,
    and will consider CmdSN 10 received because it is inside the valid CmdSN
    window.  If the target then receives "B", it will silently ignore the
    command because it is outside the valid CmdSN window.  This is obviously not
    what the initiator was trying to accomplish.

    This problem could be solved by adding a flag to the PDU for the ABORT TASK
    command to indicate if the task to be aborted is immediate or non-immediate.
    If the target does not have a matching task and the RefCmdSN is inside the
    valid command window, then the target should consider the CmdSN received
    only if the flag indicates that the task to be aborted is non-immediate.

    -----------------------

    One other thing: for draft 16, I recommend that the the following phrase:
    "... (i.e., with CmdSN not higher than the task management command CmdSN)
    ..."
    in section 9.5.1 be changed to:
    "... (i.e., with CmdSN lower than the task management command CmdSN) ...".
    This is for consistency with another sentence earlier in the same paragraph.


    Anthony J. Battersby
    Cybernetics





Home

Last updated: Tue Sep 03 17:18:59 2002
11749 messages in chronological order