FCP-2 recovery problem

Baldwin, Dave Dave.Baldwin at emulex.com
Wed Jun 28 12:44:38 PDT 2000


* From the T10 Reflector (t10 at t10.org), posted by:
* "Baldwin, Dave" <Dave.Baldwin at emulex.com>
*
Carl,

I don't believe that ABTS is a problem. Furthermore, changing ABTS to include a
LUN field would be a disaster! If there are ABTS issues you can identify, I'm
sure we could work around them.

If the initiator sends an ABTS to OX_ID 1, then there are two things that can
happen. If the target successfully received the initiator's current OX_ID 1
command, then all is well. The target and initiator are talking about the same
thing.

If the initiator sends an ABTS to OX_ID 1, and the target never received the new
command, then the target will send back a BA_ACC or BA_RJT saying either he
didn't know about the command in question, or he would be talking about a
previously completed command which already finished per the initiator's
viewpoint. The second case here would be fine for in-order delivery, as the
command frame has been dropped and will be reissued in a new exchange. If you
want this to work for Class 2, out-of-order delivery, I would require the
initiator to send REC before sending ABTS to see if the command ever made it to
the target.

So, back to my proposal. It isn't "ugly" and solves the problem.

Best regards,
Dave Baldwin

"Zeitler, Carl" wrote:

> If I issue ABTS to OX_ID 1, which LUN steps up to the plate?
>
> I do not believe any of the schemes presented to date work on a LUN basis.
> ABTS is necessary to do recovery.  It uses a bit in the Parameter field to
> distinguish between aborting the Sequence Vs the Exchange and therefore
> cannot be used for a 4 byte identifier.  The LUN or some means of making the
> OX_ID unique on a LUN basis must be part of ABTS. Otherwise the ABTS is
> ambiguous to the Target.
>
> So I think we are back to the possibilities I have proposed.  If it is
> deemed necessary to have a nexus of D_ID/S_ID/OX_ID/RX_ID/LUN, then the LUN
> (or some handle) has to be carried in the frame header that can also be used
> by ABTS, to be squeaky clean.  The only place for it is in a Device Header.
> This is necessary only for Class 3, I believe, as for Class 2, the ACK to
> FCP_Resp will wipe out the context for the completed exchange.  If the ACK
> to the response gets whacked, then ABTS/RRQ process will also wipe out the
> Exchange.  If both the ACK to FCP_RESP and the following FCP_CMD using the
> same OX_ID both get wiped, then this is a double error and the outstanding
> Exchange is aborted.  There are other double error scenarios that we have
> covered in Class 2 and believe that if we follow the rule of aborting the
> current Exchange in the event of a double error, we stay out of trouble and
> avoid data integrity problems.
>
> The other solution is to steer away from an I_T_L nexus and stick with an
> I_T nexus.  If this is the case, then the Herman bit solves the problem as
> there can only be one outstanding Exchange that is considered "open" ( See
> D.5 Class 3, T10/00-137r5).  Herman can make the distinction between the old
> "open" Exchange and the new, current Exchange having the same OX_ID.
> Another possibility is to use FCP_CONF.  FCP_CONF "closes" the Exchange.  If
> FC__CONF is not received by the Target, it issues REC.  If the OX_ID in the
> REC payload is the same as an outstanding OX_ID, the current Exchange is
> aborted. In some cases, hopefully infrequent, this could unnecessarily
> clobber a validly received command.
>
> So the saga of this seemingly trivial Class 3 error recovery corner case
> continues.  The problem appears trivial, but the solutions are downright
> ugly.
>
>
>
> Regards, Carl
>
> Carl Zeitler
> Compaq Computer Corporation
> MS 150801, 20555 SH249, Houston, TX 77070
> Phone:281-518-5258 Fax: 281-514-5270
> E-Mail: Carl.Zeitler at compaq.com
>
> -----Original Message-----
> From: Baldwin, Dave [mailto:Dave.Baldwin at emulex.com]
> Sent: Monday, June 26, 2000 10:11 PM
> To: Matt Wakeley
> Cc: T10 at t10.org; 'FC Reflector'
> Subject: Re: FCP-2 recovery problem
>
> *
> * From the fc reflector, posted by:
> * "Baldwin, Dave" <Dave.Baldwin at emulex.com>
> *
> Matt,
>
> It would not work with the change you are suggesting. If I send OX_ID=1 and
> CRN=6 to LUN=1, then send CRN 7 through CRN 5 (wraparound) to LUN=2 (maybe a
> disk), then send OX_ID=1 and CRN=6 to LUN=1 (which gets dropped), I have the
> same problem.
>
> Best regards,
> Dave Baldwin
>
> Matt Wakeley wrote:
>
> > *
> > * From the fc reflector, posted by:
> > * Matt Wakeley <matt_wakeley at agilent.com>
> > *
> > Charles,
> >
> > If the CRN was *not* based on LU, your proposal would work...
> >
> > "Binford, Charles" wrote:
> >
> > > At this point my plan is working - single LUN.  But consider Multi-LUN.
> > > IF
> > > you change the scenario only slightly we break:
> > >
> > > CMD ------OX_ID=1, CRN=6, LUN=1 ---------------------->
> > >
> > >            <---------------------------     Response
> > >
> > > CMD -----OX_ID=1, CRN=6, LUN=2 ----->  X (dropped frame)
> >
> > If the above had a CRN of 7 (since both LUs are in the same target),
> > everything would work fine.
> >
> > -Matt

*
* 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