FCP_DL too small

Robert Snively rsnively at Brocade.COM
Wed Nov 3 13:28:45 PST 2004


* From the T10 Reflector (t10 at t10.org), posted by:
* "Robert Snively" <rsnively at Brocade.COM>
*
I reluctantly agree that some kind of hack is necessary for warning
incompetent
operating systems that there has been a protocol problem fouling up
the data length indications (or anything else for that matter).

However, I disagree with the suggested solution in 03-393r3.

The FCP_DL is NOT part of the CDB and CANNOT be pointed to by
"SCSI" status and sense information.  The operating system does
not necessarily have knowledge of the FCP_DL.

I would propose instead that you invent a check condition called
something like "TRANSPORT PROTOCOL FAILURE".  That would label the
error condition so data would not be misused and would force
any operating system that really cared to go observe the 
protocol failure indications that would also be set by such
a failure.  Any that did not care would not be helped by
pointers to fields that it is not aware of anyway.

Robert Snively

Brocade Communications Systems, Inc.
1745 Technology Drive
San Jose, CA 95110

+1 408 333 8135
rsnively at brocade.com 


> -----Original Message-----
> From: owner-t10 at t10.org [mailto:owner-t10 at t10.org]On Behalf Of John
> Tyndall
> Sent: Wednesday, November 03, 2004 6:39 AM
> To: t10 at t10.org
> Subject: RE: FCP_DL too small
> 
> 
> * From the T10 Reflector (t10 at t10.org), posted by:
> * "John Tyndall" <jtyndall at Crossroads.com>
> *
> Seems like unnecessary work for the target to find problems in the
> application/host driver. Maybe the host side should just start paying
> attention to the information returned and handle it a little more
> intelligently.
> 
> - J. Tyndall
> 
> jtyndall at crossroads.com
> 512-928-7282
> 
> -----Original Message-----
> From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of Jim
> Jones (Engineering)
> Sent: Tuesday, November 02, 2004 6:47 PM
> To: 'Neil Wanamaker'; t10 at t10.org
> Subject: RE: FCP_DL too small
> 
> * From the T10 Reflector (t10 at t10.org), posted by:
> * "Jim Jones (Engineering)" <Jim.Jones at Quantum.com>
> *
> Neil,
> 
> The original intent of this proposal was to post a check 
> condition when
> the
> FCP_DL was less than the transfer length contained in the CDB.  Here's
> the
> background:
> 
> An application incorrectly specified an FCP_DL (48K) that was smaller
> than
> the transfer length on a Write CDB (64K).  48K of data was 
> written, and
> an
> overrun was reported (FCP_RESID_OVER=1, FCP_RESID=16K).  But the
> application
> also incorrectly only looked at the SCSI Status Byte and 
> because it was
> Good
> Status, continued on.  When the application tried to read the data, it
> reported a data miscompare (because it didn't receive all the 
> data that
> was
> requested).  It took us a long time to sort all this out as an
> application
> error.
> 
> I can't imagine a situation where it's valid for the FCP_DL to be less
> than
> the transfer length contained in the CDB (in particular for DATA OUT
> transfers).  For this reason we support this proposal.
> 
> This proposal wasn't meant to address any situations where the FCP_DL
> didn't
> match the media record.
> 
> Hope this clarifies things.
> 
> Jim
> __________________________
> Jim Jones
> Quantum Corporation
> 4001 Discovery Dr., Suite 2100
> Boulder, CO 80303
> Jim.Jones at quantum.com
> 720-406-5611
>  
> -----Original Message-----
> From: owner-t10 at t10.org [mailto:owner-t10 at t10.org]On Behalf Of Neil
> Wanamaker
> Sent: Monday, November 01, 2004 8:12 AM
> To: t10 at t10.org
> Subject: re: FCP_DL too small
> 
> 
> * From the T10 Reflector (t10 at t10.org), posted by:
> * "Neil Wanamaker" <neil.wanamaker at finisar.com>
> *
> 03-393r3 proposes posting a check condition when FCP_DL is too small,
> together with an indication that
> there are illegal fields in the CDB.
> 
> This proposal makes no distinction between a mismatch between the
> transfer length contained in the CDB 
> and FCP_DL, and a mismatch between FCP_DL and the media record.
> 
> While the proposed change is not unreasonable in the first case, it
> appears excessive in the second.
> 
> Since the days before there was a SCSI, a read of fewer bytes than the
> tape blocksize is legal and results
> in an indication of residual data. If  "the application 
> client often is
> not informed of the error", this seems to 
> be a failing in the driver, rather than the architecture; complete
> information is presented to the ULP.
> 
> Invoking a recovery procedure (as would be the result of a check
> condition) seems inappropriate.
> 
> 
> Neil Wanamaker
> neil.wanamaker at finisar.com
> 512.670.7303 (O) 
> 
> *
> * For T10 Reflector information, send a message with
> * 'info t10' (no quotes) in the message body to majordomo at t10.org
> *
> * For T10 Reflector information, send a message with
> * 'info t10' (no quotes) in the message body to majordomo at t10.org
> 
> *
> * For T10 Reflector information, send a message with
> * 'info t10' (no quotes) in the message body to majordomo at t10.org
> 
> 
> 
*
* 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