RE: Data in SCSI Response or SCSI Data

["Douglas Otis" <dotis@sanlight.net>]

Bob,

Explicit definitions found in FCP are better than implied.  Data transfers
should be punctuated with an end flag and all data transfers terminated with
a FCP_RSP like structure.  Any length data payload may be split across
frames, and to allow recovery without sideband signaling, a terminating
response is required.  If the goal is to save a few bytes, reduce header
size but do not change operation on some conditional basis.  Each
implementation will likely abuse intended use as it creates further
conditional implementations based on a few bits here and there and opinions
as to what small is.  I do not see elegance in an option of where to put a
response structure. (In front or behind the actual data transfer.)

The FCP residual value is to reconcile bytes indicated within the SAM
request.  The initiator is also be able to directly calculate bytes
transferred and residual.  The difficulty is with overlays by muddled
targets.  As such, let the target sort it out.

Doug


> -----Original Message-----
> From: Robert Snively [mailto:rsnively@Brocade.COM]
> Sent: Wednesday, September 13, 2000 7:39 AM
> To: Douglas Otis; Robert Reynolds; 'Stephen Bailey'; ips@ece.cmu.edu
> Subject: RE: Data in SCSI Response or SCSI Data
>
>
>
> Doug,
>
> It is true that CHECK CONDITION is not presented for transfers
> shorter than the allocation when the SILI bit is set on tape
> drives.  However, for all other cases, it is.
>
> In any case, FCP provides an independent under-run/over-run
> condition that provides the necessary information to
> "use other methods" that identify shorter or longer than
> expected transfers.  These were provided because the Fibre
> Channel transport mechanism is potentially independent from
> the SCSI command set operations.
>
> Basically, it is always there when the application program
> desires to use it.  Normally, variable length parameter fields
> such as INQUIRY data fields, are self-defining in length.  You
> merely set out a large allocation space and then parse the
> resulting data object to find out where it ends and if it
> was truncated in an undesirable way.
>
> These are interesting items to keep in mind, and are one of the
> reasons we recommend the FCP_RSP format for the actual response.
>
> Bob
>
>
>
>
> >  -----Original Message-----
> >  From: Douglas Otis [mailto:dotis@sanlight.net]
> >  Sent: Monday, September 11, 2000 11:45 AM
> >  To: Robert Reynolds; 'Stephen Bailey'; ips@ece.cmu.edu
> >  Subject: RE: Data in SCSI Response or SCSI Data
> >
> >
> >  Bob,
> >
> >  SAM-2
> >  "As specified in clause 5, the application client may
> >  request autosense
> >  service for any SCSI command. If supported
> >  by the protocol and logical unit and requested by the
> >  application client,
> >  the device server shall only return sense
> >  data in this manner coincident with the completion of a
> >  command with a
> >  status of CHECK CONDITION."
> >
> >  SSC
> >  "If the suppress incorrect length indicator (SILI) bit is
> >  one and the FIXED
> >  bit is zero, the device server shall:
> >  a) Report CHECK CONDITION status for an incorrect length
> >  condition only if
> >  the overlength condition
> >  exists and the BLOCK LENGTH field in the mode parameter
> >  block descriptor is
> >  nonzero (see SPC-2); or
> >  b) not report CHECK CONDITION status if the only error is
> >  the underlength
> >  condition, or if the only error
> >  is the overlength condition and the BLOCK LENGTH field of the mode
> >  parameters block descriptor is zero.
> >
> >  NOTE 9 Since the residue information normally provided in
> >  the INFORMATION
> >  field of the sense data may not
> >  be available when the SILI bit is set, other methods for
> >  determining the
> >  actual block length should be used
> >  (e.g. including length information in the data block).
> >
> >  If the SILI bit is one and the FIXED bit is one, the device
> >  server shall
> >  terminate the command with CHECK
> >  CONDITION status and the sense key shall be set to ILLEGAL
> >  REQUEST with an
> >  additional sense code
> >  and an additional sense code qualifier of INVALID FIELD IN CDB.
> >
> >  If the SILI bit is zero and an incorrect length block is read, CHECK
> >  CONDITION status shall be returned
> >  and the ILI and VALID bits shall be set to one in the sense
> >  data with an
> >  additional sense code and an
> >  additional sense code qualifier of NO ADDITIONAL SENSE
> >  INFORMATION. Upon
> >  termination, the
> >  logical position shall be after the incorrect length block
> >  (end-of-partition
> >  side). If the FIXED bit is one, the
> >  INFORMATION field shall be set to the requested transfer
> >  length minus the
> >  actual number of blocks read
> >  (not including the incorrect length block). If the FIXED bit
> >  is zero, the
> >  INFORMATION field shall be set to the
> >  requested transfer length minus the actual block length.
> >  Logical units that
> >  do not support negative values
> >  shall set the INFORMATION field to zero if the overlength
> >  condition exists."
> >
> >  From these definitions, do not expect any defined
> >  information regarding
> >  residue without a Check Condition.  The residue of iSCSI
> >  within the Read PDU
> >  0x45 is not SCSI related nor does it represent the
> >  underlength condition you
> >  suggest.  Such underlength data is normally lost and
> >  uninteresting if the
> >  SILI bit is set.
> >
> >  Doug
> >
> >  > -----Original Message-----
> >  > From: owner-ips@ece.cmu.edu
> [mailto:owner-ips@ece.cmu.edu]On Behalf Of
> > Robert Reynolds
> > Sent: Monday, September 11, 2000 8:07 AM
> > To: 'Stephen Bailey'; ips@ece.cmu.edu
> > Subject: RE: Data in SCSI Response or SCSI Data
> >
> >
> > > -----Original Message-----
> > > From: Stephen Bailey [mailto:steph@cs.uchicago.edu]
> > > Sent: Wednesday, September 06, 2000 5:45 PM
> > > To: ips@ece.cmu.edu
> > > Subject: Re: Data in SCSI Response or SCSI Data
> > >
> > >
> > > > When do you get GOOD status and residual counts on a read?
> > > What is causing
> > > > the target to get the length wrong?
> > >
> > > One example is an INQUIRY command.  The inquiry data length is
> > > target-specific.  Typically the CDB allocation length (and DL) are set
> > > to some arbitrary large value (0xff), and the target sends back
> > > everything it has.  The transfer ends with success status.
> > >
> > > There can certainly be transfer residual and no SCSI error status.
> >
> >
> >   Another example is on variable block tape reads.  The Initiator
> >   uses an arbitrary size to read the block and the target sends all
> >   the data in that block.  In many cases the amount of data returned
> >   is less than the amount of data asked for but it is still a GOOD
> >   status.
> >
> >     Bob
> >
>