Re: iSCSI : Initiators expected to fake CHECK CONDITIONS.

[Stephen Bailey <steph@cs.uchicago.edu>]

> It would be simple if things where that clear-cut.

It is that simple.  It is clear-cut.

> What is a format error coming from a target?  IMHO it is a target
> error and not a protocol failure.  Should target errors be reported
> in the service-response?

TARGET ERROR clearly IS one of the service response codes defined by
SAM.  Therefore, yep, go ahead and define iSCSI errors in terms of
TARGET FAILURE and SERVICE DELIVERY FAILURE.

SCSI implementations define their own set of service response status
values and whatever iSCSI specifies, a protocol driver will make its
own mapping of protocol conditions to service response status values.
For that reason, the exact specification of TF or SDF is less relevant
than that you identify the condition as a service response reported
failure.

> Except for task management that is common to all protocols I did not
> see any other thing popping up in any SCSI driver.

???  Are you saying that you don't see where the service response
status code is concretely instantiated in ANY SCSI driver?

Take CAM.  The CAM status is the service response status code.  It includes
values like:
  request completed without error
  request aborted by host
  invalid request
  target selection timeout
  command timeout
  data overrun
  unexpected bus free

etc..

Solaris and IRIX are similar (don't have the code on hand).  I can't
imagine that ANY other SCSI driver does not have this component, but I
clearly have not seen all other SCSI drivers.

Did you mean SCSI protocol rather than SCSI driver?  FCP certainly
mentions service response.

> Preaching layering won't make the issue disappear.

Perhaps we could hear a resounding chorus of other participants that
think I'm just `preaching layering' to try to make some issue
disappear.  If so, I'll agree to say no more on this topic.

Presently, it seems to be two (Santosh and me) against one (you).

It's clear how to specify and how implementations will handle, and
have historically handled these types of errors.  Where's the
conflict?  What's hard about this?

Steph