Re: iscsi: sequence timeout

["Mallikarjun C." <cbm@rose.hp.com>]

Gwendal,

Yes, "sequence timeout" should be changed to "sequence reception 
timeout" for consistency.  Thanks.

> occurs. According to the document, in 5.14.1.b), a sequence timeout is

Only the option of a recovery R2T is allowed in this case, given that
it's the description of "recovery within-command" (and 5.14.1.b makes it
clear).

>could it be mentioned the target can
> terminate the command when the Sequence reception timeout pops up

Only if it's certain that there are no more PDUs relating to the same 
ITT in the pipe *and* the initiator won't issue any further.  The only way 
to ascertain this to ensure that the task is terminated on either end.

>Should the target take more
> drastic action to prevent this situation?
 
Yes, one of the means to force a termination by the target is to drop the 
session.  And I don't recommend it on a sequence reception timeout for 
one I/O.....
--
Mallikarjun

Mallikarjun Chadalapaka
Networked Storage Architecture
Network Storage Solutions
Hewlett-Packard MS 5668 
Roseville CA 95747
cbm@rose.hp.com


----- Original Message ----- 
From: "Gwendal Grignou" <ggrignou@silverbacksystems.com>
To: <ips@ece.cmu.edu>
Sent: Tuesday, August 27, 2002 1:03 PM
Subject: iscsi: sequence timeout


> Hello,
> 
> First of all, I have a little editing comment: for coherency,
> "Sequence timeout" should be renamed "Sequence reception timeout". (2
> occurrences)
> 
> I have a question about what to do when the Sequence reception timeout
> occurs. According to the document, in 5.14.1.b), a sequence timeout is
> considered as an implicit sequence errors documented in 5.8 which is
> considered as an implicit digest errors documented in 5.7. Target can
> either send a recovery R2T or complete the command with sense data,
> indicating a "protocol service CRC error". But it is also mentioned in
> 5.7 that the target should wait for all the data, signaled with a Data
> PDU with an F bit) before finishing that command...
> 
> To solve this chicken&egg thing, could it be mentioned the target can
> terminate the command when the Sequence reception timeout pops up? But
> then it has to handle the case the final DataOut PDU arrives after the
> "Sequence reception timeout" interval. Should the target take more
> drastic action to prevent this situation?
> 
> Gwendal.
> 
>