Re: iSCSI: Logout request

[Santosh Rao <santoshr@cup.hp.com>]

I don't see why any such notes are required. This is not an
implementation guide, its a protocol wire spec. There are other SCSI
transport protocols that allow the target to do an un-solicited session
termination, without requiring any implementation notes advising the
initiators to re-establish their sessions.

All of the below discussion is implementation specific value-add. A high
end server may require its initiator to attempt persistent re-connects
to target ports that are in use by upper layers. A dumb PC driver may
not need to do this. It is outside the scope of the "scsi transport
protocol" to define what the implementation should do in this case.
These requested words fit in the realm of the implementation's design
spec or some iscsi implementator's guide, not in the core protocol spec,
IMO.

- Santosh


Paul Koning wrote:
> 
> My request is for some words that tell implementations to consider
> this possibility.
> 
> Without such words, the problem is that some initiator implementers
> think that a target logout request means "go away and NEVER come
> back", so they treat that as "device has gone offline" and start
> returning fatal errors to subsequent I/O requests.  Since it's not the
> intent of the spec that it should be viewed that way, it would help
> those of us who want to use target logout request in the manner I
> described to have the spec say this is a possibility.
> 
>         paul
> 
> 
> > ----- Original Message -----
> > From: "Paul Koning" <ni1d@arrl.net>
> > To: <cbm@rose.hp.com>
> > Cc: <ips@ece.cmu.edu>
> > Sent: Tuesday, April 30, 2002 11:51 AM
> > Subject: Re: iSCSI: Logout request
> >
> >
> > > Excerpt of message (sent 30 April 2002) by Mallikarjun C.:
> > > > >The target may have good reasons to want a
> > > > > particular connection or session to be closed but allowing a new login
> > > > > to occur right after that.
> > > >
> > > > It is a legal possibility, but I think it's unlikely for the reasons described
> > > > above.  If the NIC is being replaced on the target end, an immediate new
> > > > Login attempt will fail.
> > >
> > > Not necessarily, and that's why I bring up the issue.
> > >
> > > If the target has multiple NICs, and one of them has to be taken out
> > > of service, it may be that it can move the address (but not the open
> > > session and connection state) to another NIC.  In that case, a "target
> > > request logout" is the mechanism it would want to use to initiate that
> > > operation, and if the initiator then follows up with a new login,
> > > everything works smoothly.
> > >
> > >    paul
> > >
> > >

-- 
We have enough people who tell it like it is
Now we could use a few who tell it like it can be.

-  Robert Orben