Re: Request for review of Linux iSCSI driver version 4.0.0.1

[James Bottomley <James.Bottomley@steeleye.com>]

On Thu, 2003-12-04 at 07:30, N.C.Krishna Murthy wrote:
> Having user level utility trigger the rescan thru sysfs interfaces will call
> scsi_scan() which will inturn call scsi_scan_host_selected().
> Instead wouldn't it be preferable to call scsi_scan_host_selected()
> from the iSCSI driver itself?

Well, consider all the ramifications: what happens if the
reconfiguration is deleting an existing LUN with a mounted filesystem? 
What happens if the user wants to do something with the new LUN as soon
as it appears (like fold it into a VM volume, or format it).  The
hotplug system is designed to propagate events that others might be
interested in, which it strikes me this is.  So, if you have to trigger
a hotplug event anyway, you may as well go the whole hog and do the
reconfiguration as part of the hotplug subsystem.

> > We could do with a generic infrastructure within SCSI for generating
> > these events, anyway, if you want to go that route...
> >
> I am not much familiar with SCSI hotplug. 
> 
> If the generic infrastructure which you mentioned does not exist now and in 
> case there is a plan to do one such thing is there any framework for that?
> Wouldn't that require a kernel API which is capable of both 
> "adding new luns" and "deleting luns which no longer exist"?

Well, a hotplug event feeds certain information into the environment
when it triggers the call.  All we really need is a wrapper in SCSI to
add the SCSI information to the environment and trigger the event in a
generic fashion.

James