[Stgt-devel] uSpace Transport Patch

[FUJITA Tomonori]

From: Tom Tucker <tom at opengridcomputing.com>
Subject: Re: [Stgt-devel] uSpace Transport Patch
Date: Tue, 22 Aug 2006 07:39:53 -0500

> On Tue, 2006-08-22 at 08:11 +0900, FUJITA Tomonori wrote:
> > From: Tom Tucker <tom at opengridcomputing.com>
> > Subject: [Stgt-devel] uSpace Transport Patch
> > Date: Mon, 21 Aug 2006 16:29:40 -0500
> > 
> > > Tomo:
> > > 
> > > Enclosed is a patch that allows you to plug in multiple transports. It
> > > has a few benefits over the last approach:
> > 
> > Thanks.
> > 
> > 
> > > 1. The TCP side can remain exactly the same. i.e. user-mode connection
> > > management and login send/recv.
> > > 
> > > 2. The stgtd implementation still uses pollfd to receive I/O events. The
> > >    iser side will provide an fd that can be polled.
> > > 
> > > 
> > > I have built and run this patch with the current code and connected with
> > > a iscsi initiator over TCP. I did encounter problems, however, trying to
> > > do disk i/o.
> > 
> > The write path code is broken. I will fix it if the kernel-mode
> > approach would likely be accepted into mainline.
> > 
> > The user-space mode code should work better. As I said in the previous
> > mail, I can do mkfs, extract linux kernel tar, etc with the open-iscsi
> > default configuration.
> > 
> > 
> > > This is not done, it is a proof-of-concept/design proposal. The netlink
> > > stuff will obviously change as I flesh out the iSER side. Please let me
> > > know what you think, I'd like your opinion before I get to far down this
> > > road.
> 
> > I see. I will read this soon. Can you post iSER part code too?
> 
> Yes, I'm working on that now, I just didn't want to get too far down a
> road that lead somewhere we didn't want to go. Basically, the iSER code
> is implemented in the kernel. The user-mode interface is a file
> descriptor (used for polling), and a either a) set of methods
> implemented via read/write for listening for, accepting and
> send/receiving, or b) using the netlink interface. I'm leaning towards
> the former actually.

The previous patch uses the latter approach (similar to open-iscsi),
right? So, could you explain the formaer approach a bit further?