[Stgt-devel] vsd -> vdev is bad

[Ming Zhang]

On Tue, 2005-09-20 at 13:15 -0500, Mike Christie wrote:
> Ming Zhang wrote:
> > Hi Mike
> > 
> > I think change vsd->vdev is a bad idea.
> 
> I would agree it is a bad name, but all that device does today is the 
> reads and writes. It should probably be called something-something-IO. 
>  From your experience with iet, do you think a read or write will be 
> different for tape or disk when using the interface we are using?

yes, it is fileio in fact.

any disk based virtual object, even OSD i feel, will come to (r/w, lba,
size, buf) eventually. so yes, i feel tgat all device will use finally
do same/simialr io with this vdev. so  will have

tgt -- VD   ---- fileio
    -- VMMC 
    -- VT
    -- OSD
    -- ...

> 
> For any other non-read/write command it looks like we are going to 
> userspace for now. Originally we had the possiblity to do passthrough 

good point for Vx devices.

for pass-through/bridge like stuff, u might want CDB relay to next
level.


> using blk_execute_rq_nowait so we could do some interesting things in 
> the kernel, but not anymore - hopefully temporarily. Patches are welcome 
> though :) It was just a matter of having time to do things becuase we 
> would need to account for the destination device's limits (segments, 
> segment size, max_sectors, max cdb size for scsi, etc), and the target 
> drivers limits. And the target driver's limits are a priority because we 
> must support HW targets like qla2xxx.

yes, looking forward to that. One thing I like SCST very much is its
clear document. It has great help on writing front/back end for scst.

> 
> The reason for the concern is that we are duplicating a lot of block 
> layer code. The tgt_cmnd is starting to look like a request, the code to 
> make and use scatterlists will eventually look like the block layers, 
> and eventually for mainline I am guessing we will have to merge 
> functionality so I am trying to stay small as possible so we can push 
> common code. I am in the middle of cleaning up the SCSI ULDs to do this 
> and it is a pain in the butt :)

yes, saw u post on scsi list.


> 
> Send me a patch and I will at least change the name if you are thinking 
> a read for a tape will be different for a read for disk, or can think of 
> a better name to convey that it is only a interface to do IO. I was 
> thinking if tape or cd does become much different we will need something 
> like the SCSI ULDs to abstract this to avoid some code duplication.

no, they are almost same. one thing here is tape/mmc might generate
extra io activity while disk is a pure dumb device.


> 
> > 
> > There can be many different types of virtual device.
> > 
> > so if a virtual disk device is vdev, then a virtual tape will be vtdev?
> > i think name like these should be better.
> > 
> > vd virtual disk
> > vt virtual tape
> > vg - virtual generic
> > vmmc - virtual mmc device
> > 
> > my 2c.
> > 
> > Ming
> > 
> > 
> > --------------
> > Log:
> > rename vsd to vdev. NOTE YOU MUST CHANGE YOUR IETD.CONF TO REFLECT THIS.
> > tgt_vsd IS NOW tgt_vdev
> > 
> > _______________________________________________
> > Stgt-devel mailing list
> > Stgt-devel at lists.berlios.de
> > http://lists.berlios.de/mailman/listinfo/stgt-devel
> 
> _______________________________________________
> Stgt-devel mailing list
> Stgt-devel at lists.berlios.de
> http://lists.berlios.de/mailman/listinfo/stgt-devel