use dedicated storage for scsi_debug LUNs

use dedicated storage for scsi_debug LUNs

[Olaf Hering]

I'm using scsi_debug to implement and test tools for xen-scsiback. Today
I discovered that each LUN uses the very same storage, as described in
the docs. What would it take to optionally use dedicated storage for
each LUN?

This is how I load scsi_debug:
modprobe scsi_debug add_host=4 dev_size_mb=4 max_luns=4 num_parts=4 num_tgts=4 vpd_use_hostno=1

Olaf

Re: use dedicated storage for scsi_debug LUNs

[Douglas Gilbert]

On 15-03-05 10:33 AM, Olaf Hering wrote:
> I'm using scsi_debug to implement and test tools for xen-scsiback. Today
> I discovered that each LUN uses the very same storage, as described in
> the docs. What would it take to optionally use dedicated storage for
> each LUN?
>
> This is how I load scsi_debug:
> modprobe scsi_debug add_host=4 dev_size_mb=4 max_luns=4 num_parts=4 num_tgts=4 vpd_use_hostno=1

Olaf,
Just having a quick look: the fake_storep, dif_storep and
map_storep heaps would need to be per LU; IOWs placed in
struct sdebug_dev_info. Also atomic_rw would need to be
placed in that structure.

To be backward compatible to what the scsi_debug driver
does now (i.e. shares storage) accessor functions might
be employed to hide whether the global or per LU instance
is being handled.

Handling the inevitable ENOMEM errors might be a bit tricky
especially since extra LUs and hosts can be added dynamically
via sysfs.


So it is possible. The stock answer has typically been that
you should probably be using the target subsystem for that.
Thoughts?

Doug Gilbert

Re: use dedicated storage for scsi_debug LUNs

[Olaf Hering]

On Thu, Mar 05, Douglas Gilbert wrote:

> So it is possible. The stock answer has typically been that
> you should probably be using the target subsystem for that.
> Thoughts?

Is the target subsystem capable of stacking? What I need for
xen-scsiback is a bunch of true SCSI devices. Due to lack of hardware,
scsi_debug looks like the obvious choice.

I will see if I can come up with a patch. Thanks for your input.

Olaf

Re: use dedicated storage for scsi_debug LUNs

[Martin K. Petersen]

>>>>> "Olaf" == Olaf Hering <olaf@aepfle.de> writes:

Olaf> On Thu, Mar 05, Douglas Gilbert wrote:
>> So it is possible. The stock answer has typically been that you
>> should probably be using the target subsystem for that.  Thoughts?

Olaf> Is the target subsystem capable of stacking? What I need for
Olaf> xen-scsiback is a bunch of true SCSI devices. Due to lack of
Olaf> hardware, scsi_debug looks like the obvious choice.

I agree with Doug. The target code is much more comprehensive than
scsi_debug. You can set up file/disk/ram-backed SCSI LUNs on your
machine without the need for any hardware .

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: use dedicated storage for scsi_debug LUNs

[Olaf Hering]

On Fri, Mar 06, Martin K. Petersen wrote:

> I agree with Doug. The target code is much more comprehensive than
> scsi_debug. You can set up file/disk/ram-backed SCSI LUNs on your
> machine without the need for any hardware .

Does that target code offer what scsi_debug offers, a bunch of stuff
listable with 'lsscsi'? The latter can be passed to backstores/pscsi.

Olaf

Re: use dedicated storage for scsi_debug LUNs

[Martin K. Petersen]

>>>>> "Olaf" == Olaf Hering <olaf@aepfle.de> writes:

Olaf> Does that target code offer what scsi_debug offers, a bunch of
Olaf> stuff listable with 'lsscsi'? The latter can be passed to
Olaf> backstores/pscsi.

Yes.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: use dedicated storage for scsi_debug LUNs

[Olaf Hering]

On Fri, Mar 06, Martin K. Petersen wrote:

> >>>>> "Olaf" == Olaf Hering <olaf@aepfle.de> writes:
> Olaf> Does that target code offer what scsi_debug offers, a bunch of
> Olaf> stuff listable with 'lsscsi'? The latter can be passed to
> Olaf> backstores/pscsi.
> Yes.

Indeed, with /backstores/fileio/ and /loopback/.

It would be nice if one could specify a fixed identifier for each fileio
during creation. Right now the symlinks in /dev/disk/by-*/* are all
random. Its not easily possible to map file_or_dev=$file to the
resulting SCSI device.

Olaf