Support for CDB length > 16

Support for CDB length > 16

[Liran Schour]

Hi all,

I need a support from the Linux SCSI subsystem for CDB length > 16 byte.
Is it possible to do that without recompiling the kernel with a new
MAX_COMMAND_SIZE definition. I assume that I can find an iscsi driver that
is able to support CDB > 16 byte.

Thanks,
- Liran

Re: Support for CDB length > 16

[Douglas Gilbert]

Liran Schour wrote:
> 
> I need a support from the Linux SCSI subsystem for CDB length > 16 byte.
> Is it possible to do that without recompiling the kernel with a new
> MAX_COMMAND_SIZE definition. 

It is hard to imagine any other way at the moment (definitely
in lk 2.4 and it is probably a bit late for lk 2.6).

Is there an upper limit on the size of the cdb you are interested in
(e.g. XDWRITE EXTENDED in SBC-2 is 64 bytes)?

How were you planning to pass the cdb (length > 16) to the kernel?

Doug Gilbert

Re: Support for CDB length > 16

[Liran Schour]

Doug Gilbert wrote:

> It is hard to imagine any other way at the moment (definitely
> in lk 2.4 and it is probably a bit late for lk 2.6).

> Is there an upper limit on the size of the cdb you are interested in
> (e.g. XDWRITE EXTENDED in SBC-2 is 64 bytes)?
I have an upperlimit of 256 bytes.

> How were you planning to pass the cdb (length > 16) to the kernel?
I am planning to build a filesystem in the kernel that will be over an OSD
SCSI device. This layer will interact directly with the SCSI system.

Thanks,
- Liran

Re: Support for CDB length > 16

[Andre Hedrick]

I suspect this is midlayer injection for SAS.

Andre Hedrick
LAD Storage Consulting Group

On Tue, 18 Nov 2003, Douglas Gilbert wrote:

> Liran Schour wrote:
> > 
> > I need a support from the Linux SCSI subsystem for CDB length > 16 byte.
> > Is it possible to do that without recompiling the kernel with a new
> > MAX_COMMAND_SIZE definition. 
> 
> It is hard to imagine any other way at the moment (definitely
> in lk 2.4 and it is probably a bit late for lk 2.6).
> 
> Is there an upper limit on the size of the cdb you are interested in
> (e.g. XDWRITE EXTENDED in SBC-2 is 64 bytes)?
> 
> How were you planning to pass the cdb (length > 16) to the kernel?
> 
> Doug Gilbert
> 
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: Support for CDB length > 16

[Christoph Hellwig]

On Tue, Nov 18, 2003 at 10:49:44AM +0200, Liran Schour wrote:
> 
> 
> 
> 
> Hi all,
> 
> I need a support from the Linux SCSI subsystem for CDB length > 16 byte.
> Is it possible to do that without recompiling the kernel with a new
> MAX_COMMAND_SIZE definition. I assume that I can find an iscsi driver that
> is able to support CDB > 16 byte.

You'd have to enlarge MAX_COMMAND_SIZE, that'S the whole point of it :)
What exactly are you trying to do and for what kernel release?  We've
been through increasing it from 12 to 16 at some point in 2.4.

Re: Support for CDB length > 16

[Christoph Hellwig]

On Tue, Nov 18, 2003 at 01:15:31PM +0200, Liran Schour wrote:
> > Is there an upper limit on the size of the cdb you are interested in
> > (e.g. XDWRITE EXTENDED in SBC-2 is 64 bytes)?
> I have an upperlimit of 256 bytes.

That's far beyond what we could increase the cmnd buffers to.

> 
> > How were you planning to pass the cdb (length > 16) to the kernel?
> I am planning to build a filesystem in the kernel that will be over an OSD
> SCSI device. This layer will interact directly with the SCSI system.

I don't think there's a lot of chance to see your project supported in
mainline kernels.  Maybe we'll have to find a way to make it work when
it's ready for mainline submission, but before that you're on your own..

Re: Support for CDB length > 16

[Liran Schour]

Christoph Hellwig wrote:

> You'd have to enlarge MAX_COMMAND_SIZE, that'S the whole point of it :)
> What exactly are you trying to do and for what kernel release?  We've
> been through increasing it from 12 to 16 at some point in 2.4.

I am trying to find a way to do that without the need to recompile the
kernel.
I am trying to support an OSD device see t10 OSD draft, I think that 2.6 is
my target kernel release.

Thanks,
- Liran

Re: Support for CDB length > 16

[Douglas Gilbert]

Liran Schour wrote:
> 
> 
> 
> Christoph Hellwig wrote:
> 
> 
>>You'd have to enlarge MAX_COMMAND_SIZE, that'S the whole point of it :)
>>What exactly are you trying to do and for what kernel release?  We've
>>been through increasing it from 12 to 16 at some point in 2.4.
> 
> 
> I am trying to find a way to do that without the need to recompile the
> kernel.
> I am trying to support an OSD device see t10 OSD draft, I think that 2.6 is
> my target kernel release.

Liran,
Yep, I knew that I'd seen really big SCSI commands somewhere.
OSD, now there is a storage paradigm that doesn't fit
the block subsystem:-)

Perhaps we should starting thinking about kmalloc()ing
scsi_cmnd::cmnd (rather than having it as a fixed array).
The mid-level owns that structure so it can take care
of allocation and de-allocation. We would have to
find code that does sizeof(scsi_cmnd::cmnd) and fix it.

Doug Gilbert

Re: Support for CDB length > 16

[Christoph Hellwig]

> Perhaps we should starting thinking about kmalloc()ing
> scsi_cmnd::cmnd (rather than having it as a fixed array).
> The mid-level owns that structure so it can take care
> of allocation and de-allocation. We would have to
> find code that does sizeof(scsi_cmnd::cmnd) and fix it.

That would be the way to go if we want to support that big commands,
maybe with an pre-allocated array for 12byte commands so that the
fast-path doesn't get slowed down.

But I'm not going to do that for some vapur-ware, we'll need a real-life
application for that.