LIO Scsi Target

LIO Scsi Target

[Sadegh Ali]

Dear sir

we are considering to build SCSI Target system with ZFS filesystem
backend using Linux
I searched that two modules are available for Linux SCSI target, LIO, and SCST
but it seems LIO project that streamed to the kernel is not updated
for a while (about 7 years)
Is the LIO module project dead? or suspended?
Is any person or community available to respond to technical problems
and fix bugs or develop new features or support new hardware?

with best regards

Re: LIO Scsi Target

[Emmanuel Florac]

[-- Attachment #1: Type: text/plain, Size: 1138 bytes --]

Le Wed, 15 Jul 2020 12:18:14 +0430
Sadegh Ali <sadegh.ali.2084@gmail.com> écrivait:

> Dear sir
> 
> we are considering to build SCSI Target system with ZFS filesystem
> backend using Linux
> I searched that two modules are available for Linux SCSI target, LIO,
> and SCST but it seems LIO project that streamed to the kernel is not
> updated for a while (about 7 years)
> Is the LIO module project dead? or suspended?
> Is any person or community available to respond to technical problems
> and fix bugs or develop new features or support new hardware?

LIO has been integrated into the mainline Linux kernel. The separate
project isn't updated because it's updated together with the kernel.
All common distributions provide the LIO modules and the target-cli
tools to manage LIO.

cheers,
-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

[-- Attachment #2: Signature digitale OpenPGP --]
[-- Type: application/pgp-signature, Size: 163 bytes --]

Re: LIO Scsi Target

[Laurence Oberman]

On Wed, 2020-07-15 at 12:18 +0430, Sadegh Ali wrote:
> Dear sir
> 
> we are considering to build SCSI Target system with ZFS filesystem
> backend using Linux
> I searched that two modules are available for Linux SCSI target, LIO,
> and SCST
> but it seems LIO project that streamed to the kernel is not updated
> for a while (about 7 years)
> Is the LIO module project dead? or suspended?
> Is any person or community available to respond to technical problems
> and fix bugs or develop new features or support new hardware?
> 
> with best regards
> 

There has definitely been a pause in the LIO target code updates.
In fact we use it a lot here (qla2xxx and SRP) at Red Hat but we have
remained on the 4.5 kernel with my jammer patch in the lab because
later updates have been  seeing lots of problems.
I reported this to Marvell and they were looking into it but its been a
long time.

The ISCSI module is very stable but my issues had been mostly with the
qla2xxx target module.

Many vendors who have this in their products took a snapshot many
kernels back and I am not sure how much has been shared back to
upstream.

I have not tested recent upstream kernels for ages so I am also
intrested to know who has been using either the qla2xxx modules or the
SRP module successfully with recent kernels.

Regards
Laurence

Re: LIO Scsi Target

[Laurence Oberman]

On Wed, 2020-07-15 at 10:11 -0400, Laurence Oberman wrote:
> On Wed, 2020-07-15 at 12:18 +0430, Sadegh Ali wrote:
> > Dear sir
> > 
> > we are considering to build SCSI Target system with ZFS filesystem
> > backend using Linux
> > I searched that two modules are available for Linux SCSI target,
> > LIO,
> > and SCST
> > but it seems LIO project that streamed to the kernel is not updated
> > for a while (about 7 years)
> > Is the LIO module project dead? or suspended?
> > Is any person or community available to respond to technical
> > problems
> > and fix bugs or develop new features or support new hardware?
> > 
> > with best regards
> > 
> 
> There has definitely been a pause in the LIO target code updates.
> In fact we use it a lot here (qla2xxx and SRP) at Red Hat but we have
> remained on the 4.5 kernel with my jammer patch in the lab because
> later updates have been  seeing lots of problems.
> I reported this to Marvell and they were looking into it but its been
> a
> long time.
> 
> The ISCSI module is very stable but my issues had been mostly with
> the
> qla2xxx target module.
> 
> Many vendors who have this in their products took a snapshot many
> kernels back and I am not sure how much has been shared back to
> upstream.
> 
> I have not tested recent upstream kernels for ages so I am also
> intrested to know who has been using either the qla2xxx modules or
> the
> SRP module successfully with recent kernels.
> 
> Regards
> Laurence

I should made it clearer that this comment below meant I had not tested
recent upstream kernel LIO modules qla2xxx and SRP.
I run the 4.5 kernel with a jammer patch for both qla2xxx and SRPtarget support in my lab. 

"I have not tested recent upstream kernels for ages so I am also
intrested to know who has been using either the qla2xxx modules or the
SRP module successfully with recent kernels."

Re: LIO Scsi Target

[Lee Duncan]

On 7/15/20 12:48 AM, Sadegh Ali wrote:
> Dear sir
> 
> we are considering to build SCSI Target system with ZFS filesystem
> backend using Linux
> I searched that two modules are available for Linux SCSI target, LIO, and SCST
> but it seems LIO project that streamed to the kernel is not updated
> for a while (about 7 years)
> Is the LIO module project dead? or suspended?
> Is any person or community available to respond to technical problems
> and fix bugs or develop new features or support new hardware?
> 
> with best regards
> 


targetcli-fb and friends has replaced the lio-utils for interfacing with
the LIO subsystem in the kernel. All targetcli-fb really does is make
interacting with LIO much easier. But you could theoretically roll your
own LIO commands, since the interface is sysfs. Targetcli-fb is in
python, and quite user-friendly IMHO.

SCST is still around, but I believe it's user-space only. (Forgive me if
I'm wrong here, but it's been a while since I looked at SCST.)

I'd suggest using something like btrfs instead of zfs, since btrfs is
getting much better now -- even Debian has taken the hint from SUSE and
started using it. :)

Check out the "Kernel Target Devel" list (target-devel at vger).

Much of the kernel target development recently has been to include other
non-iSCSI target support IMHO.

There is also a cool new extension to targetcli called tcmu-runner, that
allows you to add new functionality to targetcli without having to hack
on the kernel.

-- 
Lee Duncan

Re: LIO Scsi Target

[Laurence Oberman]

On Wed, 2020-07-15 at 11:16 -0700, Lee Duncan wrote:
> On 7/15/20 12:48 AM, Sadegh Ali wrote:
> > Dear sir
> > 
> > we are considering to build SCSI Target system with ZFS filesystem
> > backend using Linux
> > I searched that two modules are available for Linux SCSI target,
> > LIO, and SCST
> > but it seems LIO project that streamed to the kernel is not updated
> > for a while (about 7 years)
> > Is the LIO module project dead? or suspended?
> > Is any person or community available to respond to technical
> > problems
> > and fix bugs or develop new features or support new hardware?
> > 
> > with best regards
> > 
> 
> 
> targetcli-fb and friends has replaced the lio-utils for interfacing
> with
> the LIO subsystem in the kernel. All targetcli-fb really does is make
> interacting with LIO much easier. But you could theoretically roll
> your
> own LIO commands, since the interface is sysfs. Targetcli-fb is in
> python, and quite user-friendly IMHO.
> 
> SCST is still around, but I believe it's user-space only. (Forgive me
> if
> I'm wrong here, but it's been a while since I looked at SCST.)
> 
> I'd suggest using something like btrfs instead of zfs, since btrfs is
> getting much better now -- even Debian has taken the hint from SUSE
> and
> started using it. :)
> 
> Check out the "Kernel Target Devel" list (target-devel at vger).
> 
> Much of the kernel target development recently has been to include
> other
> non-iSCSI target support IMHO.
> 
> There is also a cool new extension to targetcli called tcmu-runner,
> that
> allows you to add new functionality to targetcli without having to
> hack
> on the kernel.
> 

Thanks Lee, I will check it out.
The issues I was having are mainly qla2xxx target in the kernel, and I
had been working with Marvell to stabilize that in upstream.
various lockups and other workqueue relateed issues. It been a while.
I will try latest upstream and see if its any better than it was for me
a couple years back.

Re: LIO Scsi Target

[Bart Van Assche]

On 2020-07-15 11:16, Lee Duncan wrote:
> SCST is still around, but I believe it's user-space only. (Forgive me if
> I'm wrong here, but it's been a while since I looked at SCST.)

Hi Lee,

That may be a misunderstanding :-) Almost all of SCST runs in the
kernel, except the code for accepting connections from an iSCSI
initiator. The SCST configuration tool (scstadmin) is a user space tool
that interacts with the SCST kernel modules through the SCST sysfs
interface. SCST still has a significant user base despite not being
upstream.

> There is also a cool new extension to targetcli called tcmu-runner, that
> allows you to add new functionality to targetcli without having to hack
> on the kernel.

My understanding is that the design of tcmu-runner is strongly inspired
by the design of the scst_user driver. Something that's unfortunate is
that we can't ask Shaohua Li anymore about the design of tcmu-runner
(see also
https://lore.kernel.org/lkml/398a74fa-6566-8d0d-6434-e68ff3763656@kernel.dk/).

Bart.

Re: LIO Scsi Target

[Lee Duncan]

On 7/15/20 5:06 PM, Bart Van Assche wrote:
> On 2020-07-15 11:16, Lee Duncan wrote:
>> SCST is still around, but I believe it's user-space only. (Forgive me if
>> I'm wrong here, but it's been a while since I looked at SCST.)
> 
> Hi Lee,
> 
> That may be a misunderstanding :-) Almost all of SCST runs in the
> kernel, except the code for accepting connections from an iSCSI
> initiator. The SCST configuration tool (scstadmin) is a user space tool
> that interacts with the SCST kernel modules through the SCST sysfs
> interface. SCST still has a significant user base despite not being
> upstream.
> 
>> There is also a cool new extension to targetcli called tcmu-runner, that
>> allows you to add new functionality to targetcli without having to hack
>> on the kernel.
> 
> My understanding is that the design of tcmu-runner is strongly inspired
> by the design of the scst_user driver. Something that's unfortunate is
> that we can't ask Shaohua Li anymore about the design of tcmu-runner
> (see also
> https://lore.kernel.org/lkml/398a74fa-6566-8d0d-6434-e68ff3763656@kernel.dk/).
> 
> Bart.
> 
> 

Thanks Bart! Good info. I will update my personal (brain) database on SCST.
-- 
Lee