Re: [Stgt-devel] Question for pass-through target design

[Vladislav Bolkhovitin <vst@vlnb.net>]

Robert Jennings wrote:
> * Vladislav Bolkhovitin (vst@vlnb.net) wrote:
> 
>>Robert Jennings wrote:
>>
>>>>>>What I meant that is that the kernel tgt code (scsi_tgt*) receives
>>>>>>SCSI commands from one lld and send them to another lld instead of
>>>>>>sending them to user space.
>>>>>
>>>>>Although the approach of passing SCSI commands from a target LLD to an
>>>>>initiator one without any significant interventions from the target
>>>>>software looks to be nice and simple, you should realize how limited,
>>>>>unsafe and illegal it is, since it badly violates SCSI specs.
>>>>
>>>>I think that 'implemented cleanly' means that one scsi_host is assigned
>>>>to only one initiator.
>>>
>>>Vladislav listed a number of issues that are inherent in an implementation
>>>that does not have a 1:1 relationship of initiators to targets.  The vscsi
>>>architecture defines the 1:1 relationship; it's imposible to have more
>>>than one initiator per target.
>>
>>Just few small notes:
>>
>>1. As I already wrote, complete 1:1 relationship isn't practically 
>>possible, because there is always a local access on the target (i.e. one 
>>more initiator) and you can't disable it on practice.
> 
> I was proposing a 1:1 relationship of initiator to target within the
> target framework for in-kernel pass-through.  We would still have the
> case that local access on the target is possible; an administrator with
> privileges neccessary to create a target would have the responsibility
> to not then access the device locally.  
> 
> This is no different than if I create my root file system on /dev/sda1,
> I should not also 'dd' data to /dev/sda1 while the system is running.
> It's a bad idea, but nothing stops me; however this is something that
> only a root level user can do.  This would be the same, these targets in
> pass-through have permissions by default that do not allow local access
> by non-root users.

In principle, yes, but, as usually, on practice it's not so easy. In 
your file system example the device is accessed via the FS, which 
provides a shared mode, and everybody doesn't have any need to do 
anything directly with the device. But in case of non-disk devices they 
are always accessed directly, so to explain your limitation you will 
have to write it with HUGE letters everywhere. Once one SCST user 
cleared Unit Attention on his exported tape device using st driver and 
asked then me why it isn't delivered to his remote initiator.

>>2. 1:1 relationship is a serious limitation for usage cases like an SPI 
>>tape library serving backup for several servers on an FC net.
> 
> Restricting the relationship to 1:1 would be for pass-through devices
> only, this would not necessarily dictate other target types which could
> be used for such cases.

The tape library from my example is the pass-through device. You can't 
access a parallel SCSI (SPI) device on an Fibre Channel (FC) in any 
other mode, right?

Vlad