On Tue, Apr 28, 2009 at 1:20 PM, Milos Muzik <[email protected]> wrote:
>
> Andrey Kuzmin wrote:
>> IIRC (it's over a years since I last looked at user-space iSCSI target
>> code), iscsit supports pass-through mode so that running local tape
>> target should not be a problem. Of course,  whether such a mode was
>> envisaged/designed for is a totally different question.
>>
>> Regards,
>> Andrey
>>
>
> The problem is that iscsitgtd sends SCSI commands to the backing device on its
> own in raw mode during initialization phase. This is an excerpt from iscsitadm
> man page:
>
>                                         ......... raw  indicates
>         that  the emulator will use the uSCSI interface and pass
>         the command blocks directly to and from the device.  The
>         use  of raw also implies the option --backing-store will
>         be entered. The argument to  this  option  is  the  full
>         pathname  to  the device node normally found in /dev. If
>         you use --backing-store, the size of the store is deter-
>         mined by a SCSI READ_CAPACITY command or, if the backing
>         store is a regular file, by stat(2).
>
> The consequence of the text above is that tape device can't be used as a 
> backing
> store even in raw mode. The READ_CAPACITY which is always sent to the device 
> is
> a block device specific command and fails with sense key 0x5 (Illegal Request)
> when sent to a tape.
>
> It seems that the iscsitgtd was not designed for tape backing at all.
>

Right, though I'm unclear on why iscsitgtd starts with READ_CAPACITY
instead of passing through target discovery commands. Anyway, driving
tape target apparently was not envisaged.


Regards,
Andrey

> Regards,
> Milos
>
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to