Re: [PATCH] Add an array that describes which opcodes are supported by the RDWR and SHEEPDOG backends.

On Wed, 30 Oct 2013 18:44:19 -0700
ronnie sahlberg <[email protected]> wrote:

> On Sun, Oct 20, 2013 at 6:34 PM, FUJITA Tomonori
> <[email protected]> wrote:
> > On Sat, 12 Oct 2013 07:38:59 -0700
> > Ronnie Sahlberg <[email protected]> wrote:
> >
> >> While RDWR supports all SBC opcodes that TGTD implement SHEEPDOG
> >> only supports a subset and lacks the following opcodes:
> >>             WRITE_VERIFY10/12/16 VERIFY10/12/16 PREFETCH10/16
> >>             WRITE_SAME10/16 UNMAP and ORWRITE
> >>
> >> This allows backends to specify which opcodes it is prepared to process
> >> and which commands should fail with invalid op code
> >> and allows SHEEPDOG backed LUNs to respond with INVALID_OP_CODE
> >> correctly.
> >>
> >> This is most useful for block devices where we have several different backens
> >> and where some backends only support a subset of the commands
> >>
> >> Signed-off-by: Ronnie Sahlberg <[email protected]>
> >> ---
> >>  usr/bs.c          |    8 ++++++++
> >>  usr/bs_rdwr.c     |   51 +++++++++++++++++++++++++++++++++++++++++++++++++++
> >>  usr/bs_sheepdog.c |   38 ++++++++++++++++++++++++++++++++++++++
> >>  usr/scsi.c        |    6 ++++++
> >>  usr/tgtd.h        |    1 +
> >>  5 files changed, 104 insertions(+), 0 deletions(-)
> >
> > Why backend store's bs_cmd_submit cannot return a proper error and
> > sbr_rc returns that error to an initiator?
> 
> 
> It could, but this way it is more explicit which opcodes are
> available, and a simple compare of the array between opcodes
> makes it easy for non-bs_rdwr backend developers to see the delta
> between their backend and rdwr.
> 
> A second reason is for a followup patch.
> The followup patch will use this array so that it can prune which
> opcodes to report back for the REPORT_SUPPORTED_OPCODE opcode.
> So that when you issue that command to a sheepdog LUN  you get a
> pruned list back that only lists the opcodes that sheepdog supports.

I see, thanks. Using the bitmaps is simpler than the array of char if
you calculate delta and such?