On Fri, Dec 3, 2010 at 10:42 PM, Stefan Hajnoczi <stefa...@gmail.com> wrote: > On Fri, Dec 3, 2010 at 11:09 AM, <ronniesahlb...@gmail.com> wrote: >> This series of pathces adds built in iscsi support to qemu. > > Christoph asked why block/iscsi.c exists and we don't have something > like hw/iscsi-disk.c. Any thoughts on "pass-through" SCSI support so > CDBs from guest SCSI can be written straight to the iSCSI LUN? Could > we support this mode of operation?
That is very simple to add and uses an existing api i already use for my own private sg3-utils branch. ./block.iscsi/scsi-command.c offers an api where you provide a "struct scsi_task *" structure containing some basic data and the cdb. iscsi_scsi_command_async(). which later invokes the callback with status == SCSI_STATUS_GOOD in which case the buffer from the device contains the data or SCSI_STATUS_CHECK_CONDITON in which case that buffer contains the sense blob instead. So yes, there is already an api in the library that offers you to pass scsi cdb/dataout/datain/response to the device. Instrumenting this to be available to the virtual host as a "passthrough" scsi device should be relatively simple as a second step. > > I like block/iscsi.c because it allows any storage interface > (including virtio and ide) to talk to an iSCSI LUN. So I don't think > it makes sense to throw away block/iscsi.c. But the idea of a direct > iSCSI pass-through might be interesting too. That should be very easy to add once the foundation is added into qemu. I would very much like to see bus=scsi make this layer a transparent passthrough layer too, just like it is today for the special case where the file specified matches "/dev/sg" ronnie sahlberg
