Re: Pe: [PATCH v5 1/3] virtio-scsi: first version

[ronnie sahlberg]

On Tue, Feb 14, 2012 at 9:59 AM, Michael S. Tsirkin <mst@xxxxxxxxxx> wrote:
> On Tue, Feb 14, 2012 at 07:53:26AM +1100, ronnie sahlberg wrote:
>> On Tue, Feb 14, 2012 at 7:42 AM, ronnie sahlberg
>> <ronniesahlberg@xxxxxxxxx> wrote:
>> > On Tue, Feb 14, 2012 at 2:12 AM, Hannes Reinecke <hare@xxxxxxx> wrote:
>> >> On 02/13/2012 02:18 PM, Michael S. Tsirkin wrote:
>> >>> On Tue, Feb 14, 2012 at 12:13:36AM +1100, ronnie sahlberg wrote:
>> >>>> On Tue, Feb 14, 2012 at 12:00 AM, Michael S. Tsirkin <mst@xxxxxxxxxx> wrote:
>> >>>>> On Mon, Feb 13, 2012 at 02:54:03PM +0200, Dor Laor wrote:
>> >>>>>> Only if you use the pci multi-function option but that kills
>> >>>>>> standard hot unplug
>> >>>>>
>> >>>>> It doesn't kill it as such, rather you can't unplug luns individually.
>> >>>>
>> >>>> Isnt that just a consequence of the current implementation rather than
>> >>>> a SCSI limitation?
>> >>>
>> >>> Yes.
>> >>>
>> >>>> A different way to do hoplug could be to flag all devices as removable
>> >>>> in the standard inq page then
>> >>>> leave the LUN there persistently and what you remove/add is not the
>> >>>> LUN device itself but just the media in the device.
>> >>>>
>> >>>> Instead of hot-plug remove the LUN,  hot-plug becomes "media eject" or
>> >>>> "media insert".
>> >>>> The device remains present all time, you never remove it, but instead
>> >>>> hot-plug controls if the media is present or not.
>> >>>>
>> >>>>
>> >>>> This would require implementing at least START_STOP_UNIT and
>> >>>> PREVENT_ALLOW_MEDIUM_REMOVAL opcode emulation from SBC.
>> >>>>
>> >>>>
>> >>>> regards
>> >>>> ronnie sahlberg
>> >>>
>> >>> That would work.
>> >>>
>> >> Or we simply use the Peripheral Qualifier that the device is gone;
>> >> eg we could simply set PQ = 1, return sense code 0x25/00 and be done
>> >> with ...
>> >>
>> >
>> > That is still similar to "rip a device out from the guest without notice"
>> > and can cause the guest to be "surprised".
>> >
>> >
>> > Removable media is standard feature in SCSI SBC (and other commandsets).
>> > The nice part of removable media is that it activates a contract
>> > between the device and the guest
>> > to prevent removal of the media when the guest depends on the media
>> > not being removed.
>> >
>> > I.e.  If you have a SBC device with the removable-media bit set,
>> > this is used to tell the initiator "this media can be removed, be
>> > prepared that this might happen".
>> > So when you mount such a SBC device in the guest, the guest will issue
>> > a "PREVENT_ALLOW_MEDIUM_REMOVAL"
>> > to tell the device "this medium is in use and may not be removed".
>> >
>>
>> What I mean is that if /dev/sdb is removable,
>> if you mount this as   "mount /dev/sdb1 /mnt"
>> this will automatically cause the guest kernel to send a
>> PREVENT_ALLOW_MEDIUM_REMOVAL to /dev/sdb to prevent removal.
>>
>> When you "umount /dev/sdb1"   the kernel/guest will automagically send
>> PREVENT_ALLOW_MEDIUM_REMOVEAL to /dev/sdb and allow removal of the
>> media again.
>>
>>
>> If you capture this command and track the "prevent/allow removal
>> status"  you automatically get a channel where qemu will
>> know when it is safe to unplug the device  and when it is not safe to
>> unplug the device.
>> This is a nice feature.
>
> Presumably there's a way for device to notify the OS
> that user requested removal, as well?


I think that is done by responding with sense  to one of the commands,
like the every few second TEST_UNIT_READY that the
initiator/guest-kernel will send.

5Ah 01h    DT WROM BK  OPERATOR MEDIUM REMOVAL REQUEST

This sense code should be the one to use.


I dont know if linux scsi initiator honors this  or what it will do.



I guess something like this could work ?

IF device is marked as prevent-removal THEN
    send OPERATOR SEND MEDIUM REMOVAL REQUEST to the initiator
    wait xyz seconds
    IF device is still marked as prevent-removal THEN
        ask operator "guest refused to release the LUN, do you want to
forcefully remove it?"
    ELSE
        unmount the media
    FI
ELSE
   unmount the media
FI
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/