Currently, attaching zoned block devices (i.e. storage devices
compliant to ZAC/ZBC standards) using several virtio methods doesn't
work properly as zoned devices appear as regular block devices at the
guest. This may cause unexpected i/o errors and, potentially, some
data corruption.

To be more precise, attaching a zoned device via virtio-pci-blk,
virtio-scsi-pci/scsi-disk or virtio-scsi-pci/scsi-hd demonstrates the
above behavior. The virtio-scsi-pci/scsi-block method works with a
recent patch. The virtio-scsi-pci/scsi-generic method also appears to
handle zoned devices without problems.

This patchset adds code to check if the backing device that is being
opened is a zoned Host Managed device. If this is the case, the patch
prohibits virtualization of the device for all use cases lacking proper
zoned support.

Host Aware zoned block devices are designed to work as regular block
devices at a guest system that does not support ZBD. Therefeore, this
patchset doesn't prohibit attachment of Host Aware devices.

Host Aware attachments were tested using file_zbc handler of
tcmu-runner daemon. Running QEMU against zoned devices provisioned by
tcmu-runner emulator exposed another problem - upon startup, QEMU
would exit via divide by zero exception in scsi_disk_reset(). The
reason for this is that tcmu-runner zoned devices don't support
READ CAPACITY(10). This is by design, since only READ CAPACITY(16)
support is mandatory for ZBDs and READ CAPACITY(10) is optional.
The last commit of this series fixes this problem.

Considering that there is still a couple of different working ways
to virtualize a ZBD, this patchset provides a reasonable short-term
solution for this problem. What about long term?

It appears to be beneficial to add proper ZBD support to virtio-blk.
In order to support this use case properly, some virtio-blk protocol
changes will be necessary. They are needed to allow the host code to
propagate some ZBD properties that are required for virtio guest
driver to configure the guest block device as ZBD, such as zoned
device model, zone size and the total number of zones. Further, some
support needs to be added for REPORT ZONES command as well as for zone
operations, such as OPEN ZONE, CLOSE ZONE, FINISH ZONE and RESET ZONE.

These additions to the protocol are relatively straightforward, but
they need to be approved by the virtio TC and the whole process may
take some time.

ZBD support for virtio-scsi-pci/scsi-disk and virtio-scsi-pci/scsi-hd
does not seem as necessary. Users will be expected to attach zoned
block devices via virtio-scsi-pci/scsi-block instead.

This patchset contains some Linux-specific code. This code is
necessary to obtain Zoned Block Device model value from Linux sysfs.

Dmitry Fomichev (4):
  block: Add zoned device model property
  raw: Recognize zoned backing devices
  block/ide/scsi: Set BLK_PERM_SUPPORT_ZONED
  raw: Don't open ZBDs if backend can't handle them

Shin'ichiro Kawasaki (1):
  hw/scsi: Check sense key before READ CAPACITY output snoop

 block.c                   | 19 +++++++++
 block/file-posix.c        | 88 +++++++++++++++++++++++++++++++++------
 block/raw-format.c        |  8 ++++
 hw/block/block.c          |  8 +++-
 hw/block/fdc.c            |  4 +-
 hw/block/nvme.c           |  2 +-
 hw/block/virtio-blk.c     |  2 +-
 hw/block/xen-block.c      |  2 +-
 hw/ide/qdev.c             |  2 +-
 hw/scsi/scsi-disk.c       | 13 +++---
 hw/scsi/scsi-generic.c    | 33 +++++++++------
 hw/usb/dev-storage.c      |  2 +-
 include/block/block.h     | 21 +++++++++-
 include/block/block_int.h |  4 ++
 include/hw/block/block.h  |  3 +-
 15 files changed, 169 insertions(+), 42 deletions(-)

-- 
2.21.0


Reply via email to