Re: [Qemu-devel] [PATCH v3 00/19] block & scsi: write_zeroes support through the whole stack

[Stefan Hajnoczi]

On Fri, Nov 22, 2013 at 01:39:42PM +0100, Paolo Bonzini wrote:
> This series fixes the implementation of WRITE SAME in the SCSI emulation
> layer.  On top of this, it ensures that zero writes from the guest can
> be offloaded to the host device or filesystem if that's supported.
> This is a relatively important part of the thin provisioning feature,
> and builds on the unmap flag to write_zeroes, recently added by Peter
> Lieven.  The feature works with iscsi, block device and filesystem
> targets.
> 
> v2->v3
>         Drop "block/iscsi: use UNMAP to write zeroes if LBPRZ=1" [Peter]
> 
>         Patches 3/4: change tracepoint flags to %#x [Stefan]
> 
>         Patch 6: clean up conditionals for misaligned num [Stefan]
>         Patch 6: fix freeing of bounce buffer [Peter]
> 
>         Patch 7: fix bdi->cluster_size for vhdx [Stefan]
> 
>         Patch 8: do not check bs->backing_hd to set
>         bdi->unallocated_blocks_are_zero [Peter]
> 
>         Patch 19: use qemu_blockalign/qemu_vfree [Stefan]
> 
> v1->v2
>         I cleaned up the series, putting all the block layer changes
>         together and all the SCSI changes at the end.
> 
>         qcow2 and iscsi are now better covered, which caused the
>         introduction of patches 6-11.  I also downloaded the latest
>         which contributed some changes in patches 10-11.  It turns
>         out we were too cautious in treating LBPRZ as advisory when
>         using the UNMAP command.  On the other hand, the API we
>         introduced is useful for qcow2/qed/vmdk WRITE SAME, so no
>         damage done.
> 
> 
> Paolo Bonzini (17):
>   block: generalize BlockLimits handling to cover bdrv_aio_discard too
>   block: add flags to BlockRequest
>   block: add flags argument to bdrv_co_write_zeroes tracepoint
>   block: add bdrv_aio_write_zeroes
>   block: handle ENOTSUP from discard in generic code
>   block: make bdrv_co_do_write_zeroes stricter in producing aligned
>     requests
>   vpc, vhdx: add get_info
>   block drivers: add discard/write_zeroes properties to bdrv_get_info
>     implementation
>   block drivers: expose requirement for write same alignment from
>     formats
>   block/iscsi: check WRITE SAME support differently depending on
>     MAY_UNMAP
>   raw-posix: implement write_zeroes with MAY_UNMAP for files
>   raw-posix: implement write_zeroes with MAY_UNMAP for block devices
>   raw-posix: add support for write_zeroes on XFS and block devices
>   qemu-iotests: 033 is fast
>   scsi-disk: catch write protection errors in UNMAP
>   scsi-disk: reject ANCHOR=1 for UNMAP and WRITE SAME commands
>   scsi-disk: correctly implement WRITE SAME
> 
> Peter Lieven (2):
>   block/iscsi: remove .bdrv_has_zero_init
>   block/iscsi: updated copyright
> 
>  block.c                  | 145 +++++++++++++++++++++++----------------
>  block/iscsi.c            |  27 +++++---
>  block/qcow2.c            |   3 +
>  block/qed.c              |   3 +
>  block/raw-aio.h          |   3 +-
>  block/raw-posix.c        | 175 
> ++++++++++++++++++++++++++++++++++++++++++++---
>  block/vdi.c              |   1 +
>  block/vhdx.c             |  13 ++++
>  block/vmdk.c             |   4 ++
>  block/vpc.c              |  15 ++++
>  hw/scsi/scsi-disk.c      | 154 ++++++++++++++++++++++++++++++++++-------
>  include/block/block.h    |   4 ++
>  tests/qemu-iotests/group |   2 +-
>  trace-events             |   4 +-
>  14 files changed, 452 insertions(+), 101 deletions(-)

Some of it is SCSI but I've applied the whole series.

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan