[Qemu-devel] [RFC 00/24] inplace image conversion

[Devin Nakamura]

This series sets out the foundations of an api for in place image conversion, 
and implements it 
in the QED and QCOW2 drivers.

The basic flow of conversion is as follows:
The source image is openened as normal.

The source file is the queried for conversion options via 
bdrv_get_conversion_options. The conversion
options contain things like the size of the clusters, the type of encryption 
etc.

The target image is then opened with bdrv_open_conversion_target, passing the 
conversion optionons
among other things allong with it. This is basically a hybrid of bdrv_open and 
bdrv_create.

Mappings are then fetched and set using bdrv_get_mapping and bdrv_map 
respectively. They both deal 
with data in the format guest_offset, host_offset, and contiguous_bytes. For 
example if 
guest_offset = 31, host_offset = 41, and contiguous_bytes = 59 that would mean 
that offset 31 in the
virtual disk would be located at offset 41 in the image file and 32 at 42 and 
so on up to 89 at 99.

At the end of the mapping process the image is "finalized" with 
bdrv_copy_header, where the target
driver copies out the old header to a temporary file, and then writes out its 
own header into the 
image.

I've integerated inplace conversion into qemu-img in the form: 
qemu-img convert-inplace <imagefile> <target_format>

I've also integrated bdrv_get_mapping and bdrv_map into qemu-io (with the 
former replacing qemu-io's
internal mapping when then block driver supports it) 

Devin Nakamura (24):
  block: add block conversion api
  block: add bdrv_get_conversion_options()
  block: add bdrv_open_conversion_target()
  block: add bdrv_get_mapping()
  block: add bdrv_map()
  block: add bdrv_copy_header()
  qed: make qed_alloc_clusters round up offset to nearest cluster
  qed: add qed_find_cluster_sync()
  qed: add qed_bdrv_get_mapping()
  qed: add qed_bdrv_map()
  qed: add open_conversion_target()
  qed: add bdrv_qed_copy_header()
  qed: add bdrv_qed_get_conversion_options()
  qcow2: fix typo in documentation for qcow2_get_cluster_offset()
  qcow2: split up the creation of new refcount table from the act of
    checking it
  qcow2: add qcow2_drop_leaked_clusters()
  qcow2: add qcow2_get_mapping
  qcow2: add qcow2_map
  qcow2: add qcow2_copy_header()
  qcow2: add get_conversion_options()
  qcow2: add qcow2_open_conversion_target()
  qemu-io: make map command use new block mapping function
  qemu-io: add setmap command
  qemu-img: add inplace conversion to qemu-img

 Makefile               |    2 +
 block.c                |  100 +++++++++++++++++++++
 block.h                |   14 +++
 block/qcow2-cluster.c  |   53 +++++++++++-
 block/qcow2-refcount.c |   71 +++++++++++++--
 block/qcow2.c          |  233 ++++++++++++++++++++++++++++++++++++++++++++++++
 block/qcow2.h          |    5 +
 block/qed-cluster.c    |   35 +++++++
 block/qed.c            |  193 ++++++++++++++++++++++++++++++++++++---
 block/qed.h            |    4 +
 block_int.h            |   88 ++++++++++++++++++
 qemu-img-cmds.hx       |    6 ++
 qemu-img.c             |   64 +++++++++++++
 qemu-io.c              |   47 ++++++++++-
 14 files changed, 888 insertions(+), 27 deletions(-)

-- 
1.7.6.rc1