Re: [Qemu-devel] [Qemu-stable] [PATCH v7 0/8] block: Mirror discarded sectors

[Fam Zheng <famz@redhat.com>]

On Thu, 06/11 16:29, Fam Zheng wrote:
> On Mon, 06/08 14:02, Stefan Hajnoczi wrote:
> > On Mon, Jun 08, 2015 at 01:56:06PM +0800, Fam Zheng wrote:
> > > v7: Fix the lost assignment of s->unmap.
> > > 
> > > v6: Fix pnum in bdrv_get_block_status_above. [Paolo]
> > > 
> > > v5: Rewrite patch 1.
> > >     Address Eric's comments on patch 3.
> > >     Add Eric's rev-by to patches 2 & 4.
> > >     Check BDRV_BLOCK_DATA in patch 3. [Paolo]
> > > 
> > > This fixes the mirror assert failure reported by wangxiaolong:
> > > 
> > > https://lists.gnu.org/archive/html/qemu-devel/2015-04/msg04458.html
> > > 
> > > The direct cause is that hbitmap code couldn't handle unset of bits *after*
> > > iterator's current position. We could fix that, but the bdrv_reset_dirty() call
> > > is more questionable:
> > > 
> > > Before, if guest discarded some sectors during migration, it could see
> > > different data after moving to dest side, depending on block backends of the
> > > src and the dest. This is IMO worse than mirroring the actual reading as done
> > > in this series, because we don't know what the guest is doing.
> > > 
> > > For example if a guest first issues WRITE SAME to wipe out the area then issues
> > > UNMAP to discard it, just to get rid of some sensitive data completely, we may
> > > miss both operations and leave stale data on dest image.
> > > 
> > > 
> > > Fam Zheng (8):
> > >   block: Add bdrv_get_block_status_above
> > >   qmp: Add optional bool "unmap" to drive-mirror
> > >   mirror: Do zero write on target if sectors not allocated
> > >   block: Fix dirty bitmap in bdrv_co_discard
> > >   block: Remove bdrv_reset_dirty
> > >   qemu-iotests: Make block job methods common
> > >   qemu-iotests: Add test case for mirror with unmap
> > >   iotests: Use event_wait in wait_ready
> > > 
> > >  block.c                       | 12 --------
> > >  block/io.c                    | 60 ++++++++++++++++++++++++++++++---------
> > >  block/mirror.c                | 28 +++++++++++++++---
> > >  blockdev.c                    |  5 ++++
> > >  hmp.c                         |  2 +-
> > >  include/block/block.h         |  4 +++
> > >  include/block/block_int.h     |  4 +--
> > >  qapi/block-core.json          |  8 +++++-
> > >  qmp-commands.hx               |  3 ++
> > >  tests/qemu-iotests/041        | 66 ++++++++++---------------------------------
> > >  tests/qemu-iotests/132        | 59 ++++++++++++++++++++++++++++++++++++++
> > >  tests/qemu-iotests/132.out    |  5 ++++
> > >  tests/qemu-iotests/group      |  1 +
> > >  tests/qemu-iotests/iotests.py | 23 +++++++++++++++
> > >  14 files changed, 196 insertions(+), 84 deletions(-)
> > >  create mode 100644 tests/qemu-iotests/132
> > >  create mode 100644 tests/qemu-iotests/132.out
> > > 
> > > -- 
> > > 2.4.2
> > > 
> > 
> > Thanks, applied to my block tree:
> > https://github.com/stefanha/qemu/commits/block
> 
> Stefan,
> 
> The only controversial patches are the qmp/drive-mirror ones (1-3), while
> patches 4-8 are still useful on their own: they fix the mentioned crash and
> improve iotests.
> 
> Shall we merge the second half (of course none of them depend on 1-3) now that
> softfreeze is approaching?

Stefan, would you consider applying patches 4-8?

Fam