Re: [Qemu-block] [Qemu-devel] [PATCH for-2.8] qcow2: Don't strand clusters near 2G intervals during commit

[Eric Blake]

On 12/03/2016 11:34 AM, Eric Blake wrote:
> The qcow2_make_empty() function is reached during 'qemu-img commit',
> in order to clear out ALL clusters of an image.  However, if the
> image cannot use the fast code path (true if the image is format
> 0.10, or if the image contains a snapshot), the cluster size is
> larger than 512, and the image is larger than 2G in size, then our
> choice of sector_step causes problems.  Since it is not cluster
> aligned, but qcow2_discard_clusters() silently ignores an unaligned
> head or tail, we are leaving clusters allocated.
> 
> Enhance the testsuite to expose the flaw, and patch the problem by
> ensuring our step size is aligned.
> 
> [qcow2_discard_clusters() is a GROSS interface: it takes a mix of
> byte offset and sector count to perform cluster operations. But
> fixing it to use a saner byte/byte rather than byte/sector interface,
> and/or asserting that the counts are now aligned thanks to both
> this patch and commit 3482b9b, is material for another day.]
> 
> Signed-off-by: Eric Blake <address@hidden>
> ---
>  block/qcow2.c              |   3 +-
>  tests/qemu-iotests/097     |  41 +++++---
>  tests/qemu-iotests/097.out | 249 
> +++++++++++++++++++++++++++++++++------------
>  3 files changed, 210 insertions(+), 83 deletions(-)

> +++ b/block/qcow2.c
> @@ -2808,7 +2808,8 @@ static int qcow2_make_empty(BlockDriverState *bs)
>  {
>      BDRVQcow2State *s = bs->opaque;
>      uint64_t start_sector;
> -    int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
> +    int sector_step = QEMU_ALIGN_DOWN(INT_MAX / BDRV_SECTOR_SIZE,
> +                                      s->cluster_size);

Oh shoot. I got the units wrong, and made the slow path do more loop
iterations than necessary (rounding sectors to cluster size in bytes is
inappropriate - either the rounding has to occur before division, or the
rounding needs to be by secotrs instead of bytes).  I'll send a v2 that
gets the math right.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature