On 4/17/19 5:09 AM, Vladimir Sementsov-Ogievskiy wrote:
> Hi all. We faced an interesting bug, which may be simply reproduced:
>
> prepare image:
> ./qemu-img create -f qcow2 -o cluster_size=1M /ssd/test 2300M
> ./qemu-io -c 'write 100M 2000M' -c 'write 2100M 200M' -c 'write 0 100M' /ssd/test
>
> shrink:
> ./qemu-img resize --shrink /ssd/test 50M
>
> bug:
> ./qemu-img info /ssd/test
> image: /ssd/test
> file format: qcow2
> virtual size: 50M (52428800 bytes)
> disk size: 2.2G
> cluster_size: 1048576
> Format specific information:
> compat: 1.1
> lazy refcounts: false
> refcount bits: 16
> corrupt: false
>
> Virtual size is shrunk, but file - not. It is due to the fact,
> that merged qcow2 discard may exceed 2G, and then converting from
> uint64_t to int in qcow2_process_discards when we call bdrv_pdiscard
> make wrong thing.
Too late for 4.0, but also not a regression new to this release, since
the problem appears to have been present since its introduction in
commit 0b919fae (1.6.0) (that is, even back then, Qcow2DiscardRegion was
introduced with a 64-bit discard length, but qcow2_process_discards
blindly passed that through bdrv_discard() at the time, which took 'int
nb_sectors').
>
> So, here are proposal of fix and new iotest for it.
>
> Vladimir Sementsov-Ogievskiy (2):
> block/io: bdrv_pdiscard: support int64_t bytes parameter
> iotests: test big qcow2 shrink
>
> include/block/block.h | 4 +--
> block/io.c | 19 ++++++-----
> tests/qemu-iotests/249 | 69 ++++++++++++++++++++++++++++++++++++++
> tests/qemu-iotests/249.out | 30 +++++++++++++++++
> tests/qemu-iotests/group | 1 +
> 5 files changed, 112 insertions(+), 11 deletions(-)
> create mode 100755 tests/qemu-iotests/249
> create mode 100644 tests/qemu-iotests/249.out
>
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3226
Virtualization: qemu.org | libvirt.org