Hi,

As suggested by Paolo, this series drops xfsctl() calls where we have
working fallocate() alternatives.  (And thus replaces “block/file-posix:
Fix xfs_write_zeroes()”.)

Unfortunately, we also use xfsctl() to inquire the request alignment for
O_DIRECT, and this is the only way we currently have to obtain it
without trying.  Therefore, I didn’t quite like removing that call, too,
so this series doesn’t get rid of xfsctl() completely.

(If we did, we could delete 146 lines instead of these measly 76 here.)


Anyway, dropping xfs_write_zeroes() will also fix the guest corruptions
Lukáš has reported (for qcow2, but I think it should be possible to see
similar corruptions with raw, although I haven’t investigated that too
far).


Max Reitz (2):
  block/file-posix: Reduce xfsctl() use
  iotests: Test reverse sub-cluster qcow2 writes

 block/file-posix.c         | 77 +-------------------------------------
 tests/qemu-iotests/265     | 67 +++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 +++
 tests/qemu-iotests/group   |  1 +
 4 files changed, 75 insertions(+), 76 deletions(-)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out

-- 
2.21.0

Am 23.08.2019 um 15:03 hat Max Reitz geschrieben:
> Hi,
> 
> As suggested by Paolo, this series drops xfsctl() calls where we have
> working fallocate() alternatives.  (And thus replaces “block/file-posix:
> Fix xfs_write_zeroes()”.)
> 
> Unfortunately, we also use xfsctl() to inquire the request alignment for
> O_DIRECT, and this is the only way we currently have to obtain it
> without trying.  Therefore, I didn’t quite like removing that call, too,
> so this series doesn’t get rid of xfsctl() completely.
> 
> (If we did, we could delete 146 lines instead of these measly 76 here.)
> 
> 
> Anyway, dropping xfs_write_zeroes() will also fix the guest corruptions
> Lukáš has reported (for qcow2, but I think it should be possible to see
> similar corruptions with raw, although I haven’t investigated that too
> far).

Thanks, applied to the block branch.

Kevin

On Fri, Aug 23, 2019 at 03:03:39PM +0200, Max Reitz wrote:
> Hi,
> 
> As suggested by Paolo, this series drops xfsctl() calls where we have
> working fallocate() alternatives.  (And thus replaces “block/file-posix:
> Fix xfs_write_zeroes()”.)
> 
> Unfortunately, we also use xfsctl() to inquire the request alignment for
> O_DIRECT, and this is the only way we currently have to obtain it
> without trying.  Therefore, I didn’t quite like removing that call, too,
> so this series doesn’t get rid of xfsctl() completely.
> 
> (If we did, we could delete 146 lines instead of these measly 76 here.)
> 
> 
> Anyway, dropping xfs_write_zeroes() will also fix the guest corruptions
> Lukáš has reported (for qcow2, but I think it should be possible to see
> similar corruptions with raw, although I haven’t investigated that too
> far).
> 
> 
> Max Reitz (2):
>   block/file-posix: Reduce xfsctl() use
>   iotests: Test reverse sub-cluster qcow2 writes
> 
>  block/file-posix.c         | 77 +-------------------------------------
>  tests/qemu-iotests/265     | 67 +++++++++++++++++++++++++++++++++
>  tests/qemu-iotests/265.out |  6 +++
>  tests/qemu-iotests/group   |  1 +
>  4 files changed, 75 insertions(+), 76 deletions(-)
>  create mode 100755 tests/qemu-iotests/265
>  create mode 100644 tests/qemu-iotests/265.out

The patch and the test LGTM.

I tried to run the 265 test without the
"block/file-posix: Reduce xfsctl() use" patch and the failure rate is ~30% on
my system.

With the patch applied the failure rate is 0% :-)

Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Tested-by: Stefano Garzarella <sgarzare@redhat.com>

Thanks,
Stefano