Fully preallocated truncation has a 50 % chance of working on images
files over file-posix.  It works if $SIZE % 4G < 2G, and it fails
otherwise.  To make things even more interesting, often you would not
even notice because qemu reported success even though it did nothing
(because after the successful lseek(), errno was still 0, so when the
file-posix driver tried to return a negative error code, it actually
reported success).

This issue is fixed by patch 1 in this series.  Thanks to Daniel for
reporting!


Max Reitz (2):
  block/file-posix: Fix fully preallocated truncate
  iotests: Test preallocated truncate of 2G image

 block/file-posix.c         |  5 +++--
 tests/qemu-iotests/106     | 24 ++++++++++++++++++++++++
 tests/qemu-iotests/106.out | 10 ++++++++++
 3 files changed, 37 insertions(+), 2 deletions(-)

-- 
2.14.3

On 02/28/2018 07:13 AM, Max Reitz wrote:
> Fully preallocated truncation has a 50 % chance of working on images
> files over file-posix.  It works if $SIZE % 4G < 2G, and it fails
> otherwise.  To make things even more interesting, often you would not
> even notice because qemu reported success even though it did nothing
> (because after the successful lseek(), errno was still 0, so when the
> file-posix driver tried to return a negative error code, it actually
> reported success).
> 
> This issue is fixed by patch 1 in this series.  Thanks to Daniel for
> reporting!
> 
> 
> Max Reitz (2):
>    block/file-posix: Fix fully preallocated truncate
>    iotests: Test preallocated truncate of 2G image

Good catch.  Series:
Reviewed-by: Eric Blake <eblake@redhat.com>

gluster.c has similar code, but it assigns glfs_lseek() to an int64_t, 
so it does not have the bug.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

On 2018-02-28 14:13, Max Reitz wrote:
> Fully preallocated truncation has a 50 % chance of working on images
> files over file-posix.  It works if $SIZE % 4G < 2G, and it fails
> otherwise.  To make things even more interesting, often you would not
> even notice because qemu reported success even though it did nothing
> (because after the successful lseek(), errno was still 0, so when the
> file-posix driver tried to return a negative error code, it actually
> reported success).
> 
> This issue is fixed by patch 1 in this series.  Thanks to Daniel for
> reporting!
> 
> 
> Max Reitz (2):
>   block/file-posix: Fix fully preallocated truncate
>   iotests: Test preallocated truncate of 2G image
> 
>  block/file-posix.c         |  5 +++--
>  tests/qemu-iotests/106     | 24 ++++++++++++++++++++++++
>  tests/qemu-iotests/106.out | 10 ++++++++++
>  3 files changed, 37 insertions(+), 2 deletions(-)

Applied to my block tree...  (*cough* *cough* -- I'm the worst when it
comes to keeping track of my own patches.)

Max