Comment 4 for bug 1841363

Reviewed: https://review.opendev.org/682772
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=af2405e1181d70cdf60bcd0e40b3e80f2db2e3a6
Submitter: Zuul
Branch: master

commit af2405e1181d70cdf60bcd0e40b3e80f2db2e3a6
Author: Arthur Dayne <email address hidden>
Date: Tue Sep 17 19:08:59 2019 +0800

    libvirt:driver:Disallow AIO=native when 'O_DIRECT' is not available

    Because of the libvirt issue[1], there is a bug[2] that if we set cache mode
    whose write semantic is not O_DIRECT (.i.e unsafe, writeback or writethrough),
    there will be a problem with the volume drivers
    (.i.e nova.virt.libvirt.volume.LibvirtISCSIVolumeDriver,
    nova.virt.libvirt.volume.LibvirtNFSVolumeDriver and so on), which designate
    native io explicitly.

    That problem will generate a libvirt xml for the instance,
    whose content contains

    ```
    ...
    <disk ... >
      <driver ... cache='unsafe/writeback/writethrough' io='native' />
    </disk>
    ...
    ```
    In turn, it will fail to start the instance or attach the disk.

    > When qemu is configured with a block device that has aio=native set, but
    > the cache mode doesn't use O_DIRECT (i.e. isn't cache=none/directsync or any
    > unnamed mode with explicit cache.direct=on), then the raw-posix block driver
    > for local files and block devices will silently fall back to aio=threads.
    > The blockdev-add interface rejects such combinations, but qemu can't
    > change the existing legacy interfaces that libvirt uses today.

    [1]: https://github.com/libvirt/libvirt/commit/058384003db776c580d0e5a3016a6384e8eb7b92
    [2]: https://bugzilla.redhat.com/show_bug.cgi?id=1086704

    Closes-Bug: #1841363
    Change-Id: If9acc054100a6733f3659a15dd9fc2d462e84d64