do_drive_backup() acquires the AioContext lock of the corresponding
BlockDriverState. This is not a problem when it's called from
qmp_drive_backup(), but drive_backup_prepare() also acquires the lock
before calling it. The same things happens with do_blockdev_backup()
and blockdev_backup_prepare().

This patch series merges do_drive_backup() with drive_backup_prepare()
and do_blockdev_backup() with blockdev_backup_prepare(), and ensures
they're only getting called from a transaction context. This way,
there's a single code path for both transaction requests and qmp
commands, as suggested by Kevin Wolf.

We also take this opportunity to ensure we're honoring the context
acquisition semantics required by bdrv_try_set_aio_context, as
suggested by Max Reitz.

---
Changelog:

v4:
 - Unify patches 1-4 and 5-7 to avoid producing broken interim
   states. (thanks Max Reitz)
 - Include a fix for iotest 141. (thanks Kevin Wolf)

v3:
 - Rework the whole patch series to fix the issue by consolidating all
   operations in the transaction model. (thanks Kevin Wolf)

v2:
 - Honor bdrv_try_set_aio_context() context acquisition requirements
   (thanks Max Reitz).
 - Release the context at drive_backup_prepare() instead of avoiding
   re-acquiring it at do_drive_baclup(). (thanks Max Reitz)
 - Convert a single patch into a two-patch series.
---

Sergio Lopez (5):
  blockdev: fix coding style issues in drive_backup_prepare
  blockdev: unify qmp_drive_backup and drive-backup transaction paths
  blockdev: unify qmp_blockdev_backup and blockdev-backup transaction
    paths
  blockdev: honor bdrv_try_set_aio_context() context requirements
  iotests: fix 141 after qmp_drive_backup with transactions

 blockdev.c                 | 349 ++++++++++++++++++-------------------
 tests/qemu-iotests/141.out |   2 +
 2 files changed, 173 insertions(+), 178 deletions(-)

-- 
2.23.0