In swap_volume method of nova/virt/libvirt/driver.py,
before BDM was got by using the instance's UUID and
'serial' of new connection_info as the volume ID,
and driver BDM was updated by using the BDM.
('serial' has the volume ID information.)
But in _init_volume_connection method in ComputeManager class,
'serial' is passed from old connection_info to new connection_info.
It works fine in the case that cinder initiates swapping volumes
because the ID of the attached volume isn't changed after
swapping volumes.
But in the case that nova initiates swapping volumes,
the ID of the attached volume is changed.
So in the case that nova initiated swapping volumes,
after swap volume function was performed once,
BDM was got by wrong old volume id (serial)
when swap volume function was performed for the second time.
So if 'serial' of new connection_info is None,
it is set to new volume ID.
And if cinder 'migrate_volume_completion' API returns
old volume ID (the case that cinder initiates swapping volumes),
the 'serial' of new connection_info is set to old volume ID.
If cinder 'migrate_volume_completion' API returns new volume ID
(the case that nova initiated swapping volumes),
the 'serial' is left as it is (new volume ID).
Change-Id: I86b8fbb09b0f1ed4c667683de3827cd9b63bca7f
Closes-Bug: #1490236
be553fb155