For ublk servers with many ublk queues, accessing the ublk_queue in
ublk_unmap_io() is a frequent cache miss. Pass to __ublk_complete_rq()
whether the ublk server's data buffer needs to be copied to the request.
In the callers __ublk_fail_req() and ublk_ch_uring_cmd_local(), get the
flags from the ublk_device instead, as its flags have just been read.
In ublk_put_req_ref(), pass false since all the features that require
reference counting disable copying of the data buffer upon completion.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>