Use GFP_KERNEL instead of GFP_ATOMIC for allocation in open(), which may
sleep.
Signed-off-by: Johan Hovold <johan@kernel.org>