On Wed, Nov 30, 2016 at 1:16 AM, Richard Guy Briggs [off-list ref] wrote:

On 2016-11-29 09:14, Cong Wang wrote:

quoted

netlink kernel socket is protected by refcount, not RCU.
Its rcv path is neither protected by RCU. So the synchronize_net()
is just pointless.

If I understand correctly, xfrm_user_net_exit() usage of
RCU_INIT_POINTER() and synchronize_net() is similarly pointless?  Also
net/phonet/socket.c?  I probably modelled things based on the former...

Possibly, but xfrm case is slightly different, it has two copies of the pointer
to the netlink socket, also it uses exit_batch(). I need to double check.

Take a look at better examples, fib_front, genetlink, rtnetlink.