From: Manfred Spraul <manfred@colorfullife.com>

The attached patch splits netlink_unicast into three steps:

- netlink_getsock{bypid,byfilp}: lookup the destination socket.

- netlink_attachskb: perform the nonblock checks, sleep if the socket
  queue is longer than the limit, etc.

- netlink_sendskb: actually send the skb.

jamal looked over it and didn't see a problem with the netlink change.  The
actual use from ipc/mqueue.c is still open (just send back whatever the C
library passed to mq_notify, add an nlmsghdr or perhaps even make it a
specialized netlink protocol), but the attached patch is independant from
the the message queue change.

(acked by davem)