ipv6: reorder ip6_route_dev_notifier after ipv6_dev_notf
authorWANG Cong <xiyou.wangcong@gmail.com>
Mon, 8 May 2017 17:12:13 +0000 (10:12 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 14 May 2017 12:00:21 +0000 (14:00 +0200)
commit8795ee7bed243c291d30830efe925966253aa2d6
treede221ddf359bdb88394cfa56b47a386515c49802
parentecbd3ed2ddfe5fbae4db6aa3a0903af479e0543d
ipv6: reorder ip6_route_dev_notifier after ipv6_dev_notf

[ Upstream commit 242d3a49a2a1a71d8eb9f953db1bcaa9d698ce00 ]

For each netns (except init_net), we initialize its null entry
in 3 places:

1) The template itself, as we use kmemdup()
2) Code around dst_init_metrics() in ip6_route_net_init()
3) ip6_route_dev_notify(), which is supposed to initialize it after
   loopback registers

Unfortunately the last one still happens in a wrong order because
we expect to initialize net->ipv6.ip6_null_entry->rt6i_idev to
net->loopback_dev's idev, thus we have to do that after we add
idev to loopback. However, this notifier has priority == 0 same as
ipv6_dev_notf, and ipv6_dev_notf is registered after
ip6_route_dev_notifier so it is called actually after
ip6_route_dev_notifier. This is similar to commit 2f460933f58e
("ipv6: initialize route null entry in addrconf_init()") which
fixes init_net.

Fix it by picking a smaller priority for ip6_route_dev_notifier.
Also, we have to release the refcnt accordingly when unregistering
loopback_dev because device exit functions are called before subsys
exit functions.

Acked-by: David Ahern <dsahern@gmail.com>
Tested-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/net/addrconf.h
net/ipv6/addrconf.c
net/ipv6/route.c