From 799977d9aafbf0ca0b9c39b04cbfb16db71302c9 Mon Sep 17 00:00:00 2001 From: David Ahern Date: Wed, 2 Mar 2016 11:30:07 -0800 Subject: [PATCH] net: ipv6: Fix refcnt on host routes Andrew and Ying Huang's test robot both reported usage count problems that trace back to the 'keep address on ifdown' patch. >From Andrew: We execute CRIU test on linux-next. On the current linux-next kernel they hangs on creating a network namespace. The kernel log contains many massages like this: [ 1036.122108] unregister_netdevice: waiting for lo to become free. Usage count = 2 [ 1046.165156] unregister_netdevice: waiting for lo to become free. Usage count = 2 [ 1056.210287] unregister_netdevice: waiting for lo to become free. Usage count = 2 I tried to revert this patch and the bug disappeared. Here is a set of commands to reproduce this bug: [root@linux-next-test linux-next]# uname -a Linux linux-next-test 4.5.0-rc6-next-20160301+ #3 SMP Wed Mar 2 17:32:18 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux [root@linux-next-test ~]# unshare -n [root@linux-next-test ~]# ip link set up dev lo [root@linux-next-test ~]# ip a 1: lo: mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever inet6 ::1/128 scope host valid_lft forever preferred_lft forever [root@linux-next-test ~]# logout [root@linux-next-test ~]# unshare -n ----- The problem is a change made to RTM_DELADDR case in __ipv6_ifa_notify that was added in an early version of the offending patch and is no longer needed. Fixes: f1705ec197e7 ("net: ipv6: Make address flushing on ifdown optional") Cc: Andrey Wagin Cc: Ying Huang Signed-off-by: David Ahern Tested-by: Jeremiah Mahler Signed-off-by: David S. Miller --- net/ipv6/addrconf.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c index b1169d511ab4..8c0dab2de5c9 100644 --- a/net/ipv6/addrconf.c +++ b/net/ipv6/addrconf.c @@ -5319,12 +5319,10 @@ static void __ipv6_ifa_notify(int event, struct inet6_ifaddr *ifp) if (rt) ip6_del_rt(rt); } - if (ifp->rt) { - dst_hold(&ifp->rt->dst); + dst_hold(&ifp->rt->dst); + + ip6_del_rt(ifp->rt); - ip6_del_rt(ifp->rt); - ifp->rt = NULL; - } rt_genid_bump_ipv6(net); break; } -- 2.20.1