[PATCH net-next] net: ipv6: Fix refcnt on host routes

[PATCH net-next] net: ipv6: Fix refcnt on host routes

[David Ahern]

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: <LOOPBACK,UP,LOWER_UP> 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 <avagin@gmail.com>
Cc: Ying Huang <ying.huang@linux.intel.com>
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
---
 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;
 	}
-- 
2.1.4

Re: [PATCH net-next] net: ipv6: Fix refcnt on host routes

[Jeremiah Mahler]

Hi David,

On Wed, Mar 02, 2016 at 11:30:07AM -0800, David Ahern wrote:
> 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: <LOOPBACK,UP,LOWER_UP> 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 <avagin@gmail.com>
> Cc: Ying Huang <ying.huang@linux.intel.com>
> Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
> ---
>  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;
>  	}
> -- 
> 2.1.4
> 

Tested-by: Jeremiah Mahler <jmmahler@gmail.com>

-- 
- Jeremiah Mahler

Re: [PATCH net-next] net: ipv6: Fix refcnt on host routes

[Daniel Borkmann]

On 03/02/2016 08:30 PM, David Ahern wrote:
> Andrew and Ying Huang's test robot both reported usage count problems that
> trace back to the 'keep address on ifdown' patch.
>
[...]
>
> Fixes: f1705ec197e7 ("net: ipv6: Make address flushing on ifdown optional")
> Cc: Andrey Wagin <avagin@gmail.com>
> Cc: Ying Huang <ying.huang@linux.intel.com>
> Signed-off-by: David Ahern <dsa@cumulusnetworks.com>

Fixes the issue as well over here.

Re: [PATCH net-next] net: ipv6: Fix refcnt on host routes

[David Miller]

From: David Ahern <dsa@cumulusnetworks.com>
Date: Wed,  2 Mar 2016 11:30:07 -0800

> 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: <LOOPBACK,UP,LOWER_UP> 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 <avagin@gmail.com>
> Cc: Ying Huang <ying.huang@linux.intel.com>
> Signed-off-by: David Ahern <dsa@cumulusnetworks.com>

Applied, thanks David.