Message ID | 1456947007-32444-1-git-send-email-dsa@cumulusnetworks.com |
---|---|
State | Accepted, archived |
Delegated to: | David Miller |
Headers | show |
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>
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.
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.
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; }
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(-)