]> git.hungrycats.org Git - linux/commitdiff
rtnl/bond: don't send rtnl msg for unregistered iface
authorNicolas Dichtel <nicolas.dichtel@6wind.com>
Wed, 13 May 2015 12:19:42 +0000 (14:19 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 23 Jun 2015 00:03:20 +0000 (17:03 -0700)
[ Upstream commit ed2a80ab7b76f11af0b2c6255709c4ebf164b667 ]

Before the patch, the command 'ip link add bond2 type bond mode 802.3ad'
causes the kernel to send a rtnl message for the bond2 interface, with an
ifindex 0.

'ip monitor' shows:
0: bond2: <BROADCAST,MULTICAST,MASTER> mtu 1500 state DOWN group default
    link/ether 00:00:00:00:00:00 brd ff:ff:ff:ff:ff:ff
9: bond2@NONE: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN group default
    link/ether ea:3e:1f:53:92:7b brd ff:ff:ff:ff:ff:ff
[snip]

The patch fixes the spotted bug by checking in bond driver if the interface
is registered before calling the notifier chain.
It also adds a check in rtmsg_ifinfo() to prevent this kind of bug in the
future.

Fixes: d4261e565000 ("bonding: create netlink event when bonding option is changed")
CC: Jiri Pirko <jiri@resnulli.us>
Reported-by: Julien Meunier <julien.meunier@6wind.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/bonding/bond_options.c
net/core/rtnetlink.c

index 4df28943d2229035166d2bb4a72ec11c8f9671c5..e8d3c1d35453d1e182cd4b82388978f580abe946 100644 (file)
@@ -624,7 +624,7 @@ int __bond_opt_set(struct bonding *bond,
 out:
        if (ret)
                bond_opt_error_interpret(bond, opt, ret, val);
-       else
+       else if (bond->dev->reg_state == NETREG_REGISTERED)
                call_netdevice_notifiers(NETDEV_CHANGEINFODATA, bond->dev);
 
        return ret;
index 7ebed55b5f7d1b2d1faacea1a49e7f9e947f43d1..a2b90e1fc115172e15865fe82478288f09f2fe0e 100644 (file)
@@ -2337,6 +2337,9 @@ void rtmsg_ifinfo(int type, struct net_device *dev, unsigned int change,
 {
        struct sk_buff *skb;
 
+       if (dev->reg_state != NETREG_REGISTERED)
+               return;
+
        skb = rtmsg_ifinfo_build_skb(type, dev, change, flags);
        if (skb)
                rtmsg_ifinfo_send(skb, dev, flags);