bonding: emit event when bonding changes MAC
authorWeiping Pan <wpan@redhat.com>
Tue, 27 Mar 2012 19:18:24 +0000 (19:18 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 29 Mar 2012 22:11:57 +0000 (18:11 -0400)
When a bonding device is configured with fail_over_mac=active,
we expect to see the MAC address of the new active slave as the source MAC
address after failover. But we see that the source MAC address is the MAC
address of previous active slave.

Emit NETDEV_CHANGEADDR event when bonding changes its MAC address, in order
to let arp_netdev_event flush neighbour cache and route cache.

How to reproduce this bug ?

                       -----------hostB----------------
hostA ----- switch ---|-- eth0--bond0(192.168.100.2/24)|
(192.168.100.1/24  \--|-- eth1-/                       |
                       --------------------------------

1 on hostB,
modprobe bonding mode=1 miimon=500 fail_over_mac=active downdelay=1000
num_grat_arp=1
ifconfig bond0 192.168.100.2/24 up
ifenslave bond0 eth0
ifenslave bond0 eth1

then eth0 is the active slave, and MAC of bond0 is MAC of eth0.

2 on hostA, ping 192.168.100.2

3 on hostB,
tcpdump -i bond0 -p icmp -XXX
you will see bond0 uses MAC of eth0 as source MAC in icmp reply.

4 on hostB,
ifconfig eth0 down
tcpdump -i bond0 -p icmp -XXX (just keep it running in step 3)
you will see first bond0 uses MAC of eth1 as source MAC in icmp
reply, then it will use MAC of eth0 as source MAC.

Signed-off-by: Weiping Pan <wpan@redhat.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/bonding/bond_main.c

index b920d829692aa35394962f87910678bfe1d4cbb7..a20b5850e7cecd9297bc1437891cab3e7e1694e5 100644 (file)
@@ -892,9 +892,15 @@ static void bond_do_fail_over_mac(struct bonding *bond,
 
        switch (bond->params.fail_over_mac) {
        case BOND_FOM_ACTIVE:
-               if (new_active)
+               if (new_active) {
                        memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
                               new_active->dev->addr_len);
+                       write_unlock_bh(&bond->curr_slave_lock);
+                       read_unlock(&bond->lock);
+                       call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
+                       read_lock(&bond->lock);
+                       write_lock_bh(&bond->curr_slave_lock);
+               }
                break;
        case BOND_FOM_FOLLOW:
                /*