team: fix possible recursive locking when add slaves
authorHangbin Liu <liuhangbin@gmail.com>
Fri, 19 Apr 2019 06:31:00 +0000 (14:31 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 2 May 2019 07:59:00 +0000 (09:59 +0200)
[ Upstream commit 925b0c841e066b488cc3a60272472b2c56300704 ]

If we add a bond device which is already the master of the team interface,
we will hold the team->lock in team_add_slave() first and then request the
lock in team_set_mac_address() again. The functions are called like:

- team_add_slave()
 - team_port_add()
   - team_port_enter()
     - team_modeop_port_enter()
       - __set_port_dev_addr()
         - dev_set_mac_address()
           - bond_set_mac_address()
             - dev_set_mac_address()
          - team_set_mac_address

Although team_upper_dev_link() would check the upper devices but it is
called too late. Fix it by adding a checking before processing the slave.

v2: Do not split the string in netdev_err()

Fixes: 3d249d4ca7d0 ("net: introduce ethernet teaming device")
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/team/team.c

index e23eaf3..6c6230b 100644 (file)
@@ -1160,6 +1160,13 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
                return -EINVAL;
        }
 
+       if (netdev_has_upper_dev(dev, port_dev)) {
+               NL_SET_ERR_MSG(extack, "Device is already an upper device of the team interface");
+               netdev_err(dev, "Device %s is already an upper device of the team interface\n",
+                          portname);
+               return -EBUSY;
+       }
+
        if (port_dev->features & NETIF_F_VLAN_CHALLENGED &&
            vlan_uses_dev(dev)) {
                NL_SET_ERR_MSG(extack, "Device is VLAN challenged and team device has VLAN set up");