]> Pileus Git - ~andy/linux/commitdiff
bridge: Fix the way to check if a local fdb entry can be deleted
authorToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Fri, 7 Feb 2014 07:48:22 +0000 (16:48 +0900)
committerDavid S. Miller <davem@davemloft.net>
Mon, 10 Feb 2014 22:34:33 +0000 (14:34 -0800)
We should take into account the followings when deleting a local fdb
entry.

- nbp_vlan_find() can be used only when vid != 0 to check if an entry is
  deletable, because a fdb entry with vid 0 can exist at any time while
  nbp_vlan_find() always return false with vid 0.

  Example of problematic case:
    ip link set eth0 address 12:34:56:78:90:ab
    ip link set eth1 address 12:34:56:78:90:ab
    brctl addif br0 eth0
    brctl addif br0 eth1
    ip link set eth0 address aa:bb:cc:dd:ee:ff
  Then, the fdb entry 12:34:56:78:90:ab will be deleted even though the
  bridge port eth1 still has that address.

- The port to which the bridge device is attached might needs a local entry
  if its mac address is set manually.

  Example of problematic case:
    ip link set eth0 address 12:34:56:78:90:ab
    brctl addif br0 eth0
    ip link set br0 address 12:34:56:78:90:ab
    ip link set eth0 address aa:bb:cc:dd:ee:ff
  Then, the fdb still must have the entry 12:34:56:78:90:ab, but it will be
  deleted.

We can use br->dev->addr_assign_type to check if the address is manually
set or not, but I propose another approach.

Since we delete and insert local entries whenever changing mac address
of the bridge device, we can change dst of the entry to NULL regardless of
addr_assign_type when deleting an entry associated with a certain port,
and if it is found to be unnecessary later, then delete it.
That is, if changing mac address of a port, the entry might be changed
to its dst being NULL first, but is eventually deleted when recalculating
and changing bridge id.

This approach is especially useful when we want to share the code with
deleting vlan in which the bridge device might want such an entry regardless
of addr_assign_type, and makes things easy because we don't have to consider
if mac address of the bridge device will be changed or not at the time we
delete a local entry of a port, which means fdb code will not be bothered
even if the bridge id calculating logic is changed in the future.

Also, this change reduces inconsistent state, where frames whose dst is the
mac address of the bridge, can't reach the bridge because of premature fdb
entry deletion. This change reduces the possibility that the bridge device
replies unreachable mac address to arp requests, which could occur during
the short window between calling del_nbp() and br_stp_recalculate_bridge_id()
in br_del_if(). This will effective after br_fdb_delete_by_port() starts to
use the same code by following patch.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Acked-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/bridge/br_fdb.c
net/bridge/br_private.h
net/bridge/br_vlan.c

index 96ab1d1748d00d62a90a4e95ad9b3816fffe0962..b4005f5b28f48e1dd857d092c5bbabaf183239e7 100644 (file)
@@ -114,12 +114,20 @@ void br_fdb_changeaddr(struct net_bridge_port *p, const unsigned char *newaddr)
                                        if (op != p &&
                                            ether_addr_equal(op->dev->dev_addr,
                                                             f->addr.addr) &&
-                                           nbp_vlan_find(op, vid)) {
+                                           (!vid || nbp_vlan_find(op, vid))) {
                                                f->dst = op;
                                                goto skip_delete;
                                        }
                                }
 
+                               /* maybe bridge device has same hw addr? */
+                               if (ether_addr_equal(br->dev->dev_addr,
+                                                    f->addr.addr) &&
+                                   (!vid || br_vlan_find(br, vid))) {
+                                       f->dst = NULL;
+                                       goto skip_delete;
+                               }
+
                                /* delete old one */
                                fdb_delete(br, f);
 skip_delete:
index 939a59e150365e89df9c2c8079b58da083c4629b..f91e1d9c8e92713aec186744a166b6d35a9325f3 100644 (file)
@@ -585,6 +585,7 @@ struct sk_buff *br_handle_vlan(struct net_bridge *br,
 int br_vlan_add(struct net_bridge *br, u16 vid, u16 flags);
 int br_vlan_delete(struct net_bridge *br, u16 vid);
 void br_vlan_flush(struct net_bridge *br);
+bool br_vlan_find(struct net_bridge *br, u16 vid);
 int br_vlan_filter_toggle(struct net_bridge *br, unsigned long val);
 int nbp_vlan_add(struct net_bridge_port *port, u16 vid, u16 flags);
 int nbp_vlan_delete(struct net_bridge_port *port, u16 vid);
@@ -666,6 +667,11 @@ static inline void br_vlan_flush(struct net_bridge *br)
 {
 }
 
+static inline bool br_vlan_find(struct net_bridge *br, u16 vid)
+{
+       return false;
+}
+
 static inline int nbp_vlan_add(struct net_bridge_port *port, u16 vid, u16 flags)
 {
        return -EOPNOTSUPP;
index 4ca4d0a0151c49926dd7e47c6ac5bc8f6b5d28f0..233ec1c6e9db4393c378902d92158accb7715ced 100644 (file)
@@ -295,6 +295,25 @@ void br_vlan_flush(struct net_bridge *br)
        __vlan_flush(pv);
 }
 
+bool br_vlan_find(struct net_bridge *br, u16 vid)
+{
+       struct net_port_vlans *pv;
+       bool found = false;
+
+       rcu_read_lock();
+       pv = rcu_dereference(br->vlan_info);
+
+       if (!pv)
+               goto out;
+
+       if (test_bit(vid, pv->vlan_bitmap))
+               found = true;
+
+out:
+       rcu_read_unlock();
+       return found;
+}
+
 int br_vlan_filter_toggle(struct net_bridge *br, unsigned long val)
 {
        if (!rtnl_trylock())