bonding: change test for presence of VLANs
authorJay Vosburgh <fubar@us.ibm.com>
Wed, 21 Jul 2010 12:14:47 +0000 (12:14 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 22 Jul 2010 21:14:46 +0000 (14:14 -0700)
After commit ad1afb00393915a51c21b1ae8704562bf036855f
("vlan_dev: VLAN 0 should be treated as "no vlan tag" (802.1p packet)")
it is now regular practice for a VLAN "add vid" for VLAN 0 to
arrive prior to any VLAN registration or creation of a vlan_group.

This patch updates the bonding code that tests for the presence
of VLANs configured above bonding.  The new logic tests for bond->vlgrp
to determine if a registration has occured, instead of testing that
bonding's internal vlan_list is empty.

The old code would panic when vlan_list was not empty, but
vlgrp was still NULL (because only an "add vid" for VLAN 0 had occured).

Bonding still adds VLAN 0 to its internal list so that 802.1p
frames are handled correctly on transmit when non-VLAN accelerated
slaves are members of the bond.  The test against bond->vlan_list
remains in bond_dev_queue_xmit for this reason.

Modification to the bond->vlgrp now occurs under lock (in
addition to RTNL), because not all inspections of it occur under RTNL.

Additionally, because 8021q will never issue a "kill vid" for
VLAN 0, there is now logic in bond_uninit to release any remaining
entries from vlan_list.

Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Cc: Pedro Garcia <pedro.netdev@dondevamos.com>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/bonding/bond_alb.c
drivers/net/bonding/bond_main.c

index 3662d6e446a928ef544c5a45e12313c336fb2cd6..e3b35d0b4284d2c1aa7da717262dc7d082b1b09a 100644 (file)
@@ -682,7 +682,7 @@ static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bon
                        client_info->ntt = 0;
                }
 
-               if (!list_empty(&bond->vlan_list)) {
+               if (bond->vlgrp) {
                        if (!vlan_get_tag(skb, &client_info->vlan_id))
                                client_info->tag = 1;
                }
@@ -904,7 +904,7 @@ static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
                skb->priority = TC_PRIO_CONTROL;
                skb->dev = slave->dev;
 
-               if (!list_empty(&bond->vlan_list)) {
+               if (bond->vlgrp) {
                        struct vlan_entry *vlan;
 
                        vlan = bond_next_vlan(bond,
index 20f45cbf961a70c87c3bb2a5ec9f2f1ab27228d8..f3b01ce4f629d416eaca43a8e1abafcf8c592be7 100644 (file)
@@ -424,6 +424,7 @@ int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 {
        unsigned short uninitialized_var(vlan_id);
 
+       /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
        if (!list_empty(&bond->vlan_list) &&
            !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
            vlan_get_tag(skb, &vlan_id) == 0) {
@@ -487,7 +488,9 @@ static void bond_vlan_rx_register(struct net_device *bond_dev,
        struct slave *slave;
        int i;
 
+       write_lock(&bond->lock);
        bond->vlgrp = grp;
+       write_unlock(&bond->lock);
 
        bond_for_each_slave(bond, slave, i) {
                struct net_device *slave_dev = slave->dev;
@@ -569,7 +572,7 @@ static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *sla
 
        write_lock_bh(&bond->lock);
 
-       if (list_empty(&bond->vlan_list))
+       if (!bond->vlgrp)
                goto out;
 
        if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
@@ -596,7 +599,7 @@ static void bond_del_vlans_from_slave(struct bonding *bond,
 
        write_lock_bh(&bond->lock);
 
-       if (list_empty(&bond->vlan_list))
+       if (!bond->vlgrp)
                goto out;
 
        if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
@@ -604,6 +607,8 @@ static void bond_del_vlans_from_slave(struct bonding *bond,
                goto unreg;
 
        list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
+               if (!vlan->vlan_id)
+                       continue;
                /* Save and then restore vlan_dev in the grp array,
                 * since the slave's driver might clear it.
                 */
@@ -1443,7 +1448,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
        /* no need to lock since we're protected by rtnl_lock */
        if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
                pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
-               if (!list_empty(&bond->vlan_list)) {
+               if (bond->vlgrp) {
                        pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
                               bond_dev->name, slave_dev->name, bond_dev->name);
                        return -EPERM;
@@ -1942,7 +1947,7 @@ int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
                 */
                memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
 
-               if (list_empty(&bond->vlan_list)) {
+               if (!bond->vlgrp) {
                        bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
                } else {
                        pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
@@ -2134,9 +2139,9 @@ static int bond_release_all(struct net_device *bond_dev)
         */
        memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
 
-       if (list_empty(&bond->vlan_list))
+       if (!bond->vlgrp) {
                bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
-       else {
+       else {
                pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
                           bond_dev->name, bond_dev->name);
                pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
@@ -2569,7 +2574,7 @@ static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
                if (!targets[i])
                        break;
                pr_debug("basa: target %x\n", targets[i]);
-               if (list_empty(&bond->vlan_list)) {
+               if (!bond->vlgrp) {
                        pr_debug("basa: empty vlan: arp_send\n");
                        bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
                                      bond->master_ip, 0);
@@ -2658,6 +2663,9 @@ static void bond_send_gratuitous_arp(struct bonding *bond)
                                bond->master_ip, 0);
        }
 
+       if (!bond->vlgrp)
+               return;
+
        list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
                vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
                if (vlan->vlan_ip) {
@@ -3590,6 +3598,8 @@ static int bond_inetaddr_event(struct notifier_block *this, unsigned long event,
                }
 
                list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
+                       if (!bond->vlgrp)
+                               continue;
                        vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
                        if (vlan_dev == event_dev) {
                                switch (event) {
@@ -4686,6 +4696,7 @@ static void bond_work_cancel_all(struct bonding *bond)
 static void bond_uninit(struct net_device *bond_dev)
 {
        struct bonding *bond = netdev_priv(bond_dev);
+       struct vlan_entry *vlan, *tmp;
 
        bond_netpoll_cleanup(bond_dev);
 
@@ -4699,6 +4710,11 @@ static void bond_uninit(struct net_device *bond_dev)
        bond_remove_proc_entry(bond);
 
        __hw_addr_flush(&bond->mc_list);
+
+       list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
+               list_del(&vlan->vlan_list);
+               kfree(vlan);
+       }
 }
 
 /*------------------------- Module initialization ---------------------------*/