bonding: fix potential deadlock in bond_uninit()

PATCH Superseded Amerigo Wang 3 participants 5 messages Apr 1, 2010 – Apr 2, 2010

Subsystems (3)

Networking Network drivers Bonding driver

lore

Share & tools

lore

Amerigo Wang

author

2010-04-01 06:06 UTC #0 lore raw

PATCH v2 1/1 bonding: fix potential deadlock in bond_uninit()

bond_uninit() is invoked with rtnl_lock held, when it does destroy_workqueue() which will potentially flush all works in this workqueue, if we hold rtnl_lock again in the work function, it will deadlock.

So move destroy_workqueue() to destructor where rtnl_lock is not held any more, suggested by Eric.

Signed-off-by WANG Cong <[email protected]>

Cc Jay Vosburgh <[email protected]>

Cc "David S. Miller" <[email protected]>

Cc Stephen Hemminger <[email protected]>

Cc Jiri Pirko <[email protected]>

Cc "Eric W. Biederman" <[email protected]>

---

drivers/net/bonding/bond_main.c +9 -4

index 5b92fbf..9f0aaa2 100644
@@ -4450,6 +4450,14 @@ static const struct net_device_ops bond_netdev_ops = {
 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
 };
 
+static void bond_destructor(struct net_device *bond_dev)
+{
+	struct bonding *bond = netdev_priv(bond_dev);
+	if (bond->wq)
+		destroy_workqueue(bond->wq);
+	free_netdev(bond_dev);
+}
+
 static void bond_setup(struct net_device *bond_dev)
 {
 	struct bonding *bond = netdev_priv(bond_dev);

@@ -4470,7 +4478,7 @@ static void bond_setup(struct net_device *bond_dev)
 	bond_dev->ethtool_ops = &bond_ethtool_ops;
 	bond_set_mode_ops(bond, bond->params.mode);
 
-	bond_dev->destructor = free_netdev;
+	bond_dev->destructor = bond_destructor;
 
 	/* Initialize the device options */
 	bond_dev->tx_queue_len = 0;

@@ -4542,9 +4550,6 @@ static void bond_uninit(struct net_device *bond_dev)
 
 	bond_remove_proc_entry(bond);
 
-	if (bond->wq)
-		destroy_workqueue(bond->wq);
-
 	netif_addr_lock_bh(bond_dev);
 	bond_mc_list_destroy(bond);
 	netif_addr_unlock_bh(bond_dev);

Eric W. Biederman

2010-04-01 06:12 UTC #1 lore raw

Quoted 6 lines · Amerigo Wang

bond_uninit() is invoked with rtnl_lock held, when it does destroy_workqueue() which will potentially flush all works in this workqueue, if we hold rtnl_lock again in the work function, it will deadlock.
So move destroy_workqueue() to destructor where rtnl_lock is not held any more, suggested by Eric.

The error handling on creating a bond device needs to be updated as well.

Eric

Quoted 47 lines

Signed-off-by WANG Cong <[email protected]>

Cc Jay Vosburgh <[email protected]>

Cc "David S. Miller" <[email protected]>

Cc Stephen Hemminger <[email protected]>

Cc Jiri Pirko <[email protected]>

Cc "Eric W. Biederman" <[email protected]>

---

drivers/net/bonding/bond_main.c +9 -4

index 5b92fbf..9f0aaa2 100644
@@ -4450,6 +4450,14 @@ static const struct net_device_ops bond_netdev_ops = {
 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
 };
 
+static void bond_destructor(struct net_device *bond_dev)
+{
+	struct bonding *bond = netdev_priv(bond_dev);
+	if (bond->wq)
+		destroy_workqueue(bond->wq);
+	free_netdev(bond_dev);
+}
+
 static void bond_setup(struct net_device *bond_dev)
 {
 	struct bonding *bond = netdev_priv(bond_dev);

@@ -4470,7 +4478,7 @@ static void bond_setup(struct net_device *bond_dev)
 	bond_dev->ethtool_ops = &bond_ethtool_ops;
 	bond_set_mode_ops(bond, bond->params.mode);
 
-	bond_dev->destructor = free_netdev;
+	bond_dev->destructor = bond_destructor;
 
 	/* Initialize the device options */
 	bond_dev->tx_queue_len = 0;

@@ -4542,9 +4550,6 @@ static void bond_uninit(struct net_device *bond_dev)
 
 	bond_remove_proc_entry(bond);
 
-	if (bond->wq)
-		destroy_workqueue(bond->wq);
-
 	netif_addr_lock_bh(bond_dev);
 	bond_mc_list_destroy(bond);
 	netif_addr_unlock_bh(bond_dev);

Cong Wang

author

2010-04-01 06:51 UTC #2 lore raw

Quoted 11 lines · Eric W. Biederman

Quoted 6 lines · Amerigo Wang
bond_uninit() is invoked with rtnl_lock held, when it does destroy_workqueue() which will potentially flush all works in this workqueue, if we hold rtnl_lock again in the work function, it will deadlock.
So move destroy_workqueue() to destructor where rtnl_lock is not held any more, suggested by Eric.
The error handling on creating a bond device needs to be updated as well.

You're right, I missed that part. Will update it soon.

Thanks.

Cong Wang

author

2010-04-01 07:27 UTC #3 lore raw

PATCH v3 1/1 bonding: fix potential deadlock in bond_uninit()

In reply to Eric W. Biederman

Quoted 11 lines · Eric W. Biederman

Quoted 6 lines · Amerigo Wang
bond_uninit() is invoked with rtnl_lock held, when it does destroy_workqueue() which will potentially flush all works in this workqueue, if we hold rtnl_lock again in the work function, it will deadlock.
So move destroy_workqueue() to destructor where rtnl_lock is not held any more, suggested by Eric.
The error handling on creating a bond device needs to be updated as well.

Done.

V3: fix error handling path of bond_create()

So move destroy_workqueue() to destructor where rtnl_lock is not held any more, suggested by Eric.

Signed-off-by WANG Cong <[email protected]>

Cc Jay Vosburgh <[email protected]>

Cc "David S. Miller" <[email protected]>

Cc Stephen Hemminger <[email protected]>

Cc Jiri Pirko <[email protected]>

Cc "Eric W. Biederman" <[email protected]>

---

drivers/net/bonding/bond_main.c +14 ~~-12~~

index 5b92fbf..61f8c63 100644
@@ -4450,6 +4450,14 @@ static const struct net_device_ops bond_netdev_ops = {
 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
 };
 
+static void bond_destructor(struct net_device *bond_dev)
+{
+	struct bonding *bond = netdev_priv(bond_dev);
+	if (bond->wq)
+		destroy_workqueue(bond->wq);
+	free_netdev(bond_dev);
+}
+
 static void bond_setup(struct net_device *bond_dev)
 {
 	struct bonding *bond = netdev_priv(bond_dev);

@@ -4470,7 +4478,7 @@ static void bond_setup(struct net_device *bond_dev)
 	bond_dev->ethtool_ops = &bond_ethtool_ops;
 	bond_set_mode_ops(bond, bond->params.mode);
 
-	bond_dev->destructor = free_netdev;
+	bond_dev->destructor = bond_destructor;
 
 	/* Initialize the device options */
 	bond_dev->tx_queue_len = 0;

@@ -4542,9 +4550,6 @@ static void bond_uninit(struct net_device *bond_dev)
 
 	bond_remove_proc_entry(bond);
 
-	if (bond->wq)
-		destroy_workqueue(bond->wq);
-
 	netif_addr_lock_bh(bond_dev);
 	bond_mc_list_destroy(bond);
 	netif_addr_unlock_bh(bond_dev);

@@ -4956,8 +4961,8 @@ int bond_create(struct net *net, const char *name)
 				bond_setup);
 	if (!bond_dev) {
 		pr_err("%s: eek! can't alloc netdev!\n", name);
-		res = -ENOMEM;
-		goto out;
+		rtnl_unlock();
+		return -ENOMEM;
 	}
 
 	dev_net_set(bond_dev, net);

@@ -4966,19 +4971,16 @@ int bond_create(struct net *net, const char *name)
 	if (!name) {
 		res = dev_alloc_name(bond_dev, "bond%d");
 		if (res < 0)
-			goto out_netdev;
+			goto out;
 	}
 
 	res = register_netdevice(bond_dev);
-	if (res < 0)
-		goto out_netdev;
 
 out:
 	rtnl_unlock();
+	if (res < 0)
+		bond_destructor(bond_dev);
 	return res;
-out_netdev:
-	free_netdev(bond_dev);
-	goto out;
 }
 
 static int __net_init bond_net_init(struct net *net)

David Miller

maintainer Applied

2010-04-02 00:26 UTC #4 lore raw

From: Cong Wang <[email protected]>
Date: Thu, 01 Apr 2010 15:30:52 +0800

Quoted 9 lines

Quoted 8 lines · Eric W. Biederman
Quoted 6 lines · Amerigo Wang
bond_uninit() is invoked with rtnl_lock held, when it does destroy_workqueue()
which will potentially flush all works in this workqueue, if we hold rtnl_lock
again in the work function, it will deadlock.

Quoted 8 lines

Quoted 6 lines
Quoted 3 lines
So move destroy_workqueue() to destructor where rtnl_lock is not held any more,
suggested by Eric.
The error handling on creating a bond device needs to be updated as well.
Done.

Applied, thanks.