openvswitch: Simplify datapath locking.

Currently OVS uses combination of genl and rtnl lock to protect
datapath state.  This was done due to networking stack locking.
But this has complicated locking and there are few lock ordering
issues with new tunneling protocols.
Following patch simplifies locking by introducing new ovs mutex
and now this lock is used to protect entire ovs state.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>

net/openvswitch/datapath.h

@@ -57,9 +57,9 @@ struct dp_stats_percpu {
  * struct datapath - datapath for flow-based packet switching
  * @rcu: RCU callback head for deferred destruction.
  * @list_node: Element in global 'dps' list.
- * @table: Current flow table.  Protected by genl_lock and RCU.
+ * @table: Current flow table.  Protected by ovs_mutex and RCU.
  * @ports: Hash table for ports.  %OVSP_LOCAL port always exists.  Protected by
- * RTNL and RCU.
+ * ovs_mutex and RCU.
  * @stats_percpu: Per-CPU datapath statistics.
  * @net: Reference to net namespace.
  *
@@ -85,26 +85,6 @@ struct datapath {
 #endif
 };
 
-struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no);
-
-static inline struct vport *ovs_vport_rcu(const struct datapath *dp, int port_no)
-{
-	WARN_ON_ONCE(!rcu_read_lock_held());
-	return ovs_lookup_vport(dp, port_no);
-}
-
-static inline struct vport *ovs_vport_rtnl_rcu(const struct datapath *dp, int port_no)
-{
-	WARN_ON_ONCE(!rcu_read_lock_held() && !rtnl_is_locked());
-	return ovs_lookup_vport(dp, port_no);
-}
-
-static inline struct vport *ovs_vport_rtnl(const struct datapath *dp, int port_no)
-{
-	ASSERT_RTNL();
-	return ovs_lookup_vport(dp, port_no);
-}
-
 /**
  * struct ovs_skb_cb - OVS data in skb CB
  * @flow: The flow associated with this packet.  May be %NULL if no flow.
@@ -131,6 +111,30 @@ struct dp_upcall_info {
 	u32 portid;
 };
 
+/**
+ * struct ovs_net - Per net-namespace data for ovs.
+ * @dps: List of datapaths to enable dumping them all out.
+ * Protected by genl_mutex.
+ */
+struct ovs_net {
+	struct list_head dps;
+	struct work_struct dp_notify_work;
+};
+
+extern int ovs_net_id;
+void ovs_lock(void);
+void ovs_unlock(void);
+
+#ifdef CONFIG_LOCKDEP
+int lockdep_ovsl_is_held(void);
+#else
+#define lockdep_ovsl_is_held()	1
+#endif
+
+#define ASSERT_OVSL()		WARN_ON(unlikely(!lockdep_ovsl_is_held()))
+#define ovsl_dereference(p)					\
+	rcu_dereference_protected(p, lockdep_ovsl_is_held())
+
 static inline struct net *ovs_dp_get_net(struct datapath *dp)
 {
 	return read_pnet(&dp->net);
@@ -141,6 +145,26 @@ static inline void ovs_dp_set_net(struct datapath *dp, struct net *net)
 	write_pnet(&dp->net, net);
 }
 
+struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no);
+
+static inline struct vport *ovs_vport_rcu(const struct datapath *dp, int port_no)
+{
+	WARN_ON_ONCE(!rcu_read_lock_held());
+	return ovs_lookup_vport(dp, port_no);
+}
+
+static inline struct vport *ovs_vport_ovsl_rcu(const struct datapath *dp, int port_no)
+{
+	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
+	return ovs_lookup_vport(dp, port_no);
+}
+
+static inline struct vport *ovs_vport_ovsl(const struct datapath *dp, int port_no)
+{
+	ASSERT_OVSL();
+	return ovs_lookup_vport(dp, port_no);
+}
+
 extern struct notifier_block ovs_dp_device_notifier;
 extern struct genl_multicast_group ovs_dp_vport_multicast_group;
 
@@ -154,4 +178,5 @@ struct sk_buff *ovs_vport_cmd_build_info(struct vport *, u32 pid, u32 seq,
 					 u8 cmd);
 
 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb);
+void ovs_dp_notify_wq(struct work_struct *work);
 #endif /* datapath.h */

net/openvswitch/dp_notify.c

@@ -18,31 +18,18 @@
 
 #include <linux/netdevice.h>
 #include <net/genetlink.h>
+#include <net/netns/generic.h>
 
 #include "datapath.h"
 #include "vport-internal_dev.h"
 #include "vport-netdev.h"
 
-static int dp_device_event(struct notifier_block *unused, unsigned long event,
-			   void *ptr)
+static void dp_detach_port_notify(struct vport *vport)
 {
-	struct net_device *dev = ptr;
-	struct vport *vport;
-
-	if (ovs_is_internal_dev(dev))
-		vport = ovs_internal_dev_get_vport(dev);
-	else
-		vport = ovs_netdev_get_vport(dev);
-
-	if (!vport)
-		return NOTIFY_DONE;
-
-	switch (event) {
-	case NETDEV_UNREGISTER:
-		if (!ovs_is_internal_dev(dev)) {
 	struct sk_buff *notify;
-			struct datapath *dp = vport->dp;
+	struct datapath *dp;
 
+	dp = vport->dp;
 	notify = ovs_vport_cmd_build_info(vport, 0, 0,
 					  OVS_VPORT_CMD_DEL);
 	ovs_dp_detach_port(vport);
@@ -50,14 +37,59 @@ static int dp_device_event(struct notifier_block *unused, unsigned long event,
 		netlink_set_err(ovs_dp_get_net(dp)->genl_sock, 0,
 				ovs_dp_vport_multicast_group.id,
 				PTR_ERR(notify));
-				break;
+		return;
 	}
 
 	genlmsg_multicast_netns(ovs_dp_get_net(dp), notify, 0,
 				ovs_dp_vport_multicast_group.id,
 				GFP_KERNEL);
+}
+
+void ovs_dp_notify_wq(struct work_struct *work)
+{
+	struct ovs_net *ovs_net = container_of(work, struct ovs_net, dp_notify_work);
+	struct datapath *dp;
+
+	ovs_lock();
+	list_for_each_entry(dp, &ovs_net->dps, list_node) {
+		int i;
+
+		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
+			struct vport *vport;
+			struct hlist_node *n;
+
+			hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) {
+				struct netdev_vport *netdev_vport;
+
+				if (vport->ops->type != OVS_VPORT_TYPE_NETDEV)
+					continue;
+
+				netdev_vport = netdev_vport_priv(vport);
+				if (netdev_vport->dev->reg_state == NETREG_UNREGISTERED ||
+				    netdev_vport->dev->reg_state == NETREG_UNREGISTERING)
+					dp_detach_port_notify(vport);
+			}
+		}
 	}
-		break;
+	ovs_unlock();
+}
+
+static int dp_device_event(struct notifier_block *unused, unsigned long event,
+			   void *ptr)
+{
+	struct ovs_net *ovs_net;
+	struct net_device *dev = ptr;
+	struct vport *vport = NULL;
+
+	if (!ovs_is_internal_dev(dev))
+		vport = ovs_netdev_get_vport(dev);
+
+	if (!vport)
+		return NOTIFY_DONE;
+
+	if (event == NETDEV_UNREGISTER) {
+		ovs_net = net_generic(dev_net(dev), ovs_net_id);
+		queue_work(system_wq, &ovs_net->dp_notify_work);
 	}
 
 	return NOTIFY_DONE;

net/openvswitch/vport-internal_dev.c

@@ -173,16 +173,19 @@ static struct vport *internal_dev_create(const struct vport_parms *parms)
 	if (vport->port_no == OVSP_LOCAL)
 		netdev_vport->dev->features |= NETIF_F_NETNS_LOCAL;
 
+	rtnl_lock();
 	err = register_netdevice(netdev_vport->dev);
 	if (err)
 		goto error_free_netdev;
 
 	dev_set_promiscuity(netdev_vport->dev, 1);
+	rtnl_unlock();
 	netif_start_queue(netdev_vport->dev);
 
 	return vport;
 
 error_free_netdev:
+	rtnl_unlock();
 	free_netdev(netdev_vport->dev);
 error_free_vport:
 	ovs_vport_free(vport);
@@ -195,10 +198,13 @@ static void internal_dev_destroy(struct vport *vport)
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 
 	netif_stop_queue(netdev_vport->dev);
+	rtnl_lock();
 	dev_set_promiscuity(netdev_vport->dev, -1);
 
 	/* unregister_netdevice() waits for an RCU grace period. */
 	unregister_netdevice(netdev_vport->dev);
+
+	rtnl_unlock();
 }
 
 static int internal_dev_recv(struct vport *vport, struct sk_buff *skb)

net/openvswitch/vport-netdev.c

@@ -100,16 +100,20 @@ static struct vport *netdev_create(const struct vport_parms *parms)
 		goto error_put;
 	}
 
+	rtnl_lock();
 	err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 					 vport);
 	if (err)
-		goto error_put;
+		goto error_unlock;
 
 	dev_set_promiscuity(netdev_vport->dev, 1);
 	netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
+	rtnl_unlock();
 
 	return vport;
 
+error_unlock:
+	rtnl_unlock();
 error_put:
 	dev_put(netdev_vport->dev);
 error_free_vport:
@@ -131,9 +135,11 @@ static void netdev_destroy(struct vport *vport)
 {
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 
+	rtnl_lock();
 	netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 	netdev_rx_handler_unregister(netdev_vport->dev);
 	dev_set_promiscuity(netdev_vport->dev, -1);
+	rtnl_unlock();
 
 	call_rcu(&netdev_vport->rcu, free_port_rcu);
 }

net/openvswitch/vport.c

@@ -40,7 +40,7 @@ static const struct vport_ops *vport_ops_list[] = {
 	&ovs_internal_vport_ops,
 };
 
-/* Protected by RCU read lock for reading, RTNL lock for writing. */
+/* Protected by RCU read lock for reading, ovs_mutex for writing. */
 static struct hlist_head *dev_table;
 #define VPORT_HASH_BUCKETS 1024
 
@@ -80,7 +80,7 @@ static struct hlist_head *hash_bucket(struct net *net, const char *name)
  *
  * @name: name of port to find
  *
- * Must be called with RTNL or RCU read lock.
+ * Must be called with ovs or RCU read lock.
  */
 struct vport *ovs_vport_locate(struct net *net, const char *name)
 {
@@ -161,7 +161,7 @@ void ovs_vport_free(struct vport *vport)
  * @parms: Information about new vport.
  *
  * Creates a new vport with the specified configuration (which is dependent on
- * device type).  RTNL lock must be held.
+ * device type).  ovs_mutex must be held.
  */
 struct vport *ovs_vport_add(const struct vport_parms *parms)
 {
@@ -169,8 +169,6 @@ struct vport *ovs_vport_add(const struct vport_parms *parms)
 	int err = 0;
 	int i;
 
-	ASSERT_RTNL();
-
 	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
 		if (vport_ops_list[i]->type == parms->type) {
 			struct hlist_head *bucket;
@@ -201,12 +199,10 @@ struct vport *ovs_vport_add(const struct vport_parms *parms)
  * @port: New configuration.
  *
  * Modifies an existing device with the specified configuration (which is
- * dependent on device type).  RTNL lock must be held.
+ * dependent on device type).  ovs_mutex must be held.
  */
 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
 {
-	ASSERT_RTNL();
-
 	if (!vport->ops->set_options)
 		return -EOPNOTSUPP;
 	return vport->ops->set_options(vport, options);
@@ -218,11 +214,11 @@ int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
  * @vport: vport to delete.
  *
  * Detaches @vport from its datapath and destroys it.  It is possible to fail
- * for reasons such as lack of memory.  RTNL lock must be held.
+ * for reasons such as lack of memory.  ovs_mutex must be held.
  */
 void ovs_vport_del(struct vport *vport)
 {
-	ASSERT_RTNL();
+	ASSERT_OVSL();
 
 	hlist_del_rcu(&vport->hash_node);
 
@@ -237,7 +233,7 @@ void ovs_vport_del(struct vport *vport)
  *
  * Retrieves transmit, receive, and error stats for the given device.
  *
- * Must be called with RTNL lock or rcu_read_lock.
+ * Must be called with ovs_mutex or rcu_read_lock.
  */
 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 {
@@ -296,7 +292,7 @@ void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
  * negative error code if a real error occurred.  If an error occurs, @skb is
  * left unmodified.
  *
- * Must be called with RTNL lock or rcu_read_lock.
+ * Must be called with ovs_mutex or rcu_read_lock.
  */
 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
 {
@@ -348,7 +344,7 @@ void ovs_vport_receive(struct vport *vport, struct sk_buff *skb)
  * @vport: vport on which to send the packet
  * @skb: skb to send
  *
- * Sends the given packet and returns the length of data sent.  Either RTNL
+ * Sends the given packet and returns the length of data sent.  Either ovs
  * lock or rcu_read_lock must be held.
  */
 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)

net/openvswitch/vport.h

@@ -138,14 +138,14 @@ struct vport_parms {
 struct vport_ops {
 	enum ovs_vport_type type;
 
-	/* Called with RTNL lock. */
+	/* Called with ovs_mutex. */
 	struct vport *(*create)(const struct vport_parms *);
 	void (*destroy)(struct vport *);
 
 	int (*set_options)(struct vport *, struct nlattr *);
 	int (*get_options)(const struct vport *, struct sk_buff *);
 
-	/* Called with rcu_read_lock or RTNL lock. */
+	/* Called with rcu_read_lock or ovs_mutex. */
 	const char *(*get_name)(const struct vport *);
 	void (*get_config)(const struct vport *, void *);
 	int (*get_ifindex)(const struct vport *);