]> Pileus Git - ~andy/linux/blobdiff - net/openvswitch/datapath.c
openvswitch: Use parallel_ops genl.
[~andy/linux] / net / openvswitch / datapath.c
index a4b724708a1ab0b7d4aedffc07606b8fbfdac0a9..74a5fe6feae3418453a490d3214e5887394abe6f 100644 (file)
@@ -44,6 +44,7 @@
 #include <linux/netfilter_ipv4.h>
 #include <linux/inetdevice.h>
 #include <linux/list.h>
+#include <linux/lockdep.h>
 #include <linux/openvswitch.h>
 #include <linux/rculist.h>
 #include <linux/dmi.h>
 #include "flow.h"
 #include "vport-internal_dev.h"
 
-/**
- * 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;
-};
-
-static int ovs_net_id __read_mostly;
 
 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
 static void rehash_flow_table(struct work_struct *work);
 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
 
+int ovs_net_id __read_mostly;
+
+static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
+                      struct genl_multicast_group *grp)
+{
+       genl_notify(skb, genl_info_net(info), info->snd_portid,
+                   grp->id, info->nlhdr, GFP_KERNEL);
+}
+
 /**
  * DOC: Locking:
  *
- * Writes to device state (add/remove datapath, port, set operations on vports,
- * etc.) are protected by RTNL.
- *
- * Writes to other state (flow table modifications, set miscellaneous datapath
- * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
- * genl_mutex.
+ * All writes e.g. Writes to device state (add/remove datapath, port, set
+ * operations on vports, etc.), Writes to other state (flow table
+ * modifications, set miscellaneous datapath parameters, etc.) are protected
+ * by ovs_lock.
  *
  * Reads are protected by RCU.
  *
  * There are a few special cases (mostly stats) that have their own
  * synchronization but they nest under all of above and don't interact with
  * each other.
+ *
+ * The RTNL lock nests inside ovs_mutex.
  */
 
+static DEFINE_MUTEX(ovs_mutex);
+
+void ovs_lock(void)
+{
+       mutex_lock(&ovs_mutex);
+}
+
+void ovs_unlock(void)
+{
+       mutex_unlock(&ovs_mutex);
+}
+
+#ifdef CONFIG_LOCKDEP
+int lockdep_ovsl_is_held(void)
+{
+       if (debug_locks)
+               return lockdep_is_held(&ovs_mutex);
+       else
+               return 1;
+}
+#endif
+
 static struct vport *new_vport(const struct vport_parms *);
 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
                             const struct dp_upcall_info *);
@@ -95,7 +117,7 @@ static int queue_userspace_packet(struct net *, int dp_ifindex,
                                  struct sk_buff *,
                                  const struct dp_upcall_info *);
 
-/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
+/* Must be called with rcu_read_lock or ovs_mutex. */
 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 {
        struct datapath *dp = NULL;
@@ -113,10 +135,10 @@ static struct datapath *get_dp(struct net *net, int dp_ifindex)
        return dp;
 }
 
-/* Must be called with rcu_read_lock or RTNL lock. */
+/* Must be called with rcu_read_lock or ovs_mutex. */
 const char *ovs_dp_name(const struct datapath *dp)
 {
-       struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
+       struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
        return vport->ops->get_name(vport);
 }
 
@@ -168,7 +190,7 @@ struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
        return NULL;
 }
 
-/* Called with RTNL lock and genl_lock. */
+/* Called with ovs_mutex. */
 static struct vport *new_vport(const struct vport_parms *parms)
 {
        struct vport *vport;
@@ -180,14 +202,12 @@ static struct vport *new_vport(const struct vport_parms *parms)
 
                hlist_add_head_rcu(&vport->dp_hash_node, head);
        }
-
        return vport;
 }
 
-/* Called with RTNL lock. */
 void ovs_dp_detach_port(struct vport *p)
 {
-       ASSERT_RTNL();
+       ASSERT_OVSL();
 
        /* First drop references to device. */
        hlist_del_rcu(&p->dp_hash_node);
@@ -250,7 +270,8 @@ static struct genl_family dp_packet_genl_family = {
        .name = OVS_PACKET_FAMILY,
        .version = OVS_PACKET_VERSION,
        .maxattr = OVS_PACKET_ATTR_MAX,
-       .netnsok = true
+       .netnsok = true,
+       .parallel_ops = true,
 };
 
 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
@@ -337,6 +358,35 @@ static int queue_gso_packets(struct net *net, int dp_ifindex,
        return err;
 }
 
+static size_t key_attr_size(void)
+{
+       return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
+               + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
+               + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
+               + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
+               + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
+               + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
+               + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
+               + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
+               + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
+               + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
+               + nla_total_size(28); /* OVS_KEY_ATTR_ND */
+}
+
+static size_t upcall_msg_size(const struct sk_buff *skb,
+                             const struct nlattr *userdata)
+{
+       size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
+               + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
+               + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
+
+       /* OVS_PACKET_ATTR_USERDATA */
+       if (userdata)
+               size += NLA_ALIGN(userdata->nla_len);
+
+       return size;
+}
+
 static int queue_userspace_packet(struct net *net, int dp_ifindex,
                                  struct sk_buff *skb,
                                  const struct dp_upcall_info *upcall_info)
@@ -345,7 +395,6 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
        struct sk_buff *nskb = NULL;
        struct sk_buff *user_skb; /* to be queued to userspace */
        struct nlattr *nla;
-       unsigned int len;
        int err;
 
        if (vlan_tx_tag_present(skb)) {
@@ -353,7 +402,7 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
                if (!nskb)
                        return -ENOMEM;
 
-               nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
+               nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
                if (!nskb)
                        return -ENOMEM;
 
@@ -366,13 +415,7 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
                goto out;
        }
 
-       len = sizeof(struct ovs_header);
-       len += nla_total_size(skb->len);
-       len += nla_total_size(FLOW_BUFSIZE);
-       if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
-               len += nla_total_size(8);
-
-       user_skb = genlmsg_new(len, GFP_ATOMIC);
+       user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
        if (!user_skb) {
                err = -ENOMEM;
                goto out;
@@ -387,8 +430,9 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
        nla_nest_end(user_skb, nla);
 
        if (upcall_info->userdata)
-               nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
-                           nla_get_u64(upcall_info->userdata));
+               __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
+                         nla_len(upcall_info->userdata),
+                         nla_data(upcall_info->userdata));
 
        nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 
@@ -402,13 +446,13 @@ out:
        return err;
 }
 
-/* Called with genl_mutex. */
+/* Called with ovs_mutex. */
 static int flush_flows(struct datapath *dp)
 {
        struct flow_table *old_table;
        struct flow_table *new_table;
 
-       old_table = genl_dereference(dp->table);
+       old_table = ovsl_dereference(dp->table);
        new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
        if (!new_table)
                return -ENOMEM;
@@ -544,7 +588,7 @@ static int validate_userspace(const struct nlattr *attr)
 {
        static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
                [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
-               [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
+               [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
        };
        struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
        int error;
@@ -661,8 +705,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 
        err = -EINVAL;
        if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
-           !a[OVS_PACKET_ATTR_ACTIONS] ||
-           nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
+           !a[OVS_PACKET_ATTR_ACTIONS])
                goto err;
 
        len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
@@ -672,7 +715,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
                goto err;
        skb_reserve(packet, NET_IP_ALIGN);
 
-       memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
+       nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 
        skb_reset_mac_header(packet);
        eth = eth_hdr(packet);
@@ -680,7 +723,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
        /* Normally, setting the skb 'protocol' field would be handled by a
         * call to eth_type_trans(), but it assumes there's a sending
         * device, which we may not have. */
-       if (ntohs(eth->h_proto) >= 1536)
+       if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
                packet->protocol = eth->h_proto;
        else
                packet->protocol = htons(ETH_P_802_2);
@@ -743,7 +786,7 @@ err:
 }
 
 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
-       [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
+       [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
        [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
        [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 };
@@ -759,7 +802,7 @@ static struct genl_ops dp_packet_genl_ops[] = {
 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 {
        int i;
-       struct flow_table *table = genl_dereference(dp->table);
+       struct flow_table *table = ovsl_dereference(dp->table);
 
        stats->n_flows = ovs_flow_tbl_count(table);
 
@@ -794,14 +837,25 @@ static struct genl_family dp_flow_genl_family = {
        .name = OVS_FLOW_FAMILY,
        .version = OVS_FLOW_VERSION,
        .maxattr = OVS_FLOW_ATTR_MAX,
-       .netnsok = true
+       .netnsok = true,
+       .parallel_ops = true,
 };
 
 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
        .name = OVS_FLOW_MCGROUP
 };
 
-/* Called with genl_lock. */
+static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
+{
+       return NLMSG_ALIGN(sizeof(struct ovs_header))
+               + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
+               + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
+               + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
+               + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
+               + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
+}
+
+/* Called with ovs_mutex. */
 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
                                  struct sk_buff *skb, u32 portid,
                                  u32 seq, u32 flags, u8 cmd)
@@ -815,8 +869,7 @@ static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
        u8 tcp_flags;
        int err;
 
-       sf_acts = rcu_dereference_protected(flow->sf_acts,
-                                           lockdep_genl_is_held());
+       sf_acts = ovsl_dereference(flow->sf_acts);
 
        ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
        if (!ovs_header)
@@ -879,25 +932,10 @@ error:
 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 {
        const struct sw_flow_actions *sf_acts;
-       int len;
-
-       sf_acts = rcu_dereference_protected(flow->sf_acts,
-                                           lockdep_genl_is_held());
-
-       /* OVS_FLOW_ATTR_KEY */
-       len = nla_total_size(FLOW_BUFSIZE);
-       /* OVS_FLOW_ATTR_ACTIONS */
-       len += nla_total_size(sf_acts->actions_len);
-       /* OVS_FLOW_ATTR_STATS */
-       len += nla_total_size(sizeof(struct ovs_flow_stats));
-       /* OVS_FLOW_ATTR_TCP_FLAGS */
-       len += nla_total_size(1);
-       /* OVS_FLOW_ATTR_USED */
-       len += nla_total_size(8);
 
-       len += NLMSG_ALIGN(sizeof(struct ovs_header));
+       sf_acts = ovsl_dereference(flow->sf_acts);
 
-       return genlmsg_new(len, GFP_KERNEL);
+       return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 }
 
 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
@@ -946,12 +984,13 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                goto error;
        }
 
+       ovs_lock();
        dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
        error = -ENODEV;
        if (!dp)
-               goto error;
+               goto err_unlock_ovs;
 
-       table = genl_dereference(dp->table);
+       table = ovsl_dereference(dp->table);
        flow = ovs_flow_tbl_lookup(table, &key, key_len);
        if (!flow) {
                struct sw_flow_actions *acts;
@@ -959,7 +998,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                /* Bail out if we're not allowed to create a new flow. */
                error = -ENOENT;
                if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
-                       goto error;
+                       goto err_unlock_ovs;
 
                /* Expand table, if necessary, to make room. */
                if (ovs_flow_tbl_need_to_expand(table)) {
@@ -969,7 +1008,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                        if (!IS_ERR(new_table)) {
                                rcu_assign_pointer(dp->table, new_table);
                                ovs_flow_tbl_deferred_destroy(table);
-                               table = genl_dereference(dp->table);
+                               table = ovsl_dereference(dp->table);
                        }
                }
 
@@ -977,7 +1016,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                flow = ovs_flow_alloc();
                if (IS_ERR(flow)) {
                        error = PTR_ERR(flow);
-                       goto error;
+                       goto err_unlock_ovs;
                }
                flow->key = key;
                clear_stats(flow);
@@ -1010,11 +1049,10 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                error = -EEXIST;
                if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
                    info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
-                       goto error;
+                       goto err_unlock_ovs;
 
                /* Update actions. */
-               old_acts = rcu_dereference_protected(flow->sf_acts,
-                                                    lockdep_genl_is_held());
+               old_acts = ovsl_dereference(flow->sf_acts);
                acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
                if (acts_attrs &&
                   (old_acts->actions_len != nla_len(acts_attrs) ||
@@ -1025,7 +1063,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                        new_acts = ovs_flow_actions_alloc(acts_attrs);
                        error = PTR_ERR(new_acts);
                        if (IS_ERR(new_acts))
-                               goto error;
+                               goto err_unlock_ovs;
 
                        rcu_assign_pointer(flow->sf_acts, new_acts);
                        ovs_flow_deferred_free_acts(old_acts);
@@ -1041,11 +1079,10 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                        spin_unlock_bh(&flow->lock);
                }
        }
+       ovs_unlock();
 
        if (!IS_ERR(reply))
-               genl_notify(reply, genl_info_net(info), info->snd_portid,
-                          ovs_dp_flow_multicast_group.id, info->nlhdr,
-                          GFP_KERNEL);
+               ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
        else
                netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
                                ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
@@ -1053,6 +1090,8 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 
 error_free_flow:
        ovs_flow_free(flow);
+err_unlock_ovs:
+       ovs_unlock();
 error:
        return error;
 }
@@ -1075,21 +1114,32 @@ static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
        if (err)
                return err;
 
+       ovs_lock();
        dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
-       if (!dp)
-               return -ENODEV;
+       if (!dp) {
+               err = -ENODEV;
+               goto unlock;
+       }
 
-       table = genl_dereference(dp->table);
+       table = ovsl_dereference(dp->table);
        flow = ovs_flow_tbl_lookup(table, &key, key_len);
-       if (!flow)
-               return -ENOENT;
+       if (!flow) {
+               err = -ENOENT;
+               goto unlock;
+       }
 
        reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
                                        info->snd_seq, OVS_FLOW_CMD_NEW);
-       if (IS_ERR(reply))
-               return PTR_ERR(reply);
+       if (IS_ERR(reply)) {
+               err = PTR_ERR(reply);
+               goto unlock;
+       }
 
+       ovs_unlock();
        return genlmsg_reply(reply, info);
+unlock:
+       ovs_unlock();
+       return err;
 }
 
 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
@@ -1104,25 +1154,33 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
        int err;
        int key_len;
 
+       ovs_lock();
        dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
-       if (!dp)
-               return -ENODEV;
-
-       if (!a[OVS_FLOW_ATTR_KEY])
-               return flush_flows(dp);
+       if (!dp) {
+               err = -ENODEV;
+               goto unlock;
+       }
 
+       if (!a[OVS_FLOW_ATTR_KEY]) {
+               err = flush_flows(dp);
+               goto unlock;
+       }
        err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
        if (err)
-               return err;
+               goto unlock;
 
-       table = genl_dereference(dp->table);
+       table = ovsl_dereference(dp->table);
        flow = ovs_flow_tbl_lookup(table, &key, key_len);
-       if (!flow)
-               return -ENOENT;
+       if (!flow) {
+               err = -ENOENT;
+               goto unlock;
+       }
 
        reply = ovs_flow_cmd_alloc_info(flow);
-       if (!reply)
-               return -ENOMEM;
+       if (!reply) {
+               err = -ENOMEM;
+               goto unlock;
+       }
 
        ovs_flow_tbl_remove(table, flow);
 
@@ -1131,10 +1189,13 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
        BUG_ON(err < 0);
 
        ovs_flow_deferred_free(flow);
+       ovs_unlock();
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
+       ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
        return 0;
+unlock:
+       ovs_unlock();
+       return err;
 }
 
 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
@@ -1143,11 +1204,14 @@ static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
        struct datapath *dp;
        struct flow_table *table;
 
+       ovs_lock();
        dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
-       if (!dp)
+       if (!dp) {
+               ovs_unlock();
                return -ENODEV;
+       }
 
-       table = genl_dereference(dp->table);
+       table = ovsl_dereference(dp->table);
 
        for (;;) {
                struct sw_flow *flow;
@@ -1168,6 +1232,7 @@ static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
                cb->args[0] = bucket;
                cb->args[1] = obj;
        }
+       ovs_unlock();
        return skb->len;
 }
 
@@ -1206,13 +1271,24 @@ static struct genl_family dp_datapath_genl_family = {
        .name = OVS_DATAPATH_FAMILY,
        .version = OVS_DATAPATH_VERSION,
        .maxattr = OVS_DP_ATTR_MAX,
-       .netnsok = true
+       .netnsok = true,
+       .parallel_ops = true,
 };
 
 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
        .name = OVS_DATAPATH_MCGROUP
 };
 
+static size_t ovs_dp_cmd_msg_size(void)
+{
+       size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
+
+       msgsize += nla_total_size(IFNAMSIZ);
+       msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
+
+       return msgsize;
+}
+
 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
                                u32 portid, u32 seq, u32 flags, u8 cmd)
 {
@@ -1251,7 +1327,7 @@ static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
        struct sk_buff *skb;
        int retval;
 
-       skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+       skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
        if (!skb)
                return ERR_PTR(-ENOMEM);
 
@@ -1263,7 +1339,7 @@ static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
        return skb;
 }
 
-/* Called with genl_mutex and optionally with RTNL lock also. */
+/* Called with ovs_mutex. */
 static struct datapath *lookup_datapath(struct net *net,
                                        struct ovs_header *ovs_header,
                                        struct nlattr *a[OVS_DP_ATTR_MAX + 1])
@@ -1297,12 +1373,12 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
        if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
                goto err;
 
-       rtnl_lock();
+       ovs_lock();
 
        err = -ENOMEM;
        dp = kzalloc(sizeof(*dp), GFP_KERNEL);
        if (dp == NULL)
-               goto err_unlock_rtnl;
+               goto err_unlock_ovs;
 
        ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
 
@@ -1353,37 +1429,34 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 
        ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
        list_add_tail(&dp->list_node, &ovs_net->dps);
-       rtnl_unlock();
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_datapath_multicast_group.id, info->nlhdr,
-                   GFP_KERNEL);
+       ovs_unlock();
+
+       ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
        return 0;
 
 err_destroy_local_port:
-       ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
+       ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
 err_destroy_ports_array:
        kfree(dp->ports);
 err_destroy_percpu:
        free_percpu(dp->stats_percpu);
 err_destroy_table:
-       ovs_flow_tbl_destroy(genl_dereference(dp->table));
+       ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
 err_free_dp:
        release_net(ovs_dp_get_net(dp));
        kfree(dp);
-err_unlock_rtnl:
-       rtnl_unlock();
+err_unlock_ovs:
+       ovs_unlock();
 err:
        return err;
 }
 
-/* Called with genl_mutex. */
+/* Called with ovs_mutex. */
 static void __dp_destroy(struct datapath *dp)
 {
        int i;
 
-       rtnl_lock();
-
        for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
                struct vport *vport;
                struct hlist_node *n;
@@ -1394,14 +1467,11 @@ static void __dp_destroy(struct datapath *dp)
        }
 
        list_del(&dp->list_node);
-       ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
 
-       /* rtnl_unlock() will wait until all the references to devices that
-        * are pending unregistration have been dropped.  We do it here to
-        * ensure that any internal devices (which contain DP pointers) are
-        * fully destroyed before freeing the datapath.
+       /* OVSP_LOCAL is datapath internal port. We need to make sure that
+        * all port in datapath are destroyed first before freeing datapath.
         */
-       rtnl_unlock();
+       ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
 
        call_rcu(&dp->rcu, destroy_dp_rcu);
 }
@@ -1412,24 +1482,27 @@ static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
        struct datapath *dp;
        int err;
 
+       ovs_lock();
        dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
        err = PTR_ERR(dp);
        if (IS_ERR(dp))
-               return err;
+               goto unlock;
 
        reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
                                      info->snd_seq, OVS_DP_CMD_DEL);
        err = PTR_ERR(reply);
        if (IS_ERR(reply))
-               return err;
+               goto unlock;
 
        __dp_destroy(dp);
+       ovs_unlock();
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_datapath_multicast_group.id, info->nlhdr,
-                   GFP_KERNEL);
+       ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
 
        return 0;
+unlock:
+       ovs_unlock();
+       return err;
 }
 
 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
@@ -1438,9 +1511,11 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
        struct datapath *dp;
        int err;
 
+       ovs_lock();
        dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
+       err = PTR_ERR(dp);
        if (IS_ERR(dp))
-               return PTR_ERR(dp);
+               goto unlock;
 
        reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
                                      info->snd_seq, OVS_DP_CMD_NEW);
@@ -1448,31 +1523,45 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
                err = PTR_ERR(reply);
                netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
                                ovs_dp_datapath_multicast_group.id, err);
-               return 0;
+               err = 0;
+               goto unlock;
        }
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_datapath_multicast_group.id, info->nlhdr,
-                   GFP_KERNEL);
+       ovs_unlock();
+       ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
 
        return 0;
+unlock:
+       ovs_unlock();
+       return err;
 }
 
 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
 {
        struct sk_buff *reply;
        struct datapath *dp;
+       int err;
 
+       ovs_lock();
        dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
-       if (IS_ERR(dp))
-               return PTR_ERR(dp);
+       if (IS_ERR(dp)) {
+               err = PTR_ERR(dp);
+               goto unlock;
+       }
 
        reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
                                      info->snd_seq, OVS_DP_CMD_NEW);
-       if (IS_ERR(reply))
-               return PTR_ERR(reply);
+       if (IS_ERR(reply)) {
+               err = PTR_ERR(reply);
+               goto unlock;
+       }
 
+       ovs_unlock();
        return genlmsg_reply(reply, info);
+
+unlock:
+       ovs_unlock();
+       return err;
 }
 
 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
@@ -1482,6 +1571,7 @@ static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
        int skip = cb->args[0];
        int i = 0;
 
+       ovs_lock();
        list_for_each_entry(dp, &ovs_net->dps, list_node) {
                if (i >= skip &&
                    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
@@ -1490,6 +1580,7 @@ static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
                        break;
                i++;
        }
+       ovs_unlock();
 
        cb->args[0] = i;
 
@@ -1535,14 +1626,15 @@ static struct genl_family dp_vport_genl_family = {
        .name = OVS_VPORT_FAMILY,
        .version = OVS_VPORT_VERSION,
        .maxattr = OVS_VPORT_ATTR_MAX,
-       .netnsok = true
+       .netnsok = true,
+       .parallel_ops = true,
 };
 
 struct genl_multicast_group ovs_dp_vport_multicast_group = {
        .name = OVS_VPORT_MCGROUP
 };
 
-/* Called with RTNL lock or RCU read lock. */
+/* Called with ovs_mutex or RCU read lock. */
 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
                                   u32 portid, u32 seq, u32 flags, u8 cmd)
 {
@@ -1581,7 +1673,7 @@ error:
        return err;
 }
 
-/* Called with RTNL lock or RCU read lock. */
+/* Called with ovs_mutex or RCU read lock. */
 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
                                         u32 seq, u8 cmd)
 {
@@ -1593,14 +1685,12 @@ struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
                return ERR_PTR(-ENOMEM);
 
        retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
-       if (retval < 0) {
-               kfree_skb(skb);
-               return ERR_PTR(retval);
-       }
+       BUG_ON(retval < 0);
+
        return skb;
 }
 
-/* Called with RTNL lock or RCU read lock. */
+/* Called with ovs_mutex or RCU read lock. */
 static struct vport *lookup_vport(struct net *net,
                                  struct ovs_header *ovs_header,
                                  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
@@ -1626,9 +1716,9 @@ static struct vport *lookup_vport(struct net *net,
                if (!dp)
                        return ERR_PTR(-ENODEV);
 
-               vport = ovs_vport_rtnl_rcu(dp, port_no);
+               vport = ovs_vport_ovsl_rcu(dp, port_no);
                if (!vport)
-                       return ERR_PTR(-ENOENT);
+                       return ERR_PTR(-ENODEV);
                return vport;
        } else
                return ERR_PTR(-EINVAL);
@@ -1650,7 +1740,7 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
            !a[OVS_VPORT_ATTR_UPCALL_PID])
                goto exit;
 
-       rtnl_lock();
+       ovs_lock();
        dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
        err = -ENODEV;
        if (!dp)
@@ -1663,7 +1753,7 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
                if (port_no >= DP_MAX_PORTS)
                        goto exit_unlock;
 
-               vport = ovs_vport_rtnl_rcu(dp, port_no);
+               vport = ovs_vport_ovsl(dp, port_no);
                err = -EBUSY;
                if (vport)
                        goto exit_unlock;
@@ -1673,7 +1763,7 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
                                err = -EFBIG;
                                goto exit_unlock;
                        }
-                       vport = ovs_vport_rtnl(dp, port_no);
+                       vport = ovs_vport_ovsl(dp, port_no);
                        if (!vport)
                                break;
                }
@@ -1699,11 +1789,11 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
                ovs_dp_detach_port(vport);
                goto exit_unlock;
        }
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
+
+       ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
 
 exit_unlock:
-       rtnl_unlock();
+       ovs_unlock();
 exit:
        return err;
 }
@@ -1715,7 +1805,7 @@ static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
        struct vport *vport;
        int err;
 
-       rtnl_lock();
+       ovs_lock();
        vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
        err = PTR_ERR(vport);
        if (IS_ERR(vport))
@@ -1726,26 +1816,35 @@ static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
            nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
                err = -EINVAL;
 
+       reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+       if (!reply) {
+               err = -ENOMEM;
+               goto exit_unlock;
+       }
+
        if (!err && a[OVS_VPORT_ATTR_OPTIONS])
                err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
        if (err)
-               goto exit_unlock;
+               goto exit_free;
+
        if (a[OVS_VPORT_ATTR_UPCALL_PID])
                vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
 
-       reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
-                                        OVS_VPORT_CMD_NEW);
-       if (IS_ERR(reply)) {
-               netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
-                               ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
-               goto exit_unlock;
-       }
+       err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
+                                     info->snd_seq, 0, OVS_VPORT_CMD_NEW);
+       BUG_ON(err < 0);
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
+       ovs_unlock();
+       ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
+       return 0;
 
-exit_unlock:
        rtnl_unlock();
+       return 0;
+
+exit_free:
+       kfree_skb(reply);
+exit_unlock:
+       ovs_unlock();
        return err;
 }
 
@@ -1756,7 +1855,7 @@ static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
        struct vport *vport;
        int err;
 
-       rtnl_lock();
+       ovs_lock();
        vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
        err = PTR_ERR(vport);
        if (IS_ERR(vport))
@@ -1776,11 +1875,10 @@ static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
        err = 0;
        ovs_dp_detach_port(vport);
 
-       genl_notify(reply, genl_info_net(info), info->snd_portid,
-                   ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
+       ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
 
 exit_unlock:
-       rtnl_unlock();
+       ovs_unlock();
        return err;
 }
 
@@ -1940,13 +2038,13 @@ static void rehash_flow_table(struct work_struct *work)
        struct datapath *dp;
        struct net *net;
 
-       genl_lock();
+       ovs_lock();
        rtnl_lock();
        for_each_net(net) {
                struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 
                list_for_each_entry(dp, &ovs_net->dps, list_node) {
-                       struct flow_table *old_table = genl_dereference(dp->table);
+                       struct flow_table *old_table = ovsl_dereference(dp->table);
                        struct flow_table *new_table;
 
                        new_table = ovs_flow_tbl_rehash(old_table);
@@ -1957,8 +2055,7 @@ static void rehash_flow_table(struct work_struct *work)
                }
        }
        rtnl_unlock();
-       genl_unlock();
-
+       ovs_unlock();
        schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
 }
 
@@ -1967,18 +2064,21 @@ static int __net_init ovs_init_net(struct net *net)
        struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 
        INIT_LIST_HEAD(&ovs_net->dps);
+       INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
        return 0;
 }
 
 static void __net_exit ovs_exit_net(struct net *net)
 {
-       struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
        struct datapath *dp, *dp_next;
+       struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 
-       genl_lock();
+       ovs_lock();
        list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
                __dp_destroy(dp);
-       genl_unlock();
+       ovs_unlock();
+
+       cancel_work_sync(&ovs_net->dp_notify_work);
 }
 
 static struct pernet_operations ovs_net_ops = {