Use skb_copy_bits() to copy data that might not be pulled into headers.
[sliver-openvswitch.git] / datapath / datapath.c
index 639a483..07fa92d 100644 (file)
 #include <linux/random.h>
 #include <asm/system.h>
 #include <linux/netfilter_bridge.h>
+#include <linux/netfilter_ipv4.h>
 #include <linux/inetdevice.h>
 #include <linux/list.h>
+#include <linux/rculist.h>
+#include <linux/workqueue.h>
+#include <linux/dmi.h>
 
-#include "openflow-netlink.h"
+#include "openflow/nicira-ext.h"
+#include "openflow/openflow-netlink.h"
 #include "datapath.h"
+#include "nx_act_snat.h"
 #include "table.h"
 #include "chain.h"
+#include "dp_dev.h"
 #include "forward.h"
 #include "flow.h"
-#include "datapath_t.h"
 
 #include "compat.h"
 
 
+/* Strings to describe the manufacturer, hardware, and software.  This data 
+ * is queriable through the switch description stats message. */
+static char mfr_desc[DESC_STR_LEN] = "Nicira Networks, Inc.";
+static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
+static char sw_desc[DESC_STR_LEN] = VERSION BUILDNR;
+static char serial_num[SERIAL_NUM_LEN] = "None";
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
+module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
+module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
+module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
+#else
+MODULE_PARM(mfr_desc, "s");
+MODULE_PARM(hw_desc, "s");
+MODULE_PARM(sw_desc, "s");
+MODULE_PARM(serial_num, "s");
+#endif
+
+
 /* Number of milliseconds between runs of the maintenance thread. */
 #define MAINT_SLEEP_MSECS 1000
 
-#define BRIDGE_PORT_NO_FLOOD   0x00000001 
-
 #define UINT32_MAX                       4294967295U
+#define UINT16_MAX                       65535
 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
 
-struct net_bridge_port {
-       u16     port_no;
-       u32 flags;
-       struct datapath *dp;
-       struct net_device *dev;
-       struct list_head node; /* Element in datapath.ports. */
-};
-
 static struct genl_family dp_genl_family;
 static struct genl_multicast_group mc_group;
 
-int dp_dev_setup(struct net_device *dev);  
-
 /* It's hard to imagine wanting more than one datapath, but... */
 #define DP_MAX 32
 
-/* datapaths.  Protected on the read side by rcu_read_lock, on the write side
- * by dp_mutex.
+/* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
+ * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
+ * maintained by the Generic Netlink code, but the timeout path needs mutual
+ * exclusion too.
  *
  * It is safe to access the datapath and net_bridge_port structures with just
- * the dp_mutex, but to access the chain you need to take the rcu_read_lock
- * also (because dp_mutex doesn't prevent flows from being destroyed).
+ * dp_mutex.
  */
 static struct datapath *dps[DP_MAX];
-static DEFINE_MUTEX(dp_mutex);
+DEFINE_MUTEX(dp_mutex);
+EXPORT_SYMBOL(dp_mutex);
 
 static int dp_maint_func(void *data);
-static int send_port_status(struct net_bridge_port *p, uint8_t status);
-
+static void init_port_status(struct net_bridge_port *p);
+static int dp_genl_openflow_done(struct netlink_callback *);
+static struct net_bridge_port *new_nbp(struct datapath *,
+                                      struct net_device *, int port_no);
 
-/* nla_unreserve - reduce amount of space reserved by nla_reserve  
+/* nla_shrink - reduce amount of space reserved by nla_reserve
  * @skb: socket buffer from which to recover room
  * @nla: netlink attribute to adjust
- * @len: amount by which to reduce attribute payload
+ * @len: new length of attribute payload
  *
  * Reduces amount of space reserved by a call to nla_reserve.
  *
  * No other attributes may be added between calling nla_reserve and this
  * function, since it will create a hole in the message.
  */
-void nla_unreserve(struct sk_buff *skb, struct nlattr *nla, int len)
+void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 {
-       skb->tail -= len;
-       skb->len  -= len;
-
-       nla->nla_len -= len;
+       int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
+       BUG_ON(delta > 0);
+       skb->tail += delta;
+       skb->len  += delta;
+       nla->nla_len = nla_attr_size(len);
 }
 
+/* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
+ * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
+ * must specify the datapath to use.
+ *
+ * '*max_openflow_len' receives the maximum number of bytes that are available
+ * for the embedded OpenFlow message.  The caller must call
+ * resize_openflow_skb() to set the actual size of the message to this number
+ * of bytes or less.
+ *
+ * Returns the openflow header if successful, otherwise (if 'skb' is too small)
+ * an error code. */
 static void *
-alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
-                  const struct sender *sender, struct sk_buff **pskb) 
+put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
+                    const struct sender *sender, int *max_openflow_len)
 {
-       size_t genl_len;
-       struct sk_buff *skb;
-       struct nlattr *attr;
        struct ofp_header *oh;
-
-       genl_len = nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
-       genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
-       skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
-       if (!skb) {
-               if (net_ratelimit())
-                       printk("alloc_openflow_skb: genlmsg_new failed\n");
-               return NULL;
-       }
+       struct nlattr *attr;
+       int openflow_len;
 
        /* Assemble the Generic Netlink wrapper. */
        if (!genlmsg_put(skb,
                         sender ? sender->pid : 0,
                         sender ? sender->seq : 0,
                         &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
-               BUG();
+               return ERR_PTR(-ENOBUFS);
        if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
-               BUG();
+               return ERR_PTR(-ENOBUFS);
+       openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
+       if (openflow_len < sizeof *oh)
+               return ERR_PTR(-ENOBUFS);
+       *max_openflow_len = openflow_len;
        attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
        BUG_ON(!attr);
-       nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 
-       /* Fill in the header. */
+       /* Fill in the header.  The caller is responsible for the length. */
        oh = nla_data(attr);
        oh->version = OFP_VERSION;
        oh->type = type;
-       oh->length = htons(openflow_len);
        oh->xid = sender ? sender->xid : 0;
 
        return oh;
 }
 
+/* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
+ * length 'new_length' (in bytes), adjusting pointers and size values as
+ * necessary. */
 static void
 resize_openflow_skb(struct sk_buff *skb,
                    struct ofp_header *oh, size_t new_length)
 {
-       struct nlattr *attr;
-
-       BUG_ON(new_length > ntohs(oh->length));
-       attr = ((void *) oh) - NLA_HDRLEN;
-       nla_unreserve(skb, attr, ntohs(oh->length) - new_length);
+       struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
+       nla_shrink(skb, attr, new_length);
        oh->length = htons(new_length);
        nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 }
 
+/* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
+ * length.  Returns a null pointer if memory is unavailable, otherwise returns
+ * the OpenFlow header and stores a pointer to the skb in '*pskb'. 
+ *
+ * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
+ * used as the message's destination.  'dp' must specify the datapath to
+ * use.  */
+static void *
+alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
+                  const struct sender *sender, struct sk_buff **pskb) 
+{
+       struct ofp_header *oh;
+       size_t genl_len;
+       struct sk_buff *skb;
+       int max_openflow_len;
+
+       if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
+               if (net_ratelimit())
+                       printk("alloc_openflow_skb: openflow message too large: %zu\n", 
+                                       openflow_len);
+               return NULL;
+       }
+
+       genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
+       genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
+       genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
+       skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
+       if (!skb) {
+               if (net_ratelimit())
+                       printk("alloc_openflow_skb: genlmsg_new failed\n");
+               return NULL;
+       }
+
+       oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
+       BUG_ON(!oh || IS_ERR(oh));
+       resize_openflow_skb(skb, oh, openflow_len);
+
+       return oh;
+}
+
+/* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
+ * listeners. */
 static int
 send_openflow_skb(struct sk_buff *skb, const struct sender *sender) 
 {
-       int err = (sender
-                  ? genlmsg_unicast(skb, sender->pid)
-                  : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
-       if (err && net_ratelimit())
-               printk(KERN_WARNING "send_openflow_skb: send failed: %d\n",
-                      err);
-       return err;
+       return (sender
+               ? genlmsg_unicast(skb, sender->pid)
+               : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 }
 
-/* Generates a unique datapath id.  It incorporates the datapath index
- * and a hardware address, if available.  If not, it generates a random
- * one.
- */
+/* Retrieves the datapath id, which is the MAC address of the "of" device. */
 static 
-uint64_t gen_datapath_id(uint16_t dp_idx)
+uint64_t get_datapath_id(struct net_device *dev)
 {
-       uint64_t id;
+       uint64_t id = 0;
        int i;
-       struct net_device *dev;
-
-       /* The top 16 bits are used to identify the datapath.  The lower 48 bits
-        * use an interface address.  */
-       id = (uint64_t)dp_idx << 48;
-       if ((dev = dev_get_by_name(&init_net, "ctl0")) 
-                       || (dev = dev_get_by_name(&init_net, "eth0"))) {
-               for (i=0; i<ETH_ALEN; i++) {
-                       id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
-               }
-               dev_put(dev);
-       } else {
-               /* Randomly choose the lower 48 bits if we cannot find an
-                * address and mark the most significant bit to indicate that
-                * this was randomly generated. */
-               uint8_t rand[ETH_ALEN];
-               get_random_bytes(rand, ETH_ALEN);
-               id |= (uint64_t)1 << 63;
-               for (i=0; i<ETH_ALEN; i++) {
-                       id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
-               }
-       }
+
+       for (i=0; i<ETH_ALEN; i++) 
+               id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 
        return id;
 }
 
 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
- * negative error code.
- *
- * Not called with any locks. */
+ * negative error code. */
 static int new_dp(int dp_idx)
 {
        struct datapath *dp;
@@ -211,9 +248,8 @@ static int new_dp(int dp_idx)
        if (!try_module_get(THIS_MODULE))
                return -ENODEV;
 
-       mutex_lock(&dp_mutex);
-       dp = rcu_dereference(dps[dp_idx]);
-       if (dp != NULL) {
+       /* Exit early if a datapath with that number already exists. */
+       if (dps[dp_idx]) {
                err = -EEXIST;
                goto err_unlock;
        }
@@ -223,122 +259,142 @@ static int new_dp(int dp_idx)
        if (dp == NULL)
                goto err_unlock;
 
+       /* Setup our "of" device */
+       err = dp_dev_setup(dp);
+       if (err)
+               goto err_free_dp;
+
        dp->dp_idx = dp_idx;
-       dp->id = gen_datapath_id(dp_idx);
+       dp->id = get_datapath_id(dp->netdev);
        dp->chain = chain_create(dp);
        if (dp->chain == NULL)
-               goto err_free_dp;
+               goto err_destroy_dp_dev;
        INIT_LIST_HEAD(&dp->port_list);
 
-#if 0
-       /* Setup our "of" device */
-       dp->dev.priv = dp;
-       rtnl_lock();
-       err = dp_dev_setup(&dp->dev);
-       rtnl_unlock();
-       if (err != 0) 
-               printk("datapath: problem setting up 'of' device\n");
-#endif
+       dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
+       if (IS_ERR(dp->local_port)) {
+               err = PTR_ERR(dp->local_port);
+               goto err_destroy_local_port;
+       }
 
-       dp->config.flags = 0;
-       dp->config.miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
+       dp->flags = 0;
+       dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 
        dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
        if (IS_ERR(dp->dp_task))
-               goto err_free_dp;
+               goto err_destroy_chain;
 
-       rcu_assign_pointer(dps[dp_idx], dp);
-       mutex_unlock(&dp_mutex);
+       dps[dp_idx] = dp;
 
        return 0;
 
+err_destroy_local_port:
+       dp_del_switch_port(dp->local_port);
+err_destroy_chain:
+       chain_destroy(dp->chain);
+err_destroy_dp_dev:
+       dp_dev_destroy(dp);
 err_free_dp:
        kfree(dp);
 err_unlock:
-       mutex_unlock(&dp_mutex);
        module_put(THIS_MODULE);
                return err;
 }
 
-/* Find and return a free port number under 'dp'.  Called under dp_mutex. */
+/* Find and return a free port number under 'dp'. */
 static int find_portno(struct datapath *dp)
 {
        int i;
-       for (i = 0; i < OFPP_MAX; i++)
+       for (i = 0; i < DP_MAX_PORTS; i++)
                if (dp->ports[i] == NULL)
                        return i;
        return -EXFULL;
 }
 
 static struct net_bridge_port *new_nbp(struct datapath *dp,
-                                                                          struct net_device *dev)
+                                      struct net_device *dev, int port_no)
 {
        struct net_bridge_port *p;
-       int port_no;
 
-       port_no = find_portno(dp);
-       if (port_no < 0)
-               return ERR_PTR(port_no);
+       if (dev->br_port != NULL)
+               return ERR_PTR(-EBUSY);
 
        p = kzalloc(sizeof(*p), GFP_KERNEL);
        if (p == NULL)
                return ERR_PTR(-ENOMEM);
 
-       p->dp = dp;
+       rtnl_lock();
+       dev_set_promiscuity(dev, 1);
+       rtnl_unlock();
        dev_hold(dev);
+       p->dp = dp;
        p->dev = dev;
        p->port_no = port_no;
+       spin_lock_init(&p->lock);
+       INIT_WORK(&p->port_task, NULL);
+       if (port_no != OFPP_LOCAL)
+               rcu_assign_pointer(dev->br_port, p);
+       if (port_no < DP_MAX_PORTS)
+               rcu_assign_pointer(dp->ports[port_no], p); 
+       list_add_rcu(&p->node, &dp->port_list);
 
        return p;
 }
 
-/* Called with dp_mutex. */
 int add_switch_port(struct datapath *dp, struct net_device *dev)
 {
        struct net_bridge_port *p;
+       int port_no;
 
-       if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
+       if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
+           || is_dp_dev(dev))
                return -EINVAL;
 
-       if (dev->br_port != NULL)
-               return -EBUSY;
+       port_no = find_portno(dp);
+       if (port_no < 0)
+               return port_no;
 
-       p = new_nbp(dp, dev);
+       p = new_nbp(dp, dev, port_no);
        if (IS_ERR(p))
                return PTR_ERR(p);
 
-       dev_hold(dev);
-       rcu_assign_pointer(dev->br_port, p);
-       rtnl_lock();
-       dev_set_promiscuity(dev, 1);
-       rtnl_unlock();
-
-       rcu_assign_pointer(dp->ports[p->port_no], p);
-       list_add_rcu(&p->node, &dp->port_list);
+       init_port_status(p);
 
        /* Notify the ctlpath that this port has been added */
-       send_port_status(p, OFPPR_ADD);
+       dp_send_port_status(p, OFPPR_ADD);
 
        return 0;
 }
 
-/* Delete 'p' from switch.
- * Called with dp_mutex. */
-static int del_switch_port(struct net_bridge_port *p)
+/* Delete 'p' from switch. */
+int dp_del_switch_port(struct net_bridge_port *p)
 {
+#ifdef SUPPORT_SNAT
+       unsigned long flags;
+#endif
+
        /* First drop references to device. */
+       cancel_work_sync(&p->port_task);
        rtnl_lock();
        dev_set_promiscuity(p->dev, -1);
        rtnl_unlock();
        list_del_rcu(&p->node);
-       rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
+       if (p->port_no != OFPP_LOCAL)
+               rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
        rcu_assign_pointer(p->dev->br_port, NULL);
 
        /* Then wait until no one is still using it, and destroy it. */
        synchronize_rcu();
 
+#ifdef SUPPORT_SNAT
+       /* Free any SNAT configuration on the port. */
+       spin_lock_irqsave(&p->lock, flags);
+       snat_free_conf(p);
+       spin_unlock_irqrestore(&p->lock, flags);
+#endif
+
        /* Notify the ctlpath that this port no longer exists */
-       send_port_status(p, OFPPR_DELETE);
+       dp_send_port_status(p, OFPPR_DELETE);
 
        dev_put(p->dev);
        kfree(p);
@@ -346,25 +402,26 @@ static int del_switch_port(struct net_bridge_port *p)
        return 0;
 }
 
-/* Called with dp_mutex. */
 static void del_dp(struct datapath *dp)
 {
        struct net_bridge_port *p, *n;
 
-#if 0
-       /* Unregister the "of" device of this dp */
-       rtnl_lock();
-       unregister_netdevice(&dp->dev);
-       rtnl_unlock();
-#endif
-
        kthread_stop(dp->dp_task);
 
        /* Drop references to DP. */
        list_for_each_entry_safe (p, n, &dp->port_list, node)
-               del_switch_port(p);
+               dp_del_switch_port(p);
        rcu_assign_pointer(dps[dp->dp_idx], NULL);
 
+       /* Kill off local_port dev references from buffered packets that have
+        * associated dst entries. */
+       synchronize_rcu();
+       fwd_discard_all();
+
+       /* Destroy dp->netdev.  (Must follow deleting switch ports since
+        * dp->local_port has a reference to it.) */
+       dp_dev_destroy(dp);
+
        /* Wait until no longer in use, then destroy it. */
        synchronize_rcu();
        chain_destroy(dp->chain);
@@ -377,80 +434,75 @@ static int dp_maint_func(void *data)
        struct datapath *dp = (struct datapath *) data;
 
        while (!kthread_should_stop()) {
-#if 1
-               chain_timeout(dp->chain);
-#else
-               int count = chain_timeout(dp->chain);
-               chain_print_stats(dp->chain);
-               if (count)
-                       printk("%d flows timed out\n", count);
+#ifdef SUPPORT_SNAT
+               struct net_bridge_port *p;
+
+               /* Expire old SNAT entries */
+               rcu_read_lock();
+               list_for_each_entry_rcu (p, &dp->port_list, node) 
+                       snat_maint(p);
+               rcu_read_unlock();
 #endif
+
+               /* Timeout old entries */
+               chain_timeout(dp->chain);
                msleep_interruptible(MAINT_SLEEP_MSECS);
        }
                
        return 0;
 }
 
+static void
+do_port_input(struct net_bridge_port *p, struct sk_buff *skb) 
+{
+       /* Make our own copy of the packet.  Otherwise we will mangle the
+        * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
+        * (No one comes after us, since we tell handle_bridge() that we took
+        * the packet.) */
+       skb = skb_share_check(skb, GFP_ATOMIC);
+       if (!skb)
+               return;
+
+#ifdef SUPPORT_SNAT
+       /* Check if this packet needs early SNAT processing. */
+       if (snat_pre_route(skb)) {
+               kfree_skb(skb);
+               return;
+       }
+#endif
+
+       /* Push the Ethernet header back on. */
+       skb_push(skb, ETH_HLEN);
+       skb_reset_mac_header(skb);
+       fwd_port_input(p->dp->chain, skb, p);
+}
+
 /*
  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
- * different set of devices!)  Returns 0 if *pskb should be processed further,
- * 1 if *pskb is handled. */
+ * different set of devices!)
+ */
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 /* Called with rcu_read_lock. */
 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
                                         struct sk_buff *skb)
 {
-       struct ethhdr *eh = eth_hdr(skb);
-       struct sk_buff *skb_local = NULL;
-
-
-       if (compare_ether_addr(eh->h_dest, skb->dev->dev_addr) == 0) 
-               return skb;
-
-       if (is_broadcast_ether_addr(eh->h_dest)
-                               || is_multicast_ether_addr(eh->h_dest)
-                               || is_local_ether_addr(eh->h_dest)) 
-               skb_local = skb_clone(skb, GFP_ATOMIC);
-
-       /* Push the Ethernet header back on. */
-       if (skb->protocol == htons(ETH_P_8021Q))
-               skb_push(skb, VLAN_ETH_HLEN);
-       else
-               skb_push(skb, ETH_HLEN);
-
-       fwd_port_input(p->dp->chain, skb, p->port_no);
-
-       return skb_local;
+       do_port_input(p, skb);
+       return NULL;
 }
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 {
-       /* Push the Ethernet header back on. */
-       if ((*pskb)->protocol == htons(ETH_P_8021Q))
-               skb_push(*pskb, VLAN_ETH_HLEN);
-       else
-               skb_push(*pskb, ETH_HLEN);
-
-       fwd_port_input(p->dp->chain, *pskb, p->port_no);
+       do_port_input(p, *pskb);
        return 1;
 }
-#else 
+#else
 /* NB: This has only been tested on 2.4.35 */
-
-/* Called without any locks (?) */
 static void dp_frame_hook(struct sk_buff *skb)
 {
        struct net_bridge_port *p = skb->dev->br_port;
-
-       /* Push the Ethernet header back on. */
-       if (skb->protocol == htons(ETH_P_8021Q))
-               skb_push(skb, VLAN_ETH_HLEN);
-       else
-               skb_push(skb, ETH_HLEN);
-
        if (p) {
                rcu_read_lock();
-               fwd_port_input(p->dp->chain, skb, p->port_no);
+               do_port_input(p, skb);
                rcu_read_unlock();
        } else
                kfree_skb(skb);
@@ -460,17 +512,6 @@ static void dp_frame_hook(struct sk_buff *skb)
 /* Forwarding output path.
  * Based on net/bridge/br_forward.c. */
 
-/* Don't forward packets to originating port or with flooding disabled */
-static inline int should_deliver(const struct net_bridge_port *p,
-                       const struct sk_buff *skb)
-{
-       if ((skb->dev == p->dev) || (p->flags & BRIDGE_PORT_NO_FLOOD)) {
-               return 0;
-       } 
-
-       return 1;
-}
-
 static inline unsigned packet_length(const struct sk_buff *skb)
 {
        int length = skb->len - ETH_HLEN;
@@ -479,15 +520,18 @@ static inline unsigned packet_length(const struct sk_buff *skb)
        return length;
 }
 
+/* Send packets out all the ports except the originating one.  If the
+ * "flood" argument is set, only send along the minimum spanning tree.
+ */
 static int
-flood(struct datapath *dp, struct sk_buff *skb)
+output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 {
+       u32 disable = flood ? OFPPC_NO_FLOOD : 0;
        struct net_bridge_port *p;
-       int prev_port;
+       int prev_port = -1;
 
-       prev_port = -1;
        list_for_each_entry_rcu (p, &dp->port_list, node) {
-               if (!should_deliver(p, skb))
+               if (skb->dev == p->dev || p->config & disable)
                        continue;
                if (prev_port != -1) {
                        struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
@@ -495,12 +539,12 @@ flood(struct datapath *dp, struct sk_buff *skb)
                                kfree_skb(skb);
                                return -ENOMEM;
                        }
-                       dp_output_port(dp, clone, prev_port); 
+                       dp_output_port(dp, clone, prev_port, 0); 
                }
                prev_port = p->port_no;
        }
        if (prev_port != -1)
-               dp_output_port(dp, skb, prev_port);
+               dp_output_port(dp, skb, prev_port, 0);
        else
                kfree_skb(skb);
 
@@ -509,60 +553,102 @@ flood(struct datapath *dp, struct sk_buff *skb)
 
 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
    FIXME: how are devices reference counted? */
-int dp_set_origin(struct datapath *dp, uint16_t in_port,
+void dp_set_origin(struct datapath *dp, uint16_t in_port,
                           struct sk_buff *skb)
 {
-       if (in_port < OFPP_MAX && dp->ports[in_port]) {
-               skb->dev = dp->ports[in_port]->dev;
-               return 0;
+       struct net_bridge_port *p;
+       p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
+            : in_port == OFPP_LOCAL ? dp->local_port
+            : NULL);
+       if (p) 
+               skb->dev = p->dev;
+        else 
+               skb->dev = NULL;
+}
+
+int 
+dp_xmit_skb(struct sk_buff *skb)
+{
+       int len = skb->len;
+       if (packet_length(skb) > skb->dev->mtu) {
+               printk("dropped over-mtu packet: %d > %d\n",
+                          packet_length(skb), skb->dev->mtu);
+               kfree_skb(skb);
+               return -E2BIG;
        }
-       return -ENOENT;
+
+       dev_queue_xmit(skb);
+
+       return len;
 }
 
 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
  */
-int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
+int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
+                  int ignore_no_fwd)
 {
-       struct net_bridge_port *p;
-       int len = skb->len;
-
        BUG_ON(!skb);
-       if (out_port == OFPP_FLOOD)
-               return flood(dp, skb);
-       else if (out_port == OFPP_CONTROLLER)
+       switch (out_port){
+       case OFPP_IN_PORT:
+               /* Send it out the port it came in on, which is already set in
+                * the skb. */
+               if (!skb->dev) {
+                       if (net_ratelimit())
+                               printk("skb device not set forwarding to in_port\n");
+                       kfree_skb(skb);
+                       return -ESRCH;
+               }
+               return dp_xmit_skb(skb);
+               
+       case OFPP_TABLE: {
+               int retval = run_flow_through_tables(dp->chain, skb,
+                                                    skb->dev->br_port);
+               if (retval)
+                       kfree_skb(skb);
+               return retval;
+       }
+
+       case OFPP_FLOOD:
+               return output_all(dp, skb, 1);
+
+       case OFPP_ALL:
+               return output_all(dp, skb, 0);
+
+       case OFPP_CONTROLLER:
                return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
                                                  OFPR_ACTION);
-       else if (out_port == OFPP_TABLE) {
-               struct sw_flow_key key;
-               struct sw_flow *flow;
-
-               flow_extract(skb, skb->dev->br_port->port_no, &key);
-               flow = chain_lookup(dp->chain, &key);
-               if (likely(flow != NULL)) {
-                       flow_used(flow, skb);
-                       execute_actions(dp, skb, &key, flow->actions, flow->n_actions);
+
+       case OFPP_LOCAL: {
+               struct net_device *dev = dp->netdev;
+#ifdef SUPPORT_SNAT
+               snat_local_in(skb);
+#endif
+               return dev ? dp_dev_recv(dev, skb) : -ESRCH;
+       }
+
+       case 0 ... DP_MAX_PORTS - 1: {
+               struct net_bridge_port *p = dp->ports[out_port];
+               if (p == NULL)
+                       goto bad_port;
+               if (p->dev == skb->dev) {
+                       /* To send to the input port, must use OFPP_IN_PORT */
+                       kfree_skb(skb);
+                       if (net_ratelimit())
+                               printk("can't directly forward to input port\n");
+                       return -EINVAL;
+               }
+               if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
+                       kfree_skb(skb);
                        return 0;
                }
-               return -ESRCH;
-       } else if (out_port >= OFPP_MAX)
-               goto bad_port;
+               skb->dev = p->dev; 
+               return dp_xmit_skb(skb);
+       }
 
-       p = dp->ports[out_port];
-       if (p == NULL)
+       default:
                goto bad_port;
-
-       skb->dev = p->dev;
-       if (packet_length(skb) > skb->dev->mtu) {
-               printk("dropped over-mtu packet: %d > %d\n",
-                                       packet_length(skb), skb->dev->mtu);
-               kfree_skb(skb);
-               return -E2BIG;
        }
 
-       dev_queue_xmit(skb);
-
-       return len;
-
 bad_port:
        kfree_skb(skb);
        if (net_ratelimit())
@@ -593,55 +679,127 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb,
 
        opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
        opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
+       if (!opi) {
+               err = -ENOMEM;
+               goto out;
+       }
        opi->buffer_id      = htonl(buffer_id);
        opi->total_len      = htons(skb->len);
-       opi->in_port        = htons(skb->dev->br_port->port_no);
+       opi->in_port        = htons(skb->dev && skb->dev->br_port
+                                   ? skb->dev->br_port->port_no
+                                   : OFPP_LOCAL);
        opi->reason         = reason;
        opi->pad            = 0;
-       memcpy(opi->data, skb_mac_header(skb), fwd_len);
+       skb_copy_bits(skb, 0, opi->data, fwd_len);
        err = send_openflow_skb(f_skb, NULL);
 
+out:
        kfree_skb(skb);
-
        return err;
 }
 
 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 {
+       unsigned long flags;
        desc->port_no = htons(p->port_no);
        strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
        desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
        memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
-       desc->flags = htonl(p->flags);
-       desc->features = 0;
-       desc->speed = 0;
+       desc->curr = 0;
+       desc->supported = 0;
+       desc->advertised = 0;
+       desc->peer = 0;
+
+       spin_lock_irqsave(&p->lock, flags);
+       desc->config = htonl(p->config);
+       desc->state = htonl(p->state);
+       spin_unlock_irqrestore(&p->lock, flags);
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
        if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
                struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 
                if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
+                       /* Set the supported features */
                        if (ecmd.supported & SUPPORTED_10baseT_Half) 
-                               desc->features |= OFPPF_10MB_HD;
+                               desc->supported |= OFPPF_10MB_HD;
                        if (ecmd.supported & SUPPORTED_10baseT_Full)
-                               desc->features |= OFPPF_10MB_FD;
+                               desc->supported |= OFPPF_10MB_FD;
                        if (ecmd.supported & SUPPORTED_100baseT_Half) 
-                               desc->features |= OFPPF_100MB_HD;
+                               desc->supported |= OFPPF_100MB_HD;
                        if (ecmd.supported & SUPPORTED_100baseT_Full)
-                               desc->features |= OFPPF_100MB_FD;
+                               desc->supported |= OFPPF_100MB_FD;
                        if (ecmd.supported & SUPPORTED_1000baseT_Half)
-                               desc->features |= OFPPF_1GB_HD;
+                               desc->supported |= OFPPF_1GB_HD;
                        if (ecmd.supported & SUPPORTED_1000baseT_Full)
-                               desc->features |= OFPPF_1GB_FD;
-                       /* 10Gbps half-duplex doesn't exist... */
+                               desc->supported |= OFPPF_1GB_FD;
                        if (ecmd.supported & SUPPORTED_10000baseT_Full)
-                               desc->features |= OFPPF_10GB_FD;
-
-                       desc->features = htonl(desc->features);
-                       desc->speed = htonl(ecmd.speed);
+                               desc->supported |= OFPPF_10GB_FD;
+                       if (ecmd.supported & SUPPORTED_TP)
+                               desc->supported |= OFPPF_COPPER;
+                       if (ecmd.supported & SUPPORTED_FIBRE)
+                               desc->supported |= OFPPF_FIBER;
+                       if (ecmd.supported & SUPPORTED_Autoneg)
+                               desc->supported |= OFPPF_AUTONEG;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
+                       if (ecmd.supported & SUPPORTED_Pause)
+                               desc->supported |= OFPPF_PAUSE;
+                       if (ecmd.supported & SUPPORTED_Asym_Pause)
+                               desc->supported |= OFPPF_PAUSE_ASYM;
+#endif /* kernel >= 2.6.14 */
+
+                       /* Set the advertised features */
+                       if (ecmd.advertising & ADVERTISED_10baseT_Half) 
+                               desc->advertised |= OFPPF_10MB_HD;
+                       if (ecmd.advertising & ADVERTISED_10baseT_Full)
+                               desc->advertised |= OFPPF_10MB_FD;
+                       if (ecmd.advertising & ADVERTISED_100baseT_Half) 
+                               desc->advertised |= OFPPF_100MB_HD;
+                       if (ecmd.advertising & ADVERTISED_100baseT_Full)
+                               desc->advertised |= OFPPF_100MB_FD;
+                       if (ecmd.advertising & ADVERTISED_1000baseT_Half)
+                               desc->advertised |= OFPPF_1GB_HD;
+                       if (ecmd.advertising & ADVERTISED_1000baseT_Full)
+                               desc->advertised |= OFPPF_1GB_FD;
+                       if (ecmd.advertising & ADVERTISED_10000baseT_Full)
+                               desc->advertised |= OFPPF_10GB_FD;
+                       if (ecmd.advertising & ADVERTISED_TP)
+                               desc->advertised |= OFPPF_COPPER;
+                       if (ecmd.advertising & ADVERTISED_FIBRE)
+                               desc->advertised |= OFPPF_FIBER;
+                       if (ecmd.advertising & ADVERTISED_Autoneg)
+                               desc->advertised |= OFPPF_AUTONEG;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
+                       if (ecmd.advertising & ADVERTISED_Pause)
+                               desc->advertised |= OFPPF_PAUSE;
+                       if (ecmd.advertising & ADVERTISED_Asym_Pause)
+                               desc->advertised |= OFPPF_PAUSE_ASYM;
+#endif /* kernel >= 2.6.14 */
+
+                       /* Set the current features */
+                       if (ecmd.speed == SPEED_10)
+                               desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
+                       else if (ecmd.speed == SPEED_100)
+                               desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
+                       else if (ecmd.speed == SPEED_1000)
+                               desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
+                       else if (ecmd.speed == SPEED_10000)
+                               desc->curr = OFPPF_10GB_FD;
+
+                       if (ecmd.port == PORT_TP) 
+                               desc->curr |= OFPPF_COPPER;
+                       else if (ecmd.port == PORT_FIBRE) 
+                               desc->curr |= OFPPF_FIBER;
+
+                       if (ecmd.autoneg)
+                               desc->curr |= OFPPF_AUTONEG;
                }
        }
 #endif
+       desc->curr = htonl(desc->curr);
+       desc->supported = htonl(desc->supported);
+       desc->advertised = htonl(desc->advertised);
+       desc->peer = htonl(desc->peer);
 }
 
 static int 
@@ -650,16 +808,13 @@ fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
        struct net_bridge_port *p;
        int port_count = 0;
 
-       ofr->datapath_id    = cpu_to_be64(dp->id); 
+       ofr->datapath_id  = cpu_to_be64(dp->id); 
 
-       ofr->n_exact        = htonl(2 * TABLE_HASH_MAX_FLOWS);
-       ofr->n_mac_only     = htonl(TABLE_MAC_MAX_FLOWS);
-       ofr->n_compression  = 0;                                           /* Not supported */
-       ofr->n_general      = htonl(TABLE_LINEAR_MAX_FLOWS);
-       ofr->buffer_mb      = htonl(UINT32_MAX);
-       ofr->n_buffers      = htonl(N_PKT_BUFFERS);
-       ofr->capabilities   = htonl(OFP_SUPPORTED_CAPABILITIES);
-       ofr->actions        = htonl(OFP_SUPPORTED_ACTIONS);
+       ofr->n_buffers    = htonl(N_PKT_BUFFERS);
+       ofr->n_tables     = dp->chain->n_tables;
+       ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
+       ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
+       memset(ofr->pad, 0, sizeof ofr->pad);
 
        list_for_each_entry_rcu (p, &dp->port_list, node) {
                fill_port_desc(p, &ofr->ports[port_count]);
@@ -678,7 +833,7 @@ dp_send_features_reply(struct datapath *dp, const struct sender *sender)
        int port_count;
 
        /* Overallocate. */
-       port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
+       port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
        ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
                                 OFPT_FEATURES_REPLY, sender, &skb);
        if (!ofr)
@@ -699,35 +854,134 @@ dp_send_config_reply(struct datapath *dp, const struct sender *sender)
        struct sk_buff *skb;
        struct ofp_switch_config *osc;
 
-       osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_PORT_STATUS, sender,
+       osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
                                 &skb);
        if (!osc)
                return -ENOMEM;
-       memcpy(((char *)osc) + sizeof osc->header,
-              ((char *)&dp->config) + sizeof dp->config.header,
-              sizeof dp->config - sizeof dp->config.header);
+
+       osc->flags = htons(dp->flags);
+       osc->miss_send_len = htons(dp->miss_send_len);
+
        return send_openflow_skb(skb, sender);
 }
 
 int
-dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
+dp_send_hello(struct datapath *dp, const struct sender *sender,
+             const struct ofp_header *request)
 {
-       struct net_bridge_port *p;
+       if (request->version < OFP_VERSION) {
+               char err[64];
+               sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
+               dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
+                                 OFPHFC_INCOMPATIBLE, err, strlen(err));
+               return -EINVAL;
+       } else {
+               struct sk_buff *skb;
+               struct ofp_header *reply;
 
-       p = dp->ports[htons(opp->port_no)];
+               reply = alloc_openflow_skb(dp, sizeof *reply,
+                                          OFPT_HELLO, sender, &skb);
+               if (!reply)
+                       return -ENOMEM;
+
+               return send_openflow_skb(skb, sender);
+       }
+}
+
+/* Callback function for a workqueue to disable an interface */
+static void
+down_port_cb(struct work_struct *work)
+{
+       struct net_bridge_port *p = container_of(work, struct net_bridge_port, 
+                       port_task);
+
+       rtnl_lock();
+       if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
+               if (net_ratelimit())
+                       printk("problem bringing up port %s\n", p->dev->name);
+       rtnl_unlock();
+       p->config |= OFPPC_PORT_DOWN;
+}
+
+/* Callback function for a workqueue to enable an interface */
+static void
+up_port_cb(struct work_struct *work)
+{
+       struct net_bridge_port *p = container_of(work, struct net_bridge_port, 
+                       port_task);
+
+       rtnl_lock();
+       if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
+               if (net_ratelimit())
+                       printk("problem bringing down port %s\n", p->dev->name);
+       rtnl_unlock();
+       p->config &= ~OFPPC_PORT_DOWN;
+}
+
+int
+dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
+{
+       unsigned long int flags;
+       int port_no = ntohs(opm->port_no);
+       struct net_bridge_port *p;
+       p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
+            : port_no == OFPP_LOCAL ? dp->local_port
+            : NULL);
 
        /* Make sure the port id hasn't changed since this was sent */
-       if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN) != 0) 
+       if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
                return -1;
-       
-       p->flags = htonl(opp->flags);
+
+       spin_lock_irqsave(&p->lock, flags);
+       if (opm->mask) {
+               uint32_t config_mask = ntohl(opm->mask);
+               p->config &= ~config_mask;
+               p->config |= ntohl(opm->config) & config_mask;
+       }
+
+       /* Modifying the status of an interface requires taking a lock
+        * that cannot be done from here.  For this reason, we use a shared 
+        * workqueue, which will cause it to be executed from a safer 
+        * context. */
+       if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
+               if ((opm->config & htonl(OFPPC_PORT_DOWN))
+                   && (p->config & OFPPC_PORT_DOWN) == 0) {
+                       PREPARE_WORK(&p->port_task, down_port_cb);
+                       schedule_work(&p->port_task);
+               } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
+                          && (p->config & OFPPC_PORT_DOWN)) {
+                       PREPARE_WORK(&p->port_task, up_port_cb);
+                       schedule_work(&p->port_task);
+               }
+       }
+       spin_unlock_irqrestore(&p->lock, flags);
 
        return 0;
 }
 
+/* Initialize the port status field of the bridge port. */
+static void
+init_port_status(struct net_bridge_port *p)
+{
+       unsigned long int flags;
 
-static int
-send_port_status(struct net_bridge_port *p, uint8_t status)
+       spin_lock_irqsave(&p->lock, flags);
+
+       if (p->dev->flags & IFF_UP) 
+               p->config &= ~OFPPC_PORT_DOWN;
+       else
+               p->config |= OFPPC_PORT_DOWN;
+
+       if (netif_carrier_ok(p->dev))
+               p->state &= ~OFPPS_LINK_DOWN;
+       else
+               p->state |= OFPPS_LINK_DOWN;
+
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+int
+dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 {
        struct sk_buff *skb;
        struct ofp_port_status *ops;
@@ -737,173 +991,74 @@ send_port_status(struct net_bridge_port *p, uint8_t status)
        if (!ops)
                return -ENOMEM;
        ops->reason = status;
+       memset(ops->pad, 0, sizeof ops->pad);
        fill_port_desc(p, &ops->desc);
 
        return send_openflow_skb(skb, NULL);
 }
 
 int 
-dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow)
+dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
+                    enum ofp_flow_expired_reason reason)
 {
        struct sk_buff *skb;
        struct ofp_flow_expired *ofe;
-       unsigned long duration_j;
+
+       if (!(dp->flags & OFPC_SEND_FLOW_EXP))
+               return 0;
 
        ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
        if (!ofe)
                return -ENOMEM;
 
        flow_fill_match(&ofe->match, &flow->key);
-       duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
-       ofe->duration   = htonl(duration_j / HZ);
-       ofe->packet_count   = cpu_to_be64(flow->packet_count);
-       ofe->byte_count     = cpu_to_be64(flow->byte_count);
-       return send_openflow_skb(skb, NULL);
-}
 
-static void
-fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
-               int table_idx)
-{
-       ofs->match.wildcards = htons(flow->key.wildcards);
-       ofs->match.in_port   = flow->key.in_port;
-       memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
-       memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
-       ofs->match.dl_vlan   = flow->key.dl_vlan;
-       ofs->match.dl_type   = flow->key.dl_type;
-       ofs->match.nw_src    = flow->key.nw_src;
-       ofs->match.nw_dst    = flow->key.nw_dst;
-       ofs->match.nw_proto  = flow->key.nw_proto;
-       memset(ofs->match.pad, 0, sizeof ofs->match.pad);
-       ofs->match.tp_src    = flow->key.tp_src;
-       ofs->match.tp_dst    = flow->key.tp_dst;
-       ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
-       ofs->table_id        = table_idx;
-       ofs->packet_count    = cpu_to_be64(flow->packet_count);
-       ofs->byte_count      = cpu_to_be64(flow->byte_count);
-}
+       ofe->priority = htons(flow->priority);
+       ofe->reason = reason;
+       memset(ofe->pad, 0, sizeof ofe->pad);
 
-int
-dp_send_flow_stats(struct datapath *dp, const struct sender *sender,
-                  const struct ofp_match *match)
-{
-       struct sk_buff *skb;
-       struct ofp_flow_stat_reply *fsr;
-       size_t header_size, fudge, flow_size;
-       struct sw_flow_key match_key;
-       int table_idx, n_flows, max_flows;
-
-       header_size = offsetof(struct ofp_flow_stat_reply, flows);
-       fudge = 128;
-       flow_size = sizeof fsr->flows[0];
-       max_flows = (NLMSG_GOODSIZE - header_size - fudge) / flow_size;
-       fsr = alloc_openflow_skb(dp, header_size + max_flows * flow_size,
-                                OFPT_FLOW_STAT_REPLY, sender, &skb);
-       if (!fsr)
-               return -ENOMEM;
+       ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
+       memset(ofe->pad2, 0, sizeof ofe->pad2);
+       ofe->packet_count = cpu_to_be64(flow->packet_count);
+       ofe->byte_count   = cpu_to_be64(flow->byte_count);
 
-       n_flows = 0;
-       flow_extract_match(&match_key, match);
-       for (table_idx = 0; table_idx < dp->chain->n_tables; table_idx++) {
-               struct sw_table *table = dp->chain->tables[table_idx];
-               struct swt_iterator iter;
-
-               if (n_flows >= max_flows) {
-                       break;
-               }
-
-               if (!table->iterator(table, &iter)) {
-                       if (net_ratelimit())
-                               printk("iterator failed for table %d\n",
-                                      table_idx);
-                       continue;
-               }
-
-               for (; iter.flow; table->iterator_next(&iter)) {
-                       if (flow_matches(&match_key, &iter.flow->key)) {
-                               fill_flow_stats(&fsr->flows[n_flows],
-                                               iter.flow, table_idx);
-                               if (++n_flows >= max_flows) {
-                                       break;
-                               }
-                       }
-               }
-               table->iterator_destroy(&iter);
-       }
-       resize_openflow_skb(skb, &fsr->header,
-                           header_size + flow_size * n_flows);
-       return send_openflow_skb(skb, sender);
-}
-
-static int 
-fill_port_stat_reply(struct datapath *dp, struct ofp_port_stat_reply *psr)
-{
-       struct net_bridge_port *p;
-       int port_count = 0;
-
-       list_for_each_entry_rcu (p, &dp->port_list, node) {
-               struct ofp_port_stats *ps = &psr->ports[port_count++];
-               struct net_device_stats *stats = p->dev->get_stats(p->dev);
-               ps->port_no = htons(p->port_no);
-               memset(ps->pad, 0, sizeof ps->pad);
-               ps->rx_count = cpu_to_be64(stats->rx_packets);
-               ps->tx_count = cpu_to_be64(stats->tx_packets);
-               ps->drop_count = cpu_to_be64(stats->rx_dropped
-                                            + stats->tx_dropped);
-       }
-
-       return port_count;
+       return send_openflow_skb(skb, NULL);
 }
+EXPORT_SYMBOL(dp_send_flow_expired);
 
 int
-dp_send_port_stats(struct datapath *dp, const struct sender *sender)
+dp_send_error_msg(struct datapath *dp, const struct sender *sender, 
+               uint16_t type, uint16_t code, const void *data, size_t len)
 {
        struct sk_buff *skb;
-       struct ofp_port_stat_reply *psr;
-       size_t psr_len, port_max_len;
-       int port_count;
+       struct ofp_error_msg *oem;
 
-       /* Overallocate. */
-       port_max_len = sizeof(struct ofp_port_stats) * OFPP_MAX;
-       psr = alloc_openflow_skb(dp, sizeof *psr + port_max_len,
-                                OFPT_PORT_STAT_REPLY, sender, &skb);
-       if (!psr)
+
+       oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR, 
+                       sender, &skb);
+       if (!oem)
                return -ENOMEM;
 
-       /* Fill. */
-       port_count = fill_port_stat_reply(dp, psr);
+       oem->type = htons(type);
+       oem->code = htons(code);
+       memcpy(oem->data, data, len);
 
-       /* Shrink to fit. */
-       psr_len = sizeof *psr + sizeof(struct ofp_port_stats) * port_count;
-       resize_openflow_skb(skb, &psr->header, psr_len);
        return send_openflow_skb(skb, sender);
 }
 
 int
-dp_send_table_stats(struct datapath *dp, const struct sender *sender)
+dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
+                  const struct ofp_header *rq)
 {
        struct sk_buff *skb;
-       struct ofp_table_stat_reply *tsr;
-       int i, n_tables;
-
-       n_tables = dp->chain->n_tables;
-       tsr = alloc_openflow_skb(dp, (offsetof(struct ofp_table_stat_reply,
-                                              tables)
-                                     + sizeof tsr->tables[0] * n_tables),
-                                OFPT_TABLE_STAT_REPLY, sender, &skb);
-       if (!tsr)
+       struct ofp_header *reply;
+
+       reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
+                                  sender, &skb);
+       if (!reply)
                return -ENOMEM;
-       for (i = 0; i < n_tables; i++) {
-               struct ofp_table_stats *ots = &tsr->tables[i];
-               struct sw_table_stats stats;
-               dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
-               strncpy(ots->name, stats.name, sizeof ots->name);
-               ots->table_id = i;
-               ots->pad[0] = ots->pad[1] = 0;
-               ots->max_entries = htonl(stats.max_flows);
-               ots->active_count = htonl(stats.n_flows);
-               ots->matched_count = cpu_to_be64(0); /* FIXME */
-       }
+
+       memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
        return send_openflow_skb(skb, sender);
 }
 
@@ -960,7 +1115,6 @@ static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
        if (!info->attrs[DP_GENL_A_DP_IDX])
                return -EINVAL;
 
-       mutex_lock(&dp_mutex);
        dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
        if (!dp)
                err = -ENOENT;
@@ -968,7 +1122,6 @@ static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
                del_dp(dp);
                err = 0;
        }
-       mutex_unlock(&dp_mutex);
        return err;
 }
 
@@ -1007,7 +1160,7 @@ static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
                err = -ENOENT;
        else {
                void *data;
-               ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+               ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
                if (!ans_skb) {
                        err = -ENOMEM;
                        goto err;
@@ -1052,7 +1205,6 @@ static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
                return -EINVAL;
 
        /* Get datapath. */
-       mutex_lock(&dp_mutex);
        dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
        if (!dp) {
                err = -ENOENT;
@@ -1075,13 +1227,12 @@ static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
                        err = -ENOENT;
                        goto out_put;
                }
-               err = del_switch_port(port->br_port);
+               err = dp_del_switch_port(port->br_port);
        }
 
 out_put:
        dev_put(port);
 out:
-       mutex_unlock(&dp_mutex);
        return err;
 }
 
@@ -1112,26 +1263,22 @@ static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
        if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
                return -EINVAL;
 
-       rcu_read_lock();
        dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
-       if (!dp) {
-               err = -ENOENT;
-               goto out;
-       }
+       if (!dp)
+               return -ENOENT;
 
-       if (nla_len(va) < sizeof(struct ofp_header)) {
-               err = -EINVAL;
-               goto out;
-       }
+       if (nla_len(va) < sizeof(struct ofp_header))
+               return -EINVAL;
        oh = nla_data(va);
 
        sender.xid = oh->xid;
        sender.pid = info->snd_pid;
        sender.seq = info->snd_seq;
-       err = fwd_control_input(dp->chain, &sender, nla_data(va), nla_len(va));
 
-out:
-       rcu_read_unlock();
+       mutex_lock(&dp_mutex);
+       err = fwd_control_input(dp->chain, &sender,
+                               nla_data(va), nla_len(va));
+       mutex_unlock(&dp_mutex);
        return err;
 }
 
@@ -1139,26 +1286,465 @@ static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
        [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
 };
 
-static struct genl_ops dp_genl_ops_openflow = {
-       .cmd = DP_GENL_C_OPENFLOW,
-       .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
-       .policy = dp_genl_openflow_policy,
-       .doit = dp_genl_openflow,
-       .dumpit = NULL,
+static int desc_stats_dump(struct datapath *dp, void *state,
+                           void *body, int *body_len)
+{
+       struct ofp_desc_stats *ods = body;
+       int n_bytes = sizeof *ods;
+
+       if (n_bytes > *body_len) {
+               return -ENOBUFS;
+       }
+       *body_len = n_bytes;
+
+       strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
+       strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
+       strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
+       strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
+
+       return 0;
+}
+
+struct flow_stats_state {
+       int table_idx;
+       struct sw_table_position position;
+       const struct ofp_flow_stats_request *rq;
+
+       void *body;
+       int bytes_used, bytes_allocated;
 };
 
-static struct nla_policy dp_genl_benchmark_policy[DP_GENL_A_MAX + 1] = {
-       [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
-       [DP_GENL_A_NPACKETS] = { .type = NLA_U32 },
-       [DP_GENL_A_PSIZE] = { .type = NLA_U32 },
+static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
+                          void **state)
+{
+       const struct ofp_flow_stats_request *fsr = body;
+       struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
+       if (!s)
+               return -ENOMEM;
+       s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
+       memset(&s->position, 0, sizeof s->position);
+       s->rq = fsr;
+       *state = s;
+       return 0;
+}
+
+static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+       struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
+       struct flow_stats_state *s = private;
+       struct ofp_flow_stats *ofs;
+       int length;
+
+       length = sizeof *ofs + sf_acts->actions_len;
+       if (length + s->bytes_used > s->bytes_allocated)
+               return 1;
+
+       ofs = s->body + s->bytes_used;
+       ofs->length          = htons(length);
+       ofs->table_id        = s->table_idx;
+       ofs->pad             = 0;
+       ofs->match.wildcards = htonl(flow->key.wildcards);
+       ofs->match.in_port   = flow->key.in_port;
+       memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
+       memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
+       ofs->match.dl_vlan   = flow->key.dl_vlan;
+       ofs->match.dl_type   = flow->key.dl_type;
+       ofs->match.nw_src    = flow->key.nw_src;
+       ofs->match.nw_dst    = flow->key.nw_dst;
+       ofs->match.nw_proto  = flow->key.nw_proto;
+       ofs->match.pad       = 0;
+       ofs->match.tp_src    = flow->key.tp_src;
+       ofs->match.tp_dst    = flow->key.tp_dst;
+       ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
+       ofs->priority        = htons(flow->priority);
+       ofs->idle_timeout    = htons(flow->idle_timeout);
+       ofs->hard_timeout    = htons(flow->hard_timeout);
+       memset(ofs->pad2, 0, sizeof ofs->pad2);
+       ofs->packet_count    = cpu_to_be64(flow->packet_count);
+       ofs->byte_count      = cpu_to_be64(flow->byte_count);
+       memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
+
+       s->bytes_used += length;
+       return 0;
+}
+
+static int flow_stats_dump(struct datapath *dp, void *state,
+                          void *body, int *body_len)
+{
+       struct flow_stats_state *s = state;
+       struct sw_flow_key match_key;
+       int error = 0;
+
+       s->bytes_used = 0;
+       s->bytes_allocated = *body_len;
+       s->body = body;
+
+       flow_extract_match(&match_key, &s->rq->match);
+       while (s->table_idx < dp->chain->n_tables
+              && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
+       {
+               struct sw_table *table = dp->chain->tables[s->table_idx];
+
+               error = table->iterate(table, &match_key, &s->position,
+                                      flow_stats_dump_callback, s);
+               if (error)
+                       break;
+
+               s->table_idx++;
+               memset(&s->position, 0, sizeof s->position);
+       }
+       *body_len = s->bytes_used;
+
+       /* If error is 0, we're done.
+        * Otherwise, if some bytes were used, there are more flows to come.
+        * Otherwise, we were not able to fit even a single flow in the body,
+        * which indicates that we have a single flow with too many actions to
+        * fit.  We won't ever make any progress at that rate, so give up. */
+       return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
+}
+
+static void flow_stats_done(void *state)
+{
+       kfree(state);
+}
+
+static int aggregate_stats_init(struct datapath *dp,
+                               const void *body, int body_len,
+                               void **state)
+{
+       *state = (void *)body;
+       return 0;
+}
+
+static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+       struct ofp_aggregate_stats_reply *rpy = private;
+       rpy->packet_count += flow->packet_count;
+       rpy->byte_count += flow->byte_count;
+       rpy->flow_count++;
+       return 0;
+}
+
+static int aggregate_stats_dump(struct datapath *dp, void *state,
+                               void *body, int *body_len)
+{
+       struct ofp_aggregate_stats_request *rq = state;
+       struct ofp_aggregate_stats_reply *rpy;
+       struct sw_table_position position;
+       struct sw_flow_key match_key;
+       int table_idx;
+
+       if (*body_len < sizeof *rpy)
+               return -ENOBUFS;
+       rpy = body;
+       *body_len = sizeof *rpy;
+
+       memset(rpy, 0, sizeof *rpy);
+
+       flow_extract_match(&match_key, &rq->match);
+       table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
+       memset(&position, 0, sizeof position);
+       while (table_idx < dp->chain->n_tables
+              && (rq->table_id == 0xff || rq->table_id == table_idx))
+       {
+               struct sw_table *table = dp->chain->tables[table_idx];
+               int error;
+
+               error = table->iterate(table, &match_key, &position,
+                                      aggregate_stats_dump_callback, rpy);
+               if (error)
+                       return error;
+
+               table_idx++;
+               memset(&position, 0, sizeof position);
+       }
+
+       rpy->packet_count = cpu_to_be64(rpy->packet_count);
+       rpy->byte_count = cpu_to_be64(rpy->byte_count);
+       rpy->flow_count = htonl(rpy->flow_count);
+       return 0;
+}
+
+static int table_stats_dump(struct datapath *dp, void *state,
+                           void *body, int *body_len)
+{
+       struct ofp_table_stats *ots;
+       int n_bytes = dp->chain->n_tables * sizeof *ots;
+       int i;
+       if (n_bytes > *body_len)
+               return -ENOBUFS;
+       *body_len = n_bytes;
+       for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
+               struct sw_table_stats stats;
+               dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
+               strncpy(ots->name, stats.name, sizeof ots->name);
+               ots->table_id = i;
+               ots->wildcards = htonl(stats.wildcards);
+               memset(ots->pad, 0, sizeof ots->pad);
+               ots->max_entries = htonl(stats.max_flows);
+               ots->active_count = htonl(stats.n_flows);
+               ots->lookup_count = cpu_to_be64(stats.n_lookup);
+               ots->matched_count = cpu_to_be64(stats.n_matched);
+       }
+       return 0;
+}
+
+struct port_stats_state {
+       int port;
+};
+
+static int port_stats_init(struct datapath *dp, const void *body, int body_len,
+                          void **state)
+{
+       struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
+       if (!s)
+               return -ENOMEM;
+       s->port = 0;
+       *state = s;
+       return 0;
+}
+
+static int port_stats_dump(struct datapath *dp, void *state,
+                          void *body, int *body_len)
+{
+       struct port_stats_state *s = state;
+       struct ofp_port_stats *ops;
+       int n_ports, max_ports;
+       int i;
+
+       max_ports = *body_len / sizeof *ops;
+       if (!max_ports)
+               return -ENOMEM;
+       ops = body;
+
+       n_ports = 0;
+       for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
+               struct net_bridge_port *p = dp->ports[i];
+               struct net_device_stats *stats;
+               if (!p)
+                       continue;
+               stats = p->dev->get_stats(p->dev);
+               ops->port_no = htons(p->port_no);
+               memset(ops->pad, 0, sizeof ops->pad);
+               ops->rx_packets   = cpu_to_be64(stats->rx_packets);
+               ops->tx_packets   = cpu_to_be64(stats->tx_packets);
+               ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
+               ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
+               ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
+               ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
+               ops->rx_errors    = cpu_to_be64(stats->rx_errors);
+               ops->tx_errors    = cpu_to_be64(stats->tx_errors);
+               ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
+               ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
+               ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
+               ops->collisions   = cpu_to_be64(stats->collisions);
+               n_ports++;
+               ops++;
+       }
+       s->port = i;
+       *body_len = n_ports * sizeof *ops;
+       return n_ports >= max_ports;
+}
+
+static void port_stats_done(void *state)
+{
+       kfree(state);
+}
+
+struct stats_type {
+       /* Minimum and maximum acceptable number of bytes in body member of
+        * struct ofp_stats_request. */
+       size_t min_body, max_body;
+
+       /* Prepares to dump some kind of statistics on 'dp'.  'body' and
+        * 'body_len' are the 'body' member of the struct ofp_stats_request.
+        * Returns zero if successful, otherwise a negative error code.
+        * May initialize '*state' to state information.  May be null if no
+        * initialization is required.*/
+       int (*init)(struct datapath *dp, const void *body, int body_len,
+                   void **state);
+
+       /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
+        * modifies '*body_len' to reflect the number of bytes actually used.
+        * ('body' will be transmitted as the 'body' member of struct
+        * ofp_stats_reply.) */
+       int (*dump)(struct datapath *dp, void *state,
+                   void *body, int *body_len);
+
+       /* Cleans any state created by the init or dump functions.  May be null
+        * if no cleanup is required. */
+       void (*done)(void *state);
+};
+
+static const struct stats_type stats[] = {
+       [OFPST_DESC] = {
+               0,
+               0,
+               NULL,
+               desc_stats_dump,
+               NULL
+       },
+       [OFPST_FLOW] = {
+               sizeof(struct ofp_flow_stats_request),
+               sizeof(struct ofp_flow_stats_request),
+               flow_stats_init,
+               flow_stats_dump,
+               flow_stats_done
+       },
+       [OFPST_AGGREGATE] = {
+               sizeof(struct ofp_aggregate_stats_request),
+               sizeof(struct ofp_aggregate_stats_request),
+               aggregate_stats_init,
+               aggregate_stats_dump,
+               NULL
+       },
+       [OFPST_TABLE] = {
+               0,
+               0,
+               NULL,
+               table_stats_dump,
+               NULL
+       },
+       [OFPST_PORT] = {
+               0,
+               0,
+               port_stats_init,
+               port_stats_dump,
+               port_stats_done
+       },
 };
 
-static struct genl_ops dp_genl_ops_benchmark_nl = {
-       .cmd = DP_GENL_C_BENCHMARK_NL,
+static int
+dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
+{
+       struct datapath *dp;
+       struct sender sender;
+       const struct stats_type *s;
+       struct ofp_stats_reply *osr;
+       int dp_idx;
+       int max_openflow_len, body_len;
+       void *body;
+       int err;
+
+       /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
+        * support setting up cleanup functions via the .doneit member of
+        * struct genl_ops.  This kluge supports earlier versions also. */
+       cb->done = dp_genl_openflow_done;
+
+       sender.pid = NETLINK_CB(cb->skb).pid;
+       sender.seq = cb->nlh->nlmsg_seq;
+       if (!cb->args[0]) {
+               struct nlattr *attrs[DP_GENL_A_MAX + 1];
+               struct ofp_stats_request *rq;
+               struct nlattr *va;
+               size_t len, body_len;
+               int type;
+
+               err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
+                                 dp_genl_openflow_policy);
+               if (err < 0)
+                       return err;
+
+               if (!attrs[DP_GENL_A_DP_IDX])
+                       return -EINVAL;
+               dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
+               dp = dp_get(dp_idx);
+               if (!dp)
+                       return -ENOENT;
+
+               va = attrs[DP_GENL_A_OPENFLOW];
+               len = nla_len(va);
+               if (!va || len < sizeof *rq)
+                       return -EINVAL;
+
+               rq = nla_data(va);
+               sender.xid = rq->header.xid;
+               type = ntohs(rq->type);
+               if (rq->header.version != OFP_VERSION) {
+                       dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
+                                         OFPBRC_BAD_VERSION, rq, len);
+                       return -EINVAL;
+               }
+               if (rq->header.type != OFPT_STATS_REQUEST
+                   || ntohs(rq->header.length) != len)
+                       return -EINVAL;
+
+               if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
+                       dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
+                                         OFPBRC_BAD_STAT, rq, len);
+                       return -EINVAL;
+               }
+
+               s = &stats[type];
+               body_len = len - offsetof(struct ofp_stats_request, body);
+               if (body_len < s->min_body || body_len > s->max_body)
+                       return -EINVAL;
+
+               cb->args[0] = 1;
+               cb->args[1] = dp_idx;
+               cb->args[2] = type;
+               cb->args[3] = rq->header.xid;
+               if (s->init) {
+                       void *state;
+                       err = s->init(dp, rq->body, body_len, &state);
+                       if (err)
+                               return err;
+                       cb->args[4] = (long) state;
+               }
+       } else if (cb->args[0] == 1) {
+               sender.xid = cb->args[3];
+               dp_idx = cb->args[1];
+               s = &stats[cb->args[2]];
+
+               dp = dp_get(dp_idx);
+               if (!dp)
+                       return -ENOENT;
+       } else {
+               return 0;
+       }
+
+       osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
+                                  &max_openflow_len);
+       if (IS_ERR(osr))
+               return PTR_ERR(osr);
+       osr->type = htons(s - stats);
+       osr->flags = 0;
+       resize_openflow_skb(skb, &osr->header, max_openflow_len);
+       body = osr->body;
+       body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
+
+       err = s->dump(dp, (void *) cb->args[4], body, &body_len);
+       if (err >= 0) {
+               if (!err)
+                       cb->args[0] = 2;
+               else
+                       osr->flags = ntohs(OFPSF_REPLY_MORE);
+               resize_openflow_skb(skb, &osr->header,
+                                   (offsetof(struct ofp_stats_reply, body)
+                                    + body_len));
+               err = skb->len;
+       }
+
+       return err;
+}
+
+static int
+dp_genl_openflow_done(struct netlink_callback *cb)
+{
+       if (cb->args[0]) {
+               const struct stats_type *s = &stats[cb->args[2]];
+               if (s->done)
+                       s->done((void *) cb->args[4]);
+       }
+       return 0;
+}
+
+static struct genl_ops dp_genl_ops_openflow = {
+       .cmd = DP_GENL_C_OPENFLOW,
        .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
-       .policy = dp_genl_benchmark_policy,
-       .doit = dp_genl_benchmark_nl,
-       .dumpit = NULL,
+       .policy = dp_genl_openflow_policy,
+       .doit = dp_genl_openflow,
+       .dumpit = dp_genl_openflow_dumpit,
 };
 
 static struct genl_ops *dp_genl_all_ops[] = {
@@ -1172,7 +1758,6 @@ static struct genl_ops *dp_genl_all_ops[] = {
        &dp_genl_ops_query_dp,
        &dp_genl_ops_add_port,
        &dp_genl_ops_del_port,
-       &dp_genl_ops_benchmark_nl,
 };
 
 static int dp_init_netlink(void)
@@ -1207,28 +1792,56 @@ static void dp_uninit_netlink(void)
        genl_unregister_family(&dp_genl_family);
 }
 
-#define DRV_NAME               "openflow"
-#define DRV_VERSION     VERSION
-#define DRV_DESCRIPTION "OpenFlow switching datapath implementation"
-#define DRV_COPYRIGHT   "Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University"
-
+/* Set the description strings if appropriate values are available from
+ * the DMI. */
+static void set_desc(void)
+{
+       const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
+       const char *uptr = uuid + 24;
+
+       if (!uuid || *uuid == '\0' || strlen(uuid) != 36) 
+               return;
+
+       /* We are only interested version 1 UUIDs, since the last six bytes
+        * are an IEEE 802 MAC address. */
+       if (uuid[14] != '1') 
+               return;
+
+       /* Only set if the UUID is from Nicira. */
+       if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
+               return;
+
+       strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
+       snprintf(hw_desc, sizeof(hw_desc), "%s %s", 
+                       dmi_get_system_info(DMI_PRODUCT_NAME), 
+                       dmi_get_system_info(DMI_PRODUCT_VERSION));
+       strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL), 
+                       sizeof(serial_num));
+}
 
 static int __init dp_init(void)
 {
        int err;
 
-       printk(KERN_INFO DRV_NAME ": " DRV_DESCRIPTION "\n");
-       printk(KERN_INFO DRV_NAME ": " VERSION" built on "__DATE__" "__TIME__"\n");
-       printk(KERN_INFO DRV_NAME ": " DRV_COPYRIGHT "\n");
+       printk("OpenFlow %s, built "__DATE__" "__TIME__", "
+              "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
 
        err = flow_init();
        if (err)
                goto error;
 
-       err = dp_init_netlink();
+       err = register_netdevice_notifier(&dp_device_notifier);
        if (err)
                goto error_flow_exit;
 
+       err = dp_init_netlink();
+       if (err)
+               goto error_unreg_notifier;
+
+       /* Check if better descriptions of the switch are available than the
+        * defaults. */
+       set_desc();
+
        /* Hook into callback used by the bridge to intercept packets.
         * Parasites we are. */
        if (br_handle_frame_hook)
@@ -1237,6 +1850,8 @@ static int __init dp_init(void)
 
        return 0;
 
+error_unreg_notifier:
+       unregister_netdevice_notifier(&dp_device_notifier);
 error_flow_exit:
        flow_exit();
 error:
@@ -1248,6 +1863,7 @@ static void dp_cleanup(void)
 {
        fwd_exit();
        dp_uninit_netlink();
+       unregister_netdevice_notifier(&dp_device_notifier);
        flow_exit();
        br_handle_frame_hook = NULL;
 }
@@ -1255,6 +1871,6 @@ static void dp_cleanup(void)
 module_init(dp_init);
 module_exit(dp_cleanup);
 
-MODULE_DESCRIPTION(DRV_DESCRIPTION);
-MODULE_AUTHOR(DRV_COPYRIGHT);
+MODULE_DESCRIPTION("OpenFlow switching datapath");
+MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
 MODULE_LICENSE("GPL");