Make the OFPP_LOCAL port work in the kernel OpenFlow implementation.

diff --git a/datapath/datapath.c b/datapath/datapath.c
index f2d8a67d4981721d684769554833d953ff508a82..4b7064998e6781c56e24a5fffcb32c84c192faf6 100644 (file)
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -33,6 +33,7 @@
 #include "datapath.h"
 #include "table.h"
 #include "chain.h"
+#include "dp_dev.h"
 #include "forward.h"
 #include "flow.h"
 #include "datapath_t.h"
@@ -60,8 +61,6 @@ struct net_bridge_port {
 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
 
@@ -78,6 +77,9 @@ static DEFINE_MUTEX(dp_mutex);
 static int dp_maint_func(void *data);
 static int send_port_status(struct net_bridge_port *p, uint8_t status);
 static int dp_genl_openflow_done(struct netlink_callback *);
+static struct net_bridge_port *new_nbp(struct datapath *,
+                                      struct net_device *, int port_no);
+static int del_switch_port(struct net_bridge_port *);
 
 /* nla_shrink - reduce amount of space reserved by nla_reserve
  * @skb: socket buffer from which to recover room
@@ -270,35 +272,42 @@ 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->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->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);
 
        return 0;
 
+err_destroy_local_port:
+       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:
@@ -318,23 +327,29 @@ static int find_portno(struct datapath *dp)
 }
 
 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;
+       if (port_no != OFPP_LOCAL)
+               rcu_assign_pointer(dev->br_port, p);
+       if (port_no < OFPP_MAX)
+               rcu_assign_pointer(dp->ports[port_no], p); 
+       list_add_rcu(&p->node, &dp->port_list);
 
        return p;
 }
@@ -343,26 +358,20 @@ static struct net_bridge_port *new_nbp(struct datapath *dp,
 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);
-
        /* Notify the ctlpath that this port has been added */
        send_port_status(p, OFPPR_ADD);
 
@@ -396,19 +405,13 @@ static int del_switch_port(struct net_bridge_port *p)
 /* 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
+       struct net_bridge_port *p;
 
+       dp_dev_destroy(dp);
        kthread_stop(dp->dp_task);
 
        /* Drop references to DP. */
-       list_for_each_entry_safe (p, n, &dp->port_list, node)
+       list_for_each_entry_rcu (p, &dp->port_list, node)
                del_switch_port(p);
        rcu_assign_pointer(dps[dp->dp_idx], NULL);
 
@@ -479,21 +482,6 @@ 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.  If we're flooding,
- * then don't send out ports with flooding disabled.
- */
-static inline int should_deliver(const struct net_bridge_port *p,
-                       const struct sk_buff *skb, int flood)
-{
-       if (skb->dev == p->dev)
-               return 0;
-
-       if (flood && (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;
@@ -508,12 +496,12 @@ static inline unsigned packet_length(const struct sk_buff *skb)
 static int
 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 {
+       u32 disable = flood ? BRIDGE_PORT_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, flood))
+               if (skb->dev == p->dev || p->flags & disable)
                        continue;
                if (prev_port != -1) {
                        struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
@@ -538,8 +526,11 @@ output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 int 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;
+       struct net_bridge_port *p = (in_port < OFPP_MAX ? dp->ports[in_port]
+                                    : in_port == OFPP_LOCAL ? dp->local_port
+                                    : NULL);
+       if (p) {
+               skb->dev = p->dev;
                return 0;
        }
        return -ENOENT;
@@ -549,9 +540,6 @@ int dp_set_origin(struct datapath *dp, uint16_t in_port,
  */
 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
 {
-       struct net_bridge_port *p;
-       int len = skb->len;
-
        BUG_ON(!skb);
        if (out_port == OFPP_FLOOD)
                return output_all(dp, skb, 1);
@@ -561,10 +549,11 @@ int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
                return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
                                                  OFPR_ACTION);
        else if (out_port == OFPP_TABLE) {
+               struct net_bridge_port *p = skb->dev->br_port;
                struct sw_flow_key key;
                struct sw_flow *flow;
 
-               flow_extract(skb, skb->dev->br_port->port_no, &key);
+               flow_extract(skb, p ? p->port_no : OFPP_LOCAL, &key);
                flow = chain_lookup(dp->chain, &key);
                if (likely(flow != NULL)) {
                        flow_used(flow, skb);
@@ -572,25 +561,27 @@ int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
                        return 0;
                }
                return -ESRCH;
-       } else if (out_port >= OFPP_MAX)
-               goto bad_port;
+       } else if (out_port == OFPP_LOCAL) {
+               struct net_device *dev = dp->netdev;
+               return dev ? dp_dev_recv(dev, skb) : -ESRCH;
+       } else if (out_port >= 0 && out_port < OFPP_MAX) {
+               struct net_bridge_port *p = dp->ports[out_port];
+               int len = skb->len;
+               if (p == NULL)
+                       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;
+               }
 
-       p = dp->ports[out_port];
-       if (p == NULL)
-               goto bad_port;
+               dev_queue_xmit(skb);
 
-       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;
+               return len;
        }
 
-       dev_queue_xmit(skb);
-
-       return len;
-
 bad_port:
        kfree_skb(skb);
        if (net_ratelimit())
@@ -612,6 +603,7 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb,
         * forward the whole packet? */
        struct sk_buff *f_skb;
        struct ofp_packet_in *opi;
+       struct net_bridge_port *p;
        size_t fwd_len, opi_len;
        int err;
 
@@ -627,7 +619,8 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb,
        }
        opi->buffer_id      = htonl(buffer_id);
        opi->total_len      = htons(skb->len);
-       opi->in_port        = htons(skb->dev->br_port->port_no);
+       p = skb->dev->br_port;
+       opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
        opi->reason         = reason;
        opi->pad            = 0;
        memcpy(opi->data, skb_mac_header(skb), fwd_len);
@@ -744,16 +737,14 @@ dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 int
 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
 {
-       struct net_bridge_port *p;
-
-       p = dp->ports[htons(opp->port_no)];
-
+       int port_no = ntohs(opp->port_no);
+       struct net_bridge_port *p = (port_no < OFPP_MAX ? 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(opp->hw_addr, p->dev->dev_addr, ETH_ALEN))
                return -1;
-       
        p->flags = htonl(opp->flags);
-
        return 0;
 }
 

diff --git a/datapath/datapath.h b/datapath/datapath.h
index 1a884f8340958893c587e0a0ed2240d280d77b98..b342cbb5910cfd24deb3bcbec1378dc9affa8790 100644 (file)
--- a/datapath/datapath.h
+++ b/datapath/datapath.h
@@ -45,8 +45,7 @@ struct datapath {
        struct task_struct *dp_task; /* Kernel thread for maintenance. */
 
        /* Data related to the "of" device of this datapath */
-       struct net_device dev;
-       struct net_device_stats stats;
+       struct net_device *netdev;
 
        /* Configuration set from controller */
        uint16_t flags;
@@ -54,7 +53,8 @@ struct datapath {
 
        /* Switch ports. */
        struct net_bridge_port *ports[OFPP_MAX];
-       struct list_head port_list; /* List of ports, for flooding. */
+       struct net_bridge_port *local_port; /* OFPP_LOCAL port. */
+       struct list_head port_list; /* All ports, including local_port. */
 };
 
 /* Information necessary to reply to the sender of an OpenFlow message. */

diff --git a/datapath/dp_dev.c b/datapath/dp_dev.c
index e2ebf5a42912fe2589f347ab29e7c66b510513b6..4047e926794b7f5d68305d67e5f0821487375e2f 100644 (file)
--- a/datapath/dp_dev.c
+++ b/datapath/dp_dev.c
@@ -6,73 +6,129 @@
 #include "datapath.h"
 #include "forward.h"
 
-static int dp_dev_do_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+struct dp_dev {
+       struct net_device_stats stats;
+       struct datapath *dp;
+};
+
+static struct dp_dev *dp_dev_priv(struct net_device *netdev) 
 {
-       printk("xxx_do_ioctl called\n");
-       return 0;
+       return netdev_priv(netdev);
 }
 
-static struct net_device_stats *dp_dev_get_stats(struct net_device *dev)
+static struct net_device_stats *dp_dev_get_stats(struct net_device *netdev)
 {
-       struct datapath *dp = netdev_priv(dev);
-       return &dp->stats;
+       struct dp_dev *dp_dev = dp_dev_priv(netdev);
+       return &dp_dev->stats;
 }
 
-int dp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
+int dp_dev_recv(struct net_device *netdev, struct sk_buff *skb) 
 {
-       struct datapath *dp = netdev_priv(dev);
-
-       printk("xxx dp_dev_xmit not implemented yet!\n");
-       return 0;
+       int len = skb->len;
+       struct dp_dev *dp_dev = dp_dev_priv(netdev);
+       skb->pkt_type = PACKET_HOST;
+       skb->protocol = eth_type_trans(skb, netdev);
+       netif_rx(skb);
+       netdev->last_rx = jiffies;
+       dp_dev->stats.rx_packets++;
+       dp_dev->stats.rx_bytes += len;
+       return len;
+}
 
-       printk("xxx_xmit called send to dp_frame_hook\n");
+static int dp_dev_mac_addr(struct net_device *dev, void *p)
+{
+       struct sockaddr *addr = p;
 
-       rcu_read_lock();    /* xxx Only for 2.4 kernels? */
-       fwd_port_input(dp->chain, skb, OFPP_LOCAL);
-       rcu_read_unlock();  /* xxx Only for 2.4 kernels? */
-       
+       if (netif_running(dev))
+               return -EBUSY;
+       if (!is_valid_ether_addr(addr->sa_data))
+               return -EADDRNOTAVAIL;
+       memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
        return 0;
 }
 
-static int dp_dev_open(struct net_device *dev)
+static int dp_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
 {
-       netif_start_queue(dev);
+       struct dp_dev *dp_dev = dp_dev_priv(netdev);
+       struct datapath *dp;
+       rcu_read_lock();
+       dp = dp_dev->dp;
+       if (likely(dp != NULL)) {
+               dp_dev->stats.tx_packets++;
+               dp_dev->stats.tx_bytes += skb->len;
+               skb_reset_mac_header(skb);
+               fwd_port_input(dp->chain, skb, OFPP_LOCAL);
+       } else {
+               dp_dev->stats.tx_dropped++;
+               kfree_skb(skb);
+       }
+       rcu_read_unlock();
        return 0;
 }
 
-static void dp_dev_set_multicast_list(struct net_device *dev)
+static int dp_dev_open(struct net_device *netdev)
 {
-       printk("xxx_set_multi called\n");
+       netif_start_queue(netdev);
+       return 0;
 }
 
-static int dp_dev_stop(struct net_device *dev)
+static int dp_dev_stop(struct net_device *netdev)
 {
-       netif_stop_queue(dev);
+       netif_stop_queue(netdev);
        return 0;
 }
 
-int dp_dev_setup(struct net_device *dev)
+static void
+do_setup(struct net_device *netdev)
 {
+       ether_setup(netdev);
+
+       netdev->get_stats = dp_dev_get_stats;
+       netdev->hard_start_xmit = dp_dev_xmit;
+       netdev->open = dp_dev_open;
+       netdev->stop = dp_dev_stop;
+       netdev->tx_queue_len = 0;
+       netdev->set_mac_address = dp_dev_mac_addr;
+
+       netdev->flags = IFF_BROADCAST | IFF_MULTICAST;
+
+       random_ether_addr(netdev->dev_addr);
+}
+
+
+int dp_dev_setup(struct datapath *dp)
+{
+       struct dp_dev *dp_dev;
+       struct net_device *netdev;
+       char of_name[8];
        int err;
 
-       strncpy(dev->name, "of%d", IFNAMSIZ);
-       err = dev_alloc_name(dev, dev->name);
-       if (err < 0) 
-               return err;
+       snprintf(of_name, sizeof of_name, "of%d", dp->dp_idx);
+       netdev = alloc_netdev(sizeof(struct dp_dev), of_name, do_setup);
+       if (!netdev)
+               return -ENOMEM;
 
-       dev->do_ioctl = dp_dev_do_ioctl;
-       dev->get_stats = dp_dev_get_stats;
-       dev->hard_start_xmit = dp_dev_xmit;
-       dev->open = dp_dev_open;
-       dev->set_multicast_list = dp_dev_set_multicast_list;
-       dev->stop = dp_dev_stop;
-       dev->tx_queue_len = 0;
-       dev->set_mac_address = NULL;
+       err = register_netdev(netdev);
+       if (err) {
+               free_netdev(netdev);
+               return err;
+       }
 
-       dev->flags = IFF_BROADCAST | IFF_NOARP | IFF_MULTICAST;
+       dp_dev = dp_dev_priv(netdev);
+       dp_dev->dp = dp;
+       dp->netdev = netdev;
+       return 0;
+}
 
-       random_ether_addr(dev->dev_addr);
+void dp_dev_destroy(struct datapath *dp)
+{
+       struct dp_dev *dp_dev = dp_dev_priv(dp->netdev);
+       dp_dev->dp = NULL;
+       synchronize_net();
+       unregister_netdev(dp->netdev);
+}
 
-       ether_setup(dev);
-       return register_netdevice(dev);
+int is_dp_dev(struct net_device *netdev) 
+{
+       return netdev->open == dp_dev_open;
 }

diff --git a/datapath/dp_dev.h b/datapath/dp_dev.h
new file mode 100644 (file)
index 0000000..2d4453b
--- /dev/null
+++ b/datapath/dp_dev.h
@@ -0,0 +1,9 @@
+#ifndef DP_DEV_H
+#define DP_DEV_H 1
+
+int dp_dev_setup(struct datapath *);
+void dp_dev_destroy(struct datapath *);
+int dp_dev_recv(struct net_device *, struct sk_buff *);
+int is_dp_dev(struct net_device *);
+
+#endif /* dp_dev.h */

diff --git a/datapath/linux-2.4/compat-2.4/include/linux/netdevice.h b/datapath/linux-2.4/compat-2.4/include/linux/netdevice.h
index 6eba6513f6e8585f7055cd9b54fcc3ace4a6f9ab..60c81b0cc6f4bd0de2adbfc0e6e070e560d49214 100644 (file)
--- a/datapath/linux-2.4/compat-2.4/include/linux/netdevice.h
+++ b/datapath/linux-2.4/compat-2.4/include/linux/netdevice.h
@@ -2,6 +2,8 @@
 #define __LINUX_NETDEVICE_WRAPPER_H 1
 
 #include_next <linux/netdevice.h>
+#include <linux/kernel.h>
+#include <linux/rcupdate.h>
 
 /*----------------------------------------------------------------------------
  * In 2.6.24, a namespace argument became required for dev_get_by_name. 
@@ -28,4 +30,10 @@ static inline void *netdev_priv(struct net_device *dev)
 }
 #endif
 
+/* Synchronize with packet receive processing. */
+static inline void synchronize_net(void) 
+{
+       synchronize_rcu();
+}
+
 #endif

diff --git a/include/openflow.h b/include/openflow.h
index 2f5da5d87185d00dcfe8620b508c5f85945ef92d..6303b490070e837490cbda078c3b0db8f11e91da 100644 (file)
--- a/include/openflow.h
+++ b/include/openflow.h
@@ -75,7 +75,7 @@ enum ofp_port {
                                   those disabled by STP. */
     OFPP_ALL        = 0xfffc,  /* All physical ports except input port. */
     OFPP_CONTROLLER = 0xfffd,  /* Send to controller. */ 
-    OFPP_LOCAL      = 0xfffe,  /* Local openflow "port". */ /* xxx Want?! */
+    OFPP_LOCAL      = 0xfffe,  /* Local openflow "port". */
     OFPP_NONE       = 0xffff   /* Not associated with a physical port. */
 };