For SNAT, don't store the pre-fragment L2 header before actions are applied.

[sliver-openvswitch.git] / datapath / dp_dev.c

diff --git a/datapath/dp_dev.c b/datapath/dp_dev.c
index 02468eb..ec36361 100644 (file)
--- a/datapath/dp_dev.c
+++ b/datapath/dp_dev.c
@@ -3,6 +3,8 @@
 #include <linux/etherdevice.h>
 #include <linux/rcupdate.h>
 #include <linux/skbuff.h>
+#include <linux/workqueue.h>
+#include <linux/dmi.h>
 
 #include "datapath.h"
 #include "forward.h"
@@ -10,8 +12,11 @@
 struct dp_dev {
        struct net_device_stats stats;
        struct datapath *dp;
+       struct sk_buff_head xmit_queue;
+       struct work_struct xmit_work;
 };
 
+
 static struct dp_dev *dp_dev_priv(struct net_device *netdev) 
 {
        return netdev_priv(netdev);
@@ -57,17 +62,54 @@ static int dp_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
        struct dp_dev *dp_dev = dp_dev_priv(netdev);
        struct datapath *dp = dp_dev->dp;
 
+       /* By orphaning 'skb' we will screw up socket accounting slightly, but
+        * the effect is limited to the device queue length.  If we don't
+        * do this, then the sk_buff will be destructed eventually, but it is
+        * harder to predict when. */
+       skb_orphan(skb);
+
+       /* We are going to modify 'skb', by sticking it on &dp_dev->xmit_queue,
+        * so we need to have our own clone.  (At any rate, fwd_port_input()
+        * will need its own clone, so there's no benefit to queuing any other
+        * way.) */
+       skb = skb_share_check(skb, GFP_ATOMIC);
+       if (!skb)
+               return 0;
+
        dp_dev->stats.tx_packets++;
        dp_dev->stats.tx_bytes += skb->len;
 
-       skb_reset_mac_header(skb);
-       rcu_read_lock();
-       fwd_port_input(dp->chain, skb, OFPP_LOCAL);
-       rcu_read_unlock();
+       if (skb_queue_len(&dp_dev->xmit_queue) >= dp->netdev->tx_queue_len) {
+               /* Queue overflow.  Stop transmitter. */
+               netif_stop_queue(dp->netdev);
+
+               /* We won't see all dropped packets individually, so overrun
+                * error is appropriate. */
+               dp_dev->stats.tx_fifo_errors++;
+       }
+       skb_queue_tail(&dp_dev->xmit_queue, skb);
+       dp->netdev->trans_start = jiffies;
+
+       schedule_work(&dp_dev->xmit_work);
 
        return 0;
 }
 
+static void dp_dev_do_xmit(struct work_struct *work)
+{
+       struct dp_dev *dp_dev = container_of(work, struct dp_dev, xmit_work);
+       struct datapath *dp = dp_dev->dp;
+       struct sk_buff *skb;
+
+       while ((skb = skb_dequeue(&dp_dev->xmit_queue)) != NULL) {
+               skb_reset_mac_header(skb);
+               rcu_read_lock();
+               fwd_port_input(dp->chain, skb, dp->local_port);
+               rcu_read_unlock();
+       }
+       netif_wake_queue(dp->netdev);
+}
+
 static int dp_dev_open(struct net_device *netdev)
 {
        netif_start_queue(netdev);
@@ -80,6 +122,45 @@ static int dp_dev_stop(struct net_device *netdev)
        return 0;
 }
 
+/* Check if the DMI UUID contains a Nicira mac address that should be 
+ * used for this interface.  The UUID is assumed to be RFC 4122
+ * compliant. */
+static void
+set_uuid_mac(struct net_device *netdev)
+{
+       const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
+       const char *uptr;
+       uint8_t mac[ETH_ALEN];
+       int i;
+
+       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;
+
+       /* Pull out the embedded MAC address.  The kernel's sscanf doesn't
+        * support field widths on hex digits, so we use this hack. */
+       uptr = uuid + 24;
+       for (i=0; i<ETH_ALEN; i++) {
+               unsigned char d[3];
+               
+               d[0] = *uptr++;
+               d[1] = *uptr++;
+               d[2] = '\0';
+               
+               mac[i] = simple_strtoul(d, NULL, 16);
+       }
+
+       /* If this is a Nicira one, then use it. */
+       if (mac[0] != 0x00 || mac[1] != 0x23 || mac[2] != 0x20) 
+               return;
+
+       memcpy(netdev->dev_addr, mac, ETH_ALEN);
+}
+
 static void
 do_setup(struct net_device *netdev)
 {
@@ -89,12 +170,20 @@ do_setup(struct net_device *netdev)
        netdev->hard_start_xmit = dp_dev_xmit;
        netdev->open = dp_dev_open;
        netdev->stop = dp_dev_stop;
-       netdev->tx_queue_len = 0;
+       netdev->tx_queue_len = 100;
        netdev->set_mac_address = dp_dev_mac_addr;
 
        netdev->flags = IFF_BROADCAST | IFF_MULTICAST;
 
        random_ether_addr(netdev->dev_addr);
+
+       /* Set the OUI to the Nicira one. */
+       netdev->dev_addr[0] = 0x00;
+       netdev->dev_addr[1] = 0x23;
+       netdev->dev_addr[2] = 0x20;
+
+       /* Set the top bits to indicate random Nicira address. */
+       netdev->dev_addr[3] |= 0xc0;
 }
 
 
@@ -116,16 +205,26 @@ int dp_dev_setup(struct datapath *dp)
                return err;
        }
 
+       /* For "of0", we check the DMI UUID to see if a Nicira mac address
+        * is available to use instead of the random one just generated. */
+       if (dp->dp_idx == 0) 
+               set_uuid_mac(netdev);
+
        dp_dev = dp_dev_priv(netdev);
        dp_dev->dp = dp;
+       skb_queue_head_init(&dp_dev->xmit_queue);
+       INIT_WORK(&dp_dev->xmit_work, dp_dev_do_xmit);
        dp->netdev = netdev;
        return 0;
 }
 
 void dp_dev_destroy(struct datapath *dp)
 {
+       struct dp_dev *dp_dev = dp_dev_priv(dp->netdev);
+
        netif_tx_disable(dp->netdev);
        synchronize_net();
+       skb_queue_purge(&dp_dev->xmit_queue);
        unregister_netdev(dp->netdev);
        free_netdev(dp->netdev);
 }