X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fvport-netdev.c;h=4c7f1cb43692354d66cc863f1491026b2d9fa167;hb=f613a0d72c521ca3a4eeb2c29ac523f6fdf72667;hp=1773a09b98ce6418062b3220fd9b1d741be71300;hpb=67c74f7515867fb8b3a4d23af98dacd3d547ebdd;p=sliver-openvswitch.git

diff --git a/datapath/vport-netdev.c b/datapath/vport-netdev.c
index 1773a09b9..4c7f1cb43 100644
--- a/datapath/vport-netdev.c
+++ b/datapath/vport-netdev.c
@@ -1,11 +1,13 @@
 /*
- * Copyright (c) 2010 Nicira Networks.
+ * Copyright (c) 2010, 2011 Nicira Networks.
  * Distributed under the terms of the GNU GPL version 2.
  *
  * Significant portions of this file may be copied from parts of the Linux
  * kernel, by Linus Torvalds and others.
  */
 
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/if_arp.h>
 #include <linux/if_bridge.h>
 #include <linux/if_vlan.h>
@@ -16,40 +18,87 @@
 
 #include <net/llc.h>
 
+#include "checksum.h"
 #include "datapath.h"
+#include "vlan.h"
 #include "vport-internal_dev.h"
 #include "vport-netdev.h"
 
-#include "compat.h"
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
+    !defined(HAVE_VLAN_BUG_WORKAROUND)
+#include <linux/module.h>
+
+static int vlan_tso __read_mostly = 0;
+module_param(vlan_tso, int, 0644);
+MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
+#else
+#define vlan_tso true
+#endif
 
-/* If the native device stats aren't 64 bit use the vport stats tracking instead. */
-#define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64))
+static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
 
-static void netdev_port_receive(struct net_bridge_port *, struct sk_buff *);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
+/* Called with rcu_read_lock and bottom-halves disabled. */
+static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
+{
+	struct sk_buff *skb = *pskb;
+	struct vport *vport;
 
+	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
+		return RX_HANDLER_PASS;
+
+	vport = netdev_get_vport(skb->dev);
+
+	netdev_port_receive(vport, skb);
+
+	return RX_HANDLER_CONSUMED;
+}
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
+/* Called with rcu_read_lock and bottom-halves disabled. */
+static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
+{
+	struct vport *vport;
+
+	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
+		return skb;
+
+	vport = netdev_get_vport(skb->dev);
+
+	netdev_port_receive(vport, skb);
+
+	return NULL;
+}
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 /*
  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  * different set of devices!)
  */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 /* Called with rcu_read_lock and bottom-halves disabled. */
 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
 					 struct sk_buff *skb)
 {
-	netdev_port_receive(p, skb);
+	netdev_port_receive((struct vport *)p, skb);
 	return NULL;
 }
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+/*
+ * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
+ * different set of devices!)
+ */
 /* Called with rcu_read_lock and bottom-halves disabled. */
 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 {
-	netdev_port_receive(p, *pskb);
+	netdev_port_receive((struct vport *)p, *pskb);
 	return 1;
 }
 #else
 #error
 #endif
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
+static int netdev_init(void) { return 0; }
+static void netdev_exit(void) { }
+#else
 static int netdev_init(void)
 {
 	/* Hook into callback used by the bridge to intercept packets.
@@ -63,14 +112,15 @@ static void netdev_exit(void)
 {
 	br_handle_frame_hook = NULL;
 }
+#endif
 
-static struct vport *netdev_create(const char *name, const void __user *config)
+static struct vport *netdev_create(const struct vport_parms *parms)
 {
 	struct vport *vport;
 	struct netdev_vport *netdev_vport;
 	int err;
 
-	vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops);
+	vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops, parms);
 	if (IS_ERR(vport)) {
 		err = PTR_ERR(vport);
 		goto error;
@@ -78,7 +128,7 @@ static struct vport *netdev_create(const char *name, const void __user *config)
 
 	netdev_vport = netdev_vport_priv(vport);
 
-	netdev_vport->dev = dev_get_by_name(&init_net, name);
+	netdev_vport->dev = dev_get_by_name(&init_net, parms->name);
 	if (!netdev_vport->dev) {
 		err = -ENODEV;
 		goto error_free_vport;
@@ -91,20 +141,16 @@ static struct vport *netdev_create(const char *name, const void __user *config)
 		goto error_put;
 	}
 
-	if (netdev_vport->dev->br_port) {
-		err = -EBUSY;
+	err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
+					 vport);
+	if (err)
 		goto error_put;
-	}
-
-	/* If we are using the vport stats layer initialize it to the current
-	 * values so we are roughly consistent with the device stats. */
-	if (USE_VPORT_STATS) {
-		struct odp_vport_stats stats;
 
-		err = netdev_get_stats(vport, &stats);
-		if (!err)
-			vport_set_stats(vport, &stats);
-	}
+	dev_set_promiscuity(netdev_vport->dev, 1);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
+	dev_disable_lro(netdev_vport->dev);
+#endif
+	netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 
 	return vport;
 
@@ -116,41 +162,18 @@ error:
 	return ERR_PTR(err);
 }
 
-static int netdev_destroy(struct vport *vport)
-{
-	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-
-	dev_put(netdev_vport->dev);
-	vport_free(vport);
-
-	return 0;
-}
-
-static int netdev_attach(struct vport *vport)
+static void netdev_destroy(struct vport *vport)
 {
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 
-	dev_set_promiscuity(netdev_vport->dev, 1);
-	dev_disable_lro(netdev_vport->dev);
-	rcu_assign_pointer(netdev_vport->dev->br_port, (struct net_bridge_port *)vport);
-
-	return 0;
-}
-
-static int netdev_detach(struct vport *vport)
-{
-	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-
-	rcu_assign_pointer(netdev_vport->dev->br_port, NULL);
+	netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
+	netdev_rx_handler_unregister(netdev_vport->dev);
 	dev_set_promiscuity(netdev_vport->dev, -1);
 
-	return 0;
-}
+	synchronize_rcu();
 
-int netdev_set_mtu(struct vport *vport, int mtu)
-{
-	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-	return dev_set_mtu(netdev_vport->dev, mtu);
+	dev_put(netdev_vport->dev);
+	vport_free(vport);
 }
 
 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
@@ -182,29 +205,6 @@ struct kobject *netdev_get_kobj(const struct vport *vport)
 	return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 }
 
-int netdev_get_stats(const struct vport *vport, struct odp_vport_stats *stats)
-{
-	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-	const struct net_device_stats *netdev_stats;
-
-	netdev_stats = dev_get_stats(netdev_vport->dev);
-
-	stats->rx_bytes		= netdev_stats->rx_bytes;
-	stats->rx_packets	= netdev_stats->rx_packets;
-	stats->tx_bytes		= netdev_stats->tx_bytes;
-	stats->tx_packets	= netdev_stats->tx_packets;
-	stats->rx_dropped	= netdev_stats->rx_dropped;
-	stats->rx_errors	= netdev_stats->rx_errors;
-	stats->rx_frame_err	= netdev_stats->rx_frame_errors;
-	stats->rx_over_err	= netdev_stats->rx_over_errors;
-	stats->rx_crc_err	= netdev_stats->rx_crc_errors;
-	stats->tx_dropped	= netdev_stats->tx_dropped;
-	stats->tx_errors	= netdev_stats->tx_errors;
-	stats->collisions	= netdev_stats->collisions;
-
-	return 0;
-}
-
 unsigned netdev_get_dev_flags(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
@@ -229,12 +229,6 @@ int netdev_get_ifindex(const struct vport *vport)
 	return netdev_vport->dev->ifindex;
 }
 
-int netdev_get_iflink(const struct vport *vport)
-{
-	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-	return netdev_vport->dev->iflink;
-}
-
 int netdev_get_mtu(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
@@ -242,9 +236,12 @@ int netdev_get_mtu(const struct vport *vport)
 }
 
 /* Must be called with rcu_read_lock. */
-static void netdev_port_receive(struct net_bridge_port *p, struct sk_buff *skb)
+static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 {
-	struct vport *vport = (struct vport *)p;
+	if (unlikely(!vport)) {
+		kfree_skb(skb);
+		return;
+	}
 
 	/* 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).
@@ -254,63 +251,168 @@ static void netdev_port_receive(struct net_bridge_port *p, struct sk_buff *skb)
 	if (unlikely(!skb))
 		return;
 
-	skb_warn_if_lro(skb);
-
-	/* Push the Ethernet header back on. */
 	skb_push(skb, ETH_HLEN);
-	skb_reset_mac_header(skb);
-	compute_ip_summed(skb, false);
+
+	if (unlikely(compute_ip_summed(skb, false))) {
+		kfree_skb(skb);
+		return;
+	}
+	vlan_copy_skb_tci(skb);
 
 	vport_receive(vport, skb);
 }
 
+static inline unsigned packet_length(const struct sk_buff *skb)
+{
+	unsigned length = skb->len - ETH_HLEN;
+
+	if (skb->protocol == htons(ETH_P_8021Q))
+		length -= VLAN_HLEN;
+
+	return length;
+}
+
+static bool dev_supports_vlan_tx(struct net_device *dev)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
+	/* Software fallback means every device supports vlan_tci on TX. */
+	return true;
+#elif defined(HAVE_VLAN_BUG_WORKAROUND)
+	return dev->features & NETIF_F_HW_VLAN_TX;
+#else
+	/* Assume that the driver is buggy. */
+	return false;
+#endif
+}
+
 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 {
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
-	int len = skb->len;
+	int mtu = netdev_vport->dev->mtu;
+	int len;
+
+	if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
+		if (net_ratelimit())
+			pr_warn("%s: dropped over-mtu packet: %d > %d\n",
+				dp_name(vport->dp), packet_length(skb), mtu);
+		goto error;
+	}
+
+	if (unlikely(skb_warn_if_lro(skb)))
+		goto error;
 
 	skb->dev = netdev_vport->dev;
-	forward_ip_summed(skb);
+	forward_ip_summed(skb, true);
+
+	if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
+		int features;
+
+		features = netif_skb_features(skb);
+
+		if (!vlan_tso)
+			features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
+				      NETIF_F_UFO | NETIF_F_FSO);
+
+		if (netif_needs_gso(skb, features)) {
+			struct sk_buff *nskb;
+
+			nskb = skb_gso_segment(skb, features);
+			if (!nskb) {
+				if (unlikely(skb_cloned(skb) &&
+				    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
+					kfree_skb(skb);
+					return 0;
+				}
+
+				skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
+				goto tag;
+			}
+
+			if (IS_ERR(nskb)) {
+				kfree_skb(skb);
+				return 0;
+			}
+			consume_skb(skb);
+			skb = nskb;
+
+			len = 0;
+			do {
+				nskb = skb->next;
+				skb->next = NULL;
+
+				skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
+				if (likely(skb)) {
+					len += skb->len;
+					vlan_set_tci(skb, 0);
+					dev_queue_xmit(skb);
+				}
+
+				skb = nskb;
+			} while (skb);
+
+			return len;
+		}
+
+tag:
+		skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
+		if (unlikely(!skb))
+			return 0;
+		vlan_set_tci(skb, 0);
+	}
+
+	len = skb->len;
 	dev_queue_xmit(skb);
 
 	return len;
+
+error:
+	kfree_skb(skb);
+	vport_record_error(vport, VPORT_E_TX_DROPPED);
+	return 0;
 }
 
 /* Returns null if this device is not attached to a datapath. */
 struct vport *netdev_get_vport(struct net_device *dev)
 {
-	return (struct vport *)dev->br_port;
+#ifdef IFF_BRIDGE_PORT
+#if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH
+	if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
+#else
+	if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))	
+#endif
+		return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
+	else
+		return NULL;
+#else
+	return (struct vport *)rcu_dereference_rtnl(dev->br_port);
+#endif
 }
 
-struct vport_ops netdev_vport_ops = {
-	.type		= "netdev",
-	.flags          = (VPORT_F_REQUIRED |
-			  (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
+const struct vport_ops netdev_vport_ops = {
+	.type		= OVS_VPORT_TYPE_NETDEV,
+	.flags          = VPORT_F_REQUIRED,
 	.init		= netdev_init,
 	.exit		= netdev_exit,
 	.create		= netdev_create,
 	.destroy	= netdev_destroy,
-	.attach		= netdev_attach,
-	.detach		= netdev_detach,
-	.set_mtu	= netdev_set_mtu,
 	.set_addr	= netdev_set_addr,
 	.get_name	= netdev_get_name,
 	.get_addr	= netdev_get_addr,
 	.get_kobj	= netdev_get_kobj,
-	.get_stats	= netdev_get_stats,
 	.get_dev_flags	= netdev_get_dev_flags,
 	.is_running	= netdev_is_running,
 	.get_operstate	= netdev_get_operstate,
 	.get_ifindex	= netdev_get_ifindex,
-	.get_iflink	= netdev_get_iflink,
 	.get_mtu	= netdev_get_mtu,
 	.send		= netdev_send,
 };
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 /*
- * Open vSwitch cannot safely coexist with the Linux bridge module on any
- * released version of Linux, because there is only a single bridge hook
- * function and only a single br_port member in struct net_device.
+ * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
+ * the Linux bridge module on any released version of Linux, because there
+ * is only a single bridge hook function and only a single br_port member
+ * in struct net_device.
  *
  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
  * module also exports the same symbol, so the module loader will refuse to
@@ -322,3 +424,4 @@ struct vport_ops netdev_vport_ops = {
  */
 typeof(br_should_route_hook) br_should_route_hook;
 EXPORT_SYMBOL(br_should_route_hook);
+#endif