X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fvport-netdev.c;h=e45e22fbb870e87fa0d974c413515673aefaaa9f;hb=3005302426764a2f701bb3507ad9602e3fe2dbb9;hp=8cc442107339656183cf9cbef3e16182891b1e4d;hpb=f4267e344a9373a4efefff8f8f5b85f532d223e1;p=openvswitch

diff --git a/datapath/vport-netdev.c b/datapath/vport-netdev.c
index 8cc44210..e45e22fb 100644
--- a/datapath/vport-netdev.c
+++ b/datapath/vport-netdev.c
@@ -1,5 +1,5 @@
 /*
- * 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
@@ -16,107 +16,64 @@
 
 #include <net/llc.h>
 
+#include "checksum.h"
 #include "datapath.h"
 #include "vport-internal_dev.h"
 #include "vport-netdev.h"
 
-#include "compat.h"
+/* 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))
 
-struct vport_ops netdev_vport_ops;
+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(2,6,27)
-static struct llc_sap *netdev_stp_sap;
-
-static int
-netdev_stp_rcv(struct sk_buff *skb, struct net_device *dev,
-		      struct packet_type *pt, struct net_device *orig_dev)
+#if 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)
 {
-	/* We don't really care about STP packets, we just listen for them for
-	 * mutual exclusion with the bridge module, so this just discards
-	 * them. */
-	kfree_skb(skb);
-	return 0;
-}
+	struct vport *vport;
 
-static int
-netdev_avoid_bridge_init(void)
-{
-	/* Register to receive STP packets because the bridge module also
-	 * attempts to do so.  Since there can only be a single listener for a
-	 * given protocol, this provides mutual exclusion against the bridge
-	 * module, preventing both of them from being loaded at the same
-	 * time. */
-	netdev_stp_sap = llc_sap_open(LLC_SAP_BSPAN, netdev_stp_rcv);
-	if (!netdev_stp_sap) {
-		printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n");
-		return -EADDRINUSE;
-	}
-	return 0;
-}
+	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
+		return skb;
 
-static void
-netdev_avoid_bridge_exit(void)
-{
-	llc_sap_put(netdev_stp_sap);
-}
-#else  /* Linux 2.6.27 or later. */
-static int
-netdev_avoid_bridge_init(void)
-{
-	/* Linux 2.6.27 introduces a way for multiple clients to register for
-	 * STP packets, which interferes with what we try to do above.
-	 * Instead, just check whether there's a bridge hook defined.  This is
-	 * not as safe--the bridge module is willing to load over the top of
-	 * us--but it provides a little bit of protection. */
-	if (br_handle_frame_hook) {
-		printk(KERN_ERR "openvswitch: bridge module is loaded, cannot load over it\n");
-		return -EADDRINUSE;
-	}
-	return 0;
-}
+	vport = netdev_get_vport(skb->dev);
 
-static void
-netdev_avoid_bridge_exit(void)
-{
-	/* Nothing to do. */
-}
-#endif	/* Linux 2.6.27 or later */
+	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)
+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)
+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
 
-static int
-netdev_init(void)
+#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)
 {
-	int err;
-
-	err = netdev_avoid_bridge_init();
-	if (err)
-		return err;
-
 	/* Hook into callback used by the bridge to intercept packets.
 	 * Parasites we are. */
 	br_handle_frame_hook = netdev_frame_hook;
@@ -124,21 +81,19 @@ netdev_init(void)
 	return 0;
 }
 
-static void
-netdev_exit(void)
+static void netdev_exit(void)
 {
 	br_handle_frame_hook = NULL;
-	netdev_avoid_bridge_exit();
 }
+#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;
@@ -146,7 +101,7 @@ 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;
@@ -159,11 +114,25 @@ netdev_create(const char *name, const void __user *config)
 		goto error_put;
 	}
 
-	if (netdev_vport->dev->br_port) {
-		err = -EBUSY;
-		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 rtnl_link_stats64 stats;
+
+		err = netdev_get_stats(vport, &stats);
+		if (!err)
+			vport_set_stats(vport, &stats);
 	}
 
+	err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
+					 vport);
+	if (err)
+		goto error_put;
+
+	dev_set_promiscuity(netdev_vport->dev, 1);
+	dev_disable_lro(netdev_vport->dev);
+	netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
+
 	return vport;
 
 error_put:
@@ -174,49 +143,29 @@ 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 int 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);
+	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();
 
-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);
-	dev_set_promiscuity(netdev_vport->dev, -1);
+	dev_put(netdev_vport->dev);
+	vport_free(vport);
 
 	return 0;
 }
 
-int
-netdev_set_mtu(struct vport *vport, int mtu)
+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);
 }
 
-int
-netdev_set_addr(struct vport *vport, const unsigned char *addr)
+int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 {
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	struct sockaddr sa;
@@ -227,117 +176,87 @@ netdev_set_addr(struct vport *vport, const unsigned char *addr)
 	return dev_set_mac_address(netdev_vport->dev, &sa);
 }
 
-const char *
-netdev_get_name(const struct vport *vport)
+const char *netdev_get_name(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netdev_vport->dev->name;
 }
 
-const unsigned char *
-netdev_get_addr(const struct vport *vport)
+const unsigned char *netdev_get_addr(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netdev_vport->dev->dev_addr;
 }
 
-struct kobject *
-netdev_get_kobj(const struct vport *vport)
+struct kobject *netdev_get_kobj(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 }
 
-int
-netdev_get_stats(const struct vport *vport, struct odp_vport_stats *stats)
+int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *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;
-
+	dev_get_stats(netdev_vport->dev, stats);
 	return 0;
 }
 
-unsigned
-netdev_get_dev_flags(const struct vport *vport)
+unsigned netdev_get_dev_flags(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return dev_get_flags(netdev_vport->dev);
 }
 
-int
-netdev_is_running(const struct vport *vport)
+int netdev_is_running(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netif_running(netdev_vport->dev);
 }
 
-unsigned char
-netdev_get_operstate(const struct vport *vport)
+unsigned char netdev_get_operstate(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netdev_vport->dev->operstate;
 }
 
-int
-netdev_get_ifindex(const struct vport *vport)
+int netdev_get_ifindex(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netdev_vport->dev->ifindex;
 }
 
-int
-netdev_get_iflink(const struct vport *vport)
+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)
+int netdev_get_mtu(const struct vport *vport)
 {
 	const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	return netdev_vport->dev->mtu;
 }
 
 /* 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;
-
 	/* 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)
+	if (unlikely(!skb))
 		return;
 
-	/* Push the Ethernet header back on. */
+	skb_warn_if_lro(skb);
+
 	skb_push(skb, ETH_HLEN);
-	skb_reset_mac_header(skb);
 	compute_ip_summed(skb, false);
 
 	vport_receive(vport, skb);
 }
 
-static int
-netdev_send(struct vport *vport, struct sk_buff *skb)
+static int netdev_send(struct vport *vport, struct sk_buff *skb)
 {
 	struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 	int len = skb->len;
@@ -350,21 +269,30 @@ netdev_send(struct vport *vport, struct sk_buff *skb)
 }
 
 /* Returns null if this device is not attached to a datapath. */
-struct vport *
-netdev_get_vport(struct net_device *dev)
+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,
+const struct vport_ops netdev_vport_ops = {
+	.type		= ODP_VPORT_TYPE_NETDEV,
+	.flags          = (VPORT_F_REQUIRED |
+			  (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 	.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,
@@ -379,3 +307,22 @@ struct vport_ops netdev_vport_ops = {
 	.get_mtu	= netdev_get_mtu,
 	.send		= netdev_send,
 };
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
+/*
+ * 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
+ * load both modules at the same time (e.g. "bridge: exports duplicate symbol
+ * br_should_route_hook (owned by openvswitch_mod)").
+ *
+ * The use of "typeof" here avoids the need to track changes in the type of
+ * br_should_route_hook over various kernel versions.
+ */
+typeof(br_should_route_hook) br_should_route_hook;
+EXPORT_SYMBOL(br_should_route_hook);
+#endif