X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=e927b2469515938fe9f5e08f5c1bfa5debf32076;hb=cb5087caddd187e71d5f32118d8d807c9c131725;hp=3edba7ce5589a6d047311fea9a9c7592a8566c3c;hpb=a5e54d9b6f8002f34cc792df69e6eda68cf95223;p=openvswitch

diff --git a/datapath/datapath.c b/datapath/datapath.c
index 3edba7ce..e927b246 100644
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -55,22 +55,8 @@
 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
 EXPORT_SYMBOL(dp_ioctl_hook);
 
-int (*dp_add_dp_hook)(struct datapath *dp);
-EXPORT_SYMBOL(dp_add_dp_hook);
-
-int (*dp_del_dp_hook)(struct datapath *dp);
-EXPORT_SYMBOL(dp_del_dp_hook);
-
-int (*dp_add_if_hook)(struct net_bridge_port *p);
-EXPORT_SYMBOL(dp_add_if_hook);
-
-int (*dp_del_if_hook)(struct net_bridge_port *p);
-EXPORT_SYMBOL(dp_del_if_hook);
-
 /* 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.
+ * by dp_mutex.
  *
  * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
  * lock first.
@@ -184,6 +170,16 @@ errout:
 		rtnl_set_sk_err(net, RTNLGRP_LINK, err);
 }
 
+static void release_dp(struct kobject *kobj)
+{
+	struct datapath *dp = container_of(kobj, struct datapath, ifobj);
+	kfree(dp);
+}
+
+struct kobj_type dp_ktype = {
+	.release = release_dp
+};
+
 static int create_dp(int dp_idx, const char __user *devnamep)
 {
 	struct net_device *dp_dev;
@@ -225,6 +221,11 @@ static int create_dp(int dp_idx, const char __user *devnamep)
 		skb_queue_head_init(&dp->queues[i]);
 	init_waitqueue_head(&dp->waitqueue);
 
+	/* Initialize kobject for bridge.  This will be added as
+	 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
+	dp->ifobj.kset = NULL;
+	kobject_init(&dp->ifobj, &dp_ktype);
+
 	/* Allocate table. */
 	err = -ENOMEM;
 	rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE));
@@ -252,8 +253,7 @@ static int create_dp(int dp_idx, const char __user *devnamep)
 	mutex_unlock(&dp_mutex);
 	rtnl_unlock();
 
-	if (dp_add_dp_hook)
-		dp_add_dp_hook(dp);
+	dp_sysfs_add_dp(dp);
 
 	return 0;
 
@@ -281,8 +281,7 @@ static void do_destroy_dp(struct datapath *dp)
 		if (p->port_no != ODPP_LOCAL)
 			dp_del_port(p);
 
-	if (dp_del_dp_hook)
-		dp_del_dp_hook(dp);
+	dp_sysfs_del_dp(dp);
 
 	rcu_assign_pointer(dps[dp->dp_idx], NULL);
 
@@ -295,7 +294,7 @@ static void do_destroy_dp(struct datapath *dp)
 	for (i = 0; i < DP_MAX_GROUPS; i++)
 		kfree(dp->groups[i]);
 	free_percpu(dp->stats_percpu);
-	kfree(dp);
+	kobject_put(&dp->ifobj);
 	module_put(THIS_MODULE);
 }
 
@@ -320,6 +319,19 @@ err_unlock:
 	return err;
 }
 
+static void release_nbp(struct kobject *kobj)
+{
+	struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj);
+	kfree(p);
+}
+
+struct kobj_type brport_ktype = {
+#ifdef CONFIG_SYSFS
+	.sysfs_ops = &brport_sysfs_ops,
+#endif
+	.release = release_nbp
+};
+
 /* Called with RTNL lock and dp_mutex. */
 static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
 {
@@ -349,6 +361,11 @@ static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
 	list_add_rcu(&p->node, &dp->port_list);
 	dp->n_ports++;
 
+	/* Initialize kobject for bridge.  This will be added as
+	 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
+	p->kobj.kset = NULL;
+	kobject_init(&p->kobj, &brport_ktype);
+
 	dp_ifinfo_notify(RTM_NEWLINK, p);
 
 	return 0;
@@ -376,7 +393,7 @@ static int add_port(int dp_idx, struct odp_port __user *portp)
 	for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
 		if (!dp->ports[port_no])
 			goto got_port_no;
-	err = -EXFULL;
+	err = -EFBIG;
 	goto out_unlock_dp;
 
 got_port_no:
@@ -402,8 +419,7 @@ got_port_no:
 	if (err)
 		goto out_put;
 
-	if (dp_add_if_hook)
-		dp_add_if_hook(dp->ports[port_no]);
+	dp_sysfs_add_if(dp->ports[port_no]);
 
 	err = __put_user(port_no, &port.port);
 
@@ -421,13 +437,8 @@ int dp_del_port(struct net_bridge_port *p)
 {
 	ASSERT_RTNL();
 
-	if (p->port_no != ODPP_LOCAL && dp_del_if_hook) {
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25)
-		sysfs_remove_link(&p->dp->ifobj, p->dev->name);
-#else
-		sysfs_remove_link(p->dp->ifobj, p->dev->name);
-#endif
-	}
+	if (p->port_no != ODPP_LOCAL)
+		dp_sysfs_del_if(p);
 	dp_ifinfo_notify(RTM_DELLINK, p);
 
 	p->dp->n_ports--;
@@ -448,15 +459,10 @@ int dp_del_port(struct net_bridge_port *p)
 	/* Then wait until no one is still using it, and destroy it. */
 	synchronize_rcu();
 
-	if (is_dp_dev(p->dev)) {
+	if (is_dp_dev(p->dev))
 		dp_dev_destroy(p->dev);
-	}
-	if (p->port_no != ODPP_LOCAL && dp_del_if_hook) {
-		dp_del_if_hook(p);
-	} else {
-		dev_put(p->dev);
-		kfree(p);
-	}
+	dev_put(p->dev);
+	kobject_put(&p->kobj);
 
 	return 0;
 }
@@ -568,8 +574,7 @@ static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 #error
 #endif
 
-#ifdef CONFIG_XEN
-#if LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
+#if defined(CONFIG_XEN) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
 /* This code is copied verbatim from net/dev/core.c in Xen's
  * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644.  We can't call those functions
  * directly because they aren't exported. */
@@ -585,7 +590,7 @@ static int skb_pull_up_to(struct sk_buff *skb, void *ptr)
 	}
 }
 
-int skb_checksum_setup(struct sk_buff *skb)
+int vswitch_skb_checksum_setup(struct sk_buff *skb)
 {
 	if (skb->proto_csum_blank) {
 		if (skb->protocol != htons(ETH_P_IP))
@@ -616,8 +621,90 @@ int skb_checksum_setup(struct sk_buff *skb)
 out:
 	return -EPROTO;
 }
-#endif /* linux == 2.6.18 */
-#endif /* CONFIG_XEN */
+#else
+int vswitch_skb_checksum_setup(struct sk_buff *skb) { return 0; }
+#endif /* CONFIG_XEN && linux == 2.6.18 */
+
+/* Append each packet in 'skb' list to 'queue'.  There will be only one packet
+ * unless we broke up a GSO packet. */
+static int
+queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
+		      int queue_no, u32 arg)
+{
+	struct sk_buff *nskb;
+	int port_no;
+	int err;
+
+	port_no = ODPP_LOCAL;
+	if (skb->dev) {
+		if (skb->dev->br_port)
+			port_no = skb->dev->br_port->port_no;
+		else if (is_dp_dev(skb->dev))
+			port_no = dp_dev_priv(skb->dev)->port_no;
+	}
+
+	do {
+		struct odp_msg *header;
+
+		nskb = skb->next;
+		skb->next = NULL;
+
+		/* If a checksum-deferred packet is forwarded to the
+		 * controller, correct the pointers and checksum.  This happens
+		 * on a regular basis only on Xen, on which VMs can pass up
+		 * packets that do not have their checksum computed.
+		 */
+		err = vswitch_skb_checksum_setup(skb);
+		if (err)
+			goto err_kfree_skbs;
+#ifndef CHECKSUM_HW
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
+			/* Until 2.6.22, the start of the transport header was
+			 * also the start of data to be checksummed.  Linux
+			 * 2.6.22 introduced the csum_start field for this
+			 * purpose, but we should point the transport header to
+			 * it anyway for backward compatibility, as
+			 * dev_queue_xmit() does even in 2.6.28. */
+			skb_set_transport_header(skb, skb->csum_start -
+						 skb_headroom(skb));
+#endif
+			err = skb_checksum_help(skb);
+			if (err)
+				goto err_kfree_skbs;
+		}
+#else
+		if (skb->ip_summed == CHECKSUM_HW) {
+			err = skb_checksum_help(skb, 0);
+			if (err)
+				goto err_kfree_skbs;
+		}
+#endif
+
+		err = skb_cow(skb, sizeof *header);
+		if (err)
+			goto err_kfree_skbs;
+
+		header = (struct odp_msg*)__skb_push(skb, sizeof *header);
+		header->type = queue_no;
+		header->length = skb->len;
+		header->port = port_no;
+		header->reserved = 0;
+		header->arg = arg;
+		skb_queue_tail(queue, skb);
+
+		skb = nskb;
+	} while (skb);
+	return 0;
+
+err_kfree_skbs:
+	kfree_skb(skb);
+	while ((skb = nskb) != NULL) {
+		nskb = skb->next;
+		kfree_skb(skb);
+	}
+	return err;
+}
 
 int
 dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
@@ -625,7 +712,6 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
 {
 	struct dp_stats_percpu *stats;
 	struct sk_buff_head *queue;
-	int port_no;
 	int err;
 
 	WARN_ON_ONCE(skb_shared(skb));
@@ -636,40 +722,6 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
 	if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
 		goto err_kfree_skb;
 
- 	/* If a checksum-deferred packet is forwarded to the controller,
-	 * correct the pointers and checksum.  This happens on a regular basis
-	 * only on Xen (the CHECKSUM_HW case), on which VMs can pass up packets
-	 * that do not have their checksum computed.  We also implement it for
-	 * the non-Xen case, but it is difficult to trigger or test this case
-	 * there, hence the WARN_ON_ONCE().
- 	 */
-	err = skb_checksum_setup(skb);
- 	if (err)
-		goto err_kfree_skb;
-#ifndef CHECKSUM_HW
-	if (skb->ip_summed == CHECKSUM_PARTIAL) {
-		WARN_ON_ONCE(1);
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
-		/* Until 2.6.22, the start of the transport header was also the
-		 * start of data to be checksummed.  Linux 2.6.22 introduced
-		 * the csum_start field for this purpose, but we should point
-		 * the transport header to it anyway for backward
-		 * compatibility, as dev_queue_xmit() does even in 2.6.28. */
-		skb_set_transport_header(skb, skb->csum_start -
-					      skb_headroom(skb));
-#endif
-		err = skb_checksum_help(skb);
-		if (err)
-			goto err_kfree_skb;
-	}
-#else
-	if (skb->ip_summed == CHECKSUM_HW) {
-		err = skb_checksum_help(skb, 0);
-		if (err)
-			goto err_kfree_skb;
-	}
-#endif
-
 	/* Break apart GSO packets into their component pieces.  Otherwise
 	 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 	if (skb_is_gso(skb)) {
@@ -687,45 +739,9 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
 		}
 	}
 
-	/* Figure out port number. */
-	port_no = ODPP_LOCAL;
-	if (skb->dev) {
-		if (skb->dev->br_port)
-			port_no = skb->dev->br_port->port_no;
-		else if (is_dp_dev(skb->dev))
-			port_no = dp_dev_priv(skb->dev)->port_no;
-	}
-
-	/* Append each packet to queue.  There will be only one packet unless
-	 * we broke up a GSO packet above. */
-	do {
-		struct odp_msg *header;
-		struct sk_buff *nskb = skb->next;
-		skb->next = NULL;
-
-		err = skb_cow(skb, sizeof *header);
-		if (err) {
-			while (nskb) {
-				kfree_skb(skb);
-				skb = nskb;
-				nskb = skb->next;
-			}
-			goto err_kfree_skb;
-		}
-
-		header = (struct odp_msg*)__skb_push(skb, sizeof *header);
-		header->type = queue_no;
-		header->length = skb->len;
-		header->port = port_no;
-		header->reserved = 0;
-		header->arg = arg;
-		skb_queue_tail(queue, skb);
-
-		skb = nskb;
-	} while (skb);
-
+	err = queue_control_packets(skb, queue, queue_no, arg);
 	wake_up_interruptible(&dp->waitqueue);
-	return 0;
+	return err;
 
 err_kfree_skb:
 	kfree_skb(skb);
@@ -834,7 +850,7 @@ static void clear_stats(struct sw_flow *flow)
 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
 {
 	struct odp_flow_put uf;
-	struct sw_flow *flow, **bucket;
+	struct sw_flow *flow;
 	struct dp_table *table;
 	struct odp_flow_stats stats;
 	int error;
@@ -844,15 +860,10 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
 		goto error;
 	uf.flow.key.reserved = 0;
 
-retry:
 	table = rcu_dereference(dp->table);
-	bucket = dp_table_lookup_for_insert(table, &uf.flow.key);
-	if (!bucket) {
-		/* No such flow, and the slots where it could go are full. */
-		error = uf.flags & ODPPF_CREATE ? -EXFULL : -ENOENT;
-		goto error;
-	} else if (!*bucket) {
-		/* No such flow, but we found an available slot for it. */
+	flow = dp_table_lookup(table, &uf.flow.key);
+	if (!flow) {
+		/* No such flow. */
 		struct sw_flow_actions *acts;
 
 		error = -ENOENT;
@@ -860,14 +871,15 @@ retry:
 			goto error;
 
 		/* Expand table, if necessary, to make room. */
-		if (dp->n_flows * 4 >= table->n_buckets &&
-		    table->n_buckets < DP_MAX_BUCKETS) {
+		if (dp->n_flows >= table->n_buckets) {
+			error = -ENOSPC;
+			if (table->n_buckets >= DP_MAX_BUCKETS)
+				goto error;
+
 			error = dp_table_expand(dp);
 			if (error)
 				goto error;
-
-			/* The bucket's location has changed.  Try again. */
-			goto retry;
+			table = rcu_dereference(dp->table);
 		}
 
 		/* Allocate flow. */
@@ -887,12 +899,13 @@ retry:
 		rcu_assign_pointer(flow->sf_acts, acts);
 
 		/* Put flow in bucket. */
-		rcu_assign_pointer(*bucket, flow);
+		error = dp_table_insert(table, flow);
+		if (error)
+			goto error_free_flow_acts;
 		dp->n_flows++;
 		memset(&stats, 0, sizeof(struct odp_flow_stats));
 	} else {
 		/* We found a matching flow. */
-		struct sw_flow *flow = *rcu_dereference(bucket);
 		struct sw_flow_actions *old_acts, *new_acts;
 		unsigned long int flags;
 
@@ -930,6 +943,8 @@ retry:
 		return -EFAULT;
 	return 0;
 
+error_free_flow_acts:
+	kfree(flow->sf_acts);
 error_free_flow:
 	kmem_cache_free(flow_cache, flow);
 error:
@@ -1103,6 +1118,7 @@ static int do_execute(struct datapath *dp, const struct odp_execute *executep)
 	struct odp_flow_key key;
 	struct sk_buff *skb;
 	struct sw_flow_actions *actions;
+	struct ethhdr *eth;
 	int err;
 
 	err = -EFAULT;
@@ -1142,6 +1158,17 @@ static int do_execute(struct datapath *dp, const struct odp_execute *executep)
 			   execute.length))
 		goto error_free_skb;
 
+	skb_reset_mac_header(skb);
+	eth = eth_hdr(skb);
+
+    /* Normally, setting the skb 'protocol' field would be handled by a
+     * call to eth_type_trans(), but it assumes there's a sending
+     * device, which we may not have. */
+	if (ntohs(eth->h_proto) >= 1536)
+		skb->protocol = eth->h_proto;
+	else
+		skb->protocol = htons(ETH_P_802_2);
+
 	flow_extract(skb, execute.in_port, &key);
 	err = execute_actions(dp, skb, &key, actions->actions,
 			      actions->n_actions, GFP_KERNEL);
@@ -1162,8 +1189,8 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
 	int i;
 
 	stats.n_flows = dp->n_flows;
-	stats.cur_capacity = rcu_dereference(dp->table)->n_buckets * 2;
-	stats.max_capacity = DP_MAX_BUCKETS * 2;
+	stats.cur_capacity = rcu_dereference(dp->table)->n_buckets;
+	stats.max_capacity = DP_MAX_BUCKETS;
 	stats.n_ports = dp->n_ports;
 	stats.max_ports = DP_MAX_PORTS;
 	stats.max_groups = DP_MAX_GROUPS;
@@ -1181,6 +1208,29 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
 	return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
 }
 
+/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
+int dp_min_mtu(const struct datapath *dp)
+{
+	struct net_bridge_port *p;
+	int mtu = 0;
+
+	ASSERT_RTNL();
+
+	list_for_each_entry_rcu (p, &dp->port_list, node) {
+		struct net_device *dev = p->dev;
+
+		/* Skip any internal ports, since that's what we're trying to
+		 * set. */
+		if (is_dp_dev(dev))
+			continue;
+
+		if (!mtu || dev->mtu < mtu)
+			mtu = dev->mtu;
+	}
+
+	return mtu ? mtu : ETH_DATA_LEN;
+}
+
 static int
 put_port(const struct net_bridge_port *p, struct odp_port __user *uop)
 {