br_if.c revision b5ed54e94d324f17c97852296d61a143f01b227a 
1/*
2 *	Userspace interface
3 *	Linux ethernet bridge
4 *
5 *	Authors:
6 *	Lennert Buytenhek		<buytenh@gnu.org>
7 *
8 *	This program is free software; you can redistribute it and/or
9 *	modify it under the terms of the GNU General Public License
10 *	as published by the Free Software Foundation; either version
11 *	2 of the License, or (at your option) any later version.
12 */
13
14#include <linux/kernel.h>
15#include <linux/netdevice.h>
16#include <linux/netpoll.h>
17#include <linux/ethtool.h>
18#include <linux/if_arp.h>
19#include <linux/module.h>
20#include <linux/init.h>
21#include <linux/rtnetlink.h>
22#include <linux/if_ether.h>
23#include <linux/slab.h>
24#include <net/sock.h>
25
26#include "br_private.h"
27
28/*
29 * Determine initial path cost based on speed.
30 * using recommendations from 802.1d standard
31 *
32 * Since driver might sleep need to not be holding any locks.
33 */
34static int port_cost(struct net_device *dev)
35{
36	if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
37		struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
38
39		if (!dev->ethtool_ops->get_settings(dev, &ecmd)) {
40			switch(ecmd.speed) {
41			case SPEED_10000:
42				return 2;
43			case SPEED_1000:
44				return 4;
45			case SPEED_100:
46				return 19;
47			case SPEED_10:
48				return 100;
49			}
50		}
51	}
52
53	/* Old silly heuristics based on name */
54	if (!strncmp(dev->name, "lec", 3))
55		return 7;
56
57	if (!strncmp(dev->name, "plip", 4))
58		return 2500;
59
60	return 100;	/* assume old 10Mbps */
61}
62
63
64/* Check for port carrier transistions. */
65void br_port_carrier_check(struct net_bridge_port *p)
66{
67	struct net_device *dev = p->dev;
68	struct net_bridge *br = p->br;
69
70	if (netif_running(dev) && netif_carrier_ok(dev))
71		p->path_cost = port_cost(dev);
72
73	if (!netif_running(br->dev))
74		return;
75
76	spin_lock_bh(&br->lock);
77	if (netif_running(dev) && netif_carrier_ok(dev)) {
78		if (p->state == BR_STATE_DISABLED)
79			br_stp_enable_port(p);
80	} else {
81		if (p->state != BR_STATE_DISABLED)
82			br_stp_disable_port(p);
83	}
84	spin_unlock_bh(&br->lock);
85}
86
87static void release_nbp(struct kobject *kobj)
88{
89	struct net_bridge_port *p
90		= container_of(kobj, struct net_bridge_port, kobj);
91	kfree(p);
92}
93
94static struct kobj_type brport_ktype = {
95#ifdef CONFIG_SYSFS
96	.sysfs_ops = &brport_sysfs_ops,
97#endif
98	.release = release_nbp,
99};
100
101static void destroy_nbp(struct net_bridge_port *p)
102{
103	struct net_device *dev = p->dev;
104
105	p->br = NULL;
106	p->dev = NULL;
107	dev_put(dev);
108
109	kobject_put(&p->kobj);
110}
111
112static void destroy_nbp_rcu(struct rcu_head *head)
113{
114	struct net_bridge_port *p =
115			container_of(head, struct net_bridge_port, rcu);
116	destroy_nbp(p);
117}
118
119/* Delete port(interface) from bridge is done in two steps.
120 * via RCU. First step, marks device as down. That deletes
121 * all the timers and stops new packets from flowing through.
122 *
123 * Final cleanup doesn't occur until after all CPU's finished
124 * processing packets.
125 *
126 * Protected from multiple admin operations by RTNL mutex
127 */
128static void del_nbp(struct net_bridge_port *p)
129{
130	struct net_bridge *br = p->br;
131	struct net_device *dev = p->dev;
132
133	sysfs_remove_link(br->ifobj, p->dev->name);
134
135	dev_set_promiscuity(dev, -1);
136
137	spin_lock_bh(&br->lock);
138	br_stp_disable_port(p);
139	spin_unlock_bh(&br->lock);
140
141	br_ifinfo_notify(RTM_DELLINK, p);
142
143	br_fdb_delete_by_port(br, p, 1);
144
145	list_del_rcu(&p->list);
146
147	dev->priv_flags &= ~IFF_BRIDGE_PORT;
148
149	netdev_rx_handler_unregister(dev);
150
151	br_multicast_del_port(p);
152
153	kobject_uevent(&p->kobj, KOBJ_REMOVE);
154	kobject_del(&p->kobj);
155
156	br_netpoll_disable(p);
157
158	call_rcu(&p->rcu, destroy_nbp_rcu);
159}
160
161/* called with RTNL */
162static void del_br(struct net_bridge *br, struct list_head *head)
163{
164	struct net_bridge_port *p, *n;
165
166	list_for_each_entry_safe(p, n, &br->port_list, list) {
167		del_nbp(p);
168	}
169
170	del_timer_sync(&br->gc_timer);
171
172	br_sysfs_delbr(br->dev);
173	unregister_netdevice_queue(br->dev, head);
174}
175
176static struct net_device *new_bridge_dev(struct net *net, const char *name)
177{
178	struct net_bridge *br;
179	struct net_device *dev;
180
181	dev = alloc_netdev(sizeof(struct net_bridge), name,
182			   br_dev_setup);
183
184	if (!dev)
185		return NULL;
186	dev_net_set(dev, net);
187
188	br = netdev_priv(dev);
189	br->dev = dev;
190
191	br->stats = alloc_percpu(struct br_cpu_netstats);
192	if (!br->stats) {
193		free_netdev(dev);
194		return NULL;
195	}
196
197	spin_lock_init(&br->lock);
198	INIT_LIST_HEAD(&br->port_list);
199	spin_lock_init(&br->hash_lock);
200
201	br->bridge_id.prio[0] = 0x80;
202	br->bridge_id.prio[1] = 0x00;
203
204	memcpy(br->group_addr, br_group_address, ETH_ALEN);
205
206	br->feature_mask = dev->features;
207	br->stp_enabled = BR_NO_STP;
208	br->designated_root = br->bridge_id;
209	br->root_path_cost = 0;
210	br->root_port = 0;
211	br->bridge_max_age = br->max_age = 20 * HZ;
212	br->bridge_hello_time = br->hello_time = 2 * HZ;
213	br->bridge_forward_delay = br->forward_delay = 15 * HZ;
214	br->topology_change = 0;
215	br->topology_change_detected = 0;
216	br->ageing_time = 300 * HZ;
217
218	br_netfilter_rtable_init(br);
219
220	br_stp_timer_init(br);
221	br_multicast_init(br);
222
223	return dev;
224}
225
226/* find an available port number */
227static int find_portno(struct net_bridge *br)
228{
229	int index;
230	struct net_bridge_port *p;
231	unsigned long *inuse;
232
233	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
234			GFP_KERNEL);
235	if (!inuse)
236		return -ENOMEM;
237
238	set_bit(0, inuse);	/* zero is reserved */
239	list_for_each_entry(p, &br->port_list, list) {
240		set_bit(p->port_no, inuse);
241	}
242	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
243	kfree(inuse);
244
245	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
246}
247
248/* called with RTNL but without bridge lock */
249static struct net_bridge_port *new_nbp(struct net_bridge *br,
250				       struct net_device *dev)
251{
252	int index;
253	struct net_bridge_port *p;
254
255	index = find_portno(br);
256	if (index < 0)
257		return ERR_PTR(index);
258
259	p = kzalloc(sizeof(*p), GFP_KERNEL);
260	if (p == NULL)
261		return ERR_PTR(-ENOMEM);
262
263	p->br = br;
264	dev_hold(dev);
265	p->dev = dev;
266	p->path_cost = port_cost(dev);
267	p->priority = 0x8000 >> BR_PORT_BITS;
268	p->port_no = index;
269	p->flags = 0;
270	br_init_port(p);
271	p->state = BR_STATE_DISABLED;
272	br_stp_port_timer_init(p);
273	br_multicast_add_port(p);
274
275	return p;
276}
277
278static struct device_type br_type = {
279	.name	= "bridge",
280};
281
282int br_add_bridge(struct net *net, const char *name)
283{
284	struct net_device *dev;
285	int ret;
286
287	dev = new_bridge_dev(net, name);
288	if (!dev)
289		return -ENOMEM;
290
291	rtnl_lock();
292	if (strchr(dev->name, '%')) {
293		ret = dev_alloc_name(dev, dev->name);
294		if (ret < 0)
295			goto out_free;
296	}
297
298	SET_NETDEV_DEVTYPE(dev, &br_type);
299
300	ret = register_netdevice(dev);
301	if (ret)
302		goto out_free;
303
304	ret = br_sysfs_addbr(dev);
305	if (ret)
306		unregister_netdevice(dev);
307 out:
308	rtnl_unlock();
309	return ret;
310
311out_free:
312	free_netdev(dev);
313	goto out;
314}
315
316int br_del_bridge(struct net *net, const char *name)
317{
318	struct net_device *dev;
319	int ret = 0;
320
321	rtnl_lock();
322	dev = __dev_get_by_name(net, name);
323	if (dev == NULL)
324		ret =  -ENXIO; 	/* Could not find device */
325
326	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
327		/* Attempt to delete non bridge device! */
328		ret = -EPERM;
329	}
330
331	else if (dev->flags & IFF_UP) {
332		/* Not shutdown yet. */
333		ret = -EBUSY;
334	}
335
336	else
337		del_br(netdev_priv(dev), NULL);
338
339	rtnl_unlock();
340	return ret;
341}
342
343/* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
344int br_min_mtu(const struct net_bridge *br)
345{
346	const struct net_bridge_port *p;
347	int mtu = 0;
348
349	ASSERT_RTNL();
350
351	if (list_empty(&br->port_list))
352		mtu = ETH_DATA_LEN;
353	else {
354		list_for_each_entry(p, &br->port_list, list) {
355			if (!mtu  || p->dev->mtu < mtu)
356				mtu = p->dev->mtu;
357		}
358	}
359	return mtu;
360}
361
362/*
363 * Recomputes features using slave's features
364 */
365void br_features_recompute(struct net_bridge *br)
366{
367	struct net_bridge_port *p;
368	unsigned long features, mask;
369
370	features = mask = br->feature_mask;
371	if (list_empty(&br->port_list))
372		goto done;
373
374	features &= ~NETIF_F_ONE_FOR_ALL;
375
376	list_for_each_entry(p, &br->port_list, list) {
377		features = netdev_increment_features(features,
378						     p->dev->features, mask);
379	}
380
381done:
382	br->dev->features = netdev_fix_features(features, NULL);
383}
384
385/* called with RTNL */
386int br_add_if(struct net_bridge *br, struct net_device *dev)
387{
388	struct net_bridge_port *p;
389	int err = 0;
390
391	/* Don't allow bridging non-ethernet like devices */
392	if ((dev->flags & IFF_LOOPBACK) ||
393	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
394		return -EINVAL;
395
396	/* No bridging of bridges */
397	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
398		return -ELOOP;
399
400	/* Device is already being bridged */
401	if (br_port_exists(dev))
402		return -EBUSY;
403
404	/* No bridging devices that dislike that (e.g. wireless) */
405	if (dev->priv_flags & IFF_DONT_BRIDGE)
406		return -EOPNOTSUPP;
407
408	p = new_nbp(br, dev);
409	if (IS_ERR(p))
410		return PTR_ERR(p);
411
412	err = dev_set_promiscuity(dev, 1);
413	if (err)
414		goto put_back;
415
416	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
417				   SYSFS_BRIDGE_PORT_ATTR);
418	if (err)
419		goto err0;
420
421	err = br_fdb_insert(br, p, dev->dev_addr);
422	if (err)
423		goto err1;
424
425	err = br_sysfs_addif(p);
426	if (err)
427		goto err2;
428
429	if (br_netpoll_info(br) && ((err = br_netpoll_enable(p))))
430		goto err3;
431
432	err = netdev_rx_handler_register(dev, br_handle_frame, p);
433	if (err)
434		goto err3;
435
436	dev->priv_flags |= IFF_BRIDGE_PORT;
437
438	dev_disable_lro(dev);
439
440	list_add_rcu(&p->list, &br->port_list);
441
442	spin_lock_bh(&br->lock);
443	br_stp_recalculate_bridge_id(br);
444	br_features_recompute(br);
445
446	if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
447	    (br->dev->flags & IFF_UP))
448		br_stp_enable_port(p);
449	spin_unlock_bh(&br->lock);
450
451	br_ifinfo_notify(RTM_NEWLINK, p);
452
453	dev_set_mtu(br->dev, br_min_mtu(br));
454
455	kobject_uevent(&p->kobj, KOBJ_ADD);
456
457	return 0;
458err3:
459	sysfs_remove_link(br->ifobj, p->dev->name);
460err2:
461	br_fdb_delete_by_port(br, p, 1);
462err1:
463	kobject_put(&p->kobj);
464	p = NULL; /* kobject_put frees */
465err0:
466	dev_set_promiscuity(dev, -1);
467put_back:
468	dev_put(dev);
469	kfree(p);
470	return err;
471}
472
473/* called with RTNL */
474int br_del_if(struct net_bridge *br, struct net_device *dev)
475{
476	struct net_bridge_port *p;
477
478	p = br_port_get(dev);
479	if (!p || p->br != br)
480		return -EINVAL;
481
482	del_nbp(p);
483
484	spin_lock_bh(&br->lock);
485	br_stp_recalculate_bridge_id(br);
486	br_features_recompute(br);
487	spin_unlock_bh(&br->lock);
488
489	return 0;
490}
491
492void __net_exit br_net_exit(struct net *net)
493{
494	struct net_device *dev;
495	LIST_HEAD(list);
496
497	rtnl_lock();
498	for_each_netdev(net, dev)
499		if (dev->priv_flags & IFF_EBRIDGE)
500			del_br(netdev_priv(dev), &list);
501
502	unregister_netdevice_many(&list);
503	rtnl_unlock();
504
505}
506