xref: /openbmc/linux/net/bridge/br_if.c (revision 20055477)
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/etherdevice.h>
17 #include <linux/netpoll.h>
18 #include <linux/ethtool.h>
19 #include <linux/if_arp.h>
20 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <linux/rtnetlink.h>
23 #include <linux/if_ether.h>
24 #include <linux/slab.h>
25 #include <net/sock.h>
26 #include <linux/if_vlan.h>
27 
28 #include "br_private.h"
29 
30 /*
31  * Determine initial path cost based on speed.
32  * using recommendations from 802.1d standard
33  *
34  * Since driver might sleep need to not be holding any locks.
35  */
36 static int port_cost(struct net_device *dev)
37 {
38 	struct ethtool_cmd ecmd;
39 
40 	if (!__ethtool_get_settings(dev, &ecmd)) {
41 		switch (ethtool_cmd_speed(&ecmd)) {
42 		case SPEED_10000:
43 			return 2;
44 		case SPEED_1000:
45 			return 4;
46 		case SPEED_100:
47 			return 19;
48 		case SPEED_10:
49 			return 100;
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. */
65 void 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 (!(p->flags & BR_ADMIN_COST) &&
71 	    netif_running(dev) && netif_oper_up(dev))
72 		p->path_cost = port_cost(dev);
73 
74 	if (!netif_running(br->dev))
75 		return;
76 
77 	spin_lock_bh(&br->lock);
78 	if (netif_running(dev) && netif_oper_up(dev)) {
79 		if (p->state == BR_STATE_DISABLED)
80 			br_stp_enable_port(p);
81 	} else {
82 		if (p->state != BR_STATE_DISABLED)
83 			br_stp_disable_port(p);
84 	}
85 	spin_unlock_bh(&br->lock);
86 }
87 
88 static void release_nbp(struct kobject *kobj)
89 {
90 	struct net_bridge_port *p
91 		= container_of(kobj, struct net_bridge_port, kobj);
92 	kfree(p);
93 }
94 
95 static struct kobj_type brport_ktype = {
96 #ifdef CONFIG_SYSFS
97 	.sysfs_ops = &brport_sysfs_ops,
98 #endif
99 	.release = release_nbp,
100 };
101 
102 static void destroy_nbp(struct net_bridge_port *p)
103 {
104 	struct net_device *dev = p->dev;
105 
106 	p->br = NULL;
107 	p->dev = NULL;
108 	dev_put(dev);
109 
110 	kobject_put(&p->kobj);
111 }
112 
113 static void destroy_nbp_rcu(struct rcu_head *head)
114 {
115 	struct net_bridge_port *p =
116 			container_of(head, struct net_bridge_port, rcu);
117 	destroy_nbp(p);
118 }
119 
120 /* Delete port(interface) from bridge is done in two steps.
121  * via RCU. First step, marks device as down. That deletes
122  * all the timers and stops new packets from flowing through.
123  *
124  * Final cleanup doesn't occur until after all CPU's finished
125  * processing packets.
126  *
127  * Protected from multiple admin operations by RTNL mutex
128  */
129 static void del_nbp(struct net_bridge_port *p)
130 {
131 	struct net_bridge *br = p->br;
132 	struct net_device *dev = p->dev;
133 
134 	sysfs_remove_link(br->ifobj, p->dev->name);
135 
136 	dev_set_promiscuity(dev, -1);
137 
138 	spin_lock_bh(&br->lock);
139 	br_stp_disable_port(p);
140 	spin_unlock_bh(&br->lock);
141 
142 	br_ifinfo_notify(RTM_DELLINK, p);
143 
144 	nbp_vlan_flush(p);
145 	br_fdb_delete_by_port(br, p, 1);
146 
147 	list_del_rcu(&p->list);
148 
149 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
150 
151 	netdev_rx_handler_unregister(dev);
152 
153 	netdev_upper_dev_unlink(dev, br->dev);
154 
155 	br_multicast_del_port(p);
156 
157 	kobject_uevent(&p->kobj, KOBJ_REMOVE);
158 	kobject_del(&p->kobj);
159 
160 	br_netpoll_disable(p);
161 
162 	call_rcu(&p->rcu, destroy_nbp_rcu);
163 }
164 
165 /* Delete bridge device */
166 void br_dev_delete(struct net_device *dev, struct list_head *head)
167 {
168 	struct net_bridge *br = netdev_priv(dev);
169 	struct net_bridge_port *p, *n;
170 
171 	list_for_each_entry_safe(p, n, &br->port_list, list) {
172 		del_nbp(p);
173 	}
174 
175 	del_timer_sync(&br->gc_timer);
176 
177 	br_sysfs_delbr(br->dev);
178 	unregister_netdevice_queue(br->dev, head);
179 }
180 
181 /* find an available port number */
182 static int find_portno(struct net_bridge *br)
183 {
184 	int index;
185 	struct net_bridge_port *p;
186 	unsigned long *inuse;
187 
188 	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
189 			GFP_KERNEL);
190 	if (!inuse)
191 		return -ENOMEM;
192 
193 	set_bit(0, inuse);	/* zero is reserved */
194 	list_for_each_entry(p, &br->port_list, list) {
195 		set_bit(p->port_no, inuse);
196 	}
197 	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
198 	kfree(inuse);
199 
200 	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
201 }
202 
203 /* called with RTNL but without bridge lock */
204 static struct net_bridge_port *new_nbp(struct net_bridge *br,
205 				       struct net_device *dev)
206 {
207 	int index;
208 	struct net_bridge_port *p;
209 
210 	index = find_portno(br);
211 	if (index < 0)
212 		return ERR_PTR(index);
213 
214 	p = kzalloc(sizeof(*p), GFP_KERNEL);
215 	if (p == NULL)
216 		return ERR_PTR(-ENOMEM);
217 
218 	p->br = br;
219 	dev_hold(dev);
220 	p->dev = dev;
221 	p->path_cost = port_cost(dev);
222 	p->priority = 0x8000 >> BR_PORT_BITS;
223 	p->port_no = index;
224 	p->flags = BR_LEARNING | BR_FLOOD;
225 	br_init_port(p);
226 	p->state = BR_STATE_DISABLED;
227 	br_stp_port_timer_init(p);
228 	br_multicast_add_port(p);
229 
230 	return p;
231 }
232 
233 int br_add_bridge(struct net *net, const char *name)
234 {
235 	struct net_device *dev;
236 	int res;
237 
238 	dev = alloc_netdev(sizeof(struct net_bridge), name,
239 			   br_dev_setup);
240 
241 	if (!dev)
242 		return -ENOMEM;
243 
244 	dev_net_set(dev, net);
245 	dev->rtnl_link_ops = &br_link_ops;
246 
247 	res = register_netdev(dev);
248 	if (res)
249 		free_netdev(dev);
250 	return res;
251 }
252 
253 int br_del_bridge(struct net *net, const char *name)
254 {
255 	struct net_device *dev;
256 	int ret = 0;
257 
258 	rtnl_lock();
259 	dev = __dev_get_by_name(net, name);
260 	if (dev == NULL)
261 		ret =  -ENXIO; 	/* Could not find device */
262 
263 	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
264 		/* Attempt to delete non bridge device! */
265 		ret = -EPERM;
266 	}
267 
268 	else if (dev->flags & IFF_UP) {
269 		/* Not shutdown yet. */
270 		ret = -EBUSY;
271 	}
272 
273 	else
274 		br_dev_delete(dev, NULL);
275 
276 	rtnl_unlock();
277 	return ret;
278 }
279 
280 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
281 int br_min_mtu(const struct net_bridge *br)
282 {
283 	const struct net_bridge_port *p;
284 	int mtu = 0;
285 
286 	ASSERT_RTNL();
287 
288 	if (list_empty(&br->port_list))
289 		mtu = ETH_DATA_LEN;
290 	else {
291 		list_for_each_entry(p, &br->port_list, list) {
292 			if (!mtu  || p->dev->mtu < mtu)
293 				mtu = p->dev->mtu;
294 		}
295 	}
296 	return mtu;
297 }
298 
299 /*
300  * Recomputes features using slave's features
301  */
302 netdev_features_t br_features_recompute(struct net_bridge *br,
303 	netdev_features_t features)
304 {
305 	struct net_bridge_port *p;
306 	netdev_features_t mask;
307 
308 	if (list_empty(&br->port_list))
309 		return features;
310 
311 	mask = features;
312 	features &= ~NETIF_F_ONE_FOR_ALL;
313 
314 	list_for_each_entry(p, &br->port_list, list) {
315 		features = netdev_increment_features(features,
316 						     p->dev->features, mask);
317 	}
318 
319 	return features;
320 }
321 
322 /* called with RTNL */
323 int br_add_if(struct net_bridge *br, struct net_device *dev)
324 {
325 	struct net_bridge_port *p;
326 	int err = 0;
327 	bool changed_addr;
328 
329 	/* Don't allow bridging non-ethernet like devices */
330 	if ((dev->flags & IFF_LOOPBACK) ||
331 	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
332 	    !is_valid_ether_addr(dev->dev_addr))
333 		return -EINVAL;
334 
335 	/* No bridging of bridges */
336 	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
337 		return -ELOOP;
338 
339 	/* Device is already being bridged */
340 	if (br_port_exists(dev))
341 		return -EBUSY;
342 
343 	/* No bridging devices that dislike that (e.g. wireless) */
344 	if (dev->priv_flags & IFF_DONT_BRIDGE)
345 		return -EOPNOTSUPP;
346 
347 	p = new_nbp(br, dev);
348 	if (IS_ERR(p))
349 		return PTR_ERR(p);
350 
351 	call_netdevice_notifiers(NETDEV_JOIN, dev);
352 
353 	err = dev_set_promiscuity(dev, 1);
354 	if (err)
355 		goto put_back;
356 
357 	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
358 				   SYSFS_BRIDGE_PORT_ATTR);
359 	if (err)
360 		goto err1;
361 
362 	err = br_sysfs_addif(p);
363 	if (err)
364 		goto err2;
365 
366 	err = br_netpoll_enable(p, GFP_KERNEL);
367 	if (err)
368 		goto err3;
369 
370 	err = netdev_master_upper_dev_link(dev, br->dev);
371 	if (err)
372 		goto err4;
373 
374 	err = netdev_rx_handler_register(dev, br_handle_frame, p);
375 	if (err)
376 		goto err5;
377 
378 	dev->priv_flags |= IFF_BRIDGE_PORT;
379 
380 	dev_disable_lro(dev);
381 
382 	list_add_rcu(&p->list, &br->port_list);
383 
384 	netdev_update_features(br->dev);
385 
386 	if (br->dev->needed_headroom < dev->needed_headroom)
387 		br->dev->needed_headroom = dev->needed_headroom;
388 
389 	spin_lock_bh(&br->lock);
390 	changed_addr = br_stp_recalculate_bridge_id(br);
391 
392 	if (netif_running(dev) && netif_oper_up(dev) &&
393 	    (br->dev->flags & IFF_UP))
394 		br_stp_enable_port(p);
395 	spin_unlock_bh(&br->lock);
396 
397 	br_ifinfo_notify(RTM_NEWLINK, p);
398 
399 	if (changed_addr)
400 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
401 
402 	dev_set_mtu(br->dev, br_min_mtu(br));
403 
404 	if (br_fdb_insert(br, p, dev->dev_addr, 0))
405 		netdev_err(dev, "failed insert local address bridge forwarding table\n");
406 
407 	kobject_uevent(&p->kobj, KOBJ_ADD);
408 
409 	return 0;
410 
411 err5:
412 	netdev_upper_dev_unlink(dev, br->dev);
413 err4:
414 	br_netpoll_disable(p);
415 err3:
416 	sysfs_remove_link(br->ifobj, p->dev->name);
417 err2:
418 	kobject_put(&p->kobj);
419 	p = NULL; /* kobject_put frees */
420 err1:
421 	dev_set_promiscuity(dev, -1);
422 put_back:
423 	dev_put(dev);
424 	kfree(p);
425 	return err;
426 }
427 
428 /* called with RTNL */
429 int br_del_if(struct net_bridge *br, struct net_device *dev)
430 {
431 	struct net_bridge_port *p;
432 	bool changed_addr;
433 
434 	p = br_port_get_rtnl(dev);
435 	if (!p || p->br != br)
436 		return -EINVAL;
437 
438 	/* Since more than one interface can be attached to a bridge,
439 	 * there still maybe an alternate path for netconsole to use;
440 	 * therefore there is no reason for a NETDEV_RELEASE event.
441 	 */
442 	del_nbp(p);
443 
444 	spin_lock_bh(&br->lock);
445 	changed_addr = br_stp_recalculate_bridge_id(br);
446 	spin_unlock_bh(&br->lock);
447 
448 	if (changed_addr)
449 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
450 
451 	netdev_update_features(br->dev);
452 
453 	return 0;
454 }
455 
456 void __net_exit br_net_exit(struct net *net)
457 {
458 	struct net_device *dev;
459 	LIST_HEAD(list);
460 
461 	rtnl_lock();
462 	for_each_netdev(net, dev)
463 		if (dev->priv_flags & IFF_EBRIDGE)
464 			br_dev_delete(dev, &list);
465 
466 	unregister_netdevice_many(&list);
467 	rtnl_unlock();
468 
469 }
470