xref: /linux/net/bridge/br_if.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
1 /*
2  *	Userspace interface
3  *	Linux ethernet bridge
4  *
5  *	Authors:
6  *	Lennert Buytenhek		<buytenh@gnu.org>
7  *
8  *	$Id: br_if.c,v 1.7 2001/12/24 00:59:55 davem Exp $
9  *
10  *	This program is free software; you can redistribute it and/or
11  *	modify it under the terms of the GNU General Public License
12  *	as published by the Free Software Foundation; either version
13  *	2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/kernel.h>
17 #include <linux/netdevice.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 <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  * Need to simulate user ioctl because not all device's that support
33  * ethtool, use ethtool_ops.  Also, since driver might sleep need to
34  * not be holding any locks.
35  */
36 static int port_cost(struct net_device *dev)
37 {
38 	struct ethtool_cmd ecmd = { ETHTOOL_GSET };
39 	struct ifreq ifr;
40 	mm_segment_t old_fs;
41 	int err;
42 
43 	strncpy(ifr.ifr_name, dev->name, IFNAMSIZ);
44 	ifr.ifr_data = (void __user *) &ecmd;
45 
46 	old_fs = get_fs();
47 	set_fs(KERNEL_DS);
48 	err = dev_ethtool(&ifr);
49 	set_fs(old_fs);
50 
51 	if (!err) {
52 		switch(ecmd.speed) {
53 		case SPEED_100:
54 			return 19;
55 		case SPEED_1000:
56 			return 4;
57 		case SPEED_10000:
58 			return 2;
59 		case SPEED_10:
60 			return 100;
61 		}
62 	}
63 
64 	/* Old silly heuristics based on name */
65 	if (!strncmp(dev->name, "lec", 3))
66 		return 7;
67 
68 	if (!strncmp(dev->name, "plip", 4))
69 		return 2500;
70 
71 	return 100;	/* assume old 10Mbps */
72 }
73 
74 
75 /*
76  * Check for port carrier transistions.
77  * Called from work queue to allow for calling functions that
78  * might sleep (such as speed check), and to debounce.
79  */
80 void br_port_carrier_check(struct net_bridge_port *p)
81 {
82 	struct net_device *dev = p->dev;
83 	struct net_bridge *br = p->br;
84 
85 	if (netif_carrier_ok(dev))
86 		p->path_cost = port_cost(dev);
87 
88 	if (netif_running(br->dev)) {
89 		spin_lock_bh(&br->lock);
90 		if (netif_carrier_ok(dev)) {
91 			if (p->state == BR_STATE_DISABLED)
92 				br_stp_enable_port(p);
93 		} else {
94 			if (p->state != BR_STATE_DISABLED)
95 				br_stp_disable_port(p);
96 		}
97 		spin_unlock_bh(&br->lock);
98 	}
99 }
100 
101 static void release_nbp(struct kobject *kobj)
102 {
103 	struct net_bridge_port *p
104 		= container_of(kobj, struct net_bridge_port, kobj);
105 	kfree(p);
106 }
107 
108 static struct kobj_type brport_ktype = {
109 #ifdef CONFIG_SYSFS
110 	.sysfs_ops = &brport_sysfs_ops,
111 #endif
112 	.release = release_nbp,
113 };
114 
115 static void destroy_nbp(struct net_bridge_port *p)
116 {
117 	struct net_device *dev = p->dev;
118 
119 	p->br = NULL;
120 	p->dev = NULL;
121 	dev_put(dev);
122 
123 	kobject_put(&p->kobj);
124 }
125 
126 static void destroy_nbp_rcu(struct rcu_head *head)
127 {
128 	struct net_bridge_port *p =
129 			container_of(head, struct net_bridge_port, rcu);
130 	destroy_nbp(p);
131 }
132 
133 /* Delete port(interface) from bridge is done in two steps.
134  * via RCU. First step, marks device as down. That deletes
135  * all the timers and stops new packets from flowing through.
136  *
137  * Final cleanup doesn't occur until after all CPU's finished
138  * processing packets.
139  *
140  * Protected from multiple admin operations by RTNL mutex
141  */
142 static void del_nbp(struct net_bridge_port *p)
143 {
144 	struct net_bridge *br = p->br;
145 	struct net_device *dev = p->dev;
146 
147 	sysfs_remove_link(&br->ifobj, dev->name);
148 
149 	dev_set_promiscuity(dev, -1);
150 
151 	spin_lock_bh(&br->lock);
152 	br_stp_disable_port(p);
153 	spin_unlock_bh(&br->lock);
154 
155 	br_fdb_delete_by_port(br, p, 1);
156 
157 	list_del_rcu(&p->list);
158 
159 	rcu_assign_pointer(dev->br_port, NULL);
160 
161 	kobject_uevent(&p->kobj, KOBJ_REMOVE);
162 	kobject_del(&p->kobj);
163 
164 	call_rcu(&p->rcu, destroy_nbp_rcu);
165 }
166 
167 /* called with RTNL */
168 static void del_br(struct net_bridge *br)
169 {
170 	struct net_bridge_port *p, *n;
171 
172 	list_for_each_entry_safe(p, n, &br->port_list, list) {
173 		del_nbp(p);
174 	}
175 
176 	del_timer_sync(&br->gc_timer);
177 
178 	br_sysfs_delbr(br->dev);
179 	unregister_netdevice(br->dev);
180 }
181 
182 static struct net_device *new_bridge_dev(const char *name)
183 {
184 	struct net_bridge *br;
185 	struct net_device *dev;
186 
187 	dev = alloc_netdev(sizeof(struct net_bridge), name,
188 			   br_dev_setup);
189 
190 	if (!dev)
191 		return NULL;
192 
193 	br = netdev_priv(dev);
194 	br->dev = dev;
195 
196 	spin_lock_init(&br->lock);
197 	INIT_LIST_HEAD(&br->port_list);
198 	spin_lock_init(&br->hash_lock);
199 
200 	br->bridge_id.prio[0] = 0x80;
201 	br->bridge_id.prio[1] = 0x00;
202 
203 	memcpy(br->group_addr, br_group_address, ETH_ALEN);
204 
205 	br->feature_mask = dev->features;
206 	br->stp_enabled = 0;
207 	br->designated_root = br->bridge_id;
208 	br->root_path_cost = 0;
209 	br->root_port = 0;
210 	br->bridge_max_age = br->max_age = 20 * HZ;
211 	br->bridge_hello_time = br->hello_time = 2 * HZ;
212 	br->bridge_forward_delay = br->forward_delay = 15 * HZ;
213 	br->topology_change = 0;
214 	br->topology_change_detected = 0;
215 	br->ageing_time = 300 * HZ;
216 	INIT_LIST_HEAD(&br->age_list);
217 
218 	br_stp_timer_init(br);
219 
220 	return dev;
221 }
222 
223 /* find an available port number */
224 static int find_portno(struct net_bridge *br)
225 {
226 	int index;
227 	struct net_bridge_port *p;
228 	unsigned long *inuse;
229 
230 	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
231 			GFP_KERNEL);
232 	if (!inuse)
233 		return -ENOMEM;
234 
235 	set_bit(0, inuse);	/* zero is reserved */
236 	list_for_each_entry(p, &br->port_list, list) {
237 		set_bit(p->port_no, inuse);
238 	}
239 	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
240 	kfree(inuse);
241 
242 	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
243 }
244 
245 /* called with RTNL but without bridge lock */
246 static struct net_bridge_port *new_nbp(struct net_bridge *br,
247 				       struct net_device *dev)
248 {
249 	int index;
250 	struct net_bridge_port *p;
251 
252 	index = find_portno(br);
253 	if (index < 0)
254 		return ERR_PTR(index);
255 
256 	p = kzalloc(sizeof(*p), GFP_KERNEL);
257 	if (p == NULL)
258 		return ERR_PTR(-ENOMEM);
259 
260 	p->br = br;
261 	dev_hold(dev);
262 	p->dev = dev;
263 	p->path_cost = port_cost(dev);
264 	p->priority = 0x8000 >> BR_PORT_BITS;
265 	p->port_no = index;
266 	br_init_port(p);
267 	p->state = BR_STATE_DISABLED;
268 	br_stp_port_timer_init(p);
269 
270 	kobject_init(&p->kobj);
271 	kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
272 	p->kobj.ktype = &brport_ktype;
273 	p->kobj.parent = &(dev->dev.kobj);
274 	p->kobj.kset = NULL;
275 
276 	return p;
277 }
278 
279 int br_add_bridge(const char *name)
280 {
281 	struct net_device *dev;
282 	int ret;
283 
284 	dev = new_bridge_dev(name);
285 	if (!dev)
286 		return -ENOMEM;
287 
288 	rtnl_lock();
289 	if (strchr(dev->name, '%')) {
290 		ret = dev_alloc_name(dev, dev->name);
291 		if (ret < 0) {
292 			free_netdev(dev);
293 			goto out;
294 		}
295 	}
296 
297 	ret = register_netdevice(dev);
298 	if (ret)
299 		goto out;
300 
301 	ret = br_sysfs_addbr(dev);
302 	if (ret)
303 		unregister_netdevice(dev);
304  out:
305 	rtnl_unlock();
306 	return ret;
307 }
308 
309 int br_del_bridge(const char *name)
310 {
311 	struct net_device *dev;
312 	int ret = 0;
313 
314 	rtnl_lock();
315 	dev = __dev_get_by_name(name);
316 	if (dev == NULL)
317 		ret =  -ENXIO; 	/* Could not find device */
318 
319 	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
320 		/* Attempt to delete non bridge device! */
321 		ret = -EPERM;
322 	}
323 
324 	else if (dev->flags & IFF_UP) {
325 		/* Not shutdown yet. */
326 		ret = -EBUSY;
327 	}
328 
329 	else
330 		del_br(netdev_priv(dev));
331 
332 	rtnl_unlock();
333 	return ret;
334 }
335 
336 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
337 int br_min_mtu(const struct net_bridge *br)
338 {
339 	const struct net_bridge_port *p;
340 	int mtu = 0;
341 
342 	ASSERT_RTNL();
343 
344 	if (list_empty(&br->port_list))
345 		mtu = ETH_DATA_LEN;
346 	else {
347 		list_for_each_entry(p, &br->port_list, list) {
348 			if (!mtu  || p->dev->mtu < mtu)
349 				mtu = p->dev->mtu;
350 		}
351 	}
352 	return mtu;
353 }
354 
355 /*
356  * Recomputes features using slave's features
357  */
358 void br_features_recompute(struct net_bridge *br)
359 {
360 	struct net_bridge_port *p;
361 	unsigned long features, checksum;
362 
363 	checksum = br->feature_mask & NETIF_F_ALL_CSUM ? NETIF_F_NO_CSUM : 0;
364 	features = br->feature_mask & ~NETIF_F_ALL_CSUM;
365 
366 	list_for_each_entry(p, &br->port_list, list) {
367 		unsigned long feature = p->dev->features;
368 
369 		if (checksum & NETIF_F_NO_CSUM && !(feature & NETIF_F_NO_CSUM))
370 			checksum ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
371 		if (checksum & NETIF_F_HW_CSUM && !(feature & NETIF_F_HW_CSUM))
372 			checksum ^= NETIF_F_HW_CSUM | NETIF_F_IP_CSUM;
373 		if (!(feature & NETIF_F_IP_CSUM))
374 			checksum = 0;
375 
376 		if (feature & NETIF_F_GSO)
377 			feature |= NETIF_F_GSO_SOFTWARE;
378 		feature |= NETIF_F_GSO;
379 
380 		features &= feature;
381 	}
382 
383 	if (!(checksum & NETIF_F_ALL_CSUM))
384 		features &= ~NETIF_F_SG;
385 	if (!(features & NETIF_F_SG))
386 		features &= ~NETIF_F_GSO_MASK;
387 
388 	br->dev->features = features | checksum | NETIF_F_LLTX |
389 			    NETIF_F_GSO_ROBUST;
390 }
391 
392 /* called with RTNL */
393 int br_add_if(struct net_bridge *br, struct net_device *dev)
394 {
395 	struct net_bridge_port *p;
396 	int err = 0;
397 
398 	if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
399 		return -EINVAL;
400 
401 	if (dev->hard_start_xmit == br_dev_xmit)
402 		return -ELOOP;
403 
404 	if (dev->br_port != NULL)
405 		return -EBUSY;
406 
407 	p = new_nbp(br, dev);
408 	if (IS_ERR(p))
409 		return PTR_ERR(p);
410 
411 	err = kobject_add(&p->kobj);
412 	if (err)
413 		goto err0;
414 
415 	err = br_fdb_insert(br, p, dev->dev_addr);
416 	if (err)
417 		goto err1;
418 
419 	err = br_sysfs_addif(p);
420 	if (err)
421 		goto err2;
422 
423 	rcu_assign_pointer(dev->br_port, p);
424 	dev_set_promiscuity(dev, 1);
425 
426 	list_add_rcu(&p->list, &br->port_list);
427 
428 	spin_lock_bh(&br->lock);
429 	br_stp_recalculate_bridge_id(br);
430 	br_features_recompute(br);
431 
432 	if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
433 	    (br->dev->flags & IFF_UP))
434 		br_stp_enable_port(p);
435 	spin_unlock_bh(&br->lock);
436 
437 	dev_set_mtu(br->dev, br_min_mtu(br));
438 
439 	kobject_uevent(&p->kobj, KOBJ_ADD);
440 
441 	return 0;
442 err2:
443 	br_fdb_delete_by_port(br, p, 1);
444 err1:
445 	kobject_del(&p->kobj);
446 err0:
447 	kobject_put(&p->kobj);
448 	return err;
449 }
450 
451 /* called with RTNL */
452 int br_del_if(struct net_bridge *br, struct net_device *dev)
453 {
454 	struct net_bridge_port *p = dev->br_port;
455 
456 	if (!p || p->br != br)
457 		return -EINVAL;
458 
459 	del_nbp(p);
460 
461 	spin_lock_bh(&br->lock);
462 	br_stp_recalculate_bridge_id(br);
463 	br_features_recompute(br);
464 	spin_unlock_bh(&br->lock);
465 
466 	return 0;
467 }
468 
469 void __exit br_cleanup_bridges(void)
470 {
471 	struct net_device *dev, *nxt;
472 
473 	rtnl_lock();
474 	for (dev = dev_base; dev; dev = nxt) {
475 		nxt = dev->next;
476 		if (dev->priv_flags & IFF_EBRIDGE)
477 			del_br(dev->priv);
478 	}
479 	rtnl_unlock();
480 
481 }
482