xref: /openbmc/linux/net/bridge/br_if.c (revision b96fc2f3) 
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 transitions. */
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 br_port_set_promisc(struct net_bridge_port *p)
89 {
90 	int err = 0;
91 
92 	if (br_promisc_port(p))
93 		return;
94 
95 	err = dev_set_promiscuity(p->dev, 1);
96 	if (err)
97 		return;
98 
99 	br_fdb_unsync_static(p->br, p);
100 	p->flags |= BR_PROMISC;
101 }
102 
103 static void br_port_clear_promisc(struct net_bridge_port *p)
104 {
105 	int err;
106 
107 	/* Check if the port is already non-promisc or if it doesn't
108 	 * support UNICAST filtering.  Without unicast filtering support
109 	 * we'll end up re-enabling promisc mode anyway, so just check for
110 	 * it here.
111 	 */
112 	if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
113 		return;
114 
115 	/* Since we'll be clearing the promisc mode, program the port
116 	 * first so that we don't have interruption in traffic.
117 	 */
118 	err = br_fdb_sync_static(p->br, p);
119 	if (err)
120 		return;
121 
122 	dev_set_promiscuity(p->dev, -1);
123 	p->flags &= ~BR_PROMISC;
124 }
125 
126 /* When a port is added or removed or when certain port flags
127  * change, this function is called to automatically manage
128  * promiscuity setting of all the bridge ports.  We are always called
129  * under RTNL so can skip using rcu primitives.
130  */
131 void br_manage_promisc(struct net_bridge *br)
132 {
133 	struct net_bridge_port *p;
134 	bool set_all = false;
135 
136 	/* If vlan filtering is disabled or bridge interface is placed
137 	 * into promiscuous mode, place all ports in promiscuous mode.
138 	 */
139 	if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br))
140 		set_all = true;
141 
142 	list_for_each_entry(p, &br->port_list, list) {
143 		if (set_all) {
144 			br_port_set_promisc(p);
145 		} else {
146 			/* If the number of auto-ports is <= 1, then all other
147 			 * ports will have their output configuration
148 			 * statically specified through fdbs.  Since ingress
149 			 * on the auto-port becomes forwarding/egress to other
150 			 * ports and egress configuration is statically known,
151 			 * we can say that ingress configuration of the
152 			 * auto-port is also statically known.
153 			 * This lets us disable promiscuous mode and write
154 			 * this config to hw.
155 			 */
156 			if (br->auto_cnt == 0 ||
157 			    (br->auto_cnt == 1 && br_auto_port(p)))
158 				br_port_clear_promisc(p);
159 			else
160 				br_port_set_promisc(p);
161 		}
162 	}
163 }
164 
165 static void nbp_update_port_count(struct net_bridge *br)
166 {
167 	struct net_bridge_port *p;
168 	u32 cnt = 0;
169 
170 	list_for_each_entry(p, &br->port_list, list) {
171 		if (br_auto_port(p))
172 			cnt++;
173 	}
174 	if (br->auto_cnt != cnt) {
175 		br->auto_cnt = cnt;
176 		br_manage_promisc(br);
177 	}
178 }
179 
180 static void nbp_delete_promisc(struct net_bridge_port *p)
181 {
182 	/* If port is currently promiscuous, unset promiscuity.
183 	 * Otherwise, it is a static port so remove all addresses
184 	 * from it.
185 	 */
186 	dev_set_allmulti(p->dev, -1);
187 	if (br_promisc_port(p))
188 		dev_set_promiscuity(p->dev, -1);
189 	else
190 		br_fdb_unsync_static(p->br, p);
191 }
192 
193 static void release_nbp(struct kobject *kobj)
194 {
195 	struct net_bridge_port *p
196 		= container_of(kobj, struct net_bridge_port, kobj);
197 	kfree(p);
198 }
199 
200 static struct kobj_type brport_ktype = {
201 #ifdef CONFIG_SYSFS
202 	.sysfs_ops = &brport_sysfs_ops,
203 #endif
204 	.release = release_nbp,
205 };
206 
207 static void destroy_nbp(struct net_bridge_port *p)
208 {
209 	struct net_device *dev = p->dev;
210 
211 	p->br = NULL;
212 	p->dev = NULL;
213 	dev_put(dev);
214 
215 	kobject_put(&p->kobj);
216 }
217 
218 static void destroy_nbp_rcu(struct rcu_head *head)
219 {
220 	struct net_bridge_port *p =
221 			container_of(head, struct net_bridge_port, rcu);
222 	destroy_nbp(p);
223 }
224 
225 /* Delete port(interface) from bridge is done in two steps.
226  * via RCU. First step, marks device as down. That deletes
227  * all the timers and stops new packets from flowing through.
228  *
229  * Final cleanup doesn't occur until after all CPU's finished
230  * processing packets.
231  *
232  * Protected from multiple admin operations by RTNL mutex
233  */
234 static void del_nbp(struct net_bridge_port *p)
235 {
236 	struct net_bridge *br = p->br;
237 	struct net_device *dev = p->dev;
238 
239 	sysfs_remove_link(br->ifobj, p->dev->name);
240 
241 	nbp_delete_promisc(p);
242 
243 	spin_lock_bh(&br->lock);
244 	br_stp_disable_port(p);
245 	spin_unlock_bh(&br->lock);
246 
247 	br_ifinfo_notify(RTM_DELLINK, p);
248 
249 	list_del_rcu(&p->list);
250 
251 	nbp_vlan_flush(p);
252 	br_fdb_delete_by_port(br, p, 0, 1);
253 	nbp_update_port_count(br);
254 
255 	netdev_upper_dev_unlink(dev, br->dev);
256 
257 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
258 
259 	netdev_rx_handler_unregister(dev);
260 
261 	br_multicast_del_port(p);
262 
263 	kobject_uevent(&p->kobj, KOBJ_REMOVE);
264 	kobject_del(&p->kobj);
265 
266 	br_netpoll_disable(p);
267 
268 	call_rcu(&p->rcu, destroy_nbp_rcu);
269 }
270 
271 /* Delete bridge device */
272 void br_dev_delete(struct net_device *dev, struct list_head *head)
273 {
274 	struct net_bridge *br = netdev_priv(dev);
275 	struct net_bridge_port *p, *n;
276 
277 	list_for_each_entry_safe(p, n, &br->port_list, list) {
278 		del_nbp(p);
279 	}
280 
281 	br_fdb_delete_by_port(br, NULL, 0, 1);
282 
283 	br_vlan_flush(br);
284 	br_multicast_dev_del(br);
285 	del_timer_sync(&br->gc_timer);
286 
287 	br_sysfs_delbr(br->dev);
288 	unregister_netdevice_queue(br->dev, head);
289 }
290 
291 /* find an available port number */
292 static int find_portno(struct net_bridge *br)
293 {
294 	int index;
295 	struct net_bridge_port *p;
296 	unsigned long *inuse;
297 
298 	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
299 			GFP_KERNEL);
300 	if (!inuse)
301 		return -ENOMEM;
302 
303 	set_bit(0, inuse);	/* zero is reserved */
304 	list_for_each_entry(p, &br->port_list, list) {
305 		set_bit(p->port_no, inuse);
306 	}
307 	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
308 	kfree(inuse);
309 
310 	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
311 }
312 
313 /* called with RTNL but without bridge lock */
314 static struct net_bridge_port *new_nbp(struct net_bridge *br,
315 				       struct net_device *dev)
316 {
317 	int index;
318 	struct net_bridge_port *p;
319 
320 	index = find_portno(br);
321 	if (index < 0)
322 		return ERR_PTR(index);
323 
324 	p = kzalloc(sizeof(*p), GFP_KERNEL);
325 	if (p == NULL)
326 		return ERR_PTR(-ENOMEM);
327 
328 	p->br = br;
329 	dev_hold(dev);
330 	p->dev = dev;
331 	p->path_cost = port_cost(dev);
332 	p->priority = 0x8000 >> BR_PORT_BITS;
333 	p->port_no = index;
334 	p->flags = BR_LEARNING | BR_FLOOD;
335 	br_init_port(p);
336 	br_set_state(p, BR_STATE_DISABLED);
337 	br_stp_port_timer_init(p);
338 	br_multicast_add_port(p);
339 
340 	return p;
341 }
342 
343 int br_add_bridge(struct net *net, const char *name)
344 {
345 	struct net_device *dev;
346 	int res;
347 
348 	dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
349 			   br_dev_setup);
350 
351 	if (!dev)
352 		return -ENOMEM;
353 
354 	dev_net_set(dev, net);
355 	dev->rtnl_link_ops = &br_link_ops;
356 
357 	res = register_netdev(dev);
358 	if (res)
359 		free_netdev(dev);
360 	return res;
361 }
362 
363 int br_del_bridge(struct net *net, const char *name)
364 {
365 	struct net_device *dev;
366 	int ret = 0;
367 
368 	rtnl_lock();
369 	dev = __dev_get_by_name(net, name);
370 	if (dev == NULL)
371 		ret =  -ENXIO; 	/* Could not find device */
372 
373 	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
374 		/* Attempt to delete non bridge device! */
375 		ret = -EPERM;
376 	}
377 
378 	else if (dev->flags & IFF_UP) {
379 		/* Not shutdown yet. */
380 		ret = -EBUSY;
381 	}
382 
383 	else
384 		br_dev_delete(dev, NULL);
385 
386 	rtnl_unlock();
387 	return ret;
388 }
389 
390 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
391 int br_min_mtu(const struct net_bridge *br)
392 {
393 	const struct net_bridge_port *p;
394 	int mtu = 0;
395 
396 	ASSERT_RTNL();
397 
398 	if (list_empty(&br->port_list))
399 		mtu = ETH_DATA_LEN;
400 	else {
401 		list_for_each_entry(p, &br->port_list, list) {
402 			if (!mtu  || p->dev->mtu < mtu)
403 				mtu = p->dev->mtu;
404 		}
405 	}
406 	return mtu;
407 }
408 
409 /*
410  * Recomputes features using slave's features
411  */
412 netdev_features_t br_features_recompute(struct net_bridge *br,
413 	netdev_features_t features)
414 {
415 	struct net_bridge_port *p;
416 	netdev_features_t mask;
417 
418 	if (list_empty(&br->port_list))
419 		return features;
420 
421 	mask = features;
422 	features &= ~NETIF_F_ONE_FOR_ALL;
423 
424 	list_for_each_entry(p, &br->port_list, list) {
425 		features = netdev_increment_features(features,
426 						     p->dev->features, mask);
427 	}
428 	features = netdev_add_tso_features(features, mask);
429 
430 	return features;
431 }
432 
433 /* called with RTNL */
434 int br_add_if(struct net_bridge *br, struct net_device *dev)
435 {
436 	struct net_bridge_port *p;
437 	int err = 0;
438 	bool changed_addr;
439 
440 	/* Don't allow bridging non-ethernet like devices, or DSA-enabled
441 	 * master network devices since the bridge layer rx_handler prevents
442 	 * the DSA fake ethertype handler to be invoked, so we do not strip off
443 	 * the DSA switch tag protocol header and the bridge layer just return
444 	 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
445 	 */
446 	if ((dev->flags & IFF_LOOPBACK) ||
447 	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
448 	    !is_valid_ether_addr(dev->dev_addr) ||
449 	    netdev_uses_dsa(dev))
450 		return -EINVAL;
451 
452 	/* No bridging of bridges */
453 	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
454 		return -ELOOP;
455 
456 	/* Device is already being bridged */
457 	if (br_port_exists(dev))
458 		return -EBUSY;
459 
460 	/* No bridging devices that dislike that (e.g. wireless) */
461 	if (dev->priv_flags & IFF_DONT_BRIDGE)
462 		return -EOPNOTSUPP;
463 
464 	p = new_nbp(br, dev);
465 	if (IS_ERR(p))
466 		return PTR_ERR(p);
467 
468 	call_netdevice_notifiers(NETDEV_JOIN, dev);
469 
470 	err = dev_set_allmulti(dev, 1);
471 	if (err)
472 		goto put_back;
473 
474 	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
475 				   SYSFS_BRIDGE_PORT_ATTR);
476 	if (err)
477 		goto err1;
478 
479 	err = br_sysfs_addif(p);
480 	if (err)
481 		goto err2;
482 
483 	err = br_netpoll_enable(p);
484 	if (err)
485 		goto err3;
486 
487 	err = netdev_rx_handler_register(dev, br_handle_frame, p);
488 	if (err)
489 		goto err4;
490 
491 	dev->priv_flags |= IFF_BRIDGE_PORT;
492 
493 	err = netdev_master_upper_dev_link(dev, br->dev);
494 	if (err)
495 		goto err5;
496 
497 	dev_disable_lro(dev);
498 
499 	list_add_rcu(&p->list, &br->port_list);
500 
501 	nbp_update_port_count(br);
502 
503 	netdev_update_features(br->dev);
504 
505 	if (br->dev->needed_headroom < dev->needed_headroom)
506 		br->dev->needed_headroom = dev->needed_headroom;
507 
508 	if (br_fdb_insert(br, p, dev->dev_addr, 0))
509 		netdev_err(dev, "failed insert local address bridge forwarding table\n");
510 
511 	if (nbp_vlan_init(p))
512 		netdev_err(dev, "failed to initialize vlan filtering on this port\n");
513 
514 	spin_lock_bh(&br->lock);
515 	changed_addr = br_stp_recalculate_bridge_id(br);
516 
517 	if (netif_running(dev) && netif_oper_up(dev) &&
518 	    (br->dev->flags & IFF_UP))
519 		br_stp_enable_port(p);
520 	spin_unlock_bh(&br->lock);
521 
522 	br_ifinfo_notify(RTM_NEWLINK, p);
523 
524 	if (changed_addr)
525 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
526 
527 	dev_set_mtu(br->dev, br_min_mtu(br));
528 
529 	kobject_uevent(&p->kobj, KOBJ_ADD);
530 
531 	return 0;
532 
533 err5:
534 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
535 	netdev_rx_handler_unregister(dev);
536 err4:
537 	br_netpoll_disable(p);
538 err3:
539 	sysfs_remove_link(br->ifobj, p->dev->name);
540 err2:
541 	kobject_put(&p->kobj);
542 	p = NULL; /* kobject_put frees */
543 err1:
544 	dev_set_allmulti(dev, -1);
545 put_back:
546 	dev_put(dev);
547 	kfree(p);
548 	return err;
549 }
550 
551 /* called with RTNL */
552 int br_del_if(struct net_bridge *br, struct net_device *dev)
553 {
554 	struct net_bridge_port *p;
555 	bool changed_addr;
556 
557 	p = br_port_get_rtnl(dev);
558 	if (!p || p->br != br)
559 		return -EINVAL;
560 
561 	/* Since more than one interface can be attached to a bridge,
562 	 * there still maybe an alternate path for netconsole to use;
563 	 * therefore there is no reason for a NETDEV_RELEASE event.
564 	 */
565 	del_nbp(p);
566 
567 	dev_set_mtu(br->dev, br_min_mtu(br));
568 
569 	spin_lock_bh(&br->lock);
570 	changed_addr = br_stp_recalculate_bridge_id(br);
571 	spin_unlock_bh(&br->lock);
572 
573 	if (changed_addr)
574 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
575 
576 	netdev_update_features(br->dev);
577 
578 	return 0;
579 }
580 
581 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
582 {
583 	struct net_bridge *br = p->br;
584 
585 	if (mask & BR_AUTO_MASK)
586 		nbp_update_port_count(br);
587 }
588