xref: /openbmc/linux/net/bridge/br_if.c (revision a06c488d)
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 #include <net/switchdev.h>
28 
29 #include "br_private.h"
30 
31 /*
32  * Determine initial path cost based on speed.
33  * using recommendations from 802.1d standard
34  *
35  * Since driver might sleep need to not be holding any locks.
36  */
37 static int port_cost(struct net_device *dev)
38 {
39 	struct ethtool_cmd ecmd;
40 
41 	if (!__ethtool_get_settings(dev, &ecmd)) {
42 		switch (ethtool_cmd_speed(&ecmd)) {
43 		case SPEED_10000:
44 			return 2;
45 		case SPEED_1000:
46 			return 4;
47 		case SPEED_100:
48 			return 19;
49 		case SPEED_10:
50 			return 100;
51 		}
52 	}
53 
54 	/* Old silly heuristics based on name */
55 	if (!strncmp(dev->name, "lec", 3))
56 		return 7;
57 
58 	if (!strncmp(dev->name, "plip", 4))
59 		return 2500;
60 
61 	return 100;	/* assume old 10Mbps */
62 }
63 
64 
65 /* Check for port carrier transitions. */
66 void br_port_carrier_check(struct net_bridge_port *p)
67 {
68 	struct net_device *dev = p->dev;
69 	struct net_bridge *br = p->br;
70 
71 	if (!(p->flags & BR_ADMIN_COST) &&
72 	    netif_running(dev) && netif_oper_up(dev))
73 		p->path_cost = port_cost(dev);
74 
75 	if (!netif_running(br->dev))
76 		return;
77 
78 	spin_lock_bh(&br->lock);
79 	if (netif_running(dev) && netif_oper_up(dev)) {
80 		if (p->state == BR_STATE_DISABLED)
81 			br_stp_enable_port(p);
82 	} else {
83 		if (p->state != BR_STATE_DISABLED)
84 			br_stp_disable_port(p);
85 	}
86 	spin_unlock_bh(&br->lock);
87 }
88 
89 static void br_port_set_promisc(struct net_bridge_port *p)
90 {
91 	int err = 0;
92 
93 	if (br_promisc_port(p))
94 		return;
95 
96 	err = dev_set_promiscuity(p->dev, 1);
97 	if (err)
98 		return;
99 
100 	br_fdb_unsync_static(p->br, p);
101 	p->flags |= BR_PROMISC;
102 }
103 
104 static void br_port_clear_promisc(struct net_bridge_port *p)
105 {
106 	int err;
107 
108 	/* Check if the port is already non-promisc or if it doesn't
109 	 * support UNICAST filtering.  Without unicast filtering support
110 	 * we'll end up re-enabling promisc mode anyway, so just check for
111 	 * it here.
112 	 */
113 	if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
114 		return;
115 
116 	/* Since we'll be clearing the promisc mode, program the port
117 	 * first so that we don't have interruption in traffic.
118 	 */
119 	err = br_fdb_sync_static(p->br, p);
120 	if (err)
121 		return;
122 
123 	dev_set_promiscuity(p->dev, -1);
124 	p->flags &= ~BR_PROMISC;
125 }
126 
127 /* When a port is added or removed or when certain port flags
128  * change, this function is called to automatically manage
129  * promiscuity setting of all the bridge ports.  We are always called
130  * under RTNL so can skip using rcu primitives.
131  */
132 void br_manage_promisc(struct net_bridge *br)
133 {
134 	struct net_bridge_port *p;
135 	bool set_all = false;
136 
137 	/* If vlan filtering is disabled or bridge interface is placed
138 	 * into promiscuous mode, place all ports in promiscuous mode.
139 	 */
140 	if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br))
141 		set_all = true;
142 
143 	list_for_each_entry(p, &br->port_list, list) {
144 		if (set_all) {
145 			br_port_set_promisc(p);
146 		} else {
147 			/* If the number of auto-ports is <= 1, then all other
148 			 * ports will have their output configuration
149 			 * statically specified through fdbs.  Since ingress
150 			 * on the auto-port becomes forwarding/egress to other
151 			 * ports and egress configuration is statically known,
152 			 * we can say that ingress configuration of the
153 			 * auto-port is also statically known.
154 			 * This lets us disable promiscuous mode and write
155 			 * this config to hw.
156 			 */
157 			if (br->auto_cnt == 0 ||
158 			    (br->auto_cnt == 1 && br_auto_port(p)))
159 				br_port_clear_promisc(p);
160 			else
161 				br_port_set_promisc(p);
162 		}
163 	}
164 }
165 
166 static void nbp_update_port_count(struct net_bridge *br)
167 {
168 	struct net_bridge_port *p;
169 	u32 cnt = 0;
170 
171 	list_for_each_entry(p, &br->port_list, list) {
172 		if (br_auto_port(p))
173 			cnt++;
174 	}
175 	if (br->auto_cnt != cnt) {
176 		br->auto_cnt = cnt;
177 		br_manage_promisc(br);
178 	}
179 }
180 
181 static void nbp_delete_promisc(struct net_bridge_port *p)
182 {
183 	/* If port is currently promiscuous, unset promiscuity.
184 	 * Otherwise, it is a static port so remove all addresses
185 	 * from it.
186 	 */
187 	dev_set_allmulti(p->dev, -1);
188 	if (br_promisc_port(p))
189 		dev_set_promiscuity(p->dev, -1);
190 	else
191 		br_fdb_unsync_static(p->br, p);
192 }
193 
194 static void release_nbp(struct kobject *kobj)
195 {
196 	struct net_bridge_port *p
197 		= container_of(kobj, struct net_bridge_port, kobj);
198 	kfree(p);
199 }
200 
201 static struct kobj_type brport_ktype = {
202 #ifdef CONFIG_SYSFS
203 	.sysfs_ops = &brport_sysfs_ops,
204 #endif
205 	.release = release_nbp,
206 };
207 
208 static void destroy_nbp(struct net_bridge_port *p)
209 {
210 	struct net_device *dev = p->dev;
211 
212 	p->br = NULL;
213 	p->dev = NULL;
214 	dev_put(dev);
215 
216 	kobject_put(&p->kobj);
217 }
218 
219 static void destroy_nbp_rcu(struct rcu_head *head)
220 {
221 	struct net_bridge_port *p =
222 			container_of(head, struct net_bridge_port, rcu);
223 	destroy_nbp(p);
224 }
225 
226 /* Delete port(interface) from bridge is done in two steps.
227  * via RCU. First step, marks device as down. That deletes
228  * all the timers and stops new packets from flowing through.
229  *
230  * Final cleanup doesn't occur until after all CPU's finished
231  * processing packets.
232  *
233  * Protected from multiple admin operations by RTNL mutex
234  */
235 static void del_nbp(struct net_bridge_port *p)
236 {
237 	struct net_bridge *br = p->br;
238 	struct net_device *dev = p->dev;
239 
240 	sysfs_remove_link(br->ifobj, p->dev->name);
241 
242 	nbp_delete_promisc(p);
243 
244 	spin_lock_bh(&br->lock);
245 	br_stp_disable_port(p);
246 	spin_unlock_bh(&br->lock);
247 
248 	br_ifinfo_notify(RTM_DELLINK, p);
249 
250 	list_del_rcu(&p->list);
251 
252 	nbp_vlan_flush(p);
253 	br_fdb_delete_by_port(br, p, 0, 1);
254 	switchdev_deferred_process();
255 
256 	nbp_update_port_count(br);
257 
258 	netdev_upper_dev_unlink(dev, br->dev);
259 
260 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
261 
262 	netdev_rx_handler_unregister(dev);
263 
264 	br_multicast_del_port(p);
265 
266 	kobject_uevent(&p->kobj, KOBJ_REMOVE);
267 	kobject_del(&p->kobj);
268 
269 	br_netpoll_disable(p);
270 
271 	call_rcu(&p->rcu, destroy_nbp_rcu);
272 }
273 
274 /* Delete bridge device */
275 void br_dev_delete(struct net_device *dev, struct list_head *head)
276 {
277 	struct net_bridge *br = netdev_priv(dev);
278 	struct net_bridge_port *p, *n;
279 
280 	list_for_each_entry_safe(p, n, &br->port_list, list) {
281 		del_nbp(p);
282 	}
283 
284 	br_fdb_delete_by_port(br, NULL, 0, 1);
285 
286 	br_vlan_flush(br);
287 	br_multicast_dev_del(br);
288 	del_timer_sync(&br->gc_timer);
289 
290 	br_sysfs_delbr(br->dev);
291 	unregister_netdevice_queue(br->dev, head);
292 }
293 
294 /* find an available port number */
295 static int find_portno(struct net_bridge *br)
296 {
297 	int index;
298 	struct net_bridge_port *p;
299 	unsigned long *inuse;
300 
301 	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
302 			GFP_KERNEL);
303 	if (!inuse)
304 		return -ENOMEM;
305 
306 	set_bit(0, inuse);	/* zero is reserved */
307 	list_for_each_entry(p, &br->port_list, list) {
308 		set_bit(p->port_no, inuse);
309 	}
310 	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
311 	kfree(inuse);
312 
313 	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
314 }
315 
316 /* called with RTNL but without bridge lock */
317 static struct net_bridge_port *new_nbp(struct net_bridge *br,
318 				       struct net_device *dev)
319 {
320 	int index;
321 	struct net_bridge_port *p;
322 
323 	index = find_portno(br);
324 	if (index < 0)
325 		return ERR_PTR(index);
326 
327 	p = kzalloc(sizeof(*p), GFP_KERNEL);
328 	if (p == NULL)
329 		return ERR_PTR(-ENOMEM);
330 
331 	p->br = br;
332 	dev_hold(dev);
333 	p->dev = dev;
334 	p->path_cost = port_cost(dev);
335 	p->priority = 0x8000 >> BR_PORT_BITS;
336 	p->port_no = index;
337 	p->flags = BR_LEARNING | BR_FLOOD;
338 	br_init_port(p);
339 	br_set_state(p, BR_STATE_DISABLED);
340 	br_stp_port_timer_init(p);
341 	br_multicast_add_port(p);
342 
343 	return p;
344 }
345 
346 int br_add_bridge(struct net *net, const char *name)
347 {
348 	struct net_device *dev;
349 	int res;
350 
351 	dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
352 			   br_dev_setup);
353 
354 	if (!dev)
355 		return -ENOMEM;
356 
357 	dev_net_set(dev, net);
358 	dev->rtnl_link_ops = &br_link_ops;
359 
360 	res = register_netdev(dev);
361 	if (res)
362 		free_netdev(dev);
363 	return res;
364 }
365 
366 int br_del_bridge(struct net *net, const char *name)
367 {
368 	struct net_device *dev;
369 	int ret = 0;
370 
371 	rtnl_lock();
372 	dev = __dev_get_by_name(net, name);
373 	if (dev == NULL)
374 		ret =  -ENXIO; 	/* Could not find device */
375 
376 	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
377 		/* Attempt to delete non bridge device! */
378 		ret = -EPERM;
379 	}
380 
381 	else if (dev->flags & IFF_UP) {
382 		/* Not shutdown yet. */
383 		ret = -EBUSY;
384 	}
385 
386 	else
387 		br_dev_delete(dev, NULL);
388 
389 	rtnl_unlock();
390 	return ret;
391 }
392 
393 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
394 int br_min_mtu(const struct net_bridge *br)
395 {
396 	const struct net_bridge_port *p;
397 	int mtu = 0;
398 
399 	ASSERT_RTNL();
400 
401 	if (list_empty(&br->port_list))
402 		mtu = ETH_DATA_LEN;
403 	else {
404 		list_for_each_entry(p, &br->port_list, list) {
405 			if (!mtu  || p->dev->mtu < mtu)
406 				mtu = p->dev->mtu;
407 		}
408 	}
409 	return mtu;
410 }
411 
412 /*
413  * Recomputes features using slave's features
414  */
415 netdev_features_t br_features_recompute(struct net_bridge *br,
416 	netdev_features_t features)
417 {
418 	struct net_bridge_port *p;
419 	netdev_features_t mask;
420 
421 	if (list_empty(&br->port_list))
422 		return features;
423 
424 	mask = features;
425 	features &= ~NETIF_F_ONE_FOR_ALL;
426 
427 	list_for_each_entry(p, &br->port_list, list) {
428 		features = netdev_increment_features(features,
429 						     p->dev->features, mask);
430 	}
431 	features = netdev_add_tso_features(features, mask);
432 
433 	return features;
434 }
435 
436 /* called with RTNL */
437 int br_add_if(struct net_bridge *br, struct net_device *dev)
438 {
439 	struct net_bridge_port *p;
440 	int err = 0;
441 	bool changed_addr;
442 
443 	/* Don't allow bridging non-ethernet like devices, or DSA-enabled
444 	 * master network devices since the bridge layer rx_handler prevents
445 	 * the DSA fake ethertype handler to be invoked, so we do not strip off
446 	 * the DSA switch tag protocol header and the bridge layer just return
447 	 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
448 	 */
449 	if ((dev->flags & IFF_LOOPBACK) ||
450 	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
451 	    !is_valid_ether_addr(dev->dev_addr) ||
452 	    netdev_uses_dsa(dev))
453 		return -EINVAL;
454 
455 	/* No bridging of bridges */
456 	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
457 		return -ELOOP;
458 
459 	/* Device is already being bridged */
460 	if (br_port_exists(dev))
461 		return -EBUSY;
462 
463 	/* No bridging devices that dislike that (e.g. wireless) */
464 	if (dev->priv_flags & IFF_DONT_BRIDGE)
465 		return -EOPNOTSUPP;
466 
467 	p = new_nbp(br, dev);
468 	if (IS_ERR(p))
469 		return PTR_ERR(p);
470 
471 	call_netdevice_notifiers(NETDEV_JOIN, dev);
472 
473 	err = dev_set_allmulti(dev, 1);
474 	if (err)
475 		goto put_back;
476 
477 	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
478 				   SYSFS_BRIDGE_PORT_ATTR);
479 	if (err)
480 		goto err1;
481 
482 	err = br_sysfs_addif(p);
483 	if (err)
484 		goto err2;
485 
486 	err = br_netpoll_enable(p);
487 	if (err)
488 		goto err3;
489 
490 	err = netdev_rx_handler_register(dev, br_handle_frame, p);
491 	if (err)
492 		goto err4;
493 
494 	dev->priv_flags |= IFF_BRIDGE_PORT;
495 
496 	err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL);
497 	if (err)
498 		goto err5;
499 
500 	dev_disable_lro(dev);
501 
502 	list_add_rcu(&p->list, &br->port_list);
503 
504 	nbp_update_port_count(br);
505 
506 	netdev_update_features(br->dev);
507 
508 	if (br->dev->needed_headroom < dev->needed_headroom)
509 		br->dev->needed_headroom = dev->needed_headroom;
510 
511 	if (br_fdb_insert(br, p, dev->dev_addr, 0))
512 		netdev_err(dev, "failed insert local address bridge forwarding table\n");
513 
514 	err = nbp_vlan_init(p);
515 	if (err) {
516 		netdev_err(dev, "failed to initialize vlan filtering on this port\n");
517 		goto err6;
518 	}
519 
520 	spin_lock_bh(&br->lock);
521 	changed_addr = br_stp_recalculate_bridge_id(br);
522 
523 	if (netif_running(dev) && netif_oper_up(dev) &&
524 	    (br->dev->flags & IFF_UP))
525 		br_stp_enable_port(p);
526 	spin_unlock_bh(&br->lock);
527 
528 	br_ifinfo_notify(RTM_NEWLINK, p);
529 
530 	if (changed_addr)
531 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
532 
533 	dev_set_mtu(br->dev, br_min_mtu(br));
534 
535 	kobject_uevent(&p->kobj, KOBJ_ADD);
536 
537 	return 0;
538 
539 err6:
540 	list_del_rcu(&p->list);
541 	br_fdb_delete_by_port(br, p, 0, 1);
542 	nbp_update_port_count(br);
543 	netdev_upper_dev_unlink(dev, br->dev);
544 
545 err5:
546 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
547 	netdev_rx_handler_unregister(dev);
548 err4:
549 	br_netpoll_disable(p);
550 err3:
551 	sysfs_remove_link(br->ifobj, p->dev->name);
552 err2:
553 	kobject_put(&p->kobj);
554 	p = NULL; /* kobject_put frees */
555 err1:
556 	dev_set_allmulti(dev, -1);
557 put_back:
558 	dev_put(dev);
559 	kfree(p);
560 	return err;
561 }
562 
563 /* called with RTNL */
564 int br_del_if(struct net_bridge *br, struct net_device *dev)
565 {
566 	struct net_bridge_port *p;
567 	bool changed_addr;
568 
569 	p = br_port_get_rtnl(dev);
570 	if (!p || p->br != br)
571 		return -EINVAL;
572 
573 	/* Since more than one interface can be attached to a bridge,
574 	 * there still maybe an alternate path for netconsole to use;
575 	 * therefore there is no reason for a NETDEV_RELEASE event.
576 	 */
577 	del_nbp(p);
578 
579 	dev_set_mtu(br->dev, br_min_mtu(br));
580 
581 	spin_lock_bh(&br->lock);
582 	changed_addr = br_stp_recalculate_bridge_id(br);
583 	spin_unlock_bh(&br->lock);
584 
585 	if (changed_addr)
586 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
587 
588 	netdev_update_features(br->dev);
589 
590 	return 0;
591 }
592 
593 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
594 {
595 	struct net_bridge *br = p->br;
596 
597 	if (mask & BR_AUTO_MASK)
598 		nbp_update_port_count(br);
599 }
600