xref: /linux/net/bridge/br_if.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
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/dsa.h>
26 #include <net/sock.h>
27 #include <linux/if_vlan.h>
28 #include <net/switchdev.h>
29 
30 #include "br_private.h"
31 
32 /*
33  * Determine initial path cost based on speed.
34  * using recommendations from 802.1d standard
35  *
36  * Since driver might sleep need to not be holding any locks.
37  */
38 static int port_cost(struct net_device *dev)
39 {
40 	struct ethtool_link_ksettings ecmd;
41 
42 	if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
43 		switch (ecmd.base.speed) {
44 		case SPEED_10000:
45 			return 2;
46 		case SPEED_1000:
47 			return 4;
48 		case SPEED_100:
49 			return 19;
50 		case SPEED_10:
51 			return 100;
52 		}
53 	}
54 
55 	/* Old silly heuristics based on name */
56 	if (!strncmp(dev->name, "lec", 3))
57 		return 7;
58 
59 	if (!strncmp(dev->name, "plip", 4))
60 		return 2500;
61 
62 	return 100;	/* assume old 10Mbps */
63 }
64 
65 
66 /* Check for port carrier transitions. */
67 void br_port_carrier_check(struct net_bridge_port *p)
68 {
69 	struct net_device *dev = p->dev;
70 	struct net_bridge *br = p->br;
71 
72 	if (!(p->flags & BR_ADMIN_COST) &&
73 	    netif_running(dev) && netif_oper_up(dev))
74 		p->path_cost = port_cost(dev);
75 
76 	if (!netif_running(br->dev))
77 		return;
78 
79 	spin_lock_bh(&br->lock);
80 	if (netif_running(dev) && netif_oper_up(dev)) {
81 		if (p->state == BR_STATE_DISABLED)
82 			br_stp_enable_port(p);
83 	} else {
84 		if (p->state != BR_STATE_DISABLED)
85 			br_stp_disable_port(p);
86 	}
87 	spin_unlock_bh(&br->lock);
88 }
89 
90 static void br_port_set_promisc(struct net_bridge_port *p)
91 {
92 	int err = 0;
93 
94 	if (br_promisc_port(p))
95 		return;
96 
97 	err = dev_set_promiscuity(p->dev, 1);
98 	if (err)
99 		return;
100 
101 	br_fdb_unsync_static(p->br, p);
102 	p->flags |= BR_PROMISC;
103 }
104 
105 static void br_port_clear_promisc(struct net_bridge_port *p)
106 {
107 	int err;
108 
109 	/* Check if the port is already non-promisc or if it doesn't
110 	 * support UNICAST filtering.  Without unicast filtering support
111 	 * we'll end up re-enabling promisc mode anyway, so just check for
112 	 * it here.
113 	 */
114 	if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
115 		return;
116 
117 	/* Since we'll be clearing the promisc mode, program the port
118 	 * first so that we don't have interruption in traffic.
119 	 */
120 	err = br_fdb_sync_static(p->br, p);
121 	if (err)
122 		return;
123 
124 	dev_set_promiscuity(p->dev, -1);
125 	p->flags &= ~BR_PROMISC;
126 }
127 
128 /* When a port is added or removed or when certain port flags
129  * change, this function is called to automatically manage
130  * promiscuity setting of all the bridge ports.  We are always called
131  * under RTNL so can skip using rcu primitives.
132  */
133 void br_manage_promisc(struct net_bridge *br)
134 {
135 	struct net_bridge_port *p;
136 	bool set_all = false;
137 
138 	/* If vlan filtering is disabled or bridge interface is placed
139 	 * into promiscuous mode, place all ports in promiscuous mode.
140 	 */
141 	if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev))
142 		set_all = true;
143 
144 	list_for_each_entry(p, &br->port_list, list) {
145 		if (set_all) {
146 			br_port_set_promisc(p);
147 		} else {
148 			/* If the number of auto-ports is <= 1, then all other
149 			 * ports will have their output configuration
150 			 * statically specified through fdbs.  Since ingress
151 			 * on the auto-port becomes forwarding/egress to other
152 			 * ports and egress configuration is statically known,
153 			 * we can say that ingress configuration of the
154 			 * auto-port is also statically known.
155 			 * This lets us disable promiscuous mode and write
156 			 * this config to hw.
157 			 */
158 			if (br->auto_cnt == 0 ||
159 			    (br->auto_cnt == 1 && br_auto_port(p)))
160 				br_port_clear_promisc(p);
161 			else
162 				br_port_set_promisc(p);
163 		}
164 	}
165 }
166 
167 static void nbp_update_port_count(struct net_bridge *br)
168 {
169 	struct net_bridge_port *p;
170 	u32 cnt = 0;
171 
172 	list_for_each_entry(p, &br->port_list, list) {
173 		if (br_auto_port(p))
174 			cnt++;
175 	}
176 	if (br->auto_cnt != cnt) {
177 		br->auto_cnt = cnt;
178 		br_manage_promisc(br);
179 	}
180 }
181 
182 static void nbp_delete_promisc(struct net_bridge_port *p)
183 {
184 	/* If port is currently promiscuous, unset promiscuity.
185 	 * Otherwise, it is a static port so remove all addresses
186 	 * from it.
187 	 */
188 	dev_set_allmulti(p->dev, -1);
189 	if (br_promisc_port(p))
190 		dev_set_promiscuity(p->dev, -1);
191 	else
192 		br_fdb_unsync_static(p->br, p);
193 }
194 
195 static void release_nbp(struct kobject *kobj)
196 {
197 	struct net_bridge_port *p
198 		= container_of(kobj, struct net_bridge_port, kobj);
199 	kfree(p);
200 }
201 
202 static struct kobj_type brport_ktype = {
203 #ifdef CONFIG_SYSFS
204 	.sysfs_ops = &brport_sysfs_ops,
205 #endif
206 	.release = release_nbp,
207 };
208 
209 static void destroy_nbp(struct net_bridge_port *p)
210 {
211 	struct net_device *dev = p->dev;
212 
213 	p->br = NULL;
214 	p->dev = NULL;
215 	dev_put(dev);
216 
217 	kobject_put(&p->kobj);
218 }
219 
220 static void destroy_nbp_rcu(struct rcu_head *head)
221 {
222 	struct net_bridge_port *p =
223 			container_of(head, struct net_bridge_port, rcu);
224 	destroy_nbp(p);
225 }
226 
227 static unsigned get_max_headroom(struct net_bridge *br)
228 {
229 	unsigned max_headroom = 0;
230 	struct net_bridge_port *p;
231 
232 	list_for_each_entry(p, &br->port_list, list) {
233 		unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
234 
235 		if (dev_headroom > max_headroom)
236 			max_headroom = dev_headroom;
237 	}
238 
239 	return max_headroom;
240 }
241 
242 static void update_headroom(struct net_bridge *br, int new_hr)
243 {
244 	struct net_bridge_port *p;
245 
246 	list_for_each_entry(p, &br->port_list, list)
247 		netdev_set_rx_headroom(p->dev, new_hr);
248 
249 	br->dev->needed_headroom = new_hr;
250 }
251 
252 /* Delete port(interface) from bridge is done in two steps.
253  * via RCU. First step, marks device as down. That deletes
254  * all the timers and stops new packets from flowing through.
255  *
256  * Final cleanup doesn't occur until after all CPU's finished
257  * processing packets.
258  *
259  * Protected from multiple admin operations by RTNL mutex
260  */
261 static void del_nbp(struct net_bridge_port *p)
262 {
263 	struct net_bridge *br = p->br;
264 	struct net_device *dev = p->dev;
265 
266 	sysfs_remove_link(br->ifobj, p->dev->name);
267 
268 	nbp_delete_promisc(p);
269 
270 	spin_lock_bh(&br->lock);
271 	br_stp_disable_port(p);
272 	spin_unlock_bh(&br->lock);
273 
274 	br_ifinfo_notify(RTM_DELLINK, NULL, p);
275 
276 	list_del_rcu(&p->list);
277 	if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
278 		update_headroom(br, get_max_headroom(br));
279 	netdev_reset_rx_headroom(dev);
280 
281 	nbp_vlan_flush(p);
282 	br_fdb_delete_by_port(br, p, 0, 1);
283 	switchdev_deferred_process();
284 
285 	nbp_update_port_count(br);
286 
287 	netdev_upper_dev_unlink(dev, br->dev);
288 
289 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
290 
291 	netdev_rx_handler_unregister(dev);
292 
293 	br_multicast_del_port(p);
294 
295 	kobject_uevent(&p->kobj, KOBJ_REMOVE);
296 	kobject_del(&p->kobj);
297 
298 	br_netpoll_disable(p);
299 
300 	call_rcu(&p->rcu, destroy_nbp_rcu);
301 }
302 
303 /* Delete bridge device */
304 void br_dev_delete(struct net_device *dev, struct list_head *head)
305 {
306 	struct net_bridge *br = netdev_priv(dev);
307 	struct net_bridge_port *p, *n;
308 
309 	list_for_each_entry_safe(p, n, &br->port_list, list) {
310 		del_nbp(p);
311 	}
312 
313 	br_recalculate_neigh_suppress_enabled(br);
314 
315 	br_fdb_delete_by_port(br, NULL, 0, 1);
316 
317 	cancel_delayed_work_sync(&br->gc_work);
318 
319 	br_sysfs_delbr(br->dev);
320 	unregister_netdevice_queue(br->dev, head);
321 }
322 
323 /* find an available port number */
324 static int find_portno(struct net_bridge *br)
325 {
326 	int index;
327 	struct net_bridge_port *p;
328 	unsigned long *inuse;
329 
330 	inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
331 			GFP_KERNEL);
332 	if (!inuse)
333 		return -ENOMEM;
334 
335 	set_bit(0, inuse);	/* zero is reserved */
336 	list_for_each_entry(p, &br->port_list, list) {
337 		set_bit(p->port_no, inuse);
338 	}
339 	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
340 	kfree(inuse);
341 
342 	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
343 }
344 
345 /* called with RTNL but without bridge lock */
346 static struct net_bridge_port *new_nbp(struct net_bridge *br,
347 				       struct net_device *dev)
348 {
349 	struct net_bridge_port *p;
350 	int index, err;
351 
352 	index = find_portno(br);
353 	if (index < 0)
354 		return ERR_PTR(index);
355 
356 	p = kzalloc(sizeof(*p), GFP_KERNEL);
357 	if (p == NULL)
358 		return ERR_PTR(-ENOMEM);
359 
360 	p->br = br;
361 	dev_hold(dev);
362 	p->dev = dev;
363 	p->path_cost = port_cost(dev);
364 	p->priority = 0x8000 >> BR_PORT_BITS;
365 	p->port_no = index;
366 	p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
367 	br_init_port(p);
368 	br_set_state(p, BR_STATE_DISABLED);
369 	br_stp_port_timer_init(p);
370 	err = br_multicast_add_port(p);
371 	if (err) {
372 		dev_put(dev);
373 		kfree(p);
374 		p = ERR_PTR(err);
375 	}
376 
377 	return p;
378 }
379 
380 int br_add_bridge(struct net *net, const char *name)
381 {
382 	struct net_device *dev;
383 	int res;
384 
385 	dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
386 			   br_dev_setup);
387 
388 	if (!dev)
389 		return -ENOMEM;
390 
391 	dev_net_set(dev, net);
392 	dev->rtnl_link_ops = &br_link_ops;
393 
394 	res = register_netdev(dev);
395 	if (res)
396 		free_netdev(dev);
397 	return res;
398 }
399 
400 int br_del_bridge(struct net *net, const char *name)
401 {
402 	struct net_device *dev;
403 	int ret = 0;
404 
405 	rtnl_lock();
406 	dev = __dev_get_by_name(net, name);
407 	if (dev == NULL)
408 		ret =  -ENXIO; 	/* Could not find device */
409 
410 	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
411 		/* Attempt to delete non bridge device! */
412 		ret = -EPERM;
413 	}
414 
415 	else if (dev->flags & IFF_UP) {
416 		/* Not shutdown yet. */
417 		ret = -EBUSY;
418 	}
419 
420 	else
421 		br_dev_delete(dev, NULL);
422 
423 	rtnl_unlock();
424 	return ret;
425 }
426 
427 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
428 static int br_mtu_min(const struct net_bridge *br)
429 {
430 	const struct net_bridge_port *p;
431 	int ret_mtu = 0;
432 
433 	list_for_each_entry(p, &br->port_list, list)
434 		if (!ret_mtu || ret_mtu > p->dev->mtu)
435 			ret_mtu = p->dev->mtu;
436 
437 	return ret_mtu ? ret_mtu : ETH_DATA_LEN;
438 }
439 
440 void br_mtu_auto_adjust(struct net_bridge *br)
441 {
442 	ASSERT_RTNL();
443 
444 	/* if the bridge MTU was manually configured don't mess with it */
445 	if (br->mtu_set_by_user)
446 		return;
447 
448 	/* change to the minimum MTU and clear the flag which was set by
449 	 * the bridge ndo_change_mtu callback
450 	 */
451 	dev_set_mtu(br->dev, br_mtu_min(br));
452 	br->mtu_set_by_user = false;
453 }
454 
455 static void br_set_gso_limits(struct net_bridge *br)
456 {
457 	unsigned int gso_max_size = GSO_MAX_SIZE;
458 	u16 gso_max_segs = GSO_MAX_SEGS;
459 	const struct net_bridge_port *p;
460 
461 	list_for_each_entry(p, &br->port_list, list) {
462 		gso_max_size = min(gso_max_size, p->dev->gso_max_size);
463 		gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
464 	}
465 	br->dev->gso_max_size = gso_max_size;
466 	br->dev->gso_max_segs = gso_max_segs;
467 }
468 
469 /*
470  * Recomputes features using slave's features
471  */
472 netdev_features_t br_features_recompute(struct net_bridge *br,
473 	netdev_features_t features)
474 {
475 	struct net_bridge_port *p;
476 	netdev_features_t mask;
477 
478 	if (list_empty(&br->port_list))
479 		return features;
480 
481 	mask = features;
482 	features &= ~NETIF_F_ONE_FOR_ALL;
483 
484 	list_for_each_entry(p, &br->port_list, list) {
485 		features = netdev_increment_features(features,
486 						     p->dev->features, mask);
487 	}
488 	features = netdev_add_tso_features(features, mask);
489 
490 	return features;
491 }
492 
493 /* called with RTNL */
494 int br_add_if(struct net_bridge *br, struct net_device *dev,
495 	      struct netlink_ext_ack *extack)
496 {
497 	struct net_bridge_port *p;
498 	int err = 0;
499 	unsigned br_hr, dev_hr;
500 	bool changed_addr;
501 
502 	/* Don't allow bridging non-ethernet like devices, or DSA-enabled
503 	 * master network devices since the bridge layer rx_handler prevents
504 	 * the DSA fake ethertype handler to be invoked, so we do not strip off
505 	 * the DSA switch tag protocol header and the bridge layer just return
506 	 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
507 	 */
508 	if ((dev->flags & IFF_LOOPBACK) ||
509 	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
510 	    !is_valid_ether_addr(dev->dev_addr) ||
511 	    netdev_uses_dsa(dev))
512 		return -EINVAL;
513 
514 	/* No bridging of bridges */
515 	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) {
516 		NL_SET_ERR_MSG(extack,
517 			       "Can not enslave a bridge to a bridge");
518 		return -ELOOP;
519 	}
520 
521 	/* Device is already being bridged */
522 	if (br_port_exists(dev))
523 		return -EBUSY;
524 
525 	/* No bridging devices that dislike that (e.g. wireless) */
526 	if (dev->priv_flags & IFF_DONT_BRIDGE) {
527 		NL_SET_ERR_MSG(extack,
528 			       "Device does not allow enslaving to a bridge");
529 		return -EOPNOTSUPP;
530 	}
531 
532 	p = new_nbp(br, dev);
533 	if (IS_ERR(p))
534 		return PTR_ERR(p);
535 
536 	call_netdevice_notifiers(NETDEV_JOIN, dev);
537 
538 	err = dev_set_allmulti(dev, 1);
539 	if (err)
540 		goto put_back;
541 
542 	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
543 				   SYSFS_BRIDGE_PORT_ATTR);
544 	if (err)
545 		goto err1;
546 
547 	err = br_sysfs_addif(p);
548 	if (err)
549 		goto err2;
550 
551 	err = br_netpoll_enable(p);
552 	if (err)
553 		goto err3;
554 
555 	err = netdev_rx_handler_register(dev, br_handle_frame, p);
556 	if (err)
557 		goto err4;
558 
559 	dev->priv_flags |= IFF_BRIDGE_PORT;
560 
561 	err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
562 	if (err)
563 		goto err5;
564 
565 	err = nbp_switchdev_mark_set(p);
566 	if (err)
567 		goto err6;
568 
569 	dev_disable_lro(dev);
570 
571 	list_add_rcu(&p->list, &br->port_list);
572 
573 	nbp_update_port_count(br);
574 
575 	netdev_update_features(br->dev);
576 
577 	br_hr = br->dev->needed_headroom;
578 	dev_hr = netdev_get_fwd_headroom(dev);
579 	if (br_hr < dev_hr)
580 		update_headroom(br, dev_hr);
581 	else
582 		netdev_set_rx_headroom(dev, br_hr);
583 
584 	if (br_fdb_insert(br, p, dev->dev_addr, 0))
585 		netdev_err(dev, "failed insert local address bridge forwarding table\n");
586 
587 	err = nbp_vlan_init(p);
588 	if (err) {
589 		netdev_err(dev, "failed to initialize vlan filtering on this port\n");
590 		goto err7;
591 	}
592 
593 	spin_lock_bh(&br->lock);
594 	changed_addr = br_stp_recalculate_bridge_id(br);
595 
596 	if (netif_running(dev) && netif_oper_up(dev) &&
597 	    (br->dev->flags & IFF_UP))
598 		br_stp_enable_port(p);
599 	spin_unlock_bh(&br->lock);
600 
601 	br_ifinfo_notify(RTM_NEWLINK, NULL, p);
602 
603 	if (changed_addr)
604 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
605 
606 	br_mtu_auto_adjust(br);
607 	br_set_gso_limits(br);
608 
609 	kobject_uevent(&p->kobj, KOBJ_ADD);
610 
611 	return 0;
612 
613 err7:
614 	list_del_rcu(&p->list);
615 	br_fdb_delete_by_port(br, p, 0, 1);
616 	nbp_update_port_count(br);
617 err6:
618 	netdev_upper_dev_unlink(dev, br->dev);
619 err5:
620 	dev->priv_flags &= ~IFF_BRIDGE_PORT;
621 	netdev_rx_handler_unregister(dev);
622 err4:
623 	br_netpoll_disable(p);
624 err3:
625 	sysfs_remove_link(br->ifobj, p->dev->name);
626 err2:
627 	kobject_put(&p->kobj);
628 	p = NULL; /* kobject_put frees */
629 err1:
630 	dev_set_allmulti(dev, -1);
631 put_back:
632 	dev_put(dev);
633 	kfree(p);
634 	return err;
635 }
636 
637 /* called with RTNL */
638 int br_del_if(struct net_bridge *br, struct net_device *dev)
639 {
640 	struct net_bridge_port *p;
641 	bool changed_addr;
642 
643 	p = br_port_get_rtnl(dev);
644 	if (!p || p->br != br)
645 		return -EINVAL;
646 
647 	/* Since more than one interface can be attached to a bridge,
648 	 * there still maybe an alternate path for netconsole to use;
649 	 * therefore there is no reason for a NETDEV_RELEASE event.
650 	 */
651 	del_nbp(p);
652 
653 	br_mtu_auto_adjust(br);
654 	br_set_gso_limits(br);
655 
656 	spin_lock_bh(&br->lock);
657 	changed_addr = br_stp_recalculate_bridge_id(br);
658 	spin_unlock_bh(&br->lock);
659 
660 	if (changed_addr)
661 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
662 
663 	netdev_update_features(br->dev);
664 
665 	return 0;
666 }
667 
668 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
669 {
670 	struct net_bridge *br = p->br;
671 
672 	if (mask & BR_AUTO_MASK)
673 		nbp_update_port_count(br);
674 
675 	if (mask & BR_NEIGH_SUPPRESS)
676 		br_recalculate_neigh_suppress_enabled(br);
677 }
678