/net/bridge/br_if.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 *	Userspace interface
  4 *	Linux ethernet bridge
  5 *
  6 *	Authors:
  7 *	Lennert Buytenhek		<buytenh@gnu.org>
  8 */
  9
 10#include <linux/kernel.h>
 11#include <linux/netdevice.h>
 12#include <linux/etherdevice.h>
 13#include <linux/netpoll.h>
 14#include <linux/ethtool.h>
 15#include <linux/if_arp.h>
 16#include <linux/module.h>
 17#include <linux/init.h>
 18#include <linux/rtnetlink.h>
 19#include <linux/if_ether.h>
 20#include <linux/slab.h>
 21#include <net/dsa.h>
 22#include <net/sock.h>
 23#include <linux/if_vlan.h>
 24#include <net/switchdev.h>
 25#include <net/net_namespace.h>
 26
 27#include "br_private.h"
 28
 29/*
 30 * Determine initial path cost based on speed.
 31 * using recommendations from 802.1d standard
 32 *
 33 * Since driver might sleep need to not be holding any locks.
 34 */
 35static int port_cost(struct net_device *dev)
 36{
 37	struct ethtool_link_ksettings ecmd;
 38
 39	if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
 40		switch (ecmd.base.speed) {
 41		case SPEED_10000:
 42			return 2;
 43		case SPEED_1000:
 44			return 4;
 45		case SPEED_100:
 46			return 19;
 47		case SPEED_10:
 48			return 100;
 49		}
 50	}
 51
 52	/* Old silly heuristics based on name */
 53	if (!strncmp(dev->name, "lec", 3))
 54		return 7;
 55
 56	if (!strncmp(dev->name, "plip", 4))
 57		return 2500;
 58
 59	return 100;	/* assume old 10Mbps */
 60}
 61
 62
 63/* Check for port carrier transitions. */
 64void br_port_carrier_check(struct net_bridge_port *p, bool *notified)
 65{
 66	struct net_device *dev = p->dev;
 67	struct net_bridge *br = p->br;
 68
 69	if (!(p->flags & BR_ADMIN_COST) &&
 70	    netif_running(dev) && netif_oper_up(dev))
 71		p->path_cost = port_cost(dev);
 72
 73	*notified = false;
 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			*notified = true;
 82		}
 83	} else {
 84		if (p->state != BR_STATE_DISABLED) {
 85			br_stp_disable_port(p);
 86			*notified = true;
 87		}
 88	}
 89	spin_unlock_bh(&br->lock);
 90}
 91
 92static void br_port_set_promisc(struct net_bridge_port *p)
 93{
 94	int err = 0;
 95
 96	if (br_promisc_port(p))
 97		return;
 98
 99	err = dev_set_promiscuity(p->dev, 1);
100	if (err)
101		return;
102
103	br_fdb_unsync_static(p->br, p);
104	p->flags |= BR_PROMISC;
105}
106
107static void br_port_clear_promisc(struct net_bridge_port *p)
108{
109	int err;
110
111	/* Check if the port is already non-promisc or if it doesn't
112	 * support UNICAST filtering.  Without unicast filtering support
113	 * we'll end up re-enabling promisc mode anyway, so just check for
114	 * it here.
115	 */
116	if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
117		return;
118
119	/* Since we'll be clearing the promisc mode, program the port
120	 * first so that we don't have interruption in traffic.
121	 */
122	err = br_fdb_sync_static(p->br, p);
123	if (err)
124		return;
125
126	dev_set_promiscuity(p->dev, -1);
127	p->flags &= ~BR_PROMISC;
128}
129
130/* When a port is added or removed or when certain port flags
131 * change, this function is called to automatically manage
132 * promiscuity setting of all the bridge ports.  We are always called
133 * under RTNL so can skip using rcu primitives.
134 */
135void br_manage_promisc(struct net_bridge *br)
136{
137	struct net_bridge_port *p;
138	bool set_all = false;
139
140	/* If vlan filtering is disabled or bridge interface is placed
141	 * into promiscuous mode, place all ports in promiscuous mode.
142	 */
143	if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev))
144		set_all = true;
145
146	list_for_each_entry(p, &br->port_list, list) {
147		if (set_all) {
148			br_port_set_promisc(p);
149		} else {
150			/* If the number of auto-ports is <= 1, then all other
151			 * ports will have their output configuration
152			 * statically specified through fdbs.  Since ingress
153			 * on the auto-port becomes forwarding/egress to other
154			 * ports and egress configuration is statically known,
155			 * we can say that ingress configuration of the
156			 * auto-port is also statically known.
157			 * This lets us disable promiscuous mode and write
158			 * this config to hw.
159			 */
160			if (br->auto_cnt == 0 ||
161			    (br->auto_cnt == 1 && br_auto_port(p)))
162				br_port_clear_promisc(p);
163			else
164				br_port_set_promisc(p);
165		}
166	}
167}
168
169int nbp_backup_change(struct net_bridge_port *p,
170		      struct net_device *backup_dev)
171{
172	struct net_bridge_port *old_backup = rtnl_dereference(p->backup_port);
173	struct net_bridge_port *backup_p = NULL;
174
175	ASSERT_RTNL();
176
177	if (backup_dev) {
178		if (!netif_is_bridge_port(backup_dev))
179			return -ENOENT;
180
181		backup_p = br_port_get_rtnl(backup_dev);
182		if (backup_p->br != p->br)
183			return -EINVAL;
184	}
185
186	if (p == backup_p)
187		return -EINVAL;
188
189	if (old_backup == backup_p)
190		return 0;
191
192	/* if the backup link is already set, clear it */
193	if (old_backup)
194		old_backup->backup_redirected_cnt--;
195
196	if (backup_p)
197		backup_p->backup_redirected_cnt++;
198	rcu_assign_pointer(p->backup_port, backup_p);
199
200	return 0;
201}
202
203static void nbp_backup_clear(struct net_bridge_port *p)
204{
205	nbp_backup_change(p, NULL);
206	if (p->backup_redirected_cnt) {
207		struct net_bridge_port *cur_p;
208
209		list_for_each_entry(cur_p, &p->br->port_list, list) {
210			struct net_bridge_port *backup_p;
211
212			backup_p = rtnl_dereference(cur_p->backup_port);
213			if (backup_p == p)
214				nbp_backup_change(cur_p, NULL);
215		}
216	}
217
218	WARN_ON(rcu_access_pointer(p->backup_port) || p->backup_redirected_cnt);
219}
220
221static void nbp_update_port_count(struct net_bridge *br)
222{
223	struct net_bridge_port *p;
224	u32 cnt = 0;
225
226	list_for_each_entry(p, &br->port_list, list) {
227		if (br_auto_port(p))
228			cnt++;
229	}
230	if (br->auto_cnt != cnt) {
231		br->auto_cnt = cnt;
232		br_manage_promisc(br);
233	}
234}
235
236static void nbp_delete_promisc(struct net_bridge_port *p)
237{
238	/* If port is currently promiscuous, unset promiscuity.
239	 * Otherwise, it is a static port so remove all addresses
240	 * from it.
241	 */
242	dev_set_allmulti(p->dev, -1);
243	if (br_promisc_port(p))
244		dev_set_promiscuity(p->dev, -1);
245	else
246		br_fdb_unsync_static(p->br, p);
247}
248
249static void release_nbp(struct kobject *kobj)
250{
251	struct net_bridge_port *p
252		= container_of(kobj, struct net_bridge_port, kobj);
253	kfree(p);
254}
255
256static void brport_get_ownership(struct kobject *kobj, kuid_t *uid, kgid_t *gid)
257{
258	struct net_bridge_port *p = kobj_to_brport(kobj);
259
260	net_ns_get_ownership(dev_net(p->dev), uid, gid);
261}
262
263static struct kobj_type brport_ktype = {
264#ifdef CONFIG_SYSFS
265	.sysfs_ops = &brport_sysfs_ops,
266#endif
267	.release = release_nbp,
268	.get_ownership = brport_get_ownership,
269};
270
271static void destroy_nbp(struct net_bridge_port *p)
272{
273	struct net_device *dev = p->dev;
274
275	p->br = NULL;
276	p->dev = NULL;
277	dev_put(dev);
278
279	kobject_put(&p->kobj);
280}
281
282static void destroy_nbp_rcu(struct rcu_head *head)
283{
284	struct net_bridge_port *p =
285			container_of(head, struct net_bridge_port, rcu);
286	destroy_nbp(p);
287}
288
289static unsigned get_max_headroom(struct net_bridge *br)
290{
291	unsigned max_headroom = 0;
292	struct net_bridge_port *p;
293
294	list_for_each_entry(p, &br->port_list, list) {
295		unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
296
297		if (dev_headroom > max_headroom)
298			max_headroom = dev_headroom;
299	}
300
301	return max_headroom;
302}
303
304static void update_headroom(struct net_bridge *br, int new_hr)
305{
306	struct net_bridge_port *p;
307
308	list_for_each_entry(p, &br->port_list, list)
309		netdev_set_rx_headroom(p->dev, new_hr);
310
311	br->dev->needed_headroom = new_hr;
312}
313
314/* Delete port(interface) from bridge is done in two steps.
315 * via RCU. First step, marks device as down. That deletes
316 * all the timers and stops new packets from flowing through.
317 *
318 * Final cleanup doesn't occur until after all CPU's finished
319 * processing packets.
320 *
321 * Protected from multiple admin operations by RTNL mutex
322 */
323static void del_nbp(struct net_bridge_port *p)
324{
325	struct net_bridge *br = p->br;
326	struct net_device *dev = p->dev;
327
328	sysfs_remove_link(br->ifobj, p->dev->name);
329
330	nbp_delete_promisc(p);
331
332	spin_lock_bh(&br->lock);
333	br_stp_disable_port(p);
334	spin_unlock_bh(&br->lock);
335
336	br_ifinfo_notify(RTM_DELLINK, NULL, p);
337
338	list_del_rcu(&p->list);
339	if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
340		update_headroom(br, get_max_headroom(br));
341	netdev_reset_rx_headroom(dev);
342
343	nbp_vlan_flush(p);
344	br_fdb_delete_by_port(br, p, 0, 1);
345	switchdev_deferred_process();
346	nbp_backup_clear(p);
347
348	nbp_update_port_count(br);
349
350	netdev_upper_dev_unlink(dev, br->dev);
351
352	dev->priv_flags &= ~IFF_BRIDGE_PORT;
353
354	netdev_rx_handler_unregister(dev);
355
356	br_multicast_del_port(p);
357
358	kobject_uevent(&p->kobj, KOBJ_REMOVE);
359	kobject_del(&p->kobj);
360
361	br_netpoll_disable(p);
362
363	call_rcu(&p->rcu, destroy_nbp_rcu);
364}
365
366/* Delete bridge device */
367void br_dev_delete(struct net_device *dev, struct list_head *head)
368{
369	struct net_bridge *br = netdev_priv(dev);
370	struct net_bridge_port *p, *n;
371
372	list_for_each_entry_safe(p, n, &br->port_list, list) {
373		del_nbp(p);
374	}
375
376	br_recalculate_neigh_suppress_enabled(br);
377
378	br_fdb_delete_by_port(br, NULL, 0, 1);
379
380	cancel_delayed_work_sync(&br->gc_work);
381
382	br_sysfs_delbr(br->dev);
383	unregister_netdevice_queue(br->dev, head);
384}
385
386/* find an available port number */
387static int find_portno(struct net_bridge *br)
388{
389	int index;
390	struct net_bridge_port *p;
391	unsigned long *inuse;
392
393	inuse = bitmap_zalloc(BR_MAX_PORTS, GFP_KERNEL);
394	if (!inuse)
395		return -ENOMEM;
396
397	set_bit(0, inuse);	/* zero is reserved */
398	list_for_each_entry(p, &br->port_list, list) {
399		set_bit(p->port_no, inuse);
400	}
401	index = find_first_zero_bit(inuse, BR_MAX_PORTS);
402	bitmap_free(inuse);
403
404	return (index >= BR_MAX_PORTS) ? -EXFULL : index;
405}
406
407/* called with RTNL but without bridge lock */
408static struct net_bridge_port *new_nbp(struct net_bridge *br,
409				       struct net_device *dev)
410{
411	struct net_bridge_port *p;
412	int index, err;
413
414	index = find_portno(br);
415	if (index < 0)
416		return ERR_PTR(index);
417
418	p = kzalloc(sizeof(*p), GFP_KERNEL);
419	if (p == NULL)
420		return ERR_PTR(-ENOMEM);
421
422	p->br = br;
423	dev_hold(dev);
424	p->dev = dev;
425	p->path_cost = port_cost(dev);
426	p->priority = 0x8000 >> BR_PORT_BITS;
427	p->port_no = index;
428	p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
429	br_init_port(p);
430	br_set_state(p, BR_STATE_DISABLED);
431	br_stp_port_timer_init(p);
432	err = br_multicast_add_port(p);
433	if (err) {
434		dev_put(dev);
435		kfree(p);
436		p = ERR_PTR(err);
437	}
438
439	return p;
440}
441
442int br_add_bridge(struct net *net, const char *name)
443{
444	struct net_device *dev;
445	int res;
446
447	dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
448			   br_dev_setup);
449
450	if (!dev)
451		return -ENOMEM;
452
453	dev_net_set(dev, net);
454	dev->rtnl_link_ops = &br_link_ops;
455
456	res = register_netdev(dev);
457	if (res)
458		free_netdev(dev);
459	return res;
460}
461
462int br_del_bridge(struct net *net, const char *name)
463{
464	struct net_device *dev;
465	int ret = 0;
466
467	rtnl_lock();
468	dev = __dev_get_by_name(net, name);
469	if (dev == NULL)
470		ret =  -ENXIO; 	/* Could not find device */
471
472	else if (!(dev->priv_flags & IFF_EBRIDGE)) {
473		/* Attempt to delete non bridge device! */
474		ret = -EPERM;
475	}
476
477	else if (dev->flags & IFF_UP) {
478		/* Not shutdown yet. */
479		ret = -EBUSY;
480	}
481
482	else
483		br_dev_delete(dev, NULL);
484
485	rtnl_unlock();
486	return ret;
487}
488
489/* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
490static int br_mtu_min(const struct net_bridge *br)
491{
492	const struct net_bridge_port *p;
493	int ret_mtu = 0;
494
495	list_for_each_entry(p, &br->port_list, list)
496		if (!ret_mtu || ret_mtu > p->dev->mtu)
497			ret_mtu = p->dev->mtu;
498
499	return ret_mtu ? ret_mtu : ETH_DATA_LEN;
500}
501
502void br_mtu_auto_adjust(struct net_bridge *br)
503{
504	ASSERT_RTNL();
505
506	/* if the bridge MTU was manually configured don't mess with it */
507	if (br_opt_get(br, BROPT_MTU_SET_BY_USER))
508		return;
509
510	/* change to the minimum MTU and clear the flag which was set by
511	 * the bridge ndo_change_mtu callback
512	 */
513	dev_set_mtu(br->dev, br_mtu_min(br));
514	br_opt_toggle(br, BROPT_MTU_SET_BY_USER, false);
515}
516
517static void br_set_gso_limits(struct net_bridge *br)
518{
519	unsigned int gso_max_size = GSO_MAX_SIZE;
520	u16 gso_max_segs = GSO_MAX_SEGS;
521	const struct net_bridge_port *p;
522
523	list_for_each_entry(p, &br->port_list, list) {
524		gso_max_size = min(gso_max_size, p->dev->gso_max_size);
525		gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
526	}
527	br->dev->gso_max_size = gso_max_size;
528	br->dev->gso_max_segs = gso_max_segs;
529}
530
531/*
532 * Recomputes features using slave's features
533 */
534netdev_features_t br_features_recompute(struct net_bridge *br,
535	netdev_features_t features)
536{
537	struct net_bridge_port *p;
538	netdev_features_t mask;
539
540	if (list_empty(&br->port_list))
541		return features;
542
543	mask = features;
544	features &= ~NETIF_F_ONE_FOR_ALL;
545
546	list_for_each_entry(p, &br->port_list, list) {
547		features = netdev_increment_features(features,
548						     p->dev->features, mask);
549	}
550	features = netdev_add_tso_features(features, mask);
551
552	return features;
553}
554
555/* called with RTNL */
556int br_add_if(struct net_bridge *br, struct net_device *dev,
557	      struct netlink_ext_ack *extack)
558{
559	struct net_bridge_port *p;
560	int err = 0;
561	unsigned br_hr, dev_hr;
562	bool changed_addr;
563
564	/* Don't allow bridging non-ethernet like devices, or DSA-enabled
565	 * master network devices since the bridge layer rx_handler prevents
566	 * the DSA fake ethertype handler to be invoked, so we do not strip off
567	 * the DSA switch tag protocol header and the bridge layer just return
568	 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
569	 */
570	if ((dev->flags & IFF_LOOPBACK) ||
571	    dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
572	    !is_valid_ether_addr(dev->dev_addr) ||
573	    netdev_uses_dsa(dev))
574		return -EINVAL;
575
576	/* No bridging of bridges */
577	if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) {
578		NL_SET_ERR_MSG(extack,
579			       "Can not enslave a bridge to a bridge");
580		return -ELOOP;
581	}
582
583	/* Device has master upper dev */
584	if (netdev_master_upper_dev_get(dev))
585		return -EBUSY;
586
587	/* No bridging devices that dislike that (e.g. wireless) */
588	if (dev->priv_flags & IFF_DONT_BRIDGE) {
589		NL_SET_ERR_MSG(extack,
590			       "Device does not allow enslaving to a bridge");
591		return -EOPNOTSUPP;
592	}
593
594	p = new_nbp(br, dev);
595	if (IS_ERR(p))
596		return PTR_ERR(p);
597
598	call_netdevice_notifiers(NETDEV_JOIN, dev);
599
600	err = dev_set_allmulti(dev, 1);
601	if (err) {
602		kfree(p);	/* kobject not yet init'd, manually free */
603		goto err1;
604	}
605
606	err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
607				   SYSFS_BRIDGE_PORT_ATTR);
608	if (err)
609		goto err2;
610
611	err = br_sysfs_addif(p);
612	if (err)
613		goto err2;
614
615	err = br_netpoll_enable(p);
616	if (err)
617		goto err3;
618
619	err = netdev_rx_handler_register(dev, br_handle_frame, p);
620	if (err)
621		goto err4;
622
623	dev->priv_flags |= IFF_BRIDGE_PORT;
624
625	err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
626	if (err)
627		goto err5;
628
629	err = nbp_switchdev_mark_set(p);
630	if (err)
631		goto err6;
632
633	dev_disable_lro(dev);
634
635	list_add_rcu(&p->list, &br->port_list);
636
637	nbp_update_port_count(br);
638
639	netdev_update_features(br->dev);
640
641	br_hr = br->dev->needed_headroom;
642	dev_hr = netdev_get_fwd_headroom(dev);
643	if (br_hr < dev_hr)
644		update_headroom(br, dev_hr);
645	else
646		netdev_set_rx_headroom(dev, br_hr);
647
648	if (br_fdb_insert(br, p, dev->dev_addr, 0))
649		netdev_err(dev, "failed insert local address bridge forwarding table\n");
650
651	if (br->dev->addr_assign_type != NET_ADDR_SET) {
652		/* Ask for permission to use this MAC address now, even if we
653		 * don't end up choosing it below.
654		 */
655		err = dev_pre_changeaddr_notify(br->dev, dev->dev_addr, extack);
656		if (err)
657			goto err7;
658	}
659
660	err = nbp_vlan_init(p, extack);
661	if (err) {
662		netdev_err(dev, "failed to initialize vlan filtering on this port\n");
663		goto err7;
664	}
665
666	spin_lock_bh(&br->lock);
667	changed_addr = br_stp_recalculate_bridge_id(br);
668
669	if (netif_running(dev) && netif_oper_up(dev) &&
670	    (br->dev->flags & IFF_UP))
671		br_stp_enable_port(p);
672	spin_unlock_bh(&br->lock);
673
674	br_ifinfo_notify(RTM_NEWLINK, NULL, p);
675
676	if (changed_addr)
677		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
678
679	br_mtu_auto_adjust(br);
680	br_set_gso_limits(br);
681
682	kobject_uevent(&p->kobj, KOBJ_ADD);
683
684	return 0;
685
686err7:
687	list_del_rcu(&p->list);
688	br_fdb_delete_by_port(br, p, 0, 1);
689	nbp_update_port_count(br);
690err6:
691	netdev_upper_dev_unlink(dev, br->dev);
692err5:
693	dev->priv_flags &= ~IFF_BRIDGE_PORT;
694	netdev_rx_handler_unregister(dev);
695err4:
696	br_netpoll_disable(p);
697err3:
698	sysfs_remove_link(br->ifobj, p->dev->name);
699err2:
700	kobject_put(&p->kobj);
701	dev_set_allmulti(dev, -1);
702err1:
703	dev_put(dev);
704	return err;
705}
706
707/* called with RTNL */
708int br_del_if(struct net_bridge *br, struct net_device *dev)
709{
710	struct net_bridge_port *p;
711	bool changed_addr;
712
713	p = br_port_get_rtnl(dev);
714	if (!p || p->br != br)
715		return -EINVAL;
716
717	/* Since more than one interface can be attached to a bridge,
718	 * there still maybe an alternate path for netconsole to use;
719	 * therefore there is no reason for a NETDEV_RELEASE event.
720	 */
721	del_nbp(p);
722
723	br_mtu_auto_adjust(br);
724	br_set_gso_limits(br);
725
726	spin_lock_bh(&br->lock);
727	changed_addr = br_stp_recalculate_bridge_id(br);
728	spin_unlock_bh(&br->lock);
729
730	if (changed_addr)
731		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
732
733	netdev_update_features(br->dev);
734
735	return 0;
736}
737
738void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
739{
740	struct net_bridge *br = p->br;
741
742	if (mask & BR_AUTO_MASK)
743		nbp_update_port_count(br);
744
745	if (mask & BR_NEIGH_SUPPRESS)
746		br_recalculate_neigh_suppress_enabled(br);
747}
748
749bool br_port_flag_is_set(const struct net_device *dev, unsigned long flag)
750{
751	struct net_bridge_port *p;
752
753	p = br_port_get_rtnl_rcu(dev);
754	if (!p)
755		return false;
756
757	return p->flags & flag;
758}
759EXPORT_SYMBOL_GPL(br_port_flag_is_set);