Loading...
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_5000:
44 return 3;
45 case SPEED_2500:
46 return 4;
47 case SPEED_1000:
48 return 5;
49 case SPEED_100:
50 return 19;
51 case SPEED_10:
52 return 100;
53 case SPEED_UNKNOWN:
54 return 100;
55 default:
56 if (ecmd.base.speed > SPEED_10000)
57 return 1;
58 }
59 }
60
61 /* Old silly heuristics based on name */
62 if (!strncmp(dev->name, "lec", 3))
63 return 7;
64
65 if (!strncmp(dev->name, "plip", 4))
66 return 2500;
67
68 return 100; /* assume old 10Mbps */
69}
70
71
72/* Check for port carrier transitions. */
73void br_port_carrier_check(struct net_bridge_port *p, bool *notified)
74{
75 struct net_device *dev = p->dev;
76 struct net_bridge *br = p->br;
77
78 if (!(p->flags & BR_ADMIN_COST) &&
79 netif_running(dev) && netif_oper_up(dev))
80 p->path_cost = port_cost(dev);
81
82 *notified = false;
83 if (!netif_running(br->dev))
84 return;
85
86 spin_lock_bh(&br->lock);
87 if (netif_running(dev) && netif_oper_up(dev)) {
88 if (p->state == BR_STATE_DISABLED) {
89 br_stp_enable_port(p);
90 *notified = true;
91 }
92 } else {
93 if (p->state != BR_STATE_DISABLED) {
94 br_stp_disable_port(p);
95 *notified = true;
96 }
97 }
98 spin_unlock_bh(&br->lock);
99}
100
101static void br_port_set_promisc(struct net_bridge_port *p)
102{
103 int err = 0;
104
105 if (br_promisc_port(p))
106 return;
107
108 err = dev_set_promiscuity(p->dev, 1);
109 if (err)
110 return;
111
112 br_fdb_unsync_static(p->br, p);
113 p->flags |= BR_PROMISC;
114}
115
116static void br_port_clear_promisc(struct net_bridge_port *p)
117{
118 int err;
119
120 /* Check if the port is already non-promisc or if it doesn't
121 * support UNICAST filtering. Without unicast filtering support
122 * we'll end up re-enabling promisc mode anyway, so just check for
123 * it here.
124 */
125 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
126 return;
127
128 /* Since we'll be clearing the promisc mode, program the port
129 * first so that we don't have interruption in traffic.
130 */
131 err = br_fdb_sync_static(p->br, p);
132 if (err)
133 return;
134
135 dev_set_promiscuity(p->dev, -1);
136 p->flags &= ~BR_PROMISC;
137}
138
139/* When a port is added or removed or when certain port flags
140 * change, this function is called to automatically manage
141 * promiscuity setting of all the bridge ports. We are always called
142 * under RTNL so can skip using rcu primitives.
143 */
144void br_manage_promisc(struct net_bridge *br)
145{
146 struct net_bridge_port *p;
147 bool set_all = false;
148
149 /* If vlan filtering is disabled or bridge interface is placed
150 * into promiscuous mode, place all ports in promiscuous mode.
151 */
152 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev))
153 set_all = true;
154
155 list_for_each_entry(p, &br->port_list, list) {
156 if (set_all) {
157 br_port_set_promisc(p);
158 } else {
159 /* If the number of auto-ports is <= 1, then all other
160 * ports will have their output configuration
161 * statically specified through fdbs. Since ingress
162 * on the auto-port becomes forwarding/egress to other
163 * ports and egress configuration is statically known,
164 * we can say that ingress configuration of the
165 * auto-port is also statically known.
166 * This lets us disable promiscuous mode and write
167 * this config to hw.
168 */
169 if (br->auto_cnt == 0 ||
170 (br->auto_cnt == 1 && br_auto_port(p)))
171 br_port_clear_promisc(p);
172 else
173 br_port_set_promisc(p);
174 }
175 }
176}
177
178int nbp_backup_change(struct net_bridge_port *p,
179 struct net_device *backup_dev)
180{
181 struct net_bridge_port *old_backup = rtnl_dereference(p->backup_port);
182 struct net_bridge_port *backup_p = NULL;
183
184 ASSERT_RTNL();
185
186 if (backup_dev) {
187 if (!netif_is_bridge_port(backup_dev))
188 return -ENOENT;
189
190 backup_p = br_port_get_rtnl(backup_dev);
191 if (backup_p->br != p->br)
192 return -EINVAL;
193 }
194
195 if (p == backup_p)
196 return -EINVAL;
197
198 if (old_backup == backup_p)
199 return 0;
200
201 /* if the backup link is already set, clear it */
202 if (old_backup)
203 old_backup->backup_redirected_cnt--;
204
205 if (backup_p)
206 backup_p->backup_redirected_cnt++;
207 rcu_assign_pointer(p->backup_port, backup_p);
208
209 return 0;
210}
211
212static void nbp_backup_clear(struct net_bridge_port *p)
213{
214 nbp_backup_change(p, NULL);
215 if (p->backup_redirected_cnt) {
216 struct net_bridge_port *cur_p;
217
218 list_for_each_entry(cur_p, &p->br->port_list, list) {
219 struct net_bridge_port *backup_p;
220
221 backup_p = rtnl_dereference(cur_p->backup_port);
222 if (backup_p == p)
223 nbp_backup_change(cur_p, NULL);
224 }
225 }
226
227 WARN_ON(rcu_access_pointer(p->backup_port) || p->backup_redirected_cnt);
228}
229
230static void nbp_update_port_count(struct net_bridge *br)
231{
232 struct net_bridge_port *p;
233 u32 cnt = 0;
234
235 list_for_each_entry(p, &br->port_list, list) {
236 if (br_auto_port(p))
237 cnt++;
238 }
239 if (br->auto_cnt != cnt) {
240 br->auto_cnt = cnt;
241 br_manage_promisc(br);
242 }
243}
244
245static void nbp_delete_promisc(struct net_bridge_port *p)
246{
247 /* If port is currently promiscuous, unset promiscuity.
248 * Otherwise, it is a static port so remove all addresses
249 * from it.
250 */
251 dev_set_allmulti(p->dev, -1);
252 if (br_promisc_port(p))
253 dev_set_promiscuity(p->dev, -1);
254 else
255 br_fdb_unsync_static(p->br, p);
256}
257
258static void release_nbp(struct kobject *kobj)
259{
260 struct net_bridge_port *p
261 = container_of(kobj, struct net_bridge_port, kobj);
262 kfree(p);
263}
264
265static void brport_get_ownership(const struct kobject *kobj, kuid_t *uid, kgid_t *gid)
266{
267 struct net_bridge_port *p = kobj_to_brport(kobj);
268
269 net_ns_get_ownership(dev_net(p->dev), uid, gid);
270}
271
272static struct kobj_type brport_ktype = {
273#ifdef CONFIG_SYSFS
274 .sysfs_ops = &brport_sysfs_ops,
275#endif
276 .release = release_nbp,
277 .get_ownership = brport_get_ownership,
278};
279
280static void destroy_nbp(struct net_bridge_port *p)
281{
282 struct net_device *dev = p->dev;
283
284 p->br = NULL;
285 p->dev = NULL;
286 netdev_put(dev, &p->dev_tracker);
287
288 kobject_put(&p->kobj);
289}
290
291static void destroy_nbp_rcu(struct rcu_head *head)
292{
293 struct net_bridge_port *p =
294 container_of(head, struct net_bridge_port, rcu);
295 destroy_nbp(p);
296}
297
298static unsigned get_max_headroom(struct net_bridge *br)
299{
300 unsigned max_headroom = 0;
301 struct net_bridge_port *p;
302
303 list_for_each_entry(p, &br->port_list, list) {
304 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
305
306 if (dev_headroom > max_headroom)
307 max_headroom = dev_headroom;
308 }
309
310 return max_headroom;
311}
312
313static void update_headroom(struct net_bridge *br, int new_hr)
314{
315 struct net_bridge_port *p;
316
317 list_for_each_entry(p, &br->port_list, list)
318 netdev_set_rx_headroom(p->dev, new_hr);
319
320 br->dev->needed_headroom = new_hr;
321}
322
323/* Delete port(interface) from bridge is done in two steps.
324 * via RCU. First step, marks device as down. That deletes
325 * all the timers and stops new packets from flowing through.
326 *
327 * Final cleanup doesn't occur until after all CPU's finished
328 * processing packets.
329 *
330 * Protected from multiple admin operations by RTNL mutex
331 */
332static void del_nbp(struct net_bridge_port *p)
333{
334 struct net_bridge *br = p->br;
335 struct net_device *dev = p->dev;
336
337 sysfs_remove_link(br->ifobj, p->dev->name);
338
339 nbp_delete_promisc(p);
340
341 spin_lock_bh(&br->lock);
342 br_stp_disable_port(p);
343 spin_unlock_bh(&br->lock);
344
345 br_mrp_port_del(br, p);
346 br_cfm_port_del(br, p);
347
348 br_ifinfo_notify(RTM_DELLINK, NULL, p);
349
350 list_del_rcu(&p->list);
351 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
352 update_headroom(br, get_max_headroom(br));
353 netdev_reset_rx_headroom(dev);
354
355 nbp_vlan_flush(p);
356 br_fdb_delete_by_port(br, p, 0, 1);
357 switchdev_deferred_process();
358 nbp_backup_clear(p);
359
360 nbp_update_port_count(br);
361
362 netdev_upper_dev_unlink(dev, br->dev);
363
364 dev->priv_flags &= ~IFF_BRIDGE_PORT;
365
366 netdev_rx_handler_unregister(dev);
367
368 br_multicast_del_port(p);
369
370 kobject_uevent(&p->kobj, KOBJ_REMOVE);
371 kobject_del(&p->kobj);
372
373 br_netpoll_disable(p);
374
375 call_rcu(&p->rcu, destroy_nbp_rcu);
376}
377
378/* Delete bridge device */
379void br_dev_delete(struct net_device *dev, struct list_head *head)
380{
381 struct net_bridge *br = netdev_priv(dev);
382 struct net_bridge_port *p, *n;
383
384 list_for_each_entry_safe(p, n, &br->port_list, list) {
385 del_nbp(p);
386 }
387
388 br_recalculate_neigh_suppress_enabled(br);
389
390 br_fdb_delete_by_port(br, NULL, 0, 1);
391
392 cancel_delayed_work_sync(&br->gc_work);
393
394 br_sysfs_delbr(br->dev);
395 unregister_netdevice_queue(br->dev, head);
396}
397
398/* find an available port number */
399static int find_portno(struct net_bridge *br)
400{
401 int index;
402 struct net_bridge_port *p;
403 unsigned long *inuse;
404
405 inuse = bitmap_zalloc(BR_MAX_PORTS, GFP_KERNEL);
406 if (!inuse)
407 return -ENOMEM;
408
409 __set_bit(0, inuse); /* zero is reserved */
410 list_for_each_entry(p, &br->port_list, list)
411 __set_bit(p->port_no, inuse);
412
413 index = find_first_zero_bit(inuse, BR_MAX_PORTS);
414 bitmap_free(inuse);
415
416 return (index >= BR_MAX_PORTS) ? -EXFULL : index;
417}
418
419/* called with RTNL but without bridge lock */
420static struct net_bridge_port *new_nbp(struct net_bridge *br,
421 struct net_device *dev)
422{
423 struct net_bridge_port *p;
424 int index, err;
425
426 index = find_portno(br);
427 if (index < 0)
428 return ERR_PTR(index);
429
430 p = kzalloc(sizeof(*p), GFP_KERNEL);
431 if (p == NULL)
432 return ERR_PTR(-ENOMEM);
433
434 p->br = br;
435 netdev_hold(dev, &p->dev_tracker, GFP_KERNEL);
436 p->dev = dev;
437 p->path_cost = port_cost(dev);
438 p->priority = 0x8000 >> BR_PORT_BITS;
439 p->port_no = index;
440 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
441 br_init_port(p);
442 br_set_state(p, BR_STATE_DISABLED);
443 br_stp_port_timer_init(p);
444 err = br_multicast_add_port(p);
445 if (err) {
446 netdev_put(dev, &p->dev_tracker);
447 kfree(p);
448 p = ERR_PTR(err);
449 }
450
451 return p;
452}
453
454int br_add_bridge(struct net *net, const char *name)
455{
456 struct net_device *dev;
457 int res;
458
459 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
460 br_dev_setup);
461
462 if (!dev)
463 return -ENOMEM;
464
465 dev_net_set(dev, net);
466 dev->rtnl_link_ops = &br_link_ops;
467
468 res = register_netdevice(dev);
469 if (res)
470 free_netdev(dev);
471 return res;
472}
473
474int br_del_bridge(struct net *net, const char *name)
475{
476 struct net_device *dev;
477 int ret = 0;
478
479 dev = __dev_get_by_name(net, name);
480 if (dev == NULL)
481 ret = -ENXIO; /* Could not find device */
482
483 else if (!netif_is_bridge_master(dev)) {
484 /* Attempt to delete non bridge device! */
485 ret = -EPERM;
486 }
487
488 else if (dev->flags & IFF_UP) {
489 /* Not shutdown yet. */
490 ret = -EBUSY;
491 }
492
493 else
494 br_dev_delete(dev, NULL);
495
496 return ret;
497}
498
499/* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
500static int br_mtu_min(const struct net_bridge *br)
501{
502 const struct net_bridge_port *p;
503 int ret_mtu = 0;
504
505 list_for_each_entry(p, &br->port_list, list)
506 if (!ret_mtu || ret_mtu > p->dev->mtu)
507 ret_mtu = p->dev->mtu;
508
509 return ret_mtu ? ret_mtu : ETH_DATA_LEN;
510}
511
512void br_mtu_auto_adjust(struct net_bridge *br)
513{
514 ASSERT_RTNL();
515
516 /* if the bridge MTU was manually configured don't mess with it */
517 if (br_opt_get(br, BROPT_MTU_SET_BY_USER))
518 return;
519
520 /* change to the minimum MTU and clear the flag which was set by
521 * the bridge ndo_change_mtu callback
522 */
523 dev_set_mtu(br->dev, br_mtu_min(br));
524 br_opt_toggle(br, BROPT_MTU_SET_BY_USER, false);
525}
526
527static void br_set_gso_limits(struct net_bridge *br)
528{
529 unsigned int tso_max_size = TSO_MAX_SIZE;
530 const struct net_bridge_port *p;
531 u16 tso_max_segs = TSO_MAX_SEGS;
532
533 list_for_each_entry(p, &br->port_list, list) {
534 tso_max_size = min(tso_max_size, p->dev->tso_max_size);
535 tso_max_segs = min(tso_max_segs, p->dev->tso_max_segs);
536 }
537 netif_set_tso_max_size(br->dev, tso_max_size);
538 netif_set_tso_max_segs(br->dev, tso_max_segs);
539}
540
541/*
542 * Recomputes features using slave's features
543 */
544netdev_features_t br_features_recompute(struct net_bridge *br,
545 netdev_features_t features)
546{
547 struct net_bridge_port *p;
548 netdev_features_t mask;
549
550 if (list_empty(&br->port_list))
551 return features;
552
553 mask = features;
554 features &= ~NETIF_F_ONE_FOR_ALL;
555
556 list_for_each_entry(p, &br->port_list, list) {
557 features = netdev_increment_features(features,
558 p->dev->features, mask);
559 }
560 features = netdev_add_tso_features(features, mask);
561
562 return features;
563}
564
565/* called with RTNL */
566int br_add_if(struct net_bridge *br, struct net_device *dev,
567 struct netlink_ext_ack *extack)
568{
569 struct net_bridge_port *p;
570 int err = 0;
571 unsigned br_hr, dev_hr;
572 bool changed_addr, fdb_synced = false;
573
574 /* Don't allow bridging non-ethernet like devices. */
575 if ((dev->flags & IFF_LOOPBACK) ||
576 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
577 !is_valid_ether_addr(dev->dev_addr))
578 return -EINVAL;
579
580 /* No bridging of bridges */
581 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) {
582 NL_SET_ERR_MSG(extack,
583 "Can not enslave a bridge to a bridge");
584 return -ELOOP;
585 }
586
587 /* Device has master upper dev */
588 if (netdev_master_upper_dev_get(dev))
589 return -EBUSY;
590
591 /* No bridging devices that dislike that (e.g. wireless) */
592 if (dev->priv_flags & IFF_DONT_BRIDGE) {
593 NL_SET_ERR_MSG(extack,
594 "Device does not allow enslaving to a bridge");
595 return -EOPNOTSUPP;
596 }
597
598 p = new_nbp(br, dev);
599 if (IS_ERR(p))
600 return PTR_ERR(p);
601
602 call_netdevice_notifiers(NETDEV_JOIN, dev);
603
604 err = dev_set_allmulti(dev, 1);
605 if (err) {
606 br_multicast_del_port(p);
607 netdev_put(dev, &p->dev_tracker);
608 kfree(p); /* kobject not yet init'd, manually free */
609 goto err1;
610 }
611
612 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
613 SYSFS_BRIDGE_PORT_ATTR);
614 if (err)
615 goto err2;
616
617 err = br_sysfs_addif(p);
618 if (err)
619 goto err2;
620
621 err = br_netpoll_enable(p);
622 if (err)
623 goto err3;
624
625 err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p);
626 if (err)
627 goto err4;
628
629 dev->priv_flags |= IFF_BRIDGE_PORT;
630
631 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
632 if (err)
633 goto err5;
634
635 dev_disable_lro(dev);
636
637 list_add_rcu(&p->list, &br->port_list);
638
639 nbp_update_port_count(br);
640 if (!br_promisc_port(p) && (p->dev->priv_flags & IFF_UNICAST_FLT)) {
641 /* When updating the port count we also update all ports'
642 * promiscuous mode.
643 * A port leaving promiscuous mode normally gets the bridge's
644 * fdb synced to the unicast filter (if supported), however,
645 * `br_port_clear_promisc` does not distinguish between
646 * non-promiscuous ports and *new* ports, so we need to
647 * sync explicitly here.
648 */
649 fdb_synced = br_fdb_sync_static(br, p) == 0;
650 if (!fdb_synced)
651 netdev_err(dev, "failed to sync bridge static fdb addresses to this port\n");
652 }
653
654 netdev_update_features(br->dev);
655
656 br_hr = br->dev->needed_headroom;
657 dev_hr = netdev_get_fwd_headroom(dev);
658 if (br_hr < dev_hr)
659 update_headroom(br, dev_hr);
660 else
661 netdev_set_rx_headroom(dev, br_hr);
662
663 if (br_fdb_add_local(br, p, dev->dev_addr, 0))
664 netdev_err(dev, "failed insert local address bridge forwarding table\n");
665
666 if (br->dev->addr_assign_type != NET_ADDR_SET) {
667 /* Ask for permission to use this MAC address now, even if we
668 * don't end up choosing it below.
669 */
670 err = dev_pre_changeaddr_notify(br->dev, dev->dev_addr, extack);
671 if (err)
672 goto err6;
673 }
674
675 err = nbp_vlan_init(p, extack);
676 if (err) {
677 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
678 goto err6;
679 }
680
681 spin_lock_bh(&br->lock);
682 changed_addr = br_stp_recalculate_bridge_id(br);
683
684 if (netif_running(dev) && netif_oper_up(dev) &&
685 (br->dev->flags & IFF_UP))
686 br_stp_enable_port(p);
687 spin_unlock_bh(&br->lock);
688
689 br_ifinfo_notify(RTM_NEWLINK, NULL, p);
690
691 if (changed_addr)
692 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
693
694 br_mtu_auto_adjust(br);
695 br_set_gso_limits(br);
696
697 kobject_uevent(&p->kobj, KOBJ_ADD);
698
699 return 0;
700
701err6:
702 if (fdb_synced)
703 br_fdb_unsync_static(br, p);
704 list_del_rcu(&p->list);
705 br_fdb_delete_by_port(br, p, 0, 1);
706 nbp_update_port_count(br);
707 netdev_upper_dev_unlink(dev, br->dev);
708err5:
709 dev->priv_flags &= ~IFF_BRIDGE_PORT;
710 netdev_rx_handler_unregister(dev);
711err4:
712 br_netpoll_disable(p);
713err3:
714 sysfs_remove_link(br->ifobj, p->dev->name);
715err2:
716 br_multicast_del_port(p);
717 netdev_put(dev, &p->dev_tracker);
718 kobject_put(&p->kobj);
719 dev_set_allmulti(dev, -1);
720err1:
721 return err;
722}
723
724/* called with RTNL */
725int br_del_if(struct net_bridge *br, struct net_device *dev)
726{
727 struct net_bridge_port *p;
728 bool changed_addr;
729
730 p = br_port_get_rtnl(dev);
731 if (!p || p->br != br)
732 return -EINVAL;
733
734 /* Since more than one interface can be attached to a bridge,
735 * there still maybe an alternate path for netconsole to use;
736 * therefore there is no reason for a NETDEV_RELEASE event.
737 */
738 del_nbp(p);
739
740 br_mtu_auto_adjust(br);
741 br_set_gso_limits(br);
742
743 spin_lock_bh(&br->lock);
744 changed_addr = br_stp_recalculate_bridge_id(br);
745 spin_unlock_bh(&br->lock);
746
747 if (changed_addr)
748 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
749
750 netdev_update_features(br->dev);
751
752 return 0;
753}
754
755void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
756{
757 struct net_bridge *br = p->br;
758
759 if (mask & BR_AUTO_MASK)
760 nbp_update_port_count(br);
761
762 if (mask & BR_NEIGH_SUPPRESS)
763 br_recalculate_neigh_suppress_enabled(br);
764}
765
766bool br_port_flag_is_set(const struct net_device *dev, unsigned long flag)
767{
768 struct net_bridge_port *p;
769
770 p = br_port_get_rtnl_rcu(dev);
771 if (!p)
772 return false;
773
774 return p->flags & flag;
775}
776EXPORT_SYMBOL_GPL(br_port_flag_is_set);
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_mrp_port_del(br, p);
337 br_cfm_port_del(br, p);
338
339 br_ifinfo_notify(RTM_DELLINK, NULL, p);
340
341 list_del_rcu(&p->list);
342 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
343 update_headroom(br, get_max_headroom(br));
344 netdev_reset_rx_headroom(dev);
345
346 nbp_vlan_flush(p);
347 br_fdb_delete_by_port(br, p, 0, 1);
348 switchdev_deferred_process();
349 nbp_backup_clear(p);
350
351 nbp_update_port_count(br);
352
353 netdev_upper_dev_unlink(dev, br->dev);
354
355 dev->priv_flags &= ~IFF_BRIDGE_PORT;
356
357 netdev_rx_handler_unregister(dev);
358
359 br_multicast_del_port(p);
360
361 kobject_uevent(&p->kobj, KOBJ_REMOVE);
362 kobject_del(&p->kobj);
363
364 br_netpoll_disable(p);
365
366 call_rcu(&p->rcu, destroy_nbp_rcu);
367}
368
369/* Delete bridge device */
370void br_dev_delete(struct net_device *dev, struct list_head *head)
371{
372 struct net_bridge *br = netdev_priv(dev);
373 struct net_bridge_port *p, *n;
374
375 list_for_each_entry_safe(p, n, &br->port_list, list) {
376 del_nbp(p);
377 }
378
379 br_recalculate_neigh_suppress_enabled(br);
380
381 br_fdb_delete_by_port(br, NULL, 0, 1);
382
383 cancel_delayed_work_sync(&br->gc_work);
384
385 br_sysfs_delbr(br->dev);
386 unregister_netdevice_queue(br->dev, head);
387}
388
389/* find an available port number */
390static int find_portno(struct net_bridge *br)
391{
392 int index;
393 struct net_bridge_port *p;
394 unsigned long *inuse;
395
396 inuse = bitmap_zalloc(BR_MAX_PORTS, GFP_KERNEL);
397 if (!inuse)
398 return -ENOMEM;
399
400 set_bit(0, inuse); /* zero is reserved */
401 list_for_each_entry(p, &br->port_list, list) {
402 set_bit(p->port_no, inuse);
403 }
404 index = find_first_zero_bit(inuse, BR_MAX_PORTS);
405 bitmap_free(inuse);
406
407 return (index >= BR_MAX_PORTS) ? -EXFULL : index;
408}
409
410/* called with RTNL but without bridge lock */
411static struct net_bridge_port *new_nbp(struct net_bridge *br,
412 struct net_device *dev)
413{
414 struct net_bridge_port *p;
415 int index, err;
416
417 index = find_portno(br);
418 if (index < 0)
419 return ERR_PTR(index);
420
421 p = kzalloc(sizeof(*p), GFP_KERNEL);
422 if (p == NULL)
423 return ERR_PTR(-ENOMEM);
424
425 p->br = br;
426 dev_hold(dev);
427 p->dev = dev;
428 p->path_cost = port_cost(dev);
429 p->priority = 0x8000 >> BR_PORT_BITS;
430 p->port_no = index;
431 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
432 br_init_port(p);
433 br_set_state(p, BR_STATE_DISABLED);
434 br_stp_port_timer_init(p);
435 err = br_multicast_add_port(p);
436 if (err) {
437 dev_put(dev);
438 kfree(p);
439 p = ERR_PTR(err);
440 }
441
442 return p;
443}
444
445int br_add_bridge(struct net *net, const char *name)
446{
447 struct net_device *dev;
448 int res;
449
450 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
451 br_dev_setup);
452
453 if (!dev)
454 return -ENOMEM;
455
456 dev_net_set(dev, net);
457 dev->rtnl_link_ops = &br_link_ops;
458
459 res = register_netdev(dev);
460 if (res)
461 free_netdev(dev);
462 return res;
463}
464
465int br_del_bridge(struct net *net, const char *name)
466{
467 struct net_device *dev;
468 int ret = 0;
469
470 rtnl_lock();
471 dev = __dev_get_by_name(net, name);
472 if (dev == NULL)
473 ret = -ENXIO; /* Could not find device */
474
475 else if (!(dev->priv_flags & IFF_EBRIDGE)) {
476 /* Attempt to delete non bridge device! */
477 ret = -EPERM;
478 }
479
480 else if (dev->flags & IFF_UP) {
481 /* Not shutdown yet. */
482 ret = -EBUSY;
483 }
484
485 else
486 br_dev_delete(dev, NULL);
487
488 rtnl_unlock();
489 return ret;
490}
491
492/* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
493static int br_mtu_min(const struct net_bridge *br)
494{
495 const struct net_bridge_port *p;
496 int ret_mtu = 0;
497
498 list_for_each_entry(p, &br->port_list, list)
499 if (!ret_mtu || ret_mtu > p->dev->mtu)
500 ret_mtu = p->dev->mtu;
501
502 return ret_mtu ? ret_mtu : ETH_DATA_LEN;
503}
504
505void br_mtu_auto_adjust(struct net_bridge *br)
506{
507 ASSERT_RTNL();
508
509 /* if the bridge MTU was manually configured don't mess with it */
510 if (br_opt_get(br, BROPT_MTU_SET_BY_USER))
511 return;
512
513 /* change to the minimum MTU and clear the flag which was set by
514 * the bridge ndo_change_mtu callback
515 */
516 dev_set_mtu(br->dev, br_mtu_min(br));
517 br_opt_toggle(br, BROPT_MTU_SET_BY_USER, false);
518}
519
520static void br_set_gso_limits(struct net_bridge *br)
521{
522 unsigned int gso_max_size = GSO_MAX_SIZE;
523 u16 gso_max_segs = GSO_MAX_SEGS;
524 const struct net_bridge_port *p;
525
526 list_for_each_entry(p, &br->port_list, list) {
527 gso_max_size = min(gso_max_size, p->dev->gso_max_size);
528 gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
529 }
530 br->dev->gso_max_size = gso_max_size;
531 br->dev->gso_max_segs = gso_max_segs;
532}
533
534/*
535 * Recomputes features using slave's features
536 */
537netdev_features_t br_features_recompute(struct net_bridge *br,
538 netdev_features_t features)
539{
540 struct net_bridge_port *p;
541 netdev_features_t mask;
542
543 if (list_empty(&br->port_list))
544 return features;
545
546 mask = features;
547 features &= ~NETIF_F_ONE_FOR_ALL;
548
549 list_for_each_entry(p, &br->port_list, list) {
550 features = netdev_increment_features(features,
551 p->dev->features, mask);
552 }
553 features = netdev_add_tso_features(features, mask);
554
555 return features;
556}
557
558/* called with RTNL */
559int br_add_if(struct net_bridge *br, struct net_device *dev,
560 struct netlink_ext_ack *extack)
561{
562 struct net_bridge_port *p;
563 int err = 0;
564 unsigned br_hr, dev_hr;
565 bool changed_addr, fdb_synced = false;
566
567 /* Don't allow bridging non-ethernet like devices. */
568 if ((dev->flags & IFF_LOOPBACK) ||
569 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
570 !is_valid_ether_addr(dev->dev_addr))
571 return -EINVAL;
572
573 /* Also don't allow bridging of net devices that are DSA masters, since
574 * the bridge layer rx_handler prevents the DSA fake ethertype handler
575 * to be invoked, so we don't get the chance to strip off and parse the
576 * DSA switch tag protocol header (the bridge layer just returns
577 * RX_HANDLER_CONSUMED, stopping RX processing for these frames).
578 * The only case where that would not be an issue is when bridging can
579 * already be offloaded, such as when the DSA master is itself a DSA
580 * or plain switchdev port, and is bridged only with other ports from
581 * the same hardware device.
582 */
583 if (netdev_uses_dsa(dev)) {
584 list_for_each_entry(p, &br->port_list, list) {
585 if (!netdev_port_same_parent_id(dev, p->dev)) {
586 NL_SET_ERR_MSG(extack,
587 "Cannot do software bridging with a DSA master");
588 return -EINVAL;
589 }
590 }
591 }
592
593 /* No bridging of bridges */
594 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) {
595 NL_SET_ERR_MSG(extack,
596 "Can not enslave a bridge to a bridge");
597 return -ELOOP;
598 }
599
600 /* Device has master upper dev */
601 if (netdev_master_upper_dev_get(dev))
602 return -EBUSY;
603
604 /* No bridging devices that dislike that (e.g. wireless) */
605 if (dev->priv_flags & IFF_DONT_BRIDGE) {
606 NL_SET_ERR_MSG(extack,
607 "Device does not allow enslaving to a bridge");
608 return -EOPNOTSUPP;
609 }
610
611 p = new_nbp(br, dev);
612 if (IS_ERR(p))
613 return PTR_ERR(p);
614
615 call_netdevice_notifiers(NETDEV_JOIN, dev);
616
617 err = dev_set_allmulti(dev, 1);
618 if (err) {
619 br_multicast_del_port(p);
620 kfree(p); /* kobject not yet init'd, manually free */
621 goto err1;
622 }
623
624 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
625 SYSFS_BRIDGE_PORT_ATTR);
626 if (err)
627 goto err2;
628
629 err = br_sysfs_addif(p);
630 if (err)
631 goto err2;
632
633 err = br_netpoll_enable(p);
634 if (err)
635 goto err3;
636
637 err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p);
638 if (err)
639 goto err4;
640
641 dev->priv_flags |= IFF_BRIDGE_PORT;
642
643 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
644 if (err)
645 goto err5;
646
647 err = nbp_switchdev_mark_set(p);
648 if (err)
649 goto err6;
650
651 dev_disable_lro(dev);
652
653 list_add_rcu(&p->list, &br->port_list);
654
655 nbp_update_port_count(br);
656 if (!br_promisc_port(p) && (p->dev->priv_flags & IFF_UNICAST_FLT)) {
657 /* When updating the port count we also update all ports'
658 * promiscuous mode.
659 * A port leaving promiscuous mode normally gets the bridge's
660 * fdb synced to the unicast filter (if supported), however,
661 * `br_port_clear_promisc` does not distinguish between
662 * non-promiscuous ports and *new* ports, so we need to
663 * sync explicitly here.
664 */
665 fdb_synced = br_fdb_sync_static(br, p) == 0;
666 if (!fdb_synced)
667 netdev_err(dev, "failed to sync bridge static fdb addresses to this port\n");
668 }
669
670 netdev_update_features(br->dev);
671
672 br_hr = br->dev->needed_headroom;
673 dev_hr = netdev_get_fwd_headroom(dev);
674 if (br_hr < dev_hr)
675 update_headroom(br, dev_hr);
676 else
677 netdev_set_rx_headroom(dev, br_hr);
678
679 if (br_fdb_insert(br, p, dev->dev_addr, 0))
680 netdev_err(dev, "failed insert local address bridge forwarding table\n");
681
682 if (br->dev->addr_assign_type != NET_ADDR_SET) {
683 /* Ask for permission to use this MAC address now, even if we
684 * don't end up choosing it below.
685 */
686 err = dev_pre_changeaddr_notify(br->dev, dev->dev_addr, extack);
687 if (err)
688 goto err7;
689 }
690
691 err = nbp_vlan_init(p, extack);
692 if (err) {
693 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
694 goto err7;
695 }
696
697 spin_lock_bh(&br->lock);
698 changed_addr = br_stp_recalculate_bridge_id(br);
699
700 if (netif_running(dev) && netif_oper_up(dev) &&
701 (br->dev->flags & IFF_UP))
702 br_stp_enable_port(p);
703 spin_unlock_bh(&br->lock);
704
705 br_ifinfo_notify(RTM_NEWLINK, NULL, p);
706
707 if (changed_addr)
708 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
709
710 br_mtu_auto_adjust(br);
711 br_set_gso_limits(br);
712
713 kobject_uevent(&p->kobj, KOBJ_ADD);
714
715 return 0;
716
717err7:
718 if (fdb_synced)
719 br_fdb_unsync_static(br, p);
720 list_del_rcu(&p->list);
721 br_fdb_delete_by_port(br, p, 0, 1);
722 nbp_update_port_count(br);
723err6:
724 netdev_upper_dev_unlink(dev, br->dev);
725err5:
726 dev->priv_flags &= ~IFF_BRIDGE_PORT;
727 netdev_rx_handler_unregister(dev);
728err4:
729 br_netpoll_disable(p);
730err3:
731 sysfs_remove_link(br->ifobj, p->dev->name);
732err2:
733 br_multicast_del_port(p);
734 kobject_put(&p->kobj);
735 dev_set_allmulti(dev, -1);
736err1:
737 dev_put(dev);
738 return err;
739}
740
741/* called with RTNL */
742int br_del_if(struct net_bridge *br, struct net_device *dev)
743{
744 struct net_bridge_port *p;
745 bool changed_addr;
746
747 p = br_port_get_rtnl(dev);
748 if (!p || p->br != br)
749 return -EINVAL;
750
751 /* Since more than one interface can be attached to a bridge,
752 * there still maybe an alternate path for netconsole to use;
753 * therefore there is no reason for a NETDEV_RELEASE event.
754 */
755 del_nbp(p);
756
757 br_mtu_auto_adjust(br);
758 br_set_gso_limits(br);
759
760 spin_lock_bh(&br->lock);
761 changed_addr = br_stp_recalculate_bridge_id(br);
762 spin_unlock_bh(&br->lock);
763
764 if (changed_addr)
765 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
766
767 netdev_update_features(br->dev);
768
769 return 0;
770}
771
772void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
773{
774 struct net_bridge *br = p->br;
775
776 if (mask & BR_AUTO_MASK)
777 nbp_update_port_count(br);
778
779 if (mask & BR_NEIGH_SUPPRESS)
780 br_recalculate_neigh_suppress_enabled(br);
781}
782
783bool br_port_flag_is_set(const struct net_device *dev, unsigned long flag)
784{
785 struct net_bridge_port *p;
786
787 p = br_port_get_rtnl_rcu(dev);
788 if (!p)
789 return false;
790
791 return p->flags & flag;
792}
793EXPORT_SYMBOL_GPL(br_port_flag_is_set);