1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Linux ethernet bridge
7 * Lennert Buytenhek <buytenh@gnu.org>
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>
23 #include <linux/if_vlan.h>
24 #include <net/switchdev.h>
25 #include <net/net_namespace.h>
27 #include "br_private.h"
30 * Determine initial path cost based on speed.
31 * using recommendations from 802.1d standard
33 * Since driver might sleep need to not be holding any locks.
35 static int port_cost(struct net_device *dev)
37 struct ethtool_link_ksettings ecmd;
39 if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
40 switch (ecmd.base.speed) {
56 if (ecmd.base.speed > SPEED_10000)
61 /* Old silly heuristics based on name */
62 if (!strncmp(dev->name, "lec", 3))
65 if (!strncmp(dev->name, "plip", 4))
68 return 100; /* assume old 10Mbps */
72 /* Check for port carrier transitions. */
73 void br_port_carrier_check(struct net_bridge_port *p, bool *notified)
75 struct net_device *dev = p->dev;
76 struct net_bridge *br = p->br;
78 if (!(p->flags & BR_ADMIN_COST) &&
79 netif_running(dev) && netif_oper_up(dev))
80 p->path_cost = port_cost(dev);
83 if (!netif_running(br->dev))
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);
93 if (p->state != BR_STATE_DISABLED) {
94 br_stp_disable_port(p);
98 spin_unlock_bh(&br->lock);
101 static void br_port_set_promisc(struct net_bridge_port *p)
105 if (br_promisc_port(p))
108 err = dev_set_promiscuity(p->dev, 1);
112 br_fdb_unsync_static(p->br, p);
113 p->flags |= BR_PROMISC;
116 static void br_port_clear_promisc(struct net_bridge_port *p)
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
125 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
128 /* Since we'll be clearing the promisc mode, program the port
129 * first so that we don't have interruption in traffic.
131 err = br_fdb_sync_static(p->br, p);
135 dev_set_promiscuity(p->dev, -1);
136 p->flags &= ~BR_PROMISC;
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.
144 void br_manage_promisc(struct net_bridge *br)
146 struct net_bridge_port *p;
147 bool set_all = false;
149 /* If vlan filtering is disabled or bridge interface is placed
150 * into promiscuous mode, place all ports in promiscuous mode.
152 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev))
155 list_for_each_entry(p, &br->port_list, list) {
157 br_port_set_promisc(p);
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
169 if ((p->dev->priv_flags & IFF_UNICAST_FLT) &&
170 (br->auto_cnt == 0 ||
171 (br->auto_cnt == 1 && br_auto_port(p))))
172 br_port_clear_promisc(p);
174 br_port_set_promisc(p);
179 int nbp_backup_change(struct net_bridge_port *p,
180 struct net_device *backup_dev)
182 struct net_bridge_port *old_backup = rtnl_dereference(p->backup_port);
183 struct net_bridge_port *backup_p = NULL;
188 if (!netif_is_bridge_port(backup_dev))
191 backup_p = br_port_get_rtnl(backup_dev);
192 if (backup_p->br != p->br)
199 if (old_backup == backup_p)
202 /* if the backup link is already set, clear it */
204 old_backup->backup_redirected_cnt--;
207 backup_p->backup_redirected_cnt++;
208 rcu_assign_pointer(p->backup_port, backup_p);
213 static void nbp_backup_clear(struct net_bridge_port *p)
215 nbp_backup_change(p, NULL);
216 if (p->backup_redirected_cnt) {
217 struct net_bridge_port *cur_p;
219 list_for_each_entry(cur_p, &p->br->port_list, list) {
220 struct net_bridge_port *backup_p;
222 backup_p = rtnl_dereference(cur_p->backup_port);
224 nbp_backup_change(cur_p, NULL);
228 WARN_ON(rcu_access_pointer(p->backup_port) || p->backup_redirected_cnt);
231 static void nbp_update_port_count(struct net_bridge *br)
233 struct net_bridge_port *p;
236 list_for_each_entry(p, &br->port_list, list) {
240 if (br->auto_cnt != cnt) {
242 br_manage_promisc(br);
246 static void nbp_delete_promisc(struct net_bridge_port *p)
248 /* If port is currently promiscuous, unset promiscuity.
249 * Otherwise, it is a static port so remove all addresses
252 dev_set_allmulti(p->dev, -1);
253 if (br_promisc_port(p))
254 dev_set_promiscuity(p->dev, -1);
256 br_fdb_unsync_static(p->br, p);
259 static void release_nbp(struct kobject *kobj)
261 struct net_bridge_port *p
262 = container_of(kobj, struct net_bridge_port, kobj);
266 static void brport_get_ownership(const struct kobject *kobj, kuid_t *uid, kgid_t *gid)
268 struct net_bridge_port *p = kobj_to_brport(kobj);
270 net_ns_get_ownership(dev_net(p->dev), uid, gid);
273 static const struct kobj_type brport_ktype = {
275 .sysfs_ops = &brport_sysfs_ops,
277 .release = release_nbp,
278 .get_ownership = brport_get_ownership,
281 static void destroy_nbp(struct net_bridge_port *p)
283 struct net_device *dev = p->dev;
287 netdev_put(dev, &p->dev_tracker);
289 kobject_put(&p->kobj);
292 static void destroy_nbp_rcu(struct rcu_head *head)
294 struct net_bridge_port *p =
295 container_of(head, struct net_bridge_port, rcu);
299 static unsigned get_max_headroom(struct net_bridge *br)
301 unsigned max_headroom = 0;
302 struct net_bridge_port *p;
304 list_for_each_entry(p, &br->port_list, list) {
305 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
307 if (dev_headroom > max_headroom)
308 max_headroom = dev_headroom;
314 static void update_headroom(struct net_bridge *br, int new_hr)
316 struct net_bridge_port *p;
318 list_for_each_entry(p, &br->port_list, list)
319 netdev_set_rx_headroom(p->dev, new_hr);
321 br->dev->needed_headroom = new_hr;
324 /* Delete port(interface) from bridge is done in two steps.
325 * via RCU. First step, marks device as down. That deletes
326 * all the timers and stops new packets from flowing through.
328 * Final cleanup doesn't occur until after all CPU's finished
329 * processing packets.
331 * Protected from multiple admin operations by RTNL mutex
333 static void del_nbp(struct net_bridge_port *p)
335 struct net_bridge *br = p->br;
336 struct net_device *dev = p->dev;
338 sysfs_remove_link(br->ifobj, p->dev->name);
340 nbp_delete_promisc(p);
342 spin_lock_bh(&br->lock);
343 br_stp_disable_port(p);
344 spin_unlock_bh(&br->lock);
346 br_mrp_port_del(br, p);
347 br_cfm_port_del(br, p);
349 br_ifinfo_notify(RTM_DELLINK, NULL, p);
351 list_del_rcu(&p->list);
352 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
353 update_headroom(br, get_max_headroom(br));
354 netdev_reset_rx_headroom(dev);
357 br_fdb_delete_by_port(br, p, 0, 1);
358 switchdev_deferred_process();
361 nbp_update_port_count(br);
363 netdev_upper_dev_unlink(dev, br->dev);
365 dev->priv_flags &= ~IFF_BRIDGE_PORT;
367 netdev_rx_handler_unregister(dev);
369 br_multicast_del_port(p);
371 kobject_uevent(&p->kobj, KOBJ_REMOVE);
372 kobject_del(&p->kobj);
374 br_netpoll_disable(p);
376 call_rcu(&p->rcu, destroy_nbp_rcu);
379 /* Delete bridge device */
380 void br_dev_delete(struct net_device *dev, struct list_head *head)
382 struct net_bridge *br = netdev_priv(dev);
383 struct net_bridge_port *p, *n;
385 list_for_each_entry_safe(p, n, &br->port_list, list) {
389 br_recalculate_neigh_suppress_enabled(br);
391 br_fdb_delete_by_port(br, NULL, 0, 1);
393 cancel_delayed_work_sync(&br->gc_work);
395 br_sysfs_delbr(br->dev);
396 unregister_netdevice_queue(br->dev, head);
399 /* find an available port number */
400 static int find_portno(struct net_bridge *br)
403 struct net_bridge_port *p;
404 unsigned long *inuse;
406 inuse = bitmap_zalloc(BR_MAX_PORTS, GFP_KERNEL);
410 __set_bit(0, inuse); /* zero is reserved */
411 list_for_each_entry(p, &br->port_list, list)
412 __set_bit(p->port_no, inuse);
414 index = find_first_zero_bit(inuse, BR_MAX_PORTS);
417 return (index >= BR_MAX_PORTS) ? -EXFULL : index;
420 /* called with RTNL but without bridge lock */
421 static struct net_bridge_port *new_nbp(struct net_bridge *br,
422 struct net_device *dev)
424 struct net_bridge_port *p;
427 index = find_portno(br);
429 return ERR_PTR(index);
431 p = kzalloc(sizeof(*p), GFP_KERNEL);
433 return ERR_PTR(-ENOMEM);
436 netdev_hold(dev, &p->dev_tracker, GFP_KERNEL);
438 p->path_cost = port_cost(dev);
439 p->priority = 0x8000 >> BR_PORT_BITS;
441 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
443 br_set_state(p, BR_STATE_DISABLED);
444 br_stp_port_timer_init(p);
445 err = br_multicast_add_port(p);
447 netdev_put(dev, &p->dev_tracker);
455 int br_add_bridge(struct net *net, const char *name)
457 struct net_device *dev;
460 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
466 dev_net_set(dev, net);
467 dev->rtnl_link_ops = &br_link_ops;
469 res = register_netdevice(dev);
475 int br_del_bridge(struct net *net, const char *name)
477 struct net_device *dev;
480 dev = __dev_get_by_name(net, name);
482 ret = -ENXIO; /* Could not find device */
484 else if (!netif_is_bridge_master(dev)) {
485 /* Attempt to delete non bridge device! */
489 else if (dev->flags & IFF_UP) {
490 /* Not shutdown yet. */
495 br_dev_delete(dev, NULL);
500 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
501 static int br_mtu_min(const struct net_bridge *br)
503 const struct net_bridge_port *p;
506 list_for_each_entry(p, &br->port_list, list)
507 if (!ret_mtu || ret_mtu > p->dev->mtu)
508 ret_mtu = p->dev->mtu;
510 return ret_mtu ? ret_mtu : ETH_DATA_LEN;
513 void br_mtu_auto_adjust(struct net_bridge *br)
517 /* if the bridge MTU was manually configured don't mess with it */
518 if (br_opt_get(br, BROPT_MTU_SET_BY_USER))
521 /* change to the minimum MTU and clear the flag which was set by
522 * the bridge ndo_change_mtu callback
524 dev_set_mtu(br->dev, br_mtu_min(br));
525 br_opt_toggle(br, BROPT_MTU_SET_BY_USER, false);
528 static void br_set_gso_limits(struct net_bridge *br)
530 unsigned int tso_max_size = TSO_MAX_SIZE;
531 const struct net_bridge_port *p;
532 u16 tso_max_segs = TSO_MAX_SEGS;
534 list_for_each_entry(p, &br->port_list, list) {
535 tso_max_size = min(tso_max_size, p->dev->tso_max_size);
536 tso_max_segs = min(tso_max_segs, p->dev->tso_max_segs);
538 netif_set_tso_max_size(br->dev, tso_max_size);
539 netif_set_tso_max_segs(br->dev, tso_max_segs);
543 * Recomputes features using slave's features
545 netdev_features_t br_features_recompute(struct net_bridge *br,
546 netdev_features_t features)
548 struct net_bridge_port *p;
549 netdev_features_t mask;
551 if (list_empty(&br->port_list))
555 features &= ~NETIF_F_ONE_FOR_ALL;
557 list_for_each_entry(p, &br->port_list, list) {
558 features = netdev_increment_features(features,
559 p->dev->features, mask);
561 features = netdev_add_tso_features(features, mask);
566 /* called with RTNL */
567 int br_add_if(struct net_bridge *br, struct net_device *dev,
568 struct netlink_ext_ack *extack)
570 struct net_bridge_port *p;
572 unsigned br_hr, dev_hr;
573 bool changed_addr, fdb_synced = false;
575 /* Don't allow bridging non-ethernet like devices. */
576 if ((dev->flags & IFF_LOOPBACK) ||
577 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
578 !is_valid_ether_addr(dev->dev_addr))
581 /* No bridging of bridges */
582 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) {
583 NL_SET_ERR_MSG(extack,
584 "Can not enslave a bridge to a bridge");
588 /* Device has master upper dev */
589 if (netdev_master_upper_dev_get(dev))
592 /* No bridging devices that dislike that (e.g. wireless) */
593 if (dev->priv_flags & IFF_DONT_BRIDGE) {
594 NL_SET_ERR_MSG(extack,
595 "Device does not allow enslaving to a bridge");
599 p = new_nbp(br, dev);
603 call_netdevice_notifiers(NETDEV_JOIN, dev);
605 err = dev_set_allmulti(dev, 1);
607 br_multicast_del_port(p);
608 netdev_put(dev, &p->dev_tracker);
609 kfree(p); /* kobject not yet init'd, manually free */
613 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
614 SYSFS_BRIDGE_PORT_ATTR);
618 err = br_sysfs_addif(p);
622 err = br_netpoll_enable(p);
626 err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p);
630 dev->priv_flags |= IFF_BRIDGE_PORT;
632 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
636 dev_disable_lro(dev);
638 list_add_rcu(&p->list, &br->port_list);
640 nbp_update_port_count(br);
641 if (!br_promisc_port(p) && (p->dev->priv_flags & IFF_UNICAST_FLT)) {
642 /* When updating the port count we also update all ports'
644 * A port leaving promiscuous mode normally gets the bridge's
645 * fdb synced to the unicast filter (if supported), however,
646 * `br_port_clear_promisc` does not distinguish between
647 * non-promiscuous ports and *new* ports, so we need to
648 * sync explicitly here.
650 fdb_synced = br_fdb_sync_static(br, p) == 0;
652 netdev_err(dev, "failed to sync bridge static fdb addresses to this port\n");
655 netdev_update_features(br->dev);
657 br_hr = br->dev->needed_headroom;
658 dev_hr = netdev_get_fwd_headroom(dev);
660 update_headroom(br, dev_hr);
662 netdev_set_rx_headroom(dev, br_hr);
664 if (br_fdb_add_local(br, p, dev->dev_addr, 0))
665 netdev_err(dev, "failed insert local address bridge forwarding table\n");
667 if (br->dev->addr_assign_type != NET_ADDR_SET) {
668 /* Ask for permission to use this MAC address now, even if we
669 * don't end up choosing it below.
671 err = dev_pre_changeaddr_notify(br->dev, dev->dev_addr, extack);
676 err = nbp_vlan_init(p, extack);
678 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
682 spin_lock_bh(&br->lock);
683 changed_addr = br_stp_recalculate_bridge_id(br);
685 if (netif_running(dev) && netif_oper_up(dev) &&
686 (br->dev->flags & IFF_UP))
687 br_stp_enable_port(p);
688 spin_unlock_bh(&br->lock);
690 br_ifinfo_notify(RTM_NEWLINK, NULL, p);
693 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
695 br_mtu_auto_adjust(br);
696 br_set_gso_limits(br);
698 kobject_uevent(&p->kobj, KOBJ_ADD);
704 br_fdb_unsync_static(br, p);
705 list_del_rcu(&p->list);
706 br_fdb_delete_by_port(br, p, 0, 1);
707 nbp_update_port_count(br);
708 netdev_upper_dev_unlink(dev, br->dev);
710 dev->priv_flags &= ~IFF_BRIDGE_PORT;
711 netdev_rx_handler_unregister(dev);
713 br_netpoll_disable(p);
715 sysfs_remove_link(br->ifobj, p->dev->name);
717 br_multicast_del_port(p);
718 netdev_put(dev, &p->dev_tracker);
719 kobject_put(&p->kobj);
720 dev_set_allmulti(dev, -1);
725 /* called with RTNL */
726 int br_del_if(struct net_bridge *br, struct net_device *dev)
728 struct net_bridge_port *p;
731 p = br_port_get_rtnl(dev);
732 if (!p || p->br != br)
735 /* Since more than one interface can be attached to a bridge,
736 * there still maybe an alternate path for netconsole to use;
737 * therefore there is no reason for a NETDEV_RELEASE event.
741 br_mtu_auto_adjust(br);
742 br_set_gso_limits(br);
744 spin_lock_bh(&br->lock);
745 changed_addr = br_stp_recalculate_bridge_id(br);
746 spin_unlock_bh(&br->lock);
749 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
751 netdev_update_features(br->dev);
756 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
758 struct net_bridge *br = p->br;
760 if (mask & BR_AUTO_MASK)
761 nbp_update_port_count(br);
763 if (mask & (BR_NEIGH_SUPPRESS | BR_NEIGH_VLAN_SUPPRESS))
764 br_recalculate_neigh_suppress_enabled(br);
767 bool br_port_flag_is_set(const struct net_device *dev, unsigned long flag)
769 struct net_bridge_port *p;
771 p = br_port_get_rtnl_rcu(dev);
775 return p->flags & flag;
777 EXPORT_SYMBOL_GPL(br_port_flag_is_set);