3 * Linux ethernet bridge
6 * Lennert Buytenhek <buytenh@gnu.org>
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.
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>
27 #include <linux/if_vlan.h>
28 #include <net/switchdev.h>
29 #include <net/net_namespace.h>
31 #include "br_private.h"
34 * Determine initial path cost based on speed.
35 * using recommendations from 802.1d standard
37 * Since driver might sleep need to not be holding any locks.
39 static int port_cost(struct net_device *dev)
41 struct ethtool_link_ksettings ecmd;
43 if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
44 switch (ecmd.base.speed) {
56 /* Old silly heuristics based on name */
57 if (!strncmp(dev->name, "lec", 3))
60 if (!strncmp(dev->name, "plip", 4))
63 return 100; /* assume old 10Mbps */
67 /* Check for port carrier transitions. */
68 void br_port_carrier_check(struct net_bridge_port *p, bool *notified)
70 struct net_device *dev = p->dev;
71 struct net_bridge *br = p->br;
73 if (!(p->flags & BR_ADMIN_COST) &&
74 netif_running(dev) && netif_oper_up(dev))
75 p->path_cost = port_cost(dev);
78 if (!netif_running(br->dev))
81 spin_lock_bh(&br->lock);
82 if (netif_running(dev) && netif_oper_up(dev)) {
83 if (p->state == BR_STATE_DISABLED) {
84 br_stp_enable_port(p);
88 if (p->state != BR_STATE_DISABLED) {
89 br_stp_disable_port(p);
93 spin_unlock_bh(&br->lock);
96 static void br_port_set_promisc(struct net_bridge_port *p)
100 if (br_promisc_port(p))
103 err = dev_set_promiscuity(p->dev, 1);
107 br_fdb_unsync_static(p->br, p);
108 p->flags |= BR_PROMISC;
111 static void br_port_clear_promisc(struct net_bridge_port *p)
115 /* Check if the port is already non-promisc or if it doesn't
116 * support UNICAST filtering. Without unicast filtering support
117 * we'll end up re-enabling promisc mode anyway, so just check for
120 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
123 /* Since we'll be clearing the promisc mode, program the port
124 * first so that we don't have interruption in traffic.
126 err = br_fdb_sync_static(p->br, p);
130 dev_set_promiscuity(p->dev, -1);
131 p->flags &= ~BR_PROMISC;
134 /* When a port is added or removed or when certain port flags
135 * change, this function is called to automatically manage
136 * promiscuity setting of all the bridge ports. We are always called
137 * under RTNL so can skip using rcu primitives.
139 void br_manage_promisc(struct net_bridge *br)
141 struct net_bridge_port *p;
142 bool set_all = false;
144 /* If vlan filtering is disabled or bridge interface is placed
145 * into promiscuous mode, place all ports in promiscuous mode.
147 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev))
150 list_for_each_entry(p, &br->port_list, list) {
152 br_port_set_promisc(p);
154 /* If the number of auto-ports is <= 1, then all other
155 * ports will have their output configuration
156 * statically specified through fdbs. Since ingress
157 * on the auto-port becomes forwarding/egress to other
158 * ports and egress configuration is statically known,
159 * we can say that ingress configuration of the
160 * auto-port is also statically known.
161 * This lets us disable promiscuous mode and write
164 if (br->auto_cnt == 0 ||
165 (br->auto_cnt == 1 && br_auto_port(p)))
166 br_port_clear_promisc(p);
168 br_port_set_promisc(p);
173 int nbp_backup_change(struct net_bridge_port *p,
174 struct net_device *backup_dev)
176 struct net_bridge_port *old_backup = rtnl_dereference(p->backup_port);
177 struct net_bridge_port *backup_p = NULL;
182 if (!br_port_exists(backup_dev))
185 backup_p = br_port_get_rtnl(backup_dev);
186 if (backup_p->br != p->br)
193 if (old_backup == backup_p)
196 /* if the backup link is already set, clear it */
198 old_backup->backup_redirected_cnt--;
201 backup_p->backup_redirected_cnt++;
202 rcu_assign_pointer(p->backup_port, backup_p);
207 static void nbp_backup_clear(struct net_bridge_port *p)
209 nbp_backup_change(p, NULL);
210 if (p->backup_redirected_cnt) {
211 struct net_bridge_port *cur_p;
213 list_for_each_entry(cur_p, &p->br->port_list, list) {
214 struct net_bridge_port *backup_p;
216 backup_p = rtnl_dereference(cur_p->backup_port);
218 nbp_backup_change(cur_p, NULL);
222 WARN_ON(rcu_access_pointer(p->backup_port) || p->backup_redirected_cnt);
225 static void nbp_update_port_count(struct net_bridge *br)
227 struct net_bridge_port *p;
230 list_for_each_entry(p, &br->port_list, list) {
234 if (br->auto_cnt != cnt) {
236 br_manage_promisc(br);
240 static void nbp_delete_promisc(struct net_bridge_port *p)
242 /* If port is currently promiscuous, unset promiscuity.
243 * Otherwise, it is a static port so remove all addresses
246 dev_set_allmulti(p->dev, -1);
247 if (br_promisc_port(p))
248 dev_set_promiscuity(p->dev, -1);
250 br_fdb_unsync_static(p->br, p);
253 static void release_nbp(struct kobject *kobj)
255 struct net_bridge_port *p
256 = container_of(kobj, struct net_bridge_port, kobj);
260 static void brport_get_ownership(struct kobject *kobj, kuid_t *uid, kgid_t *gid)
262 struct net_bridge_port *p = kobj_to_brport(kobj);
264 net_ns_get_ownership(dev_net(p->dev), uid, gid);
267 static struct kobj_type brport_ktype = {
269 .sysfs_ops = &brport_sysfs_ops,
271 .release = release_nbp,
272 .get_ownership = brport_get_ownership,
275 static void destroy_nbp(struct net_bridge_port *p)
277 struct net_device *dev = p->dev;
283 kobject_put(&p->kobj);
286 static void destroy_nbp_rcu(struct rcu_head *head)
288 struct net_bridge_port *p =
289 container_of(head, struct net_bridge_port, rcu);
293 static unsigned get_max_headroom(struct net_bridge *br)
295 unsigned max_headroom = 0;
296 struct net_bridge_port *p;
298 list_for_each_entry(p, &br->port_list, list) {
299 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
301 if (dev_headroom > max_headroom)
302 max_headroom = dev_headroom;
308 static void update_headroom(struct net_bridge *br, int new_hr)
310 struct net_bridge_port *p;
312 list_for_each_entry(p, &br->port_list, list)
313 netdev_set_rx_headroom(p->dev, new_hr);
315 br->dev->needed_headroom = new_hr;
318 /* Delete port(interface) from bridge is done in two steps.
319 * via RCU. First step, marks device as down. That deletes
320 * all the timers and stops new packets from flowing through.
322 * Final cleanup doesn't occur until after all CPU's finished
323 * processing packets.
325 * Protected from multiple admin operations by RTNL mutex
327 static void del_nbp(struct net_bridge_port *p)
329 struct net_bridge *br = p->br;
330 struct net_device *dev = p->dev;
332 sysfs_remove_link(br->ifobj, p->dev->name);
334 nbp_delete_promisc(p);
336 spin_lock_bh(&br->lock);
337 br_stp_disable_port(p);
338 spin_unlock_bh(&br->lock);
340 br_ifinfo_notify(RTM_DELLINK, NULL, p);
342 list_del_rcu(&p->list);
343 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
344 update_headroom(br, get_max_headroom(br));
345 netdev_reset_rx_headroom(dev);
348 br_fdb_delete_by_port(br, p, 0, 1);
349 switchdev_deferred_process();
352 nbp_update_port_count(br);
354 netdev_upper_dev_unlink(dev, br->dev);
356 dev->priv_flags &= ~IFF_BRIDGE_PORT;
358 netdev_rx_handler_unregister(dev);
360 br_multicast_del_port(p);
362 kobject_uevent(&p->kobj, KOBJ_REMOVE);
363 kobject_del(&p->kobj);
365 br_netpoll_disable(p);
367 call_rcu(&p->rcu, destroy_nbp_rcu);
370 /* Delete bridge device */
371 void br_dev_delete(struct net_device *dev, struct list_head *head)
373 struct net_bridge *br = netdev_priv(dev);
374 struct net_bridge_port *p, *n;
376 list_for_each_entry_safe(p, n, &br->port_list, list) {
380 br_recalculate_neigh_suppress_enabled(br);
382 br_fdb_delete_by_port(br, NULL, 0, 1);
384 cancel_delayed_work_sync(&br->gc_work);
386 br_sysfs_delbr(br->dev);
387 unregister_netdevice_queue(br->dev, head);
390 /* find an available port number */
391 static int find_portno(struct net_bridge *br)
394 struct net_bridge_port *p;
395 unsigned long *inuse;
397 inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
402 set_bit(0, inuse); /* zero is reserved */
403 list_for_each_entry(p, &br->port_list, list) {
404 set_bit(p->port_no, inuse);
406 index = find_first_zero_bit(inuse, BR_MAX_PORTS);
409 return (index >= BR_MAX_PORTS) ? -EXFULL : index;
412 /* called with RTNL but without bridge lock */
413 static struct net_bridge_port *new_nbp(struct net_bridge *br,
414 struct net_device *dev)
416 struct net_bridge_port *p;
419 index = find_portno(br);
421 return ERR_PTR(index);
423 p = kzalloc(sizeof(*p), GFP_KERNEL);
425 return ERR_PTR(-ENOMEM);
430 p->path_cost = port_cost(dev);
431 p->priority = 0x8000 >> BR_PORT_BITS;
433 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD;
435 br_set_state(p, BR_STATE_DISABLED);
436 br_stp_port_timer_init(p);
437 err = br_multicast_add_port(p);
447 int br_add_bridge(struct net *net, const char *name)
449 struct net_device *dev;
452 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
458 dev_net_set(dev, net);
459 dev->rtnl_link_ops = &br_link_ops;
461 res = register_netdev(dev);
467 int br_del_bridge(struct net *net, const char *name)
469 struct net_device *dev;
473 dev = __dev_get_by_name(net, name);
475 ret = -ENXIO; /* Could not find device */
477 else if (!(dev->priv_flags & IFF_EBRIDGE)) {
478 /* Attempt to delete non bridge device! */
482 else if (dev->flags & IFF_UP) {
483 /* Not shutdown yet. */
488 br_dev_delete(dev, NULL);
494 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
495 static int br_mtu_min(const struct net_bridge *br)
497 const struct net_bridge_port *p;
500 list_for_each_entry(p, &br->port_list, list)
501 if (!ret_mtu || ret_mtu > p->dev->mtu)
502 ret_mtu = p->dev->mtu;
504 return ret_mtu ? ret_mtu : ETH_DATA_LEN;
507 void br_mtu_auto_adjust(struct net_bridge *br)
511 /* if the bridge MTU was manually configured don't mess with it */
512 if (br->mtu_set_by_user)
515 /* change to the minimum MTU and clear the flag which was set by
516 * the bridge ndo_change_mtu callback
518 dev_set_mtu(br->dev, br_mtu_min(br));
519 br->mtu_set_by_user = false;
522 static void br_set_gso_limits(struct net_bridge *br)
524 unsigned int gso_max_size = GSO_MAX_SIZE;
525 u16 gso_max_segs = GSO_MAX_SEGS;
526 const struct net_bridge_port *p;
528 list_for_each_entry(p, &br->port_list, list) {
529 gso_max_size = min(gso_max_size, p->dev->gso_max_size);
530 gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
532 br->dev->gso_max_size = gso_max_size;
533 br->dev->gso_max_segs = gso_max_segs;
537 * Recomputes features using slave's features
539 netdev_features_t br_features_recompute(struct net_bridge *br,
540 netdev_features_t features)
542 struct net_bridge_port *p;
543 netdev_features_t mask;
545 if (list_empty(&br->port_list))
549 features &= ~NETIF_F_ONE_FOR_ALL;
551 list_for_each_entry(p, &br->port_list, list) {
552 features = netdev_increment_features(features,
553 p->dev->features, mask);
555 features = netdev_add_tso_features(features, mask);
560 /* called with RTNL */
561 int br_add_if(struct net_bridge *br, struct net_device *dev,
562 struct netlink_ext_ack *extack)
564 struct net_bridge_port *p;
566 unsigned br_hr, dev_hr;
567 bool changed_addr, fdb_synced = false;
569 /* Don't allow bridging non-ethernet like devices, or DSA-enabled
570 * master network devices since the bridge layer rx_handler prevents
571 * the DSA fake ethertype handler to be invoked, so we do not strip off
572 * the DSA switch tag protocol header and the bridge layer just return
573 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
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 netdev_uses_dsa(dev))
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 kfree(p); /* kobject not yet init'd, manually free */
612 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
613 SYSFS_BRIDGE_PORT_ATTR);
617 err = br_sysfs_addif(p);
621 err = br_netpoll_enable(p);
625 err = netdev_rx_handler_register(dev, br_handle_frame, p);
629 dev->priv_flags |= IFF_BRIDGE_PORT;
631 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack);
635 err = nbp_switchdev_mark_set(p);
639 dev_disable_lro(dev);
641 list_add_rcu(&p->list, &br->port_list);
643 nbp_update_port_count(br);
644 if (!br_promisc_port(p) && (p->dev->priv_flags & IFF_UNICAST_FLT)) {
645 /* When updating the port count we also update all ports'
647 * A port leaving promiscuous mode normally gets the bridge's
648 * fdb synced to the unicast filter (if supported), however,
649 * `br_port_clear_promisc` does not distinguish between
650 * non-promiscuous ports and *new* ports, so we need to
651 * sync explicitly here.
653 fdb_synced = br_fdb_sync_static(br, p) == 0;
655 netdev_err(dev, "failed to sync bridge static fdb addresses to this port\n");
658 netdev_update_features(br->dev);
660 br_hr = br->dev->needed_headroom;
661 dev_hr = netdev_get_fwd_headroom(dev);
663 update_headroom(br, dev_hr);
665 netdev_set_rx_headroom(dev, br_hr);
667 if (br_fdb_insert(br, p, dev->dev_addr, 0))
668 netdev_err(dev, "failed insert local address bridge forwarding table\n");
670 err = nbp_vlan_init(p);
672 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
676 spin_lock_bh(&br->lock);
677 changed_addr = br_stp_recalculate_bridge_id(br);
679 if (netif_running(dev) && netif_oper_up(dev) &&
680 (br->dev->flags & IFF_UP))
681 br_stp_enable_port(p);
682 spin_unlock_bh(&br->lock);
684 br_ifinfo_notify(RTM_NEWLINK, NULL, p);
687 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
689 br_mtu_auto_adjust(br);
690 br_set_gso_limits(br);
692 kobject_uevent(&p->kobj, KOBJ_ADD);
698 br_fdb_unsync_static(br, p);
699 list_del_rcu(&p->list);
700 br_fdb_delete_by_port(br, p, 0, 1);
701 nbp_update_port_count(br);
703 netdev_upper_dev_unlink(dev, br->dev);
705 dev->priv_flags &= ~IFF_BRIDGE_PORT;
706 netdev_rx_handler_unregister(dev);
708 br_netpoll_disable(p);
710 sysfs_remove_link(br->ifobj, p->dev->name);
712 br_multicast_del_port(p);
713 kobject_put(&p->kobj);
714 dev_set_allmulti(dev, -1);
720 /* called with RTNL */
721 int br_del_if(struct net_bridge *br, struct net_device *dev)
723 struct net_bridge_port *p;
726 p = br_port_get_rtnl(dev);
727 if (!p || p->br != br)
730 /* Since more than one interface can be attached to a bridge,
731 * there still maybe an alternate path for netconsole to use;
732 * therefore there is no reason for a NETDEV_RELEASE event.
736 br_mtu_auto_adjust(br);
737 br_set_gso_limits(br);
739 spin_lock_bh(&br->lock);
740 changed_addr = br_stp_recalculate_bridge_id(br);
741 spin_unlock_bh(&br->lock);
744 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
746 netdev_update_features(br->dev);
751 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
753 struct net_bridge *br = p->br;
755 if (mask & BR_AUTO_MASK)
756 nbp_update_port_count(br);
758 if (mask & BR_NEIGH_SUPPRESS)
759 br_recalculate_neigh_suppress_enabled(br);