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>
26 #include <linux/if_vlan.h>
27 #include <net/switchdev.h>
29 #include "br_private.h"
32 * Determine initial path cost based on speed.
33 * using recommendations from 802.1d standard
35 * Since driver might sleep need to not be holding any locks.
37 static int port_cost(struct net_device *dev)
39 struct ethtool_link_ksettings ecmd;
41 if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
42 switch (ecmd.base.speed) {
54 /* Old silly heuristics based on name */
55 if (!strncmp(dev->name, "lec", 3))
58 if (!strncmp(dev->name, "plip", 4))
61 return 100; /* assume old 10Mbps */
65 /* Check for port carrier transitions. */
66 void br_port_carrier_check(struct net_bridge_port *p)
68 struct net_device *dev = p->dev;
69 struct net_bridge *br = p->br;
71 if (!(p->flags & BR_ADMIN_COST) &&
72 netif_running(dev) && netif_oper_up(dev))
73 p->path_cost = port_cost(dev);
75 if (!netif_running(br->dev))
78 spin_lock_bh(&br->lock);
79 if (netif_running(dev) && netif_oper_up(dev)) {
80 if (p->state == BR_STATE_DISABLED)
81 br_stp_enable_port(p);
83 if (p->state != BR_STATE_DISABLED)
84 br_stp_disable_port(p);
86 spin_unlock_bh(&br->lock);
89 static void br_port_set_promisc(struct net_bridge_port *p)
93 if (br_promisc_port(p))
96 err = dev_set_promiscuity(p->dev, 1);
100 br_fdb_unsync_static(p->br, p);
101 p->flags |= BR_PROMISC;
104 static void br_port_clear_promisc(struct net_bridge_port *p)
108 /* Check if the port is already non-promisc or if it doesn't
109 * support UNICAST filtering. Without unicast filtering support
110 * we'll end up re-enabling promisc mode anyway, so just check for
113 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
116 /* Since we'll be clearing the promisc mode, program the port
117 * first so that we don't have interruption in traffic.
119 err = br_fdb_sync_static(p->br, p);
123 dev_set_promiscuity(p->dev, -1);
124 p->flags &= ~BR_PROMISC;
127 /* When a port is added or removed or when certain port flags
128 * change, this function is called to automatically manage
129 * promiscuity setting of all the bridge ports. We are always called
130 * under RTNL so can skip using rcu primitives.
132 void br_manage_promisc(struct net_bridge *br)
134 struct net_bridge_port *p;
135 bool set_all = false;
137 /* If vlan filtering is disabled or bridge interface is placed
138 * into promiscuous mode, place all ports in promiscuous mode.
140 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br))
143 list_for_each_entry(p, &br->port_list, list) {
145 br_port_set_promisc(p);
147 /* If the number of auto-ports is <= 1, then all other
148 * ports will have their output configuration
149 * statically specified through fdbs. Since ingress
150 * on the auto-port becomes forwarding/egress to other
151 * ports and egress configuration is statically known,
152 * we can say that ingress configuration of the
153 * auto-port is also statically known.
154 * This lets us disable promiscuous mode and write
157 if (br->auto_cnt == 0 ||
158 (br->auto_cnt == 1 && br_auto_port(p)))
159 br_port_clear_promisc(p);
161 br_port_set_promisc(p);
166 static void nbp_update_port_count(struct net_bridge *br)
168 struct net_bridge_port *p;
171 list_for_each_entry(p, &br->port_list, list) {
175 if (br->auto_cnt != cnt) {
177 br_manage_promisc(br);
181 static void nbp_delete_promisc(struct net_bridge_port *p)
183 /* If port is currently promiscuous, unset promiscuity.
184 * Otherwise, it is a static port so remove all addresses
187 dev_set_allmulti(p->dev, -1);
188 if (br_promisc_port(p))
189 dev_set_promiscuity(p->dev, -1);
191 br_fdb_unsync_static(p->br, p);
194 static void release_nbp(struct kobject *kobj)
196 struct net_bridge_port *p
197 = container_of(kobj, struct net_bridge_port, kobj);
201 static struct kobj_type brport_ktype = {
203 .sysfs_ops = &brport_sysfs_ops,
205 .release = release_nbp,
208 static void destroy_nbp(struct net_bridge_port *p)
210 struct net_device *dev = p->dev;
216 kobject_put(&p->kobj);
219 static void destroy_nbp_rcu(struct rcu_head *head)
221 struct net_bridge_port *p =
222 container_of(head, struct net_bridge_port, rcu);
226 static unsigned get_max_headroom(struct net_bridge *br)
228 unsigned max_headroom = 0;
229 struct net_bridge_port *p;
231 list_for_each_entry(p, &br->port_list, list) {
232 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
234 if (dev_headroom > max_headroom)
235 max_headroom = dev_headroom;
241 static void update_headroom(struct net_bridge *br, int new_hr)
243 struct net_bridge_port *p;
245 list_for_each_entry(p, &br->port_list, list)
246 netdev_set_rx_headroom(p->dev, new_hr);
248 br->dev->needed_headroom = new_hr;
251 /* Delete port(interface) from bridge is done in two steps.
252 * via RCU. First step, marks device as down. That deletes
253 * all the timers and stops new packets from flowing through.
255 * Final cleanup doesn't occur until after all CPU's finished
256 * processing packets.
258 * Protected from multiple admin operations by RTNL mutex
260 static void del_nbp(struct net_bridge_port *p)
262 struct net_bridge *br = p->br;
263 struct net_device *dev = p->dev;
265 sysfs_remove_link(br->ifobj, p->dev->name);
267 nbp_delete_promisc(p);
269 spin_lock_bh(&br->lock);
270 br_stp_disable_port(p);
271 spin_unlock_bh(&br->lock);
273 br_ifinfo_notify(RTM_DELLINK, p);
275 list_del_rcu(&p->list);
276 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
277 update_headroom(br, get_max_headroom(br));
278 netdev_reset_rx_headroom(dev);
281 br_fdb_delete_by_port(br, p, 0, 1);
282 switchdev_deferred_process();
284 nbp_update_port_count(br);
286 netdev_upper_dev_unlink(dev, br->dev);
288 dev->priv_flags &= ~IFF_BRIDGE_PORT;
290 netdev_rx_handler_unregister(dev);
292 br_multicast_del_port(p);
294 kobject_uevent(&p->kobj, KOBJ_REMOVE);
295 kobject_del(&p->kobj);
297 br_netpoll_disable(p);
299 call_rcu(&p->rcu, destroy_nbp_rcu);
302 /* Delete bridge device */
303 void br_dev_delete(struct net_device *dev, struct list_head *head)
305 struct net_bridge *br = netdev_priv(dev);
306 struct net_bridge_port *p, *n;
308 list_for_each_entry_safe(p, n, &br->port_list, list) {
312 br_fdb_delete_by_port(br, NULL, 0, 1);
315 br_multicast_dev_del(br);
316 del_timer_sync(&br->gc_timer);
318 br_sysfs_delbr(br->dev);
319 unregister_netdevice_queue(br->dev, head);
322 /* find an available port number */
323 static int find_portno(struct net_bridge *br)
326 struct net_bridge_port *p;
327 unsigned long *inuse;
329 inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
334 set_bit(0, inuse); /* zero is reserved */
335 list_for_each_entry(p, &br->port_list, list) {
336 set_bit(p->port_no, inuse);
338 index = find_first_zero_bit(inuse, BR_MAX_PORTS);
341 return (index >= BR_MAX_PORTS) ? -EXFULL : index;
344 /* called with RTNL but without bridge lock */
345 static struct net_bridge_port *new_nbp(struct net_bridge *br,
346 struct net_device *dev)
349 struct net_bridge_port *p;
351 index = find_portno(br);
353 return ERR_PTR(index);
355 p = kzalloc(sizeof(*p), GFP_KERNEL);
357 return ERR_PTR(-ENOMEM);
362 p->path_cost = port_cost(dev);
363 p->priority = 0x8000 >> BR_PORT_BITS;
365 p->flags = BR_LEARNING | BR_FLOOD;
367 br_set_state(p, BR_STATE_DISABLED);
368 br_stp_port_timer_init(p);
369 br_multicast_add_port(p);
374 int br_add_bridge(struct net *net, const char *name)
376 struct net_device *dev;
379 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
385 dev_net_set(dev, net);
386 dev->rtnl_link_ops = &br_link_ops;
388 res = register_netdev(dev);
394 int br_del_bridge(struct net *net, const char *name)
396 struct net_device *dev;
400 dev = __dev_get_by_name(net, name);
402 ret = -ENXIO; /* Could not find device */
404 else if (!(dev->priv_flags & IFF_EBRIDGE)) {
405 /* Attempt to delete non bridge device! */
409 else if (dev->flags & IFF_UP) {
410 /* Not shutdown yet. */
415 br_dev_delete(dev, NULL);
421 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
422 int br_min_mtu(const struct net_bridge *br)
424 const struct net_bridge_port *p;
429 if (list_empty(&br->port_list))
432 list_for_each_entry(p, &br->port_list, list) {
433 if (!mtu || p->dev->mtu < mtu)
441 * Recomputes features using slave's features
443 netdev_features_t br_features_recompute(struct net_bridge *br,
444 netdev_features_t features)
446 struct net_bridge_port *p;
447 netdev_features_t mask;
449 if (list_empty(&br->port_list))
453 features &= ~NETIF_F_ONE_FOR_ALL;
455 list_for_each_entry(p, &br->port_list, list) {
456 features = netdev_increment_features(features,
457 p->dev->features, mask);
459 features = netdev_add_tso_features(features, mask);
464 /* called with RTNL */
465 int br_add_if(struct net_bridge *br, struct net_device *dev)
467 struct net_bridge_port *p;
469 unsigned br_hr, dev_hr;
472 /* Don't allow bridging non-ethernet like devices, or DSA-enabled
473 * master network devices since the bridge layer rx_handler prevents
474 * the DSA fake ethertype handler to be invoked, so we do not strip off
475 * the DSA switch tag protocol header and the bridge layer just return
476 * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
478 if ((dev->flags & IFF_LOOPBACK) ||
479 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
480 !is_valid_ether_addr(dev->dev_addr) ||
481 netdev_uses_dsa(dev))
484 /* No bridging of bridges */
485 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
488 /* Device is already being bridged */
489 if (br_port_exists(dev))
492 /* No bridging devices that dislike that (e.g. wireless) */
493 if (dev->priv_flags & IFF_DONT_BRIDGE)
496 p = new_nbp(br, dev);
500 call_netdevice_notifiers(NETDEV_JOIN, dev);
502 err = dev_set_allmulti(dev, 1);
506 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
507 SYSFS_BRIDGE_PORT_ATTR);
511 err = br_sysfs_addif(p);
515 err = br_netpoll_enable(p);
519 err = netdev_rx_handler_register(dev, br_handle_frame, p);
523 dev->priv_flags |= IFF_BRIDGE_PORT;
525 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL);
529 dev_disable_lro(dev);
531 list_add_rcu(&p->list, &br->port_list);
533 nbp_update_port_count(br);
535 netdev_update_features(br->dev);
537 br_hr = br->dev->needed_headroom;
538 dev_hr = netdev_get_fwd_headroom(dev);
540 update_headroom(br, dev_hr);
542 netdev_set_rx_headroom(dev, br_hr);
544 if (br_fdb_insert(br, p, dev->dev_addr, 0))
545 netdev_err(dev, "failed insert local address bridge forwarding table\n");
547 err = nbp_vlan_init(p);
549 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
553 spin_lock_bh(&br->lock);
554 changed_addr = br_stp_recalculate_bridge_id(br);
556 if (netif_running(dev) && netif_oper_up(dev) &&
557 (br->dev->flags & IFF_UP))
558 br_stp_enable_port(p);
559 spin_unlock_bh(&br->lock);
561 br_ifinfo_notify(RTM_NEWLINK, p);
564 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
566 dev_set_mtu(br->dev, br_min_mtu(br));
568 kobject_uevent(&p->kobj, KOBJ_ADD);
573 list_del_rcu(&p->list);
574 br_fdb_delete_by_port(br, p, 0, 1);
575 nbp_update_port_count(br);
576 netdev_upper_dev_unlink(dev, br->dev);
579 dev->priv_flags &= ~IFF_BRIDGE_PORT;
580 netdev_rx_handler_unregister(dev);
582 br_netpoll_disable(p);
584 sysfs_remove_link(br->ifobj, p->dev->name);
586 kobject_put(&p->kobj);
587 p = NULL; /* kobject_put frees */
589 dev_set_allmulti(dev, -1);
596 /* called with RTNL */
597 int br_del_if(struct net_bridge *br, struct net_device *dev)
599 struct net_bridge_port *p;
602 p = br_port_get_rtnl(dev);
603 if (!p || p->br != br)
606 /* Since more than one interface can be attached to a bridge,
607 * there still maybe an alternate path for netconsole to use;
608 * therefore there is no reason for a NETDEV_RELEASE event.
612 dev_set_mtu(br->dev, br_min_mtu(br));
614 spin_lock_bh(&br->lock);
615 changed_addr = br_stp_recalculate_bridge_id(br);
616 spin_unlock_bh(&br->lock);
619 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
621 netdev_update_features(br->dev);
626 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
628 struct net_bridge *br = p->br;
630 if (mask & BR_AUTO_MASK)
631 nbp_update_port_count(br);