1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/list.h>
4 #include <linux/netdevice.h>
5 #include <linux/rtnetlink.h>
6 #include <linux/skbuff.h>
8 #include <net/switchdev.h>
10 #include "br_private.h"
12 static struct static_key_false br_switchdev_tx_fwd_offload;
14 static bool nbp_switchdev_can_offload_tx_fwd(const struct net_bridge_port *p,
15 const struct sk_buff *skb)
17 if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
20 return (p->flags & BR_TX_FWD_OFFLOAD) &&
21 (p->hwdom != BR_INPUT_SKB_CB(skb)->src_hwdom);
24 bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb)
26 if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
29 return BR_INPUT_SKB_CB(skb)->tx_fwd_offload;
32 void br_switchdev_frame_set_offload_fwd_mark(struct sk_buff *skb)
34 skb->offload_fwd_mark = br_switchdev_frame_uses_tx_fwd_offload(skb);
37 /* Mark the frame for TX forwarding offload if this egress port supports it */
38 void nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p,
41 if (nbp_switchdev_can_offload_tx_fwd(p, skb))
42 BR_INPUT_SKB_CB(skb)->tx_fwd_offload = true;
45 /* Lazily adds the hwdom of the egress bridge port to the bit mask of hwdoms
46 * that the skb has been already forwarded to, to avoid further cloning to
47 * other ports in the same hwdom by making nbp_switchdev_allowed_egress()
50 void nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p,
53 if (nbp_switchdev_can_offload_tx_fwd(p, skb))
54 set_bit(p->hwdom, &BR_INPUT_SKB_CB(skb)->fwd_hwdoms);
57 void nbp_switchdev_frame_mark(const struct net_bridge_port *p,
61 BR_INPUT_SKB_CB(skb)->src_hwdom = p->hwdom;
64 bool nbp_switchdev_allowed_egress(const struct net_bridge_port *p,
65 const struct sk_buff *skb)
67 struct br_input_skb_cb *cb = BR_INPUT_SKB_CB(skb);
69 return !test_bit(p->hwdom, &cb->fwd_hwdoms) &&
70 (!skb->offload_fwd_mark || cb->src_hwdom != p->hwdom);
73 /* Flags that can be offloaded to hardware */
74 #define BR_PORT_FLAGS_HW_OFFLOAD (BR_LEARNING | BR_FLOOD | BR_PORT_MAB | \
75 BR_MCAST_FLOOD | BR_BCAST_FLOOD | BR_PORT_LOCKED | \
76 BR_HAIRPIN_MODE | BR_ISOLATED | BR_MULTICAST_TO_UNICAST)
78 int br_switchdev_set_port_flag(struct net_bridge_port *p,
81 struct netlink_ext_ack *extack)
83 struct switchdev_attr attr = {
86 struct switchdev_notifier_port_attr_info info = {
91 mask &= BR_PORT_FLAGS_HW_OFFLOAD;
95 attr.id = SWITCHDEV_ATTR_ID_PORT_PRE_BRIDGE_FLAGS;
96 attr.u.brport_flags.val = flags;
97 attr.u.brport_flags.mask = mask;
99 /* We run from atomic context here */
100 err = call_switchdev_notifiers(SWITCHDEV_PORT_ATTR_SET, p->dev,
102 err = notifier_to_errno(err);
103 if (err == -EOPNOTSUPP)
107 NL_SET_ERR_MSG_WEAK_MOD(extack,
108 "bridge flag offload is not supported");
112 attr.id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS;
113 attr.flags = SWITCHDEV_F_DEFER;
115 err = switchdev_port_attr_set(p->dev, &attr, extack);
117 NL_SET_ERR_MSG_WEAK_MOD(extack,
118 "error setting offload flag on port");
125 static void br_switchdev_fdb_populate(struct net_bridge *br,
126 struct switchdev_notifier_fdb_info *item,
127 const struct net_bridge_fdb_entry *fdb,
130 const struct net_bridge_port *p = READ_ONCE(fdb->dst);
132 item->addr = fdb->key.addr.addr;
133 item->vid = fdb->key.vlan_id;
134 item->added_by_user = test_bit(BR_FDB_ADDED_BY_USER, &fdb->flags);
135 item->offloaded = test_bit(BR_FDB_OFFLOADED, &fdb->flags);
136 item->is_local = test_bit(BR_FDB_LOCAL, &fdb->flags);
137 item->locked = false;
138 item->info.dev = (!p || item->is_local) ? br->dev : p->dev;
139 item->info.ctx = ctx;
143 br_switchdev_fdb_notify(struct net_bridge *br,
144 const struct net_bridge_fdb_entry *fdb, int type)
146 struct switchdev_notifier_fdb_info item;
148 if (test_bit(BR_FDB_LOCKED, &fdb->flags))
151 /* Entries with these flags were created using ndm_state == NUD_REACHABLE,
152 * ndm_flags == NTF_MASTER( | NTF_STICKY), ext_flags == 0 by something
153 * equivalent to 'bridge fdb add ... master dynamic (sticky)'.
154 * Drivers don't know how to deal with these, so don't notify them to
155 * avoid confusing them.
157 if (test_bit(BR_FDB_ADDED_BY_USER, &fdb->flags) &&
158 !test_bit(BR_FDB_STATIC, &fdb->flags) &&
159 !test_bit(BR_FDB_ADDED_BY_EXT_LEARN, &fdb->flags))
162 br_switchdev_fdb_populate(br, &item, fdb, NULL);
166 call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_DEVICE,
167 item.info.dev, &item.info, NULL);
170 call_switchdev_notifiers(SWITCHDEV_FDB_ADD_TO_DEVICE,
171 item.info.dev, &item.info, NULL);
176 int br_switchdev_port_vlan_add(struct net_device *dev, u16 vid, u16 flags,
177 bool changed, struct netlink_ext_ack *extack)
179 struct switchdev_obj_port_vlan v = {
181 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
187 return switchdev_port_obj_add(dev, &v.obj, extack);
190 int br_switchdev_port_vlan_del(struct net_device *dev, u16 vid)
192 struct switchdev_obj_port_vlan v = {
194 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
198 return switchdev_port_obj_del(dev, &v.obj);
201 static int nbp_switchdev_hwdom_set(struct net_bridge_port *joining)
203 struct net_bridge *br = joining->br;
204 struct net_bridge_port *p;
207 /* joining is yet to be added to the port list. */
208 list_for_each_entry(p, &br->port_list, list) {
209 if (netdev_phys_item_id_same(&joining->ppid, &p->ppid)) {
210 joining->hwdom = p->hwdom;
215 hwdom = find_next_zero_bit(&br->busy_hwdoms, BR_HWDOM_MAX, 1);
216 if (hwdom >= BR_HWDOM_MAX)
219 set_bit(hwdom, &br->busy_hwdoms);
220 joining->hwdom = hwdom;
224 static void nbp_switchdev_hwdom_put(struct net_bridge_port *leaving)
226 struct net_bridge *br = leaving->br;
227 struct net_bridge_port *p;
229 /* leaving is no longer in the port list. */
230 list_for_each_entry(p, &br->port_list, list) {
231 if (p->hwdom == leaving->hwdom)
235 clear_bit(leaving->hwdom, &br->busy_hwdoms);
238 static int nbp_switchdev_add(struct net_bridge_port *p,
239 struct netdev_phys_item_id ppid,
241 struct netlink_ext_ack *extack)
245 if (p->offload_count) {
246 /* Prevent unsupported configurations such as a bridge port
247 * which is a bonding interface, and the member ports are from
248 * different hardware switches.
250 if (!netdev_phys_item_id_same(&p->ppid, &ppid)) {
251 NL_SET_ERR_MSG_MOD(extack,
252 "Same bridge port cannot be offloaded by two physical switches");
256 /* Tolerate drivers that call switchdev_bridge_port_offload()
257 * more than once for the same bridge port, such as when the
258 * bridge port is an offloaded bonding/team interface.
266 p->offload_count = 1;
268 err = nbp_switchdev_hwdom_set(p);
272 if (tx_fwd_offload) {
273 p->flags |= BR_TX_FWD_OFFLOAD;
274 static_branch_inc(&br_switchdev_tx_fwd_offload);
280 static void nbp_switchdev_del(struct net_bridge_port *p)
282 if (WARN_ON(!p->offload_count))
287 if (p->offload_count)
291 nbp_switchdev_hwdom_put(p);
293 if (p->flags & BR_TX_FWD_OFFLOAD) {
294 p->flags &= ~BR_TX_FWD_OFFLOAD;
295 static_branch_dec(&br_switchdev_tx_fwd_offload);
300 br_switchdev_fdb_replay_one(struct net_bridge *br, struct notifier_block *nb,
301 const struct net_bridge_fdb_entry *fdb,
302 unsigned long action, const void *ctx)
304 struct switchdev_notifier_fdb_info item;
307 br_switchdev_fdb_populate(br, &item, fdb, ctx);
309 err = nb->notifier_call(nb, action, &item);
310 return notifier_to_errno(err);
314 br_switchdev_fdb_replay(const struct net_device *br_dev, const void *ctx,
315 bool adding, struct notifier_block *nb)
317 struct net_bridge_fdb_entry *fdb;
318 struct net_bridge *br;
319 unsigned long action;
325 if (!netif_is_bridge_master(br_dev))
328 br = netdev_priv(br_dev);
331 action = SWITCHDEV_FDB_ADD_TO_DEVICE;
333 action = SWITCHDEV_FDB_DEL_TO_DEVICE;
337 hlist_for_each_entry_rcu(fdb, &br->fdb_list, fdb_node) {
338 err = br_switchdev_fdb_replay_one(br, nb, fdb, action, ctx);
348 static int br_switchdev_vlan_attr_replay(struct net_device *br_dev,
350 struct notifier_block *nb,
351 struct netlink_ext_ack *extack)
353 struct switchdev_notifier_port_attr_info attr_info = {
360 struct net_bridge *br = netdev_priv(br_dev);
361 struct net_bridge_vlan_group *vg;
362 struct switchdev_attr attr;
363 struct net_bridge_vlan *v;
366 attr_info.attr = &attr;
367 attr.orig_dev = br_dev;
369 vg = br_vlan_group(br);
373 list_for_each_entry(v, &vg->vlan_list, vlist) {
375 attr.id = SWITCHDEV_ATTR_ID_VLAN_MSTI;
376 attr.u.vlan_msti.vid = v->vid;
377 attr.u.vlan_msti.msti = v->msti;
379 err = nb->notifier_call(nb, SWITCHDEV_PORT_ATTR_SET,
381 err = notifier_to_errno(err);
391 br_switchdev_vlan_replay_one(struct notifier_block *nb,
392 struct net_device *dev,
393 struct switchdev_obj_port_vlan *vlan,
394 const void *ctx, unsigned long action,
395 struct netlink_ext_ack *extack)
397 struct switchdev_notifier_port_obj_info obj_info = {
407 err = nb->notifier_call(nb, action, &obj_info);
408 return notifier_to_errno(err);
411 static int br_switchdev_vlan_replay_group(struct notifier_block *nb,
412 struct net_device *dev,
413 struct net_bridge_vlan_group *vg,
414 const void *ctx, unsigned long action,
415 struct netlink_ext_ack *extack)
417 struct net_bridge_vlan *v;
424 pvid = br_get_pvid(vg);
426 list_for_each_entry(v, &vg->vlan_list, vlist) {
427 struct switchdev_obj_port_vlan vlan = {
429 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
430 .flags = br_vlan_flags(v, pvid),
434 if (!br_vlan_should_use(v))
437 err = br_switchdev_vlan_replay_one(nb, dev, &vlan, ctx,
446 static int br_switchdev_vlan_replay(struct net_device *br_dev,
447 const void *ctx, bool adding,
448 struct notifier_block *nb,
449 struct netlink_ext_ack *extack)
451 struct net_bridge *br = netdev_priv(br_dev);
452 struct net_bridge_port *p;
453 unsigned long action;
461 if (!netif_is_bridge_master(br_dev))
465 action = SWITCHDEV_PORT_OBJ_ADD;
467 action = SWITCHDEV_PORT_OBJ_DEL;
469 err = br_switchdev_vlan_replay_group(nb, br_dev, br_vlan_group(br),
470 ctx, action, extack);
474 list_for_each_entry(p, &br->port_list, list) {
475 struct net_device *dev = p->dev;
477 err = br_switchdev_vlan_replay_group(nb, dev,
479 ctx, action, extack);
485 err = br_switchdev_vlan_attr_replay(br_dev, ctx, nb, extack);
493 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
494 struct br_switchdev_mdb_complete_info {
495 struct net_bridge_port *port;
499 static void br_switchdev_mdb_complete(struct net_device *dev, int err, void *priv)
501 struct br_switchdev_mdb_complete_info *data = priv;
502 struct net_bridge_port_group __rcu **pp;
503 struct net_bridge_port_group *p;
504 struct net_bridge_mdb_entry *mp;
505 struct net_bridge_port *port = data->port;
506 struct net_bridge *br = port->br;
511 spin_lock_bh(&br->multicast_lock);
512 mp = br_mdb_ip_get(br, &data->ip);
515 for (pp = &mp->ports; (p = mlock_dereference(*pp, br)) != NULL;
517 if (p->key.port != port)
519 p->flags |= MDB_PG_FLAGS_OFFLOAD;
522 spin_unlock_bh(&br->multicast_lock);
527 static void br_switchdev_mdb_populate(struct switchdev_obj_port_mdb *mdb,
528 const struct net_bridge_mdb_entry *mp)
530 if (mp->addr.proto == htons(ETH_P_IP))
531 ip_eth_mc_map(mp->addr.dst.ip4, mdb->addr);
532 #if IS_ENABLED(CONFIG_IPV6)
533 else if (mp->addr.proto == htons(ETH_P_IPV6))
534 ipv6_eth_mc_map(&mp->addr.dst.ip6, mdb->addr);
537 ether_addr_copy(mdb->addr, mp->addr.dst.mac_addr);
539 mdb->vid = mp->addr.vid;
542 static void br_switchdev_host_mdb_one(struct net_device *dev,
543 struct net_device *lower_dev,
544 struct net_bridge_mdb_entry *mp,
547 struct switchdev_obj_port_mdb mdb = {
549 .id = SWITCHDEV_OBJ_ID_HOST_MDB,
550 .flags = SWITCHDEV_F_DEFER,
555 br_switchdev_mdb_populate(&mdb, mp);
559 switchdev_port_obj_add(lower_dev, &mdb.obj, NULL);
562 switchdev_port_obj_del(lower_dev, &mdb.obj);
567 static void br_switchdev_host_mdb(struct net_device *dev,
568 struct net_bridge_mdb_entry *mp, int type)
570 struct net_device *lower_dev;
571 struct list_head *iter;
573 netdev_for_each_lower_dev(dev, lower_dev, iter)
574 br_switchdev_host_mdb_one(dev, lower_dev, mp, type);
578 br_switchdev_mdb_replay_one(struct notifier_block *nb, struct net_device *dev,
579 const struct switchdev_obj_port_mdb *mdb,
580 unsigned long action, const void *ctx,
581 struct netlink_ext_ack *extack)
583 struct switchdev_notifier_port_obj_info obj_info = {
593 err = nb->notifier_call(nb, action, &obj_info);
594 return notifier_to_errno(err);
597 static int br_switchdev_mdb_queue_one(struct list_head *mdb_list,
598 enum switchdev_obj_id id,
599 const struct net_bridge_mdb_entry *mp,
600 struct net_device *orig_dev)
602 struct switchdev_obj_port_mdb *mdb;
604 mdb = kzalloc(sizeof(*mdb), GFP_ATOMIC);
609 mdb->obj.orig_dev = orig_dev;
610 br_switchdev_mdb_populate(mdb, mp);
611 list_add_tail(&mdb->obj.list, mdb_list);
616 void br_switchdev_mdb_notify(struct net_device *dev,
617 struct net_bridge_mdb_entry *mp,
618 struct net_bridge_port_group *pg,
621 struct br_switchdev_mdb_complete_info *complete_info;
622 struct switchdev_obj_port_mdb mdb = {
624 .id = SWITCHDEV_OBJ_ID_PORT_MDB,
625 .flags = SWITCHDEV_F_DEFER,
630 return br_switchdev_host_mdb(dev, mp, type);
632 br_switchdev_mdb_populate(&mdb, mp);
634 mdb.obj.orig_dev = pg->key.port->dev;
637 complete_info = kmalloc(sizeof(*complete_info), GFP_ATOMIC);
640 complete_info->port = pg->key.port;
641 complete_info->ip = mp->addr;
642 mdb.obj.complete_priv = complete_info;
643 mdb.obj.complete = br_switchdev_mdb_complete;
644 if (switchdev_port_obj_add(pg->key.port->dev, &mdb.obj, NULL))
645 kfree(complete_info);
648 switchdev_port_obj_del(pg->key.port->dev, &mdb.obj);
655 br_switchdev_mdb_replay(struct net_device *br_dev, struct net_device *dev,
656 const void *ctx, bool adding, struct notifier_block *nb,
657 struct netlink_ext_ack *extack)
659 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
660 const struct net_bridge_mdb_entry *mp;
661 struct switchdev_obj *obj, *tmp;
662 struct net_bridge *br;
663 unsigned long action;
672 if (!netif_is_bridge_master(br_dev) || !netif_is_bridge_port(dev))
675 br = netdev_priv(br_dev);
677 if (!br_opt_get(br, BROPT_MULTICAST_ENABLED))
680 /* We cannot walk over br->mdb_list protected just by the rtnl_mutex,
681 * because the write-side protection is br->multicast_lock. But we
682 * need to emulate the [ blocking ] calling context of a regular
683 * switchdev event, so since both br->multicast_lock and RCU read side
684 * critical sections are atomic, we have no choice but to pick the RCU
685 * read side lock, queue up all our events, leave the critical section
686 * and notify switchdev from blocking context.
690 hlist_for_each_entry_rcu(mp, &br->mdb_list, mdb_node) {
691 struct net_bridge_port_group __rcu * const *pp;
692 const struct net_bridge_port_group *p;
694 if (mp->host_joined) {
695 err = br_switchdev_mdb_queue_one(&mdb_list,
696 SWITCHDEV_OBJ_ID_HOST_MDB,
704 for (pp = &mp->ports; (p = rcu_dereference(*pp)) != NULL;
706 if (p->key.port->dev != dev)
709 err = br_switchdev_mdb_queue_one(&mdb_list,
710 SWITCHDEV_OBJ_ID_PORT_MDB,
722 action = SWITCHDEV_PORT_OBJ_ADD;
724 action = SWITCHDEV_PORT_OBJ_DEL;
726 list_for_each_entry(obj, &mdb_list, list) {
727 err = br_switchdev_mdb_replay_one(nb, dev,
728 SWITCHDEV_OBJ_PORT_MDB(obj),
729 action, ctx, extack);
730 if (err == -EOPNOTSUPP)
737 list_for_each_entry_safe(obj, tmp, &mdb_list, list) {
738 list_del(&obj->list);
739 kfree(SWITCHDEV_OBJ_PORT_MDB(obj));
749 static int nbp_switchdev_sync_objs(struct net_bridge_port *p, const void *ctx,
750 struct notifier_block *atomic_nb,
751 struct notifier_block *blocking_nb,
752 struct netlink_ext_ack *extack)
754 struct net_device *br_dev = p->br->dev;
755 struct net_device *dev = p->dev;
758 err = br_switchdev_vlan_replay(br_dev, ctx, true, blocking_nb, extack);
759 if (err && err != -EOPNOTSUPP)
762 err = br_switchdev_mdb_replay(br_dev, dev, ctx, true, blocking_nb,
765 /* -EOPNOTSUPP not propagated from MDB replay. */
769 err = br_switchdev_fdb_replay(br_dev, ctx, true, atomic_nb);
770 if (err && err != -EOPNOTSUPP)
776 static void nbp_switchdev_unsync_objs(struct net_bridge_port *p,
778 struct notifier_block *atomic_nb,
779 struct notifier_block *blocking_nb)
781 struct net_device *br_dev = p->br->dev;
782 struct net_device *dev = p->dev;
784 br_switchdev_fdb_replay(br_dev, ctx, false, atomic_nb);
786 br_switchdev_mdb_replay(br_dev, dev, ctx, false, blocking_nb, NULL);
788 br_switchdev_vlan_replay(br_dev, ctx, false, blocking_nb, NULL);
791 /* Let the bridge know that this port is offloaded, so that it can assign a
792 * switchdev hardware domain to it.
794 int br_switchdev_port_offload(struct net_bridge_port *p,
795 struct net_device *dev, const void *ctx,
796 struct notifier_block *atomic_nb,
797 struct notifier_block *blocking_nb,
799 struct netlink_ext_ack *extack)
801 struct netdev_phys_item_id ppid;
804 err = dev_get_port_parent_id(dev, &ppid, false);
808 err = nbp_switchdev_add(p, ppid, tx_fwd_offload, extack);
812 err = nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);
814 goto out_switchdev_del;
819 nbp_switchdev_del(p);
824 void br_switchdev_port_unoffload(struct net_bridge_port *p, const void *ctx,
825 struct notifier_block *atomic_nb,
826 struct notifier_block *blocking_nb)
828 nbp_switchdev_unsync_objs(p, ctx, atomic_nb, blocking_nb);
830 nbp_switchdev_del(p);
833 int br_switchdev_port_replay(struct net_bridge_port *p,
834 struct net_device *dev, const void *ctx,
835 struct notifier_block *atomic_nb,
836 struct notifier_block *blocking_nb,
837 struct netlink_ext_ack *extack)
839 return nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);