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 | \
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 if (extack && !extack->_msg)
108 NL_SET_ERR_MSG_MOD(extack,
109 "bridge flag offload is not supported");
113 attr.id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS;
114 attr.flags = SWITCHDEV_F_DEFER;
116 err = switchdev_port_attr_set(p->dev, &attr, extack);
118 if (extack && !extack->_msg)
119 NL_SET_ERR_MSG_MOD(extack,
120 "error setting offload flag on port");
127 static void br_switchdev_fdb_populate(struct net_bridge *br,
128 struct switchdev_notifier_fdb_info *item,
129 const struct net_bridge_fdb_entry *fdb,
132 const struct net_bridge_port *p = READ_ONCE(fdb->dst);
134 item->addr = fdb->key.addr.addr;
135 item->vid = fdb->key.vlan_id;
136 item->added_by_user = test_bit(BR_FDB_ADDED_BY_USER, &fdb->flags);
137 item->offloaded = test_bit(BR_FDB_OFFLOADED, &fdb->flags);
138 item->is_local = test_bit(BR_FDB_LOCAL, &fdb->flags);
139 item->info.dev = (!p || item->is_local) ? br->dev : p->dev;
140 item->info.ctx = ctx;
144 br_switchdev_fdb_notify(struct net_bridge *br,
145 const struct net_bridge_fdb_entry *fdb, int type)
147 struct switchdev_notifier_fdb_info item;
149 br_switchdev_fdb_populate(br, &item, fdb, NULL);
153 call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_DEVICE,
154 item.info.dev, &item.info, NULL);
157 call_switchdev_notifiers(SWITCHDEV_FDB_ADD_TO_DEVICE,
158 item.info.dev, &item.info, NULL);
163 int br_switchdev_port_vlan_add(struct net_device *dev, u16 vid, u16 flags,
164 bool changed, struct netlink_ext_ack *extack)
166 struct switchdev_obj_port_vlan v = {
168 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
174 return switchdev_port_obj_add(dev, &v.obj, extack);
177 int br_switchdev_port_vlan_del(struct net_device *dev, u16 vid)
179 struct switchdev_obj_port_vlan v = {
181 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
185 return switchdev_port_obj_del(dev, &v.obj);
188 static int nbp_switchdev_hwdom_set(struct net_bridge_port *joining)
190 struct net_bridge *br = joining->br;
191 struct net_bridge_port *p;
194 /* joining is yet to be added to the port list. */
195 list_for_each_entry(p, &br->port_list, list) {
196 if (netdev_phys_item_id_same(&joining->ppid, &p->ppid)) {
197 joining->hwdom = p->hwdom;
202 hwdom = find_next_zero_bit(&br->busy_hwdoms, BR_HWDOM_MAX, 1);
203 if (hwdom >= BR_HWDOM_MAX)
206 set_bit(hwdom, &br->busy_hwdoms);
207 joining->hwdom = hwdom;
211 static void nbp_switchdev_hwdom_put(struct net_bridge_port *leaving)
213 struct net_bridge *br = leaving->br;
214 struct net_bridge_port *p;
216 /* leaving is no longer in the port list. */
217 list_for_each_entry(p, &br->port_list, list) {
218 if (p->hwdom == leaving->hwdom)
222 clear_bit(leaving->hwdom, &br->busy_hwdoms);
225 static int nbp_switchdev_add(struct net_bridge_port *p,
226 struct netdev_phys_item_id ppid,
228 struct netlink_ext_ack *extack)
232 if (p->offload_count) {
233 /* Prevent unsupported configurations such as a bridge port
234 * which is a bonding interface, and the member ports are from
235 * different hardware switches.
237 if (!netdev_phys_item_id_same(&p->ppid, &ppid)) {
238 NL_SET_ERR_MSG_MOD(extack,
239 "Same bridge port cannot be offloaded by two physical switches");
243 /* Tolerate drivers that call switchdev_bridge_port_offload()
244 * more than once for the same bridge port, such as when the
245 * bridge port is an offloaded bonding/team interface.
253 p->offload_count = 1;
255 err = nbp_switchdev_hwdom_set(p);
259 if (tx_fwd_offload) {
260 p->flags |= BR_TX_FWD_OFFLOAD;
261 static_branch_inc(&br_switchdev_tx_fwd_offload);
267 static void nbp_switchdev_del(struct net_bridge_port *p)
269 if (WARN_ON(!p->offload_count))
274 if (p->offload_count)
278 nbp_switchdev_hwdom_put(p);
280 if (p->flags & BR_TX_FWD_OFFLOAD) {
281 p->flags &= ~BR_TX_FWD_OFFLOAD;
282 static_branch_dec(&br_switchdev_tx_fwd_offload);
287 br_switchdev_fdb_replay_one(struct net_bridge *br, struct notifier_block *nb,
288 const struct net_bridge_fdb_entry *fdb,
289 unsigned long action, const void *ctx)
291 struct switchdev_notifier_fdb_info item;
294 br_switchdev_fdb_populate(br, &item, fdb, ctx);
296 err = nb->notifier_call(nb, action, &item);
297 return notifier_to_errno(err);
301 br_switchdev_fdb_replay(const struct net_device *br_dev, const void *ctx,
302 bool adding, struct notifier_block *nb)
304 struct net_bridge_fdb_entry *fdb;
305 struct net_bridge *br;
306 unsigned long action;
312 if (!netif_is_bridge_master(br_dev))
315 br = netdev_priv(br_dev);
318 action = SWITCHDEV_FDB_ADD_TO_DEVICE;
320 action = SWITCHDEV_FDB_DEL_TO_DEVICE;
324 hlist_for_each_entry_rcu(fdb, &br->fdb_list, fdb_node) {
325 err = br_switchdev_fdb_replay_one(br, nb, fdb, action, ctx);
335 static int br_switchdev_vlan_attr_replay(struct net_device *br_dev,
337 struct notifier_block *nb,
338 struct netlink_ext_ack *extack)
340 struct switchdev_notifier_port_attr_info attr_info = {
347 struct net_bridge *br = netdev_priv(br_dev);
348 struct net_bridge_vlan_group *vg;
349 struct switchdev_attr attr;
350 struct net_bridge_vlan *v;
353 attr_info.attr = &attr;
354 attr.orig_dev = br_dev;
356 vg = br_vlan_group(br);
360 list_for_each_entry(v, &vg->vlan_list, vlist) {
362 attr.id = SWITCHDEV_ATTR_ID_VLAN_MSTI;
363 attr.u.vlan_msti.vid = v->vid;
364 attr.u.vlan_msti.msti = v->msti;
366 err = nb->notifier_call(nb, SWITCHDEV_PORT_ATTR_SET,
368 err = notifier_to_errno(err);
378 br_switchdev_vlan_replay_one(struct notifier_block *nb,
379 struct net_device *dev,
380 struct switchdev_obj_port_vlan *vlan,
381 const void *ctx, unsigned long action,
382 struct netlink_ext_ack *extack)
384 struct switchdev_notifier_port_obj_info obj_info = {
394 err = nb->notifier_call(nb, action, &obj_info);
395 return notifier_to_errno(err);
398 static int br_switchdev_vlan_replay_group(struct notifier_block *nb,
399 struct net_device *dev,
400 struct net_bridge_vlan_group *vg,
401 const void *ctx, unsigned long action,
402 struct netlink_ext_ack *extack)
404 struct net_bridge_vlan *v;
411 pvid = br_get_pvid(vg);
413 list_for_each_entry(v, &vg->vlan_list, vlist) {
414 struct switchdev_obj_port_vlan vlan = {
416 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
417 .flags = br_vlan_flags(v, pvid),
421 if (!br_vlan_should_use(v))
424 err = br_switchdev_vlan_replay_one(nb, dev, &vlan, ctx,
433 static int br_switchdev_vlan_replay(struct net_device *br_dev,
434 const void *ctx, bool adding,
435 struct notifier_block *nb,
436 struct netlink_ext_ack *extack)
438 struct net_bridge *br = netdev_priv(br_dev);
439 struct net_bridge_port *p;
440 unsigned long action;
448 if (!netif_is_bridge_master(br_dev))
452 action = SWITCHDEV_PORT_OBJ_ADD;
454 action = SWITCHDEV_PORT_OBJ_DEL;
456 err = br_switchdev_vlan_replay_group(nb, br_dev, br_vlan_group(br),
457 ctx, action, extack);
461 list_for_each_entry(p, &br->port_list, list) {
462 struct net_device *dev = p->dev;
464 err = br_switchdev_vlan_replay_group(nb, dev,
466 ctx, action, extack);
472 err = br_switchdev_vlan_attr_replay(br_dev, ctx, nb, extack);
480 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
481 struct br_switchdev_mdb_complete_info {
482 struct net_bridge_port *port;
486 static void br_switchdev_mdb_complete(struct net_device *dev, int err, void *priv)
488 struct br_switchdev_mdb_complete_info *data = priv;
489 struct net_bridge_port_group __rcu **pp;
490 struct net_bridge_port_group *p;
491 struct net_bridge_mdb_entry *mp;
492 struct net_bridge_port *port = data->port;
493 struct net_bridge *br = port->br;
498 spin_lock_bh(&br->multicast_lock);
499 mp = br_mdb_ip_get(br, &data->ip);
502 for (pp = &mp->ports; (p = mlock_dereference(*pp, br)) != NULL;
504 if (p->key.port != port)
506 p->flags |= MDB_PG_FLAGS_OFFLOAD;
509 spin_unlock_bh(&br->multicast_lock);
514 static void br_switchdev_mdb_populate(struct switchdev_obj_port_mdb *mdb,
515 const struct net_bridge_mdb_entry *mp)
517 if (mp->addr.proto == htons(ETH_P_IP))
518 ip_eth_mc_map(mp->addr.dst.ip4, mdb->addr);
519 #if IS_ENABLED(CONFIG_IPV6)
520 else if (mp->addr.proto == htons(ETH_P_IPV6))
521 ipv6_eth_mc_map(&mp->addr.dst.ip6, mdb->addr);
524 ether_addr_copy(mdb->addr, mp->addr.dst.mac_addr);
526 mdb->vid = mp->addr.vid;
529 static void br_switchdev_host_mdb_one(struct net_device *dev,
530 struct net_device *lower_dev,
531 struct net_bridge_mdb_entry *mp,
534 struct switchdev_obj_port_mdb mdb = {
536 .id = SWITCHDEV_OBJ_ID_HOST_MDB,
537 .flags = SWITCHDEV_F_DEFER,
542 br_switchdev_mdb_populate(&mdb, mp);
546 switchdev_port_obj_add(lower_dev, &mdb.obj, NULL);
549 switchdev_port_obj_del(lower_dev, &mdb.obj);
554 static void br_switchdev_host_mdb(struct net_device *dev,
555 struct net_bridge_mdb_entry *mp, int type)
557 struct net_device *lower_dev;
558 struct list_head *iter;
560 netdev_for_each_lower_dev(dev, lower_dev, iter)
561 br_switchdev_host_mdb_one(dev, lower_dev, mp, type);
565 br_switchdev_mdb_replay_one(struct notifier_block *nb, struct net_device *dev,
566 const struct switchdev_obj_port_mdb *mdb,
567 unsigned long action, const void *ctx,
568 struct netlink_ext_ack *extack)
570 struct switchdev_notifier_port_obj_info obj_info = {
580 err = nb->notifier_call(nb, action, &obj_info);
581 return notifier_to_errno(err);
584 static int br_switchdev_mdb_queue_one(struct list_head *mdb_list,
585 enum switchdev_obj_id id,
586 const struct net_bridge_mdb_entry *mp,
587 struct net_device *orig_dev)
589 struct switchdev_obj_port_mdb *mdb;
591 mdb = kzalloc(sizeof(*mdb), GFP_ATOMIC);
596 mdb->obj.orig_dev = orig_dev;
597 br_switchdev_mdb_populate(mdb, mp);
598 list_add_tail(&mdb->obj.list, mdb_list);
603 void br_switchdev_mdb_notify(struct net_device *dev,
604 struct net_bridge_mdb_entry *mp,
605 struct net_bridge_port_group *pg,
608 struct br_switchdev_mdb_complete_info *complete_info;
609 struct switchdev_obj_port_mdb mdb = {
611 .id = SWITCHDEV_OBJ_ID_PORT_MDB,
612 .flags = SWITCHDEV_F_DEFER,
617 return br_switchdev_host_mdb(dev, mp, type);
619 br_switchdev_mdb_populate(&mdb, mp);
621 mdb.obj.orig_dev = pg->key.port->dev;
624 complete_info = kmalloc(sizeof(*complete_info), GFP_ATOMIC);
627 complete_info->port = pg->key.port;
628 complete_info->ip = mp->addr;
629 mdb.obj.complete_priv = complete_info;
630 mdb.obj.complete = br_switchdev_mdb_complete;
631 if (switchdev_port_obj_add(pg->key.port->dev, &mdb.obj, NULL))
632 kfree(complete_info);
635 switchdev_port_obj_del(pg->key.port->dev, &mdb.obj);
642 br_switchdev_mdb_replay(struct net_device *br_dev, struct net_device *dev,
643 const void *ctx, bool adding, struct notifier_block *nb,
644 struct netlink_ext_ack *extack)
646 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
647 const struct net_bridge_mdb_entry *mp;
648 struct switchdev_obj *obj, *tmp;
649 struct net_bridge *br;
650 unsigned long action;
659 if (!netif_is_bridge_master(br_dev) || !netif_is_bridge_port(dev))
662 br = netdev_priv(br_dev);
664 if (!br_opt_get(br, BROPT_MULTICAST_ENABLED))
667 /* We cannot walk over br->mdb_list protected just by the rtnl_mutex,
668 * because the write-side protection is br->multicast_lock. But we
669 * need to emulate the [ blocking ] calling context of a regular
670 * switchdev event, so since both br->multicast_lock and RCU read side
671 * critical sections are atomic, we have no choice but to pick the RCU
672 * read side lock, queue up all our events, leave the critical section
673 * and notify switchdev from blocking context.
677 hlist_for_each_entry_rcu(mp, &br->mdb_list, mdb_node) {
678 struct net_bridge_port_group __rcu * const *pp;
679 const struct net_bridge_port_group *p;
681 if (mp->host_joined) {
682 err = br_switchdev_mdb_queue_one(&mdb_list,
683 SWITCHDEV_OBJ_ID_HOST_MDB,
691 for (pp = &mp->ports; (p = rcu_dereference(*pp)) != NULL;
693 if (p->key.port->dev != dev)
696 err = br_switchdev_mdb_queue_one(&mdb_list,
697 SWITCHDEV_OBJ_ID_PORT_MDB,
709 action = SWITCHDEV_PORT_OBJ_ADD;
711 action = SWITCHDEV_PORT_OBJ_DEL;
713 list_for_each_entry(obj, &mdb_list, list) {
714 err = br_switchdev_mdb_replay_one(nb, dev,
715 SWITCHDEV_OBJ_PORT_MDB(obj),
716 action, ctx, extack);
722 list_for_each_entry_safe(obj, tmp, &mdb_list, list) {
723 list_del(&obj->list);
724 kfree(SWITCHDEV_OBJ_PORT_MDB(obj));
734 static int nbp_switchdev_sync_objs(struct net_bridge_port *p, const void *ctx,
735 struct notifier_block *atomic_nb,
736 struct notifier_block *blocking_nb,
737 struct netlink_ext_ack *extack)
739 struct net_device *br_dev = p->br->dev;
740 struct net_device *dev = p->dev;
743 err = br_switchdev_vlan_replay(br_dev, ctx, true, blocking_nb, extack);
744 if (err && err != -EOPNOTSUPP)
747 err = br_switchdev_mdb_replay(br_dev, dev, ctx, true, blocking_nb,
749 if (err && err != -EOPNOTSUPP)
752 err = br_switchdev_fdb_replay(br_dev, ctx, true, atomic_nb);
753 if (err && err != -EOPNOTSUPP)
759 static void nbp_switchdev_unsync_objs(struct net_bridge_port *p,
761 struct notifier_block *atomic_nb,
762 struct notifier_block *blocking_nb)
764 struct net_device *br_dev = p->br->dev;
765 struct net_device *dev = p->dev;
767 br_switchdev_fdb_replay(br_dev, ctx, false, atomic_nb);
769 br_switchdev_mdb_replay(br_dev, dev, ctx, false, blocking_nb, NULL);
771 br_switchdev_vlan_replay(br_dev, ctx, false, blocking_nb, NULL);
774 /* Let the bridge know that this port is offloaded, so that it can assign a
775 * switchdev hardware domain to it.
777 int br_switchdev_port_offload(struct net_bridge_port *p,
778 struct net_device *dev, const void *ctx,
779 struct notifier_block *atomic_nb,
780 struct notifier_block *blocking_nb,
782 struct netlink_ext_ack *extack)
784 struct netdev_phys_item_id ppid;
787 err = dev_get_port_parent_id(dev, &ppid, false);
791 err = nbp_switchdev_add(p, ppid, tx_fwd_offload, extack);
795 err = nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);
797 goto out_switchdev_del;
802 nbp_switchdev_del(p);
807 void br_switchdev_port_unoffload(struct net_bridge_port *p, const void *ctx,
808 struct notifier_block *atomic_nb,
809 struct notifier_block *blocking_nb)
811 nbp_switchdev_unsync_objs(p, ctx, atomic_nb, blocking_nb);
813 nbp_switchdev_del(p);