xref: /openbmc/linux/net/bridge/br_switchdev.c (revision a83856bd)
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>
7 #include <net/ip.h>
8 #include <net/switchdev.h>
9 
10 #include "br_private.h"
11 
12 static struct static_key_false br_switchdev_tx_fwd_offload;
13 
nbp_switchdev_can_offload_tx_fwd(const struct net_bridge_port * p,const struct sk_buff * skb)14 static bool nbp_switchdev_can_offload_tx_fwd(const struct net_bridge_port *p,
15 					     const struct sk_buff *skb)
16 {
17 	if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
18 		return false;
19 
20 	return (p->flags & BR_TX_FWD_OFFLOAD) &&
21 	       (p->hwdom != BR_INPUT_SKB_CB(skb)->src_hwdom);
22 }
23 
br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff * skb)24 bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb)
25 {
26 	if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
27 		return false;
28 
29 	return BR_INPUT_SKB_CB(skb)->tx_fwd_offload;
30 }
31 
br_switchdev_frame_set_offload_fwd_mark(struct sk_buff * skb)32 void br_switchdev_frame_set_offload_fwd_mark(struct sk_buff *skb)
33 {
34 	skb->offload_fwd_mark = br_switchdev_frame_uses_tx_fwd_offload(skb);
35 }
36 
37 /* Mark the frame for TX forwarding offload if this egress port supports it */
nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port * p,struct sk_buff * skb)38 void nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p,
39 					     struct sk_buff *skb)
40 {
41 	if (nbp_switchdev_can_offload_tx_fwd(p, skb))
42 		BR_INPUT_SKB_CB(skb)->tx_fwd_offload = true;
43 }
44 
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()
48  * return false.
49  */
nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port * p,struct sk_buff * skb)50 void nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p,
51 					      struct sk_buff *skb)
52 {
53 	if (nbp_switchdev_can_offload_tx_fwd(p, skb))
54 		set_bit(p->hwdom, &BR_INPUT_SKB_CB(skb)->fwd_hwdoms);
55 }
56 
nbp_switchdev_frame_mark(const struct net_bridge_port * p,struct sk_buff * skb)57 void nbp_switchdev_frame_mark(const struct net_bridge_port *p,
58 			      struct sk_buff *skb)
59 {
60 	if (p->hwdom)
61 		BR_INPUT_SKB_CB(skb)->src_hwdom = p->hwdom;
62 }
63 
nbp_switchdev_allowed_egress(const struct net_bridge_port * p,const struct sk_buff * skb)64 bool nbp_switchdev_allowed_egress(const struct net_bridge_port *p,
65 				  const struct sk_buff *skb)
66 {
67 	struct br_input_skb_cb *cb = BR_INPUT_SKB_CB(skb);
68 
69 	return !test_bit(p->hwdom, &cb->fwd_hwdoms) &&
70 		(!skb->offload_fwd_mark || cb->src_hwdom != p->hwdom);
71 }
72 
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)
77 
br_switchdev_set_port_flag(struct net_bridge_port * p,unsigned long flags,unsigned long mask,struct netlink_ext_ack * extack)78 int br_switchdev_set_port_flag(struct net_bridge_port *p,
79 			       unsigned long flags,
80 			       unsigned long mask,
81 			       struct netlink_ext_ack *extack)
82 {
83 	struct switchdev_attr attr = {
84 		.orig_dev = p->dev,
85 	};
86 	struct switchdev_notifier_port_attr_info info = {
87 		.attr = &attr,
88 	};
89 	int err;
90 
91 	mask &= BR_PORT_FLAGS_HW_OFFLOAD;
92 	if (!mask)
93 		return 0;
94 
95 	attr.id = SWITCHDEV_ATTR_ID_PORT_PRE_BRIDGE_FLAGS;
96 	attr.u.brport_flags.val = flags;
97 	attr.u.brport_flags.mask = mask;
98 
99 	/* We run from atomic context here */
100 	err = call_switchdev_notifiers(SWITCHDEV_PORT_ATTR_SET, p->dev,
101 				       &info.info, extack);
102 	err = notifier_to_errno(err);
103 	if (err == -EOPNOTSUPP)
104 		return 0;
105 
106 	if (err) {
107 		NL_SET_ERR_MSG_WEAK_MOD(extack,
108 					"bridge flag offload is not supported");
109 		return -EOPNOTSUPP;
110 	}
111 
112 	attr.id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS;
113 	attr.flags = SWITCHDEV_F_DEFER;
114 
115 	err = switchdev_port_attr_set(p->dev, &attr, extack);
116 	if (err) {
117 		NL_SET_ERR_MSG_WEAK_MOD(extack,
118 					"error setting offload flag on port");
119 		return err;
120 	}
121 
122 	return 0;
123 }
124 
br_switchdev_fdb_populate(struct net_bridge * br,struct switchdev_notifier_fdb_info * item,const struct net_bridge_fdb_entry * fdb,const void * ctx)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,
128 				      const void *ctx)
129 {
130 	const struct net_bridge_port *p = READ_ONCE(fdb->dst);
131 
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;
140 }
141 
142 void
br_switchdev_fdb_notify(struct net_bridge * br,const struct net_bridge_fdb_entry * fdb,int type)143 br_switchdev_fdb_notify(struct net_bridge *br,
144 			const struct net_bridge_fdb_entry *fdb, int type)
145 {
146 	struct switchdev_notifier_fdb_info item;
147 
148 	if (test_bit(BR_FDB_LOCKED, &fdb->flags))
149 		return;
150 
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.
156 	 */
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))
160 		return;
161 
162 	br_switchdev_fdb_populate(br, &item, fdb, NULL);
163 
164 	switch (type) {
165 	case RTM_DELNEIGH:
166 		call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_DEVICE,
167 					 item.info.dev, &item.info, NULL);
168 		break;
169 	case RTM_NEWNEIGH:
170 		call_switchdev_notifiers(SWITCHDEV_FDB_ADD_TO_DEVICE,
171 					 item.info.dev, &item.info, NULL);
172 		break;
173 	}
174 }
175 
br_switchdev_port_vlan_add(struct net_device * dev,u16 vid,u16 flags,bool changed,struct netlink_ext_ack * extack)176 int br_switchdev_port_vlan_add(struct net_device *dev, u16 vid, u16 flags,
177 			       bool changed, struct netlink_ext_ack *extack)
178 {
179 	struct switchdev_obj_port_vlan v = {
180 		.obj.orig_dev = dev,
181 		.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
182 		.flags = flags,
183 		.vid = vid,
184 		.changed = changed,
185 	};
186 
187 	return switchdev_port_obj_add(dev, &v.obj, extack);
188 }
189 
br_switchdev_port_vlan_del(struct net_device * dev,u16 vid)190 int br_switchdev_port_vlan_del(struct net_device *dev, u16 vid)
191 {
192 	struct switchdev_obj_port_vlan v = {
193 		.obj.orig_dev = dev,
194 		.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
195 		.vid = vid,
196 	};
197 
198 	return switchdev_port_obj_del(dev, &v.obj);
199 }
200 
nbp_switchdev_hwdom_set(struct net_bridge_port * joining)201 static int nbp_switchdev_hwdom_set(struct net_bridge_port *joining)
202 {
203 	struct net_bridge *br = joining->br;
204 	struct net_bridge_port *p;
205 	int hwdom;
206 
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;
211 			return 0;
212 		}
213 	}
214 
215 	hwdom = find_next_zero_bit(&br->busy_hwdoms, BR_HWDOM_MAX, 1);
216 	if (hwdom >= BR_HWDOM_MAX)
217 		return -EBUSY;
218 
219 	set_bit(hwdom, &br->busy_hwdoms);
220 	joining->hwdom = hwdom;
221 	return 0;
222 }
223 
nbp_switchdev_hwdom_put(struct net_bridge_port * leaving)224 static void nbp_switchdev_hwdom_put(struct net_bridge_port *leaving)
225 {
226 	struct net_bridge *br = leaving->br;
227 	struct net_bridge_port *p;
228 
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)
232 			return;
233 	}
234 
235 	clear_bit(leaving->hwdom, &br->busy_hwdoms);
236 }
237 
nbp_switchdev_add(struct net_bridge_port * p,struct netdev_phys_item_id ppid,bool tx_fwd_offload,struct netlink_ext_ack * extack)238 static int nbp_switchdev_add(struct net_bridge_port *p,
239 			     struct netdev_phys_item_id ppid,
240 			     bool tx_fwd_offload,
241 			     struct netlink_ext_ack *extack)
242 {
243 	int err;
244 
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.
249 		 */
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");
253 			return -EBUSY;
254 		}
255 
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.
259 		 */
260 		p->offload_count++;
261 
262 		return 0;
263 	}
264 
265 	p->ppid = ppid;
266 	p->offload_count = 1;
267 
268 	err = nbp_switchdev_hwdom_set(p);
269 	if (err)
270 		return err;
271 
272 	if (tx_fwd_offload) {
273 		p->flags |= BR_TX_FWD_OFFLOAD;
274 		static_branch_inc(&br_switchdev_tx_fwd_offload);
275 	}
276 
277 	return 0;
278 }
279 
nbp_switchdev_del(struct net_bridge_port * p)280 static void nbp_switchdev_del(struct net_bridge_port *p)
281 {
282 	if (WARN_ON(!p->offload_count))
283 		return;
284 
285 	p->offload_count--;
286 
287 	if (p->offload_count)
288 		return;
289 
290 	if (p->hwdom)
291 		nbp_switchdev_hwdom_put(p);
292 
293 	if (p->flags & BR_TX_FWD_OFFLOAD) {
294 		p->flags &= ~BR_TX_FWD_OFFLOAD;
295 		static_branch_dec(&br_switchdev_tx_fwd_offload);
296 	}
297 }
298 
299 static int
br_switchdev_fdb_replay_one(struct net_bridge * br,struct notifier_block * nb,const struct net_bridge_fdb_entry * fdb,unsigned long action,const void * ctx)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)
303 {
304 	struct switchdev_notifier_fdb_info item;
305 	int err;
306 
307 	br_switchdev_fdb_populate(br, &item, fdb, ctx);
308 
309 	err = nb->notifier_call(nb, action, &item);
310 	return notifier_to_errno(err);
311 }
312 
313 static int
br_switchdev_fdb_replay(const struct net_device * br_dev,const void * ctx,bool adding,struct notifier_block * nb)314 br_switchdev_fdb_replay(const struct net_device *br_dev, const void *ctx,
315 			bool adding, struct notifier_block *nb)
316 {
317 	struct net_bridge_fdb_entry *fdb;
318 	struct net_bridge *br;
319 	unsigned long action;
320 	int err = 0;
321 
322 	if (!nb)
323 		return 0;
324 
325 	if (!netif_is_bridge_master(br_dev))
326 		return -EINVAL;
327 
328 	br = netdev_priv(br_dev);
329 
330 	if (adding)
331 		action = SWITCHDEV_FDB_ADD_TO_DEVICE;
332 	else
333 		action = SWITCHDEV_FDB_DEL_TO_DEVICE;
334 
335 	rcu_read_lock();
336 
337 	hlist_for_each_entry_rcu(fdb, &br->fdb_list, fdb_node) {
338 		err = br_switchdev_fdb_replay_one(br, nb, fdb, action, ctx);
339 		if (err)
340 			break;
341 	}
342 
343 	rcu_read_unlock();
344 
345 	return err;
346 }
347 
br_switchdev_vlan_attr_replay(struct net_device * br_dev,const void * ctx,struct notifier_block * nb,struct netlink_ext_ack * extack)348 static int br_switchdev_vlan_attr_replay(struct net_device *br_dev,
349 					 const void *ctx,
350 					 struct notifier_block *nb,
351 					 struct netlink_ext_ack *extack)
352 {
353 	struct switchdev_notifier_port_attr_info attr_info = {
354 		.info = {
355 			.dev = br_dev,
356 			.extack = extack,
357 			.ctx = ctx,
358 		},
359 	};
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;
364 	int err;
365 
366 	attr_info.attr = &attr;
367 	attr.orig_dev = br_dev;
368 
369 	vg = br_vlan_group(br);
370 	if (!vg)
371 		return 0;
372 
373 	list_for_each_entry(v, &vg->vlan_list, vlist) {
374 		if (v->msti) {
375 			attr.id = SWITCHDEV_ATTR_ID_VLAN_MSTI;
376 			attr.u.vlan_msti.vid = v->vid;
377 			attr.u.vlan_msti.msti = v->msti;
378 
379 			err = nb->notifier_call(nb, SWITCHDEV_PORT_ATTR_SET,
380 						&attr_info);
381 			err = notifier_to_errno(err);
382 			if (err)
383 				return err;
384 		}
385 	}
386 
387 	return 0;
388 }
389 
390 static int
br_switchdev_vlan_replay_one(struct notifier_block * nb,struct net_device * dev,struct switchdev_obj_port_vlan * vlan,const void * ctx,unsigned long action,struct netlink_ext_ack * extack)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)
396 {
397 	struct switchdev_notifier_port_obj_info obj_info = {
398 		.info = {
399 			.dev = dev,
400 			.extack = extack,
401 			.ctx = ctx,
402 		},
403 		.obj = &vlan->obj,
404 	};
405 	int err;
406 
407 	err = nb->notifier_call(nb, action, &obj_info);
408 	return notifier_to_errno(err);
409 }
410 
br_switchdev_vlan_replay_group(struct notifier_block * nb,struct net_device * dev,struct net_bridge_vlan_group * vg,const void * ctx,unsigned long action,struct netlink_ext_ack * extack)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)
416 {
417 	struct net_bridge_vlan *v;
418 	int err = 0;
419 	u16 pvid;
420 
421 	if (!vg)
422 		return 0;
423 
424 	pvid = br_get_pvid(vg);
425 
426 	list_for_each_entry(v, &vg->vlan_list, vlist) {
427 		struct switchdev_obj_port_vlan vlan = {
428 			.obj.orig_dev = dev,
429 			.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
430 			.flags = br_vlan_flags(v, pvid),
431 			.vid = v->vid,
432 		};
433 
434 		if (!br_vlan_should_use(v))
435 			continue;
436 
437 		err = br_switchdev_vlan_replay_one(nb, dev, &vlan, ctx,
438 						   action, extack);
439 		if (err)
440 			return err;
441 	}
442 
443 	return 0;
444 }
445 
br_switchdev_vlan_replay(struct net_device * br_dev,const void * ctx,bool adding,struct notifier_block * nb,struct netlink_ext_ack * extack)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)
450 {
451 	struct net_bridge *br = netdev_priv(br_dev);
452 	struct net_bridge_port *p;
453 	unsigned long action;
454 	int err;
455 
456 	ASSERT_RTNL();
457 
458 	if (!nb)
459 		return 0;
460 
461 	if (!netif_is_bridge_master(br_dev))
462 		return -EINVAL;
463 
464 	if (adding)
465 		action = SWITCHDEV_PORT_OBJ_ADD;
466 	else
467 		action = SWITCHDEV_PORT_OBJ_DEL;
468 
469 	err = br_switchdev_vlan_replay_group(nb, br_dev, br_vlan_group(br),
470 					     ctx, action, extack);
471 	if (err)
472 		return err;
473 
474 	list_for_each_entry(p, &br->port_list, list) {
475 		struct net_device *dev = p->dev;
476 
477 		err = br_switchdev_vlan_replay_group(nb, dev,
478 						     nbp_vlan_group(p),
479 						     ctx, action, extack);
480 		if (err)
481 			return err;
482 	}
483 
484 	if (adding) {
485 		err = br_switchdev_vlan_attr_replay(br_dev, ctx, nb, extack);
486 		if (err)
487 			return err;
488 	}
489 
490 	return 0;
491 }
492 
493 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
494 struct br_switchdev_mdb_complete_info {
495 	struct net_bridge_port *port;
496 	struct br_ip ip;
497 };
498 
br_switchdev_mdb_complete(struct net_device * dev,int err,void * priv)499 static void br_switchdev_mdb_complete(struct net_device *dev, int err, void *priv)
500 {
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;
507 
508 	if (err)
509 		goto err;
510 
511 	spin_lock_bh(&br->multicast_lock);
512 	mp = br_mdb_ip_get(br, &data->ip);
513 	if (!mp)
514 		goto out;
515 	for (pp = &mp->ports; (p = mlock_dereference(*pp, br)) != NULL;
516 	     pp = &p->next) {
517 		if (p->key.port != port)
518 			continue;
519 		p->flags |= MDB_PG_FLAGS_OFFLOAD;
520 	}
521 out:
522 	spin_unlock_bh(&br->multicast_lock);
523 err:
524 	kfree(priv);
525 }
526 
br_switchdev_mdb_populate(struct switchdev_obj_port_mdb * mdb,const struct net_bridge_mdb_entry * mp)527 static void br_switchdev_mdb_populate(struct switchdev_obj_port_mdb *mdb,
528 				      const struct net_bridge_mdb_entry *mp)
529 {
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);
535 #endif
536 	else
537 		ether_addr_copy(mdb->addr, mp->addr.dst.mac_addr);
538 
539 	mdb->vid = mp->addr.vid;
540 }
541 
br_switchdev_host_mdb_one(struct net_device * dev,struct net_device * lower_dev,struct net_bridge_mdb_entry * mp,int type)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,
545 				      int type)
546 {
547 	struct switchdev_obj_port_mdb mdb = {
548 		.obj = {
549 			.id = SWITCHDEV_OBJ_ID_HOST_MDB,
550 			.flags = SWITCHDEV_F_DEFER,
551 			.orig_dev = dev,
552 		},
553 	};
554 
555 	br_switchdev_mdb_populate(&mdb, mp);
556 
557 	switch (type) {
558 	case RTM_NEWMDB:
559 		switchdev_port_obj_add(lower_dev, &mdb.obj, NULL);
560 		break;
561 	case RTM_DELMDB:
562 		switchdev_port_obj_del(lower_dev, &mdb.obj);
563 		break;
564 	}
565 }
566 
br_switchdev_host_mdb(struct net_device * dev,struct net_bridge_mdb_entry * mp,int type)567 static void br_switchdev_host_mdb(struct net_device *dev,
568 				  struct net_bridge_mdb_entry *mp, int type)
569 {
570 	struct net_device *lower_dev;
571 	struct list_head *iter;
572 
573 	netdev_for_each_lower_dev(dev, lower_dev, iter)
574 		br_switchdev_host_mdb_one(dev, lower_dev, mp, type);
575 }
576 
577 static int
br_switchdev_mdb_replay_one(struct notifier_block * nb,struct net_device * dev,const struct switchdev_obj_port_mdb * mdb,unsigned long action,const void * ctx,struct netlink_ext_ack * extack)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)
582 {
583 	struct switchdev_notifier_port_obj_info obj_info = {
584 		.info = {
585 			.dev = dev,
586 			.extack = extack,
587 			.ctx = ctx,
588 		},
589 		.obj = &mdb->obj,
590 	};
591 	int err;
592 
593 	err = nb->notifier_call(nb, action, &obj_info);
594 	return notifier_to_errno(err);
595 }
596 
br_switchdev_mdb_queue_one(struct list_head * mdb_list,struct net_device * dev,unsigned long action,enum switchdev_obj_id id,const struct net_bridge_mdb_entry * mp,struct net_device * orig_dev)597 static int br_switchdev_mdb_queue_one(struct list_head *mdb_list,
598 				      struct net_device *dev,
599 				      unsigned long action,
600 				      enum switchdev_obj_id id,
601 				      const struct net_bridge_mdb_entry *mp,
602 				      struct net_device *orig_dev)
603 {
604 	struct switchdev_obj_port_mdb mdb = {
605 		.obj = {
606 			.id = id,
607 			.orig_dev = orig_dev,
608 		},
609 	};
610 	struct switchdev_obj_port_mdb *pmdb;
611 
612 	br_switchdev_mdb_populate(&mdb, mp);
613 
614 	if (action == SWITCHDEV_PORT_OBJ_ADD &&
615 	    switchdev_port_obj_act_is_deferred(dev, action, &mdb.obj)) {
616 		/* This event is already in the deferred queue of
617 		 * events, so this replay must be elided, lest the
618 		 * driver receives duplicate events for it. This can
619 		 * only happen when replaying additions, since
620 		 * modifications are always immediately visible in
621 		 * br->mdb_list, whereas actual event delivery may be
622 		 * delayed.
623 		 */
624 		return 0;
625 	}
626 
627 	pmdb = kmemdup(&mdb, sizeof(mdb), GFP_ATOMIC);
628 	if (!pmdb)
629 		return -ENOMEM;
630 
631 	list_add_tail(&pmdb->obj.list, mdb_list);
632 	return 0;
633 }
634 
br_switchdev_mdb_notify(struct net_device * dev,struct net_bridge_mdb_entry * mp,struct net_bridge_port_group * pg,int type)635 void br_switchdev_mdb_notify(struct net_device *dev,
636 			     struct net_bridge_mdb_entry *mp,
637 			     struct net_bridge_port_group *pg,
638 			     int type)
639 {
640 	struct br_switchdev_mdb_complete_info *complete_info;
641 	struct switchdev_obj_port_mdb mdb = {
642 		.obj = {
643 			.id = SWITCHDEV_OBJ_ID_PORT_MDB,
644 			.flags = SWITCHDEV_F_DEFER,
645 		},
646 	};
647 
648 	if (!pg)
649 		return br_switchdev_host_mdb(dev, mp, type);
650 
651 	br_switchdev_mdb_populate(&mdb, mp);
652 
653 	mdb.obj.orig_dev = pg->key.port->dev;
654 	switch (type) {
655 	case RTM_NEWMDB:
656 		complete_info = kmalloc(sizeof(*complete_info), GFP_ATOMIC);
657 		if (!complete_info)
658 			break;
659 		complete_info->port = pg->key.port;
660 		complete_info->ip = mp->addr;
661 		mdb.obj.complete_priv = complete_info;
662 		mdb.obj.complete = br_switchdev_mdb_complete;
663 		if (switchdev_port_obj_add(pg->key.port->dev, &mdb.obj, NULL))
664 			kfree(complete_info);
665 		break;
666 	case RTM_DELMDB:
667 		switchdev_port_obj_del(pg->key.port->dev, &mdb.obj);
668 		break;
669 	}
670 }
671 #endif
672 
673 static int
br_switchdev_mdb_replay(struct net_device * br_dev,struct net_device * dev,const void * ctx,bool adding,struct notifier_block * nb,struct netlink_ext_ack * extack)674 br_switchdev_mdb_replay(struct net_device *br_dev, struct net_device *dev,
675 			const void *ctx, bool adding, struct notifier_block *nb,
676 			struct netlink_ext_ack *extack)
677 {
678 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
679 	const struct net_bridge_mdb_entry *mp;
680 	struct switchdev_obj *obj, *tmp;
681 	struct net_bridge *br;
682 	unsigned long action;
683 	LIST_HEAD(mdb_list);
684 	int err = 0;
685 
686 	ASSERT_RTNL();
687 
688 	if (!nb)
689 		return 0;
690 
691 	if (!netif_is_bridge_master(br_dev) || !netif_is_bridge_port(dev))
692 		return -EINVAL;
693 
694 	br = netdev_priv(br_dev);
695 
696 	if (!br_opt_get(br, BROPT_MULTICAST_ENABLED))
697 		return 0;
698 
699 	if (adding)
700 		action = SWITCHDEV_PORT_OBJ_ADD;
701 	else
702 		action = SWITCHDEV_PORT_OBJ_DEL;
703 
704 	/* br_switchdev_mdb_queue_one() will take care to not queue a
705 	 * replay of an event that is already pending in the switchdev
706 	 * deferred queue. In order to safely determine that, there
707 	 * must be no new deferred MDB notifications enqueued for the
708 	 * duration of the MDB scan. Therefore, grab the write-side
709 	 * lock to avoid racing with any concurrent IGMP/MLD snooping.
710 	 */
711 	spin_lock_bh(&br->multicast_lock);
712 
713 	hlist_for_each_entry(mp, &br->mdb_list, mdb_node) {
714 		struct net_bridge_port_group __rcu * const *pp;
715 		const struct net_bridge_port_group *p;
716 
717 		if (mp->host_joined) {
718 			err = br_switchdev_mdb_queue_one(&mdb_list, dev, action,
719 							 SWITCHDEV_OBJ_ID_HOST_MDB,
720 							 mp, br_dev);
721 			if (err) {
722 				spin_unlock_bh(&br->multicast_lock);
723 				goto out_free_mdb;
724 			}
725 		}
726 
727 		for (pp = &mp->ports; (p = mlock_dereference(*pp, br)) != NULL;
728 		     pp = &p->next) {
729 			if (p->key.port->dev != dev)
730 				continue;
731 
732 			err = br_switchdev_mdb_queue_one(&mdb_list, dev, action,
733 							 SWITCHDEV_OBJ_ID_PORT_MDB,
734 							 mp, dev);
735 			if (err) {
736 				spin_unlock_bh(&br->multicast_lock);
737 				goto out_free_mdb;
738 			}
739 		}
740 	}
741 
742 	spin_unlock_bh(&br->multicast_lock);
743 
744 	list_for_each_entry(obj, &mdb_list, list) {
745 		err = br_switchdev_mdb_replay_one(nb, dev,
746 						  SWITCHDEV_OBJ_PORT_MDB(obj),
747 						  action, ctx, extack);
748 		if (err == -EOPNOTSUPP)
749 			err = 0;
750 		if (err)
751 			goto out_free_mdb;
752 	}
753 
754 out_free_mdb:
755 	list_for_each_entry_safe(obj, tmp, &mdb_list, list) {
756 		list_del(&obj->list);
757 		kfree(SWITCHDEV_OBJ_PORT_MDB(obj));
758 	}
759 
760 	if (err)
761 		return err;
762 #endif
763 
764 	return 0;
765 }
766 
nbp_switchdev_sync_objs(struct net_bridge_port * p,const void * ctx,struct notifier_block * atomic_nb,struct notifier_block * blocking_nb,struct netlink_ext_ack * extack)767 static int nbp_switchdev_sync_objs(struct net_bridge_port *p, const void *ctx,
768 				   struct notifier_block *atomic_nb,
769 				   struct notifier_block *blocking_nb,
770 				   struct netlink_ext_ack *extack)
771 {
772 	struct net_device *br_dev = p->br->dev;
773 	struct net_device *dev = p->dev;
774 	int err;
775 
776 	err = br_switchdev_vlan_replay(br_dev, ctx, true, blocking_nb, extack);
777 	if (err && err != -EOPNOTSUPP)
778 		return err;
779 
780 	err = br_switchdev_mdb_replay(br_dev, dev, ctx, true, blocking_nb,
781 				      extack);
782 	if (err) {
783 		/* -EOPNOTSUPP not propagated from MDB replay. */
784 		return err;
785 	}
786 
787 	err = br_switchdev_fdb_replay(br_dev, ctx, true, atomic_nb);
788 	if (err && err != -EOPNOTSUPP)
789 		return err;
790 
791 	return 0;
792 }
793 
nbp_switchdev_unsync_objs(struct net_bridge_port * p,const void * ctx,struct notifier_block * atomic_nb,struct notifier_block * blocking_nb)794 static void nbp_switchdev_unsync_objs(struct net_bridge_port *p,
795 				      const void *ctx,
796 				      struct notifier_block *atomic_nb,
797 				      struct notifier_block *blocking_nb)
798 {
799 	struct net_device *br_dev = p->br->dev;
800 	struct net_device *dev = p->dev;
801 
802 	br_switchdev_fdb_replay(br_dev, ctx, false, atomic_nb);
803 
804 	br_switchdev_mdb_replay(br_dev, dev, ctx, false, blocking_nb, NULL);
805 
806 	br_switchdev_vlan_replay(br_dev, ctx, false, blocking_nb, NULL);
807 
808 	/* Make sure that the device leaving this bridge has seen all
809 	 * relevant events before it is disassociated. In the normal
810 	 * case, when the device is directly attached to the bridge,
811 	 * this is covered by del_nbp(). If the association was indirect
812 	 * however, e.g. via a team or bond, and the device is leaving
813 	 * that intermediate device, then the bridge port remains in
814 	 * place.
815 	 */
816 	switchdev_deferred_process();
817 }
818 
819 /* Let the bridge know that this port is offloaded, so that it can assign a
820  * switchdev hardware domain to it.
821  */
br_switchdev_port_offload(struct net_bridge_port * p,struct net_device * dev,const void * ctx,struct notifier_block * atomic_nb,struct notifier_block * blocking_nb,bool tx_fwd_offload,struct netlink_ext_ack * extack)822 int br_switchdev_port_offload(struct net_bridge_port *p,
823 			      struct net_device *dev, const void *ctx,
824 			      struct notifier_block *atomic_nb,
825 			      struct notifier_block *blocking_nb,
826 			      bool tx_fwd_offload,
827 			      struct netlink_ext_ack *extack)
828 {
829 	struct netdev_phys_item_id ppid;
830 	int err;
831 
832 	err = dev_get_port_parent_id(dev, &ppid, false);
833 	if (err)
834 		return err;
835 
836 	err = nbp_switchdev_add(p, ppid, tx_fwd_offload, extack);
837 	if (err)
838 		return err;
839 
840 	err = nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);
841 	if (err)
842 		goto out_switchdev_del;
843 
844 	return 0;
845 
846 out_switchdev_del:
847 	nbp_switchdev_del(p);
848 
849 	return err;
850 }
851 
br_switchdev_port_unoffload(struct net_bridge_port * p,const void * ctx,struct notifier_block * atomic_nb,struct notifier_block * blocking_nb)852 void br_switchdev_port_unoffload(struct net_bridge_port *p, const void *ctx,
853 				 struct notifier_block *atomic_nb,
854 				 struct notifier_block *blocking_nb)
855 {
856 	nbp_switchdev_unsync_objs(p, ctx, atomic_nb, blocking_nb);
857 
858 	nbp_switchdev_del(p);
859 }
860 
br_switchdev_port_replay(struct net_bridge_port * p,struct net_device * dev,const void * ctx,struct notifier_block * atomic_nb,struct notifier_block * blocking_nb,struct netlink_ext_ack * extack)861 int br_switchdev_port_replay(struct net_bridge_port *p,
862 			     struct net_device *dev, const void *ctx,
863 			     struct notifier_block *atomic_nb,
864 			     struct notifier_block *blocking_nb,
865 			     struct netlink_ext_ack *extack)
866 {
867 	return nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);
868 }
869