xref: /openbmc/linux/net/switchdev/switchdev.c (revision adb19164)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/switchdev/switchdev.c - Switch device API
4  * Copyright (c) 2014-2015 Jiri Pirko <jiri@resnulli.us>
5  * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/init.h>
11 #include <linux/mutex.h>
12 #include <linux/notifier.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/if_bridge.h>
16 #include <linux/list.h>
17 #include <linux/workqueue.h>
18 #include <linux/if_vlan.h>
19 #include <linux/rtnetlink.h>
20 #include <net/switchdev.h>
21 
22 static bool switchdev_obj_eq(const struct switchdev_obj *a,
23 			     const struct switchdev_obj *b)
24 {
25 	const struct switchdev_obj_port_vlan *va, *vb;
26 	const struct switchdev_obj_port_mdb *ma, *mb;
27 
28 	if (a->id != b->id || a->orig_dev != b->orig_dev)
29 		return false;
30 
31 	switch (a->id) {
32 	case SWITCHDEV_OBJ_ID_PORT_VLAN:
33 		va = SWITCHDEV_OBJ_PORT_VLAN(a);
34 		vb = SWITCHDEV_OBJ_PORT_VLAN(b);
35 		return va->flags == vb->flags &&
36 			va->vid == vb->vid &&
37 			va->changed == vb->changed;
38 	case SWITCHDEV_OBJ_ID_PORT_MDB:
39 	case SWITCHDEV_OBJ_ID_HOST_MDB:
40 		ma = SWITCHDEV_OBJ_PORT_MDB(a);
41 		mb = SWITCHDEV_OBJ_PORT_MDB(b);
42 		return ma->vid == mb->vid &&
43 			ether_addr_equal(ma->addr, mb->addr);
44 	default:
45 		break;
46 	}
47 
48 	BUG();
49 }
50 
51 static LIST_HEAD(deferred);
52 static DEFINE_SPINLOCK(deferred_lock);
53 
54 typedef void switchdev_deferred_func_t(struct net_device *dev,
55 				       const void *data);
56 
57 struct switchdev_deferred_item {
58 	struct list_head list;
59 	struct net_device *dev;
60 	netdevice_tracker dev_tracker;
61 	switchdev_deferred_func_t *func;
62 	unsigned long data[];
63 };
64 
65 static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
66 {
67 	struct switchdev_deferred_item *dfitem;
68 
69 	spin_lock_bh(&deferred_lock);
70 	if (list_empty(&deferred)) {
71 		dfitem = NULL;
72 		goto unlock;
73 	}
74 	dfitem = list_first_entry(&deferred,
75 				  struct switchdev_deferred_item, list);
76 	list_del(&dfitem->list);
77 unlock:
78 	spin_unlock_bh(&deferred_lock);
79 	return dfitem;
80 }
81 
82 /**
83  *	switchdev_deferred_process - Process ops in deferred queue
84  *
85  *	Called to flush the ops currently queued in deferred ops queue.
86  *	rtnl_lock must be held.
87  */
88 void switchdev_deferred_process(void)
89 {
90 	struct switchdev_deferred_item *dfitem;
91 
92 	ASSERT_RTNL();
93 
94 	while ((dfitem = switchdev_deferred_dequeue())) {
95 		dfitem->func(dfitem->dev, dfitem->data);
96 		netdev_put(dfitem->dev, &dfitem->dev_tracker);
97 		kfree(dfitem);
98 	}
99 }
100 EXPORT_SYMBOL_GPL(switchdev_deferred_process);
101 
102 static void switchdev_deferred_process_work(struct work_struct *work)
103 {
104 	rtnl_lock();
105 	switchdev_deferred_process();
106 	rtnl_unlock();
107 }
108 
109 static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
110 
111 static int switchdev_deferred_enqueue(struct net_device *dev,
112 				      const void *data, size_t data_len,
113 				      switchdev_deferred_func_t *func)
114 {
115 	struct switchdev_deferred_item *dfitem;
116 
117 	dfitem = kmalloc(struct_size(dfitem, data, data_len), GFP_ATOMIC);
118 	if (!dfitem)
119 		return -ENOMEM;
120 	dfitem->dev = dev;
121 	dfitem->func = func;
122 	memcpy(dfitem->data, data, data_len);
123 	netdev_hold(dev, &dfitem->dev_tracker, GFP_ATOMIC);
124 	spin_lock_bh(&deferred_lock);
125 	list_add_tail(&dfitem->list, &deferred);
126 	spin_unlock_bh(&deferred_lock);
127 	schedule_work(&deferred_process_work);
128 	return 0;
129 }
130 
131 static int switchdev_port_attr_notify(enum switchdev_notifier_type nt,
132 				      struct net_device *dev,
133 				      const struct switchdev_attr *attr,
134 				      struct netlink_ext_ack *extack)
135 {
136 	int err;
137 	int rc;
138 
139 	struct switchdev_notifier_port_attr_info attr_info = {
140 		.attr = attr,
141 		.handled = false,
142 	};
143 
144 	rc = call_switchdev_blocking_notifiers(nt, dev,
145 					       &attr_info.info, extack);
146 	err = notifier_to_errno(rc);
147 	if (err) {
148 		WARN_ON(!attr_info.handled);
149 		return err;
150 	}
151 
152 	if (!attr_info.handled)
153 		return -EOPNOTSUPP;
154 
155 	return 0;
156 }
157 
158 static int switchdev_port_attr_set_now(struct net_device *dev,
159 				       const struct switchdev_attr *attr,
160 				       struct netlink_ext_ack *extack)
161 {
162 	return switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
163 					  extack);
164 }
165 
166 static void switchdev_port_attr_set_deferred(struct net_device *dev,
167 					     const void *data)
168 {
169 	const struct switchdev_attr *attr = data;
170 	int err;
171 
172 	err = switchdev_port_attr_set_now(dev, attr, NULL);
173 	if (err && err != -EOPNOTSUPP)
174 		netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
175 			   err, attr->id);
176 	if (attr->complete)
177 		attr->complete(dev, err, attr->complete_priv);
178 }
179 
180 static int switchdev_port_attr_set_defer(struct net_device *dev,
181 					 const struct switchdev_attr *attr)
182 {
183 	return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
184 					  switchdev_port_attr_set_deferred);
185 }
186 
187 /**
188  *	switchdev_port_attr_set - Set port attribute
189  *
190  *	@dev: port device
191  *	@attr: attribute to set
192  *	@extack: netlink extended ack, for error message propagation
193  *
194  *	rtnl_lock must be held and must not be in atomic section,
195  *	in case SWITCHDEV_F_DEFER flag is not set.
196  */
197 int switchdev_port_attr_set(struct net_device *dev,
198 			    const struct switchdev_attr *attr,
199 			    struct netlink_ext_ack *extack)
200 {
201 	if (attr->flags & SWITCHDEV_F_DEFER)
202 		return switchdev_port_attr_set_defer(dev, attr);
203 	ASSERT_RTNL();
204 	return switchdev_port_attr_set_now(dev, attr, extack);
205 }
206 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
207 
208 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
209 {
210 	switch (obj->id) {
211 	case SWITCHDEV_OBJ_ID_PORT_VLAN:
212 		return sizeof(struct switchdev_obj_port_vlan);
213 	case SWITCHDEV_OBJ_ID_PORT_MDB:
214 		return sizeof(struct switchdev_obj_port_mdb);
215 	case SWITCHDEV_OBJ_ID_HOST_MDB:
216 		return sizeof(struct switchdev_obj_port_mdb);
217 	default:
218 		BUG();
219 	}
220 	return 0;
221 }
222 
223 static int switchdev_port_obj_notify(enum switchdev_notifier_type nt,
224 				     struct net_device *dev,
225 				     const struct switchdev_obj *obj,
226 				     struct netlink_ext_ack *extack)
227 {
228 	int rc;
229 	int err;
230 
231 	struct switchdev_notifier_port_obj_info obj_info = {
232 		.obj = obj,
233 		.handled = false,
234 	};
235 
236 	rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack);
237 	err = notifier_to_errno(rc);
238 	if (err) {
239 		WARN_ON(!obj_info.handled);
240 		return err;
241 	}
242 	if (!obj_info.handled)
243 		return -EOPNOTSUPP;
244 	return 0;
245 }
246 
247 static void switchdev_port_obj_add_deferred(struct net_device *dev,
248 					    const void *data)
249 {
250 	const struct switchdev_obj *obj = data;
251 	int err;
252 
253 	ASSERT_RTNL();
254 	err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
255 					dev, obj, NULL);
256 	if (err && err != -EOPNOTSUPP)
257 		netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
258 			   err, obj->id);
259 	if (obj->complete)
260 		obj->complete(dev, err, obj->complete_priv);
261 }
262 
263 static int switchdev_port_obj_add_defer(struct net_device *dev,
264 					const struct switchdev_obj *obj)
265 {
266 	return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
267 					  switchdev_port_obj_add_deferred);
268 }
269 
270 /**
271  *	switchdev_port_obj_add - Add port object
272  *
273  *	@dev: port device
274  *	@obj: object to add
275  *	@extack: netlink extended ack
276  *
277  *	rtnl_lock must be held and must not be in atomic section,
278  *	in case SWITCHDEV_F_DEFER flag is not set.
279  */
280 int switchdev_port_obj_add(struct net_device *dev,
281 			   const struct switchdev_obj *obj,
282 			   struct netlink_ext_ack *extack)
283 {
284 	if (obj->flags & SWITCHDEV_F_DEFER)
285 		return switchdev_port_obj_add_defer(dev, obj);
286 	ASSERT_RTNL();
287 	return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
288 					 dev, obj, extack);
289 }
290 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
291 
292 static int switchdev_port_obj_del_now(struct net_device *dev,
293 				      const struct switchdev_obj *obj)
294 {
295 	return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL,
296 					 dev, obj, NULL);
297 }
298 
299 static void switchdev_port_obj_del_deferred(struct net_device *dev,
300 					    const void *data)
301 {
302 	const struct switchdev_obj *obj = data;
303 	int err;
304 
305 	err = switchdev_port_obj_del_now(dev, obj);
306 	if (err && err != -EOPNOTSUPP)
307 		netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
308 			   err, obj->id);
309 	if (obj->complete)
310 		obj->complete(dev, err, obj->complete_priv);
311 }
312 
313 static int switchdev_port_obj_del_defer(struct net_device *dev,
314 					const struct switchdev_obj *obj)
315 {
316 	return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
317 					  switchdev_port_obj_del_deferred);
318 }
319 
320 /**
321  *	switchdev_port_obj_del - Delete port object
322  *
323  *	@dev: port device
324  *	@obj: object to delete
325  *
326  *	rtnl_lock must be held and must not be in atomic section,
327  *	in case SWITCHDEV_F_DEFER flag is not set.
328  */
329 int switchdev_port_obj_del(struct net_device *dev,
330 			   const struct switchdev_obj *obj)
331 {
332 	if (obj->flags & SWITCHDEV_F_DEFER)
333 		return switchdev_port_obj_del_defer(dev, obj);
334 	ASSERT_RTNL();
335 	return switchdev_port_obj_del_now(dev, obj);
336 }
337 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
338 
339 /**
340  *	switchdev_port_obj_act_is_deferred - Is object action pending?
341  *
342  *	@dev: port device
343  *	@nt: type of action; add or delete
344  *	@obj: object to test
345  *
346  *	Returns true if a deferred item is pending, which is
347  *	equivalent to the action @nt on an object @obj.
348  *
349  *	rtnl_lock must be held.
350  */
351 bool switchdev_port_obj_act_is_deferred(struct net_device *dev,
352 					enum switchdev_notifier_type nt,
353 					const struct switchdev_obj *obj)
354 {
355 	struct switchdev_deferred_item *dfitem;
356 	bool found = false;
357 
358 	ASSERT_RTNL();
359 
360 	spin_lock_bh(&deferred_lock);
361 
362 	list_for_each_entry(dfitem, &deferred, list) {
363 		if (dfitem->dev != dev)
364 			continue;
365 
366 		if ((dfitem->func == switchdev_port_obj_add_deferred &&
367 		     nt == SWITCHDEV_PORT_OBJ_ADD) ||
368 		    (dfitem->func == switchdev_port_obj_del_deferred &&
369 		     nt == SWITCHDEV_PORT_OBJ_DEL)) {
370 			if (switchdev_obj_eq((const void *)dfitem->data, obj)) {
371 				found = true;
372 				break;
373 			}
374 		}
375 	}
376 
377 	spin_unlock_bh(&deferred_lock);
378 
379 	return found;
380 }
381 EXPORT_SYMBOL_GPL(switchdev_port_obj_act_is_deferred);
382 
383 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
384 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
385 
386 /**
387  *	register_switchdev_notifier - Register notifier
388  *	@nb: notifier_block
389  *
390  *	Register switch device notifier.
391  */
392 int register_switchdev_notifier(struct notifier_block *nb)
393 {
394 	return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
395 }
396 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
397 
398 /**
399  *	unregister_switchdev_notifier - Unregister notifier
400  *	@nb: notifier_block
401  *
402  *	Unregister switch device notifier.
403  */
404 int unregister_switchdev_notifier(struct notifier_block *nb)
405 {
406 	return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
407 }
408 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
409 
410 /**
411  *	call_switchdev_notifiers - Call notifiers
412  *	@val: value passed unmodified to notifier function
413  *	@dev: port device
414  *	@info: notifier information data
415  *	@extack: netlink extended ack
416  *	Call all network notifier blocks.
417  */
418 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
419 			     struct switchdev_notifier_info *info,
420 			     struct netlink_ext_ack *extack)
421 {
422 	info->dev = dev;
423 	info->extack = extack;
424 	return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
425 }
426 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
427 
428 int register_switchdev_blocking_notifier(struct notifier_block *nb)
429 {
430 	struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
431 
432 	return blocking_notifier_chain_register(chain, nb);
433 }
434 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
435 
436 int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
437 {
438 	struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
439 
440 	return blocking_notifier_chain_unregister(chain, nb);
441 }
442 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
443 
444 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
445 				      struct switchdev_notifier_info *info,
446 				      struct netlink_ext_ack *extack)
447 {
448 	info->dev = dev;
449 	info->extack = extack;
450 	return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
451 					    val, info);
452 }
453 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
454 
455 struct switchdev_nested_priv {
456 	bool (*check_cb)(const struct net_device *dev);
457 	bool (*foreign_dev_check_cb)(const struct net_device *dev,
458 				     const struct net_device *foreign_dev);
459 	const struct net_device *dev;
460 	struct net_device *lower_dev;
461 };
462 
463 static int switchdev_lower_dev_walk(struct net_device *lower_dev,
464 				    struct netdev_nested_priv *priv)
465 {
466 	struct switchdev_nested_priv *switchdev_priv = priv->data;
467 	bool (*foreign_dev_check_cb)(const struct net_device *dev,
468 				     const struct net_device *foreign_dev);
469 	bool (*check_cb)(const struct net_device *dev);
470 	const struct net_device *dev;
471 
472 	check_cb = switchdev_priv->check_cb;
473 	foreign_dev_check_cb = switchdev_priv->foreign_dev_check_cb;
474 	dev = switchdev_priv->dev;
475 
476 	if (check_cb(lower_dev) && !foreign_dev_check_cb(lower_dev, dev)) {
477 		switchdev_priv->lower_dev = lower_dev;
478 		return 1;
479 	}
480 
481 	return 0;
482 }
483 
484 static struct net_device *
485 switchdev_lower_dev_find_rcu(struct net_device *dev,
486 			     bool (*check_cb)(const struct net_device *dev),
487 			     bool (*foreign_dev_check_cb)(const struct net_device *dev,
488 							  const struct net_device *foreign_dev))
489 {
490 	struct switchdev_nested_priv switchdev_priv = {
491 		.check_cb = check_cb,
492 		.foreign_dev_check_cb = foreign_dev_check_cb,
493 		.dev = dev,
494 		.lower_dev = NULL,
495 	};
496 	struct netdev_nested_priv priv = {
497 		.data = &switchdev_priv,
498 	};
499 
500 	netdev_walk_all_lower_dev_rcu(dev, switchdev_lower_dev_walk, &priv);
501 
502 	return switchdev_priv.lower_dev;
503 }
504 
505 static struct net_device *
506 switchdev_lower_dev_find(struct net_device *dev,
507 			 bool (*check_cb)(const struct net_device *dev),
508 			 bool (*foreign_dev_check_cb)(const struct net_device *dev,
509 						      const struct net_device *foreign_dev))
510 {
511 	struct switchdev_nested_priv switchdev_priv = {
512 		.check_cb = check_cb,
513 		.foreign_dev_check_cb = foreign_dev_check_cb,
514 		.dev = dev,
515 		.lower_dev = NULL,
516 	};
517 	struct netdev_nested_priv priv = {
518 		.data = &switchdev_priv,
519 	};
520 
521 	netdev_walk_all_lower_dev(dev, switchdev_lower_dev_walk, &priv);
522 
523 	return switchdev_priv.lower_dev;
524 }
525 
526 static int __switchdev_handle_fdb_event_to_device(struct net_device *dev,
527 		struct net_device *orig_dev, unsigned long event,
528 		const struct switchdev_notifier_fdb_info *fdb_info,
529 		bool (*check_cb)(const struct net_device *dev),
530 		bool (*foreign_dev_check_cb)(const struct net_device *dev,
531 					     const struct net_device *foreign_dev),
532 		int (*mod_cb)(struct net_device *dev, struct net_device *orig_dev,
533 			      unsigned long event, const void *ctx,
534 			      const struct switchdev_notifier_fdb_info *fdb_info))
535 {
536 	const struct switchdev_notifier_info *info = &fdb_info->info;
537 	struct net_device *br, *lower_dev, *switchdev;
538 	struct list_head *iter;
539 	int err = -EOPNOTSUPP;
540 
541 	if (check_cb(dev))
542 		return mod_cb(dev, orig_dev, event, info->ctx, fdb_info);
543 
544 	/* Recurse through lower interfaces in case the FDB entry is pointing
545 	 * towards a bridge or a LAG device.
546 	 */
547 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
548 		/* Do not propagate FDB entries across bridges */
549 		if (netif_is_bridge_master(lower_dev))
550 			continue;
551 
552 		/* Bridge ports might be either us, or LAG interfaces
553 		 * that we offload.
554 		 */
555 		if (!check_cb(lower_dev) &&
556 		    !switchdev_lower_dev_find_rcu(lower_dev, check_cb,
557 						  foreign_dev_check_cb))
558 			continue;
559 
560 		err = __switchdev_handle_fdb_event_to_device(lower_dev, orig_dev,
561 							     event, fdb_info, check_cb,
562 							     foreign_dev_check_cb,
563 							     mod_cb);
564 		if (err && err != -EOPNOTSUPP)
565 			return err;
566 	}
567 
568 	/* Event is neither on a bridge nor a LAG. Check whether it is on an
569 	 * interface that is in a bridge with us.
570 	 */
571 	br = netdev_master_upper_dev_get_rcu(dev);
572 	if (!br || !netif_is_bridge_master(br))
573 		return 0;
574 
575 	switchdev = switchdev_lower_dev_find_rcu(br, check_cb, foreign_dev_check_cb);
576 	if (!switchdev)
577 		return 0;
578 
579 	if (!foreign_dev_check_cb(switchdev, dev))
580 		return err;
581 
582 	return __switchdev_handle_fdb_event_to_device(br, orig_dev, event, fdb_info,
583 						      check_cb, foreign_dev_check_cb,
584 						      mod_cb);
585 }
586 
587 int switchdev_handle_fdb_event_to_device(struct net_device *dev, unsigned long event,
588 		const struct switchdev_notifier_fdb_info *fdb_info,
589 		bool (*check_cb)(const struct net_device *dev),
590 		bool (*foreign_dev_check_cb)(const struct net_device *dev,
591 					     const struct net_device *foreign_dev),
592 		int (*mod_cb)(struct net_device *dev, struct net_device *orig_dev,
593 			      unsigned long event, const void *ctx,
594 			      const struct switchdev_notifier_fdb_info *fdb_info))
595 {
596 	int err;
597 
598 	err = __switchdev_handle_fdb_event_to_device(dev, dev, event, fdb_info,
599 						     check_cb, foreign_dev_check_cb,
600 						     mod_cb);
601 	if (err == -EOPNOTSUPP)
602 		err = 0;
603 
604 	return err;
605 }
606 EXPORT_SYMBOL_GPL(switchdev_handle_fdb_event_to_device);
607 
608 static int __switchdev_handle_port_obj_add(struct net_device *dev,
609 			struct switchdev_notifier_port_obj_info *port_obj_info,
610 			bool (*check_cb)(const struct net_device *dev),
611 			bool (*foreign_dev_check_cb)(const struct net_device *dev,
612 						     const struct net_device *foreign_dev),
613 			int (*add_cb)(struct net_device *dev, const void *ctx,
614 				      const struct switchdev_obj *obj,
615 				      struct netlink_ext_ack *extack))
616 {
617 	struct switchdev_notifier_info *info = &port_obj_info->info;
618 	struct net_device *br, *lower_dev, *switchdev;
619 	struct netlink_ext_ack *extack;
620 	struct list_head *iter;
621 	int err = -EOPNOTSUPP;
622 
623 	extack = switchdev_notifier_info_to_extack(info);
624 
625 	if (check_cb(dev)) {
626 		err = add_cb(dev, info->ctx, port_obj_info->obj, extack);
627 		if (err != -EOPNOTSUPP)
628 			port_obj_info->handled = true;
629 		return err;
630 	}
631 
632 	/* Switch ports might be stacked under e.g. a LAG. Ignore the
633 	 * unsupported devices, another driver might be able to handle them. But
634 	 * propagate to the callers any hard errors.
635 	 *
636 	 * If the driver does its own bookkeeping of stacked ports, it's not
637 	 * necessary to go through this helper.
638 	 */
639 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
640 		if (netif_is_bridge_master(lower_dev))
641 			continue;
642 
643 		/* When searching for switchdev interfaces that are neighbors
644 		 * of foreign ones, and @dev is a bridge, do not recurse on the
645 		 * foreign interface again, it was already visited.
646 		 */
647 		if (foreign_dev_check_cb && !check_cb(lower_dev) &&
648 		    !switchdev_lower_dev_find(lower_dev, check_cb, foreign_dev_check_cb))
649 			continue;
650 
651 		err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
652 						      check_cb, foreign_dev_check_cb,
653 						      add_cb);
654 		if (err && err != -EOPNOTSUPP)
655 			return err;
656 	}
657 
658 	/* Event is neither on a bridge nor a LAG. Check whether it is on an
659 	 * interface that is in a bridge with us.
660 	 */
661 	if (!foreign_dev_check_cb)
662 		return err;
663 
664 	br = netdev_master_upper_dev_get(dev);
665 	if (!br || !netif_is_bridge_master(br))
666 		return err;
667 
668 	switchdev = switchdev_lower_dev_find(br, check_cb, foreign_dev_check_cb);
669 	if (!switchdev)
670 		return err;
671 
672 	if (!foreign_dev_check_cb(switchdev, dev))
673 		return err;
674 
675 	return __switchdev_handle_port_obj_add(br, port_obj_info, check_cb,
676 					       foreign_dev_check_cb, add_cb);
677 }
678 
679 /* Pass through a port object addition, if @dev passes @check_cb, or replicate
680  * it towards all lower interfaces of @dev that pass @check_cb, if @dev is a
681  * bridge or a LAG.
682  */
683 int switchdev_handle_port_obj_add(struct net_device *dev,
684 			struct switchdev_notifier_port_obj_info *port_obj_info,
685 			bool (*check_cb)(const struct net_device *dev),
686 			int (*add_cb)(struct net_device *dev, const void *ctx,
687 				      const struct switchdev_obj *obj,
688 				      struct netlink_ext_ack *extack))
689 {
690 	int err;
691 
692 	err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
693 					      NULL, add_cb);
694 	if (err == -EOPNOTSUPP)
695 		err = 0;
696 	return err;
697 }
698 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
699 
700 /* Same as switchdev_handle_port_obj_add(), except if object is notified on a
701  * @dev that passes @foreign_dev_check_cb, it is replicated towards all devices
702  * that pass @check_cb and are in the same bridge as @dev.
703  */
704 int switchdev_handle_port_obj_add_foreign(struct net_device *dev,
705 			struct switchdev_notifier_port_obj_info *port_obj_info,
706 			bool (*check_cb)(const struct net_device *dev),
707 			bool (*foreign_dev_check_cb)(const struct net_device *dev,
708 						     const struct net_device *foreign_dev),
709 			int (*add_cb)(struct net_device *dev, const void *ctx,
710 				      const struct switchdev_obj *obj,
711 				      struct netlink_ext_ack *extack))
712 {
713 	int err;
714 
715 	err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
716 					      foreign_dev_check_cb, add_cb);
717 	if (err == -EOPNOTSUPP)
718 		err = 0;
719 	return err;
720 }
721 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add_foreign);
722 
723 static int __switchdev_handle_port_obj_del(struct net_device *dev,
724 			struct switchdev_notifier_port_obj_info *port_obj_info,
725 			bool (*check_cb)(const struct net_device *dev),
726 			bool (*foreign_dev_check_cb)(const struct net_device *dev,
727 						     const struct net_device *foreign_dev),
728 			int (*del_cb)(struct net_device *dev, const void *ctx,
729 				      const struct switchdev_obj *obj))
730 {
731 	struct switchdev_notifier_info *info = &port_obj_info->info;
732 	struct net_device *br, *lower_dev, *switchdev;
733 	struct list_head *iter;
734 	int err = -EOPNOTSUPP;
735 
736 	if (check_cb(dev)) {
737 		err = del_cb(dev, info->ctx, port_obj_info->obj);
738 		if (err != -EOPNOTSUPP)
739 			port_obj_info->handled = true;
740 		return err;
741 	}
742 
743 	/* Switch ports might be stacked under e.g. a LAG. Ignore the
744 	 * unsupported devices, another driver might be able to handle them. But
745 	 * propagate to the callers any hard errors.
746 	 *
747 	 * If the driver does its own bookkeeping of stacked ports, it's not
748 	 * necessary to go through this helper.
749 	 */
750 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
751 		if (netif_is_bridge_master(lower_dev))
752 			continue;
753 
754 		/* When searching for switchdev interfaces that are neighbors
755 		 * of foreign ones, and @dev is a bridge, do not recurse on the
756 		 * foreign interface again, it was already visited.
757 		 */
758 		if (foreign_dev_check_cb && !check_cb(lower_dev) &&
759 		    !switchdev_lower_dev_find(lower_dev, check_cb, foreign_dev_check_cb))
760 			continue;
761 
762 		err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
763 						      check_cb, foreign_dev_check_cb,
764 						      del_cb);
765 		if (err && err != -EOPNOTSUPP)
766 			return err;
767 	}
768 
769 	/* Event is neither on a bridge nor a LAG. Check whether it is on an
770 	 * interface that is in a bridge with us.
771 	 */
772 	if (!foreign_dev_check_cb)
773 		return err;
774 
775 	br = netdev_master_upper_dev_get(dev);
776 	if (!br || !netif_is_bridge_master(br))
777 		return err;
778 
779 	switchdev = switchdev_lower_dev_find(br, check_cb, foreign_dev_check_cb);
780 	if (!switchdev)
781 		return err;
782 
783 	if (!foreign_dev_check_cb(switchdev, dev))
784 		return err;
785 
786 	return __switchdev_handle_port_obj_del(br, port_obj_info, check_cb,
787 					       foreign_dev_check_cb, del_cb);
788 }
789 
790 /* Pass through a port object deletion, if @dev passes @check_cb, or replicate
791  * it towards all lower interfaces of @dev that pass @check_cb, if @dev is a
792  * bridge or a LAG.
793  */
794 int switchdev_handle_port_obj_del(struct net_device *dev,
795 			struct switchdev_notifier_port_obj_info *port_obj_info,
796 			bool (*check_cb)(const struct net_device *dev),
797 			int (*del_cb)(struct net_device *dev, const void *ctx,
798 				      const struct switchdev_obj *obj))
799 {
800 	int err;
801 
802 	err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
803 					      NULL, del_cb);
804 	if (err == -EOPNOTSUPP)
805 		err = 0;
806 	return err;
807 }
808 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
809 
810 /* Same as switchdev_handle_port_obj_del(), except if object is notified on a
811  * @dev that passes @foreign_dev_check_cb, it is replicated towards all devices
812  * that pass @check_cb and are in the same bridge as @dev.
813  */
814 int switchdev_handle_port_obj_del_foreign(struct net_device *dev,
815 			struct switchdev_notifier_port_obj_info *port_obj_info,
816 			bool (*check_cb)(const struct net_device *dev),
817 			bool (*foreign_dev_check_cb)(const struct net_device *dev,
818 						     const struct net_device *foreign_dev),
819 			int (*del_cb)(struct net_device *dev, const void *ctx,
820 				      const struct switchdev_obj *obj))
821 {
822 	int err;
823 
824 	err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
825 					      foreign_dev_check_cb, del_cb);
826 	if (err == -EOPNOTSUPP)
827 		err = 0;
828 	return err;
829 }
830 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del_foreign);
831 
832 static int __switchdev_handle_port_attr_set(struct net_device *dev,
833 			struct switchdev_notifier_port_attr_info *port_attr_info,
834 			bool (*check_cb)(const struct net_device *dev),
835 			int (*set_cb)(struct net_device *dev, const void *ctx,
836 				      const struct switchdev_attr *attr,
837 				      struct netlink_ext_ack *extack))
838 {
839 	struct switchdev_notifier_info *info = &port_attr_info->info;
840 	struct netlink_ext_ack *extack;
841 	struct net_device *lower_dev;
842 	struct list_head *iter;
843 	int err = -EOPNOTSUPP;
844 
845 	extack = switchdev_notifier_info_to_extack(info);
846 
847 	if (check_cb(dev)) {
848 		err = set_cb(dev, info->ctx, port_attr_info->attr, extack);
849 		if (err != -EOPNOTSUPP)
850 			port_attr_info->handled = true;
851 		return err;
852 	}
853 
854 	/* Switch ports might be stacked under e.g. a LAG. Ignore the
855 	 * unsupported devices, another driver might be able to handle them. But
856 	 * propagate to the callers any hard errors.
857 	 *
858 	 * If the driver does its own bookkeeping of stacked ports, it's not
859 	 * necessary to go through this helper.
860 	 */
861 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
862 		if (netif_is_bridge_master(lower_dev))
863 			continue;
864 
865 		err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
866 						       check_cb, set_cb);
867 		if (err && err != -EOPNOTSUPP)
868 			return err;
869 	}
870 
871 	return err;
872 }
873 
874 int switchdev_handle_port_attr_set(struct net_device *dev,
875 			struct switchdev_notifier_port_attr_info *port_attr_info,
876 			bool (*check_cb)(const struct net_device *dev),
877 			int (*set_cb)(struct net_device *dev, const void *ctx,
878 				      const struct switchdev_attr *attr,
879 				      struct netlink_ext_ack *extack))
880 {
881 	int err;
882 
883 	err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
884 					       set_cb);
885 	if (err == -EOPNOTSUPP)
886 		err = 0;
887 	return err;
888 }
889 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);
890 
891 int switchdev_bridge_port_offload(struct net_device *brport_dev,
892 				  struct net_device *dev, const void *ctx,
893 				  struct notifier_block *atomic_nb,
894 				  struct notifier_block *blocking_nb,
895 				  bool tx_fwd_offload,
896 				  struct netlink_ext_ack *extack)
897 {
898 	struct switchdev_notifier_brport_info brport_info = {
899 		.brport = {
900 			.dev = dev,
901 			.ctx = ctx,
902 			.atomic_nb = atomic_nb,
903 			.blocking_nb = blocking_nb,
904 			.tx_fwd_offload = tx_fwd_offload,
905 		},
906 	};
907 	int err;
908 
909 	ASSERT_RTNL();
910 
911 	err = call_switchdev_blocking_notifiers(SWITCHDEV_BRPORT_OFFLOADED,
912 						brport_dev, &brport_info.info,
913 						extack);
914 	return notifier_to_errno(err);
915 }
916 EXPORT_SYMBOL_GPL(switchdev_bridge_port_offload);
917 
918 void switchdev_bridge_port_unoffload(struct net_device *brport_dev,
919 				     const void *ctx,
920 				     struct notifier_block *atomic_nb,
921 				     struct notifier_block *blocking_nb)
922 {
923 	struct switchdev_notifier_brport_info brport_info = {
924 		.brport = {
925 			.ctx = ctx,
926 			.atomic_nb = atomic_nb,
927 			.blocking_nb = blocking_nb,
928 		},
929 	};
930 
931 	ASSERT_RTNL();
932 
933 	call_switchdev_blocking_notifiers(SWITCHDEV_BRPORT_UNOFFLOADED,
934 					  brport_dev, &brport_info.info,
935 					  NULL);
936 }
937 EXPORT_SYMBOL_GPL(switchdev_bridge_port_unoffload);
938 
939 int switchdev_bridge_port_replay(struct net_device *brport_dev,
940 				 struct net_device *dev, const void *ctx,
941 				 struct notifier_block *atomic_nb,
942 				 struct notifier_block *blocking_nb,
943 				 struct netlink_ext_ack *extack)
944 {
945 	struct switchdev_notifier_brport_info brport_info = {
946 		.brport = {
947 			.dev = dev,
948 			.ctx = ctx,
949 			.atomic_nb = atomic_nb,
950 			.blocking_nb = blocking_nb,
951 		},
952 	};
953 	int err;
954 
955 	ASSERT_RTNL();
956 
957 	err = call_switchdev_blocking_notifiers(SWITCHDEV_BRPORT_REPLAY,
958 						brport_dev, &brport_info.info,
959 						extack);
960 	return notifier_to_errno(err);
961 }
962 EXPORT_SYMBOL_GPL(switchdev_bridge_port_replay);
963