Loading...
1/*
2 * net/switchdev/switchdev.c - Switch device API
3 * Copyright (c) 2014-2015 Jiri Pirko <jiri@resnulli.us>
4 * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 */
11
12#include <linux/kernel.h>
13#include <linux/types.h>
14#include <linux/init.h>
15#include <linux/mutex.h>
16#include <linux/notifier.h>
17#include <linux/netdevice.h>
18#include <linux/etherdevice.h>
19#include <linux/if_bridge.h>
20#include <linux/list.h>
21#include <linux/workqueue.h>
22#include <linux/if_vlan.h>
23#include <linux/rtnetlink.h>
24#include <net/switchdev.h>
25
26/**
27 * switchdev_trans_item_enqueue - Enqueue data item to transaction queue
28 *
29 * @trans: transaction
30 * @data: pointer to data being queued
31 * @destructor: data destructor
32 * @tritem: transaction item being queued
33 *
34 * Enqeueue data item to transaction queue. tritem is typically placed in
35 * cointainter pointed at by data pointer. Destructor is called on
36 * transaction abort and after successful commit phase in case
37 * the caller did not dequeue the item before.
38 */
39void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
40 void *data, void (*destructor)(void const *),
41 struct switchdev_trans_item *tritem)
42{
43 tritem->data = data;
44 tritem->destructor = destructor;
45 list_add_tail(&tritem->list, &trans->item_list);
46}
47EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
48
49static struct switchdev_trans_item *
50__switchdev_trans_item_dequeue(struct switchdev_trans *trans)
51{
52 struct switchdev_trans_item *tritem;
53
54 if (list_empty(&trans->item_list))
55 return NULL;
56 tritem = list_first_entry(&trans->item_list,
57 struct switchdev_trans_item, list);
58 list_del(&tritem->list);
59 return tritem;
60}
61
62/**
63 * switchdev_trans_item_dequeue - Dequeue data item from transaction queue
64 *
65 * @trans: transaction
66 */
67void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
68{
69 struct switchdev_trans_item *tritem;
70
71 tritem = __switchdev_trans_item_dequeue(trans);
72 BUG_ON(!tritem);
73 return tritem->data;
74}
75EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
76
77static void switchdev_trans_init(struct switchdev_trans *trans)
78{
79 INIT_LIST_HEAD(&trans->item_list);
80}
81
82static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
83{
84 struct switchdev_trans_item *tritem;
85
86 while ((tritem = __switchdev_trans_item_dequeue(trans)))
87 tritem->destructor(tritem->data);
88}
89
90static void switchdev_trans_items_warn_destroy(struct net_device *dev,
91 struct switchdev_trans *trans)
92{
93 WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
94 dev->name);
95 switchdev_trans_items_destroy(trans);
96}
97
98static LIST_HEAD(deferred);
99static DEFINE_SPINLOCK(deferred_lock);
100
101typedef void switchdev_deferred_func_t(struct net_device *dev,
102 const void *data);
103
104struct switchdev_deferred_item {
105 struct list_head list;
106 struct net_device *dev;
107 switchdev_deferred_func_t *func;
108 unsigned long data[0];
109};
110
111static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
112{
113 struct switchdev_deferred_item *dfitem;
114
115 spin_lock_bh(&deferred_lock);
116 if (list_empty(&deferred)) {
117 dfitem = NULL;
118 goto unlock;
119 }
120 dfitem = list_first_entry(&deferred,
121 struct switchdev_deferred_item, list);
122 list_del(&dfitem->list);
123unlock:
124 spin_unlock_bh(&deferred_lock);
125 return dfitem;
126}
127
128/**
129 * switchdev_deferred_process - Process ops in deferred queue
130 *
131 * Called to flush the ops currently queued in deferred ops queue.
132 * rtnl_lock must be held.
133 */
134void switchdev_deferred_process(void)
135{
136 struct switchdev_deferred_item *dfitem;
137
138 ASSERT_RTNL();
139
140 while ((dfitem = switchdev_deferred_dequeue())) {
141 dfitem->func(dfitem->dev, dfitem->data);
142 dev_put(dfitem->dev);
143 kfree(dfitem);
144 }
145}
146EXPORT_SYMBOL_GPL(switchdev_deferred_process);
147
148static void switchdev_deferred_process_work(struct work_struct *work)
149{
150 rtnl_lock();
151 switchdev_deferred_process();
152 rtnl_unlock();
153}
154
155static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
156
157static int switchdev_deferred_enqueue(struct net_device *dev,
158 const void *data, size_t data_len,
159 switchdev_deferred_func_t *func)
160{
161 struct switchdev_deferred_item *dfitem;
162
163 dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
164 if (!dfitem)
165 return -ENOMEM;
166 dfitem->dev = dev;
167 dfitem->func = func;
168 memcpy(dfitem->data, data, data_len);
169 dev_hold(dev);
170 spin_lock_bh(&deferred_lock);
171 list_add_tail(&dfitem->list, &deferred);
172 spin_unlock_bh(&deferred_lock);
173 schedule_work(&deferred_process_work);
174 return 0;
175}
176
177/**
178 * switchdev_port_attr_get - Get port attribute
179 *
180 * @dev: port device
181 * @attr: attribute to get
182 */
183int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
184{
185 const struct switchdev_ops *ops = dev->switchdev_ops;
186 struct net_device *lower_dev;
187 struct list_head *iter;
188 struct switchdev_attr first = {
189 .id = SWITCHDEV_ATTR_ID_UNDEFINED
190 };
191 int err = -EOPNOTSUPP;
192
193 if (ops && ops->switchdev_port_attr_get)
194 return ops->switchdev_port_attr_get(dev, attr);
195
196 if (attr->flags & SWITCHDEV_F_NO_RECURSE)
197 return err;
198
199 /* Switch device port(s) may be stacked under
200 * bond/team/vlan dev, so recurse down to get attr on
201 * each port. Return -ENODATA if attr values don't
202 * compare across ports.
203 */
204
205 netdev_for_each_lower_dev(dev, lower_dev, iter) {
206 err = switchdev_port_attr_get(lower_dev, attr);
207 if (err)
208 break;
209 if (first.id == SWITCHDEV_ATTR_ID_UNDEFINED)
210 first = *attr;
211 else if (memcmp(&first, attr, sizeof(*attr)))
212 return -ENODATA;
213 }
214
215 return err;
216}
217EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
218
219static int __switchdev_port_attr_set(struct net_device *dev,
220 const struct switchdev_attr *attr,
221 struct switchdev_trans *trans)
222{
223 const struct switchdev_ops *ops = dev->switchdev_ops;
224 struct net_device *lower_dev;
225 struct list_head *iter;
226 int err = -EOPNOTSUPP;
227
228 if (ops && ops->switchdev_port_attr_set) {
229 err = ops->switchdev_port_attr_set(dev, attr, trans);
230 goto done;
231 }
232
233 if (attr->flags & SWITCHDEV_F_NO_RECURSE)
234 goto done;
235
236 /* Switch device port(s) may be stacked under
237 * bond/team/vlan dev, so recurse down to set attr on
238 * each port.
239 */
240
241 netdev_for_each_lower_dev(dev, lower_dev, iter) {
242 err = __switchdev_port_attr_set(lower_dev, attr, trans);
243 if (err)
244 break;
245 }
246
247done:
248 if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
249 err = 0;
250
251 return err;
252}
253
254static int switchdev_port_attr_set_now(struct net_device *dev,
255 const struct switchdev_attr *attr)
256{
257 struct switchdev_trans trans;
258 int err;
259
260 switchdev_trans_init(&trans);
261
262 /* Phase I: prepare for attr set. Driver/device should fail
263 * here if there are going to be issues in the commit phase,
264 * such as lack of resources or support. The driver/device
265 * should reserve resources needed for the commit phase here,
266 * but should not commit the attr.
267 */
268
269 trans.ph_prepare = true;
270 err = __switchdev_port_attr_set(dev, attr, &trans);
271 if (err) {
272 /* Prepare phase failed: abort the transaction. Any
273 * resources reserved in the prepare phase are
274 * released.
275 */
276
277 if (err != -EOPNOTSUPP)
278 switchdev_trans_items_destroy(&trans);
279
280 return err;
281 }
282
283 /* Phase II: commit attr set. This cannot fail as a fault
284 * of driver/device. If it does, it's a bug in the driver/device
285 * because the driver said everythings was OK in phase I.
286 */
287
288 trans.ph_prepare = false;
289 err = __switchdev_port_attr_set(dev, attr, &trans);
290 WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
291 dev->name, attr->id);
292 switchdev_trans_items_warn_destroy(dev, &trans);
293
294 return err;
295}
296
297static void switchdev_port_attr_set_deferred(struct net_device *dev,
298 const void *data)
299{
300 const struct switchdev_attr *attr = data;
301 int err;
302
303 err = switchdev_port_attr_set_now(dev, attr);
304 if (err && err != -EOPNOTSUPP)
305 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
306 err, attr->id);
307 if (attr->complete)
308 attr->complete(dev, err, attr->complete_priv);
309}
310
311static int switchdev_port_attr_set_defer(struct net_device *dev,
312 const struct switchdev_attr *attr)
313{
314 return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
315 switchdev_port_attr_set_deferred);
316}
317
318/**
319 * switchdev_port_attr_set - Set port attribute
320 *
321 * @dev: port device
322 * @attr: attribute to set
323 *
324 * Use a 2-phase prepare-commit transaction model to ensure
325 * system is not left in a partially updated state due to
326 * failure from driver/device.
327 *
328 * rtnl_lock must be held and must not be in atomic section,
329 * in case SWITCHDEV_F_DEFER flag is not set.
330 */
331int switchdev_port_attr_set(struct net_device *dev,
332 const struct switchdev_attr *attr)
333{
334 if (attr->flags & SWITCHDEV_F_DEFER)
335 return switchdev_port_attr_set_defer(dev, attr);
336 ASSERT_RTNL();
337 return switchdev_port_attr_set_now(dev, attr);
338}
339EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
340
341static size_t switchdev_obj_size(const struct switchdev_obj *obj)
342{
343 switch (obj->id) {
344 case SWITCHDEV_OBJ_ID_PORT_VLAN:
345 return sizeof(struct switchdev_obj_port_vlan);
346 case SWITCHDEV_OBJ_ID_PORT_FDB:
347 return sizeof(struct switchdev_obj_port_fdb);
348 case SWITCHDEV_OBJ_ID_PORT_MDB:
349 return sizeof(struct switchdev_obj_port_mdb);
350 default:
351 BUG();
352 }
353 return 0;
354}
355
356static int __switchdev_port_obj_add(struct net_device *dev,
357 const struct switchdev_obj *obj,
358 struct switchdev_trans *trans)
359{
360 const struct switchdev_ops *ops = dev->switchdev_ops;
361 struct net_device *lower_dev;
362 struct list_head *iter;
363 int err = -EOPNOTSUPP;
364
365 if (ops && ops->switchdev_port_obj_add)
366 return ops->switchdev_port_obj_add(dev, obj, trans);
367
368 /* Switch device port(s) may be stacked under
369 * bond/team/vlan dev, so recurse down to add object on
370 * each port.
371 */
372
373 netdev_for_each_lower_dev(dev, lower_dev, iter) {
374 err = __switchdev_port_obj_add(lower_dev, obj, trans);
375 if (err)
376 break;
377 }
378
379 return err;
380}
381
382static int switchdev_port_obj_add_now(struct net_device *dev,
383 const struct switchdev_obj *obj)
384{
385 struct switchdev_trans trans;
386 int err;
387
388 ASSERT_RTNL();
389
390 switchdev_trans_init(&trans);
391
392 /* Phase I: prepare for obj add. Driver/device should fail
393 * here if there are going to be issues in the commit phase,
394 * such as lack of resources or support. The driver/device
395 * should reserve resources needed for the commit phase here,
396 * but should not commit the obj.
397 */
398
399 trans.ph_prepare = true;
400 err = __switchdev_port_obj_add(dev, obj, &trans);
401 if (err) {
402 /* Prepare phase failed: abort the transaction. Any
403 * resources reserved in the prepare phase are
404 * released.
405 */
406
407 if (err != -EOPNOTSUPP)
408 switchdev_trans_items_destroy(&trans);
409
410 return err;
411 }
412
413 /* Phase II: commit obj add. This cannot fail as a fault
414 * of driver/device. If it does, it's a bug in the driver/device
415 * because the driver said everythings was OK in phase I.
416 */
417
418 trans.ph_prepare = false;
419 err = __switchdev_port_obj_add(dev, obj, &trans);
420 WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
421 switchdev_trans_items_warn_destroy(dev, &trans);
422
423 return err;
424}
425
426static void switchdev_port_obj_add_deferred(struct net_device *dev,
427 const void *data)
428{
429 const struct switchdev_obj *obj = data;
430 int err;
431
432 err = switchdev_port_obj_add_now(dev, obj);
433 if (err && err != -EOPNOTSUPP)
434 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
435 err, obj->id);
436 if (obj->complete)
437 obj->complete(dev, err, obj->complete_priv);
438}
439
440static int switchdev_port_obj_add_defer(struct net_device *dev,
441 const struct switchdev_obj *obj)
442{
443 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
444 switchdev_port_obj_add_deferred);
445}
446
447/**
448 * switchdev_port_obj_add - Add port object
449 *
450 * @dev: port device
451 * @id: object ID
452 * @obj: object to add
453 *
454 * Use a 2-phase prepare-commit transaction model to ensure
455 * system is not left in a partially updated state due to
456 * failure from driver/device.
457 *
458 * rtnl_lock must be held and must not be in atomic section,
459 * in case SWITCHDEV_F_DEFER flag is not set.
460 */
461int switchdev_port_obj_add(struct net_device *dev,
462 const struct switchdev_obj *obj)
463{
464 if (obj->flags & SWITCHDEV_F_DEFER)
465 return switchdev_port_obj_add_defer(dev, obj);
466 ASSERT_RTNL();
467 return switchdev_port_obj_add_now(dev, obj);
468}
469EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
470
471static int switchdev_port_obj_del_now(struct net_device *dev,
472 const struct switchdev_obj *obj)
473{
474 const struct switchdev_ops *ops = dev->switchdev_ops;
475 struct net_device *lower_dev;
476 struct list_head *iter;
477 int err = -EOPNOTSUPP;
478
479 if (ops && ops->switchdev_port_obj_del)
480 return ops->switchdev_port_obj_del(dev, obj);
481
482 /* Switch device port(s) may be stacked under
483 * bond/team/vlan dev, so recurse down to delete object on
484 * each port.
485 */
486
487 netdev_for_each_lower_dev(dev, lower_dev, iter) {
488 err = switchdev_port_obj_del_now(lower_dev, obj);
489 if (err)
490 break;
491 }
492
493 return err;
494}
495
496static void switchdev_port_obj_del_deferred(struct net_device *dev,
497 const void *data)
498{
499 const struct switchdev_obj *obj = data;
500 int err;
501
502 err = switchdev_port_obj_del_now(dev, obj);
503 if (err && err != -EOPNOTSUPP)
504 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
505 err, obj->id);
506 if (obj->complete)
507 obj->complete(dev, err, obj->complete_priv);
508}
509
510static int switchdev_port_obj_del_defer(struct net_device *dev,
511 const struct switchdev_obj *obj)
512{
513 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
514 switchdev_port_obj_del_deferred);
515}
516
517/**
518 * switchdev_port_obj_del - Delete port object
519 *
520 * @dev: port device
521 * @id: object ID
522 * @obj: object to delete
523 *
524 * rtnl_lock must be held and must not be in atomic section,
525 * in case SWITCHDEV_F_DEFER flag is not set.
526 */
527int switchdev_port_obj_del(struct net_device *dev,
528 const struct switchdev_obj *obj)
529{
530 if (obj->flags & SWITCHDEV_F_DEFER)
531 return switchdev_port_obj_del_defer(dev, obj);
532 ASSERT_RTNL();
533 return switchdev_port_obj_del_now(dev, obj);
534}
535EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
536
537/**
538 * switchdev_port_obj_dump - Dump port objects
539 *
540 * @dev: port device
541 * @id: object ID
542 * @obj: object to dump
543 * @cb: function to call with a filled object
544 *
545 * rtnl_lock must be held.
546 */
547int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
548 switchdev_obj_dump_cb_t *cb)
549{
550 const struct switchdev_ops *ops = dev->switchdev_ops;
551 struct net_device *lower_dev;
552 struct list_head *iter;
553 int err = -EOPNOTSUPP;
554
555 ASSERT_RTNL();
556
557 if (ops && ops->switchdev_port_obj_dump)
558 return ops->switchdev_port_obj_dump(dev, obj, cb);
559
560 /* Switch device port(s) may be stacked under
561 * bond/team/vlan dev, so recurse down to dump objects on
562 * first port at bottom of stack.
563 */
564
565 netdev_for_each_lower_dev(dev, lower_dev, iter) {
566 err = switchdev_port_obj_dump(lower_dev, obj, cb);
567 break;
568 }
569
570 return err;
571}
572EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
573
574static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
575
576/**
577 * register_switchdev_notifier - Register notifier
578 * @nb: notifier_block
579 *
580 * Register switch device notifier. This should be used by code
581 * which needs to monitor events happening in particular device.
582 * Return values are same as for atomic_notifier_chain_register().
583 */
584int register_switchdev_notifier(struct notifier_block *nb)
585{
586 int err;
587
588 rtnl_lock();
589 err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
590 rtnl_unlock();
591 return err;
592}
593EXPORT_SYMBOL_GPL(register_switchdev_notifier);
594
595/**
596 * unregister_switchdev_notifier - Unregister notifier
597 * @nb: notifier_block
598 *
599 * Unregister switch device notifier.
600 * Return values are same as for atomic_notifier_chain_unregister().
601 */
602int unregister_switchdev_notifier(struct notifier_block *nb)
603{
604 int err;
605
606 rtnl_lock();
607 err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
608 rtnl_unlock();
609 return err;
610}
611EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
612
613/**
614 * call_switchdev_notifiers - Call notifiers
615 * @val: value passed unmodified to notifier function
616 * @dev: port device
617 * @info: notifier information data
618 *
619 * Call all network notifier blocks. This should be called by driver
620 * when it needs to propagate hardware event.
621 * Return values are same as for atomic_notifier_call_chain().
622 * rtnl_lock must be held.
623 */
624int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
625 struct switchdev_notifier_info *info)
626{
627 ASSERT_RTNL();
628
629 info->dev = dev;
630 return raw_notifier_call_chain(&switchdev_notif_chain, val, info);
631}
632EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
633
634struct switchdev_vlan_dump {
635 struct switchdev_obj_port_vlan vlan;
636 struct sk_buff *skb;
637 u32 filter_mask;
638 u16 flags;
639 u16 begin;
640 u16 end;
641};
642
643static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
644{
645 struct bridge_vlan_info vinfo;
646
647 vinfo.flags = dump->flags;
648
649 if (dump->begin == 0 && dump->end == 0) {
650 return 0;
651 } else if (dump->begin == dump->end) {
652 vinfo.vid = dump->begin;
653 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
654 sizeof(vinfo), &vinfo))
655 return -EMSGSIZE;
656 } else {
657 vinfo.vid = dump->begin;
658 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
659 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
660 sizeof(vinfo), &vinfo))
661 return -EMSGSIZE;
662 vinfo.vid = dump->end;
663 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
664 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
665 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
666 sizeof(vinfo), &vinfo))
667 return -EMSGSIZE;
668 }
669
670 return 0;
671}
672
673static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
674{
675 struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
676 struct switchdev_vlan_dump *dump =
677 container_of(vlan, struct switchdev_vlan_dump, vlan);
678 int err = 0;
679
680 if (vlan->vid_begin > vlan->vid_end)
681 return -EINVAL;
682
683 if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
684 dump->flags = vlan->flags;
685 for (dump->begin = dump->end = vlan->vid_begin;
686 dump->begin <= vlan->vid_end;
687 dump->begin++, dump->end++) {
688 err = switchdev_port_vlan_dump_put(dump);
689 if (err)
690 return err;
691 }
692 } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
693 if (dump->begin > vlan->vid_begin &&
694 dump->begin >= vlan->vid_end) {
695 if ((dump->begin - 1) == vlan->vid_end &&
696 dump->flags == vlan->flags) {
697 /* prepend */
698 dump->begin = vlan->vid_begin;
699 } else {
700 err = switchdev_port_vlan_dump_put(dump);
701 dump->flags = vlan->flags;
702 dump->begin = vlan->vid_begin;
703 dump->end = vlan->vid_end;
704 }
705 } else if (dump->end <= vlan->vid_begin &&
706 dump->end < vlan->vid_end) {
707 if ((dump->end + 1) == vlan->vid_begin &&
708 dump->flags == vlan->flags) {
709 /* append */
710 dump->end = vlan->vid_end;
711 } else {
712 err = switchdev_port_vlan_dump_put(dump);
713 dump->flags = vlan->flags;
714 dump->begin = vlan->vid_begin;
715 dump->end = vlan->vid_end;
716 }
717 } else {
718 err = -EINVAL;
719 }
720 }
721
722 return err;
723}
724
725static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
726 u32 filter_mask)
727{
728 struct switchdev_vlan_dump dump = {
729 .vlan.obj.orig_dev = dev,
730 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
731 .skb = skb,
732 .filter_mask = filter_mask,
733 };
734 int err = 0;
735
736 if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
737 (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
738 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
739 switchdev_port_vlan_dump_cb);
740 if (err)
741 goto err_out;
742 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
743 /* last one */
744 err = switchdev_port_vlan_dump_put(&dump);
745 }
746
747err_out:
748 return err == -EOPNOTSUPP ? 0 : err;
749}
750
751/**
752 * switchdev_port_bridge_getlink - Get bridge port attributes
753 *
754 * @dev: port device
755 *
756 * Called for SELF on rtnl_bridge_getlink to get bridge port
757 * attributes.
758 */
759int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
760 struct net_device *dev, u32 filter_mask,
761 int nlflags)
762{
763 struct switchdev_attr attr = {
764 .orig_dev = dev,
765 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
766 };
767 u16 mode = BRIDGE_MODE_UNDEF;
768 u32 mask = BR_LEARNING | BR_LEARNING_SYNC | BR_FLOOD;
769 int err;
770
771 if (!netif_is_bridge_port(dev))
772 return -EOPNOTSUPP;
773
774 err = switchdev_port_attr_get(dev, &attr);
775 if (err && err != -EOPNOTSUPP)
776 return err;
777
778 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
779 attr.u.brport_flags, mask, nlflags,
780 filter_mask, switchdev_port_vlan_fill);
781}
782EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
783
784static int switchdev_port_br_setflag(struct net_device *dev,
785 struct nlattr *nlattr,
786 unsigned long brport_flag)
787{
788 struct switchdev_attr attr = {
789 .orig_dev = dev,
790 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
791 };
792 u8 flag = nla_get_u8(nlattr);
793 int err;
794
795 err = switchdev_port_attr_get(dev, &attr);
796 if (err)
797 return err;
798
799 if (flag)
800 attr.u.brport_flags |= brport_flag;
801 else
802 attr.u.brport_flags &= ~brport_flag;
803
804 return switchdev_port_attr_set(dev, &attr);
805}
806
807static const struct nla_policy
808switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
809 [IFLA_BRPORT_STATE] = { .type = NLA_U8 },
810 [IFLA_BRPORT_COST] = { .type = NLA_U32 },
811 [IFLA_BRPORT_PRIORITY] = { .type = NLA_U16 },
812 [IFLA_BRPORT_MODE] = { .type = NLA_U8 },
813 [IFLA_BRPORT_GUARD] = { .type = NLA_U8 },
814 [IFLA_BRPORT_PROTECT] = { .type = NLA_U8 },
815 [IFLA_BRPORT_FAST_LEAVE] = { .type = NLA_U8 },
816 [IFLA_BRPORT_LEARNING] = { .type = NLA_U8 },
817 [IFLA_BRPORT_LEARNING_SYNC] = { .type = NLA_U8 },
818 [IFLA_BRPORT_UNICAST_FLOOD] = { .type = NLA_U8 },
819};
820
821static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
822 struct nlattr *protinfo)
823{
824 struct nlattr *attr;
825 int rem;
826 int err;
827
828 err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
829 switchdev_port_bridge_policy);
830 if (err)
831 return err;
832
833 nla_for_each_nested(attr, protinfo, rem) {
834 switch (nla_type(attr)) {
835 case IFLA_BRPORT_LEARNING:
836 err = switchdev_port_br_setflag(dev, attr,
837 BR_LEARNING);
838 break;
839 case IFLA_BRPORT_LEARNING_SYNC:
840 err = switchdev_port_br_setflag(dev, attr,
841 BR_LEARNING_SYNC);
842 break;
843 case IFLA_BRPORT_UNICAST_FLOOD:
844 err = switchdev_port_br_setflag(dev, attr, BR_FLOOD);
845 break;
846 default:
847 err = -EOPNOTSUPP;
848 break;
849 }
850 if (err)
851 return err;
852 }
853
854 return 0;
855}
856
857static int switchdev_port_br_afspec(struct net_device *dev,
858 struct nlattr *afspec,
859 int (*f)(struct net_device *dev,
860 const struct switchdev_obj *obj))
861{
862 struct nlattr *attr;
863 struct bridge_vlan_info *vinfo;
864 struct switchdev_obj_port_vlan vlan = {
865 .obj.orig_dev = dev,
866 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
867 };
868 int rem;
869 int err;
870
871 nla_for_each_nested(attr, afspec, rem) {
872 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
873 continue;
874 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
875 return -EINVAL;
876 vinfo = nla_data(attr);
877 if (!vinfo->vid || vinfo->vid >= VLAN_VID_MASK)
878 return -EINVAL;
879 vlan.flags = vinfo->flags;
880 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
881 if (vlan.vid_begin)
882 return -EINVAL;
883 vlan.vid_begin = vinfo->vid;
884 /* don't allow range of pvids */
885 if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
886 return -EINVAL;
887 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
888 if (!vlan.vid_begin)
889 return -EINVAL;
890 vlan.vid_end = vinfo->vid;
891 if (vlan.vid_end <= vlan.vid_begin)
892 return -EINVAL;
893 err = f(dev, &vlan.obj);
894 if (err)
895 return err;
896 vlan.vid_begin = 0;
897 } else {
898 if (vlan.vid_begin)
899 return -EINVAL;
900 vlan.vid_begin = vinfo->vid;
901 vlan.vid_end = vinfo->vid;
902 err = f(dev, &vlan.obj);
903 if (err)
904 return err;
905 vlan.vid_begin = 0;
906 }
907 }
908
909 return 0;
910}
911
912/**
913 * switchdev_port_bridge_setlink - Set bridge port attributes
914 *
915 * @dev: port device
916 * @nlh: netlink header
917 * @flags: netlink flags
918 *
919 * Called for SELF on rtnl_bridge_setlink to set bridge port
920 * attributes.
921 */
922int switchdev_port_bridge_setlink(struct net_device *dev,
923 struct nlmsghdr *nlh, u16 flags)
924{
925 struct nlattr *protinfo;
926 struct nlattr *afspec;
927 int err = 0;
928
929 if (!netif_is_bridge_port(dev))
930 return -EOPNOTSUPP;
931
932 protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
933 IFLA_PROTINFO);
934 if (protinfo) {
935 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
936 if (err)
937 return err;
938 }
939
940 afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
941 IFLA_AF_SPEC);
942 if (afspec)
943 err = switchdev_port_br_afspec(dev, afspec,
944 switchdev_port_obj_add);
945
946 return err;
947}
948EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
949
950/**
951 * switchdev_port_bridge_dellink - Set bridge port attributes
952 *
953 * @dev: port device
954 * @nlh: netlink header
955 * @flags: netlink flags
956 *
957 * Called for SELF on rtnl_bridge_dellink to set bridge port
958 * attributes.
959 */
960int switchdev_port_bridge_dellink(struct net_device *dev,
961 struct nlmsghdr *nlh, u16 flags)
962{
963 struct nlattr *afspec;
964
965 if (!netif_is_bridge_port(dev))
966 return -EOPNOTSUPP;
967
968 afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
969 IFLA_AF_SPEC);
970 if (afspec)
971 return switchdev_port_br_afspec(dev, afspec,
972 switchdev_port_obj_del);
973
974 return 0;
975}
976EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
977
978/**
979 * switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
980 *
981 * @ndmsg: netlink hdr
982 * @nlattr: netlink attributes
983 * @dev: port device
984 * @addr: MAC address to add
985 * @vid: VLAN to add
986 *
987 * Add FDB entry to switch device.
988 */
989int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
990 struct net_device *dev, const unsigned char *addr,
991 u16 vid, u16 nlm_flags)
992{
993 struct switchdev_obj_port_fdb fdb = {
994 .obj.orig_dev = dev,
995 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
996 .vid = vid,
997 };
998
999 ether_addr_copy(fdb.addr, addr);
1000 return switchdev_port_obj_add(dev, &fdb.obj);
1001}
1002EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
1003
1004/**
1005 * switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
1006 *
1007 * @ndmsg: netlink hdr
1008 * @nlattr: netlink attributes
1009 * @dev: port device
1010 * @addr: MAC address to delete
1011 * @vid: VLAN to delete
1012 *
1013 * Delete FDB entry from switch device.
1014 */
1015int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1016 struct net_device *dev, const unsigned char *addr,
1017 u16 vid)
1018{
1019 struct switchdev_obj_port_fdb fdb = {
1020 .obj.orig_dev = dev,
1021 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1022 .vid = vid,
1023 };
1024
1025 ether_addr_copy(fdb.addr, addr);
1026 return switchdev_port_obj_del(dev, &fdb.obj);
1027}
1028EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
1029
1030struct switchdev_fdb_dump {
1031 struct switchdev_obj_port_fdb fdb;
1032 struct net_device *dev;
1033 struct sk_buff *skb;
1034 struct netlink_callback *cb;
1035 int idx;
1036};
1037
1038static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
1039{
1040 struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
1041 struct switchdev_fdb_dump *dump =
1042 container_of(fdb, struct switchdev_fdb_dump, fdb);
1043 u32 portid = NETLINK_CB(dump->cb->skb).portid;
1044 u32 seq = dump->cb->nlh->nlmsg_seq;
1045 struct nlmsghdr *nlh;
1046 struct ndmsg *ndm;
1047
1048 if (dump->idx < dump->cb->args[2])
1049 goto skip;
1050
1051 nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
1052 sizeof(*ndm), NLM_F_MULTI);
1053 if (!nlh)
1054 return -EMSGSIZE;
1055
1056 ndm = nlmsg_data(nlh);
1057 ndm->ndm_family = AF_BRIDGE;
1058 ndm->ndm_pad1 = 0;
1059 ndm->ndm_pad2 = 0;
1060 ndm->ndm_flags = NTF_SELF;
1061 ndm->ndm_type = 0;
1062 ndm->ndm_ifindex = dump->dev->ifindex;
1063 ndm->ndm_state = fdb->ndm_state;
1064
1065 if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
1066 goto nla_put_failure;
1067
1068 if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
1069 goto nla_put_failure;
1070
1071 nlmsg_end(dump->skb, nlh);
1072
1073skip:
1074 dump->idx++;
1075 return 0;
1076
1077nla_put_failure:
1078 nlmsg_cancel(dump->skb, nlh);
1079 return -EMSGSIZE;
1080}
1081
1082/**
1083 * switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
1084 *
1085 * @skb: netlink skb
1086 * @cb: netlink callback
1087 * @dev: port device
1088 * @filter_dev: filter device
1089 * @idx:
1090 *
1091 * Dump FDB entries from switch device.
1092 */
1093int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1094 struct net_device *dev,
1095 struct net_device *filter_dev, int *idx)
1096{
1097 struct switchdev_fdb_dump dump = {
1098 .fdb.obj.orig_dev = dev,
1099 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1100 .dev = dev,
1101 .skb = skb,
1102 .cb = cb,
1103 .idx = *idx,
1104 };
1105 int err;
1106
1107 err = switchdev_port_obj_dump(dev, &dump.fdb.obj,
1108 switchdev_port_fdb_dump_cb);
1109 *idx = dump.idx;
1110 return err;
1111}
1112EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
1113
1114bool switchdev_port_same_parent_id(struct net_device *a,
1115 struct net_device *b)
1116{
1117 struct switchdev_attr a_attr = {
1118 .orig_dev = a,
1119 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1120 };
1121 struct switchdev_attr b_attr = {
1122 .orig_dev = b,
1123 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1124 };
1125
1126 if (switchdev_port_attr_get(a, &a_attr) ||
1127 switchdev_port_attr_get(b, &b_attr))
1128 return false;
1129
1130 return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1131}
1132EXPORT_SYMBOL_GPL(switchdev_port_same_parent_id);
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
22static LIST_HEAD(deferred);
23static DEFINE_SPINLOCK(deferred_lock);
24
25typedef void switchdev_deferred_func_t(struct net_device *dev,
26 const void *data);
27
28struct switchdev_deferred_item {
29 struct list_head list;
30 struct net_device *dev;
31 netdevice_tracker dev_tracker;
32 switchdev_deferred_func_t *func;
33 unsigned long data[];
34};
35
36static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
37{
38 struct switchdev_deferred_item *dfitem;
39
40 spin_lock_bh(&deferred_lock);
41 if (list_empty(&deferred)) {
42 dfitem = NULL;
43 goto unlock;
44 }
45 dfitem = list_first_entry(&deferred,
46 struct switchdev_deferred_item, list);
47 list_del(&dfitem->list);
48unlock:
49 spin_unlock_bh(&deferred_lock);
50 return dfitem;
51}
52
53/**
54 * switchdev_deferred_process - Process ops in deferred queue
55 *
56 * Called to flush the ops currently queued in deferred ops queue.
57 * rtnl_lock must be held.
58 */
59void switchdev_deferred_process(void)
60{
61 struct switchdev_deferred_item *dfitem;
62
63 ASSERT_RTNL();
64
65 while ((dfitem = switchdev_deferred_dequeue())) {
66 dfitem->func(dfitem->dev, dfitem->data);
67 netdev_put(dfitem->dev, &dfitem->dev_tracker);
68 kfree(dfitem);
69 }
70}
71EXPORT_SYMBOL_GPL(switchdev_deferred_process);
72
73static void switchdev_deferred_process_work(struct work_struct *work)
74{
75 rtnl_lock();
76 switchdev_deferred_process();
77 rtnl_unlock();
78}
79
80static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
81
82static int switchdev_deferred_enqueue(struct net_device *dev,
83 const void *data, size_t data_len,
84 switchdev_deferred_func_t *func)
85{
86 struct switchdev_deferred_item *dfitem;
87
88 dfitem = kmalloc(struct_size(dfitem, data, data_len), GFP_ATOMIC);
89 if (!dfitem)
90 return -ENOMEM;
91 dfitem->dev = dev;
92 dfitem->func = func;
93 memcpy(dfitem->data, data, data_len);
94 netdev_hold(dev, &dfitem->dev_tracker, GFP_ATOMIC);
95 spin_lock_bh(&deferred_lock);
96 list_add_tail(&dfitem->list, &deferred);
97 spin_unlock_bh(&deferred_lock);
98 schedule_work(&deferred_process_work);
99 return 0;
100}
101
102static int switchdev_port_attr_notify(enum switchdev_notifier_type nt,
103 struct net_device *dev,
104 const struct switchdev_attr *attr,
105 struct netlink_ext_ack *extack)
106{
107 int err;
108 int rc;
109
110 struct switchdev_notifier_port_attr_info attr_info = {
111 .attr = attr,
112 .handled = false,
113 };
114
115 rc = call_switchdev_blocking_notifiers(nt, dev,
116 &attr_info.info, extack);
117 err = notifier_to_errno(rc);
118 if (err) {
119 WARN_ON(!attr_info.handled);
120 return err;
121 }
122
123 if (!attr_info.handled)
124 return -EOPNOTSUPP;
125
126 return 0;
127}
128
129static int switchdev_port_attr_set_now(struct net_device *dev,
130 const struct switchdev_attr *attr,
131 struct netlink_ext_ack *extack)
132{
133 return switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
134 extack);
135}
136
137static void switchdev_port_attr_set_deferred(struct net_device *dev,
138 const void *data)
139{
140 const struct switchdev_attr *attr = data;
141 int err;
142
143 err = switchdev_port_attr_set_now(dev, attr, NULL);
144 if (err && err != -EOPNOTSUPP)
145 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
146 err, attr->id);
147 if (attr->complete)
148 attr->complete(dev, err, attr->complete_priv);
149}
150
151static int switchdev_port_attr_set_defer(struct net_device *dev,
152 const struct switchdev_attr *attr)
153{
154 return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
155 switchdev_port_attr_set_deferred);
156}
157
158/**
159 * switchdev_port_attr_set - Set port attribute
160 *
161 * @dev: port device
162 * @attr: attribute to set
163 * @extack: netlink extended ack, for error message propagation
164 *
165 * rtnl_lock must be held and must not be in atomic section,
166 * in case SWITCHDEV_F_DEFER flag is not set.
167 */
168int switchdev_port_attr_set(struct net_device *dev,
169 const struct switchdev_attr *attr,
170 struct netlink_ext_ack *extack)
171{
172 if (attr->flags & SWITCHDEV_F_DEFER)
173 return switchdev_port_attr_set_defer(dev, attr);
174 ASSERT_RTNL();
175 return switchdev_port_attr_set_now(dev, attr, extack);
176}
177EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
178
179static size_t switchdev_obj_size(const struct switchdev_obj *obj)
180{
181 switch (obj->id) {
182 case SWITCHDEV_OBJ_ID_PORT_VLAN:
183 return sizeof(struct switchdev_obj_port_vlan);
184 case SWITCHDEV_OBJ_ID_PORT_MDB:
185 return sizeof(struct switchdev_obj_port_mdb);
186 case SWITCHDEV_OBJ_ID_HOST_MDB:
187 return sizeof(struct switchdev_obj_port_mdb);
188 default:
189 BUG();
190 }
191 return 0;
192}
193
194static int switchdev_port_obj_notify(enum switchdev_notifier_type nt,
195 struct net_device *dev,
196 const struct switchdev_obj *obj,
197 struct netlink_ext_ack *extack)
198{
199 int rc;
200 int err;
201
202 struct switchdev_notifier_port_obj_info obj_info = {
203 .obj = obj,
204 .handled = false,
205 };
206
207 rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack);
208 err = notifier_to_errno(rc);
209 if (err) {
210 WARN_ON(!obj_info.handled);
211 return err;
212 }
213 if (!obj_info.handled)
214 return -EOPNOTSUPP;
215 return 0;
216}
217
218static void switchdev_port_obj_add_deferred(struct net_device *dev,
219 const void *data)
220{
221 const struct switchdev_obj *obj = data;
222 int err;
223
224 ASSERT_RTNL();
225 err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
226 dev, obj, NULL);
227 if (err && err != -EOPNOTSUPP)
228 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
229 err, obj->id);
230 if (obj->complete)
231 obj->complete(dev, err, obj->complete_priv);
232}
233
234static int switchdev_port_obj_add_defer(struct net_device *dev,
235 const struct switchdev_obj *obj)
236{
237 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
238 switchdev_port_obj_add_deferred);
239}
240
241/**
242 * switchdev_port_obj_add - Add port object
243 *
244 * @dev: port device
245 * @obj: object to add
246 * @extack: netlink extended ack
247 *
248 * rtnl_lock must be held and must not be in atomic section,
249 * in case SWITCHDEV_F_DEFER flag is not set.
250 */
251int switchdev_port_obj_add(struct net_device *dev,
252 const struct switchdev_obj *obj,
253 struct netlink_ext_ack *extack)
254{
255 if (obj->flags & SWITCHDEV_F_DEFER)
256 return switchdev_port_obj_add_defer(dev, obj);
257 ASSERT_RTNL();
258 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
259 dev, obj, extack);
260}
261EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
262
263static int switchdev_port_obj_del_now(struct net_device *dev,
264 const struct switchdev_obj *obj)
265{
266 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL,
267 dev, obj, NULL);
268}
269
270static void switchdev_port_obj_del_deferred(struct net_device *dev,
271 const void *data)
272{
273 const struct switchdev_obj *obj = data;
274 int err;
275
276 err = switchdev_port_obj_del_now(dev, obj);
277 if (err && err != -EOPNOTSUPP)
278 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
279 err, obj->id);
280 if (obj->complete)
281 obj->complete(dev, err, obj->complete_priv);
282}
283
284static int switchdev_port_obj_del_defer(struct net_device *dev,
285 const struct switchdev_obj *obj)
286{
287 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
288 switchdev_port_obj_del_deferred);
289}
290
291/**
292 * switchdev_port_obj_del - Delete port object
293 *
294 * @dev: port device
295 * @obj: object to delete
296 *
297 * rtnl_lock must be held and must not be in atomic section,
298 * in case SWITCHDEV_F_DEFER flag is not set.
299 */
300int switchdev_port_obj_del(struct net_device *dev,
301 const struct switchdev_obj *obj)
302{
303 if (obj->flags & SWITCHDEV_F_DEFER)
304 return switchdev_port_obj_del_defer(dev, obj);
305 ASSERT_RTNL();
306 return switchdev_port_obj_del_now(dev, obj);
307}
308EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
309
310static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
311static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
312
313/**
314 * register_switchdev_notifier - Register notifier
315 * @nb: notifier_block
316 *
317 * Register switch device notifier.
318 */
319int register_switchdev_notifier(struct notifier_block *nb)
320{
321 return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
322}
323EXPORT_SYMBOL_GPL(register_switchdev_notifier);
324
325/**
326 * unregister_switchdev_notifier - Unregister notifier
327 * @nb: notifier_block
328 *
329 * Unregister switch device notifier.
330 */
331int unregister_switchdev_notifier(struct notifier_block *nb)
332{
333 return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
334}
335EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
336
337/**
338 * call_switchdev_notifiers - Call notifiers
339 * @val: value passed unmodified to notifier function
340 * @dev: port device
341 * @info: notifier information data
342 * @extack: netlink extended ack
343 * Call all network notifier blocks.
344 */
345int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
346 struct switchdev_notifier_info *info,
347 struct netlink_ext_ack *extack)
348{
349 info->dev = dev;
350 info->extack = extack;
351 return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
352}
353EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
354
355int register_switchdev_blocking_notifier(struct notifier_block *nb)
356{
357 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
358
359 return blocking_notifier_chain_register(chain, nb);
360}
361EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
362
363int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
364{
365 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
366
367 return blocking_notifier_chain_unregister(chain, nb);
368}
369EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
370
371int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
372 struct switchdev_notifier_info *info,
373 struct netlink_ext_ack *extack)
374{
375 info->dev = dev;
376 info->extack = extack;
377 return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
378 val, info);
379}
380EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
381
382struct switchdev_nested_priv {
383 bool (*check_cb)(const struct net_device *dev);
384 bool (*foreign_dev_check_cb)(const struct net_device *dev,
385 const struct net_device *foreign_dev);
386 const struct net_device *dev;
387 struct net_device *lower_dev;
388};
389
390static int switchdev_lower_dev_walk(struct net_device *lower_dev,
391 struct netdev_nested_priv *priv)
392{
393 struct switchdev_nested_priv *switchdev_priv = priv->data;
394 bool (*foreign_dev_check_cb)(const struct net_device *dev,
395 const struct net_device *foreign_dev);
396 bool (*check_cb)(const struct net_device *dev);
397 const struct net_device *dev;
398
399 check_cb = switchdev_priv->check_cb;
400 foreign_dev_check_cb = switchdev_priv->foreign_dev_check_cb;
401 dev = switchdev_priv->dev;
402
403 if (check_cb(lower_dev) && !foreign_dev_check_cb(lower_dev, dev)) {
404 switchdev_priv->lower_dev = lower_dev;
405 return 1;
406 }
407
408 return 0;
409}
410
411static struct net_device *
412switchdev_lower_dev_find_rcu(struct net_device *dev,
413 bool (*check_cb)(const struct net_device *dev),
414 bool (*foreign_dev_check_cb)(const struct net_device *dev,
415 const struct net_device *foreign_dev))
416{
417 struct switchdev_nested_priv switchdev_priv = {
418 .check_cb = check_cb,
419 .foreign_dev_check_cb = foreign_dev_check_cb,
420 .dev = dev,
421 .lower_dev = NULL,
422 };
423 struct netdev_nested_priv priv = {
424 .data = &switchdev_priv,
425 };
426
427 netdev_walk_all_lower_dev_rcu(dev, switchdev_lower_dev_walk, &priv);
428
429 return switchdev_priv.lower_dev;
430}
431
432static struct net_device *
433switchdev_lower_dev_find(struct net_device *dev,
434 bool (*check_cb)(const struct net_device *dev),
435 bool (*foreign_dev_check_cb)(const struct net_device *dev,
436 const struct net_device *foreign_dev))
437{
438 struct switchdev_nested_priv switchdev_priv = {
439 .check_cb = check_cb,
440 .foreign_dev_check_cb = foreign_dev_check_cb,
441 .dev = dev,
442 .lower_dev = NULL,
443 };
444 struct netdev_nested_priv priv = {
445 .data = &switchdev_priv,
446 };
447
448 netdev_walk_all_lower_dev(dev, switchdev_lower_dev_walk, &priv);
449
450 return switchdev_priv.lower_dev;
451}
452
453static int __switchdev_handle_fdb_event_to_device(struct net_device *dev,
454 struct net_device *orig_dev, unsigned long event,
455 const struct switchdev_notifier_fdb_info *fdb_info,
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 int (*mod_cb)(struct net_device *dev, struct net_device *orig_dev,
460 unsigned long event, const void *ctx,
461 const struct switchdev_notifier_fdb_info *fdb_info))
462{
463 const struct switchdev_notifier_info *info = &fdb_info->info;
464 struct net_device *br, *lower_dev, *switchdev;
465 struct list_head *iter;
466 int err = -EOPNOTSUPP;
467
468 if (check_cb(dev))
469 return mod_cb(dev, orig_dev, event, info->ctx, fdb_info);
470
471 /* Recurse through lower interfaces in case the FDB entry is pointing
472 * towards a bridge or a LAG device.
473 */
474 netdev_for_each_lower_dev(dev, lower_dev, iter) {
475 /* Do not propagate FDB entries across bridges */
476 if (netif_is_bridge_master(lower_dev))
477 continue;
478
479 /* Bridge ports might be either us, or LAG interfaces
480 * that we offload.
481 */
482 if (!check_cb(lower_dev) &&
483 !switchdev_lower_dev_find_rcu(lower_dev, check_cb,
484 foreign_dev_check_cb))
485 continue;
486
487 err = __switchdev_handle_fdb_event_to_device(lower_dev, orig_dev,
488 event, fdb_info, check_cb,
489 foreign_dev_check_cb,
490 mod_cb);
491 if (err && err != -EOPNOTSUPP)
492 return err;
493 }
494
495 /* Event is neither on a bridge nor a LAG. Check whether it is on an
496 * interface that is in a bridge with us.
497 */
498 br = netdev_master_upper_dev_get_rcu(dev);
499 if (!br || !netif_is_bridge_master(br))
500 return 0;
501
502 switchdev = switchdev_lower_dev_find_rcu(br, check_cb, foreign_dev_check_cb);
503 if (!switchdev)
504 return 0;
505
506 if (!foreign_dev_check_cb(switchdev, dev))
507 return err;
508
509 return __switchdev_handle_fdb_event_to_device(br, orig_dev, event, fdb_info,
510 check_cb, foreign_dev_check_cb,
511 mod_cb);
512}
513
514int switchdev_handle_fdb_event_to_device(struct net_device *dev, unsigned long event,
515 const struct switchdev_notifier_fdb_info *fdb_info,
516 bool (*check_cb)(const struct net_device *dev),
517 bool (*foreign_dev_check_cb)(const struct net_device *dev,
518 const struct net_device *foreign_dev),
519 int (*mod_cb)(struct net_device *dev, struct net_device *orig_dev,
520 unsigned long event, const void *ctx,
521 const struct switchdev_notifier_fdb_info *fdb_info))
522{
523 int err;
524
525 err = __switchdev_handle_fdb_event_to_device(dev, dev, event, fdb_info,
526 check_cb, foreign_dev_check_cb,
527 mod_cb);
528 if (err == -EOPNOTSUPP)
529 err = 0;
530
531 return err;
532}
533EXPORT_SYMBOL_GPL(switchdev_handle_fdb_event_to_device);
534
535static int __switchdev_handle_port_obj_add(struct net_device *dev,
536 struct switchdev_notifier_port_obj_info *port_obj_info,
537 bool (*check_cb)(const struct net_device *dev),
538 bool (*foreign_dev_check_cb)(const struct net_device *dev,
539 const struct net_device *foreign_dev),
540 int (*add_cb)(struct net_device *dev, const void *ctx,
541 const struct switchdev_obj *obj,
542 struct netlink_ext_ack *extack))
543{
544 struct switchdev_notifier_info *info = &port_obj_info->info;
545 struct net_device *br, *lower_dev, *switchdev;
546 struct netlink_ext_ack *extack;
547 struct list_head *iter;
548 int err = -EOPNOTSUPP;
549
550 extack = switchdev_notifier_info_to_extack(info);
551
552 if (check_cb(dev)) {
553 err = add_cb(dev, info->ctx, port_obj_info->obj, extack);
554 if (err != -EOPNOTSUPP)
555 port_obj_info->handled = true;
556 return err;
557 }
558
559 /* Switch ports might be stacked under e.g. a LAG. Ignore the
560 * unsupported devices, another driver might be able to handle them. But
561 * propagate to the callers any hard errors.
562 *
563 * If the driver does its own bookkeeping of stacked ports, it's not
564 * necessary to go through this helper.
565 */
566 netdev_for_each_lower_dev(dev, lower_dev, iter) {
567 if (netif_is_bridge_master(lower_dev))
568 continue;
569
570 /* When searching for switchdev interfaces that are neighbors
571 * of foreign ones, and @dev is a bridge, do not recurse on the
572 * foreign interface again, it was already visited.
573 */
574 if (foreign_dev_check_cb && !check_cb(lower_dev) &&
575 !switchdev_lower_dev_find(lower_dev, check_cb, foreign_dev_check_cb))
576 continue;
577
578 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
579 check_cb, foreign_dev_check_cb,
580 add_cb);
581 if (err && err != -EOPNOTSUPP)
582 return err;
583 }
584
585 /* Event is neither on a bridge nor a LAG. Check whether it is on an
586 * interface that is in a bridge with us.
587 */
588 if (!foreign_dev_check_cb)
589 return err;
590
591 br = netdev_master_upper_dev_get(dev);
592 if (!br || !netif_is_bridge_master(br))
593 return err;
594
595 switchdev = switchdev_lower_dev_find(br, check_cb, foreign_dev_check_cb);
596 if (!switchdev)
597 return err;
598
599 if (!foreign_dev_check_cb(switchdev, dev))
600 return err;
601
602 return __switchdev_handle_port_obj_add(br, port_obj_info, check_cb,
603 foreign_dev_check_cb, add_cb);
604}
605
606/* Pass through a port object addition, if @dev passes @check_cb, or replicate
607 * it towards all lower interfaces of @dev that pass @check_cb, if @dev is a
608 * bridge or a LAG.
609 */
610int switchdev_handle_port_obj_add(struct net_device *dev,
611 struct switchdev_notifier_port_obj_info *port_obj_info,
612 bool (*check_cb)(const struct net_device *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 int err;
618
619 err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
620 NULL, add_cb);
621 if (err == -EOPNOTSUPP)
622 err = 0;
623 return err;
624}
625EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
626
627/* Same as switchdev_handle_port_obj_add(), except if object is notified on a
628 * @dev that passes @foreign_dev_check_cb, it is replicated towards all devices
629 * that pass @check_cb and are in the same bridge as @dev.
630 */
631int switchdev_handle_port_obj_add_foreign(struct net_device *dev,
632 struct switchdev_notifier_port_obj_info *port_obj_info,
633 bool (*check_cb)(const struct net_device *dev),
634 bool (*foreign_dev_check_cb)(const struct net_device *dev,
635 const struct net_device *foreign_dev),
636 int (*add_cb)(struct net_device *dev, const void *ctx,
637 const struct switchdev_obj *obj,
638 struct netlink_ext_ack *extack))
639{
640 int err;
641
642 err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
643 foreign_dev_check_cb, add_cb);
644 if (err == -EOPNOTSUPP)
645 err = 0;
646 return err;
647}
648EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add_foreign);
649
650static int __switchdev_handle_port_obj_del(struct net_device *dev,
651 struct switchdev_notifier_port_obj_info *port_obj_info,
652 bool (*check_cb)(const struct net_device *dev),
653 bool (*foreign_dev_check_cb)(const struct net_device *dev,
654 const struct net_device *foreign_dev),
655 int (*del_cb)(struct net_device *dev, const void *ctx,
656 const struct switchdev_obj *obj))
657{
658 struct switchdev_notifier_info *info = &port_obj_info->info;
659 struct net_device *br, *lower_dev, *switchdev;
660 struct list_head *iter;
661 int err = -EOPNOTSUPP;
662
663 if (check_cb(dev)) {
664 err = del_cb(dev, info->ctx, port_obj_info->obj);
665 if (err != -EOPNOTSUPP)
666 port_obj_info->handled = true;
667 return err;
668 }
669
670 /* Switch ports might be stacked under e.g. a LAG. Ignore the
671 * unsupported devices, another driver might be able to handle them. But
672 * propagate to the callers any hard errors.
673 *
674 * If the driver does its own bookkeeping of stacked ports, it's not
675 * necessary to go through this helper.
676 */
677 netdev_for_each_lower_dev(dev, lower_dev, iter) {
678 if (netif_is_bridge_master(lower_dev))
679 continue;
680
681 /* When searching for switchdev interfaces that are neighbors
682 * of foreign ones, and @dev is a bridge, do not recurse on the
683 * foreign interface again, it was already visited.
684 */
685 if (foreign_dev_check_cb && !check_cb(lower_dev) &&
686 !switchdev_lower_dev_find(lower_dev, check_cb, foreign_dev_check_cb))
687 continue;
688
689 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
690 check_cb, foreign_dev_check_cb,
691 del_cb);
692 if (err && err != -EOPNOTSUPP)
693 return err;
694 }
695
696 /* Event is neither on a bridge nor a LAG. Check whether it is on an
697 * interface that is in a bridge with us.
698 */
699 if (!foreign_dev_check_cb)
700 return err;
701
702 br = netdev_master_upper_dev_get(dev);
703 if (!br || !netif_is_bridge_master(br))
704 return err;
705
706 switchdev = switchdev_lower_dev_find(br, check_cb, foreign_dev_check_cb);
707 if (!switchdev)
708 return err;
709
710 if (!foreign_dev_check_cb(switchdev, dev))
711 return err;
712
713 return __switchdev_handle_port_obj_del(br, port_obj_info, check_cb,
714 foreign_dev_check_cb, del_cb);
715}
716
717/* Pass through a port object deletion, if @dev passes @check_cb, or replicate
718 * it towards all lower interfaces of @dev that pass @check_cb, if @dev is a
719 * bridge or a LAG.
720 */
721int switchdev_handle_port_obj_del(struct net_device *dev,
722 struct switchdev_notifier_port_obj_info *port_obj_info,
723 bool (*check_cb)(const struct net_device *dev),
724 int (*del_cb)(struct net_device *dev, const void *ctx,
725 const struct switchdev_obj *obj))
726{
727 int err;
728
729 err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
730 NULL, del_cb);
731 if (err == -EOPNOTSUPP)
732 err = 0;
733 return err;
734}
735EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
736
737/* Same as switchdev_handle_port_obj_del(), except if object is notified on a
738 * @dev that passes @foreign_dev_check_cb, it is replicated towards all devices
739 * that pass @check_cb and are in the same bridge as @dev.
740 */
741int switchdev_handle_port_obj_del_foreign(struct net_device *dev,
742 struct switchdev_notifier_port_obj_info *port_obj_info,
743 bool (*check_cb)(const struct net_device *dev),
744 bool (*foreign_dev_check_cb)(const struct net_device *dev,
745 const struct net_device *foreign_dev),
746 int (*del_cb)(struct net_device *dev, const void *ctx,
747 const struct switchdev_obj *obj))
748{
749 int err;
750
751 err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
752 foreign_dev_check_cb, del_cb);
753 if (err == -EOPNOTSUPP)
754 err = 0;
755 return err;
756}
757EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del_foreign);
758
759static int __switchdev_handle_port_attr_set(struct net_device *dev,
760 struct switchdev_notifier_port_attr_info *port_attr_info,
761 bool (*check_cb)(const struct net_device *dev),
762 int (*set_cb)(struct net_device *dev, const void *ctx,
763 const struct switchdev_attr *attr,
764 struct netlink_ext_ack *extack))
765{
766 struct switchdev_notifier_info *info = &port_attr_info->info;
767 struct netlink_ext_ack *extack;
768 struct net_device *lower_dev;
769 struct list_head *iter;
770 int err = -EOPNOTSUPP;
771
772 extack = switchdev_notifier_info_to_extack(info);
773
774 if (check_cb(dev)) {
775 err = set_cb(dev, info->ctx, port_attr_info->attr, extack);
776 if (err != -EOPNOTSUPP)
777 port_attr_info->handled = true;
778 return err;
779 }
780
781 /* Switch ports might be stacked under e.g. a LAG. Ignore the
782 * unsupported devices, another driver might be able to handle them. But
783 * propagate to the callers any hard errors.
784 *
785 * If the driver does its own bookkeeping of stacked ports, it's not
786 * necessary to go through this helper.
787 */
788 netdev_for_each_lower_dev(dev, lower_dev, iter) {
789 if (netif_is_bridge_master(lower_dev))
790 continue;
791
792 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
793 check_cb, set_cb);
794 if (err && err != -EOPNOTSUPP)
795 return err;
796 }
797
798 return err;
799}
800
801int switchdev_handle_port_attr_set(struct net_device *dev,
802 struct switchdev_notifier_port_attr_info *port_attr_info,
803 bool (*check_cb)(const struct net_device *dev),
804 int (*set_cb)(struct net_device *dev, const void *ctx,
805 const struct switchdev_attr *attr,
806 struct netlink_ext_ack *extack))
807{
808 int err;
809
810 err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
811 set_cb);
812 if (err == -EOPNOTSUPP)
813 err = 0;
814 return err;
815}
816EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);
817
818int switchdev_bridge_port_offload(struct net_device *brport_dev,
819 struct net_device *dev, const void *ctx,
820 struct notifier_block *atomic_nb,
821 struct notifier_block *blocking_nb,
822 bool tx_fwd_offload,
823 struct netlink_ext_ack *extack)
824{
825 struct switchdev_notifier_brport_info brport_info = {
826 .brport = {
827 .dev = dev,
828 .ctx = ctx,
829 .atomic_nb = atomic_nb,
830 .blocking_nb = blocking_nb,
831 .tx_fwd_offload = tx_fwd_offload,
832 },
833 };
834 int err;
835
836 ASSERT_RTNL();
837
838 err = call_switchdev_blocking_notifiers(SWITCHDEV_BRPORT_OFFLOADED,
839 brport_dev, &brport_info.info,
840 extack);
841 return notifier_to_errno(err);
842}
843EXPORT_SYMBOL_GPL(switchdev_bridge_port_offload);
844
845void switchdev_bridge_port_unoffload(struct net_device *brport_dev,
846 const void *ctx,
847 struct notifier_block *atomic_nb,
848 struct notifier_block *blocking_nb)
849{
850 struct switchdev_notifier_brport_info brport_info = {
851 .brport = {
852 .ctx = ctx,
853 .atomic_nb = atomic_nb,
854 .blocking_nb = blocking_nb,
855 },
856 };
857
858 ASSERT_RTNL();
859
860 call_switchdev_blocking_notifiers(SWITCHDEV_BRPORT_UNOFFLOADED,
861 brport_dev, &brport_info.info,
862 NULL);
863}
864EXPORT_SYMBOL_GPL(switchdev_bridge_port_unoffload);