projects/linux·blob v4.6 · refs

1395 lines · 34.8 KiB

1/*
2 * net/switchdev/switchdev.c - Switch device API
3 * Copyright (c) 2014-2015 Jiri Pirko <[email protected]>
4 * Copyright (c) 2014-2015 Scott Feldman <[email protected]>
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/ip_fib.h>
25#include <net/switchdev.h>
26
27/**
28 * switchdev_trans_item_enqueue - Enqueue data item to transaction queue
29 *
30 * @trans: transaction
31 * @data: pointer to data being queued
32 * @destructor: data destructor
33 * @tritem: transaction item being queued
34 *
35 * Enqeueue data item to transaction queue. tritem is typically placed in
36 * cointainter pointed at by data pointer. Destructor is called on
37 * transaction abort and after successful commit phase in case
38 * the caller did not dequeue the item before.
39 */
40void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
41 void *data, void (*destructor)(void const *),
42 struct switchdev_trans_item *tritem)
43{
44 tritem->data = data;
45 tritem->destructor = destructor;
46 list_add_tail(&tritem->list, &trans->item_list);
47}
48EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
49
50static struct switchdev_trans_item *
51__switchdev_trans_item_dequeue(struct switchdev_trans *trans)
52{
53 struct switchdev_trans_item *tritem;
54
55 if (list_empty(&trans->item_list))
56 return NULL;
57 tritem = list_first_entry(&trans->item_list,
58 struct switchdev_trans_item, list);
59 list_del(&tritem->list);
60 return tritem;
61}
62
63/**
64 * switchdev_trans_item_dequeue - Dequeue data item from transaction queue
65 *
66 * @trans: transaction
67 */
68void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
69{
70 struct switchdev_trans_item *tritem;
71
72 tritem = __switchdev_trans_item_dequeue(trans);
73 BUG_ON(!tritem);
74 return tritem->data;
75}
76EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
77
78static void switchdev_trans_init(struct switchdev_trans *trans)
79{
80 INIT_LIST_HEAD(&trans->item_list);
81}
82
83static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
84{
85 struct switchdev_trans_item *tritem;
86
87 while ((tritem = __switchdev_trans_item_dequeue(trans)))
88 tritem->destructor(tritem->data);
89}
90
91static void switchdev_trans_items_warn_destroy(struct net_device *dev,
92 struct switchdev_trans *trans)
93{
94 WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
95 dev->name);
96 switchdev_trans_items_destroy(trans);
97}
98
99static LIST_HEAD(deferred);
100static DEFINE_SPINLOCK(deferred_lock);
101
102typedef void switchdev_deferred_func_t(struct net_device *dev,
103 const void *data);
104
105struct switchdev_deferred_item {
106 struct list_head list;
107 struct net_device *dev;
108 switchdev_deferred_func_t *func;
109 unsigned long data[0];
110};
111
112static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
113{
114 struct switchdev_deferred_item *dfitem;
115
116 spin_lock_bh(&deferred_lock);
117 if (list_empty(&deferred)) {
118 dfitem = NULL;
119 goto unlock;
120 }
121 dfitem = list_first_entry(&deferred,
122 struct switchdev_deferred_item, list);
123 list_del(&dfitem->list);
124unlock:
125 spin_unlock_bh(&deferred_lock);
126 return dfitem;
127}
128
129/**
130 * switchdev_deferred_process - Process ops in deferred queue
131 *
132 * Called to flush the ops currently queued in deferred ops queue.
133 * rtnl_lock must be held.
134 */
135void switchdev_deferred_process(void)
136{
137 struct switchdev_deferred_item *dfitem;
138
139 ASSERT_RTNL();
140
141 while ((dfitem = switchdev_deferred_dequeue())) {
142 dfitem->func(dfitem->dev, dfitem->data);
143 dev_put(dfitem->dev);
144 kfree(dfitem);
145 }
146}
147EXPORT_SYMBOL_GPL(switchdev_deferred_process);
148
149static void switchdev_deferred_process_work(struct work_struct *work)
150{
151 rtnl_lock();
152 switchdev_deferred_process();
153 rtnl_unlock();
154}
155
156static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
157
158static int switchdev_deferred_enqueue(struct net_device *dev,
159 const void *data, size_t data_len,
160 switchdev_deferred_func_t *func)
161{
162 struct switchdev_deferred_item *dfitem;
163
164 dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
165 if (!dfitem)
166 return -ENOMEM;
167 dfitem->dev = dev;
168 dfitem->func = func;
169 memcpy(dfitem->data, data, data_len);
170 dev_hold(dev);
171 spin_lock_bh(&deferred_lock);
172 list_add_tail(&dfitem->list, &deferred);
173 spin_unlock_bh(&deferred_lock);
174 schedule_work(&deferred_process_work);
175 return 0;
176}
177
178/**
179 * switchdev_port_attr_get - Get port attribute
180 *
181 * @dev: port device
182 * @attr: attribute to get
183 */
184int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
185{
186 const struct switchdev_ops *ops = dev->switchdev_ops;
187 struct net_device *lower_dev;
188 struct list_head *iter;
189 struct switchdev_attr first = {
190 .id = SWITCHDEV_ATTR_ID_UNDEFINED
191 };
192 int err = -EOPNOTSUPP;
193
194 if (ops && ops->switchdev_port_attr_get)
195 return ops->switchdev_port_attr_get(dev, attr);
196
197 if (attr->flags & SWITCHDEV_F_NO_RECURSE)
198 return err;
199
200 /* Switch device port(s) may be stacked under
201 * bond/team/vlan dev, so recurse down to get attr on
202 * each port. Return -ENODATA if attr values don't
203 * compare across ports.
204 */
205
206 netdev_for_each_lower_dev(dev, lower_dev, iter) {
207 err = switchdev_port_attr_get(lower_dev, attr);
208 if (err)
209 break;
210 if (first.id == SWITCHDEV_ATTR_ID_UNDEFINED)
211 first = *attr;
212 else if (memcmp(&first, attr, sizeof(*attr)))
213 return -ENODATA;
214 }
215
216 return err;
217}
218EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
219
220static int __switchdev_port_attr_set(struct net_device *dev,
221 const struct switchdev_attr *attr,
222 struct switchdev_trans *trans)
223{
224 const struct switchdev_ops *ops = dev->switchdev_ops;
225 struct net_device *lower_dev;
226 struct list_head *iter;
227 int err = -EOPNOTSUPP;
228
229 if (ops && ops->switchdev_port_attr_set) {
230 err = ops->switchdev_port_attr_set(dev, attr, trans);
231 goto done;
232 }
233
234 if (attr->flags & SWITCHDEV_F_NO_RECURSE)
235 goto done;
236
237 /* Switch device port(s) may be stacked under
238 * bond/team/vlan dev, so recurse down to set attr on
239 * each port.
240 */
241
242 netdev_for_each_lower_dev(dev, lower_dev, iter) {
243 err = __switchdev_port_attr_set(lower_dev, attr, trans);
244 if (err)
245 break;
246 }
247
248done:
249 if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
250 err = 0;
251
252 return err;
253}
254
255static int switchdev_port_attr_set_now(struct net_device *dev,
256 const struct switchdev_attr *attr)
257{
258 struct switchdev_trans trans;
259 int err;
260
261 switchdev_trans_init(&trans);
262
263 /* Phase I: prepare for attr set. Driver/device should fail
264 * here if there are going to be issues in the commit phase,
265 * such as lack of resources or support. The driver/device
266 * should reserve resources needed for the commit phase here,
267 * but should not commit the attr.
268 */
269
270 trans.ph_prepare = true;
271 err = __switchdev_port_attr_set(dev, attr, &trans);
272 if (err) {
273 /* Prepare phase failed: abort the transaction. Any
274 * resources reserved in the prepare phase are
275 * released.
276 */
277
278 if (err != -EOPNOTSUPP)
279 switchdev_trans_items_destroy(&trans);
280
281 return err;
282 }
283
284 /* Phase II: commit attr set. This cannot fail as a fault
285 * of driver/device. If it does, it's a bug in the driver/device
286 * because the driver said everythings was OK in phase I.
287 */
288
289 trans.ph_prepare = false;
290 err = __switchdev_port_attr_set(dev, attr, &trans);
291 WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
292 dev->name, attr->id);
293 switchdev_trans_items_warn_destroy(dev, &trans);
294
295 return err;
296}
297
298static void switchdev_port_attr_set_deferred(struct net_device *dev,
299 const void *data)
300{
301 const struct switchdev_attr *attr = data;
302 int err;
303
304 err = switchdev_port_attr_set_now(dev, attr);
305 if (err && err != -EOPNOTSUPP)
306 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
307 err, attr->id);
308 if (attr->complete)
309 attr->complete(dev, err, attr->complete_priv);
310}
311
312static int switchdev_port_attr_set_defer(struct net_device *dev,
313 const struct switchdev_attr *attr)
314{
315 return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
316 switchdev_port_attr_set_deferred);
317}
318
319/**
320 * switchdev_port_attr_set - Set port attribute
321 *
322 * @dev: port device
323 * @attr: attribute to set
324 *
325 * Use a 2-phase prepare-commit transaction model to ensure
326 * system is not left in a partially updated state due to
327 * failure from driver/device.
328 *
329 * rtnl_lock must be held and must not be in atomic section,
330 * in case SWITCHDEV_F_DEFER flag is not set.
331 */
332int switchdev_port_attr_set(struct net_device *dev,
333 const struct switchdev_attr *attr)
334{
335 if (attr->flags & SWITCHDEV_F_DEFER)
336 return switchdev_port_attr_set_defer(dev, attr);
337 ASSERT_RTNL();
338 return switchdev_port_attr_set_now(dev, attr);
339}
340EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
341
342static size_t switchdev_obj_size(const struct switchdev_obj *obj)
343{
344 switch (obj->id) {
345 case SWITCHDEV_OBJ_ID_PORT_VLAN:
346 return sizeof(struct switchdev_obj_port_vlan);
347 case SWITCHDEV_OBJ_ID_IPV4_FIB:
348 return sizeof(struct switchdev_obj_ipv4_fib);
349 case SWITCHDEV_OBJ_ID_PORT_FDB:
350 return sizeof(struct switchdev_obj_port_fdb);
351 case SWITCHDEV_OBJ_ID_PORT_MDB:
352 return sizeof(struct switchdev_obj_port_mdb);
353 default:
354 BUG();
355 }
356 return 0;
357}
358
359static int __switchdev_port_obj_add(struct net_device *dev,
360 const struct switchdev_obj *obj,
361 struct switchdev_trans *trans)
362{
363 const struct switchdev_ops *ops = dev->switchdev_ops;
364 struct net_device *lower_dev;
365 struct list_head *iter;
366 int err = -EOPNOTSUPP;
367
368 if (ops && ops->switchdev_port_obj_add)
369 return ops->switchdev_port_obj_add(dev, obj, trans);
370
371 /* Switch device port(s) may be stacked under
372 * bond/team/vlan dev, so recurse down to add object on
373 * each port.
374 */
375
376 netdev_for_each_lower_dev(dev, lower_dev, iter) {
377 err = __switchdev_port_obj_add(lower_dev, obj, trans);
378 if (err)
379 break;
380 }
381
382 return err;
383}
384
385static int switchdev_port_obj_add_now(struct net_device *dev,
386 const struct switchdev_obj *obj)
387{
388 struct switchdev_trans trans;
389 int err;
390
391 ASSERT_RTNL();
392
393 switchdev_trans_init(&trans);
394
395 /* Phase I: prepare for obj add. Driver/device should fail
396 * here if there are going to be issues in the commit phase,
397 * such as lack of resources or support. The driver/device
398 * should reserve resources needed for the commit phase here,
399 * but should not commit the obj.
400 */
401
402 trans.ph_prepare = true;
403 err = __switchdev_port_obj_add(dev, obj, &trans);
404 if (err) {
405 /* Prepare phase failed: abort the transaction. Any
406 * resources reserved in the prepare phase are
407 * released.
408 */
409
410 if (err != -EOPNOTSUPP)
411 switchdev_trans_items_destroy(&trans);
412
413 return err;
414 }
415
416 /* Phase II: commit obj add. This cannot fail as a fault
417 * of driver/device. If it does, it's a bug in the driver/device
418 * because the driver said everythings was OK in phase I.
419 */
420
421 trans.ph_prepare = false;
422 err = __switchdev_port_obj_add(dev, obj, &trans);
423 WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
424 switchdev_trans_items_warn_destroy(dev, &trans);
425
426 return err;
427}
428
429static void switchdev_port_obj_add_deferred(struct net_device *dev,
430 const void *data)
431{
432 const struct switchdev_obj *obj = data;
433 int err;
434
435 err = switchdev_port_obj_add_now(dev, obj);
436 if (err && err != -EOPNOTSUPP)
437 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
438 err, obj->id);
439 if (obj->complete)
440 obj->complete(dev, err, obj->complete_priv);
441}
442
443static int switchdev_port_obj_add_defer(struct net_device *dev,
444 const struct switchdev_obj *obj)
445{
446 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
447 switchdev_port_obj_add_deferred);
448}
449
450/**
451 * switchdev_port_obj_add - Add port object
452 *
453 * @dev: port device
454 * @id: object ID
455 * @obj: object to add
456 *
457 * Use a 2-phase prepare-commit transaction model to ensure
458 * system is not left in a partially updated state due to
459 * failure from driver/device.
460 *
461 * rtnl_lock must be held and must not be in atomic section,
462 * in case SWITCHDEV_F_DEFER flag is not set.
463 */
464int switchdev_port_obj_add(struct net_device *dev,
465 const struct switchdev_obj *obj)
466{
467 if (obj->flags & SWITCHDEV_F_DEFER)
468 return switchdev_port_obj_add_defer(dev, obj);
469 ASSERT_RTNL();
470 return switchdev_port_obj_add_now(dev, obj);
471}
472EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
473
474static int switchdev_port_obj_del_now(struct net_device *dev,
475 const struct switchdev_obj *obj)
476{
477 const struct switchdev_ops *ops = dev->switchdev_ops;
478 struct net_device *lower_dev;
479 struct list_head *iter;
480 int err = -EOPNOTSUPP;
481
482 if (ops && ops->switchdev_port_obj_del)
483 return ops->switchdev_port_obj_del(dev, obj);
484
485 /* Switch device port(s) may be stacked under
486 * bond/team/vlan dev, so recurse down to delete object on
487 * each port.
488 */
489
490 netdev_for_each_lower_dev(dev, lower_dev, iter) {
491 err = switchdev_port_obj_del_now(lower_dev, obj);
492 if (err)
493 break;
494 }
495
496 return err;
497}
498
499static void switchdev_port_obj_del_deferred(struct net_device *dev,
500 const void *data)
501{
502 const struct switchdev_obj *obj = data;
503 int err;
504
505 err = switchdev_port_obj_del_now(dev, obj);
506 if (err && err != -EOPNOTSUPP)
507 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
508 err, obj->id);
509 if (obj->complete)
510 obj->complete(dev, err, obj->complete_priv);
511}
512
513static int switchdev_port_obj_del_defer(struct net_device *dev,
514 const struct switchdev_obj *obj)
515{
516 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
517 switchdev_port_obj_del_deferred);
518}
519
520/**
521 * switchdev_port_obj_del - Delete port object
522 *
523 * @dev: port device
524 * @id: object ID
525 * @obj: object to delete
526 *
527 * rtnl_lock must be held and must not be in atomic section,
528 * in case SWITCHDEV_F_DEFER flag is not set.
529 */
530int switchdev_port_obj_del(struct net_device *dev,
531 const struct switchdev_obj *obj)
532{
533 if (obj->flags & SWITCHDEV_F_DEFER)
534 return switchdev_port_obj_del_defer(dev, obj);
535 ASSERT_RTNL();
536 return switchdev_port_obj_del_now(dev, obj);
537}
538EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
539
540/**
541 * switchdev_port_obj_dump - Dump port objects
542 *
543 * @dev: port device
544 * @id: object ID
545 * @obj: object to dump
546 * @cb: function to call with a filled object
547 *
548 * rtnl_lock must be held.
549 */
550int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
551 switchdev_obj_dump_cb_t *cb)
552{
553 const struct switchdev_ops *ops = dev->switchdev_ops;
554 struct net_device *lower_dev;
555 struct list_head *iter;
556 int err = -EOPNOTSUPP;
557
558 ASSERT_RTNL();
559
560 if (ops && ops->switchdev_port_obj_dump)
561 return ops->switchdev_port_obj_dump(dev, obj, cb);
562
563 /* Switch device port(s) may be stacked under
564 * bond/team/vlan dev, so recurse down to dump objects on
565 * first port at bottom of stack.
566 */
567
568 netdev_for_each_lower_dev(dev, lower_dev, iter) {
569 err = switchdev_port_obj_dump(lower_dev, obj, cb);
570 break;
571 }
572
573 return err;
574}
575EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
576
577static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
578
579/**
580 * register_switchdev_notifier - Register notifier
581 * @nb: notifier_block
582 *
583 * Register switch device notifier. This should be used by code
584 * which needs to monitor events happening in particular device.
585 * Return values are same as for atomic_notifier_chain_register().
586 */
587int register_switchdev_notifier(struct notifier_block *nb)
588{
589 int err;
590
591 rtnl_lock();
592 err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
593 rtnl_unlock();
594 return err;
595}
596EXPORT_SYMBOL_GPL(register_switchdev_notifier);
597
598/**
599 * unregister_switchdev_notifier - Unregister notifier
600 * @nb: notifier_block
601 *
602 * Unregister switch device notifier.
603 * Return values are same as for atomic_notifier_chain_unregister().
604 */
605int unregister_switchdev_notifier(struct notifier_block *nb)
606{
607 int err;
608
609 rtnl_lock();
610 err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
611 rtnl_unlock();
612 return err;
613}
614EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
615
616/**
617 * call_switchdev_notifiers - Call notifiers
618 * @val: value passed unmodified to notifier function
619 * @dev: port device
620 * @info: notifier information data
621 *
622 * Call all network notifier blocks. This should be called by driver
623 * when it needs to propagate hardware event.
624 * Return values are same as for atomic_notifier_call_chain().
625 * rtnl_lock must be held.
626 */
627int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
628 struct switchdev_notifier_info *info)
629{
630 int err;
631
632 ASSERT_RTNL();
633
634 info->dev = dev;
635 err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
636 return err;
637}
638EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
639
640struct switchdev_vlan_dump {
641 struct switchdev_obj_port_vlan vlan;
642 struct sk_buff *skb;
643 u32 filter_mask;
644 u16 flags;
645 u16 begin;
646 u16 end;
647};
648
649static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
650{
651 struct bridge_vlan_info vinfo;
652
653 vinfo.flags = dump->flags;
654
655 if (dump->begin == 0 && dump->end == 0) {
656 return 0;
657 } else if (dump->begin == dump->end) {
658 vinfo.vid = dump->begin;
659 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
660 sizeof(vinfo), &vinfo))
661 return -EMSGSIZE;
662 } else {
663 vinfo.vid = dump->begin;
664 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
665 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
666 sizeof(vinfo), &vinfo))
667 return -EMSGSIZE;
668 vinfo.vid = dump->end;
669 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
670 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
671 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
672 sizeof(vinfo), &vinfo))
673 return -EMSGSIZE;
674 }
675
676 return 0;
677}
678
679static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
680{
681 struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
682 struct switchdev_vlan_dump *dump =
683 container_of(vlan, struct switchdev_vlan_dump, vlan);
684 int err = 0;
685
686 if (vlan->vid_begin > vlan->vid_end)
687 return -EINVAL;
688
689 if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
690 dump->flags = vlan->flags;
691 for (dump->begin = dump->end = vlan->vid_begin;
692 dump->begin <= vlan->vid_end;
693 dump->begin++, dump->end++) {
694 err = switchdev_port_vlan_dump_put(dump);
695 if (err)
696 return err;
697 }
698 } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
699 if (dump->begin > vlan->vid_begin &&
700 dump->begin >= vlan->vid_end) {
701 if ((dump->begin - 1) == vlan->vid_end &&
702 dump->flags == vlan->flags) {
703 /* prepend */
704 dump->begin = vlan->vid_begin;
705 } else {
706 err = switchdev_port_vlan_dump_put(dump);
707 dump->flags = vlan->flags;
708 dump->begin = vlan->vid_begin;
709 dump->end = vlan->vid_end;
710 }
711 } else if (dump->end <= vlan->vid_begin &&
712 dump->end < vlan->vid_end) {
713 if ((dump->end + 1) == vlan->vid_begin &&
714 dump->flags == vlan->flags) {
715 /* append */
716 dump->end = vlan->vid_end;
717 } else {
718 err = switchdev_port_vlan_dump_put(dump);
719 dump->flags = vlan->flags;
720 dump->begin = vlan->vid_begin;
721 dump->end = vlan->vid_end;
722 }
723 } else {
724 err = -EINVAL;
725 }
726 }
727
728 return err;
729}
730
731static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
732 u32 filter_mask)
733{
734 struct switchdev_vlan_dump dump = {
735 .vlan.obj.orig_dev = dev,
736 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
737 .skb = skb,
738 .filter_mask = filter_mask,
739 };
740 int err = 0;
741
742 if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
743 (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
744 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
745 switchdev_port_vlan_dump_cb);
746 if (err)
747 goto err_out;
748 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
749 /* last one */
750 err = switchdev_port_vlan_dump_put(&dump);
751 }
752
753err_out:
754 return err == -EOPNOTSUPP ? 0 : err;
755}
756
757/**
758 * switchdev_port_bridge_getlink - Get bridge port attributes
759 *
760 * @dev: port device
761 *
762 * Called for SELF on rtnl_bridge_getlink to get bridge port
763 * attributes.
764 */
765int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
766 struct net_device *dev, u32 filter_mask,
767 int nlflags)
768{
769 struct switchdev_attr attr = {
770 .orig_dev = dev,
771 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
772 };
773 u16 mode = BRIDGE_MODE_UNDEF;
774 u32 mask = BR_LEARNING | BR_LEARNING_SYNC | BR_FLOOD;
775 int err;
776
777 err = switchdev_port_attr_get(dev, &attr);
778 if (err && err != -EOPNOTSUPP)
779 return err;
780
781 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
782 attr.u.brport_flags, mask, nlflags,
783 filter_mask, switchdev_port_vlan_fill);
784}
785EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
786
787static int switchdev_port_br_setflag(struct net_device *dev,
788 struct nlattr *nlattr,
789 unsigned long brport_flag)
790{
791 struct switchdev_attr attr = {
792 .orig_dev = dev,
793 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
794 };
795 u8 flag = nla_get_u8(nlattr);
796 int err;
797
798 err = switchdev_port_attr_get(dev, &attr);
799 if (err)
800 return err;
801
802 if (flag)
803 attr.u.brport_flags |= brport_flag;
804 else
805 attr.u.brport_flags &= ~brport_flag;
806
807 return switchdev_port_attr_set(dev, &attr);
808}
809
810static const struct nla_policy
811switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
812 [IFLA_BRPORT_STATE] = { .type = NLA_U8 },
813 [IFLA_BRPORT_COST] = { .type = NLA_U32 },
814 [IFLA_BRPORT_PRIORITY] = { .type = NLA_U16 },
815 [IFLA_BRPORT_MODE] = { .type = NLA_U8 },
816 [IFLA_BRPORT_GUARD] = { .type = NLA_U8 },
817 [IFLA_BRPORT_PROTECT] = { .type = NLA_U8 },
818 [IFLA_BRPORT_FAST_LEAVE] = { .type = NLA_U8 },
819 [IFLA_BRPORT_LEARNING] = { .type = NLA_U8 },
820 [IFLA_BRPORT_LEARNING_SYNC] = { .type = NLA_U8 },
821 [IFLA_BRPORT_UNICAST_FLOOD] = { .type = NLA_U8 },
822};
823
824static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
825 struct nlattr *protinfo)
826{
827 struct nlattr *attr;
828 int rem;
829 int err;
830
831 err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
832 switchdev_port_bridge_policy);
833 if (err)
834 return err;
835
836 nla_for_each_nested(attr, protinfo, rem) {
837 switch (nla_type(attr)) {
838 case IFLA_BRPORT_LEARNING:
839 err = switchdev_port_br_setflag(dev, attr,
840 BR_LEARNING);
841 break;
842 case IFLA_BRPORT_LEARNING_SYNC:
843 err = switchdev_port_br_setflag(dev, attr,
844 BR_LEARNING_SYNC);
845 break;
846 case IFLA_BRPORT_UNICAST_FLOOD:
847 err = switchdev_port_br_setflag(dev, attr, BR_FLOOD);
848 break;
849 default:
850 err = -EOPNOTSUPP;
851 break;
852 }
853 if (err)
854 return err;
855 }
856
857 return 0;
858}
859
860static int switchdev_port_br_afspec(struct net_device *dev,
861 struct nlattr *afspec,
862 int (*f)(struct net_device *dev,
863 const struct switchdev_obj *obj))
864{
865 struct nlattr *attr;
866 struct bridge_vlan_info *vinfo;
867 struct switchdev_obj_port_vlan vlan = {
868 .obj.orig_dev = dev,
869 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
870 };
871 int rem;
872 int err;
873
874 nla_for_each_nested(attr, afspec, rem) {
875 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
876 continue;
877 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
878 return -EINVAL;
879 vinfo = nla_data(attr);
880 if (!vinfo->vid || vinfo->vid >= VLAN_VID_MASK)
881 return -EINVAL;
882 vlan.flags = vinfo->flags;
883 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
884 if (vlan.vid_begin)
885 return -EINVAL;
886 vlan.vid_begin = vinfo->vid;
887 /* don't allow range of pvids */
888 if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
889 return -EINVAL;
890 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
891 if (!vlan.vid_begin)
892 return -EINVAL;
893 vlan.vid_end = vinfo->vid;
894 if (vlan.vid_end <= vlan.vid_begin)
895 return -EINVAL;
896 err = f(dev, &vlan.obj);
897 if (err)
898 return err;
899 vlan.vid_begin = 0;
900 } else {
901 if (vlan.vid_begin)
902 return -EINVAL;
903 vlan.vid_begin = vinfo->vid;
904 vlan.vid_end = vinfo->vid;
905 err = f(dev, &vlan.obj);
906 if (err)
907 return err;
908 vlan.vid_begin = 0;
909 }
910 }
911
912 return 0;
913}
914
915/**
916 * switchdev_port_bridge_setlink - Set bridge port attributes
917 *
918 * @dev: port device
919 * @nlh: netlink header
920 * @flags: netlink flags
921 *
922 * Called for SELF on rtnl_bridge_setlink to set bridge port
923 * attributes.
924 */
925int switchdev_port_bridge_setlink(struct net_device *dev,
926 struct nlmsghdr *nlh, u16 flags)
927{
928 struct nlattr *protinfo;
929 struct nlattr *afspec;
930 int err = 0;
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 afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
966 IFLA_AF_SPEC);
967 if (afspec)
968 return switchdev_port_br_afspec(dev, afspec,
969 switchdev_port_obj_del);
970
971 return 0;
972}
973EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
974
975/**
976 * switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
977 *
978 * @ndmsg: netlink hdr
979 * @nlattr: netlink attributes
980 * @dev: port device
981 * @addr: MAC address to add
982 * @vid: VLAN to add
983 *
984 * Add FDB entry to switch device.
985 */
986int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
987 struct net_device *dev, const unsigned char *addr,
988 u16 vid, u16 nlm_flags)
989{
990 struct switchdev_obj_port_fdb fdb = {
991 .obj.orig_dev = dev,
992 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
993 .vid = vid,
994 };
995
996 ether_addr_copy(fdb.addr, addr);
997 return switchdev_port_obj_add(dev, &fdb.obj);
998}
999EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
1000
1001/**
1002 * switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
1003 *
1004 * @ndmsg: netlink hdr
1005 * @nlattr: netlink attributes
1006 * @dev: port device
1007 * @addr: MAC address to delete
1008 * @vid: VLAN to delete
1009 *
1010 * Delete FDB entry from switch device.
1011 */
1012int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1013 struct net_device *dev, const unsigned char *addr,
1014 u16 vid)
1015{
1016 struct switchdev_obj_port_fdb fdb = {
1017 .obj.orig_dev = dev,
1018 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1019 .vid = vid,
1020 };
1021
1022 ether_addr_copy(fdb.addr, addr);
1023 return switchdev_port_obj_del(dev, &fdb.obj);
1024}
1025EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
1026
1027struct switchdev_fdb_dump {
1028 struct switchdev_obj_port_fdb fdb;
1029 struct net_device *dev;
1030 struct sk_buff *skb;
1031 struct netlink_callback *cb;
1032 int idx;
1033};
1034
1035static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
1036{
1037 struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
1038 struct switchdev_fdb_dump *dump =
1039 container_of(fdb, struct switchdev_fdb_dump, fdb);
1040 u32 portid = NETLINK_CB(dump->cb->skb).portid;
1041 u32 seq = dump->cb->nlh->nlmsg_seq;
1042 struct nlmsghdr *nlh;
1043 struct ndmsg *ndm;
1044
1045 if (dump->idx < dump->cb->args[0])
1046 goto skip;
1047
1048 nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
1049 sizeof(*ndm), NLM_F_MULTI);
1050 if (!nlh)
1051 return -EMSGSIZE;
1052
1053 ndm = nlmsg_data(nlh);
1054 ndm->ndm_family = AF_BRIDGE;
1055 ndm->ndm_pad1 = 0;
1056 ndm->ndm_pad2 = 0;
1057 ndm->ndm_flags = NTF_SELF;
1058 ndm->ndm_type = 0;
1059 ndm->ndm_ifindex = dump->dev->ifindex;
1060 ndm->ndm_state = fdb->ndm_state;
1061
1062 if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
1063 goto nla_put_failure;
1064
1065 if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
1066 goto nla_put_failure;
1067
1068 nlmsg_end(dump->skb, nlh);
1069
1070skip:
1071 dump->idx++;
1072 return 0;
1073
1074nla_put_failure:
1075 nlmsg_cancel(dump->skb, nlh);
1076 return -EMSGSIZE;
1077}
1078
1079/**
1080 * switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
1081 *
1082 * @skb: netlink skb
1083 * @cb: netlink callback
1084 * @dev: port device
1085 * @filter_dev: filter device
1086 * @idx:
1087 *
1088 * Dump FDB entries from switch device.
1089 */
1090int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1091 struct net_device *dev,
1092 struct net_device *filter_dev, int idx)
1093{
1094 struct switchdev_fdb_dump dump = {
1095 .fdb.obj.orig_dev = dev,
1096 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1097 .dev = dev,
1098 .skb = skb,
1099 .cb = cb,
1100 .idx = idx,
1101 };
1102 int err;
1103
1104 err = switchdev_port_obj_dump(dev, &dump.fdb.obj,
1105 switchdev_port_fdb_dump_cb);
1106 cb->args[1] = err;
1107 return dump.idx;
1108}
1109EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
1110
1111static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
1112{
1113 const struct switchdev_ops *ops = dev->switchdev_ops;
1114 struct net_device *lower_dev;
1115 struct net_device *port_dev;
1116 struct list_head *iter;
1117
1118 /* Recusively search down until we find a sw port dev.
1119 * (A sw port dev supports switchdev_port_attr_get).
1120 */
1121
1122 if (ops && ops->switchdev_port_attr_get)
1123 return dev;
1124
1125 netdev_for_each_lower_dev(dev, lower_dev, iter) {
1126 port_dev = switchdev_get_lowest_dev(lower_dev);
1127 if (port_dev)
1128 return port_dev;
1129 }
1130
1131 return NULL;
1132}
1133
1134static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
1135{
1136 struct switchdev_attr attr = {
1137 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1138 };
1139 struct switchdev_attr prev_attr;
1140 struct net_device *dev = NULL;
1141 int nhsel;
1142
1143 ASSERT_RTNL();
1144
1145 /* For this route, all nexthop devs must be on the same switch. */
1146
1147 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1148 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1149
1150 if (!nh->nh_dev)
1151 return NULL;
1152
1153 dev = switchdev_get_lowest_dev(nh->nh_dev);
1154 if (!dev)
1155 return NULL;
1156
1157 attr.orig_dev = dev;
1158 if (switchdev_port_attr_get(dev, &attr))
1159 return NULL;
1160
1161 if (nhsel > 0 &&
1162 !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
1163 return NULL;
1164
1165 prev_attr = attr;
1166 }
1167
1168 return dev;
1169}
1170
1171/**
1172 * switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
1173 *
1174 * @dst: route's IPv4 destination address
1175 * @dst_len: destination address length (prefix length)
1176 * @fi: route FIB info structure
1177 * @tos: route TOS
1178 * @type: route type
1179 * @nlflags: netlink flags passed in (NLM_F_*)
1180 * @tb_id: route table ID
1181 *
1182 * Add/modify switch IPv4 route entry.
1183 */
1184int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1185 u8 tos, u8 type, u32 nlflags, u32 tb_id)
1186{
1187 struct switchdev_obj_ipv4_fib ipv4_fib = {
1188 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1189 .dst = dst,
1190 .dst_len = dst_len,
1191 .tos = tos,
1192 .type = type,
1193 .nlflags = nlflags,
1194 .tb_id = tb_id,
1195 };
1196 struct net_device *dev;
1197 int err = 0;
1198
1199 memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1200
1201 /* Don't offload route if using custom ip rules or if
1202 * IPv4 FIB offloading has been disabled completely.
1203 */
1204
1205#ifdef CONFIG_IP_MULTIPLE_TABLES
1206 if (fi->fib_net->ipv4.fib_has_custom_rules)
1207 return 0;
1208#endif
1209
1210 if (fi->fib_net->ipv4.fib_offload_disabled)
1211 return 0;
1212
1213 dev = switchdev_get_dev_by_nhs(fi);
1214 if (!dev)
1215 return 0;
1216
1217 ipv4_fib.obj.orig_dev = dev;
1218 err = switchdev_port_obj_add(dev, &ipv4_fib.obj);
1219 if (!err)
1220 fi->fib_flags |= RTNH_F_OFFLOAD;
1221
1222 return err == -EOPNOTSUPP ? 0 : err;
1223}
1224EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1225
1226/**
1227 * switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1228 *
1229 * @dst: route's IPv4 destination address
1230 * @dst_len: destination address length (prefix length)
1231 * @fi: route FIB info structure
1232 * @tos: route TOS
1233 * @type: route type
1234 * @tb_id: route table ID
1235 *
1236 * Delete IPv4 route entry from switch device.
1237 */
1238int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1239 u8 tos, u8 type, u32 tb_id)
1240{
1241 struct switchdev_obj_ipv4_fib ipv4_fib = {
1242 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1243 .dst = dst,
1244 .dst_len = dst_len,
1245 .tos = tos,
1246 .type = type,
1247 .nlflags = 0,
1248 .tb_id = tb_id,
1249 };
1250 struct net_device *dev;
1251 int err = 0;
1252
1253 memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1254
1255 if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1256 return 0;
1257
1258 dev = switchdev_get_dev_by_nhs(fi);
1259 if (!dev)
1260 return 0;
1261
1262 ipv4_fib.obj.orig_dev = dev;
1263 err = switchdev_port_obj_del(dev, &ipv4_fib.obj);
1264 if (!err)
1265 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1266
1267 return err == -EOPNOTSUPP ? 0 : err;
1268}
1269EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1270
1271/**
1272 * switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1273 *
1274 * @fi: route FIB info structure
1275 */
1276void switchdev_fib_ipv4_abort(struct fib_info *fi)
1277{
1278 /* There was a problem installing this route to the offload
1279 * device. For now, until we come up with more refined
1280 * policy handling, abruptly end IPv4 fib offloading for
1281 * for entire net by flushing offload device(s) of all
1282 * IPv4 routes, and mark IPv4 fib offloading broken from
1283 * this point forward.
1284 */
1285
1286 fib_flush_external(fi->fib_net);
1287 fi->fib_net->ipv4.fib_offload_disabled = true;
1288}
1289EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1290
1291static bool switchdev_port_same_parent_id(struct net_device *a,
1292 struct net_device *b)
1293{
1294 struct switchdev_attr a_attr = {
1295 .orig_dev = a,
1296 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1297 .flags = SWITCHDEV_F_NO_RECURSE,
1298 };
1299 struct switchdev_attr b_attr = {
1300 .orig_dev = b,
1301 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1302 .flags = SWITCHDEV_F_NO_RECURSE,
1303 };
1304
1305 if (switchdev_port_attr_get(a, &a_attr) ||
1306 switchdev_port_attr_get(b, &b_attr))
1307 return false;
1308
1309 return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1310}
1311
1312static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1313 struct net_device *group_dev)
1314{
1315 struct net_device *lower_dev;
1316 struct list_head *iter;
1317
1318 netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1319 if (lower_dev == dev)
1320 continue;
1321 if (switchdev_port_same_parent_id(dev, lower_dev))
1322 return lower_dev->offload_fwd_mark;
1323 return switchdev_port_fwd_mark_get(dev, lower_dev);
1324 }
1325
1326 return dev->ifindex;
1327}
1328
1329static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1330 u32 old_mark, u32 *reset_mark)
1331{
1332 struct net_device *lower_dev;
1333 struct list_head *iter;
1334
1335 netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1336 if (lower_dev->offload_fwd_mark == old_mark) {
1337 if (!*reset_mark)
1338 *reset_mark = lower_dev->ifindex;
1339 lower_dev->offload_fwd_mark = *reset_mark;
1340 }
1341 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1342 }
1343}
1344
1345/**
1346 * switchdev_port_fwd_mark_set - Set port offload forwarding mark
1347 *
1348 * @dev: port device
1349 * @group_dev: containing device
1350 * @joining: true if dev is joining group; false if leaving group
1351 *
1352 * An ungrouped port's offload mark is just its ifindex. A grouped
1353 * port's (member of a bridge, for example) offload mark is the ifindex
1354 * of one of the ports in the group with the same parent (switch) ID.
1355 * Ports on the same device in the same group will have the same mark.
1356 *
1357 * Example:
1358 *
1359 * br0 ifindex=9
1360 * sw1p1 ifindex=2 mark=2
1361 * sw1p2 ifindex=3 mark=2
1362 * sw2p1 ifindex=4 mark=5
1363 * sw2p2 ifindex=5 mark=5
1364 *
1365 * If sw2p2 leaves the bridge, we'll have:
1366 *
1367 * br0 ifindex=9
1368 * sw1p1 ifindex=2 mark=2
1369 * sw1p2 ifindex=3 mark=2
1370 * sw2p1 ifindex=4 mark=4
1371 * sw2p2 ifindex=5 mark=5
1372 */
1373void switchdev_port_fwd_mark_set(struct net_device *dev,
1374 struct net_device *group_dev,
1375 bool joining)
1376{
1377 u32 mark = dev->ifindex;
1378 u32 reset_mark = 0;
1379
1380 if (group_dev) {
1381 ASSERT_RTNL();
1382 if (joining)
1383 mark = switchdev_port_fwd_mark_get(dev, group_dev);
1384 else if (dev->offload_fwd_mark == mark)
1385 /* Ohoh, this port was the mark reference port,
1386 * but it's leaving the group, so reset the
1387 * mark for the remaining ports in the group.
1388 */
1389 switchdev_port_fwd_mark_reset(group_dev, mark,
1390 &reset_mark);
1391 }
1392
1393 dev->offload_fwd_mark = mark;
1394}
1395EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);