projects/linux·blob v5.14-rc3 · refs

551 lines · 15.0 KiB

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 <[email protected]>
5 * Copyright (c) 2014-2015 Scott Feldman <[email protected]>
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 switchdev_deferred_func_t *func;
32 unsigned long data[];
33};
34
35static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
36{
37 struct switchdev_deferred_item *dfitem;
38
39 spin_lock_bh(&deferred_lock);
40 if (list_empty(&deferred)) {
41 dfitem = NULL;
42 goto unlock;
43 }
44 dfitem = list_first_entry(&deferred,
45 struct switchdev_deferred_item, list);
46 list_del(&dfitem->list);
47unlock:
48 spin_unlock_bh(&deferred_lock);
49 return dfitem;
50}
51
52/**
53 * switchdev_deferred_process - Process ops in deferred queue
54 *
55 * Called to flush the ops currently queued in deferred ops queue.
56 * rtnl_lock must be held.
57 */
58void switchdev_deferred_process(void)
59{
60 struct switchdev_deferred_item *dfitem;
61
62 ASSERT_RTNL();
63
64 while ((dfitem = switchdev_deferred_dequeue())) {
65 dfitem->func(dfitem->dev, dfitem->data);
66 dev_put(dfitem->dev);
67 kfree(dfitem);
68 }
69}
70EXPORT_SYMBOL_GPL(switchdev_deferred_process);
71
72static void switchdev_deferred_process_work(struct work_struct *work)
73{
74 rtnl_lock();
75 switchdev_deferred_process();
76 rtnl_unlock();
77}
78
79static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
80
81static int switchdev_deferred_enqueue(struct net_device *dev,
82 const void *data, size_t data_len,
83 switchdev_deferred_func_t *func)
84{
85 struct switchdev_deferred_item *dfitem;
86
87 dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
88 if (!dfitem)
89 return -ENOMEM;
90 dfitem->dev = dev;
91 dfitem->func = func;
92 memcpy(dfitem->data, data, data_len);
93 dev_hold(dev);
94 spin_lock_bh(&deferred_lock);
95 list_add_tail(&dfitem->list, &deferred);
96 spin_unlock_bh(&deferred_lock);
97 schedule_work(&deferred_process_work);
98 return 0;
99}
100
101static int switchdev_port_attr_notify(enum switchdev_notifier_type nt,
102 struct net_device *dev,
103 const struct switchdev_attr *attr,
104 struct netlink_ext_ack *extack)
105{
106 int err;
107 int rc;
108
109 struct switchdev_notifier_port_attr_info attr_info = {
110 .attr = attr,
111 .handled = false,
112 };
113
114 rc = call_switchdev_blocking_notifiers(nt, dev,
115 &attr_info.info, extack);
116 err = notifier_to_errno(rc);
117 if (err) {
118 WARN_ON(!attr_info.handled);
119 return err;
120 }
121
122 if (!attr_info.handled)
123 return -EOPNOTSUPP;
124
125 return 0;
126}
127
128static int switchdev_port_attr_set_now(struct net_device *dev,
129 const struct switchdev_attr *attr,
130 struct netlink_ext_ack *extack)
131{
132 return switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
133 extack);
134}
135
136static void switchdev_port_attr_set_deferred(struct net_device *dev,
137 const void *data)
138{
139 const struct switchdev_attr *attr = data;
140 int err;
141
142 err = switchdev_port_attr_set_now(dev, attr, NULL);
143 if (err && err != -EOPNOTSUPP)
144 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
145 err, attr->id);
146 if (attr->complete)
147 attr->complete(dev, err, attr->complete_priv);
148}
149
150static int switchdev_port_attr_set_defer(struct net_device *dev,
151 const struct switchdev_attr *attr)
152{
153 return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
154 switchdev_port_attr_set_deferred);
155}
156
157/**
158 * switchdev_port_attr_set - Set port attribute
159 *
160 * @dev: port device
161 * @attr: attribute to set
162 * @extack: netlink extended ack, for error message propagation
163 *
164 * rtnl_lock must be held and must not be in atomic section,
165 * in case SWITCHDEV_F_DEFER flag is not set.
166 */
167int switchdev_port_attr_set(struct net_device *dev,
168 const struct switchdev_attr *attr,
169 struct netlink_ext_ack *extack)
170{
171 if (attr->flags & SWITCHDEV_F_DEFER)
172 return switchdev_port_attr_set_defer(dev, attr);
173 ASSERT_RTNL();
174 return switchdev_port_attr_set_now(dev, attr, extack);
175}
176EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
177
178static size_t switchdev_obj_size(const struct switchdev_obj *obj)
179{
180 switch (obj->id) {
181 case SWITCHDEV_OBJ_ID_PORT_VLAN:
182 return sizeof(struct switchdev_obj_port_vlan);
183 case SWITCHDEV_OBJ_ID_PORT_MDB:
184 return sizeof(struct switchdev_obj_port_mdb);
185 case SWITCHDEV_OBJ_ID_HOST_MDB:
186 return sizeof(struct switchdev_obj_port_mdb);
187 default:
188 BUG();
189 }
190 return 0;
191}
192
193static int switchdev_port_obj_notify(enum switchdev_notifier_type nt,
194 struct net_device *dev,
195 const struct switchdev_obj *obj,
196 struct netlink_ext_ack *extack)
197{
198 int rc;
199 int err;
200
201 struct switchdev_notifier_port_obj_info obj_info = {
202 .obj = obj,
203 .handled = false,
204 };
205
206 rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack);
207 err = notifier_to_errno(rc);
208 if (err) {
209 WARN_ON(!obj_info.handled);
210 return err;
211 }
212 if (!obj_info.handled)
213 return -EOPNOTSUPP;
214 return 0;
215}
216
217static void switchdev_port_obj_add_deferred(struct net_device *dev,
218 const void *data)
219{
220 const struct switchdev_obj *obj = data;
221 int err;
222
223 ASSERT_RTNL();
224 err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
225 dev, obj, NULL);
226 if (err && err != -EOPNOTSUPP)
227 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
228 err, obj->id);
229 if (obj->complete)
230 obj->complete(dev, err, obj->complete_priv);
231}
232
233static int switchdev_port_obj_add_defer(struct net_device *dev,
234 const struct switchdev_obj *obj)
235{
236 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
237 switchdev_port_obj_add_deferred);
238}
239
240/**
241 * switchdev_port_obj_add - Add port object
242 *
243 * @dev: port device
244 * @obj: object to add
245 * @extack: netlink extended ack
246 *
247 * rtnl_lock must be held and must not be in atomic section,
248 * in case SWITCHDEV_F_DEFER flag is not set.
249 */
250int switchdev_port_obj_add(struct net_device *dev,
251 const struct switchdev_obj *obj,
252 struct netlink_ext_ack *extack)
253{
254 if (obj->flags & SWITCHDEV_F_DEFER)
255 return switchdev_port_obj_add_defer(dev, obj);
256 ASSERT_RTNL();
257 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
258 dev, obj, extack);
259}
260EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
261
262static int switchdev_port_obj_del_now(struct net_device *dev,
263 const struct switchdev_obj *obj)
264{
265 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL,
266 dev, obj, NULL);
267}
268
269static void switchdev_port_obj_del_deferred(struct net_device *dev,
270 const void *data)
271{
272 const struct switchdev_obj *obj = data;
273 int err;
274
275 err = switchdev_port_obj_del_now(dev, obj);
276 if (err && err != -EOPNOTSUPP)
277 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
278 err, obj->id);
279 if (obj->complete)
280 obj->complete(dev, err, obj->complete_priv);
281}
282
283static int switchdev_port_obj_del_defer(struct net_device *dev,
284 const struct switchdev_obj *obj)
285{
286 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
287 switchdev_port_obj_del_deferred);
288}
289
290/**
291 * switchdev_port_obj_del - Delete port object
292 *
293 * @dev: port device
294 * @obj: object to delete
295 *
296 * rtnl_lock must be held and must not be in atomic section,
297 * in case SWITCHDEV_F_DEFER flag is not set.
298 */
299int switchdev_port_obj_del(struct net_device *dev,
300 const struct switchdev_obj *obj)
301{
302 if (obj->flags & SWITCHDEV_F_DEFER)
303 return switchdev_port_obj_del_defer(dev, obj);
304 ASSERT_RTNL();
305 return switchdev_port_obj_del_now(dev, obj);
306}
307EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
308
309static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
310static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
311
312/**
313 * register_switchdev_notifier - Register notifier
314 * @nb: notifier_block
315 *
316 * Register switch device notifier.
317 */
318int register_switchdev_notifier(struct notifier_block *nb)
319{
320 return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
321}
322EXPORT_SYMBOL_GPL(register_switchdev_notifier);
323
324/**
325 * unregister_switchdev_notifier - Unregister notifier
326 * @nb: notifier_block
327 *
328 * Unregister switch device notifier.
329 */
330int unregister_switchdev_notifier(struct notifier_block *nb)
331{
332 return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
333}
334EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
335
336/**
337 * call_switchdev_notifiers - Call notifiers
338 * @val: value passed unmodified to notifier function
339 * @dev: port device
340 * @info: notifier information data
341 * @extack: netlink extended ack
342 * Call all network notifier blocks.
343 */
344int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
345 struct switchdev_notifier_info *info,
346 struct netlink_ext_ack *extack)
347{
348 info->dev = dev;
349 info->extack = extack;
350 return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
351}
352EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
353
354int register_switchdev_blocking_notifier(struct notifier_block *nb)
355{
356 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
357
358 return blocking_notifier_chain_register(chain, nb);
359}
360EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
361
362int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
363{
364 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
365
366 return blocking_notifier_chain_unregister(chain, nb);
367}
368EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
369
370int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
371 struct switchdev_notifier_info *info,
372 struct netlink_ext_ack *extack)
373{
374 info->dev = dev;
375 info->extack = extack;
376 return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
377 val, info);
378}
379EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
380
381static int __switchdev_handle_port_obj_add(struct net_device *dev,
382 struct switchdev_notifier_port_obj_info *port_obj_info,
383 bool (*check_cb)(const struct net_device *dev),
384 int (*add_cb)(struct net_device *dev, const void *ctx,
385 const struct switchdev_obj *obj,
386 struct netlink_ext_ack *extack))
387{
388 struct switchdev_notifier_info *info = &port_obj_info->info;
389 struct netlink_ext_ack *extack;
390 struct net_device *lower_dev;
391 struct list_head *iter;
392 int err = -EOPNOTSUPP;
393
394 extack = switchdev_notifier_info_to_extack(info);
395
396 if (check_cb(dev)) {
397 err = add_cb(dev, info->ctx, port_obj_info->obj, extack);
398 if (err != -EOPNOTSUPP)
399 port_obj_info->handled = true;
400 return err;
401 }
402
403 /* Switch ports might be stacked under e.g. a LAG. Ignore the
404 * unsupported devices, another driver might be able to handle them. But
405 * propagate to the callers any hard errors.
406 *
407 * If the driver does its own bookkeeping of stacked ports, it's not
408 * necessary to go through this helper.
409 */
410 netdev_for_each_lower_dev(dev, lower_dev, iter) {
411 if (netif_is_bridge_master(lower_dev))
412 continue;
413
414 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
415 check_cb, add_cb);
416 if (err && err != -EOPNOTSUPP)
417 return err;
418 }
419
420 return err;
421}
422
423int switchdev_handle_port_obj_add(struct net_device *dev,
424 struct switchdev_notifier_port_obj_info *port_obj_info,
425 bool (*check_cb)(const struct net_device *dev),
426 int (*add_cb)(struct net_device *dev, const void *ctx,
427 const struct switchdev_obj *obj,
428 struct netlink_ext_ack *extack))
429{
430 int err;
431
432 err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
433 add_cb);
434 if (err == -EOPNOTSUPP)
435 err = 0;
436 return err;
437}
438EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
439
440static int __switchdev_handle_port_obj_del(struct net_device *dev,
441 struct switchdev_notifier_port_obj_info *port_obj_info,
442 bool (*check_cb)(const struct net_device *dev),
443 int (*del_cb)(struct net_device *dev, const void *ctx,
444 const struct switchdev_obj *obj))
445{
446 struct switchdev_notifier_info *info = &port_obj_info->info;
447 struct net_device *lower_dev;
448 struct list_head *iter;
449 int err = -EOPNOTSUPP;
450
451 if (check_cb(dev)) {
452 err = del_cb(dev, info->ctx, port_obj_info->obj);
453 if (err != -EOPNOTSUPP)
454 port_obj_info->handled = true;
455 return err;
456 }
457
458 /* Switch ports might be stacked under e.g. a LAG. Ignore the
459 * unsupported devices, another driver might be able to handle them. But
460 * propagate to the callers any hard errors.
461 *
462 * If the driver does its own bookkeeping of stacked ports, it's not
463 * necessary to go through this helper.
464 */
465 netdev_for_each_lower_dev(dev, lower_dev, iter) {
466 if (netif_is_bridge_master(lower_dev))
467 continue;
468
469 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
470 check_cb, del_cb);
471 if (err && err != -EOPNOTSUPP)
472 return err;
473 }
474
475 return err;
476}
477
478int switchdev_handle_port_obj_del(struct net_device *dev,
479 struct switchdev_notifier_port_obj_info *port_obj_info,
480 bool (*check_cb)(const struct net_device *dev),
481 int (*del_cb)(struct net_device *dev, const void *ctx,
482 const struct switchdev_obj *obj))
483{
484 int err;
485
486 err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
487 del_cb);
488 if (err == -EOPNOTSUPP)
489 err = 0;
490 return err;
491}
492EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
493
494static int __switchdev_handle_port_attr_set(struct net_device *dev,
495 struct switchdev_notifier_port_attr_info *port_attr_info,
496 bool (*check_cb)(const struct net_device *dev),
497 int (*set_cb)(struct net_device *dev, const void *ctx,
498 const struct switchdev_attr *attr,
499 struct netlink_ext_ack *extack))
500{
501 struct switchdev_notifier_info *info = &port_attr_info->info;
502 struct netlink_ext_ack *extack;
503 struct net_device *lower_dev;
504 struct list_head *iter;
505 int err = -EOPNOTSUPP;
506
507 extack = switchdev_notifier_info_to_extack(info);
508
509 if (check_cb(dev)) {
510 err = set_cb(dev, info->ctx, port_attr_info->attr, extack);
511 if (err != -EOPNOTSUPP)
512 port_attr_info->handled = true;
513 return err;
514 }
515
516 /* Switch ports might be stacked under e.g. a LAG. Ignore the
517 * unsupported devices, another driver might be able to handle them. But
518 * propagate to the callers any hard errors.
519 *
520 * If the driver does its own bookkeeping of stacked ports, it's not
521 * necessary to go through this helper.
522 */
523 netdev_for_each_lower_dev(dev, lower_dev, iter) {
524 if (netif_is_bridge_master(lower_dev))
525 continue;
526
527 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
528 check_cb, set_cb);
529 if (err && err != -EOPNOTSUPP)
530 return err;
531 }
532
533 return err;
534}
535
536int switchdev_handle_port_attr_set(struct net_device *dev,
537 struct switchdev_notifier_port_attr_info *port_attr_info,
538 bool (*check_cb)(const struct net_device *dev),
539 int (*set_cb)(struct net_device *dev, const void *ctx,
540 const struct switchdev_attr *attr,
541 struct netlink_ext_ack *extack))
542{
543 int err;
544
545 err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
546 set_cb);
547 if (err == -EOPNOTSUPP)
548 err = 0;
549 return err;
550}
551EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);