xref: /linux/net/devlink/core.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2016 Mellanox Technologies. All rights reserved.
4  * Copyright (c) 2016 Jiri Pirko <jiri@mellanox.com>
5  */
6 
7 #include <net/genetlink.h>
8 #define CREATE_TRACE_POINTS
9 #include <trace/events/devlink.h>
10 
11 #include "devl_internal.h"
12 
13 EXPORT_TRACEPOINT_SYMBOL_GPL(devlink_hwmsg);
14 EXPORT_TRACEPOINT_SYMBOL_GPL(devlink_hwerr);
15 EXPORT_TRACEPOINT_SYMBOL_GPL(devlink_trap_report);
16 
17 DEFINE_XARRAY_FLAGS(devlinks, XA_FLAGS_ALLOC);
18 
19 static struct devlink *devlinks_xa_get(unsigned long index)
20 {
21 	struct devlink *devlink;
22 
23 	rcu_read_lock();
24 	devlink = xa_find(&devlinks, &index, index, DEVLINK_REGISTERED);
25 	if (!devlink || !devlink_try_get(devlink))
26 		devlink = NULL;
27 	rcu_read_unlock();
28 	return devlink;
29 }
30 
31 /* devlink_rels xarray contains 1:1 relationships between
32  * devlink object and related nested devlink instance.
33  * The xarray index is used to get the nested object from
34  * the nested-in object code.
35  */
36 static DEFINE_XARRAY_FLAGS(devlink_rels, XA_FLAGS_ALLOC1);
37 
38 #define DEVLINK_REL_IN_USE XA_MARK_0
39 
40 struct devlink_rel {
41 	u32 index;
42 	refcount_t refcount;
43 	u32 devlink_index;
44 	struct {
45 		u32 devlink_index;
46 		u32 obj_index;
47 		devlink_rel_notify_cb_t *notify_cb;
48 		devlink_rel_cleanup_cb_t *cleanup_cb;
49 		struct delayed_work notify_work;
50 	} nested_in;
51 };
52 
53 static void devlink_rel_free(struct devlink_rel *rel)
54 {
55 	xa_erase(&devlink_rels, rel->index);
56 	kfree(rel);
57 }
58 
59 static void __devlink_rel_get(struct devlink_rel *rel)
60 {
61 	refcount_inc(&rel->refcount);
62 }
63 
64 static void __devlink_rel_put(struct devlink_rel *rel)
65 {
66 	if (refcount_dec_and_test(&rel->refcount))
67 		devlink_rel_free(rel);
68 }
69 
70 static void devlink_rel_nested_in_notify_work(struct work_struct *work)
71 {
72 	struct devlink_rel *rel = container_of(work, struct devlink_rel,
73 					       nested_in.notify_work.work);
74 	struct devlink *devlink;
75 
76 	devlink = devlinks_xa_get(rel->nested_in.devlink_index);
77 	if (!devlink)
78 		goto rel_put;
79 	if (!devl_trylock(devlink)) {
80 		devlink_put(devlink);
81 		goto reschedule_work;
82 	}
83 	if (!devl_is_registered(devlink)) {
84 		devl_unlock(devlink);
85 		devlink_put(devlink);
86 		goto rel_put;
87 	}
88 	if (!xa_get_mark(&devlink_rels, rel->index, DEVLINK_REL_IN_USE))
89 		rel->nested_in.cleanup_cb(devlink, rel->nested_in.obj_index, rel->index);
90 	rel->nested_in.notify_cb(devlink, rel->nested_in.obj_index);
91 	devl_unlock(devlink);
92 	devlink_put(devlink);
93 
94 rel_put:
95 	__devlink_rel_put(rel);
96 	return;
97 
98 reschedule_work:
99 	schedule_delayed_work(&rel->nested_in.notify_work, 1);
100 }
101 
102 static void devlink_rel_nested_in_notify_work_schedule(struct devlink_rel *rel)
103 {
104 	__devlink_rel_get(rel);
105 	schedule_delayed_work(&rel->nested_in.notify_work, 0);
106 }
107 
108 static struct devlink_rel *devlink_rel_alloc(void)
109 {
110 	struct devlink_rel *rel;
111 	static u32 next;
112 	int err;
113 
114 	rel = kzalloc(sizeof(*rel), GFP_KERNEL);
115 	if (!rel)
116 		return ERR_PTR(-ENOMEM);
117 
118 	err = xa_alloc_cyclic(&devlink_rels, &rel->index, rel,
119 			      xa_limit_32b, &next, GFP_KERNEL);
120 	if (err) {
121 		kfree(rel);
122 		return ERR_PTR(err);
123 	}
124 
125 	refcount_set(&rel->refcount, 1);
126 	INIT_DELAYED_WORK(&rel->nested_in.notify_work,
127 			  &devlink_rel_nested_in_notify_work);
128 	return rel;
129 }
130 
131 static void devlink_rel_put(struct devlink *devlink)
132 {
133 	struct devlink_rel *rel = devlink->rel;
134 
135 	if (!rel)
136 		return;
137 	xa_clear_mark(&devlink_rels, rel->index, DEVLINK_REL_IN_USE);
138 	devlink_rel_nested_in_notify_work_schedule(rel);
139 	__devlink_rel_put(rel);
140 	devlink->rel = NULL;
141 }
142 
143 void devlink_rel_nested_in_clear(u32 rel_index)
144 {
145 	xa_clear_mark(&devlink_rels, rel_index, DEVLINK_REL_IN_USE);
146 }
147 
148 int devlink_rel_nested_in_add(u32 *rel_index, u32 devlink_index,
149 			      u32 obj_index, devlink_rel_notify_cb_t *notify_cb,
150 			      devlink_rel_cleanup_cb_t *cleanup_cb,
151 			      struct devlink *devlink)
152 {
153 	struct devlink_rel *rel = devlink_rel_alloc();
154 
155 	ASSERT_DEVLINK_NOT_REGISTERED(devlink);
156 
157 	if (IS_ERR(rel))
158 		return PTR_ERR(rel);
159 
160 	rel->devlink_index = devlink->index;
161 	rel->nested_in.devlink_index = devlink_index;
162 	rel->nested_in.obj_index = obj_index;
163 	rel->nested_in.notify_cb = notify_cb;
164 	rel->nested_in.cleanup_cb = cleanup_cb;
165 	*rel_index = rel->index;
166 	xa_set_mark(&devlink_rels, rel->index, DEVLINK_REL_IN_USE);
167 	devlink->rel = rel;
168 	return 0;
169 }
170 
171 /**
172  * devlink_rel_nested_in_notify - Notify the object this devlink
173  *				  instance is nested in.
174  * @devlink: devlink
175  *
176  * This is called upon network namespace change of devlink instance.
177  * In case this devlink instance is nested in another devlink object,
178  * a notification of a change of this object should be sent
179  * over netlink. The parent devlink instance lock needs to be
180  * taken during the notification preparation.
181  * However, since the devlink lock of nested instance is held here,
182  * we would end with wrong devlink instance lock ordering and
183  * deadlock. Therefore the work is utilized to avoid that.
184  */
185 void devlink_rel_nested_in_notify(struct devlink *devlink)
186 {
187 	struct devlink_rel *rel = devlink->rel;
188 
189 	if (!rel)
190 		return;
191 	devlink_rel_nested_in_notify_work_schedule(rel);
192 }
193 
194 static struct devlink_rel *devlink_rel_find(unsigned long rel_index)
195 {
196 	return xa_find(&devlink_rels, &rel_index, rel_index,
197 		       DEVLINK_REL_IN_USE);
198 }
199 
200 static struct devlink *devlink_rel_devlink_get(u32 rel_index)
201 {
202 	struct devlink_rel *rel;
203 	u32 devlink_index;
204 
205 	if (!rel_index)
206 		return NULL;
207 	xa_lock(&devlink_rels);
208 	rel = devlink_rel_find(rel_index);
209 	if (rel)
210 		devlink_index = rel->devlink_index;
211 	xa_unlock(&devlink_rels);
212 	if (!rel)
213 		return NULL;
214 	return devlinks_xa_get(devlink_index);
215 }
216 
217 int devlink_rel_devlink_handle_put(struct sk_buff *msg, struct devlink *devlink,
218 				   u32 rel_index, int attrtype,
219 				   bool *msg_updated)
220 {
221 	struct net *net = devlink_net(devlink);
222 	struct devlink *rel_devlink;
223 	int err;
224 
225 	rel_devlink = devlink_rel_devlink_get(rel_index);
226 	if (!rel_devlink)
227 		return 0;
228 	err = devlink_nl_put_nested_handle(msg, net, rel_devlink, attrtype);
229 	devlink_put(rel_devlink);
230 	if (!err && msg_updated)
231 		*msg_updated = true;
232 	return err;
233 }
234 
235 void *devlink_priv(struct devlink *devlink)
236 {
237 	return &devlink->priv;
238 }
239 EXPORT_SYMBOL_GPL(devlink_priv);
240 
241 struct devlink *priv_to_devlink(void *priv)
242 {
243 	return container_of(priv, struct devlink, priv);
244 }
245 EXPORT_SYMBOL_GPL(priv_to_devlink);
246 
247 struct device *devlink_to_dev(const struct devlink *devlink)
248 {
249 	return devlink->dev;
250 }
251 EXPORT_SYMBOL_GPL(devlink_to_dev);
252 
253 struct net *devlink_net(const struct devlink *devlink)
254 {
255 	return read_pnet(&devlink->_net);
256 }
257 EXPORT_SYMBOL_GPL(devlink_net);
258 
259 void devl_assert_locked(struct devlink *devlink)
260 {
261 	lockdep_assert_held(&devlink->lock);
262 }
263 EXPORT_SYMBOL_GPL(devl_assert_locked);
264 
265 #ifdef CONFIG_LOCKDEP
266 /* For use in conjunction with LOCKDEP only e.g. rcu_dereference_protected() */
267 bool devl_lock_is_held(struct devlink *devlink)
268 {
269 	return lockdep_is_held(&devlink->lock);
270 }
271 EXPORT_SYMBOL_GPL(devl_lock_is_held);
272 #endif
273 
274 void devl_lock(struct devlink *devlink)
275 {
276 	mutex_lock(&devlink->lock);
277 }
278 EXPORT_SYMBOL_GPL(devl_lock);
279 
280 int devl_trylock(struct devlink *devlink)
281 {
282 	return mutex_trylock(&devlink->lock);
283 }
284 EXPORT_SYMBOL_GPL(devl_trylock);
285 
286 void devl_unlock(struct devlink *devlink)
287 {
288 	mutex_unlock(&devlink->lock);
289 }
290 EXPORT_SYMBOL_GPL(devl_unlock);
291 
292 /**
293  * devlink_try_get() - try to obtain a reference on a devlink instance
294  * @devlink: instance to reference
295  *
296  * Obtain a reference on a devlink instance. A reference on a devlink instance
297  * only implies that it's safe to take the instance lock. It does not imply
298  * that the instance is registered, use devl_is_registered() after taking
299  * the instance lock to check registration status.
300  */
301 struct devlink *__must_check devlink_try_get(struct devlink *devlink)
302 {
303 	if (refcount_inc_not_zero(&devlink->refcount))
304 		return devlink;
305 	return NULL;
306 }
307 
308 static void devlink_release(struct work_struct *work)
309 {
310 	struct devlink *devlink;
311 
312 	devlink = container_of(to_rcu_work(work), struct devlink, rwork);
313 
314 	mutex_destroy(&devlink->lock);
315 	lockdep_unregister_key(&devlink->lock_key);
316 	put_device(devlink->dev);
317 	kvfree(devlink);
318 }
319 
320 void devlink_put(struct devlink *devlink)
321 {
322 	if (refcount_dec_and_test(&devlink->refcount))
323 		queue_rcu_work(system_wq, &devlink->rwork);
324 }
325 
326 struct devlink *devlinks_xa_find_get(struct net *net, unsigned long *indexp)
327 {
328 	struct devlink *devlink = NULL;
329 
330 	rcu_read_lock();
331 retry:
332 	devlink = xa_find(&devlinks, indexp, ULONG_MAX, DEVLINK_REGISTERED);
333 	if (!devlink)
334 		goto unlock;
335 
336 	if (!devlink_try_get(devlink))
337 		goto next;
338 	if (!net_eq(devlink_net(devlink), net)) {
339 		devlink_put(devlink);
340 		goto next;
341 	}
342 unlock:
343 	rcu_read_unlock();
344 	return devlink;
345 
346 next:
347 	(*indexp)++;
348 	goto retry;
349 }
350 
351 /**
352  * devl_register - Register devlink instance
353  * @devlink: devlink
354  */
355 int devl_register(struct devlink *devlink)
356 {
357 	ASSERT_DEVLINK_NOT_REGISTERED(devlink);
358 	devl_assert_locked(devlink);
359 
360 	xa_set_mark(&devlinks, devlink->index, DEVLINK_REGISTERED);
361 	devlink_notify_register(devlink);
362 	devlink_rel_nested_in_notify(devlink);
363 
364 	return 0;
365 }
366 EXPORT_SYMBOL_GPL(devl_register);
367 
368 void devlink_register(struct devlink *devlink)
369 {
370 	devl_lock(devlink);
371 	devl_register(devlink);
372 	devl_unlock(devlink);
373 }
374 EXPORT_SYMBOL_GPL(devlink_register);
375 
376 /**
377  * devl_unregister - Unregister devlink instance
378  * @devlink: devlink
379  */
380 void devl_unregister(struct devlink *devlink)
381 {
382 	ASSERT_DEVLINK_REGISTERED(devlink);
383 	devl_assert_locked(devlink);
384 
385 	devlink_notify_unregister(devlink);
386 	xa_clear_mark(&devlinks, devlink->index, DEVLINK_REGISTERED);
387 	devlink_rel_put(devlink);
388 }
389 EXPORT_SYMBOL_GPL(devl_unregister);
390 
391 void devlink_unregister(struct devlink *devlink)
392 {
393 	devl_lock(devlink);
394 	devl_unregister(devlink);
395 	devl_unlock(devlink);
396 }
397 EXPORT_SYMBOL_GPL(devlink_unregister);
398 
399 /**
400  *	devlink_alloc_ns - Allocate new devlink instance resources
401  *	in specific namespace
402  *
403  *	@ops: ops
404  *	@priv_size: size of user private data
405  *	@net: net namespace
406  *	@dev: parent device
407  *
408  *	Allocate new devlink instance resources, including devlink index
409  *	and name.
410  */
411 struct devlink *devlink_alloc_ns(const struct devlink_ops *ops,
412 				 size_t priv_size, struct net *net,
413 				 struct device *dev)
414 {
415 	struct devlink *devlink;
416 	static u32 last_id;
417 	int ret;
418 
419 	WARN_ON(!ops || !dev);
420 	if (!devlink_reload_actions_valid(ops))
421 		return NULL;
422 
423 	devlink = kvzalloc(struct_size(devlink, priv, priv_size), GFP_KERNEL);
424 	if (!devlink)
425 		return NULL;
426 
427 	ret = xa_alloc_cyclic(&devlinks, &devlink->index, devlink, xa_limit_31b,
428 			      &last_id, GFP_KERNEL);
429 	if (ret < 0)
430 		goto err_xa_alloc;
431 
432 	devlink->dev = get_device(dev);
433 	devlink->ops = ops;
434 	xa_init_flags(&devlink->ports, XA_FLAGS_ALLOC);
435 	xa_init_flags(&devlink->params, XA_FLAGS_ALLOC);
436 	xa_init_flags(&devlink->snapshot_ids, XA_FLAGS_ALLOC);
437 	xa_init_flags(&devlink->nested_rels, XA_FLAGS_ALLOC);
438 	write_pnet(&devlink->_net, net);
439 	INIT_LIST_HEAD(&devlink->rate_list);
440 	INIT_LIST_HEAD(&devlink->linecard_list);
441 	INIT_LIST_HEAD(&devlink->sb_list);
442 	INIT_LIST_HEAD_RCU(&devlink->dpipe_table_list);
443 	INIT_LIST_HEAD(&devlink->resource_list);
444 	INIT_LIST_HEAD(&devlink->region_list);
445 	INIT_LIST_HEAD(&devlink->reporter_list);
446 	INIT_LIST_HEAD(&devlink->trap_list);
447 	INIT_LIST_HEAD(&devlink->trap_group_list);
448 	INIT_LIST_HEAD(&devlink->trap_policer_list);
449 	INIT_RCU_WORK(&devlink->rwork, devlink_release);
450 	lockdep_register_key(&devlink->lock_key);
451 	mutex_init(&devlink->lock);
452 	lockdep_set_class(&devlink->lock, &devlink->lock_key);
453 	refcount_set(&devlink->refcount, 1);
454 
455 	return devlink;
456 
457 err_xa_alloc:
458 	kvfree(devlink);
459 	return NULL;
460 }
461 EXPORT_SYMBOL_GPL(devlink_alloc_ns);
462 
463 /**
464  *	devlink_free - Free devlink instance resources
465  *
466  *	@devlink: devlink
467  */
468 void devlink_free(struct devlink *devlink)
469 {
470 	ASSERT_DEVLINK_NOT_REGISTERED(devlink);
471 
472 	WARN_ON(!list_empty(&devlink->trap_policer_list));
473 	WARN_ON(!list_empty(&devlink->trap_group_list));
474 	WARN_ON(!list_empty(&devlink->trap_list));
475 	WARN_ON(!list_empty(&devlink->reporter_list));
476 	WARN_ON(!list_empty(&devlink->region_list));
477 	WARN_ON(!list_empty(&devlink->resource_list));
478 	WARN_ON(!list_empty(&devlink->dpipe_table_list));
479 	WARN_ON(!list_empty(&devlink->sb_list));
480 	WARN_ON(!list_empty(&devlink->rate_list));
481 	WARN_ON(!list_empty(&devlink->linecard_list));
482 	WARN_ON(!xa_empty(&devlink->ports));
483 
484 	xa_destroy(&devlink->nested_rels);
485 	xa_destroy(&devlink->snapshot_ids);
486 	xa_destroy(&devlink->params);
487 	xa_destroy(&devlink->ports);
488 
489 	xa_erase(&devlinks, devlink->index);
490 
491 	devlink_put(devlink);
492 }
493 EXPORT_SYMBOL_GPL(devlink_free);
494 
495 static void __net_exit devlink_pernet_pre_exit(struct net *net)
496 {
497 	struct devlink *devlink;
498 	u32 actions_performed;
499 	unsigned long index;
500 	int err;
501 
502 	/* In case network namespace is getting destroyed, reload
503 	 * all devlink instances from this namespace into init_net.
504 	 */
505 	devlinks_xa_for_each_registered_get(net, index, devlink) {
506 		devl_dev_lock(devlink, true);
507 		err = 0;
508 		if (devl_is_registered(devlink))
509 			err = devlink_reload(devlink, &init_net,
510 					     DEVLINK_RELOAD_ACTION_DRIVER_REINIT,
511 					     DEVLINK_RELOAD_LIMIT_UNSPEC,
512 					     &actions_performed, NULL);
513 		devl_dev_unlock(devlink, true);
514 		devlink_put(devlink);
515 		if (err && err != -EOPNOTSUPP)
516 			pr_warn("Failed to reload devlink instance into init_net\n");
517 	}
518 }
519 
520 static struct pernet_operations devlink_pernet_ops __net_initdata = {
521 	.pre_exit = devlink_pernet_pre_exit,
522 };
523 
524 static struct notifier_block devlink_port_netdevice_nb = {
525 	.notifier_call = devlink_port_netdevice_event,
526 };
527 
528 static int __init devlink_init(void)
529 {
530 	int err;
531 
532 	err = register_pernet_subsys(&devlink_pernet_ops);
533 	if (err)
534 		goto out;
535 	err = genl_register_family(&devlink_nl_family);
536 	if (err)
537 		goto out_unreg_pernet_subsys;
538 	err = register_netdevice_notifier(&devlink_port_netdevice_nb);
539 	if (!err)
540 		return 0;
541 
542 	genl_unregister_family(&devlink_nl_family);
543 
544 out_unreg_pernet_subsys:
545 	unregister_pernet_subsys(&devlink_pernet_ops);
546 out:
547 	WARN_ON(err);
548 	return err;
549 }
550 
551 subsys_initcall(devlink_init);
552