xref: /linux/net/devlink/health.c (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
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 #include <net/sock.h>
9 #include <trace/events/devlink.h>
10 #include "devl_internal.h"
11 
12 struct devlink_fmsg_item {
13 	struct list_head list;
14 	int attrtype;
15 	u8 nla_type;
16 	u16 len;
17 	int value[];
18 };
19 
20 struct devlink_fmsg {
21 	struct list_head item_list;
22 	int err; /* first error encountered on some devlink_fmsg_XXX() call */
23 	bool putting_binary; /* This flag forces enclosing of binary data
24 			      * in an array brackets. It forces using
25 			      * of designated API:
26 			      * devlink_fmsg_binary_pair_nest_start()
27 			      * devlink_fmsg_binary_pair_nest_end()
28 			      */
29 };
30 
31 static struct devlink_fmsg *devlink_fmsg_alloc(void)
32 {
33 	struct devlink_fmsg *fmsg;
34 
35 	fmsg = kzalloc(sizeof(*fmsg), GFP_KERNEL);
36 	if (!fmsg)
37 		return NULL;
38 
39 	INIT_LIST_HEAD(&fmsg->item_list);
40 
41 	return fmsg;
42 }
43 
44 static void devlink_fmsg_free(struct devlink_fmsg *fmsg)
45 {
46 	struct devlink_fmsg_item *item, *tmp;
47 
48 	list_for_each_entry_safe(item, tmp, &fmsg->item_list, list) {
49 		list_del(&item->list);
50 		kfree(item);
51 	}
52 	kfree(fmsg);
53 }
54 
55 struct devlink_health_reporter {
56 	struct list_head list;
57 	void *priv;
58 	const struct devlink_health_reporter_ops *ops;
59 	struct devlink *devlink;
60 	struct devlink_port *devlink_port;
61 	struct devlink_fmsg *dump_fmsg;
62 	u64 graceful_period;
63 	bool auto_recover;
64 	bool auto_dump;
65 	u8 health_state;
66 	u64 dump_ts;
67 	u64 dump_real_ts;
68 	u64 error_count;
69 	u64 recovery_count;
70 	u64 last_recovery_ts;
71 };
72 
73 void *
74 devlink_health_reporter_priv(struct devlink_health_reporter *reporter)
75 {
76 	return reporter->priv;
77 }
78 EXPORT_SYMBOL_GPL(devlink_health_reporter_priv);
79 
80 static struct devlink_health_reporter *
81 __devlink_health_reporter_find_by_name(struct list_head *reporter_list,
82 				       const char *reporter_name)
83 {
84 	struct devlink_health_reporter *reporter;
85 
86 	list_for_each_entry(reporter, reporter_list, list)
87 		if (!strcmp(reporter->ops->name, reporter_name))
88 			return reporter;
89 	return NULL;
90 }
91 
92 static struct devlink_health_reporter *
93 devlink_health_reporter_find_by_name(struct devlink *devlink,
94 				     const char *reporter_name)
95 {
96 	return __devlink_health_reporter_find_by_name(&devlink->reporter_list,
97 						      reporter_name);
98 }
99 
100 static struct devlink_health_reporter *
101 devlink_port_health_reporter_find_by_name(struct devlink_port *devlink_port,
102 					  const char *reporter_name)
103 {
104 	return __devlink_health_reporter_find_by_name(&devlink_port->reporter_list,
105 						      reporter_name);
106 }
107 
108 static struct devlink_health_reporter *
109 __devlink_health_reporter_create(struct devlink *devlink,
110 				 const struct devlink_health_reporter_ops *ops,
111 				 u64 graceful_period, void *priv)
112 {
113 	struct devlink_health_reporter *reporter;
114 
115 	if (WARN_ON(graceful_period && !ops->recover))
116 		return ERR_PTR(-EINVAL);
117 
118 	reporter = kzalloc(sizeof(*reporter), GFP_KERNEL);
119 	if (!reporter)
120 		return ERR_PTR(-ENOMEM);
121 
122 	reporter->priv = priv;
123 	reporter->ops = ops;
124 	reporter->devlink = devlink;
125 	reporter->graceful_period = graceful_period;
126 	reporter->auto_recover = !!ops->recover;
127 	reporter->auto_dump = !!ops->dump;
128 	return reporter;
129 }
130 
131 /**
132  * devl_port_health_reporter_create() - create devlink health reporter for
133  *                                      specified port instance
134  *
135  * @port: devlink_port to which health reports will relate
136  * @ops: devlink health reporter ops
137  * @graceful_period: min time (in msec) between recovery attempts
138  * @priv: driver priv pointer
139  */
140 struct devlink_health_reporter *
141 devl_port_health_reporter_create(struct devlink_port *port,
142 				 const struct devlink_health_reporter_ops *ops,
143 				 u64 graceful_period, void *priv)
144 {
145 	struct devlink_health_reporter *reporter;
146 
147 	devl_assert_locked(port->devlink);
148 
149 	if (__devlink_health_reporter_find_by_name(&port->reporter_list,
150 						   ops->name))
151 		return ERR_PTR(-EEXIST);
152 
153 	reporter = __devlink_health_reporter_create(port->devlink, ops,
154 						    graceful_period, priv);
155 	if (IS_ERR(reporter))
156 		return reporter;
157 
158 	reporter->devlink_port = port;
159 	list_add_tail(&reporter->list, &port->reporter_list);
160 	return reporter;
161 }
162 EXPORT_SYMBOL_GPL(devl_port_health_reporter_create);
163 
164 struct devlink_health_reporter *
165 devlink_port_health_reporter_create(struct devlink_port *port,
166 				    const struct devlink_health_reporter_ops *ops,
167 				    u64 graceful_period, void *priv)
168 {
169 	struct devlink_health_reporter *reporter;
170 	struct devlink *devlink = port->devlink;
171 
172 	devl_lock(devlink);
173 	reporter = devl_port_health_reporter_create(port, ops,
174 						    graceful_period, priv);
175 	devl_unlock(devlink);
176 	return reporter;
177 }
178 EXPORT_SYMBOL_GPL(devlink_port_health_reporter_create);
179 
180 /**
181  * devl_health_reporter_create - create devlink health reporter
182  *
183  * @devlink: devlink instance which the health reports will relate
184  * @ops: devlink health reporter ops
185  * @graceful_period: min time (in msec) between recovery attempts
186  * @priv: driver priv pointer
187  */
188 struct devlink_health_reporter *
189 devl_health_reporter_create(struct devlink *devlink,
190 			    const struct devlink_health_reporter_ops *ops,
191 			    u64 graceful_period, void *priv)
192 {
193 	struct devlink_health_reporter *reporter;
194 
195 	devl_assert_locked(devlink);
196 
197 	if (devlink_health_reporter_find_by_name(devlink, ops->name))
198 		return ERR_PTR(-EEXIST);
199 
200 	reporter = __devlink_health_reporter_create(devlink, ops,
201 						    graceful_period, priv);
202 	if (IS_ERR(reporter))
203 		return reporter;
204 
205 	list_add_tail(&reporter->list, &devlink->reporter_list);
206 	return reporter;
207 }
208 EXPORT_SYMBOL_GPL(devl_health_reporter_create);
209 
210 struct devlink_health_reporter *
211 devlink_health_reporter_create(struct devlink *devlink,
212 			       const struct devlink_health_reporter_ops *ops,
213 			       u64 graceful_period, void *priv)
214 {
215 	struct devlink_health_reporter *reporter;
216 
217 	devl_lock(devlink);
218 	reporter = devl_health_reporter_create(devlink, ops,
219 					       graceful_period, priv);
220 	devl_unlock(devlink);
221 	return reporter;
222 }
223 EXPORT_SYMBOL_GPL(devlink_health_reporter_create);
224 
225 static void
226 devlink_health_reporter_free(struct devlink_health_reporter *reporter)
227 {
228 	if (reporter->dump_fmsg)
229 		devlink_fmsg_free(reporter->dump_fmsg);
230 	kfree(reporter);
231 }
232 
233 /**
234  * devl_health_reporter_destroy() - destroy devlink health reporter
235  *
236  * @reporter: devlink health reporter to destroy
237  */
238 void
239 devl_health_reporter_destroy(struct devlink_health_reporter *reporter)
240 {
241 	devl_assert_locked(reporter->devlink);
242 
243 	list_del(&reporter->list);
244 	devlink_health_reporter_free(reporter);
245 }
246 EXPORT_SYMBOL_GPL(devl_health_reporter_destroy);
247 
248 void
249 devlink_health_reporter_destroy(struct devlink_health_reporter *reporter)
250 {
251 	struct devlink *devlink = reporter->devlink;
252 
253 	devl_lock(devlink);
254 	devl_health_reporter_destroy(reporter);
255 	devl_unlock(devlink);
256 }
257 EXPORT_SYMBOL_GPL(devlink_health_reporter_destroy);
258 
259 static int
260 devlink_nl_health_reporter_fill(struct sk_buff *msg,
261 				struct devlink_health_reporter *reporter,
262 				enum devlink_command cmd, u32 portid,
263 				u32 seq, int flags)
264 {
265 	struct devlink *devlink = reporter->devlink;
266 	struct nlattr *reporter_attr;
267 	void *hdr;
268 
269 	hdr = genlmsg_put(msg, portid, seq, &devlink_nl_family, flags, cmd);
270 	if (!hdr)
271 		return -EMSGSIZE;
272 
273 	if (devlink_nl_put_handle(msg, devlink))
274 		goto genlmsg_cancel;
275 
276 	if (reporter->devlink_port) {
277 		if (nla_put_u32(msg, DEVLINK_ATTR_PORT_INDEX, reporter->devlink_port->index))
278 			goto genlmsg_cancel;
279 	}
280 	reporter_attr = nla_nest_start_noflag(msg,
281 					      DEVLINK_ATTR_HEALTH_REPORTER);
282 	if (!reporter_attr)
283 		goto genlmsg_cancel;
284 	if (nla_put_string(msg, DEVLINK_ATTR_HEALTH_REPORTER_NAME,
285 			   reporter->ops->name))
286 		goto reporter_nest_cancel;
287 	if (nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_STATE,
288 		       reporter->health_state))
289 		goto reporter_nest_cancel;
290 	if (nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_ERR_COUNT,
291 			      reporter->error_count, DEVLINK_ATTR_PAD))
292 		goto reporter_nest_cancel;
293 	if (nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_RECOVER_COUNT,
294 			      reporter->recovery_count, DEVLINK_ATTR_PAD))
295 		goto reporter_nest_cancel;
296 	if (reporter->ops->recover &&
297 	    nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD,
298 			      reporter->graceful_period,
299 			      DEVLINK_ATTR_PAD))
300 		goto reporter_nest_cancel;
301 	if (reporter->ops->recover &&
302 	    nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER,
303 		       reporter->auto_recover))
304 		goto reporter_nest_cancel;
305 	if (reporter->dump_fmsg &&
306 	    nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS,
307 			      jiffies_to_msecs(reporter->dump_ts),
308 			      DEVLINK_ATTR_PAD))
309 		goto reporter_nest_cancel;
310 	if (reporter->dump_fmsg &&
311 	    nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS_NS,
312 			      reporter->dump_real_ts, DEVLINK_ATTR_PAD))
313 		goto reporter_nest_cancel;
314 	if (reporter->ops->dump &&
315 	    nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP,
316 		       reporter->auto_dump))
317 		goto reporter_nest_cancel;
318 
319 	nla_nest_end(msg, reporter_attr);
320 	genlmsg_end(msg, hdr);
321 	return 0;
322 
323 reporter_nest_cancel:
324 	nla_nest_cancel(msg, reporter_attr);
325 genlmsg_cancel:
326 	genlmsg_cancel(msg, hdr);
327 	return -EMSGSIZE;
328 }
329 
330 static struct devlink_health_reporter *
331 devlink_health_reporter_get_from_attrs(struct devlink *devlink,
332 				       struct nlattr **attrs)
333 {
334 	struct devlink_port *devlink_port;
335 	char *reporter_name;
336 
337 	if (!attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME])
338 		return NULL;
339 
340 	reporter_name = nla_data(attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME]);
341 	devlink_port = devlink_port_get_from_attrs(devlink, attrs);
342 	if (IS_ERR(devlink_port))
343 		return devlink_health_reporter_find_by_name(devlink,
344 							    reporter_name);
345 	else
346 		return devlink_port_health_reporter_find_by_name(devlink_port,
347 								 reporter_name);
348 }
349 
350 static struct devlink_health_reporter *
351 devlink_health_reporter_get_from_info(struct devlink *devlink,
352 				      struct genl_info *info)
353 {
354 	return devlink_health_reporter_get_from_attrs(devlink, info->attrs);
355 }
356 
357 int devlink_nl_health_reporter_get_doit(struct sk_buff *skb,
358 					struct genl_info *info)
359 {
360 	struct devlink *devlink = info->user_ptr[0];
361 	struct devlink_health_reporter *reporter;
362 	struct sk_buff *msg;
363 	int err;
364 
365 	reporter = devlink_health_reporter_get_from_info(devlink, info);
366 	if (!reporter)
367 		return -EINVAL;
368 
369 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
370 	if (!msg)
371 		return -ENOMEM;
372 
373 	err = devlink_nl_health_reporter_fill(msg, reporter,
374 					      DEVLINK_CMD_HEALTH_REPORTER_GET,
375 					      info->snd_portid, info->snd_seq,
376 					      0);
377 	if (err) {
378 		nlmsg_free(msg);
379 		return err;
380 	}
381 
382 	return genlmsg_reply(msg, info);
383 }
384 
385 static int devlink_nl_health_reporter_get_dump_one(struct sk_buff *msg,
386 						   struct devlink *devlink,
387 						   struct netlink_callback *cb,
388 						   int flags)
389 {
390 	struct devlink_nl_dump_state *state = devlink_dump_state(cb);
391 	const struct genl_info *info = genl_info_dump(cb);
392 	struct devlink_health_reporter *reporter;
393 	unsigned long port_index_end = ULONG_MAX;
394 	struct nlattr **attrs = info->attrs;
395 	unsigned long port_index_start = 0;
396 	struct devlink_port *port;
397 	unsigned long port_index;
398 	int idx = 0;
399 	int err;
400 
401 	if (attrs && attrs[DEVLINK_ATTR_PORT_INDEX]) {
402 		port_index_start = nla_get_u32(attrs[DEVLINK_ATTR_PORT_INDEX]);
403 		port_index_end = port_index_start;
404 		flags |= NLM_F_DUMP_FILTERED;
405 		goto per_port_dump;
406 	}
407 
408 	list_for_each_entry(reporter, &devlink->reporter_list, list) {
409 		if (idx < state->idx) {
410 			idx++;
411 			continue;
412 		}
413 		err = devlink_nl_health_reporter_fill(msg, reporter,
414 						      DEVLINK_CMD_HEALTH_REPORTER_GET,
415 						      NETLINK_CB(cb->skb).portid,
416 						      cb->nlh->nlmsg_seq,
417 						      flags);
418 		if (err) {
419 			state->idx = idx;
420 			return err;
421 		}
422 		idx++;
423 	}
424 per_port_dump:
425 	xa_for_each_range(&devlink->ports, port_index, port,
426 			  port_index_start, port_index_end) {
427 		list_for_each_entry(reporter, &port->reporter_list, list) {
428 			if (idx < state->idx) {
429 				idx++;
430 				continue;
431 			}
432 			err = devlink_nl_health_reporter_fill(msg, reporter,
433 							      DEVLINK_CMD_HEALTH_REPORTER_GET,
434 							      NETLINK_CB(cb->skb).portid,
435 							      cb->nlh->nlmsg_seq,
436 							      flags);
437 			if (err) {
438 				state->idx = idx;
439 				return err;
440 			}
441 			idx++;
442 		}
443 	}
444 
445 	return 0;
446 }
447 
448 int devlink_nl_health_reporter_get_dumpit(struct sk_buff *skb,
449 					  struct netlink_callback *cb)
450 {
451 	return devlink_nl_dumpit(skb, cb,
452 				 devlink_nl_health_reporter_get_dump_one);
453 }
454 
455 int devlink_nl_health_reporter_set_doit(struct sk_buff *skb,
456 					struct genl_info *info)
457 {
458 	struct devlink *devlink = info->user_ptr[0];
459 	struct devlink_health_reporter *reporter;
460 
461 	reporter = devlink_health_reporter_get_from_info(devlink, info);
462 	if (!reporter)
463 		return -EINVAL;
464 
465 	if (!reporter->ops->recover &&
466 	    (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD] ||
467 	     info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]))
468 		return -EOPNOTSUPP;
469 
470 	if (!reporter->ops->dump &&
471 	    info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
472 		return -EOPNOTSUPP;
473 
474 	if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD])
475 		reporter->graceful_period =
476 			nla_get_u64(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD]);
477 
478 	if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER])
479 		reporter->auto_recover =
480 			nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]);
481 
482 	if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
483 		reporter->auto_dump =
484 		nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP]);
485 
486 	return 0;
487 }
488 
489 static void devlink_recover_notify(struct devlink_health_reporter *reporter,
490 				   enum devlink_command cmd)
491 {
492 	struct devlink *devlink = reporter->devlink;
493 	struct devlink_obj_desc desc;
494 	struct sk_buff *msg;
495 	int err;
496 
497 	WARN_ON(cmd != DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
498 	ASSERT_DEVLINK_REGISTERED(devlink);
499 
500 	if (!devlink_nl_notify_need(devlink))
501 		return;
502 
503 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
504 	if (!msg)
505 		return;
506 
507 	err = devlink_nl_health_reporter_fill(msg, reporter, cmd, 0, 0, 0);
508 	if (err) {
509 		nlmsg_free(msg);
510 		return;
511 	}
512 
513 	devlink_nl_obj_desc_init(&desc, devlink);
514 	if (reporter->devlink_port)
515 		devlink_nl_obj_desc_port_set(&desc, reporter->devlink_port);
516 	devlink_nl_notify_send_desc(devlink, msg, &desc);
517 }
518 
519 void
520 devlink_health_reporter_recovery_done(struct devlink_health_reporter *reporter)
521 {
522 	reporter->recovery_count++;
523 	reporter->last_recovery_ts = jiffies;
524 }
525 EXPORT_SYMBOL_GPL(devlink_health_reporter_recovery_done);
526 
527 static int
528 devlink_health_reporter_recover(struct devlink_health_reporter *reporter,
529 				void *priv_ctx, struct netlink_ext_ack *extack)
530 {
531 	int err;
532 
533 	if (reporter->health_state == DEVLINK_HEALTH_REPORTER_STATE_HEALTHY)
534 		return 0;
535 
536 	if (!reporter->ops->recover)
537 		return -EOPNOTSUPP;
538 
539 	err = reporter->ops->recover(reporter, priv_ctx, extack);
540 	if (err)
541 		return err;
542 
543 	devlink_health_reporter_recovery_done(reporter);
544 	reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_HEALTHY;
545 	devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
546 
547 	return 0;
548 }
549 
550 static void
551 devlink_health_dump_clear(struct devlink_health_reporter *reporter)
552 {
553 	if (!reporter->dump_fmsg)
554 		return;
555 	devlink_fmsg_free(reporter->dump_fmsg);
556 	reporter->dump_fmsg = NULL;
557 }
558 
559 static int devlink_health_do_dump(struct devlink_health_reporter *reporter,
560 				  void *priv_ctx,
561 				  struct netlink_ext_ack *extack)
562 {
563 	int err;
564 
565 	if (!reporter->ops->dump)
566 		return 0;
567 
568 	if (reporter->dump_fmsg)
569 		return 0;
570 
571 	reporter->dump_fmsg = devlink_fmsg_alloc();
572 	if (!reporter->dump_fmsg)
573 		return -ENOMEM;
574 
575 	devlink_fmsg_obj_nest_start(reporter->dump_fmsg);
576 
577 	err = reporter->ops->dump(reporter, reporter->dump_fmsg,
578 				  priv_ctx, extack);
579 	if (err)
580 		goto dump_err;
581 
582 	devlink_fmsg_obj_nest_end(reporter->dump_fmsg);
583 	err = reporter->dump_fmsg->err;
584 	if (err)
585 		goto dump_err;
586 
587 	reporter->dump_ts = jiffies;
588 	reporter->dump_real_ts = ktime_get_real_ns();
589 
590 	return 0;
591 
592 dump_err:
593 	devlink_health_dump_clear(reporter);
594 	return err;
595 }
596 
597 int devlink_health_report(struct devlink_health_reporter *reporter,
598 			  const char *msg, void *priv_ctx)
599 {
600 	enum devlink_health_reporter_state prev_health_state;
601 	struct devlink *devlink = reporter->devlink;
602 	unsigned long recover_ts_threshold;
603 	int ret;
604 
605 	/* write a log message of the current error */
606 	WARN_ON(!msg);
607 	trace_devlink_health_report(devlink, reporter->ops->name, msg);
608 	reporter->error_count++;
609 	prev_health_state = reporter->health_state;
610 	reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_ERROR;
611 	devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
612 
613 	/* abort if the previous error wasn't recovered */
614 	recover_ts_threshold = reporter->last_recovery_ts +
615 			       msecs_to_jiffies(reporter->graceful_period);
616 	if (reporter->auto_recover &&
617 	    (prev_health_state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY ||
618 	     (reporter->last_recovery_ts && reporter->recovery_count &&
619 	      time_is_after_jiffies(recover_ts_threshold)))) {
620 		trace_devlink_health_recover_aborted(devlink,
621 						     reporter->ops->name,
622 						     reporter->health_state,
623 						     jiffies -
624 						     reporter->last_recovery_ts);
625 		return -ECANCELED;
626 	}
627 
628 	if (reporter->auto_dump) {
629 		devl_lock(devlink);
630 		/* store current dump of current error, for later analysis */
631 		devlink_health_do_dump(reporter, priv_ctx, NULL);
632 		devl_unlock(devlink);
633 	}
634 
635 	if (!reporter->auto_recover)
636 		return 0;
637 
638 	devl_lock(devlink);
639 	ret = devlink_health_reporter_recover(reporter, priv_ctx, NULL);
640 	devl_unlock(devlink);
641 
642 	return ret;
643 }
644 EXPORT_SYMBOL_GPL(devlink_health_report);
645 
646 void
647 devlink_health_reporter_state_update(struct devlink_health_reporter *reporter,
648 				     enum devlink_health_reporter_state state)
649 {
650 	if (WARN_ON(state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY &&
651 		    state != DEVLINK_HEALTH_REPORTER_STATE_ERROR))
652 		return;
653 
654 	if (reporter->health_state == state)
655 		return;
656 
657 	reporter->health_state = state;
658 	trace_devlink_health_reporter_state_update(reporter->devlink,
659 						   reporter->ops->name, state);
660 	devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
661 }
662 EXPORT_SYMBOL_GPL(devlink_health_reporter_state_update);
663 
664 int devlink_nl_health_reporter_recover_doit(struct sk_buff *skb,
665 					    struct genl_info *info)
666 {
667 	struct devlink *devlink = info->user_ptr[0];
668 	struct devlink_health_reporter *reporter;
669 
670 	reporter = devlink_health_reporter_get_from_info(devlink, info);
671 	if (!reporter)
672 		return -EINVAL;
673 
674 	return devlink_health_reporter_recover(reporter, NULL, info->extack);
675 }
676 
677 static void devlink_fmsg_err_if_binary(struct devlink_fmsg *fmsg)
678 {
679 	if (!fmsg->err && fmsg->putting_binary)
680 		fmsg->err = -EINVAL;
681 }
682 
683 static void devlink_fmsg_nest_common(struct devlink_fmsg *fmsg, int attrtype)
684 {
685 	struct devlink_fmsg_item *item;
686 
687 	if (fmsg->err)
688 		return;
689 
690 	item = kzalloc(sizeof(*item), GFP_KERNEL);
691 	if (!item) {
692 		fmsg->err = -ENOMEM;
693 		return;
694 	}
695 
696 	item->attrtype = attrtype;
697 	list_add_tail(&item->list, &fmsg->item_list);
698 }
699 
700 void devlink_fmsg_obj_nest_start(struct devlink_fmsg *fmsg)
701 {
702 	devlink_fmsg_err_if_binary(fmsg);
703 	devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_OBJ_NEST_START);
704 }
705 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_start);
706 
707 static void devlink_fmsg_nest_end(struct devlink_fmsg *fmsg)
708 {
709 	devlink_fmsg_err_if_binary(fmsg);
710 	devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_NEST_END);
711 }
712 
713 void devlink_fmsg_obj_nest_end(struct devlink_fmsg *fmsg)
714 {
715 	devlink_fmsg_nest_end(fmsg);
716 }
717 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_end);
718 
719 #define DEVLINK_FMSG_MAX_SIZE (GENLMSG_DEFAULT_SIZE - GENL_HDRLEN - NLA_HDRLEN)
720 
721 static void devlink_fmsg_put_name(struct devlink_fmsg *fmsg, const char *name)
722 {
723 	struct devlink_fmsg_item *item;
724 
725 	devlink_fmsg_err_if_binary(fmsg);
726 	if (fmsg->err)
727 		return;
728 
729 	if (strlen(name) + 1 > DEVLINK_FMSG_MAX_SIZE) {
730 		fmsg->err = -EMSGSIZE;
731 		return;
732 	}
733 
734 	item = kzalloc(sizeof(*item) + strlen(name) + 1, GFP_KERNEL);
735 	if (!item) {
736 		fmsg->err = -ENOMEM;
737 		return;
738 	}
739 
740 	item->nla_type = NLA_NUL_STRING;
741 	item->len = strlen(name) + 1;
742 	item->attrtype = DEVLINK_ATTR_FMSG_OBJ_NAME;
743 	memcpy(&item->value, name, item->len);
744 	list_add_tail(&item->list, &fmsg->item_list);
745 }
746 
747 void devlink_fmsg_pair_nest_start(struct devlink_fmsg *fmsg, const char *name)
748 {
749 	devlink_fmsg_err_if_binary(fmsg);
750 	devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_PAIR_NEST_START);
751 	devlink_fmsg_put_name(fmsg, name);
752 }
753 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_start);
754 
755 void devlink_fmsg_pair_nest_end(struct devlink_fmsg *fmsg)
756 {
757 	devlink_fmsg_nest_end(fmsg);
758 }
759 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_end);
760 
761 void devlink_fmsg_arr_pair_nest_start(struct devlink_fmsg *fmsg,
762 				      const char *name)
763 {
764 	devlink_fmsg_pair_nest_start(fmsg, name);
765 	devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_ARR_NEST_START);
766 }
767 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_start);
768 
769 void devlink_fmsg_arr_pair_nest_end(struct devlink_fmsg *fmsg)
770 {
771 	devlink_fmsg_nest_end(fmsg);
772 	devlink_fmsg_nest_end(fmsg);
773 }
774 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_end);
775 
776 void devlink_fmsg_binary_pair_nest_start(struct devlink_fmsg *fmsg,
777 					 const char *name)
778 {
779 	devlink_fmsg_arr_pair_nest_start(fmsg, name);
780 	fmsg->putting_binary = true;
781 }
782 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_start);
783 
784 void devlink_fmsg_binary_pair_nest_end(struct devlink_fmsg *fmsg)
785 {
786 	if (fmsg->err)
787 		return;
788 
789 	if (!fmsg->putting_binary)
790 		fmsg->err = -EINVAL;
791 
792 	fmsg->putting_binary = false;
793 	devlink_fmsg_arr_pair_nest_end(fmsg);
794 }
795 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_end);
796 
797 static void devlink_fmsg_put_value(struct devlink_fmsg *fmsg,
798 				   const void *value, u16 value_len,
799 				   u8 value_nla_type)
800 {
801 	struct devlink_fmsg_item *item;
802 
803 	if (fmsg->err)
804 		return;
805 
806 	if (value_len > DEVLINK_FMSG_MAX_SIZE) {
807 		fmsg->err = -EMSGSIZE;
808 		return;
809 	}
810 
811 	item = kzalloc(sizeof(*item) + value_len, GFP_KERNEL);
812 	if (!item) {
813 		fmsg->err = -ENOMEM;
814 		return;
815 	}
816 
817 	item->nla_type = value_nla_type;
818 	item->len = value_len;
819 	item->attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
820 	memcpy(&item->value, value, item->len);
821 	list_add_tail(&item->list, &fmsg->item_list);
822 }
823 
824 static void devlink_fmsg_bool_put(struct devlink_fmsg *fmsg, bool value)
825 {
826 	devlink_fmsg_err_if_binary(fmsg);
827 	devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_FLAG);
828 }
829 
830 static void devlink_fmsg_u8_put(struct devlink_fmsg *fmsg, u8 value)
831 {
832 	devlink_fmsg_err_if_binary(fmsg);
833 	devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U8);
834 }
835 
836 void devlink_fmsg_u32_put(struct devlink_fmsg *fmsg, u32 value)
837 {
838 	devlink_fmsg_err_if_binary(fmsg);
839 	devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U32);
840 }
841 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_put);
842 
843 static void devlink_fmsg_u64_put(struct devlink_fmsg *fmsg, u64 value)
844 {
845 	devlink_fmsg_err_if_binary(fmsg);
846 	devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U64);
847 }
848 
849 void devlink_fmsg_string_put(struct devlink_fmsg *fmsg, const char *value)
850 {
851 	devlink_fmsg_err_if_binary(fmsg);
852 	devlink_fmsg_put_value(fmsg, value, strlen(value) + 1, NLA_NUL_STRING);
853 }
854 EXPORT_SYMBOL_GPL(devlink_fmsg_string_put);
855 
856 void devlink_fmsg_binary_put(struct devlink_fmsg *fmsg, const void *value,
857 			     u16 value_len)
858 {
859 	if (!fmsg->err && !fmsg->putting_binary)
860 		fmsg->err = -EINVAL;
861 
862 	devlink_fmsg_put_value(fmsg, value, value_len, NLA_BINARY);
863 }
864 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_put);
865 
866 void devlink_fmsg_bool_pair_put(struct devlink_fmsg *fmsg, const char *name,
867 				bool value)
868 {
869 	devlink_fmsg_pair_nest_start(fmsg, name);
870 	devlink_fmsg_bool_put(fmsg, value);
871 	devlink_fmsg_pair_nest_end(fmsg);
872 }
873 EXPORT_SYMBOL_GPL(devlink_fmsg_bool_pair_put);
874 
875 void devlink_fmsg_u8_pair_put(struct devlink_fmsg *fmsg, const char *name,
876 			      u8 value)
877 {
878 	devlink_fmsg_pair_nest_start(fmsg, name);
879 	devlink_fmsg_u8_put(fmsg, value);
880 	devlink_fmsg_pair_nest_end(fmsg);
881 }
882 EXPORT_SYMBOL_GPL(devlink_fmsg_u8_pair_put);
883 
884 void devlink_fmsg_u32_pair_put(struct devlink_fmsg *fmsg, const char *name,
885 			       u32 value)
886 {
887 	devlink_fmsg_pair_nest_start(fmsg, name);
888 	devlink_fmsg_u32_put(fmsg, value);
889 	devlink_fmsg_pair_nest_end(fmsg);
890 }
891 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_pair_put);
892 
893 void devlink_fmsg_u64_pair_put(struct devlink_fmsg *fmsg, const char *name,
894 			       u64 value)
895 {
896 	devlink_fmsg_pair_nest_start(fmsg, name);
897 	devlink_fmsg_u64_put(fmsg, value);
898 	devlink_fmsg_pair_nest_end(fmsg);
899 }
900 EXPORT_SYMBOL_GPL(devlink_fmsg_u64_pair_put);
901 
902 void devlink_fmsg_string_pair_put(struct devlink_fmsg *fmsg, const char *name,
903 				  const char *value)
904 {
905 	devlink_fmsg_pair_nest_start(fmsg, name);
906 	devlink_fmsg_string_put(fmsg, value);
907 	devlink_fmsg_pair_nest_end(fmsg);
908 }
909 EXPORT_SYMBOL_GPL(devlink_fmsg_string_pair_put);
910 
911 void devlink_fmsg_binary_pair_put(struct devlink_fmsg *fmsg, const char *name,
912 				  const void *value, u32 value_len)
913 {
914 	u32 data_size;
915 	u32 offset;
916 
917 	devlink_fmsg_binary_pair_nest_start(fmsg, name);
918 
919 	for (offset = 0; offset < value_len; offset += data_size) {
920 		data_size = value_len - offset;
921 		if (data_size > DEVLINK_FMSG_MAX_SIZE)
922 			data_size = DEVLINK_FMSG_MAX_SIZE;
923 
924 		devlink_fmsg_binary_put(fmsg, value + offset, data_size);
925 	}
926 
927 	devlink_fmsg_binary_pair_nest_end(fmsg);
928 	fmsg->putting_binary = false;
929 }
930 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_put);
931 
932 static int
933 devlink_fmsg_item_fill_type(struct devlink_fmsg_item *msg, struct sk_buff *skb)
934 {
935 	switch (msg->nla_type) {
936 	case NLA_FLAG:
937 	case NLA_U8:
938 	case NLA_U32:
939 	case NLA_U64:
940 	case NLA_NUL_STRING:
941 	case NLA_BINARY:
942 		return nla_put_u8(skb, DEVLINK_ATTR_FMSG_OBJ_VALUE_TYPE,
943 				  msg->nla_type);
944 	default:
945 		return -EINVAL;
946 	}
947 }
948 
949 static int
950 devlink_fmsg_item_fill_data(struct devlink_fmsg_item *msg, struct sk_buff *skb)
951 {
952 	int attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
953 	u8 tmp;
954 
955 	switch (msg->nla_type) {
956 	case NLA_FLAG:
957 		/* Always provide flag data, regardless of its value */
958 		tmp = *(bool *)msg->value;
959 
960 		return nla_put_u8(skb, attrtype, tmp);
961 	case NLA_U8:
962 		return nla_put_u8(skb, attrtype, *(u8 *)msg->value);
963 	case NLA_U32:
964 		return nla_put_u32(skb, attrtype, *(u32 *)msg->value);
965 	case NLA_U64:
966 		return nla_put_u64_64bit(skb, attrtype, *(u64 *)msg->value,
967 					 DEVLINK_ATTR_PAD);
968 	case NLA_NUL_STRING:
969 		return nla_put_string(skb, attrtype, (char *)&msg->value);
970 	case NLA_BINARY:
971 		return nla_put(skb, attrtype, msg->len, (void *)&msg->value);
972 	default:
973 		return -EINVAL;
974 	}
975 }
976 
977 static int
978 devlink_fmsg_prepare_skb(struct devlink_fmsg *fmsg, struct sk_buff *skb,
979 			 int *start)
980 {
981 	struct devlink_fmsg_item *item;
982 	struct nlattr *fmsg_nlattr;
983 	int err = 0;
984 	int i = 0;
985 
986 	fmsg_nlattr = nla_nest_start_noflag(skb, DEVLINK_ATTR_FMSG);
987 	if (!fmsg_nlattr)
988 		return -EMSGSIZE;
989 
990 	list_for_each_entry(item, &fmsg->item_list, list) {
991 		if (i < *start) {
992 			i++;
993 			continue;
994 		}
995 
996 		switch (item->attrtype) {
997 		case DEVLINK_ATTR_FMSG_OBJ_NEST_START:
998 		case DEVLINK_ATTR_FMSG_PAIR_NEST_START:
999 		case DEVLINK_ATTR_FMSG_ARR_NEST_START:
1000 		case DEVLINK_ATTR_FMSG_NEST_END:
1001 			err = nla_put_flag(skb, item->attrtype);
1002 			break;
1003 		case DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA:
1004 			err = devlink_fmsg_item_fill_type(item, skb);
1005 			if (err)
1006 				break;
1007 			err = devlink_fmsg_item_fill_data(item, skb);
1008 			break;
1009 		case DEVLINK_ATTR_FMSG_OBJ_NAME:
1010 			err = nla_put_string(skb, item->attrtype,
1011 					     (char *)&item->value);
1012 			break;
1013 		default:
1014 			err = -EINVAL;
1015 			break;
1016 		}
1017 		if (!err)
1018 			*start = ++i;
1019 		else
1020 			break;
1021 	}
1022 
1023 	nla_nest_end(skb, fmsg_nlattr);
1024 	return err;
1025 }
1026 
1027 static int devlink_fmsg_snd(struct devlink_fmsg *fmsg,
1028 			    struct genl_info *info,
1029 			    enum devlink_command cmd, int flags)
1030 {
1031 	struct nlmsghdr *nlh;
1032 	struct sk_buff *skb;
1033 	bool last = false;
1034 	int index = 0;
1035 	void *hdr;
1036 	int err;
1037 
1038 	if (fmsg->err)
1039 		return fmsg->err;
1040 
1041 	while (!last) {
1042 		int tmp_index = index;
1043 
1044 		skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1045 		if (!skb)
1046 			return -ENOMEM;
1047 
1048 		hdr = genlmsg_put(skb, info->snd_portid, info->snd_seq,
1049 				  &devlink_nl_family, flags | NLM_F_MULTI, cmd);
1050 		if (!hdr) {
1051 			err = -EMSGSIZE;
1052 			goto nla_put_failure;
1053 		}
1054 
1055 		err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1056 		if (!err)
1057 			last = true;
1058 		else if (err != -EMSGSIZE || tmp_index == index)
1059 			goto nla_put_failure;
1060 
1061 		genlmsg_end(skb, hdr);
1062 		err = genlmsg_reply(skb, info);
1063 		if (err)
1064 			return err;
1065 	}
1066 
1067 	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1068 	if (!skb)
1069 		return -ENOMEM;
1070 	nlh = nlmsg_put(skb, info->snd_portid, info->snd_seq,
1071 			NLMSG_DONE, 0, flags | NLM_F_MULTI);
1072 	if (!nlh) {
1073 		err = -EMSGSIZE;
1074 		goto nla_put_failure;
1075 	}
1076 
1077 	return genlmsg_reply(skb, info);
1078 
1079 nla_put_failure:
1080 	nlmsg_free(skb);
1081 	return err;
1082 }
1083 
1084 static int devlink_fmsg_dumpit(struct devlink_fmsg *fmsg, struct sk_buff *skb,
1085 			       struct netlink_callback *cb,
1086 			       enum devlink_command cmd)
1087 {
1088 	struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1089 	int index = state->idx;
1090 	int tmp_index = index;
1091 	void *hdr;
1092 	int err;
1093 
1094 	if (fmsg->err)
1095 		return fmsg->err;
1096 
1097 	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1098 			  &devlink_nl_family, NLM_F_ACK | NLM_F_MULTI, cmd);
1099 	if (!hdr) {
1100 		err = -EMSGSIZE;
1101 		goto nla_put_failure;
1102 	}
1103 
1104 	err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1105 	if ((err && err != -EMSGSIZE) || tmp_index == index)
1106 		goto nla_put_failure;
1107 
1108 	state->idx = index;
1109 	genlmsg_end(skb, hdr);
1110 	return skb->len;
1111 
1112 nla_put_failure:
1113 	genlmsg_cancel(skb, hdr);
1114 	return err;
1115 }
1116 
1117 int devlink_nl_health_reporter_diagnose_doit(struct sk_buff *skb,
1118 					     struct genl_info *info)
1119 {
1120 	struct devlink *devlink = info->user_ptr[0];
1121 	struct devlink_health_reporter *reporter;
1122 	struct devlink_fmsg *fmsg;
1123 	int err;
1124 
1125 	reporter = devlink_health_reporter_get_from_info(devlink, info);
1126 	if (!reporter)
1127 		return -EINVAL;
1128 
1129 	if (!reporter->ops->diagnose)
1130 		return -EOPNOTSUPP;
1131 
1132 	fmsg = devlink_fmsg_alloc();
1133 	if (!fmsg)
1134 		return -ENOMEM;
1135 
1136 	devlink_fmsg_obj_nest_start(fmsg);
1137 
1138 	err = reporter->ops->diagnose(reporter, fmsg, info->extack);
1139 	if (err)
1140 		goto out;
1141 
1142 	devlink_fmsg_obj_nest_end(fmsg);
1143 
1144 	err = devlink_fmsg_snd(fmsg, info,
1145 			       DEVLINK_CMD_HEALTH_REPORTER_DIAGNOSE, 0);
1146 
1147 out:
1148 	devlink_fmsg_free(fmsg);
1149 	return err;
1150 }
1151 
1152 static struct devlink_health_reporter *
1153 devlink_health_reporter_get_from_cb_lock(struct netlink_callback *cb)
1154 {
1155 	const struct genl_info *info = genl_info_dump(cb);
1156 	struct devlink_health_reporter *reporter;
1157 	struct nlattr **attrs = info->attrs;
1158 	struct devlink *devlink;
1159 
1160 	devlink = devlink_get_from_attrs_lock(sock_net(cb->skb->sk), attrs,
1161 					      false);
1162 	if (IS_ERR(devlink))
1163 		return NULL;
1164 
1165 	reporter = devlink_health_reporter_get_from_attrs(devlink, attrs);
1166 	if (!reporter) {
1167 		devl_unlock(devlink);
1168 		devlink_put(devlink);
1169 	}
1170 	return reporter;
1171 }
1172 
1173 int devlink_nl_health_reporter_dump_get_dumpit(struct sk_buff *skb,
1174 					       struct netlink_callback *cb)
1175 {
1176 	struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1177 	struct devlink_health_reporter *reporter;
1178 	struct devlink *devlink;
1179 	int err;
1180 
1181 	reporter = devlink_health_reporter_get_from_cb_lock(cb);
1182 	if (!reporter)
1183 		return -EINVAL;
1184 
1185 	devlink = reporter->devlink;
1186 	if (!reporter->ops->dump) {
1187 		devl_unlock(devlink);
1188 		devlink_put(devlink);
1189 		return -EOPNOTSUPP;
1190 	}
1191 
1192 	if (!state->idx) {
1193 		err = devlink_health_do_dump(reporter, NULL, cb->extack);
1194 		if (err)
1195 			goto unlock;
1196 		state->dump_ts = reporter->dump_ts;
1197 	}
1198 	if (!reporter->dump_fmsg || state->dump_ts != reporter->dump_ts) {
1199 		NL_SET_ERR_MSG(cb->extack, "Dump trampled, please retry");
1200 		err = -EAGAIN;
1201 		goto unlock;
1202 	}
1203 
1204 	err = devlink_fmsg_dumpit(reporter->dump_fmsg, skb, cb,
1205 				  DEVLINK_CMD_HEALTH_REPORTER_DUMP_GET);
1206 unlock:
1207 	devl_unlock(devlink);
1208 	devlink_put(devlink);
1209 	return err;
1210 }
1211 
1212 int devlink_nl_health_reporter_dump_clear_doit(struct sk_buff *skb,
1213 					       struct genl_info *info)
1214 {
1215 	struct devlink *devlink = info->user_ptr[0];
1216 	struct devlink_health_reporter *reporter;
1217 
1218 	reporter = devlink_health_reporter_get_from_info(devlink, info);
1219 	if (!reporter)
1220 		return -EINVAL;
1221 
1222 	if (!reporter->ops->dump)
1223 		return -EOPNOTSUPP;
1224 
1225 	devlink_health_dump_clear(reporter);
1226 	return 0;
1227 }
1228 
1229 int devlink_nl_health_reporter_test_doit(struct sk_buff *skb,
1230 					 struct genl_info *info)
1231 {
1232 	struct devlink *devlink = info->user_ptr[0];
1233 	struct devlink_health_reporter *reporter;
1234 
1235 	reporter = devlink_health_reporter_get_from_info(devlink, info);
1236 	if (!reporter)
1237 		return -EINVAL;
1238 
1239 	if (!reporter->ops->test)
1240 		return -EOPNOTSUPP;
1241 
1242 	return reporter->ops->test(reporter, info->extack);
1243 }
1244