xref: /linux/net/core/drop_monitor.c (revision 439f392084f8f7f59ab9d47a9579185accefe1d8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Monitoring code for network dropped packet alerts
4  *
5  * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/string.h>
13 #include <linux/if_arp.h>
14 #include <linux/inetdevice.h>
15 #include <linux/inet.h>
16 #include <linux/interrupt.h>
17 #include <linux/netpoll.h>
18 #include <linux/sched.h>
19 #include <linux/delay.h>
20 #include <linux/types.h>
21 #include <linux/workqueue.h>
22 #include <linux/netlink.h>
23 #include <linux/net_dropmon.h>
24 #include <linux/bitfield.h>
25 #include <linux/percpu.h>
26 #include <linux/timer.h>
27 #include <linux/bitops.h>
28 #include <linux/slab.h>
29 #include <linux/module.h>
30 #include <net/genetlink.h>
31 #include <net/netevent.h>
32 #include <net/flow_offload.h>
33 #include <net/dropreason.h>
34 #include <net/devlink.h>
35 
36 #include <trace/events/skb.h>
37 #include <trace/events/napi.h>
38 #include <trace/events/devlink.h>
39 
40 #include <linux/unaligned.h>
41 
42 #define TRACE_ON 1
43 #define TRACE_OFF 0
44 
45 /*
46  * Globals, our netlink socket pointer
47  * and the work handle that will send up
48  * netlink alerts
49  */
50 static int trace_state = TRACE_OFF;
51 static bool monitor_hw;
52 
53 /* net_dm_mutex
54  *
55  * An overall lock guarding every operation coming from userspace.
56  */
57 static DEFINE_MUTEX(net_dm_mutex);
58 
59 struct net_dm_stats {
60 	u64_stats_t dropped;
61 	struct u64_stats_sync syncp;
62 };
63 
64 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
65 
66 struct net_dm_hw_entry {
67 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
68 	u32 count;
69 };
70 
71 struct net_dm_hw_entries {
72 	u32 num_entries;
73 	struct net_dm_hw_entry entries[];
74 };
75 
76 struct per_cpu_dm_data {
77 	raw_spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
78 					 * 'send_timer'
79 					 */
80 	union {
81 		struct sk_buff			*skb;
82 		struct net_dm_hw_entries	*hw_entries;
83 	};
84 	struct sk_buff_head	drop_queue;
85 	struct work_struct	dm_alert_work;
86 	struct timer_list	send_timer;
87 	struct net_dm_stats	stats;
88 };
89 
90 struct dm_hw_stat_delta {
91 	unsigned long last_rx;
92 	unsigned long last_drop_val;
93 	struct rcu_head rcu;
94 };
95 
96 static struct genl_family net_drop_monitor_family;
97 
98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100 
101 static int dm_hit_limit = 64;
102 static int dm_delay = 1;
103 static unsigned long dm_hw_check_delta = 2*HZ;
104 
105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106 static u32 net_dm_trunc_len;
107 static u32 net_dm_queue_len = 1000;
108 
109 struct net_dm_alert_ops {
110 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
111 				void *location,
112 				enum skb_drop_reason reason,
113 				const struct sock *rx_sk);
114 	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
115 				int work, int budget);
116 	void (*work_item_func)(struct work_struct *work);
117 	void (*hw_work_item_func)(struct work_struct *work);
118 	void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
119 			      struct sk_buff *skb,
120 			      const struct devlink_trap_metadata *metadata);
121 };
122 
123 struct net_dm_skb_cb {
124 	union {
125 		struct devlink_trap_metadata *hw_metadata;
126 		void *pc;
127 	};
128 	enum skb_drop_reason reason;
129 };
130 
131 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
132 
133 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
134 {
135 	size_t al;
136 	struct net_dm_alert_msg *msg;
137 	struct nlattr *nla;
138 	struct sk_buff *skb;
139 	unsigned long flags;
140 	void *msg_header;
141 
142 	al = sizeof(struct net_dm_alert_msg);
143 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
144 
145 	skb = genlmsg_new(nla_total_size(al), GFP_KERNEL);
146 
147 	if (!skb)
148 		goto err;
149 
150 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
151 				 0, NET_DM_CMD_ALERT);
152 	if (!msg_header) {
153 		nlmsg_free(skb);
154 		skb = NULL;
155 		goto err;
156 	}
157 	nla = nla_reserve(skb, NLA_UNSPEC,
158 			  sizeof(struct net_dm_alert_msg));
159 	if (!nla) {
160 		nlmsg_free(skb);
161 		skb = NULL;
162 		goto err;
163 	}
164 	msg = nla_data(nla);
165 	memset(msg, 0, al);
166 	goto out;
167 
168 err:
169 	mod_timer(&data->send_timer, jiffies + HZ / 10);
170 out:
171 	raw_spin_lock_irqsave(&data->lock, flags);
172 	swap(data->skb, skb);
173 	raw_spin_unlock_irqrestore(&data->lock, flags);
174 
175 	if (skb) {
176 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
177 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
178 
179 		genlmsg_end(skb, genlmsg_data(gnlh));
180 	}
181 
182 	return skb;
183 }
184 
185 static const struct genl_multicast_group dropmon_mcgrps[] = {
186 	{ .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, },
187 };
188 
189 static void send_dm_alert(struct work_struct *work)
190 {
191 	struct sk_buff *skb;
192 	struct per_cpu_dm_data *data;
193 
194 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
195 
196 	skb = reset_per_cpu_data(data);
197 
198 	if (skb)
199 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
200 				  0, GFP_KERNEL);
201 }
202 
203 /*
204  * This is the timer function to delay the sending of an alert
205  * in the event that more drops will arrive during the
206  * hysteresis period.
207  */
208 static void sched_send_work(struct timer_list *t)
209 {
210 	struct per_cpu_dm_data *data = timer_container_of(data, t, send_timer);
211 
212 	schedule_work(&data->dm_alert_work);
213 }
214 
215 static void trace_drop_common(struct sk_buff *skb, void *location)
216 {
217 	struct net_dm_alert_msg *msg;
218 	struct net_dm_drop_point *point;
219 	struct nlmsghdr *nlh;
220 	struct nlattr *nla;
221 	int i;
222 	struct sk_buff *dskb;
223 	struct per_cpu_dm_data *data;
224 	unsigned long flags;
225 
226 	local_irq_save(flags);
227 	data = this_cpu_ptr(&dm_cpu_data);
228 	raw_spin_lock(&data->lock);
229 	dskb = data->skb;
230 
231 	if (!dskb)
232 		goto out;
233 
234 	nlh = (struct nlmsghdr *)dskb->data;
235 	nla = genlmsg_data(nlmsg_data(nlh));
236 	msg = nla_data(nla);
237 	point = msg->points;
238 	for (i = 0; i < msg->entries; i++) {
239 		if (!memcmp(&location, &point->pc, sizeof(void *))) {
240 			point->count++;
241 			goto out;
242 		}
243 		point++;
244 	}
245 	if (msg->entries == dm_hit_limit)
246 		goto out;
247 	/*
248 	 * We need to create a new entry
249 	 */
250 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
251 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
252 	memcpy(point->pc, &location, sizeof(void *));
253 	point->count = 1;
254 	msg->entries++;
255 
256 	if (!timer_pending(&data->send_timer)) {
257 		data->send_timer.expires = jiffies + dm_delay * HZ;
258 		add_timer(&data->send_timer);
259 	}
260 
261 out:
262 	raw_spin_unlock_irqrestore(&data->lock, flags);
263 }
264 
265 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb,
266 				void *location,
267 				enum skb_drop_reason reason,
268 				const struct sock *rx_sk)
269 {
270 	trace_drop_common(skb, location);
271 }
272 
273 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
274 				int work, int budget)
275 {
276 	struct net_device *dev = napi->dev;
277 	struct dm_hw_stat_delta *stat;
278 	/*
279 	 * Don't check napi structures with no associated device
280 	 */
281 	if (!dev)
282 		return;
283 
284 	rcu_read_lock();
285 	stat = rcu_dereference(dev->dm_private);
286 	if (stat) {
287 		/*
288 		 * only add a note to our monitor buffer if:
289 		 * 1) its after the last_rx delta
290 		 * 2) our rx_dropped count has gone up
291 		 */
292 		if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
293 		    (dev->stats.rx_dropped != stat->last_drop_val)) {
294 			trace_drop_common(NULL, NULL);
295 			stat->last_drop_val = dev->stats.rx_dropped;
296 			stat->last_rx = jiffies;
297 		}
298 	}
299 	rcu_read_unlock();
300 }
301 
302 static struct net_dm_hw_entries *
303 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
304 {
305 	struct net_dm_hw_entries *hw_entries;
306 	unsigned long flags;
307 
308 	hw_entries = kzalloc_flex(*hw_entries, entries, dm_hit_limit);
309 	if (!hw_entries) {
310 		/* If the memory allocation failed, we try to perform another
311 		 * allocation in 1/10 second. Otherwise, the probe function
312 		 * will constantly bail out.
313 		 */
314 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
315 	}
316 
317 	raw_spin_lock_irqsave(&hw_data->lock, flags);
318 	swap(hw_data->hw_entries, hw_entries);
319 	raw_spin_unlock_irqrestore(&hw_data->lock, flags);
320 
321 	return hw_entries;
322 }
323 
324 static int net_dm_hw_entry_put(struct sk_buff *msg,
325 			       const struct net_dm_hw_entry *hw_entry)
326 {
327 	struct nlattr *attr;
328 
329 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
330 	if (!attr)
331 		return -EMSGSIZE;
332 
333 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
334 		goto nla_put_failure;
335 
336 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
337 		goto nla_put_failure;
338 
339 	nla_nest_end(msg, attr);
340 
341 	return 0;
342 
343 nla_put_failure:
344 	nla_nest_cancel(msg, attr);
345 	return -EMSGSIZE;
346 }
347 
348 static int net_dm_hw_entries_put(struct sk_buff *msg,
349 				 const struct net_dm_hw_entries *hw_entries)
350 {
351 	struct nlattr *attr;
352 	int i;
353 
354 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
355 	if (!attr)
356 		return -EMSGSIZE;
357 
358 	for (i = 0; i < hw_entries->num_entries; i++) {
359 		int rc;
360 
361 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
362 		if (rc)
363 			goto nla_put_failure;
364 	}
365 
366 	nla_nest_end(msg, attr);
367 
368 	return 0;
369 
370 nla_put_failure:
371 	nla_nest_cancel(msg, attr);
372 	return -EMSGSIZE;
373 }
374 
375 static int
376 net_dm_hw_summary_report_fill(struct sk_buff *msg,
377 			      const struct net_dm_hw_entries *hw_entries)
378 {
379 	struct net_dm_alert_msg anc_hdr = { 0 };
380 	void *hdr;
381 	int rc;
382 
383 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
384 			  NET_DM_CMD_ALERT);
385 	if (!hdr)
386 		return -EMSGSIZE;
387 
388 	/* We need to put the ancillary header in order not to break user
389 	 * space.
390 	 */
391 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
392 		goto nla_put_failure;
393 
394 	rc = net_dm_hw_entries_put(msg, hw_entries);
395 	if (rc)
396 		goto nla_put_failure;
397 
398 	genlmsg_end(msg, hdr);
399 
400 	return 0;
401 
402 nla_put_failure:
403 	genlmsg_cancel(msg, hdr);
404 	return -EMSGSIZE;
405 }
406 
407 static void net_dm_hw_summary_work(struct work_struct *work)
408 {
409 	struct net_dm_hw_entries *hw_entries;
410 	struct per_cpu_dm_data *hw_data;
411 	struct sk_buff *msg;
412 	int rc;
413 
414 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
415 
416 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
417 	if (!hw_entries)
418 		return;
419 
420 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
421 	if (!msg)
422 		goto out;
423 
424 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
425 	if (rc) {
426 		nlmsg_free(msg);
427 		goto out;
428 	}
429 
430 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
431 
432 out:
433 	kfree(hw_entries);
434 }
435 
436 static void
437 net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
438 			     struct sk_buff *skb,
439 			     const struct devlink_trap_metadata *metadata)
440 {
441 	struct net_dm_hw_entries *hw_entries;
442 	struct net_dm_hw_entry *hw_entry;
443 	struct per_cpu_dm_data *hw_data;
444 	unsigned long flags;
445 	int i;
446 
447 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
448 		return;
449 
450 	hw_data = raw_cpu_ptr(&dm_hw_cpu_data);
451 	raw_spin_lock_irqsave(&hw_data->lock, flags);
452 	hw_entries = hw_data->hw_entries;
453 
454 	if (!hw_entries)
455 		goto out;
456 
457 	for (i = 0; i < hw_entries->num_entries; i++) {
458 		hw_entry = &hw_entries->entries[i];
459 		if (!strncmp(hw_entry->trap_name, metadata->trap_name,
460 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
461 			hw_entry->count++;
462 			goto out;
463 		}
464 	}
465 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
466 		goto out;
467 
468 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
469 	strscpy(hw_entry->trap_name, metadata->trap_name,
470 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
471 	hw_entry->count = 1;
472 	hw_entries->num_entries++;
473 
474 	if (!timer_pending(&hw_data->send_timer)) {
475 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
476 		add_timer(&hw_data->send_timer);
477 	}
478 
479 out:
480 	raw_spin_unlock_irqrestore(&hw_data->lock, flags);
481 }
482 
483 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
484 	.kfree_skb_probe	= trace_kfree_skb_hit,
485 	.napi_poll_probe	= trace_napi_poll_hit,
486 	.work_item_func		= send_dm_alert,
487 	.hw_work_item_func	= net_dm_hw_summary_work,
488 	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
489 };
490 
491 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
492 					      struct sk_buff *skb,
493 					      void *location,
494 					      enum skb_drop_reason reason,
495 					      const struct sock *rx_sk)
496 {
497 	ktime_t tstamp = ktime_get_real();
498 	struct per_cpu_dm_data *data;
499 	struct net_dm_skb_cb *cb;
500 	struct sk_buff *nskb;
501 	unsigned long flags;
502 
503 	if (!skb_mac_header_was_set(skb))
504 		return;
505 
506 	nskb = skb_clone(skb, GFP_ATOMIC);
507 	if (!nskb)
508 		return;
509 
510 	cb = NET_DM_SKB_CB(nskb);
511 	cb->reason = reason;
512 	cb->pc = location;
513 	/* Override the timestamp because we care about the time when the
514 	 * packet was dropped.
515 	 */
516 	nskb->tstamp = tstamp;
517 
518 	data = raw_cpu_ptr(&dm_cpu_data);
519 
520 	spin_lock_irqsave(&data->drop_queue.lock, flags);
521 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
522 		__skb_queue_tail(&data->drop_queue, nskb);
523 	else
524 		goto unlock_free;
525 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
526 
527 	schedule_work(&data->dm_alert_work);
528 
529 	return;
530 
531 unlock_free:
532 	u64_stats_update_begin(&data->stats.syncp);
533 	u64_stats_inc(&data->stats.dropped);
534 	u64_stats_update_end(&data->stats.syncp);
535 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
536 	consume_skb(nskb);
537 }
538 
539 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
540 					      struct napi_struct *napi,
541 					      int work, int budget)
542 {
543 }
544 
545 static size_t net_dm_in_port_size(void)
546 {
547 	       /* NET_DM_ATTR_IN_PORT nest */
548 	return nla_total_size(0) +
549 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
550 	       nla_total_size(sizeof(u32)) +
551 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
552 	       nla_total_size(IFNAMSIZ + 1);
553 }
554 
555 #define NET_DM_MAX_SYMBOL_LEN 40
556 #define NET_DM_MAX_REASON_LEN 50
557 
558 static size_t net_dm_packet_report_size(size_t payload_len)
559 {
560 	size_t size;
561 
562 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
563 
564 	return NLMSG_ALIGN(size) +
565 	       /* NET_DM_ATTR_ORIGIN */
566 	       nla_total_size(sizeof(u16)) +
567 	       /* NET_DM_ATTR_PC */
568 	       nla_total_size_64bit(sizeof(u64)) +
569 	       /* NET_DM_ATTR_SYMBOL */
570 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
571 	       /* NET_DM_ATTR_IN_PORT */
572 	       net_dm_in_port_size() +
573 	       /* NET_DM_ATTR_TIMESTAMP */
574 	       nla_total_size_64bit(sizeof(u64)) +
575 	       /* NET_DM_ATTR_ORIG_LEN */
576 	       nla_total_size(sizeof(u32)) +
577 	       /* NET_DM_ATTR_PROTO */
578 	       nla_total_size(sizeof(u16)) +
579 	       /* NET_DM_ATTR_REASON */
580 	       nla_total_size(NET_DM_MAX_REASON_LEN + 1) +
581 	       /* NET_DM_ATTR_PAYLOAD */
582 	       nla_total_size(payload_len);
583 }
584 
585 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
586 					    const char *name)
587 {
588 	struct nlattr *attr;
589 
590 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
591 	if (!attr)
592 		return -EMSGSIZE;
593 
594 	if (ifindex &&
595 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
596 		goto nla_put_failure;
597 
598 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
599 		goto nla_put_failure;
600 
601 	nla_nest_end(msg, attr);
602 
603 	return 0;
604 
605 nla_put_failure:
606 	nla_nest_cancel(msg, attr);
607 	return -EMSGSIZE;
608 }
609 
610 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
611 				     size_t payload_len)
612 {
613 	struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb);
614 	const struct drop_reason_list *list = NULL;
615 	unsigned int subsys, subsys_reason;
616 	char buf[NET_DM_MAX_SYMBOL_LEN];
617 	struct nlattr *attr;
618 	void *hdr;
619 	int rc;
620 
621 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
622 			  NET_DM_CMD_PACKET_ALERT);
623 	if (!hdr)
624 		return -EMSGSIZE;
625 
626 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
627 		goto nla_put_failure;
628 
629 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
630 			      NET_DM_ATTR_PAD))
631 		goto nla_put_failure;
632 
633 	rcu_read_lock();
634 	subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK);
635 	if (subsys < SKB_DROP_REASON_SUBSYS_NUM)
636 		list = rcu_dereference(drop_reasons_by_subsys[subsys]);
637 	subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK;
638 	if (!list ||
639 	    subsys_reason >= list->n_reasons ||
640 	    !list->reasons[subsys_reason] ||
641 	    strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) {
642 		list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]);
643 		subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED;
644 	}
645 	if (nla_put_string(msg, NET_DM_ATTR_REASON,
646 			   list->reasons[subsys_reason])) {
647 		rcu_read_unlock();
648 		goto nla_put_failure;
649 	}
650 	rcu_read_unlock();
651 
652 	snprintf(buf, sizeof(buf), "%pS", cb->pc);
653 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
654 		goto nla_put_failure;
655 
656 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
657 	if (rc)
658 		goto nla_put_failure;
659 
660 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
661 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
662 		goto nla_put_failure;
663 
664 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
665 		goto nla_put_failure;
666 
667 	if (!payload_len)
668 		goto out;
669 
670 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
671 		goto nla_put_failure;
672 
673 	attr = __nla_reserve(msg, NET_DM_ATTR_PAYLOAD, payload_len);
674 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
675 		goto nla_put_failure;
676 
677 out:
678 	genlmsg_end(msg, hdr);
679 
680 	return 0;
681 
682 nla_put_failure:
683 	genlmsg_cancel(msg, hdr);
684 	return -EMSGSIZE;
685 }
686 
687 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
688 
689 static void net_dm_packet_report(struct sk_buff *skb)
690 {
691 	struct sk_buff *msg;
692 	size_t payload_len;
693 	int rc;
694 
695 	/* Make sure we start copying the packet from the MAC header */
696 	if (skb->data > skb_mac_header(skb))
697 		skb_push(skb, skb->data - skb_mac_header(skb));
698 	else
699 		skb_pull(skb, skb_mac_header(skb) - skb->data);
700 
701 	/* Ensure packet fits inside a single netlink attribute */
702 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
703 	if (net_dm_trunc_len)
704 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
705 
706 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
707 	if (!msg)
708 		goto out;
709 
710 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
711 	if (rc) {
712 		nlmsg_free(msg);
713 		goto out;
714 	}
715 
716 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
717 
718 out:
719 	consume_skb(skb);
720 }
721 
722 static void net_dm_packet_work(struct work_struct *work)
723 {
724 	struct per_cpu_dm_data *data;
725 	struct sk_buff_head list;
726 	struct sk_buff *skb;
727 	unsigned long flags;
728 
729 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
730 
731 	__skb_queue_head_init(&list);
732 
733 	spin_lock_irqsave(&data->drop_queue.lock, flags);
734 	skb_queue_splice_tail_init(&data->drop_queue, &list);
735 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
736 
737 	while ((skb = __skb_dequeue(&list)))
738 		net_dm_packet_report(skb);
739 }
740 
741 static size_t
742 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
743 {
744 	return hw_metadata->fa_cookie ?
745 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
746 }
747 
748 static size_t
749 net_dm_hw_packet_report_size(size_t payload_len,
750 			     const struct devlink_trap_metadata *hw_metadata)
751 {
752 	size_t size;
753 
754 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
755 
756 	return NLMSG_ALIGN(size) +
757 	       /* NET_DM_ATTR_ORIGIN */
758 	       nla_total_size(sizeof(u16)) +
759 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
760 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
761 	       /* NET_DM_ATTR_HW_TRAP_NAME */
762 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
763 	       /* NET_DM_ATTR_IN_PORT */
764 	       net_dm_in_port_size() +
765 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
766 	       net_dm_flow_action_cookie_size(hw_metadata) +
767 	       /* NET_DM_ATTR_TIMESTAMP */
768 	       nla_total_size_64bit(sizeof(u64)) +
769 	       /* NET_DM_ATTR_ORIG_LEN */
770 	       nla_total_size(sizeof(u32)) +
771 	       /* NET_DM_ATTR_PROTO */
772 	       nla_total_size(sizeof(u16)) +
773 	       /* NET_DM_ATTR_PAYLOAD */
774 	       nla_total_size(payload_len);
775 }
776 
777 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
778 					struct sk_buff *skb, size_t payload_len)
779 {
780 	struct devlink_trap_metadata *hw_metadata;
781 	struct nlattr *attr;
782 	void *hdr;
783 
784 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
785 
786 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
787 			  NET_DM_CMD_PACKET_ALERT);
788 	if (!hdr)
789 		return -EMSGSIZE;
790 
791 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
792 		goto nla_put_failure;
793 
794 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
795 			   hw_metadata->trap_group_name))
796 		goto nla_put_failure;
797 
798 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
799 			   hw_metadata->trap_name))
800 		goto nla_put_failure;
801 
802 	if (hw_metadata->input_dev) {
803 		struct net_device *dev = hw_metadata->input_dev;
804 		int rc;
805 
806 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
807 						      dev->name);
808 		if (rc)
809 			goto nla_put_failure;
810 	}
811 
812 	if (hw_metadata->fa_cookie &&
813 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
814 		    hw_metadata->fa_cookie->cookie_len,
815 		    hw_metadata->fa_cookie->cookie))
816 		goto nla_put_failure;
817 
818 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
819 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
820 		goto nla_put_failure;
821 
822 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
823 		goto nla_put_failure;
824 
825 	if (!payload_len)
826 		goto out;
827 
828 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
829 		goto nla_put_failure;
830 
831 	attr = __nla_reserve(msg, NET_DM_ATTR_PAYLOAD, payload_len);
832 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
833 		goto nla_put_failure;
834 
835 out:
836 	genlmsg_end(msg, hdr);
837 
838 	return 0;
839 
840 nla_put_failure:
841 	genlmsg_cancel(msg, hdr);
842 	return -EMSGSIZE;
843 }
844 
845 static struct devlink_trap_metadata *
846 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
847 {
848 	const struct flow_action_cookie *fa_cookie;
849 	struct devlink_trap_metadata *hw_metadata;
850 	const char *trap_group_name;
851 	const char *trap_name;
852 
853 	hw_metadata = kzalloc_obj(*hw_metadata, GFP_ATOMIC);
854 	if (!hw_metadata)
855 		return NULL;
856 
857 	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
858 	if (!trap_group_name)
859 		goto free_hw_metadata;
860 	hw_metadata->trap_group_name = trap_group_name;
861 
862 	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
863 	if (!trap_name)
864 		goto free_trap_group;
865 	hw_metadata->trap_name = trap_name;
866 
867 	if (metadata->fa_cookie) {
868 		size_t cookie_size = sizeof(*fa_cookie) +
869 				     metadata->fa_cookie->cookie_len;
870 
871 		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
872 				    GFP_ATOMIC);
873 		if (!fa_cookie)
874 			goto free_trap_name;
875 		hw_metadata->fa_cookie = fa_cookie;
876 	}
877 
878 	hw_metadata->input_dev = metadata->input_dev;
879 	netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
880 		    GFP_ATOMIC);
881 
882 	return hw_metadata;
883 
884 free_trap_name:
885 	kfree(trap_name);
886 free_trap_group:
887 	kfree(trap_group_name);
888 free_hw_metadata:
889 	kfree(hw_metadata);
890 	return NULL;
891 }
892 
893 static void
894 net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
895 {
896 	netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
897 	kfree(hw_metadata->fa_cookie);
898 	kfree(hw_metadata->trap_name);
899 	kfree(hw_metadata->trap_group_name);
900 	kfree(hw_metadata);
901 }
902 
903 static void net_dm_hw_packet_report(struct sk_buff *skb)
904 {
905 	struct devlink_trap_metadata *hw_metadata;
906 	struct sk_buff *msg;
907 	size_t payload_len;
908 	int rc;
909 
910 	if (skb->data > skb_mac_header(skb))
911 		skb_push(skb, skb->data - skb_mac_header(skb));
912 	else
913 		skb_pull(skb, skb_mac_header(skb) - skb->data);
914 
915 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
916 	if (net_dm_trunc_len)
917 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
918 
919 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
920 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
921 			GFP_KERNEL);
922 	if (!msg)
923 		goto out;
924 
925 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
926 	if (rc) {
927 		nlmsg_free(msg);
928 		goto out;
929 	}
930 
931 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
932 
933 out:
934 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
935 	consume_skb(skb);
936 }
937 
938 static void net_dm_hw_packet_work(struct work_struct *work)
939 {
940 	struct per_cpu_dm_data *hw_data;
941 	struct sk_buff_head list;
942 	struct sk_buff *skb;
943 	unsigned long flags;
944 
945 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
946 
947 	__skb_queue_head_init(&list);
948 
949 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
950 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
951 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
952 
953 	while ((skb = __skb_dequeue(&list)))
954 		net_dm_hw_packet_report(skb);
955 }
956 
957 static void
958 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
959 			    struct sk_buff *skb,
960 			    const struct devlink_trap_metadata *metadata)
961 {
962 	struct devlink_trap_metadata *n_hw_metadata;
963 	ktime_t tstamp = ktime_get_real();
964 	struct per_cpu_dm_data *hw_data;
965 	struct sk_buff *nskb;
966 	unsigned long flags;
967 
968 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
969 		return;
970 
971 	if (!skb_mac_header_was_set(skb))
972 		return;
973 
974 	nskb = skb_clone(skb, GFP_ATOMIC);
975 	if (!nskb)
976 		return;
977 
978 	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
979 	if (!n_hw_metadata)
980 		goto free;
981 
982 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
983 	nskb->tstamp = tstamp;
984 
985 	hw_data = raw_cpu_ptr(&dm_hw_cpu_data);
986 
987 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
988 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
989 		__skb_queue_tail(&hw_data->drop_queue, nskb);
990 	else
991 		goto unlock_free;
992 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
993 
994 	schedule_work(&hw_data->dm_alert_work);
995 
996 	return;
997 
998 unlock_free:
999 	u64_stats_update_begin(&hw_data->stats.syncp);
1000 	u64_stats_inc(&hw_data->stats.dropped);
1001 	u64_stats_update_end(&hw_data->stats.syncp);
1002 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
1003 	net_dm_hw_metadata_free(n_hw_metadata);
1004 free:
1005 	consume_skb(nskb);
1006 }
1007 
1008 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
1009 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
1010 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
1011 	.work_item_func		= net_dm_packet_work,
1012 	.hw_work_item_func	= net_dm_hw_packet_work,
1013 	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
1014 };
1015 
1016 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1017 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
1018 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
1019 };
1020 
1021 #if IS_ENABLED(CONFIG_NET_DEVLINK)
1022 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1023 {
1024 	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1025 }
1026 
1027 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1028 {
1029 	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1030 	tracepoint_synchronize_unregister();
1031 }
1032 #else
1033 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1034 {
1035 	return -EOPNOTSUPP;
1036 }
1037 
1038 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1039 {
1040 }
1041 #endif
1042 
1043 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1044 {
1045 	const struct net_dm_alert_ops *ops;
1046 	int cpu, rc;
1047 
1048 	if (monitor_hw) {
1049 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1050 		return -EAGAIN;
1051 	}
1052 
1053 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1054 
1055 	if (!try_module_get(THIS_MODULE)) {
1056 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1057 		return -ENODEV;
1058 	}
1059 
1060 	for_each_possible_cpu(cpu) {
1061 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1062 		struct net_dm_hw_entries *hw_entries;
1063 
1064 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1065 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1066 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1067 		kfree(hw_entries);
1068 	}
1069 
1070 	rc = net_dm_hw_probe_register(ops);
1071 	if (rc) {
1072 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1073 		goto err_module_put;
1074 	}
1075 
1076 	monitor_hw = true;
1077 
1078 	return 0;
1079 
1080 err_module_put:
1081 	for_each_possible_cpu(cpu) {
1082 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1083 		struct sk_buff *skb;
1084 
1085 		timer_shutdown_sync(&hw_data->send_timer);
1086 		cancel_work_sync(&hw_data->dm_alert_work);
1087 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1088 			struct devlink_trap_metadata *hw_metadata;
1089 
1090 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1091 			net_dm_hw_metadata_free(hw_metadata);
1092 			consume_skb(skb);
1093 		}
1094 	}
1095 	module_put(THIS_MODULE);
1096 	return rc;
1097 }
1098 
1099 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1100 {
1101 	const struct net_dm_alert_ops *ops;
1102 	int cpu;
1103 
1104 	if (!monitor_hw) {
1105 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1106 		return;
1107 	}
1108 
1109 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1110 
1111 	monitor_hw = false;
1112 
1113 	net_dm_hw_probe_unregister(ops);
1114 
1115 	for_each_possible_cpu(cpu) {
1116 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1117 		struct sk_buff *skb;
1118 
1119 		timer_shutdown_sync(&hw_data->send_timer);
1120 		cancel_work_sync(&hw_data->dm_alert_work);
1121 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1122 			struct devlink_trap_metadata *hw_metadata;
1123 
1124 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1125 			net_dm_hw_metadata_free(hw_metadata);
1126 			consume_skb(skb);
1127 		}
1128 	}
1129 
1130 	module_put(THIS_MODULE);
1131 }
1132 
1133 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1134 {
1135 	const struct net_dm_alert_ops *ops;
1136 	int cpu, rc;
1137 
1138 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1139 
1140 	if (!try_module_get(THIS_MODULE)) {
1141 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1142 		return -ENODEV;
1143 	}
1144 
1145 	for_each_possible_cpu(cpu) {
1146 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1147 		struct sk_buff *skb;
1148 
1149 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1150 		timer_setup(&data->send_timer, sched_send_work, 0);
1151 		/* Allocate a new per-CPU skb for the summary alert message and
1152 		 * free the old one which might contain stale data from
1153 		 * previous tracing.
1154 		 */
1155 		skb = reset_per_cpu_data(data);
1156 		consume_skb(skb);
1157 	}
1158 
1159 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1160 	if (rc) {
1161 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1162 		goto err_module_put;
1163 	}
1164 
1165 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1166 	if (rc) {
1167 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1168 		goto err_unregister_trace;
1169 	}
1170 
1171 	return 0;
1172 
1173 err_unregister_trace:
1174 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1175 	tracepoint_synchronize_unregister();
1176 err_module_put:
1177 	for_each_possible_cpu(cpu) {
1178 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1179 		struct sk_buff *skb;
1180 
1181 		timer_shutdown_sync(&data->send_timer);
1182 		cancel_work_sync(&data->dm_alert_work);
1183 		while ((skb = __skb_dequeue(&data->drop_queue)))
1184 			consume_skb(skb);
1185 	}
1186 	module_put(THIS_MODULE);
1187 	return rc;
1188 }
1189 
1190 static void net_dm_trace_off_set(void)
1191 {
1192 	const struct net_dm_alert_ops *ops;
1193 	int cpu;
1194 
1195 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1196 
1197 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1198 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1199 
1200 	tracepoint_synchronize_unregister();
1201 
1202 	/* Make sure we do not send notifications to user space after request
1203 	 * to stop tracing returns.
1204 	 */
1205 	for_each_possible_cpu(cpu) {
1206 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1207 		struct sk_buff *skb;
1208 
1209 		timer_shutdown_sync(&data->send_timer);
1210 		cancel_work_sync(&data->dm_alert_work);
1211 		while ((skb = __skb_dequeue(&data->drop_queue)))
1212 			consume_skb(skb);
1213 	}
1214 
1215 	module_put(THIS_MODULE);
1216 }
1217 
1218 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1219 {
1220 	int rc = 0;
1221 
1222 	if (state == trace_state) {
1223 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1224 		return -EAGAIN;
1225 	}
1226 
1227 	switch (state) {
1228 	case TRACE_ON:
1229 		rc = net_dm_trace_on_set(extack);
1230 		break;
1231 	case TRACE_OFF:
1232 		net_dm_trace_off_set();
1233 		break;
1234 	default:
1235 		rc = 1;
1236 		break;
1237 	}
1238 
1239 	if (!rc)
1240 		trace_state = state;
1241 	else
1242 		rc = -EINPROGRESS;
1243 
1244 	return rc;
1245 }
1246 
1247 static bool net_dm_is_monitoring(void)
1248 {
1249 	return trace_state == TRACE_ON || monitor_hw;
1250 }
1251 
1252 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1253 					   enum net_dm_alert_mode *p_alert_mode)
1254 {
1255 	u8 val;
1256 
1257 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1258 
1259 	switch (val) {
1260 	case NET_DM_ALERT_MODE_SUMMARY:
1261 	case NET_DM_ALERT_MODE_PACKET:
1262 		*p_alert_mode = val;
1263 		break;
1264 	default:
1265 		return -EINVAL;
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static int net_dm_alert_mode_set(struct genl_info *info)
1272 {
1273 	struct netlink_ext_ack *extack = info->extack;
1274 	enum net_dm_alert_mode alert_mode;
1275 	int rc;
1276 
1277 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1278 		return 0;
1279 
1280 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1281 	if (rc) {
1282 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1283 		return -EINVAL;
1284 	}
1285 
1286 	net_dm_alert_mode = alert_mode;
1287 
1288 	return 0;
1289 }
1290 
1291 static void net_dm_trunc_len_set(struct genl_info *info)
1292 {
1293 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1294 		return;
1295 
1296 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1297 }
1298 
1299 static void net_dm_queue_len_set(struct genl_info *info)
1300 {
1301 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1302 		return;
1303 
1304 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1305 }
1306 
1307 static int net_dm_cmd_config(struct sk_buff *skb,
1308 			struct genl_info *info)
1309 {
1310 	struct netlink_ext_ack *extack = info->extack;
1311 	int rc;
1312 
1313 	if (net_dm_is_monitoring()) {
1314 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1315 		return -EBUSY;
1316 	}
1317 
1318 	rc = net_dm_alert_mode_set(info);
1319 	if (rc)
1320 		return rc;
1321 
1322 	net_dm_trunc_len_set(info);
1323 
1324 	net_dm_queue_len_set(info);
1325 
1326 	return 0;
1327 }
1328 
1329 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1330 				struct netlink_ext_ack *extack)
1331 {
1332 	bool sw_set = false;
1333 	int rc;
1334 
1335 	if (set_sw) {
1336 		rc = set_all_monitor_traces(TRACE_ON, extack);
1337 		if (rc)
1338 			return rc;
1339 		sw_set = true;
1340 	}
1341 
1342 	if (set_hw) {
1343 		rc = net_dm_hw_monitor_start(extack);
1344 		if (rc)
1345 			goto err_monitor_hw;
1346 	}
1347 
1348 	return 0;
1349 
1350 err_monitor_hw:
1351 	if (sw_set)
1352 		set_all_monitor_traces(TRACE_OFF, extack);
1353 	return rc;
1354 }
1355 
1356 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1357 				struct netlink_ext_ack *extack)
1358 {
1359 	if (set_hw)
1360 		net_dm_hw_monitor_stop(extack);
1361 	if (set_sw)
1362 		set_all_monitor_traces(TRACE_OFF, extack);
1363 }
1364 
1365 static int net_dm_cmd_trace(struct sk_buff *skb,
1366 			struct genl_info *info)
1367 {
1368 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1369 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1370 	struct netlink_ext_ack *extack = info->extack;
1371 
1372 	/* To maintain backward compatibility, we start / stop monitoring of
1373 	 * software drops if no flag is specified.
1374 	 */
1375 	if (!set_sw && !set_hw)
1376 		set_sw = true;
1377 
1378 	switch (info->genlhdr->cmd) {
1379 	case NET_DM_CMD_START:
1380 		return net_dm_monitor_start(set_sw, set_hw, extack);
1381 	case NET_DM_CMD_STOP:
1382 		net_dm_monitor_stop(set_sw, set_hw, extack);
1383 		return 0;
1384 	}
1385 
1386 	return -EOPNOTSUPP;
1387 }
1388 
1389 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1390 {
1391 	void *hdr;
1392 
1393 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1394 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1395 	if (!hdr)
1396 		return -EMSGSIZE;
1397 
1398 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1399 		goto nla_put_failure;
1400 
1401 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1402 		goto nla_put_failure;
1403 
1404 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1405 		goto nla_put_failure;
1406 
1407 	genlmsg_end(msg, hdr);
1408 
1409 	return 0;
1410 
1411 nla_put_failure:
1412 	genlmsg_cancel(msg, hdr);
1413 	return -EMSGSIZE;
1414 }
1415 
1416 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1417 {
1418 	struct sk_buff *msg;
1419 	int rc;
1420 
1421 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1422 	if (!msg)
1423 		return -ENOMEM;
1424 
1425 	rc = net_dm_config_fill(msg, info);
1426 	if (rc)
1427 		goto free_msg;
1428 
1429 	return genlmsg_reply(msg, info);
1430 
1431 free_msg:
1432 	nlmsg_free(msg);
1433 	return rc;
1434 }
1435 
1436 static void net_dm_stats_read(struct net_dm_stats *stats)
1437 {
1438 	int cpu;
1439 
1440 	memset(stats, 0, sizeof(*stats));
1441 	for_each_possible_cpu(cpu) {
1442 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1443 		struct net_dm_stats *cpu_stats = &data->stats;
1444 		unsigned int start;
1445 		u64 dropped;
1446 
1447 		do {
1448 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1449 			dropped = u64_stats_read(&cpu_stats->dropped);
1450 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1451 
1452 		u64_stats_add(&stats->dropped, dropped);
1453 	}
1454 }
1455 
1456 static int net_dm_stats_put(struct sk_buff *msg)
1457 {
1458 	struct net_dm_stats stats;
1459 	struct nlattr *attr;
1460 
1461 	net_dm_stats_read(&stats);
1462 
1463 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1464 	if (!attr)
1465 		return -EMSGSIZE;
1466 
1467 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1468 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1469 		goto nla_put_failure;
1470 
1471 	nla_nest_end(msg, attr);
1472 
1473 	return 0;
1474 
1475 nla_put_failure:
1476 	nla_nest_cancel(msg, attr);
1477 	return -EMSGSIZE;
1478 }
1479 
1480 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1481 {
1482 	int cpu;
1483 
1484 	memset(stats, 0, sizeof(*stats));
1485 	for_each_possible_cpu(cpu) {
1486 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1487 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1488 		unsigned int start;
1489 		u64 dropped;
1490 
1491 		do {
1492 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1493 			dropped = u64_stats_read(&cpu_stats->dropped);
1494 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1495 
1496 		u64_stats_add(&stats->dropped, dropped);
1497 	}
1498 }
1499 
1500 static int net_dm_hw_stats_put(struct sk_buff *msg)
1501 {
1502 	struct net_dm_stats stats;
1503 	struct nlattr *attr;
1504 
1505 	net_dm_hw_stats_read(&stats);
1506 
1507 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1508 	if (!attr)
1509 		return -EMSGSIZE;
1510 
1511 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1512 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1513 		goto nla_put_failure;
1514 
1515 	nla_nest_end(msg, attr);
1516 
1517 	return 0;
1518 
1519 nla_put_failure:
1520 	nla_nest_cancel(msg, attr);
1521 	return -EMSGSIZE;
1522 }
1523 
1524 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1525 {
1526 	void *hdr;
1527 	int rc;
1528 
1529 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1530 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1531 	if (!hdr)
1532 		return -EMSGSIZE;
1533 
1534 	rc = net_dm_stats_put(msg);
1535 	if (rc)
1536 		goto nla_put_failure;
1537 
1538 	rc = net_dm_hw_stats_put(msg);
1539 	if (rc)
1540 		goto nla_put_failure;
1541 
1542 	genlmsg_end(msg, hdr);
1543 
1544 	return 0;
1545 
1546 nla_put_failure:
1547 	genlmsg_cancel(msg, hdr);
1548 	return -EMSGSIZE;
1549 }
1550 
1551 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1552 {
1553 	struct sk_buff *msg;
1554 	int rc;
1555 
1556 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1557 	if (!msg)
1558 		return -ENOMEM;
1559 
1560 	rc = net_dm_stats_fill(msg, info);
1561 	if (rc)
1562 		goto free_msg;
1563 
1564 	return genlmsg_reply(msg, info);
1565 
1566 free_msg:
1567 	nlmsg_free(msg);
1568 	return rc;
1569 }
1570 
1571 static int dropmon_net_event(struct notifier_block *ev_block,
1572 			     unsigned long event, void *ptr)
1573 {
1574 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1575 	struct dm_hw_stat_delta *stat;
1576 
1577 	switch (event) {
1578 	case NETDEV_REGISTER:
1579 		if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1580 			break;
1581 		stat = kzalloc_obj(*stat);
1582 		if (!stat)
1583 			break;
1584 
1585 		stat->last_rx = jiffies;
1586 		rcu_assign_pointer(dev->dm_private, stat);
1587 
1588 		break;
1589 	case NETDEV_UNREGISTER:
1590 		stat = rtnl_dereference(dev->dm_private);
1591 		if (stat) {
1592 			rcu_assign_pointer(dev->dm_private, NULL);
1593 			kfree_rcu(stat, rcu);
1594 		}
1595 		break;
1596 	}
1597 	return NOTIFY_DONE;
1598 }
1599 
1600 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1601 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1602 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1603 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1604 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1605 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1606 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1607 };
1608 
1609 static const struct genl_small_ops dropmon_ops[] = {
1610 	{
1611 		.cmd = NET_DM_CMD_CONFIG,
1612 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1613 		.doit = net_dm_cmd_config,
1614 		.flags = GENL_ADMIN_PERM,
1615 	},
1616 	{
1617 		.cmd = NET_DM_CMD_START,
1618 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1619 		.doit = net_dm_cmd_trace,
1620 		.flags = GENL_ADMIN_PERM,
1621 	},
1622 	{
1623 		.cmd = NET_DM_CMD_STOP,
1624 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1625 		.doit = net_dm_cmd_trace,
1626 		.flags = GENL_ADMIN_PERM,
1627 	},
1628 	{
1629 		.cmd = NET_DM_CMD_CONFIG_GET,
1630 		.doit = net_dm_cmd_config_get,
1631 	},
1632 	{
1633 		.cmd = NET_DM_CMD_STATS_GET,
1634 		.doit = net_dm_cmd_stats_get,
1635 	},
1636 };
1637 
1638 static int net_dm_nl_pre_doit(const struct genl_split_ops *ops,
1639 			      struct sk_buff *skb, struct genl_info *info)
1640 {
1641 	mutex_lock(&net_dm_mutex);
1642 
1643 	return 0;
1644 }
1645 
1646 static void net_dm_nl_post_doit(const struct genl_split_ops *ops,
1647 				struct sk_buff *skb, struct genl_info *info)
1648 {
1649 	mutex_unlock(&net_dm_mutex);
1650 }
1651 
1652 static struct genl_family net_drop_monitor_family __ro_after_init = {
1653 	.hdrsize        = 0,
1654 	.name           = "NET_DM",
1655 	.version        = 2,
1656 	.maxattr	= NET_DM_ATTR_MAX,
1657 	.policy		= net_dm_nl_policy,
1658 	.pre_doit	= net_dm_nl_pre_doit,
1659 	.post_doit	= net_dm_nl_post_doit,
1660 	.module		= THIS_MODULE,
1661 	.small_ops	= dropmon_ops,
1662 	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1663 	.resv_start_op	= NET_DM_CMD_STATS_GET + 1,
1664 	.mcgrps		= dropmon_mcgrps,
1665 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1666 };
1667 
1668 static struct notifier_block dropmon_net_notifier = {
1669 	.notifier_call = dropmon_net_event
1670 };
1671 
1672 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1673 {
1674 	raw_spin_lock_init(&data->lock);
1675 	skb_queue_head_init(&data->drop_queue);
1676 	u64_stats_init(&data->stats.syncp);
1677 }
1678 
1679 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1680 {
1681 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1682 }
1683 
1684 static void net_dm_cpu_data_init(int cpu)
1685 {
1686 	struct per_cpu_dm_data *data;
1687 
1688 	data = &per_cpu(dm_cpu_data, cpu);
1689 	__net_dm_cpu_data_init(data);
1690 }
1691 
1692 static void net_dm_cpu_data_fini(int cpu)
1693 {
1694 	struct per_cpu_dm_data *data;
1695 
1696 	data = &per_cpu(dm_cpu_data, cpu);
1697 	/* At this point, we should have exclusive access
1698 	 * to this struct and can free the skb inside it.
1699 	 */
1700 	consume_skb(data->skb);
1701 	__net_dm_cpu_data_fini(data);
1702 }
1703 
1704 static void net_dm_hw_cpu_data_init(int cpu)
1705 {
1706 	struct per_cpu_dm_data *hw_data;
1707 
1708 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1709 	__net_dm_cpu_data_init(hw_data);
1710 }
1711 
1712 static void net_dm_hw_cpu_data_fini(int cpu)
1713 {
1714 	struct per_cpu_dm_data *hw_data;
1715 
1716 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1717 	kfree(hw_data->hw_entries);
1718 	__net_dm_cpu_data_fini(hw_data);
1719 }
1720 
1721 static int __init init_net_drop_monitor(void)
1722 {
1723 	int cpu, rc;
1724 
1725 	pr_info("Initializing network drop monitor service\n");
1726 
1727 	if (sizeof(void *) > 8) {
1728 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1729 		return -ENOSPC;
1730 	}
1731 
1732 	for_each_possible_cpu(cpu) {
1733 		net_dm_cpu_data_init(cpu);
1734 		net_dm_hw_cpu_data_init(cpu);
1735 	}
1736 
1737 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1738 	if (rc < 0) {
1739 		pr_crit("Failed to register netdevice notifier\n");
1740 		return rc;
1741 	}
1742 
1743 	rc = genl_register_family(&net_drop_monitor_family);
1744 	if (rc) {
1745 		pr_err("Could not create drop monitor netlink family\n");
1746 		goto out_unreg;
1747 	}
1748 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1749 
1750 	rc = 0;
1751 
1752 	goto out;
1753 
1754 out_unreg:
1755 	WARN_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1756 out:
1757 	return rc;
1758 }
1759 
1760 static void exit_net_drop_monitor(void)
1761 {
1762 	int cpu;
1763 
1764 	/*
1765 	 * Because of the module_get/put we do in the trace state change path
1766 	 * we are guaranteed not to have any current users when we get here
1767 	 */
1768 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1769 
1770 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1771 
1772 	for_each_possible_cpu(cpu) {
1773 		net_dm_hw_cpu_data_fini(cpu);
1774 		net_dm_cpu_data_fini(cpu);
1775 	}
1776 }
1777 
1778 module_init(init_net_drop_monitor);
1779 module_exit(exit_net_drop_monitor);
1780 
1781 MODULE_LICENSE("GPL v2");
1782 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1783 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1784 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1785