xref: /openbmc/linux/net/core/drop_monitor.c (revision 15e3ae36)
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/percpu.h>
25 #include <linux/timer.h>
26 #include <linux/bitops.h>
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <net/drop_monitor.h>
30 #include <net/genetlink.h>
31 #include <net/netevent.h>
32 #include <net/flow_offload.h>
33 
34 #include <trace/events/skb.h>
35 #include <trace/events/napi.h>
36 
37 #include <asm/unaligned.h>
38 
39 #define TRACE_ON 1
40 #define TRACE_OFF 0
41 
42 /*
43  * Globals, our netlink socket pointer
44  * and the work handle that will send up
45  * netlink alerts
46  */
47 static int trace_state = TRACE_OFF;
48 static bool monitor_hw;
49 
50 /* net_dm_mutex
51  *
52  * An overall lock guarding every operation coming from userspace.
53  * It also guards the global 'hw_stats_list' list.
54  */
55 static DEFINE_MUTEX(net_dm_mutex);
56 
57 struct net_dm_stats {
58 	u64 dropped;
59 	struct u64_stats_sync syncp;
60 };
61 
62 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
63 
64 struct net_dm_hw_entry {
65 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
66 	u32 count;
67 };
68 
69 struct net_dm_hw_entries {
70 	u32 num_entries;
71 	struct net_dm_hw_entry entries[];
72 };
73 
74 struct per_cpu_dm_data {
75 	spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
76 					 * 'send_timer'
77 					 */
78 	union {
79 		struct sk_buff			*skb;
80 		struct net_dm_hw_entries	*hw_entries;
81 	};
82 	struct sk_buff_head	drop_queue;
83 	struct work_struct	dm_alert_work;
84 	struct timer_list	send_timer;
85 	struct net_dm_stats	stats;
86 };
87 
88 struct dm_hw_stat_delta {
89 	struct net_device *dev;
90 	unsigned long last_rx;
91 	struct list_head list;
92 	struct rcu_head rcu;
93 	unsigned long last_drop_val;
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 static LIST_HEAD(hw_stats_list);
105 
106 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
107 static u32 net_dm_trunc_len;
108 static u32 net_dm_queue_len = 1000;
109 
110 struct net_dm_alert_ops {
111 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
112 				void *location);
113 	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
114 				int work, int budget);
115 	void (*work_item_func)(struct work_struct *work);
116 	void (*hw_work_item_func)(struct work_struct *work);
117 	void (*hw_probe)(struct sk_buff *skb,
118 			 const struct net_dm_hw_metadata *hw_metadata);
119 };
120 
121 struct net_dm_skb_cb {
122 	union {
123 		struct net_dm_hw_metadata *hw_metadata;
124 		void *pc;
125 	};
126 };
127 
128 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
129 
130 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
131 {
132 	size_t al;
133 	struct net_dm_alert_msg *msg;
134 	struct nlattr *nla;
135 	struct sk_buff *skb;
136 	unsigned long flags;
137 	void *msg_header;
138 
139 	al = sizeof(struct net_dm_alert_msg);
140 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
141 	al += sizeof(struct nlattr);
142 
143 	skb = genlmsg_new(al, GFP_KERNEL);
144 
145 	if (!skb)
146 		goto err;
147 
148 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
149 				 0, NET_DM_CMD_ALERT);
150 	if (!msg_header) {
151 		nlmsg_free(skb);
152 		skb = NULL;
153 		goto err;
154 	}
155 	nla = nla_reserve(skb, NLA_UNSPEC,
156 			  sizeof(struct net_dm_alert_msg));
157 	if (!nla) {
158 		nlmsg_free(skb);
159 		skb = NULL;
160 		goto err;
161 	}
162 	msg = nla_data(nla);
163 	memset(msg, 0, al);
164 	goto out;
165 
166 err:
167 	mod_timer(&data->send_timer, jiffies + HZ / 10);
168 out:
169 	spin_lock_irqsave(&data->lock, flags);
170 	swap(data->skb, skb);
171 	spin_unlock_irqrestore(&data->lock, flags);
172 
173 	if (skb) {
174 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
175 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
176 
177 		genlmsg_end(skb, genlmsg_data(gnlh));
178 	}
179 
180 	return skb;
181 }
182 
183 static const struct genl_multicast_group dropmon_mcgrps[] = {
184 	{ .name = "events", },
185 };
186 
187 static void send_dm_alert(struct work_struct *work)
188 {
189 	struct sk_buff *skb;
190 	struct per_cpu_dm_data *data;
191 
192 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
193 
194 	skb = reset_per_cpu_data(data);
195 
196 	if (skb)
197 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
198 				  0, GFP_KERNEL);
199 }
200 
201 /*
202  * This is the timer function to delay the sending of an alert
203  * in the event that more drops will arrive during the
204  * hysteresis period.
205  */
206 static void sched_send_work(struct timer_list *t)
207 {
208 	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
209 
210 	schedule_work(&data->dm_alert_work);
211 }
212 
213 static void trace_drop_common(struct sk_buff *skb, void *location)
214 {
215 	struct net_dm_alert_msg *msg;
216 	struct nlmsghdr *nlh;
217 	struct nlattr *nla;
218 	int i;
219 	struct sk_buff *dskb;
220 	struct per_cpu_dm_data *data;
221 	unsigned long flags;
222 
223 	local_irq_save(flags);
224 	data = this_cpu_ptr(&dm_cpu_data);
225 	spin_lock(&data->lock);
226 	dskb = data->skb;
227 
228 	if (!dskb)
229 		goto out;
230 
231 	nlh = (struct nlmsghdr *)dskb->data;
232 	nla = genlmsg_data(nlmsg_data(nlh));
233 	msg = nla_data(nla);
234 	for (i = 0; i < msg->entries; i++) {
235 		if (!memcmp(&location, msg->points[i].pc, sizeof(void *))) {
236 			msg->points[i].count++;
237 			goto out;
238 		}
239 	}
240 	if (msg->entries == dm_hit_limit)
241 		goto out;
242 	/*
243 	 * We need to create a new entry
244 	 */
245 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
246 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
247 	memcpy(msg->points[msg->entries].pc, &location, sizeof(void *));
248 	msg->points[msg->entries].count = 1;
249 	msg->entries++;
250 
251 	if (!timer_pending(&data->send_timer)) {
252 		data->send_timer.expires = jiffies + dm_delay * HZ;
253 		add_timer(&data->send_timer);
254 	}
255 
256 out:
257 	spin_unlock_irqrestore(&data->lock, flags);
258 }
259 
260 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
261 {
262 	trace_drop_common(skb, location);
263 }
264 
265 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
266 				int work, int budget)
267 {
268 	struct dm_hw_stat_delta *new_stat;
269 
270 	/*
271 	 * Don't check napi structures with no associated device
272 	 */
273 	if (!napi->dev)
274 		return;
275 
276 	rcu_read_lock();
277 	list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
278 		/*
279 		 * only add a note to our monitor buffer if:
280 		 * 1) this is the dev we received on
281 		 * 2) its after the last_rx delta
282 		 * 3) our rx_dropped count has gone up
283 		 */
284 		if ((new_stat->dev == napi->dev)  &&
285 		    (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
286 		    (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
287 			trace_drop_common(NULL, NULL);
288 			new_stat->last_drop_val = napi->dev->stats.rx_dropped;
289 			new_stat->last_rx = jiffies;
290 			break;
291 		}
292 	}
293 	rcu_read_unlock();
294 }
295 
296 static struct net_dm_hw_entries *
297 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
298 {
299 	struct net_dm_hw_entries *hw_entries;
300 	unsigned long flags;
301 
302 	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
303 			     GFP_KERNEL);
304 	if (!hw_entries) {
305 		/* If the memory allocation failed, we try to perform another
306 		 * allocation in 1/10 second. Otherwise, the probe function
307 		 * will constantly bail out.
308 		 */
309 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
310 	}
311 
312 	spin_lock_irqsave(&hw_data->lock, flags);
313 	swap(hw_data->hw_entries, hw_entries);
314 	spin_unlock_irqrestore(&hw_data->lock, flags);
315 
316 	return hw_entries;
317 }
318 
319 static int net_dm_hw_entry_put(struct sk_buff *msg,
320 			       const struct net_dm_hw_entry *hw_entry)
321 {
322 	struct nlattr *attr;
323 
324 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
325 	if (!attr)
326 		return -EMSGSIZE;
327 
328 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
329 		goto nla_put_failure;
330 
331 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
332 		goto nla_put_failure;
333 
334 	nla_nest_end(msg, attr);
335 
336 	return 0;
337 
338 nla_put_failure:
339 	nla_nest_cancel(msg, attr);
340 	return -EMSGSIZE;
341 }
342 
343 static int net_dm_hw_entries_put(struct sk_buff *msg,
344 				 const struct net_dm_hw_entries *hw_entries)
345 {
346 	struct nlattr *attr;
347 	int i;
348 
349 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
350 	if (!attr)
351 		return -EMSGSIZE;
352 
353 	for (i = 0; i < hw_entries->num_entries; i++) {
354 		int rc;
355 
356 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
357 		if (rc)
358 			goto nla_put_failure;
359 	}
360 
361 	nla_nest_end(msg, attr);
362 
363 	return 0;
364 
365 nla_put_failure:
366 	nla_nest_cancel(msg, attr);
367 	return -EMSGSIZE;
368 }
369 
370 static int
371 net_dm_hw_summary_report_fill(struct sk_buff *msg,
372 			      const struct net_dm_hw_entries *hw_entries)
373 {
374 	struct net_dm_alert_msg anc_hdr = { 0 };
375 	void *hdr;
376 	int rc;
377 
378 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
379 			  NET_DM_CMD_ALERT);
380 	if (!hdr)
381 		return -EMSGSIZE;
382 
383 	/* We need to put the ancillary header in order not to break user
384 	 * space.
385 	 */
386 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
387 		goto nla_put_failure;
388 
389 	rc = net_dm_hw_entries_put(msg, hw_entries);
390 	if (rc)
391 		goto nla_put_failure;
392 
393 	genlmsg_end(msg, hdr);
394 
395 	return 0;
396 
397 nla_put_failure:
398 	genlmsg_cancel(msg, hdr);
399 	return -EMSGSIZE;
400 }
401 
402 static void net_dm_hw_summary_work(struct work_struct *work)
403 {
404 	struct net_dm_hw_entries *hw_entries;
405 	struct per_cpu_dm_data *hw_data;
406 	struct sk_buff *msg;
407 	int rc;
408 
409 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
410 
411 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
412 	if (!hw_entries)
413 		return;
414 
415 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
416 	if (!msg)
417 		goto out;
418 
419 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
420 	if (rc) {
421 		nlmsg_free(msg);
422 		goto out;
423 	}
424 
425 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
426 
427 out:
428 	kfree(hw_entries);
429 }
430 
431 static void
432 net_dm_hw_summary_probe(struct sk_buff *skb,
433 			const struct net_dm_hw_metadata *hw_metadata)
434 {
435 	struct net_dm_hw_entries *hw_entries;
436 	struct net_dm_hw_entry *hw_entry;
437 	struct per_cpu_dm_data *hw_data;
438 	unsigned long flags;
439 	int i;
440 
441 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
442 	spin_lock_irqsave(&hw_data->lock, flags);
443 	hw_entries = hw_data->hw_entries;
444 
445 	if (!hw_entries)
446 		goto out;
447 
448 	for (i = 0; i < hw_entries->num_entries; i++) {
449 		hw_entry = &hw_entries->entries[i];
450 		if (!strncmp(hw_entry->trap_name, hw_metadata->trap_name,
451 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
452 			hw_entry->count++;
453 			goto out;
454 		}
455 	}
456 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
457 		goto out;
458 
459 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
460 	strlcpy(hw_entry->trap_name, hw_metadata->trap_name,
461 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
462 	hw_entry->count = 1;
463 	hw_entries->num_entries++;
464 
465 	if (!timer_pending(&hw_data->send_timer)) {
466 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
467 		add_timer(&hw_data->send_timer);
468 	}
469 
470 out:
471 	spin_unlock_irqrestore(&hw_data->lock, flags);
472 }
473 
474 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
475 	.kfree_skb_probe	= trace_kfree_skb_hit,
476 	.napi_poll_probe	= trace_napi_poll_hit,
477 	.work_item_func		= send_dm_alert,
478 	.hw_work_item_func	= net_dm_hw_summary_work,
479 	.hw_probe		= net_dm_hw_summary_probe,
480 };
481 
482 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
483 					      struct sk_buff *skb,
484 					      void *location)
485 {
486 	ktime_t tstamp = ktime_get_real();
487 	struct per_cpu_dm_data *data;
488 	struct sk_buff *nskb;
489 	unsigned long flags;
490 
491 	if (!skb_mac_header_was_set(skb))
492 		return;
493 
494 	nskb = skb_clone(skb, GFP_ATOMIC);
495 	if (!nskb)
496 		return;
497 
498 	NET_DM_SKB_CB(nskb)->pc = location;
499 	/* Override the timestamp because we care about the time when the
500 	 * packet was dropped.
501 	 */
502 	nskb->tstamp = tstamp;
503 
504 	data = this_cpu_ptr(&dm_cpu_data);
505 
506 	spin_lock_irqsave(&data->drop_queue.lock, flags);
507 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
508 		__skb_queue_tail(&data->drop_queue, nskb);
509 	else
510 		goto unlock_free;
511 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
512 
513 	schedule_work(&data->dm_alert_work);
514 
515 	return;
516 
517 unlock_free:
518 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
519 	u64_stats_update_begin(&data->stats.syncp);
520 	data->stats.dropped++;
521 	u64_stats_update_end(&data->stats.syncp);
522 	consume_skb(nskb);
523 }
524 
525 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
526 					      struct napi_struct *napi,
527 					      int work, int budget)
528 {
529 }
530 
531 static size_t net_dm_in_port_size(void)
532 {
533 	       /* NET_DM_ATTR_IN_PORT nest */
534 	return nla_total_size(0) +
535 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
536 	       nla_total_size(sizeof(u32)) +
537 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
538 	       nla_total_size(IFNAMSIZ + 1);
539 }
540 
541 #define NET_DM_MAX_SYMBOL_LEN 40
542 
543 static size_t net_dm_packet_report_size(size_t payload_len)
544 {
545 	size_t size;
546 
547 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
548 
549 	return NLMSG_ALIGN(size) +
550 	       /* NET_DM_ATTR_ORIGIN */
551 	       nla_total_size(sizeof(u16)) +
552 	       /* NET_DM_ATTR_PC */
553 	       nla_total_size(sizeof(u64)) +
554 	       /* NET_DM_ATTR_SYMBOL */
555 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
556 	       /* NET_DM_ATTR_IN_PORT */
557 	       net_dm_in_port_size() +
558 	       /* NET_DM_ATTR_TIMESTAMP */
559 	       nla_total_size(sizeof(u64)) +
560 	       /* NET_DM_ATTR_ORIG_LEN */
561 	       nla_total_size(sizeof(u32)) +
562 	       /* NET_DM_ATTR_PROTO */
563 	       nla_total_size(sizeof(u16)) +
564 	       /* NET_DM_ATTR_PAYLOAD */
565 	       nla_total_size(payload_len);
566 }
567 
568 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
569 					    const char *name)
570 {
571 	struct nlattr *attr;
572 
573 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
574 	if (!attr)
575 		return -EMSGSIZE;
576 
577 	if (ifindex &&
578 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
579 		goto nla_put_failure;
580 
581 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
582 		goto nla_put_failure;
583 
584 	nla_nest_end(msg, attr);
585 
586 	return 0;
587 
588 nla_put_failure:
589 	nla_nest_cancel(msg, attr);
590 	return -EMSGSIZE;
591 }
592 
593 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
594 				     size_t payload_len)
595 {
596 	u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
597 	char buf[NET_DM_MAX_SYMBOL_LEN];
598 	struct nlattr *attr;
599 	void *hdr;
600 	int rc;
601 
602 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
603 			  NET_DM_CMD_PACKET_ALERT);
604 	if (!hdr)
605 		return -EMSGSIZE;
606 
607 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
608 		goto nla_put_failure;
609 
610 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
611 		goto nla_put_failure;
612 
613 	snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
614 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
615 		goto nla_put_failure;
616 
617 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
618 	if (rc)
619 		goto nla_put_failure;
620 
621 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
622 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
623 		goto nla_put_failure;
624 
625 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
626 		goto nla_put_failure;
627 
628 	if (!payload_len)
629 		goto out;
630 
631 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
632 		goto nla_put_failure;
633 
634 	attr = skb_put(msg, nla_total_size(payload_len));
635 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
636 	attr->nla_len = nla_attr_size(payload_len);
637 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
638 		goto nla_put_failure;
639 
640 out:
641 	genlmsg_end(msg, hdr);
642 
643 	return 0;
644 
645 nla_put_failure:
646 	genlmsg_cancel(msg, hdr);
647 	return -EMSGSIZE;
648 }
649 
650 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
651 
652 static void net_dm_packet_report(struct sk_buff *skb)
653 {
654 	struct sk_buff *msg;
655 	size_t payload_len;
656 	int rc;
657 
658 	/* Make sure we start copying the packet from the MAC header */
659 	if (skb->data > skb_mac_header(skb))
660 		skb_push(skb, skb->data - skb_mac_header(skb));
661 	else
662 		skb_pull(skb, skb_mac_header(skb) - skb->data);
663 
664 	/* Ensure packet fits inside a single netlink attribute */
665 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
666 	if (net_dm_trunc_len)
667 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
668 
669 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
670 	if (!msg)
671 		goto out;
672 
673 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
674 	if (rc) {
675 		nlmsg_free(msg);
676 		goto out;
677 	}
678 
679 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
680 
681 out:
682 	consume_skb(skb);
683 }
684 
685 static void net_dm_packet_work(struct work_struct *work)
686 {
687 	struct per_cpu_dm_data *data;
688 	struct sk_buff_head list;
689 	struct sk_buff *skb;
690 	unsigned long flags;
691 
692 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
693 
694 	__skb_queue_head_init(&list);
695 
696 	spin_lock_irqsave(&data->drop_queue.lock, flags);
697 	skb_queue_splice_tail_init(&data->drop_queue, &list);
698 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
699 
700 	while ((skb = __skb_dequeue(&list)))
701 		net_dm_packet_report(skb);
702 }
703 
704 static size_t
705 net_dm_flow_action_cookie_size(const struct net_dm_hw_metadata *hw_metadata)
706 {
707 	return hw_metadata->fa_cookie ?
708 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
709 }
710 
711 static size_t
712 net_dm_hw_packet_report_size(size_t payload_len,
713 			     const struct net_dm_hw_metadata *hw_metadata)
714 {
715 	size_t size;
716 
717 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
718 
719 	return NLMSG_ALIGN(size) +
720 	       /* NET_DM_ATTR_ORIGIN */
721 	       nla_total_size(sizeof(u16)) +
722 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
723 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
724 	       /* NET_DM_ATTR_HW_TRAP_NAME */
725 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
726 	       /* NET_DM_ATTR_IN_PORT */
727 	       net_dm_in_port_size() +
728 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
729 	       net_dm_flow_action_cookie_size(hw_metadata) +
730 	       /* NET_DM_ATTR_TIMESTAMP */
731 	       nla_total_size(sizeof(u64)) +
732 	       /* NET_DM_ATTR_ORIG_LEN */
733 	       nla_total_size(sizeof(u32)) +
734 	       /* NET_DM_ATTR_PROTO */
735 	       nla_total_size(sizeof(u16)) +
736 	       /* NET_DM_ATTR_PAYLOAD */
737 	       nla_total_size(payload_len);
738 }
739 
740 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
741 					struct sk_buff *skb, size_t payload_len)
742 {
743 	struct net_dm_hw_metadata *hw_metadata;
744 	struct nlattr *attr;
745 	void *hdr;
746 
747 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
748 
749 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
750 			  NET_DM_CMD_PACKET_ALERT);
751 	if (!hdr)
752 		return -EMSGSIZE;
753 
754 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
755 		goto nla_put_failure;
756 
757 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
758 			   hw_metadata->trap_group_name))
759 		goto nla_put_failure;
760 
761 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
762 			   hw_metadata->trap_name))
763 		goto nla_put_failure;
764 
765 	if (hw_metadata->input_dev) {
766 		struct net_device *dev = hw_metadata->input_dev;
767 		int rc;
768 
769 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
770 						      dev->name);
771 		if (rc)
772 			goto nla_put_failure;
773 	}
774 
775 	if (hw_metadata->fa_cookie &&
776 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
777 		    hw_metadata->fa_cookie->cookie_len,
778 		    hw_metadata->fa_cookie->cookie))
779 		goto nla_put_failure;
780 
781 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
782 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
783 		goto nla_put_failure;
784 
785 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
786 		goto nla_put_failure;
787 
788 	if (!payload_len)
789 		goto out;
790 
791 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
792 		goto nla_put_failure;
793 
794 	attr = skb_put(msg, nla_total_size(payload_len));
795 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
796 	attr->nla_len = nla_attr_size(payload_len);
797 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
798 		goto nla_put_failure;
799 
800 out:
801 	genlmsg_end(msg, hdr);
802 
803 	return 0;
804 
805 nla_put_failure:
806 	genlmsg_cancel(msg, hdr);
807 	return -EMSGSIZE;
808 }
809 
810 static struct net_dm_hw_metadata *
811 net_dm_hw_metadata_clone(const struct net_dm_hw_metadata *hw_metadata)
812 {
813 	const struct flow_action_cookie *fa_cookie;
814 	struct net_dm_hw_metadata *n_hw_metadata;
815 	const char *trap_group_name;
816 	const char *trap_name;
817 
818 	n_hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
819 	if (!n_hw_metadata)
820 		return NULL;
821 
822 	trap_group_name = kstrdup(hw_metadata->trap_group_name, GFP_ATOMIC);
823 	if (!trap_group_name)
824 		goto free_hw_metadata;
825 	n_hw_metadata->trap_group_name = trap_group_name;
826 
827 	trap_name = kstrdup(hw_metadata->trap_name, GFP_ATOMIC);
828 	if (!trap_name)
829 		goto free_trap_group;
830 	n_hw_metadata->trap_name = trap_name;
831 
832 	if (hw_metadata->fa_cookie) {
833 		size_t cookie_size = sizeof(*fa_cookie) +
834 				     hw_metadata->fa_cookie->cookie_len;
835 
836 		fa_cookie = kmemdup(hw_metadata->fa_cookie, cookie_size,
837 				    GFP_ATOMIC);
838 		if (!fa_cookie)
839 			goto free_trap_name;
840 		n_hw_metadata->fa_cookie = fa_cookie;
841 	}
842 
843 	n_hw_metadata->input_dev = hw_metadata->input_dev;
844 	if (n_hw_metadata->input_dev)
845 		dev_hold(n_hw_metadata->input_dev);
846 
847 	return n_hw_metadata;
848 
849 free_trap_name:
850 	kfree(trap_name);
851 free_trap_group:
852 	kfree(trap_group_name);
853 free_hw_metadata:
854 	kfree(n_hw_metadata);
855 	return NULL;
856 }
857 
858 static void
859 net_dm_hw_metadata_free(const struct net_dm_hw_metadata *hw_metadata)
860 {
861 	if (hw_metadata->input_dev)
862 		dev_put(hw_metadata->input_dev);
863 	kfree(hw_metadata->fa_cookie);
864 	kfree(hw_metadata->trap_name);
865 	kfree(hw_metadata->trap_group_name);
866 	kfree(hw_metadata);
867 }
868 
869 static void net_dm_hw_packet_report(struct sk_buff *skb)
870 {
871 	struct net_dm_hw_metadata *hw_metadata;
872 	struct sk_buff *msg;
873 	size_t payload_len;
874 	int rc;
875 
876 	if (skb->data > skb_mac_header(skb))
877 		skb_push(skb, skb->data - skb_mac_header(skb));
878 	else
879 		skb_pull(skb, skb_mac_header(skb) - skb->data);
880 
881 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
882 	if (net_dm_trunc_len)
883 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
884 
885 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
886 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
887 			GFP_KERNEL);
888 	if (!msg)
889 		goto out;
890 
891 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
892 	if (rc) {
893 		nlmsg_free(msg);
894 		goto out;
895 	}
896 
897 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
898 
899 out:
900 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
901 	consume_skb(skb);
902 }
903 
904 static void net_dm_hw_packet_work(struct work_struct *work)
905 {
906 	struct per_cpu_dm_data *hw_data;
907 	struct sk_buff_head list;
908 	struct sk_buff *skb;
909 	unsigned long flags;
910 
911 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
912 
913 	__skb_queue_head_init(&list);
914 
915 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
916 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
917 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
918 
919 	while ((skb = __skb_dequeue(&list)))
920 		net_dm_hw_packet_report(skb);
921 }
922 
923 static void
924 net_dm_hw_packet_probe(struct sk_buff *skb,
925 		       const struct net_dm_hw_metadata *hw_metadata)
926 {
927 	struct net_dm_hw_metadata *n_hw_metadata;
928 	ktime_t tstamp = ktime_get_real();
929 	struct per_cpu_dm_data *hw_data;
930 	struct sk_buff *nskb;
931 	unsigned long flags;
932 
933 	if (!skb_mac_header_was_set(skb))
934 		return;
935 
936 	nskb = skb_clone(skb, GFP_ATOMIC);
937 	if (!nskb)
938 		return;
939 
940 	n_hw_metadata = net_dm_hw_metadata_clone(hw_metadata);
941 	if (!n_hw_metadata)
942 		goto free;
943 
944 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
945 	nskb->tstamp = tstamp;
946 
947 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
948 
949 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
950 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
951 		__skb_queue_tail(&hw_data->drop_queue, nskb);
952 	else
953 		goto unlock_free;
954 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
955 
956 	schedule_work(&hw_data->dm_alert_work);
957 
958 	return;
959 
960 unlock_free:
961 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
962 	u64_stats_update_begin(&hw_data->stats.syncp);
963 	hw_data->stats.dropped++;
964 	u64_stats_update_end(&hw_data->stats.syncp);
965 	net_dm_hw_metadata_free(n_hw_metadata);
966 free:
967 	consume_skb(nskb);
968 }
969 
970 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
971 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
972 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
973 	.work_item_func		= net_dm_packet_work,
974 	.hw_work_item_func	= net_dm_hw_packet_work,
975 	.hw_probe		= net_dm_hw_packet_probe,
976 };
977 
978 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
979 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
980 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
981 };
982 
983 void net_dm_hw_report(struct sk_buff *skb,
984 		      const struct net_dm_hw_metadata *hw_metadata)
985 {
986 	rcu_read_lock();
987 
988 	if (!monitor_hw)
989 		goto out;
990 
991 	net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
992 
993 out:
994 	rcu_read_unlock();
995 }
996 EXPORT_SYMBOL_GPL(net_dm_hw_report);
997 
998 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
999 {
1000 	const struct net_dm_alert_ops *ops;
1001 	int cpu;
1002 
1003 	if (monitor_hw) {
1004 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1005 		return -EAGAIN;
1006 	}
1007 
1008 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1009 
1010 	if (!try_module_get(THIS_MODULE)) {
1011 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1012 		return -ENODEV;
1013 	}
1014 
1015 	for_each_possible_cpu(cpu) {
1016 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1017 		struct net_dm_hw_entries *hw_entries;
1018 
1019 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1020 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1021 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1022 		kfree(hw_entries);
1023 	}
1024 
1025 	monitor_hw = true;
1026 
1027 	return 0;
1028 }
1029 
1030 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1031 {
1032 	int cpu;
1033 
1034 	if (!monitor_hw) {
1035 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1036 		return;
1037 	}
1038 
1039 	monitor_hw = false;
1040 
1041 	/* After this call returns we are guaranteed that no CPU is processing
1042 	 * any hardware drops.
1043 	 */
1044 	synchronize_rcu();
1045 
1046 	for_each_possible_cpu(cpu) {
1047 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1048 		struct sk_buff *skb;
1049 
1050 		del_timer_sync(&hw_data->send_timer);
1051 		cancel_work_sync(&hw_data->dm_alert_work);
1052 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1053 			struct net_dm_hw_metadata *hw_metadata;
1054 
1055 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1056 			net_dm_hw_metadata_free(hw_metadata);
1057 			consume_skb(skb);
1058 		}
1059 	}
1060 
1061 	module_put(THIS_MODULE);
1062 }
1063 
1064 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1065 {
1066 	const struct net_dm_alert_ops *ops;
1067 	int cpu, rc;
1068 
1069 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1070 
1071 	if (!try_module_get(THIS_MODULE)) {
1072 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1073 		return -ENODEV;
1074 	}
1075 
1076 	for_each_possible_cpu(cpu) {
1077 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1078 		struct sk_buff *skb;
1079 
1080 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1081 		timer_setup(&data->send_timer, sched_send_work, 0);
1082 		/* Allocate a new per-CPU skb for the summary alert message and
1083 		 * free the old one which might contain stale data from
1084 		 * previous tracing.
1085 		 */
1086 		skb = reset_per_cpu_data(data);
1087 		consume_skb(skb);
1088 	}
1089 
1090 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1091 	if (rc) {
1092 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1093 		goto err_module_put;
1094 	}
1095 
1096 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1097 	if (rc) {
1098 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1099 		goto err_unregister_trace;
1100 	}
1101 
1102 	return 0;
1103 
1104 err_unregister_trace:
1105 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1106 err_module_put:
1107 	module_put(THIS_MODULE);
1108 	return rc;
1109 }
1110 
1111 static void net_dm_trace_off_set(void)
1112 {
1113 	struct dm_hw_stat_delta *new_stat, *temp;
1114 	const struct net_dm_alert_ops *ops;
1115 	int cpu;
1116 
1117 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1118 
1119 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1120 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1121 
1122 	tracepoint_synchronize_unregister();
1123 
1124 	/* Make sure we do not send notifications to user space after request
1125 	 * to stop tracing returns.
1126 	 */
1127 	for_each_possible_cpu(cpu) {
1128 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1129 		struct sk_buff *skb;
1130 
1131 		del_timer_sync(&data->send_timer);
1132 		cancel_work_sync(&data->dm_alert_work);
1133 		while ((skb = __skb_dequeue(&data->drop_queue)))
1134 			consume_skb(skb);
1135 	}
1136 
1137 	list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1138 		if (new_stat->dev == NULL) {
1139 			list_del_rcu(&new_stat->list);
1140 			kfree_rcu(new_stat, rcu);
1141 		}
1142 	}
1143 
1144 	module_put(THIS_MODULE);
1145 }
1146 
1147 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1148 {
1149 	int rc = 0;
1150 
1151 	if (state == trace_state) {
1152 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1153 		return -EAGAIN;
1154 	}
1155 
1156 	switch (state) {
1157 	case TRACE_ON:
1158 		rc = net_dm_trace_on_set(extack);
1159 		break;
1160 	case TRACE_OFF:
1161 		net_dm_trace_off_set();
1162 		break;
1163 	default:
1164 		rc = 1;
1165 		break;
1166 	}
1167 
1168 	if (!rc)
1169 		trace_state = state;
1170 	else
1171 		rc = -EINPROGRESS;
1172 
1173 	return rc;
1174 }
1175 
1176 static bool net_dm_is_monitoring(void)
1177 {
1178 	return trace_state == TRACE_ON || monitor_hw;
1179 }
1180 
1181 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1182 					   enum net_dm_alert_mode *p_alert_mode)
1183 {
1184 	u8 val;
1185 
1186 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1187 
1188 	switch (val) {
1189 	case NET_DM_ALERT_MODE_SUMMARY: /* fall-through */
1190 	case NET_DM_ALERT_MODE_PACKET:
1191 		*p_alert_mode = val;
1192 		break;
1193 	default:
1194 		return -EINVAL;
1195 	}
1196 
1197 	return 0;
1198 }
1199 
1200 static int net_dm_alert_mode_set(struct genl_info *info)
1201 {
1202 	struct netlink_ext_ack *extack = info->extack;
1203 	enum net_dm_alert_mode alert_mode;
1204 	int rc;
1205 
1206 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1207 		return 0;
1208 
1209 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1210 	if (rc) {
1211 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1212 		return -EINVAL;
1213 	}
1214 
1215 	net_dm_alert_mode = alert_mode;
1216 
1217 	return 0;
1218 }
1219 
1220 static void net_dm_trunc_len_set(struct genl_info *info)
1221 {
1222 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1223 		return;
1224 
1225 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1226 }
1227 
1228 static void net_dm_queue_len_set(struct genl_info *info)
1229 {
1230 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1231 		return;
1232 
1233 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1234 }
1235 
1236 static int net_dm_cmd_config(struct sk_buff *skb,
1237 			struct genl_info *info)
1238 {
1239 	struct netlink_ext_ack *extack = info->extack;
1240 	int rc;
1241 
1242 	if (net_dm_is_monitoring()) {
1243 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1244 		return -EBUSY;
1245 	}
1246 
1247 	rc = net_dm_alert_mode_set(info);
1248 	if (rc)
1249 		return rc;
1250 
1251 	net_dm_trunc_len_set(info);
1252 
1253 	net_dm_queue_len_set(info);
1254 
1255 	return 0;
1256 }
1257 
1258 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1259 				struct netlink_ext_ack *extack)
1260 {
1261 	bool sw_set = false;
1262 	int rc;
1263 
1264 	if (set_sw) {
1265 		rc = set_all_monitor_traces(TRACE_ON, extack);
1266 		if (rc)
1267 			return rc;
1268 		sw_set = true;
1269 	}
1270 
1271 	if (set_hw) {
1272 		rc = net_dm_hw_monitor_start(extack);
1273 		if (rc)
1274 			goto err_monitor_hw;
1275 	}
1276 
1277 	return 0;
1278 
1279 err_monitor_hw:
1280 	if (sw_set)
1281 		set_all_monitor_traces(TRACE_OFF, extack);
1282 	return rc;
1283 }
1284 
1285 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1286 				struct netlink_ext_ack *extack)
1287 {
1288 	if (set_hw)
1289 		net_dm_hw_monitor_stop(extack);
1290 	if (set_sw)
1291 		set_all_monitor_traces(TRACE_OFF, extack);
1292 }
1293 
1294 static int net_dm_cmd_trace(struct sk_buff *skb,
1295 			struct genl_info *info)
1296 {
1297 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1298 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1299 	struct netlink_ext_ack *extack = info->extack;
1300 
1301 	/* To maintain backward compatibility, we start / stop monitoring of
1302 	 * software drops if no flag is specified.
1303 	 */
1304 	if (!set_sw && !set_hw)
1305 		set_sw = true;
1306 
1307 	switch (info->genlhdr->cmd) {
1308 	case NET_DM_CMD_START:
1309 		return net_dm_monitor_start(set_sw, set_hw, extack);
1310 	case NET_DM_CMD_STOP:
1311 		net_dm_monitor_stop(set_sw, set_hw, extack);
1312 		return 0;
1313 	}
1314 
1315 	return -EOPNOTSUPP;
1316 }
1317 
1318 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1319 {
1320 	void *hdr;
1321 
1322 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1323 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1324 	if (!hdr)
1325 		return -EMSGSIZE;
1326 
1327 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1328 		goto nla_put_failure;
1329 
1330 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1331 		goto nla_put_failure;
1332 
1333 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1334 		goto nla_put_failure;
1335 
1336 	genlmsg_end(msg, hdr);
1337 
1338 	return 0;
1339 
1340 nla_put_failure:
1341 	genlmsg_cancel(msg, hdr);
1342 	return -EMSGSIZE;
1343 }
1344 
1345 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1346 {
1347 	struct sk_buff *msg;
1348 	int rc;
1349 
1350 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1351 	if (!msg)
1352 		return -ENOMEM;
1353 
1354 	rc = net_dm_config_fill(msg, info);
1355 	if (rc)
1356 		goto free_msg;
1357 
1358 	return genlmsg_reply(msg, info);
1359 
1360 free_msg:
1361 	nlmsg_free(msg);
1362 	return rc;
1363 }
1364 
1365 static void net_dm_stats_read(struct net_dm_stats *stats)
1366 {
1367 	int cpu;
1368 
1369 	memset(stats, 0, sizeof(*stats));
1370 	for_each_possible_cpu(cpu) {
1371 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1372 		struct net_dm_stats *cpu_stats = &data->stats;
1373 		unsigned int start;
1374 		u64 dropped;
1375 
1376 		do {
1377 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1378 			dropped = cpu_stats->dropped;
1379 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1380 
1381 		stats->dropped += dropped;
1382 	}
1383 }
1384 
1385 static int net_dm_stats_put(struct sk_buff *msg)
1386 {
1387 	struct net_dm_stats stats;
1388 	struct nlattr *attr;
1389 
1390 	net_dm_stats_read(&stats);
1391 
1392 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1393 	if (!attr)
1394 		return -EMSGSIZE;
1395 
1396 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1397 			      stats.dropped, NET_DM_ATTR_PAD))
1398 		goto nla_put_failure;
1399 
1400 	nla_nest_end(msg, attr);
1401 
1402 	return 0;
1403 
1404 nla_put_failure:
1405 	nla_nest_cancel(msg, attr);
1406 	return -EMSGSIZE;
1407 }
1408 
1409 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1410 {
1411 	int cpu;
1412 
1413 	memset(stats, 0, sizeof(*stats));
1414 	for_each_possible_cpu(cpu) {
1415 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1416 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1417 		unsigned int start;
1418 		u64 dropped;
1419 
1420 		do {
1421 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1422 			dropped = cpu_stats->dropped;
1423 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1424 
1425 		stats->dropped += dropped;
1426 	}
1427 }
1428 
1429 static int net_dm_hw_stats_put(struct sk_buff *msg)
1430 {
1431 	struct net_dm_stats stats;
1432 	struct nlattr *attr;
1433 
1434 	net_dm_hw_stats_read(&stats);
1435 
1436 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1437 	if (!attr)
1438 		return -EMSGSIZE;
1439 
1440 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1441 			      stats.dropped, NET_DM_ATTR_PAD))
1442 		goto nla_put_failure;
1443 
1444 	nla_nest_end(msg, attr);
1445 
1446 	return 0;
1447 
1448 nla_put_failure:
1449 	nla_nest_cancel(msg, attr);
1450 	return -EMSGSIZE;
1451 }
1452 
1453 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1454 {
1455 	void *hdr;
1456 	int rc;
1457 
1458 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1459 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1460 	if (!hdr)
1461 		return -EMSGSIZE;
1462 
1463 	rc = net_dm_stats_put(msg);
1464 	if (rc)
1465 		goto nla_put_failure;
1466 
1467 	rc = net_dm_hw_stats_put(msg);
1468 	if (rc)
1469 		goto nla_put_failure;
1470 
1471 	genlmsg_end(msg, hdr);
1472 
1473 	return 0;
1474 
1475 nla_put_failure:
1476 	genlmsg_cancel(msg, hdr);
1477 	return -EMSGSIZE;
1478 }
1479 
1480 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1481 {
1482 	struct sk_buff *msg;
1483 	int rc;
1484 
1485 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1486 	if (!msg)
1487 		return -ENOMEM;
1488 
1489 	rc = net_dm_stats_fill(msg, info);
1490 	if (rc)
1491 		goto free_msg;
1492 
1493 	return genlmsg_reply(msg, info);
1494 
1495 free_msg:
1496 	nlmsg_free(msg);
1497 	return rc;
1498 }
1499 
1500 static int dropmon_net_event(struct notifier_block *ev_block,
1501 			     unsigned long event, void *ptr)
1502 {
1503 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1504 	struct dm_hw_stat_delta *new_stat = NULL;
1505 	struct dm_hw_stat_delta *tmp;
1506 
1507 	switch (event) {
1508 	case NETDEV_REGISTER:
1509 		new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
1510 
1511 		if (!new_stat)
1512 			goto out;
1513 
1514 		new_stat->dev = dev;
1515 		new_stat->last_rx = jiffies;
1516 		mutex_lock(&net_dm_mutex);
1517 		list_add_rcu(&new_stat->list, &hw_stats_list);
1518 		mutex_unlock(&net_dm_mutex);
1519 		break;
1520 	case NETDEV_UNREGISTER:
1521 		mutex_lock(&net_dm_mutex);
1522 		list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1523 			if (new_stat->dev == dev) {
1524 				new_stat->dev = NULL;
1525 				if (trace_state == TRACE_OFF) {
1526 					list_del_rcu(&new_stat->list);
1527 					kfree_rcu(new_stat, rcu);
1528 					break;
1529 				}
1530 			}
1531 		}
1532 		mutex_unlock(&net_dm_mutex);
1533 		break;
1534 	}
1535 out:
1536 	return NOTIFY_DONE;
1537 }
1538 
1539 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1540 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1541 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1542 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1543 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1544 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1545 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1546 };
1547 
1548 static const struct genl_ops dropmon_ops[] = {
1549 	{
1550 		.cmd = NET_DM_CMD_CONFIG,
1551 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1552 		.doit = net_dm_cmd_config,
1553 		.flags = GENL_ADMIN_PERM,
1554 	},
1555 	{
1556 		.cmd = NET_DM_CMD_START,
1557 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1558 		.doit = net_dm_cmd_trace,
1559 	},
1560 	{
1561 		.cmd = NET_DM_CMD_STOP,
1562 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1563 		.doit = net_dm_cmd_trace,
1564 	},
1565 	{
1566 		.cmd = NET_DM_CMD_CONFIG_GET,
1567 		.doit = net_dm_cmd_config_get,
1568 	},
1569 	{
1570 		.cmd = NET_DM_CMD_STATS_GET,
1571 		.doit = net_dm_cmd_stats_get,
1572 	},
1573 };
1574 
1575 static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1576 			      struct sk_buff *skb, struct genl_info *info)
1577 {
1578 	mutex_lock(&net_dm_mutex);
1579 
1580 	return 0;
1581 }
1582 
1583 static void net_dm_nl_post_doit(const struct genl_ops *ops,
1584 				struct sk_buff *skb, struct genl_info *info)
1585 {
1586 	mutex_unlock(&net_dm_mutex);
1587 }
1588 
1589 static struct genl_family net_drop_monitor_family __ro_after_init = {
1590 	.hdrsize        = 0,
1591 	.name           = "NET_DM",
1592 	.version        = 2,
1593 	.maxattr	= NET_DM_ATTR_MAX,
1594 	.policy		= net_dm_nl_policy,
1595 	.pre_doit	= net_dm_nl_pre_doit,
1596 	.post_doit	= net_dm_nl_post_doit,
1597 	.module		= THIS_MODULE,
1598 	.ops		= dropmon_ops,
1599 	.n_ops		= ARRAY_SIZE(dropmon_ops),
1600 	.mcgrps		= dropmon_mcgrps,
1601 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1602 };
1603 
1604 static struct notifier_block dropmon_net_notifier = {
1605 	.notifier_call = dropmon_net_event
1606 };
1607 
1608 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1609 {
1610 	spin_lock_init(&data->lock);
1611 	skb_queue_head_init(&data->drop_queue);
1612 	u64_stats_init(&data->stats.syncp);
1613 }
1614 
1615 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1616 {
1617 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1618 }
1619 
1620 static void net_dm_cpu_data_init(int cpu)
1621 {
1622 	struct per_cpu_dm_data *data;
1623 
1624 	data = &per_cpu(dm_cpu_data, cpu);
1625 	__net_dm_cpu_data_init(data);
1626 }
1627 
1628 static void net_dm_cpu_data_fini(int cpu)
1629 {
1630 	struct per_cpu_dm_data *data;
1631 
1632 	data = &per_cpu(dm_cpu_data, cpu);
1633 	/* At this point, we should have exclusive access
1634 	 * to this struct and can free the skb inside it.
1635 	 */
1636 	consume_skb(data->skb);
1637 	__net_dm_cpu_data_fini(data);
1638 }
1639 
1640 static void net_dm_hw_cpu_data_init(int cpu)
1641 {
1642 	struct per_cpu_dm_data *hw_data;
1643 
1644 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1645 	__net_dm_cpu_data_init(hw_data);
1646 }
1647 
1648 static void net_dm_hw_cpu_data_fini(int cpu)
1649 {
1650 	struct per_cpu_dm_data *hw_data;
1651 
1652 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1653 	kfree(hw_data->hw_entries);
1654 	__net_dm_cpu_data_fini(hw_data);
1655 }
1656 
1657 static int __init init_net_drop_monitor(void)
1658 {
1659 	int cpu, rc;
1660 
1661 	pr_info("Initializing network drop monitor service\n");
1662 
1663 	if (sizeof(void *) > 8) {
1664 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1665 		return -ENOSPC;
1666 	}
1667 
1668 	rc = genl_register_family(&net_drop_monitor_family);
1669 	if (rc) {
1670 		pr_err("Could not create drop monitor netlink family\n");
1671 		return rc;
1672 	}
1673 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1674 
1675 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1676 	if (rc < 0) {
1677 		pr_crit("Failed to register netdevice notifier\n");
1678 		goto out_unreg;
1679 	}
1680 
1681 	rc = 0;
1682 
1683 	for_each_possible_cpu(cpu) {
1684 		net_dm_cpu_data_init(cpu);
1685 		net_dm_hw_cpu_data_init(cpu);
1686 	}
1687 
1688 	goto out;
1689 
1690 out_unreg:
1691 	genl_unregister_family(&net_drop_monitor_family);
1692 out:
1693 	return rc;
1694 }
1695 
1696 static void exit_net_drop_monitor(void)
1697 {
1698 	int cpu;
1699 
1700 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1701 
1702 	/*
1703 	 * Because of the module_get/put we do in the trace state change path
1704 	 * we are guarnateed not to have any current users when we get here
1705 	 */
1706 
1707 	for_each_possible_cpu(cpu) {
1708 		net_dm_hw_cpu_data_fini(cpu);
1709 		net_dm_cpu_data_fini(cpu);
1710 	}
1711 
1712 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1713 }
1714 
1715 module_init(init_net_drop_monitor);
1716 module_exit(exit_net_drop_monitor);
1717 
1718 MODULE_LICENSE("GPL v2");
1719 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1720 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1721