xref: /openbmc/linux/net/core/drop_monitor.c (revision 04b3c795)
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 net_dm_drop_point *point;
217 	struct nlmsghdr *nlh;
218 	struct nlattr *nla;
219 	int i;
220 	struct sk_buff *dskb;
221 	struct per_cpu_dm_data *data;
222 	unsigned long flags;
223 
224 	local_irq_save(flags);
225 	data = this_cpu_ptr(&dm_cpu_data);
226 	spin_lock(&data->lock);
227 	dskb = data->skb;
228 
229 	if (!dskb)
230 		goto out;
231 
232 	nlh = (struct nlmsghdr *)dskb->data;
233 	nla = genlmsg_data(nlmsg_data(nlh));
234 	msg = nla_data(nla);
235 	point = msg->points;
236 	for (i = 0; i < msg->entries; i++) {
237 		if (!memcmp(&location, &point->pc, sizeof(void *))) {
238 			point->count++;
239 			goto out;
240 		}
241 		point++;
242 	}
243 	if (msg->entries == dm_hit_limit)
244 		goto out;
245 	/*
246 	 * We need to create a new entry
247 	 */
248 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
249 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
250 	memcpy(point->pc, &location, sizeof(void *));
251 	point->count = 1;
252 	msg->entries++;
253 
254 	if (!timer_pending(&data->send_timer)) {
255 		data->send_timer.expires = jiffies + dm_delay * HZ;
256 		add_timer(&data->send_timer);
257 	}
258 
259 out:
260 	spin_unlock_irqrestore(&data->lock, flags);
261 }
262 
263 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
264 {
265 	trace_drop_common(skb, location);
266 }
267 
268 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
269 				int work, int budget)
270 {
271 	struct dm_hw_stat_delta *new_stat;
272 
273 	/*
274 	 * Don't check napi structures with no associated device
275 	 */
276 	if (!napi->dev)
277 		return;
278 
279 	rcu_read_lock();
280 	list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
281 		/*
282 		 * only add a note to our monitor buffer if:
283 		 * 1) this is the dev we received on
284 		 * 2) its after the last_rx delta
285 		 * 3) our rx_dropped count has gone up
286 		 */
287 		if ((new_stat->dev == napi->dev)  &&
288 		    (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
289 		    (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
290 			trace_drop_common(NULL, NULL);
291 			new_stat->last_drop_val = napi->dev->stats.rx_dropped;
292 			new_stat->last_rx = jiffies;
293 			break;
294 		}
295 	}
296 	rcu_read_unlock();
297 }
298 
299 static struct net_dm_hw_entries *
300 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
301 {
302 	struct net_dm_hw_entries *hw_entries;
303 	unsigned long flags;
304 
305 	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
306 			     GFP_KERNEL);
307 	if (!hw_entries) {
308 		/* If the memory allocation failed, we try to perform another
309 		 * allocation in 1/10 second. Otherwise, the probe function
310 		 * will constantly bail out.
311 		 */
312 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
313 	}
314 
315 	spin_lock_irqsave(&hw_data->lock, flags);
316 	swap(hw_data->hw_entries, hw_entries);
317 	spin_unlock_irqrestore(&hw_data->lock, flags);
318 
319 	return hw_entries;
320 }
321 
322 static int net_dm_hw_entry_put(struct sk_buff *msg,
323 			       const struct net_dm_hw_entry *hw_entry)
324 {
325 	struct nlattr *attr;
326 
327 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
328 	if (!attr)
329 		return -EMSGSIZE;
330 
331 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
332 		goto nla_put_failure;
333 
334 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
335 		goto nla_put_failure;
336 
337 	nla_nest_end(msg, attr);
338 
339 	return 0;
340 
341 nla_put_failure:
342 	nla_nest_cancel(msg, attr);
343 	return -EMSGSIZE;
344 }
345 
346 static int net_dm_hw_entries_put(struct sk_buff *msg,
347 				 const struct net_dm_hw_entries *hw_entries)
348 {
349 	struct nlattr *attr;
350 	int i;
351 
352 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
353 	if (!attr)
354 		return -EMSGSIZE;
355 
356 	for (i = 0; i < hw_entries->num_entries; i++) {
357 		int rc;
358 
359 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
360 		if (rc)
361 			goto nla_put_failure;
362 	}
363 
364 	nla_nest_end(msg, attr);
365 
366 	return 0;
367 
368 nla_put_failure:
369 	nla_nest_cancel(msg, attr);
370 	return -EMSGSIZE;
371 }
372 
373 static int
374 net_dm_hw_summary_report_fill(struct sk_buff *msg,
375 			      const struct net_dm_hw_entries *hw_entries)
376 {
377 	struct net_dm_alert_msg anc_hdr = { 0 };
378 	void *hdr;
379 	int rc;
380 
381 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
382 			  NET_DM_CMD_ALERT);
383 	if (!hdr)
384 		return -EMSGSIZE;
385 
386 	/* We need to put the ancillary header in order not to break user
387 	 * space.
388 	 */
389 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
390 		goto nla_put_failure;
391 
392 	rc = net_dm_hw_entries_put(msg, hw_entries);
393 	if (rc)
394 		goto nla_put_failure;
395 
396 	genlmsg_end(msg, hdr);
397 
398 	return 0;
399 
400 nla_put_failure:
401 	genlmsg_cancel(msg, hdr);
402 	return -EMSGSIZE;
403 }
404 
405 static void net_dm_hw_summary_work(struct work_struct *work)
406 {
407 	struct net_dm_hw_entries *hw_entries;
408 	struct per_cpu_dm_data *hw_data;
409 	struct sk_buff *msg;
410 	int rc;
411 
412 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
413 
414 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
415 	if (!hw_entries)
416 		return;
417 
418 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
419 	if (!msg)
420 		goto out;
421 
422 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
423 	if (rc) {
424 		nlmsg_free(msg);
425 		goto out;
426 	}
427 
428 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
429 
430 out:
431 	kfree(hw_entries);
432 }
433 
434 static void
435 net_dm_hw_summary_probe(struct sk_buff *skb,
436 			const struct net_dm_hw_metadata *hw_metadata)
437 {
438 	struct net_dm_hw_entries *hw_entries;
439 	struct net_dm_hw_entry *hw_entry;
440 	struct per_cpu_dm_data *hw_data;
441 	unsigned long flags;
442 	int i;
443 
444 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
445 	spin_lock_irqsave(&hw_data->lock, flags);
446 	hw_entries = hw_data->hw_entries;
447 
448 	if (!hw_entries)
449 		goto out;
450 
451 	for (i = 0; i < hw_entries->num_entries; i++) {
452 		hw_entry = &hw_entries->entries[i];
453 		if (!strncmp(hw_entry->trap_name, hw_metadata->trap_name,
454 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
455 			hw_entry->count++;
456 			goto out;
457 		}
458 	}
459 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
460 		goto out;
461 
462 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
463 	strlcpy(hw_entry->trap_name, hw_metadata->trap_name,
464 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
465 	hw_entry->count = 1;
466 	hw_entries->num_entries++;
467 
468 	if (!timer_pending(&hw_data->send_timer)) {
469 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
470 		add_timer(&hw_data->send_timer);
471 	}
472 
473 out:
474 	spin_unlock_irqrestore(&hw_data->lock, flags);
475 }
476 
477 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
478 	.kfree_skb_probe	= trace_kfree_skb_hit,
479 	.napi_poll_probe	= trace_napi_poll_hit,
480 	.work_item_func		= send_dm_alert,
481 	.hw_work_item_func	= net_dm_hw_summary_work,
482 	.hw_probe		= net_dm_hw_summary_probe,
483 };
484 
485 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
486 					      struct sk_buff *skb,
487 					      void *location)
488 {
489 	ktime_t tstamp = ktime_get_real();
490 	struct per_cpu_dm_data *data;
491 	struct sk_buff *nskb;
492 	unsigned long flags;
493 
494 	if (!skb_mac_header_was_set(skb))
495 		return;
496 
497 	nskb = skb_clone(skb, GFP_ATOMIC);
498 	if (!nskb)
499 		return;
500 
501 	NET_DM_SKB_CB(nskb)->pc = location;
502 	/* Override the timestamp because we care about the time when the
503 	 * packet was dropped.
504 	 */
505 	nskb->tstamp = tstamp;
506 
507 	data = this_cpu_ptr(&dm_cpu_data);
508 
509 	spin_lock_irqsave(&data->drop_queue.lock, flags);
510 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
511 		__skb_queue_tail(&data->drop_queue, nskb);
512 	else
513 		goto unlock_free;
514 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
515 
516 	schedule_work(&data->dm_alert_work);
517 
518 	return;
519 
520 unlock_free:
521 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
522 	u64_stats_update_begin(&data->stats.syncp);
523 	data->stats.dropped++;
524 	u64_stats_update_end(&data->stats.syncp);
525 	consume_skb(nskb);
526 }
527 
528 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
529 					      struct napi_struct *napi,
530 					      int work, int budget)
531 {
532 }
533 
534 static size_t net_dm_in_port_size(void)
535 {
536 	       /* NET_DM_ATTR_IN_PORT nest */
537 	return nla_total_size(0) +
538 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
539 	       nla_total_size(sizeof(u32)) +
540 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
541 	       nla_total_size(IFNAMSIZ + 1);
542 }
543 
544 #define NET_DM_MAX_SYMBOL_LEN 40
545 
546 static size_t net_dm_packet_report_size(size_t payload_len)
547 {
548 	size_t size;
549 
550 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
551 
552 	return NLMSG_ALIGN(size) +
553 	       /* NET_DM_ATTR_ORIGIN */
554 	       nla_total_size(sizeof(u16)) +
555 	       /* NET_DM_ATTR_PC */
556 	       nla_total_size(sizeof(u64)) +
557 	       /* NET_DM_ATTR_SYMBOL */
558 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
559 	       /* NET_DM_ATTR_IN_PORT */
560 	       net_dm_in_port_size() +
561 	       /* NET_DM_ATTR_TIMESTAMP */
562 	       nla_total_size(sizeof(u64)) +
563 	       /* NET_DM_ATTR_ORIG_LEN */
564 	       nla_total_size(sizeof(u32)) +
565 	       /* NET_DM_ATTR_PROTO */
566 	       nla_total_size(sizeof(u16)) +
567 	       /* NET_DM_ATTR_PAYLOAD */
568 	       nla_total_size(payload_len);
569 }
570 
571 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
572 					    const char *name)
573 {
574 	struct nlattr *attr;
575 
576 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
577 	if (!attr)
578 		return -EMSGSIZE;
579 
580 	if (ifindex &&
581 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
582 		goto nla_put_failure;
583 
584 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
585 		goto nla_put_failure;
586 
587 	nla_nest_end(msg, attr);
588 
589 	return 0;
590 
591 nla_put_failure:
592 	nla_nest_cancel(msg, attr);
593 	return -EMSGSIZE;
594 }
595 
596 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
597 				     size_t payload_len)
598 {
599 	u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
600 	char buf[NET_DM_MAX_SYMBOL_LEN];
601 	struct nlattr *attr;
602 	void *hdr;
603 	int rc;
604 
605 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
606 			  NET_DM_CMD_PACKET_ALERT);
607 	if (!hdr)
608 		return -EMSGSIZE;
609 
610 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
611 		goto nla_put_failure;
612 
613 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
614 		goto nla_put_failure;
615 
616 	snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
617 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
618 		goto nla_put_failure;
619 
620 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
621 	if (rc)
622 		goto nla_put_failure;
623 
624 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
625 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
626 		goto nla_put_failure;
627 
628 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
629 		goto nla_put_failure;
630 
631 	if (!payload_len)
632 		goto out;
633 
634 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
635 		goto nla_put_failure;
636 
637 	attr = skb_put(msg, nla_total_size(payload_len));
638 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
639 	attr->nla_len = nla_attr_size(payload_len);
640 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
641 		goto nla_put_failure;
642 
643 out:
644 	genlmsg_end(msg, hdr);
645 
646 	return 0;
647 
648 nla_put_failure:
649 	genlmsg_cancel(msg, hdr);
650 	return -EMSGSIZE;
651 }
652 
653 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
654 
655 static void net_dm_packet_report(struct sk_buff *skb)
656 {
657 	struct sk_buff *msg;
658 	size_t payload_len;
659 	int rc;
660 
661 	/* Make sure we start copying the packet from the MAC header */
662 	if (skb->data > skb_mac_header(skb))
663 		skb_push(skb, skb->data - skb_mac_header(skb));
664 	else
665 		skb_pull(skb, skb_mac_header(skb) - skb->data);
666 
667 	/* Ensure packet fits inside a single netlink attribute */
668 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
669 	if (net_dm_trunc_len)
670 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
671 
672 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
673 	if (!msg)
674 		goto out;
675 
676 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
677 	if (rc) {
678 		nlmsg_free(msg);
679 		goto out;
680 	}
681 
682 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
683 
684 out:
685 	consume_skb(skb);
686 }
687 
688 static void net_dm_packet_work(struct work_struct *work)
689 {
690 	struct per_cpu_dm_data *data;
691 	struct sk_buff_head list;
692 	struct sk_buff *skb;
693 	unsigned long flags;
694 
695 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
696 
697 	__skb_queue_head_init(&list);
698 
699 	spin_lock_irqsave(&data->drop_queue.lock, flags);
700 	skb_queue_splice_tail_init(&data->drop_queue, &list);
701 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
702 
703 	while ((skb = __skb_dequeue(&list)))
704 		net_dm_packet_report(skb);
705 }
706 
707 static size_t
708 net_dm_flow_action_cookie_size(const struct net_dm_hw_metadata *hw_metadata)
709 {
710 	return hw_metadata->fa_cookie ?
711 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
712 }
713 
714 static size_t
715 net_dm_hw_packet_report_size(size_t payload_len,
716 			     const struct net_dm_hw_metadata *hw_metadata)
717 {
718 	size_t size;
719 
720 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
721 
722 	return NLMSG_ALIGN(size) +
723 	       /* NET_DM_ATTR_ORIGIN */
724 	       nla_total_size(sizeof(u16)) +
725 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
726 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
727 	       /* NET_DM_ATTR_HW_TRAP_NAME */
728 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
729 	       /* NET_DM_ATTR_IN_PORT */
730 	       net_dm_in_port_size() +
731 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
732 	       net_dm_flow_action_cookie_size(hw_metadata) +
733 	       /* NET_DM_ATTR_TIMESTAMP */
734 	       nla_total_size(sizeof(u64)) +
735 	       /* NET_DM_ATTR_ORIG_LEN */
736 	       nla_total_size(sizeof(u32)) +
737 	       /* NET_DM_ATTR_PROTO */
738 	       nla_total_size(sizeof(u16)) +
739 	       /* NET_DM_ATTR_PAYLOAD */
740 	       nla_total_size(payload_len);
741 }
742 
743 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
744 					struct sk_buff *skb, size_t payload_len)
745 {
746 	struct net_dm_hw_metadata *hw_metadata;
747 	struct nlattr *attr;
748 	void *hdr;
749 
750 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
751 
752 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
753 			  NET_DM_CMD_PACKET_ALERT);
754 	if (!hdr)
755 		return -EMSGSIZE;
756 
757 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
758 		goto nla_put_failure;
759 
760 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
761 			   hw_metadata->trap_group_name))
762 		goto nla_put_failure;
763 
764 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
765 			   hw_metadata->trap_name))
766 		goto nla_put_failure;
767 
768 	if (hw_metadata->input_dev) {
769 		struct net_device *dev = hw_metadata->input_dev;
770 		int rc;
771 
772 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
773 						      dev->name);
774 		if (rc)
775 			goto nla_put_failure;
776 	}
777 
778 	if (hw_metadata->fa_cookie &&
779 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
780 		    hw_metadata->fa_cookie->cookie_len,
781 		    hw_metadata->fa_cookie->cookie))
782 		goto nla_put_failure;
783 
784 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
785 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
786 		goto nla_put_failure;
787 
788 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
789 		goto nla_put_failure;
790 
791 	if (!payload_len)
792 		goto out;
793 
794 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
795 		goto nla_put_failure;
796 
797 	attr = skb_put(msg, nla_total_size(payload_len));
798 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
799 	attr->nla_len = nla_attr_size(payload_len);
800 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
801 		goto nla_put_failure;
802 
803 out:
804 	genlmsg_end(msg, hdr);
805 
806 	return 0;
807 
808 nla_put_failure:
809 	genlmsg_cancel(msg, hdr);
810 	return -EMSGSIZE;
811 }
812 
813 static struct net_dm_hw_metadata *
814 net_dm_hw_metadata_clone(const struct net_dm_hw_metadata *hw_metadata)
815 {
816 	const struct flow_action_cookie *fa_cookie;
817 	struct net_dm_hw_metadata *n_hw_metadata;
818 	const char *trap_group_name;
819 	const char *trap_name;
820 
821 	n_hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
822 	if (!n_hw_metadata)
823 		return NULL;
824 
825 	trap_group_name = kstrdup(hw_metadata->trap_group_name, GFP_ATOMIC);
826 	if (!trap_group_name)
827 		goto free_hw_metadata;
828 	n_hw_metadata->trap_group_name = trap_group_name;
829 
830 	trap_name = kstrdup(hw_metadata->trap_name, GFP_ATOMIC);
831 	if (!trap_name)
832 		goto free_trap_group;
833 	n_hw_metadata->trap_name = trap_name;
834 
835 	if (hw_metadata->fa_cookie) {
836 		size_t cookie_size = sizeof(*fa_cookie) +
837 				     hw_metadata->fa_cookie->cookie_len;
838 
839 		fa_cookie = kmemdup(hw_metadata->fa_cookie, cookie_size,
840 				    GFP_ATOMIC);
841 		if (!fa_cookie)
842 			goto free_trap_name;
843 		n_hw_metadata->fa_cookie = fa_cookie;
844 	}
845 
846 	n_hw_metadata->input_dev = hw_metadata->input_dev;
847 	if (n_hw_metadata->input_dev)
848 		dev_hold(n_hw_metadata->input_dev);
849 
850 	return n_hw_metadata;
851 
852 free_trap_name:
853 	kfree(trap_name);
854 free_trap_group:
855 	kfree(trap_group_name);
856 free_hw_metadata:
857 	kfree(n_hw_metadata);
858 	return NULL;
859 }
860 
861 static void
862 net_dm_hw_metadata_free(const struct net_dm_hw_metadata *hw_metadata)
863 {
864 	if (hw_metadata->input_dev)
865 		dev_put(hw_metadata->input_dev);
866 	kfree(hw_metadata->fa_cookie);
867 	kfree(hw_metadata->trap_name);
868 	kfree(hw_metadata->trap_group_name);
869 	kfree(hw_metadata);
870 }
871 
872 static void net_dm_hw_packet_report(struct sk_buff *skb)
873 {
874 	struct net_dm_hw_metadata *hw_metadata;
875 	struct sk_buff *msg;
876 	size_t payload_len;
877 	int rc;
878 
879 	if (skb->data > skb_mac_header(skb))
880 		skb_push(skb, skb->data - skb_mac_header(skb));
881 	else
882 		skb_pull(skb, skb_mac_header(skb) - skb->data);
883 
884 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
885 	if (net_dm_trunc_len)
886 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
887 
888 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
889 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
890 			GFP_KERNEL);
891 	if (!msg)
892 		goto out;
893 
894 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
895 	if (rc) {
896 		nlmsg_free(msg);
897 		goto out;
898 	}
899 
900 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
901 
902 out:
903 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
904 	consume_skb(skb);
905 }
906 
907 static void net_dm_hw_packet_work(struct work_struct *work)
908 {
909 	struct per_cpu_dm_data *hw_data;
910 	struct sk_buff_head list;
911 	struct sk_buff *skb;
912 	unsigned long flags;
913 
914 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
915 
916 	__skb_queue_head_init(&list);
917 
918 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
919 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
920 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
921 
922 	while ((skb = __skb_dequeue(&list)))
923 		net_dm_hw_packet_report(skb);
924 }
925 
926 static void
927 net_dm_hw_packet_probe(struct sk_buff *skb,
928 		       const struct net_dm_hw_metadata *hw_metadata)
929 {
930 	struct net_dm_hw_metadata *n_hw_metadata;
931 	ktime_t tstamp = ktime_get_real();
932 	struct per_cpu_dm_data *hw_data;
933 	struct sk_buff *nskb;
934 	unsigned long flags;
935 
936 	if (!skb_mac_header_was_set(skb))
937 		return;
938 
939 	nskb = skb_clone(skb, GFP_ATOMIC);
940 	if (!nskb)
941 		return;
942 
943 	n_hw_metadata = net_dm_hw_metadata_clone(hw_metadata);
944 	if (!n_hw_metadata)
945 		goto free;
946 
947 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
948 	nskb->tstamp = tstamp;
949 
950 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
951 
952 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
953 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
954 		__skb_queue_tail(&hw_data->drop_queue, nskb);
955 	else
956 		goto unlock_free;
957 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
958 
959 	schedule_work(&hw_data->dm_alert_work);
960 
961 	return;
962 
963 unlock_free:
964 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
965 	u64_stats_update_begin(&hw_data->stats.syncp);
966 	hw_data->stats.dropped++;
967 	u64_stats_update_end(&hw_data->stats.syncp);
968 	net_dm_hw_metadata_free(n_hw_metadata);
969 free:
970 	consume_skb(nskb);
971 }
972 
973 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
974 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
975 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
976 	.work_item_func		= net_dm_packet_work,
977 	.hw_work_item_func	= net_dm_hw_packet_work,
978 	.hw_probe		= net_dm_hw_packet_probe,
979 };
980 
981 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
982 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
983 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
984 };
985 
986 void net_dm_hw_report(struct sk_buff *skb,
987 		      const struct net_dm_hw_metadata *hw_metadata)
988 {
989 	rcu_read_lock();
990 
991 	if (!monitor_hw)
992 		goto out;
993 
994 	net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
995 
996 out:
997 	rcu_read_unlock();
998 }
999 EXPORT_SYMBOL_GPL(net_dm_hw_report);
1000 
1001 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1002 {
1003 	const struct net_dm_alert_ops *ops;
1004 	int cpu;
1005 
1006 	if (monitor_hw) {
1007 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1008 		return -EAGAIN;
1009 	}
1010 
1011 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1012 
1013 	if (!try_module_get(THIS_MODULE)) {
1014 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1015 		return -ENODEV;
1016 	}
1017 
1018 	for_each_possible_cpu(cpu) {
1019 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1020 		struct net_dm_hw_entries *hw_entries;
1021 
1022 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1023 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1024 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1025 		kfree(hw_entries);
1026 	}
1027 
1028 	monitor_hw = true;
1029 
1030 	return 0;
1031 }
1032 
1033 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1034 {
1035 	int cpu;
1036 
1037 	if (!monitor_hw) {
1038 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1039 		return;
1040 	}
1041 
1042 	monitor_hw = false;
1043 
1044 	/* After this call returns we are guaranteed that no CPU is processing
1045 	 * any hardware drops.
1046 	 */
1047 	synchronize_rcu();
1048 
1049 	for_each_possible_cpu(cpu) {
1050 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1051 		struct sk_buff *skb;
1052 
1053 		del_timer_sync(&hw_data->send_timer);
1054 		cancel_work_sync(&hw_data->dm_alert_work);
1055 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1056 			struct net_dm_hw_metadata *hw_metadata;
1057 
1058 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1059 			net_dm_hw_metadata_free(hw_metadata);
1060 			consume_skb(skb);
1061 		}
1062 	}
1063 
1064 	module_put(THIS_MODULE);
1065 }
1066 
1067 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1068 {
1069 	const struct net_dm_alert_ops *ops;
1070 	int cpu, rc;
1071 
1072 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1073 
1074 	if (!try_module_get(THIS_MODULE)) {
1075 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1076 		return -ENODEV;
1077 	}
1078 
1079 	for_each_possible_cpu(cpu) {
1080 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1081 		struct sk_buff *skb;
1082 
1083 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1084 		timer_setup(&data->send_timer, sched_send_work, 0);
1085 		/* Allocate a new per-CPU skb for the summary alert message and
1086 		 * free the old one which might contain stale data from
1087 		 * previous tracing.
1088 		 */
1089 		skb = reset_per_cpu_data(data);
1090 		consume_skb(skb);
1091 	}
1092 
1093 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1094 	if (rc) {
1095 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1096 		goto err_module_put;
1097 	}
1098 
1099 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1100 	if (rc) {
1101 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1102 		goto err_unregister_trace;
1103 	}
1104 
1105 	return 0;
1106 
1107 err_unregister_trace:
1108 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1109 err_module_put:
1110 	module_put(THIS_MODULE);
1111 	return rc;
1112 }
1113 
1114 static void net_dm_trace_off_set(void)
1115 {
1116 	struct dm_hw_stat_delta *new_stat, *temp;
1117 	const struct net_dm_alert_ops *ops;
1118 	int cpu;
1119 
1120 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1121 
1122 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1123 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1124 
1125 	tracepoint_synchronize_unregister();
1126 
1127 	/* Make sure we do not send notifications to user space after request
1128 	 * to stop tracing returns.
1129 	 */
1130 	for_each_possible_cpu(cpu) {
1131 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1132 		struct sk_buff *skb;
1133 
1134 		del_timer_sync(&data->send_timer);
1135 		cancel_work_sync(&data->dm_alert_work);
1136 		while ((skb = __skb_dequeue(&data->drop_queue)))
1137 			consume_skb(skb);
1138 	}
1139 
1140 	list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1141 		if (new_stat->dev == NULL) {
1142 			list_del_rcu(&new_stat->list);
1143 			kfree_rcu(new_stat, rcu);
1144 		}
1145 	}
1146 
1147 	module_put(THIS_MODULE);
1148 }
1149 
1150 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1151 {
1152 	int rc = 0;
1153 
1154 	if (state == trace_state) {
1155 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1156 		return -EAGAIN;
1157 	}
1158 
1159 	switch (state) {
1160 	case TRACE_ON:
1161 		rc = net_dm_trace_on_set(extack);
1162 		break;
1163 	case TRACE_OFF:
1164 		net_dm_trace_off_set();
1165 		break;
1166 	default:
1167 		rc = 1;
1168 		break;
1169 	}
1170 
1171 	if (!rc)
1172 		trace_state = state;
1173 	else
1174 		rc = -EINPROGRESS;
1175 
1176 	return rc;
1177 }
1178 
1179 static bool net_dm_is_monitoring(void)
1180 {
1181 	return trace_state == TRACE_ON || monitor_hw;
1182 }
1183 
1184 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1185 					   enum net_dm_alert_mode *p_alert_mode)
1186 {
1187 	u8 val;
1188 
1189 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1190 
1191 	switch (val) {
1192 	case NET_DM_ALERT_MODE_SUMMARY: /* fall-through */
1193 	case NET_DM_ALERT_MODE_PACKET:
1194 		*p_alert_mode = val;
1195 		break;
1196 	default:
1197 		return -EINVAL;
1198 	}
1199 
1200 	return 0;
1201 }
1202 
1203 static int net_dm_alert_mode_set(struct genl_info *info)
1204 {
1205 	struct netlink_ext_ack *extack = info->extack;
1206 	enum net_dm_alert_mode alert_mode;
1207 	int rc;
1208 
1209 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1210 		return 0;
1211 
1212 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1213 	if (rc) {
1214 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1215 		return -EINVAL;
1216 	}
1217 
1218 	net_dm_alert_mode = alert_mode;
1219 
1220 	return 0;
1221 }
1222 
1223 static void net_dm_trunc_len_set(struct genl_info *info)
1224 {
1225 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1226 		return;
1227 
1228 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1229 }
1230 
1231 static void net_dm_queue_len_set(struct genl_info *info)
1232 {
1233 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1234 		return;
1235 
1236 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1237 }
1238 
1239 static int net_dm_cmd_config(struct sk_buff *skb,
1240 			struct genl_info *info)
1241 {
1242 	struct netlink_ext_ack *extack = info->extack;
1243 	int rc;
1244 
1245 	if (net_dm_is_monitoring()) {
1246 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1247 		return -EBUSY;
1248 	}
1249 
1250 	rc = net_dm_alert_mode_set(info);
1251 	if (rc)
1252 		return rc;
1253 
1254 	net_dm_trunc_len_set(info);
1255 
1256 	net_dm_queue_len_set(info);
1257 
1258 	return 0;
1259 }
1260 
1261 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1262 				struct netlink_ext_ack *extack)
1263 {
1264 	bool sw_set = false;
1265 	int rc;
1266 
1267 	if (set_sw) {
1268 		rc = set_all_monitor_traces(TRACE_ON, extack);
1269 		if (rc)
1270 			return rc;
1271 		sw_set = true;
1272 	}
1273 
1274 	if (set_hw) {
1275 		rc = net_dm_hw_monitor_start(extack);
1276 		if (rc)
1277 			goto err_monitor_hw;
1278 	}
1279 
1280 	return 0;
1281 
1282 err_monitor_hw:
1283 	if (sw_set)
1284 		set_all_monitor_traces(TRACE_OFF, extack);
1285 	return rc;
1286 }
1287 
1288 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1289 				struct netlink_ext_ack *extack)
1290 {
1291 	if (set_hw)
1292 		net_dm_hw_monitor_stop(extack);
1293 	if (set_sw)
1294 		set_all_monitor_traces(TRACE_OFF, extack);
1295 }
1296 
1297 static int net_dm_cmd_trace(struct sk_buff *skb,
1298 			struct genl_info *info)
1299 {
1300 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1301 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1302 	struct netlink_ext_ack *extack = info->extack;
1303 
1304 	/* To maintain backward compatibility, we start / stop monitoring of
1305 	 * software drops if no flag is specified.
1306 	 */
1307 	if (!set_sw && !set_hw)
1308 		set_sw = true;
1309 
1310 	switch (info->genlhdr->cmd) {
1311 	case NET_DM_CMD_START:
1312 		return net_dm_monitor_start(set_sw, set_hw, extack);
1313 	case NET_DM_CMD_STOP:
1314 		net_dm_monitor_stop(set_sw, set_hw, extack);
1315 		return 0;
1316 	}
1317 
1318 	return -EOPNOTSUPP;
1319 }
1320 
1321 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1322 {
1323 	void *hdr;
1324 
1325 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1326 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1327 	if (!hdr)
1328 		return -EMSGSIZE;
1329 
1330 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1331 		goto nla_put_failure;
1332 
1333 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1334 		goto nla_put_failure;
1335 
1336 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1337 		goto nla_put_failure;
1338 
1339 	genlmsg_end(msg, hdr);
1340 
1341 	return 0;
1342 
1343 nla_put_failure:
1344 	genlmsg_cancel(msg, hdr);
1345 	return -EMSGSIZE;
1346 }
1347 
1348 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1349 {
1350 	struct sk_buff *msg;
1351 	int rc;
1352 
1353 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1354 	if (!msg)
1355 		return -ENOMEM;
1356 
1357 	rc = net_dm_config_fill(msg, info);
1358 	if (rc)
1359 		goto free_msg;
1360 
1361 	return genlmsg_reply(msg, info);
1362 
1363 free_msg:
1364 	nlmsg_free(msg);
1365 	return rc;
1366 }
1367 
1368 static void net_dm_stats_read(struct net_dm_stats *stats)
1369 {
1370 	int cpu;
1371 
1372 	memset(stats, 0, sizeof(*stats));
1373 	for_each_possible_cpu(cpu) {
1374 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1375 		struct net_dm_stats *cpu_stats = &data->stats;
1376 		unsigned int start;
1377 		u64 dropped;
1378 
1379 		do {
1380 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1381 			dropped = cpu_stats->dropped;
1382 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1383 
1384 		stats->dropped += dropped;
1385 	}
1386 }
1387 
1388 static int net_dm_stats_put(struct sk_buff *msg)
1389 {
1390 	struct net_dm_stats stats;
1391 	struct nlattr *attr;
1392 
1393 	net_dm_stats_read(&stats);
1394 
1395 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1396 	if (!attr)
1397 		return -EMSGSIZE;
1398 
1399 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1400 			      stats.dropped, NET_DM_ATTR_PAD))
1401 		goto nla_put_failure;
1402 
1403 	nla_nest_end(msg, attr);
1404 
1405 	return 0;
1406 
1407 nla_put_failure:
1408 	nla_nest_cancel(msg, attr);
1409 	return -EMSGSIZE;
1410 }
1411 
1412 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1413 {
1414 	int cpu;
1415 
1416 	memset(stats, 0, sizeof(*stats));
1417 	for_each_possible_cpu(cpu) {
1418 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1419 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1420 		unsigned int start;
1421 		u64 dropped;
1422 
1423 		do {
1424 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1425 			dropped = cpu_stats->dropped;
1426 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1427 
1428 		stats->dropped += dropped;
1429 	}
1430 }
1431 
1432 static int net_dm_hw_stats_put(struct sk_buff *msg)
1433 {
1434 	struct net_dm_stats stats;
1435 	struct nlattr *attr;
1436 
1437 	net_dm_hw_stats_read(&stats);
1438 
1439 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1440 	if (!attr)
1441 		return -EMSGSIZE;
1442 
1443 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1444 			      stats.dropped, NET_DM_ATTR_PAD))
1445 		goto nla_put_failure;
1446 
1447 	nla_nest_end(msg, attr);
1448 
1449 	return 0;
1450 
1451 nla_put_failure:
1452 	nla_nest_cancel(msg, attr);
1453 	return -EMSGSIZE;
1454 }
1455 
1456 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1457 {
1458 	void *hdr;
1459 	int rc;
1460 
1461 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1462 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1463 	if (!hdr)
1464 		return -EMSGSIZE;
1465 
1466 	rc = net_dm_stats_put(msg);
1467 	if (rc)
1468 		goto nla_put_failure;
1469 
1470 	rc = net_dm_hw_stats_put(msg);
1471 	if (rc)
1472 		goto nla_put_failure;
1473 
1474 	genlmsg_end(msg, hdr);
1475 
1476 	return 0;
1477 
1478 nla_put_failure:
1479 	genlmsg_cancel(msg, hdr);
1480 	return -EMSGSIZE;
1481 }
1482 
1483 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1484 {
1485 	struct sk_buff *msg;
1486 	int rc;
1487 
1488 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1489 	if (!msg)
1490 		return -ENOMEM;
1491 
1492 	rc = net_dm_stats_fill(msg, info);
1493 	if (rc)
1494 		goto free_msg;
1495 
1496 	return genlmsg_reply(msg, info);
1497 
1498 free_msg:
1499 	nlmsg_free(msg);
1500 	return rc;
1501 }
1502 
1503 static int dropmon_net_event(struct notifier_block *ev_block,
1504 			     unsigned long event, void *ptr)
1505 {
1506 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1507 	struct dm_hw_stat_delta *new_stat = NULL;
1508 	struct dm_hw_stat_delta *tmp;
1509 
1510 	switch (event) {
1511 	case NETDEV_REGISTER:
1512 		new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
1513 
1514 		if (!new_stat)
1515 			goto out;
1516 
1517 		new_stat->dev = dev;
1518 		new_stat->last_rx = jiffies;
1519 		mutex_lock(&net_dm_mutex);
1520 		list_add_rcu(&new_stat->list, &hw_stats_list);
1521 		mutex_unlock(&net_dm_mutex);
1522 		break;
1523 	case NETDEV_UNREGISTER:
1524 		mutex_lock(&net_dm_mutex);
1525 		list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1526 			if (new_stat->dev == dev) {
1527 				new_stat->dev = NULL;
1528 				if (trace_state == TRACE_OFF) {
1529 					list_del_rcu(&new_stat->list);
1530 					kfree_rcu(new_stat, rcu);
1531 					break;
1532 				}
1533 			}
1534 		}
1535 		mutex_unlock(&net_dm_mutex);
1536 		break;
1537 	}
1538 out:
1539 	return NOTIFY_DONE;
1540 }
1541 
1542 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1543 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1544 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1545 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1546 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1547 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1548 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1549 };
1550 
1551 static const struct genl_ops dropmon_ops[] = {
1552 	{
1553 		.cmd = NET_DM_CMD_CONFIG,
1554 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1555 		.doit = net_dm_cmd_config,
1556 		.flags = GENL_ADMIN_PERM,
1557 	},
1558 	{
1559 		.cmd = NET_DM_CMD_START,
1560 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1561 		.doit = net_dm_cmd_trace,
1562 	},
1563 	{
1564 		.cmd = NET_DM_CMD_STOP,
1565 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1566 		.doit = net_dm_cmd_trace,
1567 	},
1568 	{
1569 		.cmd = NET_DM_CMD_CONFIG_GET,
1570 		.doit = net_dm_cmd_config_get,
1571 	},
1572 	{
1573 		.cmd = NET_DM_CMD_STATS_GET,
1574 		.doit = net_dm_cmd_stats_get,
1575 	},
1576 };
1577 
1578 static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1579 			      struct sk_buff *skb, struct genl_info *info)
1580 {
1581 	mutex_lock(&net_dm_mutex);
1582 
1583 	return 0;
1584 }
1585 
1586 static void net_dm_nl_post_doit(const struct genl_ops *ops,
1587 				struct sk_buff *skb, struct genl_info *info)
1588 {
1589 	mutex_unlock(&net_dm_mutex);
1590 }
1591 
1592 static struct genl_family net_drop_monitor_family __ro_after_init = {
1593 	.hdrsize        = 0,
1594 	.name           = "NET_DM",
1595 	.version        = 2,
1596 	.maxattr	= NET_DM_ATTR_MAX,
1597 	.policy		= net_dm_nl_policy,
1598 	.pre_doit	= net_dm_nl_pre_doit,
1599 	.post_doit	= net_dm_nl_post_doit,
1600 	.module		= THIS_MODULE,
1601 	.ops		= dropmon_ops,
1602 	.n_ops		= ARRAY_SIZE(dropmon_ops),
1603 	.mcgrps		= dropmon_mcgrps,
1604 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1605 };
1606 
1607 static struct notifier_block dropmon_net_notifier = {
1608 	.notifier_call = dropmon_net_event
1609 };
1610 
1611 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1612 {
1613 	spin_lock_init(&data->lock);
1614 	skb_queue_head_init(&data->drop_queue);
1615 	u64_stats_init(&data->stats.syncp);
1616 }
1617 
1618 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1619 {
1620 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1621 }
1622 
1623 static void net_dm_cpu_data_init(int cpu)
1624 {
1625 	struct per_cpu_dm_data *data;
1626 
1627 	data = &per_cpu(dm_cpu_data, cpu);
1628 	__net_dm_cpu_data_init(data);
1629 }
1630 
1631 static void net_dm_cpu_data_fini(int cpu)
1632 {
1633 	struct per_cpu_dm_data *data;
1634 
1635 	data = &per_cpu(dm_cpu_data, cpu);
1636 	/* At this point, we should have exclusive access
1637 	 * to this struct and can free the skb inside it.
1638 	 */
1639 	consume_skb(data->skb);
1640 	__net_dm_cpu_data_fini(data);
1641 }
1642 
1643 static void net_dm_hw_cpu_data_init(int cpu)
1644 {
1645 	struct per_cpu_dm_data *hw_data;
1646 
1647 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1648 	__net_dm_cpu_data_init(hw_data);
1649 }
1650 
1651 static void net_dm_hw_cpu_data_fini(int cpu)
1652 {
1653 	struct per_cpu_dm_data *hw_data;
1654 
1655 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1656 	kfree(hw_data->hw_entries);
1657 	__net_dm_cpu_data_fini(hw_data);
1658 }
1659 
1660 static int __init init_net_drop_monitor(void)
1661 {
1662 	int cpu, rc;
1663 
1664 	pr_info("Initializing network drop monitor service\n");
1665 
1666 	if (sizeof(void *) > 8) {
1667 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1668 		return -ENOSPC;
1669 	}
1670 
1671 	rc = genl_register_family(&net_drop_monitor_family);
1672 	if (rc) {
1673 		pr_err("Could not create drop monitor netlink family\n");
1674 		return rc;
1675 	}
1676 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1677 
1678 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1679 	if (rc < 0) {
1680 		pr_crit("Failed to register netdevice notifier\n");
1681 		goto out_unreg;
1682 	}
1683 
1684 	rc = 0;
1685 
1686 	for_each_possible_cpu(cpu) {
1687 		net_dm_cpu_data_init(cpu);
1688 		net_dm_hw_cpu_data_init(cpu);
1689 	}
1690 
1691 	goto out;
1692 
1693 out_unreg:
1694 	genl_unregister_family(&net_drop_monitor_family);
1695 out:
1696 	return rc;
1697 }
1698 
1699 static void exit_net_drop_monitor(void)
1700 {
1701 	int cpu;
1702 
1703 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1704 
1705 	/*
1706 	 * Because of the module_get/put we do in the trace state change path
1707 	 * we are guarnateed not to have any current users when we get here
1708 	 */
1709 
1710 	for_each_possible_cpu(cpu) {
1711 		net_dm_hw_cpu_data_fini(cpu);
1712 		net_dm_cpu_data_fini(cpu);
1713 	}
1714 
1715 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1716 }
1717 
1718 module_init(init_net_drop_monitor);
1719 module_exit(exit_net_drop_monitor);
1720 
1721 MODULE_LICENSE("GPL v2");
1722 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1723 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1724