xref: /linux/net/core/drop_monitor.c (revision 067012974c8ae31a8886046df082aeba93592972)
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/genetlink.h>
30 #include <net/netevent.h>
31 #include <net/flow_offload.h>
32 #include <net/devlink.h>
33 
34 #include <trace/events/skb.h>
35 #include <trace/events/napi.h>
36 #include <trace/events/devlink.h>
37 
38 #include <asm/unaligned.h>
39 
40 #define TRACE_ON 1
41 #define TRACE_OFF 0
42 
43 /*
44  * Globals, our netlink socket pointer
45  * and the work handle that will send up
46  * netlink alerts
47  */
48 static int trace_state = TRACE_OFF;
49 static bool monitor_hw;
50 
51 /* net_dm_mutex
52  *
53  * An overall lock guarding every operation coming from userspace.
54  * It also guards the global 'hw_stats_list' list.
55  */
56 static DEFINE_MUTEX(net_dm_mutex);
57 
58 struct net_dm_stats {
59 	u64 dropped;
60 	struct u64_stats_sync syncp;
61 };
62 
63 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
64 
65 struct net_dm_hw_entry {
66 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
67 	u32 count;
68 };
69 
70 struct net_dm_hw_entries {
71 	u32 num_entries;
72 	struct net_dm_hw_entry entries[];
73 };
74 
75 struct per_cpu_dm_data {
76 	spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
77 					 * 'send_timer'
78 					 */
79 	union {
80 		struct sk_buff			*skb;
81 		struct net_dm_hw_entries	*hw_entries;
82 	};
83 	struct sk_buff_head	drop_queue;
84 	struct work_struct	dm_alert_work;
85 	struct timer_list	send_timer;
86 	struct net_dm_stats	stats;
87 };
88 
89 struct dm_hw_stat_delta {
90 	struct net_device *dev;
91 	unsigned long last_rx;
92 	struct list_head list;
93 	struct rcu_head rcu;
94 	unsigned long last_drop_val;
95 };
96 
97 static struct genl_family net_drop_monitor_family;
98 
99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
100 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
101 
102 static int dm_hit_limit = 64;
103 static int dm_delay = 1;
104 static unsigned long dm_hw_check_delta = 2*HZ;
105 static LIST_HEAD(hw_stats_list);
106 
107 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
108 static u32 net_dm_trunc_len;
109 static u32 net_dm_queue_len = 1000;
110 
111 struct net_dm_alert_ops {
112 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
113 				void *location);
114 	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
115 				int work, int budget);
116 	void (*work_item_func)(struct work_struct *work);
117 	void (*hw_work_item_func)(struct work_struct *work);
118 	void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
119 			      struct sk_buff *skb,
120 			      const struct devlink_trap_metadata *metadata);
121 };
122 
123 struct net_dm_skb_cb {
124 	union {
125 		struct devlink_trap_metadata *hw_metadata;
126 		void *pc;
127 	};
128 };
129 
130 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
131 
132 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
133 {
134 	size_t al;
135 	struct net_dm_alert_msg *msg;
136 	struct nlattr *nla;
137 	struct sk_buff *skb;
138 	unsigned long flags;
139 	void *msg_header;
140 
141 	al = sizeof(struct net_dm_alert_msg);
142 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
143 	al += sizeof(struct nlattr);
144 
145 	skb = genlmsg_new(al, GFP_KERNEL);
146 
147 	if (!skb)
148 		goto err;
149 
150 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
151 				 0, NET_DM_CMD_ALERT);
152 	if (!msg_header) {
153 		nlmsg_free(skb);
154 		skb = NULL;
155 		goto err;
156 	}
157 	nla = nla_reserve(skb, NLA_UNSPEC,
158 			  sizeof(struct net_dm_alert_msg));
159 	if (!nla) {
160 		nlmsg_free(skb);
161 		skb = NULL;
162 		goto err;
163 	}
164 	msg = nla_data(nla);
165 	memset(msg, 0, al);
166 	goto out;
167 
168 err:
169 	mod_timer(&data->send_timer, jiffies + HZ / 10);
170 out:
171 	spin_lock_irqsave(&data->lock, flags);
172 	swap(data->skb, skb);
173 	spin_unlock_irqrestore(&data->lock, flags);
174 
175 	if (skb) {
176 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
177 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
178 
179 		genlmsg_end(skb, genlmsg_data(gnlh));
180 	}
181 
182 	return skb;
183 }
184 
185 static const struct genl_multicast_group dropmon_mcgrps[] = {
186 	{ .name = "events", },
187 };
188 
189 static void send_dm_alert(struct work_struct *work)
190 {
191 	struct sk_buff *skb;
192 	struct per_cpu_dm_data *data;
193 
194 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
195 
196 	skb = reset_per_cpu_data(data);
197 
198 	if (skb)
199 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
200 				  0, GFP_KERNEL);
201 }
202 
203 /*
204  * This is the timer function to delay the sending of an alert
205  * in the event that more drops will arrive during the
206  * hysteresis period.
207  */
208 static void sched_send_work(struct timer_list *t)
209 {
210 	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
211 
212 	schedule_work(&data->dm_alert_work);
213 }
214 
215 static void trace_drop_common(struct sk_buff *skb, void *location)
216 {
217 	struct net_dm_alert_msg *msg;
218 	struct net_dm_drop_point *point;
219 	struct nlmsghdr *nlh;
220 	struct nlattr *nla;
221 	int i;
222 	struct sk_buff *dskb;
223 	struct per_cpu_dm_data *data;
224 	unsigned long flags;
225 
226 	local_irq_save(flags);
227 	data = this_cpu_ptr(&dm_cpu_data);
228 	spin_lock(&data->lock);
229 	dskb = data->skb;
230 
231 	if (!dskb)
232 		goto out;
233 
234 	nlh = (struct nlmsghdr *)dskb->data;
235 	nla = genlmsg_data(nlmsg_data(nlh));
236 	msg = nla_data(nla);
237 	point = msg->points;
238 	for (i = 0; i < msg->entries; i++) {
239 		if (!memcmp(&location, &point->pc, sizeof(void *))) {
240 			point->count++;
241 			goto out;
242 		}
243 		point++;
244 	}
245 	if (msg->entries == dm_hit_limit)
246 		goto out;
247 	/*
248 	 * We need to create a new entry
249 	 */
250 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
251 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
252 	memcpy(point->pc, &location, sizeof(void *));
253 	point->count = 1;
254 	msg->entries++;
255 
256 	if (!timer_pending(&data->send_timer)) {
257 		data->send_timer.expires = jiffies + dm_delay * HZ;
258 		add_timer(&data->send_timer);
259 	}
260 
261 out:
262 	spin_unlock_irqrestore(&data->lock, flags);
263 }
264 
265 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
266 {
267 	trace_drop_common(skb, location);
268 }
269 
270 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
271 				int work, int budget)
272 {
273 	struct dm_hw_stat_delta *new_stat;
274 
275 	/*
276 	 * Don't check napi structures with no associated device
277 	 */
278 	if (!napi->dev)
279 		return;
280 
281 	rcu_read_lock();
282 	list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
283 		/*
284 		 * only add a note to our monitor buffer if:
285 		 * 1) this is the dev we received on
286 		 * 2) its after the last_rx delta
287 		 * 3) our rx_dropped count has gone up
288 		 */
289 		if ((new_stat->dev == napi->dev)  &&
290 		    (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
291 		    (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
292 			trace_drop_common(NULL, NULL);
293 			new_stat->last_drop_val = napi->dev->stats.rx_dropped;
294 			new_stat->last_rx = jiffies;
295 			break;
296 		}
297 	}
298 	rcu_read_unlock();
299 }
300 
301 static struct net_dm_hw_entries *
302 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
303 {
304 	struct net_dm_hw_entries *hw_entries;
305 	unsigned long flags;
306 
307 	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
308 			     GFP_KERNEL);
309 	if (!hw_entries) {
310 		/* If the memory allocation failed, we try to perform another
311 		 * allocation in 1/10 second. Otherwise, the probe function
312 		 * will constantly bail out.
313 		 */
314 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
315 	}
316 
317 	spin_lock_irqsave(&hw_data->lock, flags);
318 	swap(hw_data->hw_entries, hw_entries);
319 	spin_unlock_irqrestore(&hw_data->lock, flags);
320 
321 	return hw_entries;
322 }
323 
324 static int net_dm_hw_entry_put(struct sk_buff *msg,
325 			       const struct net_dm_hw_entry *hw_entry)
326 {
327 	struct nlattr *attr;
328 
329 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
330 	if (!attr)
331 		return -EMSGSIZE;
332 
333 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
334 		goto nla_put_failure;
335 
336 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
337 		goto nla_put_failure;
338 
339 	nla_nest_end(msg, attr);
340 
341 	return 0;
342 
343 nla_put_failure:
344 	nla_nest_cancel(msg, attr);
345 	return -EMSGSIZE;
346 }
347 
348 static int net_dm_hw_entries_put(struct sk_buff *msg,
349 				 const struct net_dm_hw_entries *hw_entries)
350 {
351 	struct nlattr *attr;
352 	int i;
353 
354 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
355 	if (!attr)
356 		return -EMSGSIZE;
357 
358 	for (i = 0; i < hw_entries->num_entries; i++) {
359 		int rc;
360 
361 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
362 		if (rc)
363 			goto nla_put_failure;
364 	}
365 
366 	nla_nest_end(msg, attr);
367 
368 	return 0;
369 
370 nla_put_failure:
371 	nla_nest_cancel(msg, attr);
372 	return -EMSGSIZE;
373 }
374 
375 static int
376 net_dm_hw_summary_report_fill(struct sk_buff *msg,
377 			      const struct net_dm_hw_entries *hw_entries)
378 {
379 	struct net_dm_alert_msg anc_hdr = { 0 };
380 	void *hdr;
381 	int rc;
382 
383 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
384 			  NET_DM_CMD_ALERT);
385 	if (!hdr)
386 		return -EMSGSIZE;
387 
388 	/* We need to put the ancillary header in order not to break user
389 	 * space.
390 	 */
391 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
392 		goto nla_put_failure;
393 
394 	rc = net_dm_hw_entries_put(msg, hw_entries);
395 	if (rc)
396 		goto nla_put_failure;
397 
398 	genlmsg_end(msg, hdr);
399 
400 	return 0;
401 
402 nla_put_failure:
403 	genlmsg_cancel(msg, hdr);
404 	return -EMSGSIZE;
405 }
406 
407 static void net_dm_hw_summary_work(struct work_struct *work)
408 {
409 	struct net_dm_hw_entries *hw_entries;
410 	struct per_cpu_dm_data *hw_data;
411 	struct sk_buff *msg;
412 	int rc;
413 
414 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
415 
416 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
417 	if (!hw_entries)
418 		return;
419 
420 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
421 	if (!msg)
422 		goto out;
423 
424 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
425 	if (rc) {
426 		nlmsg_free(msg);
427 		goto out;
428 	}
429 
430 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
431 
432 out:
433 	kfree(hw_entries);
434 }
435 
436 static void
437 net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
438 			     struct sk_buff *skb,
439 			     const struct devlink_trap_metadata *metadata)
440 {
441 	struct net_dm_hw_entries *hw_entries;
442 	struct net_dm_hw_entry *hw_entry;
443 	struct per_cpu_dm_data *hw_data;
444 	unsigned long flags;
445 	int i;
446 
447 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
448 		return;
449 
450 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
451 	spin_lock_irqsave(&hw_data->lock, flags);
452 	hw_entries = hw_data->hw_entries;
453 
454 	if (!hw_entries)
455 		goto out;
456 
457 	for (i = 0; i < hw_entries->num_entries; i++) {
458 		hw_entry = &hw_entries->entries[i];
459 		if (!strncmp(hw_entry->trap_name, metadata->trap_name,
460 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
461 			hw_entry->count++;
462 			goto out;
463 		}
464 	}
465 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
466 		goto out;
467 
468 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
469 	strlcpy(hw_entry->trap_name, metadata->trap_name,
470 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
471 	hw_entry->count = 1;
472 	hw_entries->num_entries++;
473 
474 	if (!timer_pending(&hw_data->send_timer)) {
475 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
476 		add_timer(&hw_data->send_timer);
477 	}
478 
479 out:
480 	spin_unlock_irqrestore(&hw_data->lock, flags);
481 }
482 
483 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
484 	.kfree_skb_probe	= trace_kfree_skb_hit,
485 	.napi_poll_probe	= trace_napi_poll_hit,
486 	.work_item_func		= send_dm_alert,
487 	.hw_work_item_func	= net_dm_hw_summary_work,
488 	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
489 };
490 
491 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
492 					      struct sk_buff *skb,
493 					      void *location)
494 {
495 	ktime_t tstamp = ktime_get_real();
496 	struct per_cpu_dm_data *data;
497 	struct sk_buff *nskb;
498 	unsigned long flags;
499 
500 	if (!skb_mac_header_was_set(skb))
501 		return;
502 
503 	nskb = skb_clone(skb, GFP_ATOMIC);
504 	if (!nskb)
505 		return;
506 
507 	NET_DM_SKB_CB(nskb)->pc = location;
508 	/* Override the timestamp because we care about the time when the
509 	 * packet was dropped.
510 	 */
511 	nskb->tstamp = tstamp;
512 
513 	data = this_cpu_ptr(&dm_cpu_data);
514 
515 	spin_lock_irqsave(&data->drop_queue.lock, flags);
516 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
517 		__skb_queue_tail(&data->drop_queue, nskb);
518 	else
519 		goto unlock_free;
520 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
521 
522 	schedule_work(&data->dm_alert_work);
523 
524 	return;
525 
526 unlock_free:
527 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
528 	u64_stats_update_begin(&data->stats.syncp);
529 	data->stats.dropped++;
530 	u64_stats_update_end(&data->stats.syncp);
531 	consume_skb(nskb);
532 }
533 
534 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
535 					      struct napi_struct *napi,
536 					      int work, int budget)
537 {
538 }
539 
540 static size_t net_dm_in_port_size(void)
541 {
542 	       /* NET_DM_ATTR_IN_PORT nest */
543 	return nla_total_size(0) +
544 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
545 	       nla_total_size(sizeof(u32)) +
546 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
547 	       nla_total_size(IFNAMSIZ + 1);
548 }
549 
550 #define NET_DM_MAX_SYMBOL_LEN 40
551 
552 static size_t net_dm_packet_report_size(size_t payload_len)
553 {
554 	size_t size;
555 
556 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
557 
558 	return NLMSG_ALIGN(size) +
559 	       /* NET_DM_ATTR_ORIGIN */
560 	       nla_total_size(sizeof(u16)) +
561 	       /* NET_DM_ATTR_PC */
562 	       nla_total_size(sizeof(u64)) +
563 	       /* NET_DM_ATTR_SYMBOL */
564 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
565 	       /* NET_DM_ATTR_IN_PORT */
566 	       net_dm_in_port_size() +
567 	       /* NET_DM_ATTR_TIMESTAMP */
568 	       nla_total_size(sizeof(u64)) +
569 	       /* NET_DM_ATTR_ORIG_LEN */
570 	       nla_total_size(sizeof(u32)) +
571 	       /* NET_DM_ATTR_PROTO */
572 	       nla_total_size(sizeof(u16)) +
573 	       /* NET_DM_ATTR_PAYLOAD */
574 	       nla_total_size(payload_len);
575 }
576 
577 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
578 					    const char *name)
579 {
580 	struct nlattr *attr;
581 
582 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
583 	if (!attr)
584 		return -EMSGSIZE;
585 
586 	if (ifindex &&
587 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
588 		goto nla_put_failure;
589 
590 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
591 		goto nla_put_failure;
592 
593 	nla_nest_end(msg, attr);
594 
595 	return 0;
596 
597 nla_put_failure:
598 	nla_nest_cancel(msg, attr);
599 	return -EMSGSIZE;
600 }
601 
602 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
603 				     size_t payload_len)
604 {
605 	u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
606 	char buf[NET_DM_MAX_SYMBOL_LEN];
607 	struct nlattr *attr;
608 	void *hdr;
609 	int rc;
610 
611 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
612 			  NET_DM_CMD_PACKET_ALERT);
613 	if (!hdr)
614 		return -EMSGSIZE;
615 
616 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
617 		goto nla_put_failure;
618 
619 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
620 		goto nla_put_failure;
621 
622 	snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
623 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
624 		goto nla_put_failure;
625 
626 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
627 	if (rc)
628 		goto nla_put_failure;
629 
630 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
631 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
632 		goto nla_put_failure;
633 
634 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
635 		goto nla_put_failure;
636 
637 	if (!payload_len)
638 		goto out;
639 
640 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
641 		goto nla_put_failure;
642 
643 	attr = skb_put(msg, nla_total_size(payload_len));
644 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
645 	attr->nla_len = nla_attr_size(payload_len);
646 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
647 		goto nla_put_failure;
648 
649 out:
650 	genlmsg_end(msg, hdr);
651 
652 	return 0;
653 
654 nla_put_failure:
655 	genlmsg_cancel(msg, hdr);
656 	return -EMSGSIZE;
657 }
658 
659 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
660 
661 static void net_dm_packet_report(struct sk_buff *skb)
662 {
663 	struct sk_buff *msg;
664 	size_t payload_len;
665 	int rc;
666 
667 	/* Make sure we start copying the packet from the MAC header */
668 	if (skb->data > skb_mac_header(skb))
669 		skb_push(skb, skb->data - skb_mac_header(skb));
670 	else
671 		skb_pull(skb, skb_mac_header(skb) - skb->data);
672 
673 	/* Ensure packet fits inside a single netlink attribute */
674 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
675 	if (net_dm_trunc_len)
676 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
677 
678 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
679 	if (!msg)
680 		goto out;
681 
682 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
683 	if (rc) {
684 		nlmsg_free(msg);
685 		goto out;
686 	}
687 
688 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
689 
690 out:
691 	consume_skb(skb);
692 }
693 
694 static void net_dm_packet_work(struct work_struct *work)
695 {
696 	struct per_cpu_dm_data *data;
697 	struct sk_buff_head list;
698 	struct sk_buff *skb;
699 	unsigned long flags;
700 
701 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
702 
703 	__skb_queue_head_init(&list);
704 
705 	spin_lock_irqsave(&data->drop_queue.lock, flags);
706 	skb_queue_splice_tail_init(&data->drop_queue, &list);
707 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
708 
709 	while ((skb = __skb_dequeue(&list)))
710 		net_dm_packet_report(skb);
711 }
712 
713 static size_t
714 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
715 {
716 	return hw_metadata->fa_cookie ?
717 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
718 }
719 
720 static size_t
721 net_dm_hw_packet_report_size(size_t payload_len,
722 			     const struct devlink_trap_metadata *hw_metadata)
723 {
724 	size_t size;
725 
726 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
727 
728 	return NLMSG_ALIGN(size) +
729 	       /* NET_DM_ATTR_ORIGIN */
730 	       nla_total_size(sizeof(u16)) +
731 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
732 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
733 	       /* NET_DM_ATTR_HW_TRAP_NAME */
734 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
735 	       /* NET_DM_ATTR_IN_PORT */
736 	       net_dm_in_port_size() +
737 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
738 	       net_dm_flow_action_cookie_size(hw_metadata) +
739 	       /* NET_DM_ATTR_TIMESTAMP */
740 	       nla_total_size(sizeof(u64)) +
741 	       /* NET_DM_ATTR_ORIG_LEN */
742 	       nla_total_size(sizeof(u32)) +
743 	       /* NET_DM_ATTR_PROTO */
744 	       nla_total_size(sizeof(u16)) +
745 	       /* NET_DM_ATTR_PAYLOAD */
746 	       nla_total_size(payload_len);
747 }
748 
749 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
750 					struct sk_buff *skb, size_t payload_len)
751 {
752 	struct devlink_trap_metadata *hw_metadata;
753 	struct nlattr *attr;
754 	void *hdr;
755 
756 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
757 
758 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
759 			  NET_DM_CMD_PACKET_ALERT);
760 	if (!hdr)
761 		return -EMSGSIZE;
762 
763 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
764 		goto nla_put_failure;
765 
766 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
767 			   hw_metadata->trap_group_name))
768 		goto nla_put_failure;
769 
770 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
771 			   hw_metadata->trap_name))
772 		goto nla_put_failure;
773 
774 	if (hw_metadata->input_dev) {
775 		struct net_device *dev = hw_metadata->input_dev;
776 		int rc;
777 
778 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
779 						      dev->name);
780 		if (rc)
781 			goto nla_put_failure;
782 	}
783 
784 	if (hw_metadata->fa_cookie &&
785 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
786 		    hw_metadata->fa_cookie->cookie_len,
787 		    hw_metadata->fa_cookie->cookie))
788 		goto nla_put_failure;
789 
790 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
791 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
792 		goto nla_put_failure;
793 
794 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
795 		goto nla_put_failure;
796 
797 	if (!payload_len)
798 		goto out;
799 
800 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
801 		goto nla_put_failure;
802 
803 	attr = skb_put(msg, nla_total_size(payload_len));
804 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
805 	attr->nla_len = nla_attr_size(payload_len);
806 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
807 		goto nla_put_failure;
808 
809 out:
810 	genlmsg_end(msg, hdr);
811 
812 	return 0;
813 
814 nla_put_failure:
815 	genlmsg_cancel(msg, hdr);
816 	return -EMSGSIZE;
817 }
818 
819 static struct devlink_trap_metadata *
820 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
821 {
822 	const struct flow_action_cookie *fa_cookie;
823 	struct devlink_trap_metadata *hw_metadata;
824 	const char *trap_group_name;
825 	const char *trap_name;
826 
827 	hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
828 	if (!hw_metadata)
829 		return NULL;
830 
831 	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
832 	if (!trap_group_name)
833 		goto free_hw_metadata;
834 	hw_metadata->trap_group_name = trap_group_name;
835 
836 	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
837 	if (!trap_name)
838 		goto free_trap_group;
839 	hw_metadata->trap_name = trap_name;
840 
841 	if (metadata->fa_cookie) {
842 		size_t cookie_size = sizeof(*fa_cookie) +
843 				     metadata->fa_cookie->cookie_len;
844 
845 		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
846 				    GFP_ATOMIC);
847 		if (!fa_cookie)
848 			goto free_trap_name;
849 		hw_metadata->fa_cookie = fa_cookie;
850 	}
851 
852 	hw_metadata->input_dev = metadata->input_dev;
853 	if (hw_metadata->input_dev)
854 		dev_hold(hw_metadata->input_dev);
855 
856 	return hw_metadata;
857 
858 free_trap_name:
859 	kfree(trap_name);
860 free_trap_group:
861 	kfree(trap_group_name);
862 free_hw_metadata:
863 	kfree(hw_metadata);
864 	return NULL;
865 }
866 
867 static void
868 net_dm_hw_metadata_free(const struct devlink_trap_metadata *hw_metadata)
869 {
870 	if (hw_metadata->input_dev)
871 		dev_put(hw_metadata->input_dev);
872 	kfree(hw_metadata->fa_cookie);
873 	kfree(hw_metadata->trap_name);
874 	kfree(hw_metadata->trap_group_name);
875 	kfree(hw_metadata);
876 }
877 
878 static void net_dm_hw_packet_report(struct sk_buff *skb)
879 {
880 	struct devlink_trap_metadata *hw_metadata;
881 	struct sk_buff *msg;
882 	size_t payload_len;
883 	int rc;
884 
885 	if (skb->data > skb_mac_header(skb))
886 		skb_push(skb, skb->data - skb_mac_header(skb));
887 	else
888 		skb_pull(skb, skb_mac_header(skb) - skb->data);
889 
890 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
891 	if (net_dm_trunc_len)
892 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
893 
894 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
895 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
896 			GFP_KERNEL);
897 	if (!msg)
898 		goto out;
899 
900 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
901 	if (rc) {
902 		nlmsg_free(msg);
903 		goto out;
904 	}
905 
906 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
907 
908 out:
909 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
910 	consume_skb(skb);
911 }
912 
913 static void net_dm_hw_packet_work(struct work_struct *work)
914 {
915 	struct per_cpu_dm_data *hw_data;
916 	struct sk_buff_head list;
917 	struct sk_buff *skb;
918 	unsigned long flags;
919 
920 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
921 
922 	__skb_queue_head_init(&list);
923 
924 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
925 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
926 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
927 
928 	while ((skb = __skb_dequeue(&list)))
929 		net_dm_hw_packet_report(skb);
930 }
931 
932 static void
933 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
934 			    struct sk_buff *skb,
935 			    const struct devlink_trap_metadata *metadata)
936 {
937 	struct devlink_trap_metadata *n_hw_metadata;
938 	ktime_t tstamp = ktime_get_real();
939 	struct per_cpu_dm_data *hw_data;
940 	struct sk_buff *nskb;
941 	unsigned long flags;
942 
943 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
944 		return;
945 
946 	if (!skb_mac_header_was_set(skb))
947 		return;
948 
949 	nskb = skb_clone(skb, GFP_ATOMIC);
950 	if (!nskb)
951 		return;
952 
953 	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
954 	if (!n_hw_metadata)
955 		goto free;
956 
957 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
958 	nskb->tstamp = tstamp;
959 
960 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
961 
962 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
963 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
964 		__skb_queue_tail(&hw_data->drop_queue, nskb);
965 	else
966 		goto unlock_free;
967 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
968 
969 	schedule_work(&hw_data->dm_alert_work);
970 
971 	return;
972 
973 unlock_free:
974 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
975 	u64_stats_update_begin(&hw_data->stats.syncp);
976 	hw_data->stats.dropped++;
977 	u64_stats_update_end(&hw_data->stats.syncp);
978 	net_dm_hw_metadata_free(n_hw_metadata);
979 free:
980 	consume_skb(nskb);
981 }
982 
983 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
984 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
985 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
986 	.work_item_func		= net_dm_packet_work,
987 	.hw_work_item_func	= net_dm_hw_packet_work,
988 	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
989 };
990 
991 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
992 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
993 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
994 };
995 
996 #if IS_ENABLED(CONFIG_NET_DEVLINK)
997 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
998 {
999 	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1000 }
1001 
1002 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1003 {
1004 	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1005 	tracepoint_synchronize_unregister();
1006 }
1007 #else
1008 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1009 {
1010 	return -EOPNOTSUPP;
1011 }
1012 
1013 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1014 {
1015 }
1016 #endif
1017 
1018 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1019 {
1020 	const struct net_dm_alert_ops *ops;
1021 	int cpu, rc;
1022 
1023 	if (monitor_hw) {
1024 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1025 		return -EAGAIN;
1026 	}
1027 
1028 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1029 
1030 	if (!try_module_get(THIS_MODULE)) {
1031 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1032 		return -ENODEV;
1033 	}
1034 
1035 	for_each_possible_cpu(cpu) {
1036 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1037 		struct net_dm_hw_entries *hw_entries;
1038 
1039 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1040 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1041 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1042 		kfree(hw_entries);
1043 	}
1044 
1045 	rc = net_dm_hw_probe_register(ops);
1046 	if (rc) {
1047 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1048 		goto err_module_put;
1049 	}
1050 
1051 	monitor_hw = true;
1052 
1053 	return 0;
1054 
1055 err_module_put:
1056 	module_put(THIS_MODULE);
1057 	return rc;
1058 }
1059 
1060 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1061 {
1062 	const struct net_dm_alert_ops *ops;
1063 	int cpu;
1064 
1065 	if (!monitor_hw) {
1066 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1067 		return;
1068 	}
1069 
1070 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1071 
1072 	monitor_hw = false;
1073 
1074 	net_dm_hw_probe_unregister(ops);
1075 
1076 	for_each_possible_cpu(cpu) {
1077 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1078 		struct sk_buff *skb;
1079 
1080 		del_timer_sync(&hw_data->send_timer);
1081 		cancel_work_sync(&hw_data->dm_alert_work);
1082 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1083 			struct devlink_trap_metadata *hw_metadata;
1084 
1085 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1086 			net_dm_hw_metadata_free(hw_metadata);
1087 			consume_skb(skb);
1088 		}
1089 	}
1090 
1091 	module_put(THIS_MODULE);
1092 }
1093 
1094 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1095 {
1096 	const struct net_dm_alert_ops *ops;
1097 	int cpu, rc;
1098 
1099 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1100 
1101 	if (!try_module_get(THIS_MODULE)) {
1102 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1103 		return -ENODEV;
1104 	}
1105 
1106 	for_each_possible_cpu(cpu) {
1107 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1108 		struct sk_buff *skb;
1109 
1110 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1111 		timer_setup(&data->send_timer, sched_send_work, 0);
1112 		/* Allocate a new per-CPU skb for the summary alert message and
1113 		 * free the old one which might contain stale data from
1114 		 * previous tracing.
1115 		 */
1116 		skb = reset_per_cpu_data(data);
1117 		consume_skb(skb);
1118 	}
1119 
1120 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1121 	if (rc) {
1122 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1123 		goto err_module_put;
1124 	}
1125 
1126 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1127 	if (rc) {
1128 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1129 		goto err_unregister_trace;
1130 	}
1131 
1132 	return 0;
1133 
1134 err_unregister_trace:
1135 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1136 err_module_put:
1137 	module_put(THIS_MODULE);
1138 	return rc;
1139 }
1140 
1141 static void net_dm_trace_off_set(void)
1142 {
1143 	struct dm_hw_stat_delta *new_stat, *temp;
1144 	const struct net_dm_alert_ops *ops;
1145 	int cpu;
1146 
1147 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1148 
1149 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1150 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1151 
1152 	tracepoint_synchronize_unregister();
1153 
1154 	/* Make sure we do not send notifications to user space after request
1155 	 * to stop tracing returns.
1156 	 */
1157 	for_each_possible_cpu(cpu) {
1158 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1159 		struct sk_buff *skb;
1160 
1161 		del_timer_sync(&data->send_timer);
1162 		cancel_work_sync(&data->dm_alert_work);
1163 		while ((skb = __skb_dequeue(&data->drop_queue)))
1164 			consume_skb(skb);
1165 	}
1166 
1167 	list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1168 		if (new_stat->dev == NULL) {
1169 			list_del_rcu(&new_stat->list);
1170 			kfree_rcu(new_stat, rcu);
1171 		}
1172 	}
1173 
1174 	module_put(THIS_MODULE);
1175 }
1176 
1177 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1178 {
1179 	int rc = 0;
1180 
1181 	if (state == trace_state) {
1182 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1183 		return -EAGAIN;
1184 	}
1185 
1186 	switch (state) {
1187 	case TRACE_ON:
1188 		rc = net_dm_trace_on_set(extack);
1189 		break;
1190 	case TRACE_OFF:
1191 		net_dm_trace_off_set();
1192 		break;
1193 	default:
1194 		rc = 1;
1195 		break;
1196 	}
1197 
1198 	if (!rc)
1199 		trace_state = state;
1200 	else
1201 		rc = -EINPROGRESS;
1202 
1203 	return rc;
1204 }
1205 
1206 static bool net_dm_is_monitoring(void)
1207 {
1208 	return trace_state == TRACE_ON || monitor_hw;
1209 }
1210 
1211 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1212 					   enum net_dm_alert_mode *p_alert_mode)
1213 {
1214 	u8 val;
1215 
1216 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1217 
1218 	switch (val) {
1219 	case NET_DM_ALERT_MODE_SUMMARY:
1220 	case NET_DM_ALERT_MODE_PACKET:
1221 		*p_alert_mode = val;
1222 		break;
1223 	default:
1224 		return -EINVAL;
1225 	}
1226 
1227 	return 0;
1228 }
1229 
1230 static int net_dm_alert_mode_set(struct genl_info *info)
1231 {
1232 	struct netlink_ext_ack *extack = info->extack;
1233 	enum net_dm_alert_mode alert_mode;
1234 	int rc;
1235 
1236 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1237 		return 0;
1238 
1239 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1240 	if (rc) {
1241 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1242 		return -EINVAL;
1243 	}
1244 
1245 	net_dm_alert_mode = alert_mode;
1246 
1247 	return 0;
1248 }
1249 
1250 static void net_dm_trunc_len_set(struct genl_info *info)
1251 {
1252 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1253 		return;
1254 
1255 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1256 }
1257 
1258 static void net_dm_queue_len_set(struct genl_info *info)
1259 {
1260 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1261 		return;
1262 
1263 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1264 }
1265 
1266 static int net_dm_cmd_config(struct sk_buff *skb,
1267 			struct genl_info *info)
1268 {
1269 	struct netlink_ext_ack *extack = info->extack;
1270 	int rc;
1271 
1272 	if (net_dm_is_monitoring()) {
1273 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1274 		return -EBUSY;
1275 	}
1276 
1277 	rc = net_dm_alert_mode_set(info);
1278 	if (rc)
1279 		return rc;
1280 
1281 	net_dm_trunc_len_set(info);
1282 
1283 	net_dm_queue_len_set(info);
1284 
1285 	return 0;
1286 }
1287 
1288 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1289 				struct netlink_ext_ack *extack)
1290 {
1291 	bool sw_set = false;
1292 	int rc;
1293 
1294 	if (set_sw) {
1295 		rc = set_all_monitor_traces(TRACE_ON, extack);
1296 		if (rc)
1297 			return rc;
1298 		sw_set = true;
1299 	}
1300 
1301 	if (set_hw) {
1302 		rc = net_dm_hw_monitor_start(extack);
1303 		if (rc)
1304 			goto err_monitor_hw;
1305 	}
1306 
1307 	return 0;
1308 
1309 err_monitor_hw:
1310 	if (sw_set)
1311 		set_all_monitor_traces(TRACE_OFF, extack);
1312 	return rc;
1313 }
1314 
1315 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1316 				struct netlink_ext_ack *extack)
1317 {
1318 	if (set_hw)
1319 		net_dm_hw_monitor_stop(extack);
1320 	if (set_sw)
1321 		set_all_monitor_traces(TRACE_OFF, extack);
1322 }
1323 
1324 static int net_dm_cmd_trace(struct sk_buff *skb,
1325 			struct genl_info *info)
1326 {
1327 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1328 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1329 	struct netlink_ext_ack *extack = info->extack;
1330 
1331 	/* To maintain backward compatibility, we start / stop monitoring of
1332 	 * software drops if no flag is specified.
1333 	 */
1334 	if (!set_sw && !set_hw)
1335 		set_sw = true;
1336 
1337 	switch (info->genlhdr->cmd) {
1338 	case NET_DM_CMD_START:
1339 		return net_dm_monitor_start(set_sw, set_hw, extack);
1340 	case NET_DM_CMD_STOP:
1341 		net_dm_monitor_stop(set_sw, set_hw, extack);
1342 		return 0;
1343 	}
1344 
1345 	return -EOPNOTSUPP;
1346 }
1347 
1348 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1349 {
1350 	void *hdr;
1351 
1352 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1353 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1354 	if (!hdr)
1355 		return -EMSGSIZE;
1356 
1357 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1358 		goto nla_put_failure;
1359 
1360 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1361 		goto nla_put_failure;
1362 
1363 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1364 		goto nla_put_failure;
1365 
1366 	genlmsg_end(msg, hdr);
1367 
1368 	return 0;
1369 
1370 nla_put_failure:
1371 	genlmsg_cancel(msg, hdr);
1372 	return -EMSGSIZE;
1373 }
1374 
1375 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1376 {
1377 	struct sk_buff *msg;
1378 	int rc;
1379 
1380 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1381 	if (!msg)
1382 		return -ENOMEM;
1383 
1384 	rc = net_dm_config_fill(msg, info);
1385 	if (rc)
1386 		goto free_msg;
1387 
1388 	return genlmsg_reply(msg, info);
1389 
1390 free_msg:
1391 	nlmsg_free(msg);
1392 	return rc;
1393 }
1394 
1395 static void net_dm_stats_read(struct net_dm_stats *stats)
1396 {
1397 	int cpu;
1398 
1399 	memset(stats, 0, sizeof(*stats));
1400 	for_each_possible_cpu(cpu) {
1401 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1402 		struct net_dm_stats *cpu_stats = &data->stats;
1403 		unsigned int start;
1404 		u64 dropped;
1405 
1406 		do {
1407 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1408 			dropped = cpu_stats->dropped;
1409 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1410 
1411 		stats->dropped += dropped;
1412 	}
1413 }
1414 
1415 static int net_dm_stats_put(struct sk_buff *msg)
1416 {
1417 	struct net_dm_stats stats;
1418 	struct nlattr *attr;
1419 
1420 	net_dm_stats_read(&stats);
1421 
1422 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1423 	if (!attr)
1424 		return -EMSGSIZE;
1425 
1426 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1427 			      stats.dropped, NET_DM_ATTR_PAD))
1428 		goto nla_put_failure;
1429 
1430 	nla_nest_end(msg, attr);
1431 
1432 	return 0;
1433 
1434 nla_put_failure:
1435 	nla_nest_cancel(msg, attr);
1436 	return -EMSGSIZE;
1437 }
1438 
1439 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1440 {
1441 	int cpu;
1442 
1443 	memset(stats, 0, sizeof(*stats));
1444 	for_each_possible_cpu(cpu) {
1445 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1446 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1447 		unsigned int start;
1448 		u64 dropped;
1449 
1450 		do {
1451 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1452 			dropped = cpu_stats->dropped;
1453 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1454 
1455 		stats->dropped += dropped;
1456 	}
1457 }
1458 
1459 static int net_dm_hw_stats_put(struct sk_buff *msg)
1460 {
1461 	struct net_dm_stats stats;
1462 	struct nlattr *attr;
1463 
1464 	net_dm_hw_stats_read(&stats);
1465 
1466 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1467 	if (!attr)
1468 		return -EMSGSIZE;
1469 
1470 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1471 			      stats.dropped, NET_DM_ATTR_PAD))
1472 		goto nla_put_failure;
1473 
1474 	nla_nest_end(msg, attr);
1475 
1476 	return 0;
1477 
1478 nla_put_failure:
1479 	nla_nest_cancel(msg, attr);
1480 	return -EMSGSIZE;
1481 }
1482 
1483 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1484 {
1485 	void *hdr;
1486 	int rc;
1487 
1488 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1489 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1490 	if (!hdr)
1491 		return -EMSGSIZE;
1492 
1493 	rc = net_dm_stats_put(msg);
1494 	if (rc)
1495 		goto nla_put_failure;
1496 
1497 	rc = net_dm_hw_stats_put(msg);
1498 	if (rc)
1499 		goto nla_put_failure;
1500 
1501 	genlmsg_end(msg, hdr);
1502 
1503 	return 0;
1504 
1505 nla_put_failure:
1506 	genlmsg_cancel(msg, hdr);
1507 	return -EMSGSIZE;
1508 }
1509 
1510 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1511 {
1512 	struct sk_buff *msg;
1513 	int rc;
1514 
1515 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1516 	if (!msg)
1517 		return -ENOMEM;
1518 
1519 	rc = net_dm_stats_fill(msg, info);
1520 	if (rc)
1521 		goto free_msg;
1522 
1523 	return genlmsg_reply(msg, info);
1524 
1525 free_msg:
1526 	nlmsg_free(msg);
1527 	return rc;
1528 }
1529 
1530 static int dropmon_net_event(struct notifier_block *ev_block,
1531 			     unsigned long event, void *ptr)
1532 {
1533 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1534 	struct dm_hw_stat_delta *new_stat = NULL;
1535 	struct dm_hw_stat_delta *tmp;
1536 
1537 	switch (event) {
1538 	case NETDEV_REGISTER:
1539 		new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
1540 
1541 		if (!new_stat)
1542 			goto out;
1543 
1544 		new_stat->dev = dev;
1545 		new_stat->last_rx = jiffies;
1546 		mutex_lock(&net_dm_mutex);
1547 		list_add_rcu(&new_stat->list, &hw_stats_list);
1548 		mutex_unlock(&net_dm_mutex);
1549 		break;
1550 	case NETDEV_UNREGISTER:
1551 		mutex_lock(&net_dm_mutex);
1552 		list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1553 			if (new_stat->dev == dev) {
1554 				new_stat->dev = NULL;
1555 				if (trace_state == TRACE_OFF) {
1556 					list_del_rcu(&new_stat->list);
1557 					kfree_rcu(new_stat, rcu);
1558 					break;
1559 				}
1560 			}
1561 		}
1562 		mutex_unlock(&net_dm_mutex);
1563 		break;
1564 	}
1565 out:
1566 	return NOTIFY_DONE;
1567 }
1568 
1569 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1570 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1571 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1572 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1573 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1574 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1575 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1576 };
1577 
1578 static const struct genl_small_ops dropmon_ops[] = {
1579 	{
1580 		.cmd = NET_DM_CMD_CONFIG,
1581 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1582 		.doit = net_dm_cmd_config,
1583 		.flags = GENL_ADMIN_PERM,
1584 	},
1585 	{
1586 		.cmd = NET_DM_CMD_START,
1587 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1588 		.doit = net_dm_cmd_trace,
1589 	},
1590 	{
1591 		.cmd = NET_DM_CMD_STOP,
1592 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1593 		.doit = net_dm_cmd_trace,
1594 	},
1595 	{
1596 		.cmd = NET_DM_CMD_CONFIG_GET,
1597 		.doit = net_dm_cmd_config_get,
1598 	},
1599 	{
1600 		.cmd = NET_DM_CMD_STATS_GET,
1601 		.doit = net_dm_cmd_stats_get,
1602 	},
1603 };
1604 
1605 static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1606 			      struct sk_buff *skb, struct genl_info *info)
1607 {
1608 	mutex_lock(&net_dm_mutex);
1609 
1610 	return 0;
1611 }
1612 
1613 static void net_dm_nl_post_doit(const struct genl_ops *ops,
1614 				struct sk_buff *skb, struct genl_info *info)
1615 {
1616 	mutex_unlock(&net_dm_mutex);
1617 }
1618 
1619 static struct genl_family net_drop_monitor_family __ro_after_init = {
1620 	.hdrsize        = 0,
1621 	.name           = "NET_DM",
1622 	.version        = 2,
1623 	.maxattr	= NET_DM_ATTR_MAX,
1624 	.policy		= net_dm_nl_policy,
1625 	.pre_doit	= net_dm_nl_pre_doit,
1626 	.post_doit	= net_dm_nl_post_doit,
1627 	.module		= THIS_MODULE,
1628 	.small_ops	= dropmon_ops,
1629 	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1630 	.mcgrps		= dropmon_mcgrps,
1631 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1632 };
1633 
1634 static struct notifier_block dropmon_net_notifier = {
1635 	.notifier_call = dropmon_net_event
1636 };
1637 
1638 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1639 {
1640 	spin_lock_init(&data->lock);
1641 	skb_queue_head_init(&data->drop_queue);
1642 	u64_stats_init(&data->stats.syncp);
1643 }
1644 
1645 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1646 {
1647 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1648 }
1649 
1650 static void net_dm_cpu_data_init(int cpu)
1651 {
1652 	struct per_cpu_dm_data *data;
1653 
1654 	data = &per_cpu(dm_cpu_data, cpu);
1655 	__net_dm_cpu_data_init(data);
1656 }
1657 
1658 static void net_dm_cpu_data_fini(int cpu)
1659 {
1660 	struct per_cpu_dm_data *data;
1661 
1662 	data = &per_cpu(dm_cpu_data, cpu);
1663 	/* At this point, we should have exclusive access
1664 	 * to this struct and can free the skb inside it.
1665 	 */
1666 	consume_skb(data->skb);
1667 	__net_dm_cpu_data_fini(data);
1668 }
1669 
1670 static void net_dm_hw_cpu_data_init(int cpu)
1671 {
1672 	struct per_cpu_dm_data *hw_data;
1673 
1674 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1675 	__net_dm_cpu_data_init(hw_data);
1676 }
1677 
1678 static void net_dm_hw_cpu_data_fini(int cpu)
1679 {
1680 	struct per_cpu_dm_data *hw_data;
1681 
1682 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1683 	kfree(hw_data->hw_entries);
1684 	__net_dm_cpu_data_fini(hw_data);
1685 }
1686 
1687 static int __init init_net_drop_monitor(void)
1688 {
1689 	int cpu, rc;
1690 
1691 	pr_info("Initializing network drop monitor service\n");
1692 
1693 	if (sizeof(void *) > 8) {
1694 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1695 		return -ENOSPC;
1696 	}
1697 
1698 	rc = genl_register_family(&net_drop_monitor_family);
1699 	if (rc) {
1700 		pr_err("Could not create drop monitor netlink family\n");
1701 		return rc;
1702 	}
1703 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1704 
1705 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1706 	if (rc < 0) {
1707 		pr_crit("Failed to register netdevice notifier\n");
1708 		goto out_unreg;
1709 	}
1710 
1711 	rc = 0;
1712 
1713 	for_each_possible_cpu(cpu) {
1714 		net_dm_cpu_data_init(cpu);
1715 		net_dm_hw_cpu_data_init(cpu);
1716 	}
1717 
1718 	goto out;
1719 
1720 out_unreg:
1721 	genl_unregister_family(&net_drop_monitor_family);
1722 out:
1723 	return rc;
1724 }
1725 
1726 static void exit_net_drop_monitor(void)
1727 {
1728 	int cpu;
1729 
1730 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1731 
1732 	/*
1733 	 * Because of the module_get/put we do in the trace state change path
1734 	 * we are guarnateed not to have any current users when we get here
1735 	 */
1736 
1737 	for_each_possible_cpu(cpu) {
1738 		net_dm_hw_cpu_data_fini(cpu);
1739 		net_dm_cpu_data_fini(cpu);
1740 	}
1741 
1742 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1743 }
1744 
1745 module_init(init_net_drop_monitor);
1746 module_exit(exit_net_drop_monitor);
1747 
1748 MODULE_LICENSE("GPL v2");
1749 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1750 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1751 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1752