xref: /openbmc/linux/net/core/drop_monitor.c (revision adb19164)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Monitoring code for network dropped packet alerts
4  *
5  * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/string.h>
13 #include <linux/if_arp.h>
14 #include <linux/inetdevice.h>
15 #include <linux/inet.h>
16 #include <linux/interrupt.h>
17 #include <linux/netpoll.h>
18 #include <linux/sched.h>
19 #include <linux/delay.h>
20 #include <linux/types.h>
21 #include <linux/workqueue.h>
22 #include <linux/netlink.h>
23 #include <linux/net_dropmon.h>
24 #include <linux/bitfield.h>
25 #include <linux/percpu.h>
26 #include <linux/timer.h>
27 #include <linux/bitops.h>
28 #include <linux/slab.h>
29 #include <linux/module.h>
30 #include <net/genetlink.h>
31 #include <net/netevent.h>
32 #include <net/flow_offload.h>
33 #include <net/dropreason.h>
34 #include <net/devlink.h>
35 
36 #include <trace/events/skb.h>
37 #include <trace/events/napi.h>
38 #include <trace/events/devlink.h>
39 
40 #include <asm/unaligned.h>
41 
42 #define TRACE_ON 1
43 #define TRACE_OFF 0
44 
45 /*
46  * Globals, our netlink socket pointer
47  * and the work handle that will send up
48  * netlink alerts
49  */
50 static int trace_state = TRACE_OFF;
51 static bool monitor_hw;
52 
53 /* net_dm_mutex
54  *
55  * An overall lock guarding every operation coming from userspace.
56  */
57 static DEFINE_MUTEX(net_dm_mutex);
58 
59 struct net_dm_stats {
60 	u64_stats_t dropped;
61 	struct u64_stats_sync syncp;
62 };
63 
64 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
65 
66 struct net_dm_hw_entry {
67 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
68 	u32 count;
69 };
70 
71 struct net_dm_hw_entries {
72 	u32 num_entries;
73 	struct net_dm_hw_entry entries[];
74 };
75 
76 struct per_cpu_dm_data {
77 	spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
78 					 * 'send_timer'
79 					 */
80 	union {
81 		struct sk_buff			*skb;
82 		struct net_dm_hw_entries	*hw_entries;
83 	};
84 	struct sk_buff_head	drop_queue;
85 	struct work_struct	dm_alert_work;
86 	struct timer_list	send_timer;
87 	struct net_dm_stats	stats;
88 };
89 
90 struct dm_hw_stat_delta {
91 	unsigned long last_rx;
92 	unsigned long last_drop_val;
93 	struct rcu_head rcu;
94 };
95 
96 static struct genl_family net_drop_monitor_family;
97 
98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100 
101 static int dm_hit_limit = 64;
102 static int dm_delay = 1;
103 static unsigned long dm_hw_check_delta = 2*HZ;
104 
105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106 static u32 net_dm_trunc_len;
107 static u32 net_dm_queue_len = 1000;
108 
109 struct net_dm_alert_ops {
110 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
111 				void *location,
112 				enum skb_drop_reason reason);
113 	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_trap_probe)(void *ignore, const struct devlink *devlink,
118 			      struct sk_buff *skb,
119 			      const struct devlink_trap_metadata *metadata);
120 };
121 
122 struct net_dm_skb_cb {
123 	union {
124 		struct devlink_trap_metadata *hw_metadata;
125 		void *pc;
126 	};
127 	enum skb_drop_reason reason;
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", .cap_sys_admin = 1 },
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,
266 				void *location,
267 				enum skb_drop_reason reason)
268 {
269 	trace_drop_common(skb, location);
270 }
271 
272 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
273 				int work, int budget)
274 {
275 	struct net_device *dev = napi->dev;
276 	struct dm_hw_stat_delta *stat;
277 	/*
278 	 * Don't check napi structures with no associated device
279 	 */
280 	if (!dev)
281 		return;
282 
283 	rcu_read_lock();
284 	stat = rcu_dereference(dev->dm_private);
285 	if (stat) {
286 		/*
287 		 * only add a note to our monitor buffer if:
288 		 * 1) its after the last_rx delta
289 		 * 2) our rx_dropped count has gone up
290 		 */
291 		if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
292 		    (dev->stats.rx_dropped != stat->last_drop_val)) {
293 			trace_drop_common(NULL, NULL);
294 			stat->last_drop_val = dev->stats.rx_dropped;
295 			stat->last_rx = jiffies;
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 	strscpy(hw_entry->trap_name, metadata->trap_name,
470 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
471 	hw_entry->count = 1;
472 	hw_entries->num_entries++;
473 
474 	if (!timer_pending(&hw_data->send_timer)) {
475 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
476 		add_timer(&hw_data->send_timer);
477 	}
478 
479 out:
480 	spin_unlock_irqrestore(&hw_data->lock, flags);
481 }
482 
483 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
484 	.kfree_skb_probe	= trace_kfree_skb_hit,
485 	.napi_poll_probe	= trace_napi_poll_hit,
486 	.work_item_func		= send_dm_alert,
487 	.hw_work_item_func	= net_dm_hw_summary_work,
488 	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
489 };
490 
491 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
492 					      struct sk_buff *skb,
493 					      void *location,
494 					      enum skb_drop_reason reason)
495 {
496 	ktime_t tstamp = ktime_get_real();
497 	struct per_cpu_dm_data *data;
498 	struct net_dm_skb_cb *cb;
499 	struct sk_buff *nskb;
500 	unsigned long flags;
501 
502 	if (!skb_mac_header_was_set(skb))
503 		return;
504 
505 	nskb = skb_clone(skb, GFP_ATOMIC);
506 	if (!nskb)
507 		return;
508 
509 	cb = NET_DM_SKB_CB(nskb);
510 	cb->reason = reason;
511 	cb->pc = location;
512 	/* Override the timestamp because we care about the time when the
513 	 * packet was dropped.
514 	 */
515 	nskb->tstamp = tstamp;
516 
517 	data = this_cpu_ptr(&dm_cpu_data);
518 
519 	spin_lock_irqsave(&data->drop_queue.lock, flags);
520 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
521 		__skb_queue_tail(&data->drop_queue, nskb);
522 	else
523 		goto unlock_free;
524 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
525 
526 	schedule_work(&data->dm_alert_work);
527 
528 	return;
529 
530 unlock_free:
531 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
532 	u64_stats_update_begin(&data->stats.syncp);
533 	u64_stats_inc(&data->stats.dropped);
534 	u64_stats_update_end(&data->stats.syncp);
535 	consume_skb(nskb);
536 }
537 
538 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
539 					      struct napi_struct *napi,
540 					      int work, int budget)
541 {
542 }
543 
544 static size_t net_dm_in_port_size(void)
545 {
546 	       /* NET_DM_ATTR_IN_PORT nest */
547 	return nla_total_size(0) +
548 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
549 	       nla_total_size(sizeof(u32)) +
550 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
551 	       nla_total_size(IFNAMSIZ + 1);
552 }
553 
554 #define NET_DM_MAX_SYMBOL_LEN 40
555 #define NET_DM_MAX_REASON_LEN 50
556 
557 static size_t net_dm_packet_report_size(size_t payload_len)
558 {
559 	size_t size;
560 
561 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
562 
563 	return NLMSG_ALIGN(size) +
564 	       /* NET_DM_ATTR_ORIGIN */
565 	       nla_total_size(sizeof(u16)) +
566 	       /* NET_DM_ATTR_PC */
567 	       nla_total_size(sizeof(u64)) +
568 	       /* NET_DM_ATTR_SYMBOL */
569 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
570 	       /* NET_DM_ATTR_IN_PORT */
571 	       net_dm_in_port_size() +
572 	       /* NET_DM_ATTR_TIMESTAMP */
573 	       nla_total_size(sizeof(u64)) +
574 	       /* NET_DM_ATTR_ORIG_LEN */
575 	       nla_total_size(sizeof(u32)) +
576 	       /* NET_DM_ATTR_PROTO */
577 	       nla_total_size(sizeof(u16)) +
578 	       /* NET_DM_ATTR_REASON */
579 	       nla_total_size(NET_DM_MAX_REASON_LEN + 1) +
580 	       /* NET_DM_ATTR_PAYLOAD */
581 	       nla_total_size(payload_len);
582 }
583 
584 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
585 					    const char *name)
586 {
587 	struct nlattr *attr;
588 
589 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
590 	if (!attr)
591 		return -EMSGSIZE;
592 
593 	if (ifindex &&
594 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
595 		goto nla_put_failure;
596 
597 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
598 		goto nla_put_failure;
599 
600 	nla_nest_end(msg, attr);
601 
602 	return 0;
603 
604 nla_put_failure:
605 	nla_nest_cancel(msg, attr);
606 	return -EMSGSIZE;
607 }
608 
609 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
610 				     size_t payload_len)
611 {
612 	struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb);
613 	const struct drop_reason_list *list = NULL;
614 	unsigned int subsys, subsys_reason;
615 	char buf[NET_DM_MAX_SYMBOL_LEN];
616 	struct nlattr *attr;
617 	void *hdr;
618 	int rc;
619 
620 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
621 			  NET_DM_CMD_PACKET_ALERT);
622 	if (!hdr)
623 		return -EMSGSIZE;
624 
625 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
626 		goto nla_put_failure;
627 
628 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
629 			      NET_DM_ATTR_PAD))
630 		goto nla_put_failure;
631 
632 	rcu_read_lock();
633 	subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK);
634 	if (subsys < SKB_DROP_REASON_SUBSYS_NUM)
635 		list = rcu_dereference(drop_reasons_by_subsys[subsys]);
636 	subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK;
637 	if (!list ||
638 	    subsys_reason >= list->n_reasons ||
639 	    !list->reasons[subsys_reason] ||
640 	    strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) {
641 		list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]);
642 		subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED;
643 	}
644 	if (nla_put_string(msg, NET_DM_ATTR_REASON,
645 			   list->reasons[subsys_reason])) {
646 		rcu_read_unlock();
647 		goto nla_put_failure;
648 	}
649 	rcu_read_unlock();
650 
651 	snprintf(buf, sizeof(buf), "%pS", cb->pc);
652 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
653 		goto nla_put_failure;
654 
655 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
656 	if (rc)
657 		goto nla_put_failure;
658 
659 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
660 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
661 		goto nla_put_failure;
662 
663 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
664 		goto nla_put_failure;
665 
666 	if (!payload_len)
667 		goto out;
668 
669 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
670 		goto nla_put_failure;
671 
672 	attr = skb_put(msg, nla_total_size(payload_len));
673 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
674 	attr->nla_len = nla_attr_size(payload_len);
675 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
676 		goto nla_put_failure;
677 
678 out:
679 	genlmsg_end(msg, hdr);
680 
681 	return 0;
682 
683 nla_put_failure:
684 	genlmsg_cancel(msg, hdr);
685 	return -EMSGSIZE;
686 }
687 
688 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
689 
690 static void net_dm_packet_report(struct sk_buff *skb)
691 {
692 	struct sk_buff *msg;
693 	size_t payload_len;
694 	int rc;
695 
696 	/* Make sure we start copying the packet from the MAC header */
697 	if (skb->data > skb_mac_header(skb))
698 		skb_push(skb, skb->data - skb_mac_header(skb));
699 	else
700 		skb_pull(skb, skb_mac_header(skb) - skb->data);
701 
702 	/* Ensure packet fits inside a single netlink attribute */
703 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
704 	if (net_dm_trunc_len)
705 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
706 
707 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
708 	if (!msg)
709 		goto out;
710 
711 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
712 	if (rc) {
713 		nlmsg_free(msg);
714 		goto out;
715 	}
716 
717 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
718 
719 out:
720 	consume_skb(skb);
721 }
722 
723 static void net_dm_packet_work(struct work_struct *work)
724 {
725 	struct per_cpu_dm_data *data;
726 	struct sk_buff_head list;
727 	struct sk_buff *skb;
728 	unsigned long flags;
729 
730 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
731 
732 	__skb_queue_head_init(&list);
733 
734 	spin_lock_irqsave(&data->drop_queue.lock, flags);
735 	skb_queue_splice_tail_init(&data->drop_queue, &list);
736 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
737 
738 	while ((skb = __skb_dequeue(&list)))
739 		net_dm_packet_report(skb);
740 }
741 
742 static size_t
743 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
744 {
745 	return hw_metadata->fa_cookie ?
746 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
747 }
748 
749 static size_t
750 net_dm_hw_packet_report_size(size_t payload_len,
751 			     const struct devlink_trap_metadata *hw_metadata)
752 {
753 	size_t size;
754 
755 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
756 
757 	return NLMSG_ALIGN(size) +
758 	       /* NET_DM_ATTR_ORIGIN */
759 	       nla_total_size(sizeof(u16)) +
760 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
761 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
762 	       /* NET_DM_ATTR_HW_TRAP_NAME */
763 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
764 	       /* NET_DM_ATTR_IN_PORT */
765 	       net_dm_in_port_size() +
766 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
767 	       net_dm_flow_action_cookie_size(hw_metadata) +
768 	       /* NET_DM_ATTR_TIMESTAMP */
769 	       nla_total_size(sizeof(u64)) +
770 	       /* NET_DM_ATTR_ORIG_LEN */
771 	       nla_total_size(sizeof(u32)) +
772 	       /* NET_DM_ATTR_PROTO */
773 	       nla_total_size(sizeof(u16)) +
774 	       /* NET_DM_ATTR_PAYLOAD */
775 	       nla_total_size(payload_len);
776 }
777 
778 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
779 					struct sk_buff *skb, size_t payload_len)
780 {
781 	struct devlink_trap_metadata *hw_metadata;
782 	struct nlattr *attr;
783 	void *hdr;
784 
785 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
786 
787 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
788 			  NET_DM_CMD_PACKET_ALERT);
789 	if (!hdr)
790 		return -EMSGSIZE;
791 
792 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
793 		goto nla_put_failure;
794 
795 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
796 			   hw_metadata->trap_group_name))
797 		goto nla_put_failure;
798 
799 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
800 			   hw_metadata->trap_name))
801 		goto nla_put_failure;
802 
803 	if (hw_metadata->input_dev) {
804 		struct net_device *dev = hw_metadata->input_dev;
805 		int rc;
806 
807 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
808 						      dev->name);
809 		if (rc)
810 			goto nla_put_failure;
811 	}
812 
813 	if (hw_metadata->fa_cookie &&
814 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
815 		    hw_metadata->fa_cookie->cookie_len,
816 		    hw_metadata->fa_cookie->cookie))
817 		goto nla_put_failure;
818 
819 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
820 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
821 		goto nla_put_failure;
822 
823 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
824 		goto nla_put_failure;
825 
826 	if (!payload_len)
827 		goto out;
828 
829 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
830 		goto nla_put_failure;
831 
832 	attr = skb_put(msg, nla_total_size(payload_len));
833 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
834 	attr->nla_len = nla_attr_size(payload_len);
835 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
836 		goto nla_put_failure;
837 
838 out:
839 	genlmsg_end(msg, hdr);
840 
841 	return 0;
842 
843 nla_put_failure:
844 	genlmsg_cancel(msg, hdr);
845 	return -EMSGSIZE;
846 }
847 
848 static struct devlink_trap_metadata *
849 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
850 {
851 	const struct flow_action_cookie *fa_cookie;
852 	struct devlink_trap_metadata *hw_metadata;
853 	const char *trap_group_name;
854 	const char *trap_name;
855 
856 	hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
857 	if (!hw_metadata)
858 		return NULL;
859 
860 	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
861 	if (!trap_group_name)
862 		goto free_hw_metadata;
863 	hw_metadata->trap_group_name = trap_group_name;
864 
865 	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
866 	if (!trap_name)
867 		goto free_trap_group;
868 	hw_metadata->trap_name = trap_name;
869 
870 	if (metadata->fa_cookie) {
871 		size_t cookie_size = sizeof(*fa_cookie) +
872 				     metadata->fa_cookie->cookie_len;
873 
874 		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
875 				    GFP_ATOMIC);
876 		if (!fa_cookie)
877 			goto free_trap_name;
878 		hw_metadata->fa_cookie = fa_cookie;
879 	}
880 
881 	hw_metadata->input_dev = metadata->input_dev;
882 	netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
883 		    GFP_ATOMIC);
884 
885 	return hw_metadata;
886 
887 free_trap_name:
888 	kfree(trap_name);
889 free_trap_group:
890 	kfree(trap_group_name);
891 free_hw_metadata:
892 	kfree(hw_metadata);
893 	return NULL;
894 }
895 
896 static void
897 net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
898 {
899 	netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
900 	kfree(hw_metadata->fa_cookie);
901 	kfree(hw_metadata->trap_name);
902 	kfree(hw_metadata->trap_group_name);
903 	kfree(hw_metadata);
904 }
905 
906 static void net_dm_hw_packet_report(struct sk_buff *skb)
907 {
908 	struct devlink_trap_metadata *hw_metadata;
909 	struct sk_buff *msg;
910 	size_t payload_len;
911 	int rc;
912 
913 	if (skb->data > skb_mac_header(skb))
914 		skb_push(skb, skb->data - skb_mac_header(skb));
915 	else
916 		skb_pull(skb, skb_mac_header(skb) - skb->data);
917 
918 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
919 	if (net_dm_trunc_len)
920 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
921 
922 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
923 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
924 			GFP_KERNEL);
925 	if (!msg)
926 		goto out;
927 
928 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
929 	if (rc) {
930 		nlmsg_free(msg);
931 		goto out;
932 	}
933 
934 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
935 
936 out:
937 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
938 	consume_skb(skb);
939 }
940 
941 static void net_dm_hw_packet_work(struct work_struct *work)
942 {
943 	struct per_cpu_dm_data *hw_data;
944 	struct sk_buff_head list;
945 	struct sk_buff *skb;
946 	unsigned long flags;
947 
948 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
949 
950 	__skb_queue_head_init(&list);
951 
952 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
953 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
954 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
955 
956 	while ((skb = __skb_dequeue(&list)))
957 		net_dm_hw_packet_report(skb);
958 }
959 
960 static void
961 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
962 			    struct sk_buff *skb,
963 			    const struct devlink_trap_metadata *metadata)
964 {
965 	struct devlink_trap_metadata *n_hw_metadata;
966 	ktime_t tstamp = ktime_get_real();
967 	struct per_cpu_dm_data *hw_data;
968 	struct sk_buff *nskb;
969 	unsigned long flags;
970 
971 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
972 		return;
973 
974 	if (!skb_mac_header_was_set(skb))
975 		return;
976 
977 	nskb = skb_clone(skb, GFP_ATOMIC);
978 	if (!nskb)
979 		return;
980 
981 	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
982 	if (!n_hw_metadata)
983 		goto free;
984 
985 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
986 	nskb->tstamp = tstamp;
987 
988 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
989 
990 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
991 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
992 		__skb_queue_tail(&hw_data->drop_queue, nskb);
993 	else
994 		goto unlock_free;
995 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
996 
997 	schedule_work(&hw_data->dm_alert_work);
998 
999 	return;
1000 
1001 unlock_free:
1002 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
1003 	u64_stats_update_begin(&hw_data->stats.syncp);
1004 	u64_stats_inc(&hw_data->stats.dropped);
1005 	u64_stats_update_end(&hw_data->stats.syncp);
1006 	net_dm_hw_metadata_free(n_hw_metadata);
1007 free:
1008 	consume_skb(nskb);
1009 }
1010 
1011 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
1012 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
1013 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
1014 	.work_item_func		= net_dm_packet_work,
1015 	.hw_work_item_func	= net_dm_hw_packet_work,
1016 	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
1017 };
1018 
1019 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1020 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
1021 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
1022 };
1023 
1024 #if IS_ENABLED(CONFIG_NET_DEVLINK)
1025 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1026 {
1027 	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1028 }
1029 
1030 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1031 {
1032 	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1033 	tracepoint_synchronize_unregister();
1034 }
1035 #else
1036 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1037 {
1038 	return -EOPNOTSUPP;
1039 }
1040 
1041 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1042 {
1043 }
1044 #endif
1045 
1046 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1047 {
1048 	const struct net_dm_alert_ops *ops;
1049 	int cpu, rc;
1050 
1051 	if (monitor_hw) {
1052 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1053 		return -EAGAIN;
1054 	}
1055 
1056 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1057 
1058 	if (!try_module_get(THIS_MODULE)) {
1059 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1060 		return -ENODEV;
1061 	}
1062 
1063 	for_each_possible_cpu(cpu) {
1064 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1065 		struct net_dm_hw_entries *hw_entries;
1066 
1067 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1068 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1069 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1070 		kfree(hw_entries);
1071 	}
1072 
1073 	rc = net_dm_hw_probe_register(ops);
1074 	if (rc) {
1075 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1076 		goto err_module_put;
1077 	}
1078 
1079 	monitor_hw = true;
1080 
1081 	return 0;
1082 
1083 err_module_put:
1084 	for_each_possible_cpu(cpu) {
1085 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1086 		struct sk_buff *skb;
1087 
1088 		del_timer_sync(&hw_data->send_timer);
1089 		cancel_work_sync(&hw_data->dm_alert_work);
1090 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1091 			struct devlink_trap_metadata *hw_metadata;
1092 
1093 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1094 			net_dm_hw_metadata_free(hw_metadata);
1095 			consume_skb(skb);
1096 		}
1097 	}
1098 	module_put(THIS_MODULE);
1099 	return rc;
1100 }
1101 
1102 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1103 {
1104 	const struct net_dm_alert_ops *ops;
1105 	int cpu;
1106 
1107 	if (!monitor_hw) {
1108 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1109 		return;
1110 	}
1111 
1112 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1113 
1114 	monitor_hw = false;
1115 
1116 	net_dm_hw_probe_unregister(ops);
1117 
1118 	for_each_possible_cpu(cpu) {
1119 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1120 		struct sk_buff *skb;
1121 
1122 		del_timer_sync(&hw_data->send_timer);
1123 		cancel_work_sync(&hw_data->dm_alert_work);
1124 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1125 			struct devlink_trap_metadata *hw_metadata;
1126 
1127 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1128 			net_dm_hw_metadata_free(hw_metadata);
1129 			consume_skb(skb);
1130 		}
1131 	}
1132 
1133 	module_put(THIS_MODULE);
1134 }
1135 
1136 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1137 {
1138 	const struct net_dm_alert_ops *ops;
1139 	int cpu, rc;
1140 
1141 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1142 
1143 	if (!try_module_get(THIS_MODULE)) {
1144 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1145 		return -ENODEV;
1146 	}
1147 
1148 	for_each_possible_cpu(cpu) {
1149 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1150 		struct sk_buff *skb;
1151 
1152 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1153 		timer_setup(&data->send_timer, sched_send_work, 0);
1154 		/* Allocate a new per-CPU skb for the summary alert message and
1155 		 * free the old one which might contain stale data from
1156 		 * previous tracing.
1157 		 */
1158 		skb = reset_per_cpu_data(data);
1159 		consume_skb(skb);
1160 	}
1161 
1162 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1163 	if (rc) {
1164 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1165 		goto err_module_put;
1166 	}
1167 
1168 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1169 	if (rc) {
1170 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1171 		goto err_unregister_trace;
1172 	}
1173 
1174 	return 0;
1175 
1176 err_unregister_trace:
1177 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1178 err_module_put:
1179 	for_each_possible_cpu(cpu) {
1180 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1181 		struct sk_buff *skb;
1182 
1183 		del_timer_sync(&data->send_timer);
1184 		cancel_work_sync(&data->dm_alert_work);
1185 		while ((skb = __skb_dequeue(&data->drop_queue)))
1186 			consume_skb(skb);
1187 	}
1188 	module_put(THIS_MODULE);
1189 	return rc;
1190 }
1191 
1192 static void net_dm_trace_off_set(void)
1193 {
1194 	const struct net_dm_alert_ops *ops;
1195 	int cpu;
1196 
1197 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1198 
1199 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1200 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1201 
1202 	tracepoint_synchronize_unregister();
1203 
1204 	/* Make sure we do not send notifications to user space after request
1205 	 * to stop tracing returns.
1206 	 */
1207 	for_each_possible_cpu(cpu) {
1208 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1209 		struct sk_buff *skb;
1210 
1211 		del_timer_sync(&data->send_timer);
1212 		cancel_work_sync(&data->dm_alert_work);
1213 		while ((skb = __skb_dequeue(&data->drop_queue)))
1214 			consume_skb(skb);
1215 	}
1216 
1217 	module_put(THIS_MODULE);
1218 }
1219 
1220 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1221 {
1222 	int rc = 0;
1223 
1224 	if (state == trace_state) {
1225 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1226 		return -EAGAIN;
1227 	}
1228 
1229 	switch (state) {
1230 	case TRACE_ON:
1231 		rc = net_dm_trace_on_set(extack);
1232 		break;
1233 	case TRACE_OFF:
1234 		net_dm_trace_off_set();
1235 		break;
1236 	default:
1237 		rc = 1;
1238 		break;
1239 	}
1240 
1241 	if (!rc)
1242 		trace_state = state;
1243 	else
1244 		rc = -EINPROGRESS;
1245 
1246 	return rc;
1247 }
1248 
1249 static bool net_dm_is_monitoring(void)
1250 {
1251 	return trace_state == TRACE_ON || monitor_hw;
1252 }
1253 
1254 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1255 					   enum net_dm_alert_mode *p_alert_mode)
1256 {
1257 	u8 val;
1258 
1259 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1260 
1261 	switch (val) {
1262 	case NET_DM_ALERT_MODE_SUMMARY:
1263 	case NET_DM_ALERT_MODE_PACKET:
1264 		*p_alert_mode = val;
1265 		break;
1266 	default:
1267 		return -EINVAL;
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 static int net_dm_alert_mode_set(struct genl_info *info)
1274 {
1275 	struct netlink_ext_ack *extack = info->extack;
1276 	enum net_dm_alert_mode alert_mode;
1277 	int rc;
1278 
1279 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1280 		return 0;
1281 
1282 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1283 	if (rc) {
1284 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1285 		return -EINVAL;
1286 	}
1287 
1288 	net_dm_alert_mode = alert_mode;
1289 
1290 	return 0;
1291 }
1292 
1293 static void net_dm_trunc_len_set(struct genl_info *info)
1294 {
1295 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1296 		return;
1297 
1298 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1299 }
1300 
1301 static void net_dm_queue_len_set(struct genl_info *info)
1302 {
1303 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1304 		return;
1305 
1306 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1307 }
1308 
1309 static int net_dm_cmd_config(struct sk_buff *skb,
1310 			struct genl_info *info)
1311 {
1312 	struct netlink_ext_ack *extack = info->extack;
1313 	int rc;
1314 
1315 	if (net_dm_is_monitoring()) {
1316 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1317 		return -EBUSY;
1318 	}
1319 
1320 	rc = net_dm_alert_mode_set(info);
1321 	if (rc)
1322 		return rc;
1323 
1324 	net_dm_trunc_len_set(info);
1325 
1326 	net_dm_queue_len_set(info);
1327 
1328 	return 0;
1329 }
1330 
1331 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1332 				struct netlink_ext_ack *extack)
1333 {
1334 	bool sw_set = false;
1335 	int rc;
1336 
1337 	if (set_sw) {
1338 		rc = set_all_monitor_traces(TRACE_ON, extack);
1339 		if (rc)
1340 			return rc;
1341 		sw_set = true;
1342 	}
1343 
1344 	if (set_hw) {
1345 		rc = net_dm_hw_monitor_start(extack);
1346 		if (rc)
1347 			goto err_monitor_hw;
1348 	}
1349 
1350 	return 0;
1351 
1352 err_monitor_hw:
1353 	if (sw_set)
1354 		set_all_monitor_traces(TRACE_OFF, extack);
1355 	return rc;
1356 }
1357 
1358 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1359 				struct netlink_ext_ack *extack)
1360 {
1361 	if (set_hw)
1362 		net_dm_hw_monitor_stop(extack);
1363 	if (set_sw)
1364 		set_all_monitor_traces(TRACE_OFF, extack);
1365 }
1366 
1367 static int net_dm_cmd_trace(struct sk_buff *skb,
1368 			struct genl_info *info)
1369 {
1370 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1371 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1372 	struct netlink_ext_ack *extack = info->extack;
1373 
1374 	/* To maintain backward compatibility, we start / stop monitoring of
1375 	 * software drops if no flag is specified.
1376 	 */
1377 	if (!set_sw && !set_hw)
1378 		set_sw = true;
1379 
1380 	switch (info->genlhdr->cmd) {
1381 	case NET_DM_CMD_START:
1382 		return net_dm_monitor_start(set_sw, set_hw, extack);
1383 	case NET_DM_CMD_STOP:
1384 		net_dm_monitor_stop(set_sw, set_hw, extack);
1385 		return 0;
1386 	}
1387 
1388 	return -EOPNOTSUPP;
1389 }
1390 
1391 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1392 {
1393 	void *hdr;
1394 
1395 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1396 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1397 	if (!hdr)
1398 		return -EMSGSIZE;
1399 
1400 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1401 		goto nla_put_failure;
1402 
1403 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1404 		goto nla_put_failure;
1405 
1406 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1407 		goto nla_put_failure;
1408 
1409 	genlmsg_end(msg, hdr);
1410 
1411 	return 0;
1412 
1413 nla_put_failure:
1414 	genlmsg_cancel(msg, hdr);
1415 	return -EMSGSIZE;
1416 }
1417 
1418 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1419 {
1420 	struct sk_buff *msg;
1421 	int rc;
1422 
1423 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1424 	if (!msg)
1425 		return -ENOMEM;
1426 
1427 	rc = net_dm_config_fill(msg, info);
1428 	if (rc)
1429 		goto free_msg;
1430 
1431 	return genlmsg_reply(msg, info);
1432 
1433 free_msg:
1434 	nlmsg_free(msg);
1435 	return rc;
1436 }
1437 
1438 static void net_dm_stats_read(struct net_dm_stats *stats)
1439 {
1440 	int cpu;
1441 
1442 	memset(stats, 0, sizeof(*stats));
1443 	for_each_possible_cpu(cpu) {
1444 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1445 		struct net_dm_stats *cpu_stats = &data->stats;
1446 		unsigned int start;
1447 		u64 dropped;
1448 
1449 		do {
1450 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1451 			dropped = u64_stats_read(&cpu_stats->dropped);
1452 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1453 
1454 		u64_stats_add(&stats->dropped, dropped);
1455 	}
1456 }
1457 
1458 static int net_dm_stats_put(struct sk_buff *msg)
1459 {
1460 	struct net_dm_stats stats;
1461 	struct nlattr *attr;
1462 
1463 	net_dm_stats_read(&stats);
1464 
1465 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1466 	if (!attr)
1467 		return -EMSGSIZE;
1468 
1469 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1470 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1471 		goto nla_put_failure;
1472 
1473 	nla_nest_end(msg, attr);
1474 
1475 	return 0;
1476 
1477 nla_put_failure:
1478 	nla_nest_cancel(msg, attr);
1479 	return -EMSGSIZE;
1480 }
1481 
1482 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1483 {
1484 	int cpu;
1485 
1486 	memset(stats, 0, sizeof(*stats));
1487 	for_each_possible_cpu(cpu) {
1488 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1489 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1490 		unsigned int start;
1491 		u64 dropped;
1492 
1493 		do {
1494 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1495 			dropped = u64_stats_read(&cpu_stats->dropped);
1496 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1497 
1498 		u64_stats_add(&stats->dropped, dropped);
1499 	}
1500 }
1501 
1502 static int net_dm_hw_stats_put(struct sk_buff *msg)
1503 {
1504 	struct net_dm_stats stats;
1505 	struct nlattr *attr;
1506 
1507 	net_dm_hw_stats_read(&stats);
1508 
1509 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1510 	if (!attr)
1511 		return -EMSGSIZE;
1512 
1513 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1514 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1515 		goto nla_put_failure;
1516 
1517 	nla_nest_end(msg, attr);
1518 
1519 	return 0;
1520 
1521 nla_put_failure:
1522 	nla_nest_cancel(msg, attr);
1523 	return -EMSGSIZE;
1524 }
1525 
1526 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1527 {
1528 	void *hdr;
1529 	int rc;
1530 
1531 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1532 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1533 	if (!hdr)
1534 		return -EMSGSIZE;
1535 
1536 	rc = net_dm_stats_put(msg);
1537 	if (rc)
1538 		goto nla_put_failure;
1539 
1540 	rc = net_dm_hw_stats_put(msg);
1541 	if (rc)
1542 		goto nla_put_failure;
1543 
1544 	genlmsg_end(msg, hdr);
1545 
1546 	return 0;
1547 
1548 nla_put_failure:
1549 	genlmsg_cancel(msg, hdr);
1550 	return -EMSGSIZE;
1551 }
1552 
1553 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1554 {
1555 	struct sk_buff *msg;
1556 	int rc;
1557 
1558 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1559 	if (!msg)
1560 		return -ENOMEM;
1561 
1562 	rc = net_dm_stats_fill(msg, info);
1563 	if (rc)
1564 		goto free_msg;
1565 
1566 	return genlmsg_reply(msg, info);
1567 
1568 free_msg:
1569 	nlmsg_free(msg);
1570 	return rc;
1571 }
1572 
1573 static int dropmon_net_event(struct notifier_block *ev_block,
1574 			     unsigned long event, void *ptr)
1575 {
1576 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1577 	struct dm_hw_stat_delta *stat;
1578 
1579 	switch (event) {
1580 	case NETDEV_REGISTER:
1581 		if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1582 			break;
1583 		stat = kzalloc(sizeof(*stat), GFP_KERNEL);
1584 		if (!stat)
1585 			break;
1586 
1587 		stat->last_rx = jiffies;
1588 		rcu_assign_pointer(dev->dm_private, stat);
1589 
1590 		break;
1591 	case NETDEV_UNREGISTER:
1592 		stat = rtnl_dereference(dev->dm_private);
1593 		if (stat) {
1594 			rcu_assign_pointer(dev->dm_private, NULL);
1595 			kfree_rcu(stat, rcu);
1596 		}
1597 		break;
1598 	}
1599 	return NOTIFY_DONE;
1600 }
1601 
1602 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1603 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1604 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1605 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1606 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1607 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1608 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1609 };
1610 
1611 static const struct genl_small_ops dropmon_ops[] = {
1612 	{
1613 		.cmd = NET_DM_CMD_CONFIG,
1614 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1615 		.doit = net_dm_cmd_config,
1616 		.flags = GENL_ADMIN_PERM,
1617 	},
1618 	{
1619 		.cmd = NET_DM_CMD_START,
1620 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1621 		.doit = net_dm_cmd_trace,
1622 		.flags = GENL_ADMIN_PERM,
1623 	},
1624 	{
1625 		.cmd = NET_DM_CMD_STOP,
1626 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1627 		.doit = net_dm_cmd_trace,
1628 		.flags = GENL_ADMIN_PERM,
1629 	},
1630 	{
1631 		.cmd = NET_DM_CMD_CONFIG_GET,
1632 		.doit = net_dm_cmd_config_get,
1633 	},
1634 	{
1635 		.cmd = NET_DM_CMD_STATS_GET,
1636 		.doit = net_dm_cmd_stats_get,
1637 	},
1638 };
1639 
1640 static int net_dm_nl_pre_doit(const struct genl_split_ops *ops,
1641 			      struct sk_buff *skb, struct genl_info *info)
1642 {
1643 	mutex_lock(&net_dm_mutex);
1644 
1645 	return 0;
1646 }
1647 
1648 static void net_dm_nl_post_doit(const struct genl_split_ops *ops,
1649 				struct sk_buff *skb, struct genl_info *info)
1650 {
1651 	mutex_unlock(&net_dm_mutex);
1652 }
1653 
1654 static struct genl_family net_drop_monitor_family __ro_after_init = {
1655 	.hdrsize        = 0,
1656 	.name           = "NET_DM",
1657 	.version        = 2,
1658 	.maxattr	= NET_DM_ATTR_MAX,
1659 	.policy		= net_dm_nl_policy,
1660 	.pre_doit	= net_dm_nl_pre_doit,
1661 	.post_doit	= net_dm_nl_post_doit,
1662 	.module		= THIS_MODULE,
1663 	.small_ops	= dropmon_ops,
1664 	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1665 	.resv_start_op	= NET_DM_CMD_STATS_GET + 1,
1666 	.mcgrps		= dropmon_mcgrps,
1667 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1668 };
1669 
1670 static struct notifier_block dropmon_net_notifier = {
1671 	.notifier_call = dropmon_net_event
1672 };
1673 
1674 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1675 {
1676 	spin_lock_init(&data->lock);
1677 	skb_queue_head_init(&data->drop_queue);
1678 	u64_stats_init(&data->stats.syncp);
1679 }
1680 
1681 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1682 {
1683 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1684 }
1685 
1686 static void net_dm_cpu_data_init(int cpu)
1687 {
1688 	struct per_cpu_dm_data *data;
1689 
1690 	data = &per_cpu(dm_cpu_data, cpu);
1691 	__net_dm_cpu_data_init(data);
1692 }
1693 
1694 static void net_dm_cpu_data_fini(int cpu)
1695 {
1696 	struct per_cpu_dm_data *data;
1697 
1698 	data = &per_cpu(dm_cpu_data, cpu);
1699 	/* At this point, we should have exclusive access
1700 	 * to this struct and can free the skb inside it.
1701 	 */
1702 	consume_skb(data->skb);
1703 	__net_dm_cpu_data_fini(data);
1704 }
1705 
1706 static void net_dm_hw_cpu_data_init(int cpu)
1707 {
1708 	struct per_cpu_dm_data *hw_data;
1709 
1710 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1711 	__net_dm_cpu_data_init(hw_data);
1712 }
1713 
1714 static void net_dm_hw_cpu_data_fini(int cpu)
1715 {
1716 	struct per_cpu_dm_data *hw_data;
1717 
1718 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1719 	kfree(hw_data->hw_entries);
1720 	__net_dm_cpu_data_fini(hw_data);
1721 }
1722 
1723 static int __init init_net_drop_monitor(void)
1724 {
1725 	int cpu, rc;
1726 
1727 	pr_info("Initializing network drop monitor service\n");
1728 
1729 	if (sizeof(void *) > 8) {
1730 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1731 		return -ENOSPC;
1732 	}
1733 
1734 	rc = genl_register_family(&net_drop_monitor_family);
1735 	if (rc) {
1736 		pr_err("Could not create drop monitor netlink family\n");
1737 		return rc;
1738 	}
1739 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1740 
1741 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1742 	if (rc < 0) {
1743 		pr_crit("Failed to register netdevice notifier\n");
1744 		goto out_unreg;
1745 	}
1746 
1747 	rc = 0;
1748 
1749 	for_each_possible_cpu(cpu) {
1750 		net_dm_cpu_data_init(cpu);
1751 		net_dm_hw_cpu_data_init(cpu);
1752 	}
1753 
1754 	goto out;
1755 
1756 out_unreg:
1757 	genl_unregister_family(&net_drop_monitor_family);
1758 out:
1759 	return rc;
1760 }
1761 
1762 static void exit_net_drop_monitor(void)
1763 {
1764 	int cpu;
1765 
1766 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1767 
1768 	/*
1769 	 * Because of the module_get/put we do in the trace state change path
1770 	 * we are guaranteed not to have any current users when we get here
1771 	 */
1772 
1773 	for_each_possible_cpu(cpu) {
1774 		net_dm_hw_cpu_data_fini(cpu);
1775 		net_dm_cpu_data_fini(cpu);
1776 	}
1777 
1778 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1779 }
1780 
1781 module_init(init_net_drop_monitor);
1782 module_exit(exit_net_drop_monitor);
1783 
1784 MODULE_LICENSE("GPL v2");
1785 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1786 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1787 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1788