xref: /linux/net/core/drop_monitor.c (revision 32a92f8c89326985e05dce8b22d3f0aa07a3e1bd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Monitoring code for network dropped packet alerts
4  *
5  * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/string.h>
13 #include <linux/if_arp.h>
14 #include <linux/inetdevice.h>
15 #include <linux/inet.h>
16 #include <linux/interrupt.h>
17 #include <linux/netpoll.h>
18 #include <linux/sched.h>
19 #include <linux/delay.h>
20 #include <linux/types.h>
21 #include <linux/workqueue.h>
22 #include <linux/netlink.h>
23 #include <linux/net_dropmon.h>
24 #include <linux/bitfield.h>
25 #include <linux/percpu.h>
26 #include <linux/timer.h>
27 #include <linux/bitops.h>
28 #include <linux/slab.h>
29 #include <linux/module.h>
30 #include <net/genetlink.h>
31 #include <net/netevent.h>
32 #include <net/flow_offload.h>
33 #include <net/dropreason.h>
34 #include <net/devlink.h>
35 
36 #include <trace/events/skb.h>
37 #include <trace/events/napi.h>
38 #include <trace/events/devlink.h>
39 
40 #include <linux/unaligned.h>
41 
42 #define TRACE_ON 1
43 #define TRACE_OFF 0
44 
45 /*
46  * Globals, our netlink socket pointer
47  * and the work handle that will send up
48  * netlink alerts
49  */
50 static int trace_state = TRACE_OFF;
51 static bool monitor_hw;
52 
53 /* net_dm_mutex
54  *
55  * An overall lock guarding every operation coming from userspace.
56  */
57 static DEFINE_MUTEX(net_dm_mutex);
58 
59 struct net_dm_stats {
60 	u64_stats_t dropped;
61 	struct u64_stats_sync syncp;
62 };
63 
64 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
65 
66 struct net_dm_hw_entry {
67 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
68 	u32 count;
69 };
70 
71 struct net_dm_hw_entries {
72 	u32 num_entries;
73 	struct net_dm_hw_entry entries[];
74 };
75 
76 struct per_cpu_dm_data {
77 	raw_spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
78 					 * 'send_timer'
79 					 */
80 	union {
81 		struct sk_buff			*skb;
82 		struct net_dm_hw_entries	*hw_entries;
83 	};
84 	struct sk_buff_head	drop_queue;
85 	struct work_struct	dm_alert_work;
86 	struct timer_list	send_timer;
87 	struct net_dm_stats	stats;
88 };
89 
90 struct dm_hw_stat_delta {
91 	unsigned long last_rx;
92 	unsigned long last_drop_val;
93 	struct rcu_head rcu;
94 };
95 
96 static struct genl_family net_drop_monitor_family;
97 
98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100 
101 static int dm_hit_limit = 64;
102 static int dm_delay = 1;
103 static unsigned long dm_hw_check_delta = 2*HZ;
104 
105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106 static u32 net_dm_trunc_len;
107 static u32 net_dm_queue_len = 1000;
108 
109 struct net_dm_alert_ops {
110 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
111 				void *location,
112 				enum skb_drop_reason reason,
113 				struct sock *rx_sk);
114 	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
115 				int work, int budget);
116 	void (*work_item_func)(struct work_struct *work);
117 	void (*hw_work_item_func)(struct work_struct *work);
118 	void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
119 			      struct sk_buff *skb,
120 			      const struct devlink_trap_metadata *metadata);
121 };
122 
123 struct net_dm_skb_cb {
124 	union {
125 		struct devlink_trap_metadata *hw_metadata;
126 		void *pc;
127 	};
128 	enum skb_drop_reason reason;
129 };
130 
131 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
132 
133 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
134 {
135 	size_t al;
136 	struct net_dm_alert_msg *msg;
137 	struct nlattr *nla;
138 	struct sk_buff *skb;
139 	unsigned long flags;
140 	void *msg_header;
141 
142 	al = sizeof(struct net_dm_alert_msg);
143 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
144 	al += sizeof(struct nlattr);
145 
146 	skb = genlmsg_new(al, GFP_KERNEL);
147 
148 	if (!skb)
149 		goto err;
150 
151 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
152 				 0, NET_DM_CMD_ALERT);
153 	if (!msg_header) {
154 		nlmsg_free(skb);
155 		skb = NULL;
156 		goto err;
157 	}
158 	nla = nla_reserve(skb, NLA_UNSPEC,
159 			  sizeof(struct net_dm_alert_msg));
160 	if (!nla) {
161 		nlmsg_free(skb);
162 		skb = NULL;
163 		goto err;
164 	}
165 	msg = nla_data(nla);
166 	memset(msg, 0, al);
167 	goto out;
168 
169 err:
170 	mod_timer(&data->send_timer, jiffies + HZ / 10);
171 out:
172 	raw_spin_lock_irqsave(&data->lock, flags);
173 	swap(data->skb, skb);
174 	raw_spin_unlock_irqrestore(&data->lock, flags);
175 
176 	if (skb) {
177 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
178 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
179 
180 		genlmsg_end(skb, genlmsg_data(gnlh));
181 	}
182 
183 	return skb;
184 }
185 
186 static const struct genl_multicast_group dropmon_mcgrps[] = {
187 	{ .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, },
188 };
189 
190 static void send_dm_alert(struct work_struct *work)
191 {
192 	struct sk_buff *skb;
193 	struct per_cpu_dm_data *data;
194 
195 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
196 
197 	skb = reset_per_cpu_data(data);
198 
199 	if (skb)
200 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
201 				  0, GFP_KERNEL);
202 }
203 
204 /*
205  * This is the timer function to delay the sending of an alert
206  * in the event that more drops will arrive during the
207  * hysteresis period.
208  */
209 static void sched_send_work(struct timer_list *t)
210 {
211 	struct per_cpu_dm_data *data = timer_container_of(data, t, send_timer);
212 
213 	schedule_work(&data->dm_alert_work);
214 }
215 
216 static void trace_drop_common(struct sk_buff *skb, void *location)
217 {
218 	struct net_dm_alert_msg *msg;
219 	struct net_dm_drop_point *point;
220 	struct nlmsghdr *nlh;
221 	struct nlattr *nla;
222 	int i;
223 	struct sk_buff *dskb;
224 	struct per_cpu_dm_data *data;
225 	unsigned long flags;
226 
227 	local_irq_save(flags);
228 	data = this_cpu_ptr(&dm_cpu_data);
229 	raw_spin_lock(&data->lock);
230 	dskb = data->skb;
231 
232 	if (!dskb)
233 		goto out;
234 
235 	nlh = (struct nlmsghdr *)dskb->data;
236 	nla = genlmsg_data(nlmsg_data(nlh));
237 	msg = nla_data(nla);
238 	point = msg->points;
239 	for (i = 0; i < msg->entries; i++) {
240 		if (!memcmp(&location, &point->pc, sizeof(void *))) {
241 			point->count++;
242 			goto out;
243 		}
244 		point++;
245 	}
246 	if (msg->entries == dm_hit_limit)
247 		goto out;
248 	/*
249 	 * We need to create a new entry
250 	 */
251 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
252 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
253 	memcpy(point->pc, &location, sizeof(void *));
254 	point->count = 1;
255 	msg->entries++;
256 
257 	if (!timer_pending(&data->send_timer)) {
258 		data->send_timer.expires = jiffies + dm_delay * HZ;
259 		add_timer(&data->send_timer);
260 	}
261 
262 out:
263 	raw_spin_unlock_irqrestore(&data->lock, flags);
264 }
265 
266 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb,
267 				void *location,
268 				enum skb_drop_reason reason,
269 				struct sock *rx_sk)
270 {
271 	trace_drop_common(skb, location);
272 }
273 
274 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
275 				int work, int budget)
276 {
277 	struct net_device *dev = napi->dev;
278 	struct dm_hw_stat_delta *stat;
279 	/*
280 	 * Don't check napi structures with no associated device
281 	 */
282 	if (!dev)
283 		return;
284 
285 	rcu_read_lock();
286 	stat = rcu_dereference(dev->dm_private);
287 	if (stat) {
288 		/*
289 		 * only add a note to our monitor buffer if:
290 		 * 1) its after the last_rx delta
291 		 * 2) our rx_dropped count has gone up
292 		 */
293 		if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
294 		    (dev->stats.rx_dropped != stat->last_drop_val)) {
295 			trace_drop_common(NULL, NULL);
296 			stat->last_drop_val = dev->stats.rx_dropped;
297 			stat->last_rx = jiffies;
298 		}
299 	}
300 	rcu_read_unlock();
301 }
302 
303 static struct net_dm_hw_entries *
304 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
305 {
306 	struct net_dm_hw_entries *hw_entries;
307 	unsigned long flags;
308 
309 	hw_entries = kzalloc_flex(*hw_entries, entries, dm_hit_limit);
310 	if (!hw_entries) {
311 		/* If the memory allocation failed, we try to perform another
312 		 * allocation in 1/10 second. Otherwise, the probe function
313 		 * will constantly bail out.
314 		 */
315 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
316 	}
317 
318 	raw_spin_lock_irqsave(&hw_data->lock, flags);
319 	swap(hw_data->hw_entries, hw_entries);
320 	raw_spin_unlock_irqrestore(&hw_data->lock, flags);
321 
322 	return hw_entries;
323 }
324 
325 static int net_dm_hw_entry_put(struct sk_buff *msg,
326 			       const struct net_dm_hw_entry *hw_entry)
327 {
328 	struct nlattr *attr;
329 
330 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
331 	if (!attr)
332 		return -EMSGSIZE;
333 
334 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
335 		goto nla_put_failure;
336 
337 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
338 		goto nla_put_failure;
339 
340 	nla_nest_end(msg, attr);
341 
342 	return 0;
343 
344 nla_put_failure:
345 	nla_nest_cancel(msg, attr);
346 	return -EMSGSIZE;
347 }
348 
349 static int net_dm_hw_entries_put(struct sk_buff *msg,
350 				 const struct net_dm_hw_entries *hw_entries)
351 {
352 	struct nlattr *attr;
353 	int i;
354 
355 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
356 	if (!attr)
357 		return -EMSGSIZE;
358 
359 	for (i = 0; i < hw_entries->num_entries; i++) {
360 		int rc;
361 
362 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
363 		if (rc)
364 			goto nla_put_failure;
365 	}
366 
367 	nla_nest_end(msg, attr);
368 
369 	return 0;
370 
371 nla_put_failure:
372 	nla_nest_cancel(msg, attr);
373 	return -EMSGSIZE;
374 }
375 
376 static int
377 net_dm_hw_summary_report_fill(struct sk_buff *msg,
378 			      const struct net_dm_hw_entries *hw_entries)
379 {
380 	struct net_dm_alert_msg anc_hdr = { 0 };
381 	void *hdr;
382 	int rc;
383 
384 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
385 			  NET_DM_CMD_ALERT);
386 	if (!hdr)
387 		return -EMSGSIZE;
388 
389 	/* We need to put the ancillary header in order not to break user
390 	 * space.
391 	 */
392 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
393 		goto nla_put_failure;
394 
395 	rc = net_dm_hw_entries_put(msg, hw_entries);
396 	if (rc)
397 		goto nla_put_failure;
398 
399 	genlmsg_end(msg, hdr);
400 
401 	return 0;
402 
403 nla_put_failure:
404 	genlmsg_cancel(msg, hdr);
405 	return -EMSGSIZE;
406 }
407 
408 static void net_dm_hw_summary_work(struct work_struct *work)
409 {
410 	struct net_dm_hw_entries *hw_entries;
411 	struct per_cpu_dm_data *hw_data;
412 	struct sk_buff *msg;
413 	int rc;
414 
415 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
416 
417 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
418 	if (!hw_entries)
419 		return;
420 
421 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
422 	if (!msg)
423 		goto out;
424 
425 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
426 	if (rc) {
427 		nlmsg_free(msg);
428 		goto out;
429 	}
430 
431 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
432 
433 out:
434 	kfree(hw_entries);
435 }
436 
437 static void
438 net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
439 			     struct sk_buff *skb,
440 			     const struct devlink_trap_metadata *metadata)
441 {
442 	struct net_dm_hw_entries *hw_entries;
443 	struct net_dm_hw_entry *hw_entry;
444 	struct per_cpu_dm_data *hw_data;
445 	unsigned long flags;
446 	int i;
447 
448 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
449 		return;
450 
451 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
452 	raw_spin_lock_irqsave(&hw_data->lock, flags);
453 	hw_entries = hw_data->hw_entries;
454 
455 	if (!hw_entries)
456 		goto out;
457 
458 	for (i = 0; i < hw_entries->num_entries; i++) {
459 		hw_entry = &hw_entries->entries[i];
460 		if (!strncmp(hw_entry->trap_name, metadata->trap_name,
461 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
462 			hw_entry->count++;
463 			goto out;
464 		}
465 	}
466 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
467 		goto out;
468 
469 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
470 	strscpy(hw_entry->trap_name, metadata->trap_name,
471 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
472 	hw_entry->count = 1;
473 	hw_entries->num_entries++;
474 
475 	if (!timer_pending(&hw_data->send_timer)) {
476 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
477 		add_timer(&hw_data->send_timer);
478 	}
479 
480 out:
481 	raw_spin_unlock_irqrestore(&hw_data->lock, flags);
482 }
483 
484 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
485 	.kfree_skb_probe	= trace_kfree_skb_hit,
486 	.napi_poll_probe	= trace_napi_poll_hit,
487 	.work_item_func		= send_dm_alert,
488 	.hw_work_item_func	= net_dm_hw_summary_work,
489 	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
490 };
491 
492 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
493 					      struct sk_buff *skb,
494 					      void *location,
495 					      enum skb_drop_reason reason,
496 					      struct sock *rx_sk)
497 {
498 	ktime_t tstamp = ktime_get_real();
499 	struct per_cpu_dm_data *data;
500 	struct net_dm_skb_cb *cb;
501 	struct sk_buff *nskb;
502 	unsigned long flags;
503 
504 	if (!skb_mac_header_was_set(skb))
505 		return;
506 
507 	nskb = skb_clone(skb, GFP_ATOMIC);
508 	if (!nskb)
509 		return;
510 
511 	cb = NET_DM_SKB_CB(nskb);
512 	cb->reason = reason;
513 	cb->pc = location;
514 	/* Override the timestamp because we care about the time when the
515 	 * packet was dropped.
516 	 */
517 	nskb->tstamp = tstamp;
518 
519 	data = this_cpu_ptr(&dm_cpu_data);
520 
521 	spin_lock_irqsave(&data->drop_queue.lock, flags);
522 	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
523 		__skb_queue_tail(&data->drop_queue, nskb);
524 	else
525 		goto unlock_free;
526 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
527 
528 	schedule_work(&data->dm_alert_work);
529 
530 	return;
531 
532 unlock_free:
533 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
534 	u64_stats_update_begin(&data->stats.syncp);
535 	u64_stats_inc(&data->stats.dropped);
536 	u64_stats_update_end(&data->stats.syncp);
537 	consume_skb(nskb);
538 }
539 
540 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
541 					      struct napi_struct *napi,
542 					      int work, int budget)
543 {
544 }
545 
546 static size_t net_dm_in_port_size(void)
547 {
548 	       /* NET_DM_ATTR_IN_PORT nest */
549 	return nla_total_size(0) +
550 	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
551 	       nla_total_size(sizeof(u32)) +
552 	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
553 	       nla_total_size(IFNAMSIZ + 1);
554 }
555 
556 #define NET_DM_MAX_SYMBOL_LEN 40
557 #define NET_DM_MAX_REASON_LEN 50
558 
559 static size_t net_dm_packet_report_size(size_t payload_len)
560 {
561 	size_t size;
562 
563 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
564 
565 	return NLMSG_ALIGN(size) +
566 	       /* NET_DM_ATTR_ORIGIN */
567 	       nla_total_size(sizeof(u16)) +
568 	       /* NET_DM_ATTR_PC */
569 	       nla_total_size(sizeof(u64)) +
570 	       /* NET_DM_ATTR_SYMBOL */
571 	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
572 	       /* NET_DM_ATTR_IN_PORT */
573 	       net_dm_in_port_size() +
574 	       /* NET_DM_ATTR_TIMESTAMP */
575 	       nla_total_size(sizeof(u64)) +
576 	       /* NET_DM_ATTR_ORIG_LEN */
577 	       nla_total_size(sizeof(u32)) +
578 	       /* NET_DM_ATTR_PROTO */
579 	       nla_total_size(sizeof(u16)) +
580 	       /* NET_DM_ATTR_REASON */
581 	       nla_total_size(NET_DM_MAX_REASON_LEN + 1) +
582 	       /* NET_DM_ATTR_PAYLOAD */
583 	       nla_total_size(payload_len);
584 }
585 
586 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
587 					    const char *name)
588 {
589 	struct nlattr *attr;
590 
591 	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
592 	if (!attr)
593 		return -EMSGSIZE;
594 
595 	if (ifindex &&
596 	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
597 		goto nla_put_failure;
598 
599 	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
600 		goto nla_put_failure;
601 
602 	nla_nest_end(msg, attr);
603 
604 	return 0;
605 
606 nla_put_failure:
607 	nla_nest_cancel(msg, attr);
608 	return -EMSGSIZE;
609 }
610 
611 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
612 				     size_t payload_len)
613 {
614 	struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb);
615 	const struct drop_reason_list *list = NULL;
616 	unsigned int subsys, subsys_reason;
617 	char buf[NET_DM_MAX_SYMBOL_LEN];
618 	struct nlattr *attr;
619 	void *hdr;
620 	int rc;
621 
622 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
623 			  NET_DM_CMD_PACKET_ALERT);
624 	if (!hdr)
625 		return -EMSGSIZE;
626 
627 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
628 		goto nla_put_failure;
629 
630 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
631 			      NET_DM_ATTR_PAD))
632 		goto nla_put_failure;
633 
634 	rcu_read_lock();
635 	subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK);
636 	if (subsys < SKB_DROP_REASON_SUBSYS_NUM)
637 		list = rcu_dereference(drop_reasons_by_subsys[subsys]);
638 	subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK;
639 	if (!list ||
640 	    subsys_reason >= list->n_reasons ||
641 	    !list->reasons[subsys_reason] ||
642 	    strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) {
643 		list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]);
644 		subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED;
645 	}
646 	if (nla_put_string(msg, NET_DM_ATTR_REASON,
647 			   list->reasons[subsys_reason])) {
648 		rcu_read_unlock();
649 		goto nla_put_failure;
650 	}
651 	rcu_read_unlock();
652 
653 	snprintf(buf, sizeof(buf), "%pS", cb->pc);
654 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
655 		goto nla_put_failure;
656 
657 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
658 	if (rc)
659 		goto nla_put_failure;
660 
661 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
662 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
663 		goto nla_put_failure;
664 
665 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
666 		goto nla_put_failure;
667 
668 	if (!payload_len)
669 		goto out;
670 
671 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
672 		goto nla_put_failure;
673 
674 	attr = skb_put(msg, nla_total_size(payload_len));
675 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
676 	attr->nla_len = nla_attr_size(payload_len);
677 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
678 		goto nla_put_failure;
679 
680 out:
681 	genlmsg_end(msg, hdr);
682 
683 	return 0;
684 
685 nla_put_failure:
686 	genlmsg_cancel(msg, hdr);
687 	return -EMSGSIZE;
688 }
689 
690 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
691 
692 static void net_dm_packet_report(struct sk_buff *skb)
693 {
694 	struct sk_buff *msg;
695 	size_t payload_len;
696 	int rc;
697 
698 	/* Make sure we start copying the packet from the MAC header */
699 	if (skb->data > skb_mac_header(skb))
700 		skb_push(skb, skb->data - skb_mac_header(skb));
701 	else
702 		skb_pull(skb, skb_mac_header(skb) - skb->data);
703 
704 	/* Ensure packet fits inside a single netlink attribute */
705 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
706 	if (net_dm_trunc_len)
707 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
708 
709 	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
710 	if (!msg)
711 		goto out;
712 
713 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
714 	if (rc) {
715 		nlmsg_free(msg);
716 		goto out;
717 	}
718 
719 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
720 
721 out:
722 	consume_skb(skb);
723 }
724 
725 static void net_dm_packet_work(struct work_struct *work)
726 {
727 	struct per_cpu_dm_data *data;
728 	struct sk_buff_head list;
729 	struct sk_buff *skb;
730 	unsigned long flags;
731 
732 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
733 
734 	__skb_queue_head_init(&list);
735 
736 	spin_lock_irqsave(&data->drop_queue.lock, flags);
737 	skb_queue_splice_tail_init(&data->drop_queue, &list);
738 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
739 
740 	while ((skb = __skb_dequeue(&list)))
741 		net_dm_packet_report(skb);
742 }
743 
744 static size_t
745 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
746 {
747 	return hw_metadata->fa_cookie ?
748 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
749 }
750 
751 static size_t
752 net_dm_hw_packet_report_size(size_t payload_len,
753 			     const struct devlink_trap_metadata *hw_metadata)
754 {
755 	size_t size;
756 
757 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
758 
759 	return NLMSG_ALIGN(size) +
760 	       /* NET_DM_ATTR_ORIGIN */
761 	       nla_total_size(sizeof(u16)) +
762 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
763 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
764 	       /* NET_DM_ATTR_HW_TRAP_NAME */
765 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
766 	       /* NET_DM_ATTR_IN_PORT */
767 	       net_dm_in_port_size() +
768 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
769 	       net_dm_flow_action_cookie_size(hw_metadata) +
770 	       /* NET_DM_ATTR_TIMESTAMP */
771 	       nla_total_size(sizeof(u64)) +
772 	       /* NET_DM_ATTR_ORIG_LEN */
773 	       nla_total_size(sizeof(u32)) +
774 	       /* NET_DM_ATTR_PROTO */
775 	       nla_total_size(sizeof(u16)) +
776 	       /* NET_DM_ATTR_PAYLOAD */
777 	       nla_total_size(payload_len);
778 }
779 
780 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
781 					struct sk_buff *skb, size_t payload_len)
782 {
783 	struct devlink_trap_metadata *hw_metadata;
784 	struct nlattr *attr;
785 	void *hdr;
786 
787 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
788 
789 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
790 			  NET_DM_CMD_PACKET_ALERT);
791 	if (!hdr)
792 		return -EMSGSIZE;
793 
794 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
795 		goto nla_put_failure;
796 
797 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
798 			   hw_metadata->trap_group_name))
799 		goto nla_put_failure;
800 
801 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
802 			   hw_metadata->trap_name))
803 		goto nla_put_failure;
804 
805 	if (hw_metadata->input_dev) {
806 		struct net_device *dev = hw_metadata->input_dev;
807 		int rc;
808 
809 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
810 						      dev->name);
811 		if (rc)
812 			goto nla_put_failure;
813 	}
814 
815 	if (hw_metadata->fa_cookie &&
816 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
817 		    hw_metadata->fa_cookie->cookie_len,
818 		    hw_metadata->fa_cookie->cookie))
819 		goto nla_put_failure;
820 
821 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
822 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
823 		goto nla_put_failure;
824 
825 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
826 		goto nla_put_failure;
827 
828 	if (!payload_len)
829 		goto out;
830 
831 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
832 		goto nla_put_failure;
833 
834 	attr = skb_put(msg, nla_total_size(payload_len));
835 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
836 	attr->nla_len = nla_attr_size(payload_len);
837 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
838 		goto nla_put_failure;
839 
840 out:
841 	genlmsg_end(msg, hdr);
842 
843 	return 0;
844 
845 nla_put_failure:
846 	genlmsg_cancel(msg, hdr);
847 	return -EMSGSIZE;
848 }
849 
850 static struct devlink_trap_metadata *
851 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
852 {
853 	const struct flow_action_cookie *fa_cookie;
854 	struct devlink_trap_metadata *hw_metadata;
855 	const char *trap_group_name;
856 	const char *trap_name;
857 
858 	hw_metadata = kzalloc_obj(*hw_metadata, GFP_ATOMIC);
859 	if (!hw_metadata)
860 		return NULL;
861 
862 	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
863 	if (!trap_group_name)
864 		goto free_hw_metadata;
865 	hw_metadata->trap_group_name = trap_group_name;
866 
867 	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
868 	if (!trap_name)
869 		goto free_trap_group;
870 	hw_metadata->trap_name = trap_name;
871 
872 	if (metadata->fa_cookie) {
873 		size_t cookie_size = sizeof(*fa_cookie) +
874 				     metadata->fa_cookie->cookie_len;
875 
876 		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
877 				    GFP_ATOMIC);
878 		if (!fa_cookie)
879 			goto free_trap_name;
880 		hw_metadata->fa_cookie = fa_cookie;
881 	}
882 
883 	hw_metadata->input_dev = metadata->input_dev;
884 	netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
885 		    GFP_ATOMIC);
886 
887 	return hw_metadata;
888 
889 free_trap_name:
890 	kfree(trap_name);
891 free_trap_group:
892 	kfree(trap_group_name);
893 free_hw_metadata:
894 	kfree(hw_metadata);
895 	return NULL;
896 }
897 
898 static void
899 net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
900 {
901 	netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
902 	kfree(hw_metadata->fa_cookie);
903 	kfree(hw_metadata->trap_name);
904 	kfree(hw_metadata->trap_group_name);
905 	kfree(hw_metadata);
906 }
907 
908 static void net_dm_hw_packet_report(struct sk_buff *skb)
909 {
910 	struct devlink_trap_metadata *hw_metadata;
911 	struct sk_buff *msg;
912 	size_t payload_len;
913 	int rc;
914 
915 	if (skb->data > skb_mac_header(skb))
916 		skb_push(skb, skb->data - skb_mac_header(skb));
917 	else
918 		skb_pull(skb, skb_mac_header(skb) - skb->data);
919 
920 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
921 	if (net_dm_trunc_len)
922 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
923 
924 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
925 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
926 			GFP_KERNEL);
927 	if (!msg)
928 		goto out;
929 
930 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
931 	if (rc) {
932 		nlmsg_free(msg);
933 		goto out;
934 	}
935 
936 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
937 
938 out:
939 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
940 	consume_skb(skb);
941 }
942 
943 static void net_dm_hw_packet_work(struct work_struct *work)
944 {
945 	struct per_cpu_dm_data *hw_data;
946 	struct sk_buff_head list;
947 	struct sk_buff *skb;
948 	unsigned long flags;
949 
950 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
951 
952 	__skb_queue_head_init(&list);
953 
954 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
955 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
956 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
957 
958 	while ((skb = __skb_dequeue(&list)))
959 		net_dm_hw_packet_report(skb);
960 }
961 
962 static void
963 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
964 			    struct sk_buff *skb,
965 			    const struct devlink_trap_metadata *metadata)
966 {
967 	struct devlink_trap_metadata *n_hw_metadata;
968 	ktime_t tstamp = ktime_get_real();
969 	struct per_cpu_dm_data *hw_data;
970 	struct sk_buff *nskb;
971 	unsigned long flags;
972 
973 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
974 		return;
975 
976 	if (!skb_mac_header_was_set(skb))
977 		return;
978 
979 	nskb = skb_clone(skb, GFP_ATOMIC);
980 	if (!nskb)
981 		return;
982 
983 	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
984 	if (!n_hw_metadata)
985 		goto free;
986 
987 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
988 	nskb->tstamp = tstamp;
989 
990 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
991 
992 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
993 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
994 		__skb_queue_tail(&hw_data->drop_queue, nskb);
995 	else
996 		goto unlock_free;
997 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
998 
999 	schedule_work(&hw_data->dm_alert_work);
1000 
1001 	return;
1002 
1003 unlock_free:
1004 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
1005 	u64_stats_update_begin(&hw_data->stats.syncp);
1006 	u64_stats_inc(&hw_data->stats.dropped);
1007 	u64_stats_update_end(&hw_data->stats.syncp);
1008 	net_dm_hw_metadata_free(n_hw_metadata);
1009 free:
1010 	consume_skb(nskb);
1011 }
1012 
1013 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
1014 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
1015 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
1016 	.work_item_func		= net_dm_packet_work,
1017 	.hw_work_item_func	= net_dm_hw_packet_work,
1018 	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
1019 };
1020 
1021 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1022 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
1023 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
1024 };
1025 
1026 #if IS_ENABLED(CONFIG_NET_DEVLINK)
1027 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1028 {
1029 	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1030 }
1031 
1032 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1033 {
1034 	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1035 	tracepoint_synchronize_unregister();
1036 }
1037 #else
1038 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1039 {
1040 	return -EOPNOTSUPP;
1041 }
1042 
1043 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1044 {
1045 }
1046 #endif
1047 
1048 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1049 {
1050 	const struct net_dm_alert_ops *ops;
1051 	int cpu, rc;
1052 
1053 	if (monitor_hw) {
1054 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1055 		return -EAGAIN;
1056 	}
1057 
1058 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1059 
1060 	if (!try_module_get(THIS_MODULE)) {
1061 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1062 		return -ENODEV;
1063 	}
1064 
1065 	for_each_possible_cpu(cpu) {
1066 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1067 		struct net_dm_hw_entries *hw_entries;
1068 
1069 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1070 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1071 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1072 		kfree(hw_entries);
1073 	}
1074 
1075 	rc = net_dm_hw_probe_register(ops);
1076 	if (rc) {
1077 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1078 		goto err_module_put;
1079 	}
1080 
1081 	monitor_hw = true;
1082 
1083 	return 0;
1084 
1085 err_module_put:
1086 	for_each_possible_cpu(cpu) {
1087 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1088 		struct sk_buff *skb;
1089 
1090 		timer_delete_sync(&hw_data->send_timer);
1091 		cancel_work_sync(&hw_data->dm_alert_work);
1092 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1093 			struct devlink_trap_metadata *hw_metadata;
1094 
1095 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1096 			net_dm_hw_metadata_free(hw_metadata);
1097 			consume_skb(skb);
1098 		}
1099 	}
1100 	module_put(THIS_MODULE);
1101 	return rc;
1102 }
1103 
1104 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1105 {
1106 	const struct net_dm_alert_ops *ops;
1107 	int cpu;
1108 
1109 	if (!monitor_hw) {
1110 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1111 		return;
1112 	}
1113 
1114 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1115 
1116 	monitor_hw = false;
1117 
1118 	net_dm_hw_probe_unregister(ops);
1119 
1120 	for_each_possible_cpu(cpu) {
1121 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1122 		struct sk_buff *skb;
1123 
1124 		timer_delete_sync(&hw_data->send_timer);
1125 		cancel_work_sync(&hw_data->dm_alert_work);
1126 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1127 			struct devlink_trap_metadata *hw_metadata;
1128 
1129 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1130 			net_dm_hw_metadata_free(hw_metadata);
1131 			consume_skb(skb);
1132 		}
1133 	}
1134 
1135 	module_put(THIS_MODULE);
1136 }
1137 
1138 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1139 {
1140 	const struct net_dm_alert_ops *ops;
1141 	int cpu, rc;
1142 
1143 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1144 
1145 	if (!try_module_get(THIS_MODULE)) {
1146 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1147 		return -ENODEV;
1148 	}
1149 
1150 	for_each_possible_cpu(cpu) {
1151 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1152 		struct sk_buff *skb;
1153 
1154 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1155 		timer_setup(&data->send_timer, sched_send_work, 0);
1156 		/* Allocate a new per-CPU skb for the summary alert message and
1157 		 * free the old one which might contain stale data from
1158 		 * previous tracing.
1159 		 */
1160 		skb = reset_per_cpu_data(data);
1161 		consume_skb(skb);
1162 	}
1163 
1164 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1165 	if (rc) {
1166 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1167 		goto err_module_put;
1168 	}
1169 
1170 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1171 	if (rc) {
1172 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1173 		goto err_unregister_trace;
1174 	}
1175 
1176 	return 0;
1177 
1178 err_unregister_trace:
1179 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1180 err_module_put:
1181 	for_each_possible_cpu(cpu) {
1182 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1183 		struct sk_buff *skb;
1184 
1185 		timer_delete_sync(&data->send_timer);
1186 		cancel_work_sync(&data->dm_alert_work);
1187 		while ((skb = __skb_dequeue(&data->drop_queue)))
1188 			consume_skb(skb);
1189 	}
1190 	module_put(THIS_MODULE);
1191 	return rc;
1192 }
1193 
1194 static void net_dm_trace_off_set(void)
1195 {
1196 	const struct net_dm_alert_ops *ops;
1197 	int cpu;
1198 
1199 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1200 
1201 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1202 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1203 
1204 	tracepoint_synchronize_unregister();
1205 
1206 	/* Make sure we do not send notifications to user space after request
1207 	 * to stop tracing returns.
1208 	 */
1209 	for_each_possible_cpu(cpu) {
1210 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1211 		struct sk_buff *skb;
1212 
1213 		timer_delete_sync(&data->send_timer);
1214 		cancel_work_sync(&data->dm_alert_work);
1215 		while ((skb = __skb_dequeue(&data->drop_queue)))
1216 			consume_skb(skb);
1217 	}
1218 
1219 	module_put(THIS_MODULE);
1220 }
1221 
1222 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1223 {
1224 	int rc = 0;
1225 
1226 	if (state == trace_state) {
1227 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1228 		return -EAGAIN;
1229 	}
1230 
1231 	switch (state) {
1232 	case TRACE_ON:
1233 		rc = net_dm_trace_on_set(extack);
1234 		break;
1235 	case TRACE_OFF:
1236 		net_dm_trace_off_set();
1237 		break;
1238 	default:
1239 		rc = 1;
1240 		break;
1241 	}
1242 
1243 	if (!rc)
1244 		trace_state = state;
1245 	else
1246 		rc = -EINPROGRESS;
1247 
1248 	return rc;
1249 }
1250 
1251 static bool net_dm_is_monitoring(void)
1252 {
1253 	return trace_state == TRACE_ON || monitor_hw;
1254 }
1255 
1256 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1257 					   enum net_dm_alert_mode *p_alert_mode)
1258 {
1259 	u8 val;
1260 
1261 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1262 
1263 	switch (val) {
1264 	case NET_DM_ALERT_MODE_SUMMARY:
1265 	case NET_DM_ALERT_MODE_PACKET:
1266 		*p_alert_mode = val;
1267 		break;
1268 	default:
1269 		return -EINVAL;
1270 	}
1271 
1272 	return 0;
1273 }
1274 
1275 static int net_dm_alert_mode_set(struct genl_info *info)
1276 {
1277 	struct netlink_ext_ack *extack = info->extack;
1278 	enum net_dm_alert_mode alert_mode;
1279 	int rc;
1280 
1281 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1282 		return 0;
1283 
1284 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1285 	if (rc) {
1286 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1287 		return -EINVAL;
1288 	}
1289 
1290 	net_dm_alert_mode = alert_mode;
1291 
1292 	return 0;
1293 }
1294 
1295 static void net_dm_trunc_len_set(struct genl_info *info)
1296 {
1297 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1298 		return;
1299 
1300 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1301 }
1302 
1303 static void net_dm_queue_len_set(struct genl_info *info)
1304 {
1305 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1306 		return;
1307 
1308 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1309 }
1310 
1311 static int net_dm_cmd_config(struct sk_buff *skb,
1312 			struct genl_info *info)
1313 {
1314 	struct netlink_ext_ack *extack = info->extack;
1315 	int rc;
1316 
1317 	if (net_dm_is_monitoring()) {
1318 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1319 		return -EBUSY;
1320 	}
1321 
1322 	rc = net_dm_alert_mode_set(info);
1323 	if (rc)
1324 		return rc;
1325 
1326 	net_dm_trunc_len_set(info);
1327 
1328 	net_dm_queue_len_set(info);
1329 
1330 	return 0;
1331 }
1332 
1333 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1334 				struct netlink_ext_ack *extack)
1335 {
1336 	bool sw_set = false;
1337 	int rc;
1338 
1339 	if (set_sw) {
1340 		rc = set_all_monitor_traces(TRACE_ON, extack);
1341 		if (rc)
1342 			return rc;
1343 		sw_set = true;
1344 	}
1345 
1346 	if (set_hw) {
1347 		rc = net_dm_hw_monitor_start(extack);
1348 		if (rc)
1349 			goto err_monitor_hw;
1350 	}
1351 
1352 	return 0;
1353 
1354 err_monitor_hw:
1355 	if (sw_set)
1356 		set_all_monitor_traces(TRACE_OFF, extack);
1357 	return rc;
1358 }
1359 
1360 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1361 				struct netlink_ext_ack *extack)
1362 {
1363 	if (set_hw)
1364 		net_dm_hw_monitor_stop(extack);
1365 	if (set_sw)
1366 		set_all_monitor_traces(TRACE_OFF, extack);
1367 }
1368 
1369 static int net_dm_cmd_trace(struct sk_buff *skb,
1370 			struct genl_info *info)
1371 {
1372 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1373 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1374 	struct netlink_ext_ack *extack = info->extack;
1375 
1376 	/* To maintain backward compatibility, we start / stop monitoring of
1377 	 * software drops if no flag is specified.
1378 	 */
1379 	if (!set_sw && !set_hw)
1380 		set_sw = true;
1381 
1382 	switch (info->genlhdr->cmd) {
1383 	case NET_DM_CMD_START:
1384 		return net_dm_monitor_start(set_sw, set_hw, extack);
1385 	case NET_DM_CMD_STOP:
1386 		net_dm_monitor_stop(set_sw, set_hw, extack);
1387 		return 0;
1388 	}
1389 
1390 	return -EOPNOTSUPP;
1391 }
1392 
1393 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1394 {
1395 	void *hdr;
1396 
1397 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1398 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1399 	if (!hdr)
1400 		return -EMSGSIZE;
1401 
1402 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1403 		goto nla_put_failure;
1404 
1405 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1406 		goto nla_put_failure;
1407 
1408 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1409 		goto nla_put_failure;
1410 
1411 	genlmsg_end(msg, hdr);
1412 
1413 	return 0;
1414 
1415 nla_put_failure:
1416 	genlmsg_cancel(msg, hdr);
1417 	return -EMSGSIZE;
1418 }
1419 
1420 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1421 {
1422 	struct sk_buff *msg;
1423 	int rc;
1424 
1425 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1426 	if (!msg)
1427 		return -ENOMEM;
1428 
1429 	rc = net_dm_config_fill(msg, info);
1430 	if (rc)
1431 		goto free_msg;
1432 
1433 	return genlmsg_reply(msg, info);
1434 
1435 free_msg:
1436 	nlmsg_free(msg);
1437 	return rc;
1438 }
1439 
1440 static void net_dm_stats_read(struct net_dm_stats *stats)
1441 {
1442 	int cpu;
1443 
1444 	memset(stats, 0, sizeof(*stats));
1445 	for_each_possible_cpu(cpu) {
1446 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1447 		struct net_dm_stats *cpu_stats = &data->stats;
1448 		unsigned int start;
1449 		u64 dropped;
1450 
1451 		do {
1452 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1453 			dropped = u64_stats_read(&cpu_stats->dropped);
1454 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1455 
1456 		u64_stats_add(&stats->dropped, dropped);
1457 	}
1458 }
1459 
1460 static int net_dm_stats_put(struct sk_buff *msg)
1461 {
1462 	struct net_dm_stats stats;
1463 	struct nlattr *attr;
1464 
1465 	net_dm_stats_read(&stats);
1466 
1467 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1468 	if (!attr)
1469 		return -EMSGSIZE;
1470 
1471 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1472 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1473 		goto nla_put_failure;
1474 
1475 	nla_nest_end(msg, attr);
1476 
1477 	return 0;
1478 
1479 nla_put_failure:
1480 	nla_nest_cancel(msg, attr);
1481 	return -EMSGSIZE;
1482 }
1483 
1484 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1485 {
1486 	int cpu;
1487 
1488 	memset(stats, 0, sizeof(*stats));
1489 	for_each_possible_cpu(cpu) {
1490 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1491 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1492 		unsigned int start;
1493 		u64 dropped;
1494 
1495 		do {
1496 			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1497 			dropped = u64_stats_read(&cpu_stats->dropped);
1498 		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1499 
1500 		u64_stats_add(&stats->dropped, dropped);
1501 	}
1502 }
1503 
1504 static int net_dm_hw_stats_put(struct sk_buff *msg)
1505 {
1506 	struct net_dm_stats stats;
1507 	struct nlattr *attr;
1508 
1509 	net_dm_hw_stats_read(&stats);
1510 
1511 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1512 	if (!attr)
1513 		return -EMSGSIZE;
1514 
1515 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1516 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1517 		goto nla_put_failure;
1518 
1519 	nla_nest_end(msg, attr);
1520 
1521 	return 0;
1522 
1523 nla_put_failure:
1524 	nla_nest_cancel(msg, attr);
1525 	return -EMSGSIZE;
1526 }
1527 
1528 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1529 {
1530 	void *hdr;
1531 	int rc;
1532 
1533 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1534 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1535 	if (!hdr)
1536 		return -EMSGSIZE;
1537 
1538 	rc = net_dm_stats_put(msg);
1539 	if (rc)
1540 		goto nla_put_failure;
1541 
1542 	rc = net_dm_hw_stats_put(msg);
1543 	if (rc)
1544 		goto nla_put_failure;
1545 
1546 	genlmsg_end(msg, hdr);
1547 
1548 	return 0;
1549 
1550 nla_put_failure:
1551 	genlmsg_cancel(msg, hdr);
1552 	return -EMSGSIZE;
1553 }
1554 
1555 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1556 {
1557 	struct sk_buff *msg;
1558 	int rc;
1559 
1560 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1561 	if (!msg)
1562 		return -ENOMEM;
1563 
1564 	rc = net_dm_stats_fill(msg, info);
1565 	if (rc)
1566 		goto free_msg;
1567 
1568 	return genlmsg_reply(msg, info);
1569 
1570 free_msg:
1571 	nlmsg_free(msg);
1572 	return rc;
1573 }
1574 
1575 static int dropmon_net_event(struct notifier_block *ev_block,
1576 			     unsigned long event, void *ptr)
1577 {
1578 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1579 	struct dm_hw_stat_delta *stat;
1580 
1581 	switch (event) {
1582 	case NETDEV_REGISTER:
1583 		if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1584 			break;
1585 		stat = kzalloc_obj(*stat);
1586 		if (!stat)
1587 			break;
1588 
1589 		stat->last_rx = jiffies;
1590 		rcu_assign_pointer(dev->dm_private, stat);
1591 
1592 		break;
1593 	case NETDEV_UNREGISTER:
1594 		stat = rtnl_dereference(dev->dm_private);
1595 		if (stat) {
1596 			rcu_assign_pointer(dev->dm_private, NULL);
1597 			kfree_rcu(stat, rcu);
1598 		}
1599 		break;
1600 	}
1601 	return NOTIFY_DONE;
1602 }
1603 
1604 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1605 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1606 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1607 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1608 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1609 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1610 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1611 };
1612 
1613 static const struct genl_small_ops dropmon_ops[] = {
1614 	{
1615 		.cmd = NET_DM_CMD_CONFIG,
1616 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1617 		.doit = net_dm_cmd_config,
1618 		.flags = GENL_ADMIN_PERM,
1619 	},
1620 	{
1621 		.cmd = NET_DM_CMD_START,
1622 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1623 		.doit = net_dm_cmd_trace,
1624 		.flags = GENL_ADMIN_PERM,
1625 	},
1626 	{
1627 		.cmd = NET_DM_CMD_STOP,
1628 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1629 		.doit = net_dm_cmd_trace,
1630 		.flags = GENL_ADMIN_PERM,
1631 	},
1632 	{
1633 		.cmd = NET_DM_CMD_CONFIG_GET,
1634 		.doit = net_dm_cmd_config_get,
1635 	},
1636 	{
1637 		.cmd = NET_DM_CMD_STATS_GET,
1638 		.doit = net_dm_cmd_stats_get,
1639 	},
1640 };
1641 
1642 static int net_dm_nl_pre_doit(const struct genl_split_ops *ops,
1643 			      struct sk_buff *skb, struct genl_info *info)
1644 {
1645 	mutex_lock(&net_dm_mutex);
1646 
1647 	return 0;
1648 }
1649 
1650 static void net_dm_nl_post_doit(const struct genl_split_ops *ops,
1651 				struct sk_buff *skb, struct genl_info *info)
1652 {
1653 	mutex_unlock(&net_dm_mutex);
1654 }
1655 
1656 static struct genl_family net_drop_monitor_family __ro_after_init = {
1657 	.hdrsize        = 0,
1658 	.name           = "NET_DM",
1659 	.version        = 2,
1660 	.maxattr	= NET_DM_ATTR_MAX,
1661 	.policy		= net_dm_nl_policy,
1662 	.pre_doit	= net_dm_nl_pre_doit,
1663 	.post_doit	= net_dm_nl_post_doit,
1664 	.module		= THIS_MODULE,
1665 	.small_ops	= dropmon_ops,
1666 	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1667 	.resv_start_op	= NET_DM_CMD_STATS_GET + 1,
1668 	.mcgrps		= dropmon_mcgrps,
1669 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1670 };
1671 
1672 static struct notifier_block dropmon_net_notifier = {
1673 	.notifier_call = dropmon_net_event
1674 };
1675 
1676 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1677 {
1678 	raw_spin_lock_init(&data->lock);
1679 	skb_queue_head_init(&data->drop_queue);
1680 	u64_stats_init(&data->stats.syncp);
1681 }
1682 
1683 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1684 {
1685 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1686 }
1687 
1688 static void net_dm_cpu_data_init(int cpu)
1689 {
1690 	struct per_cpu_dm_data *data;
1691 
1692 	data = &per_cpu(dm_cpu_data, cpu);
1693 	__net_dm_cpu_data_init(data);
1694 }
1695 
1696 static void net_dm_cpu_data_fini(int cpu)
1697 {
1698 	struct per_cpu_dm_data *data;
1699 
1700 	data = &per_cpu(dm_cpu_data, cpu);
1701 	/* At this point, we should have exclusive access
1702 	 * to this struct and can free the skb inside it.
1703 	 */
1704 	consume_skb(data->skb);
1705 	__net_dm_cpu_data_fini(data);
1706 }
1707 
1708 static void net_dm_hw_cpu_data_init(int cpu)
1709 {
1710 	struct per_cpu_dm_data *hw_data;
1711 
1712 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1713 	__net_dm_cpu_data_init(hw_data);
1714 }
1715 
1716 static void net_dm_hw_cpu_data_fini(int cpu)
1717 {
1718 	struct per_cpu_dm_data *hw_data;
1719 
1720 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1721 	kfree(hw_data->hw_entries);
1722 	__net_dm_cpu_data_fini(hw_data);
1723 }
1724 
1725 static int __init init_net_drop_monitor(void)
1726 {
1727 	int cpu, rc;
1728 
1729 	pr_info("Initializing network drop monitor service\n");
1730 
1731 	if (sizeof(void *) > 8) {
1732 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1733 		return -ENOSPC;
1734 	}
1735 
1736 	for_each_possible_cpu(cpu) {
1737 		net_dm_cpu_data_init(cpu);
1738 		net_dm_hw_cpu_data_init(cpu);
1739 	}
1740 
1741 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1742 	if (rc < 0) {
1743 		pr_crit("Failed to register netdevice notifier\n");
1744 		return rc;
1745 	}
1746 
1747 	rc = genl_register_family(&net_drop_monitor_family);
1748 	if (rc) {
1749 		pr_err("Could not create drop monitor netlink family\n");
1750 		goto out_unreg;
1751 	}
1752 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1753 
1754 	rc = 0;
1755 
1756 	goto out;
1757 
1758 out_unreg:
1759 	WARN_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1760 out:
1761 	return rc;
1762 }
1763 
1764 static void exit_net_drop_monitor(void)
1765 {
1766 	int cpu;
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 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1773 
1774 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1775 
1776 	for_each_possible_cpu(cpu) {
1777 		net_dm_hw_cpu_data_fini(cpu);
1778 		net_dm_cpu_data_fini(cpu);
1779 	}
1780 }
1781 
1782 module_init(init_net_drop_monitor);
1783 module_exit(exit_net_drop_monitor);
1784 
1785 MODULE_LICENSE("GPL v2");
1786 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1787 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1788 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1789