xref: /linux/net/core/drop_monitor.c (revision 3c4fc7bf4c9e66fe71abcbf93f62f4ddb89b7f15)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Monitoring code for network dropped packet alerts
4  *
5  * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/string.h>
13 #include <linux/if_arp.h>
14 #include <linux/inetdevice.h>
15 #include <linux/inet.h>
16 #include <linux/interrupt.h>
17 #include <linux/netpoll.h>
18 #include <linux/sched.h>
19 #include <linux/delay.h>
20 #include <linux/types.h>
21 #include <linux/workqueue.h>
22 #include <linux/netlink.h>
23 #include <linux/net_dropmon.h>
24 #include <linux/percpu.h>
25 #include <linux/timer.h>
26 #include <linux/bitops.h>
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <net/genetlink.h>
30 #include <net/netevent.h>
31 #include <net/flow_offload.h>
32 #include <net/devlink.h>
33 
34 #include <trace/events/skb.h>
35 #include <trace/events/napi.h>
36 #include <trace/events/devlink.h>
37 
38 #include <asm/unaligned.h>
39 
40 #define TRACE_ON 1
41 #define TRACE_OFF 0
42 
43 /*
44  * Globals, our netlink socket pointer
45  * and the work handle that will send up
46  * netlink alerts
47  */
48 static int trace_state = TRACE_OFF;
49 static bool monitor_hw;
50 
51 /* net_dm_mutex
52  *
53  * An overall lock guarding every operation coming from userspace.
54  */
55 static DEFINE_MUTEX(net_dm_mutex);
56 
57 struct net_dm_stats {
58 	u64_stats_t dropped;
59 	struct u64_stats_sync syncp;
60 };
61 
62 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
63 
64 struct net_dm_hw_entry {
65 	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
66 	u32 count;
67 };
68 
69 struct net_dm_hw_entries {
70 	u32 num_entries;
71 	struct net_dm_hw_entry entries[];
72 };
73 
74 struct per_cpu_dm_data {
75 	spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
76 					 * 'send_timer'
77 					 */
78 	union {
79 		struct sk_buff			*skb;
80 		struct net_dm_hw_entries	*hw_entries;
81 	};
82 	struct sk_buff_head	drop_queue;
83 	struct work_struct	dm_alert_work;
84 	struct timer_list	send_timer;
85 	struct net_dm_stats	stats;
86 };
87 
88 struct dm_hw_stat_delta {
89 	unsigned long last_rx;
90 	unsigned long last_drop_val;
91 	struct rcu_head rcu;
92 };
93 
94 static struct genl_family net_drop_monitor_family;
95 
96 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
97 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
98 
99 static int dm_hit_limit = 64;
100 static int dm_delay = 1;
101 static unsigned long dm_hw_check_delta = 2*HZ;
102 
103 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
104 static u32 net_dm_trunc_len;
105 static u32 net_dm_queue_len = 1000;
106 
107 struct net_dm_alert_ops {
108 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
109 				void *location,
110 				enum skb_drop_reason reason);
111 	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
112 				int work, int budget);
113 	void (*work_item_func)(struct work_struct *work);
114 	void (*hw_work_item_func)(struct work_struct *work);
115 	void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
116 			      struct sk_buff *skb,
117 			      const struct devlink_trap_metadata *metadata);
118 };
119 
120 struct net_dm_skb_cb {
121 	union {
122 		struct devlink_trap_metadata *hw_metadata;
123 		void *pc;
124 	};
125 	enum skb_drop_reason reason;
126 };
127 
128 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
129 
130 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
131 {
132 	size_t al;
133 	struct net_dm_alert_msg *msg;
134 	struct nlattr *nla;
135 	struct sk_buff *skb;
136 	unsigned long flags;
137 	void *msg_header;
138 
139 	al = sizeof(struct net_dm_alert_msg);
140 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
141 	al += sizeof(struct nlattr);
142 
143 	skb = genlmsg_new(al, GFP_KERNEL);
144 
145 	if (!skb)
146 		goto err;
147 
148 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
149 				 0, NET_DM_CMD_ALERT);
150 	if (!msg_header) {
151 		nlmsg_free(skb);
152 		skb = NULL;
153 		goto err;
154 	}
155 	nla = nla_reserve(skb, NLA_UNSPEC,
156 			  sizeof(struct net_dm_alert_msg));
157 	if (!nla) {
158 		nlmsg_free(skb);
159 		skb = NULL;
160 		goto err;
161 	}
162 	msg = nla_data(nla);
163 	memset(msg, 0, al);
164 	goto out;
165 
166 err:
167 	mod_timer(&data->send_timer, jiffies + HZ / 10);
168 out:
169 	spin_lock_irqsave(&data->lock, flags);
170 	swap(data->skb, skb);
171 	spin_unlock_irqrestore(&data->lock, flags);
172 
173 	if (skb) {
174 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
175 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
176 
177 		genlmsg_end(skb, genlmsg_data(gnlh));
178 	}
179 
180 	return skb;
181 }
182 
183 static const struct genl_multicast_group dropmon_mcgrps[] = {
184 	{ .name = "events", },
185 };
186 
187 static void send_dm_alert(struct work_struct *work)
188 {
189 	struct sk_buff *skb;
190 	struct per_cpu_dm_data *data;
191 
192 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
193 
194 	skb = reset_per_cpu_data(data);
195 
196 	if (skb)
197 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
198 				  0, GFP_KERNEL);
199 }
200 
201 /*
202  * This is the timer function to delay the sending of an alert
203  * in the event that more drops will arrive during the
204  * hysteresis period.
205  */
206 static void sched_send_work(struct timer_list *t)
207 {
208 	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
209 
210 	schedule_work(&data->dm_alert_work);
211 }
212 
213 static void trace_drop_common(struct sk_buff *skb, void *location)
214 {
215 	struct net_dm_alert_msg *msg;
216 	struct net_dm_drop_point *point;
217 	struct nlmsghdr *nlh;
218 	struct nlattr *nla;
219 	int i;
220 	struct sk_buff *dskb;
221 	struct per_cpu_dm_data *data;
222 	unsigned long flags;
223 
224 	local_irq_save(flags);
225 	data = this_cpu_ptr(&dm_cpu_data);
226 	spin_lock(&data->lock);
227 	dskb = data->skb;
228 
229 	if (!dskb)
230 		goto out;
231 
232 	nlh = (struct nlmsghdr *)dskb->data;
233 	nla = genlmsg_data(nlmsg_data(nlh));
234 	msg = nla_data(nla);
235 	point = msg->points;
236 	for (i = 0; i < msg->entries; i++) {
237 		if (!memcmp(&location, &point->pc, sizeof(void *))) {
238 			point->count++;
239 			goto out;
240 		}
241 		point++;
242 	}
243 	if (msg->entries == dm_hit_limit)
244 		goto out;
245 	/*
246 	 * We need to create a new entry
247 	 */
248 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
249 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
250 	memcpy(point->pc, &location, sizeof(void *));
251 	point->count = 1;
252 	msg->entries++;
253 
254 	if (!timer_pending(&data->send_timer)) {
255 		data->send_timer.expires = jiffies + dm_delay * HZ;
256 		add_timer(&data->send_timer);
257 	}
258 
259 out:
260 	spin_unlock_irqrestore(&data->lock, flags);
261 }
262 
263 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb,
264 				void *location,
265 				enum skb_drop_reason reason)
266 {
267 	trace_drop_common(skb, location);
268 }
269 
270 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
271 				int work, int budget)
272 {
273 	struct net_device *dev = napi->dev;
274 	struct dm_hw_stat_delta *stat;
275 	/*
276 	 * Don't check napi structures with no associated device
277 	 */
278 	if (!dev)
279 		return;
280 
281 	rcu_read_lock();
282 	stat = rcu_dereference(dev->dm_private);
283 	if (stat) {
284 		/*
285 		 * only add a note to our monitor buffer if:
286 		 * 1) its after the last_rx delta
287 		 * 2) our rx_dropped count has gone up
288 		 */
289 		if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
290 		    (dev->stats.rx_dropped != stat->last_drop_val)) {
291 			trace_drop_common(NULL, NULL);
292 			stat->last_drop_val = dev->stats.rx_dropped;
293 			stat->last_rx = jiffies;
294 		}
295 	}
296 	rcu_read_unlock();
297 }
298 
299 static struct net_dm_hw_entries *
300 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
301 {
302 	struct net_dm_hw_entries *hw_entries;
303 	unsigned long flags;
304 
305 	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
306 			     GFP_KERNEL);
307 	if (!hw_entries) {
308 		/* If the memory allocation failed, we try to perform another
309 		 * allocation in 1/10 second. Otherwise, the probe function
310 		 * will constantly bail out.
311 		 */
312 		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
313 	}
314 
315 	spin_lock_irqsave(&hw_data->lock, flags);
316 	swap(hw_data->hw_entries, hw_entries);
317 	spin_unlock_irqrestore(&hw_data->lock, flags);
318 
319 	return hw_entries;
320 }
321 
322 static int net_dm_hw_entry_put(struct sk_buff *msg,
323 			       const struct net_dm_hw_entry *hw_entry)
324 {
325 	struct nlattr *attr;
326 
327 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
328 	if (!attr)
329 		return -EMSGSIZE;
330 
331 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
332 		goto nla_put_failure;
333 
334 	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
335 		goto nla_put_failure;
336 
337 	nla_nest_end(msg, attr);
338 
339 	return 0;
340 
341 nla_put_failure:
342 	nla_nest_cancel(msg, attr);
343 	return -EMSGSIZE;
344 }
345 
346 static int net_dm_hw_entries_put(struct sk_buff *msg,
347 				 const struct net_dm_hw_entries *hw_entries)
348 {
349 	struct nlattr *attr;
350 	int i;
351 
352 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
353 	if (!attr)
354 		return -EMSGSIZE;
355 
356 	for (i = 0; i < hw_entries->num_entries; i++) {
357 		int rc;
358 
359 		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
360 		if (rc)
361 			goto nla_put_failure;
362 	}
363 
364 	nla_nest_end(msg, attr);
365 
366 	return 0;
367 
368 nla_put_failure:
369 	nla_nest_cancel(msg, attr);
370 	return -EMSGSIZE;
371 }
372 
373 static int
374 net_dm_hw_summary_report_fill(struct sk_buff *msg,
375 			      const struct net_dm_hw_entries *hw_entries)
376 {
377 	struct net_dm_alert_msg anc_hdr = { 0 };
378 	void *hdr;
379 	int rc;
380 
381 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
382 			  NET_DM_CMD_ALERT);
383 	if (!hdr)
384 		return -EMSGSIZE;
385 
386 	/* We need to put the ancillary header in order not to break user
387 	 * space.
388 	 */
389 	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
390 		goto nla_put_failure;
391 
392 	rc = net_dm_hw_entries_put(msg, hw_entries);
393 	if (rc)
394 		goto nla_put_failure;
395 
396 	genlmsg_end(msg, hdr);
397 
398 	return 0;
399 
400 nla_put_failure:
401 	genlmsg_cancel(msg, hdr);
402 	return -EMSGSIZE;
403 }
404 
405 static void net_dm_hw_summary_work(struct work_struct *work)
406 {
407 	struct net_dm_hw_entries *hw_entries;
408 	struct per_cpu_dm_data *hw_data;
409 	struct sk_buff *msg;
410 	int rc;
411 
412 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
413 
414 	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
415 	if (!hw_entries)
416 		return;
417 
418 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
419 	if (!msg)
420 		goto out;
421 
422 	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
423 	if (rc) {
424 		nlmsg_free(msg);
425 		goto out;
426 	}
427 
428 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
429 
430 out:
431 	kfree(hw_entries);
432 }
433 
434 static void
435 net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
436 			     struct sk_buff *skb,
437 			     const struct devlink_trap_metadata *metadata)
438 {
439 	struct net_dm_hw_entries *hw_entries;
440 	struct net_dm_hw_entry *hw_entry;
441 	struct per_cpu_dm_data *hw_data;
442 	unsigned long flags;
443 	int i;
444 
445 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
446 		return;
447 
448 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
449 	spin_lock_irqsave(&hw_data->lock, flags);
450 	hw_entries = hw_data->hw_entries;
451 
452 	if (!hw_entries)
453 		goto out;
454 
455 	for (i = 0; i < hw_entries->num_entries; i++) {
456 		hw_entry = &hw_entries->entries[i];
457 		if (!strncmp(hw_entry->trap_name, metadata->trap_name,
458 			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
459 			hw_entry->count++;
460 			goto out;
461 		}
462 	}
463 	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
464 		goto out;
465 
466 	hw_entry = &hw_entries->entries[hw_entries->num_entries];
467 	strlcpy(hw_entry->trap_name, metadata->trap_name,
468 		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
469 	hw_entry->count = 1;
470 	hw_entries->num_entries++;
471 
472 	if (!timer_pending(&hw_data->send_timer)) {
473 		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
474 		add_timer(&hw_data->send_timer);
475 	}
476 
477 out:
478 	spin_unlock_irqrestore(&hw_data->lock, flags);
479 }
480 
481 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
482 	.kfree_skb_probe	= trace_kfree_skb_hit,
483 	.napi_poll_probe	= trace_napi_poll_hit,
484 	.work_item_func		= send_dm_alert,
485 	.hw_work_item_func	= net_dm_hw_summary_work,
486 	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
487 };
488 
489 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
490 					      struct sk_buff *skb,
491 					      void *location,
492 					      enum skb_drop_reason reason)
493 {
494 	ktime_t tstamp = ktime_get_real();
495 	struct per_cpu_dm_data *data;
496 	struct net_dm_skb_cb *cb;
497 	struct sk_buff *nskb;
498 	unsigned long flags;
499 
500 	if (!skb_mac_header_was_set(skb))
501 		return;
502 
503 	nskb = skb_clone(skb, GFP_ATOMIC);
504 	if (!nskb)
505 		return;
506 
507 	if (unlikely(reason >= SKB_DROP_REASON_MAX || reason <= 0))
508 		reason = SKB_DROP_REASON_NOT_SPECIFIED;
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 
556 static size_t net_dm_packet_report_size(size_t payload_len,
557 					enum skb_drop_reason reason)
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(strlen(drop_reasons[reason]) + 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 	char buf[NET_DM_MAX_SYMBOL_LEN];
614 	struct nlattr *attr;
615 	void *hdr;
616 	int rc;
617 
618 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
619 			  NET_DM_CMD_PACKET_ALERT);
620 	if (!hdr)
621 		return -EMSGSIZE;
622 
623 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
624 		goto nla_put_failure;
625 
626 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
627 			      NET_DM_ATTR_PAD))
628 		goto nla_put_failure;
629 
630 	if (nla_put_string(msg, NET_DM_ATTR_REASON,
631 			   drop_reasons[cb->reason]))
632 		goto nla_put_failure;
633 
634 	snprintf(buf, sizeof(buf), "%pS", cb->pc);
635 	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
636 		goto nla_put_failure;
637 
638 	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
639 	if (rc)
640 		goto nla_put_failure;
641 
642 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
643 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
644 		goto nla_put_failure;
645 
646 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
647 		goto nla_put_failure;
648 
649 	if (!payload_len)
650 		goto out;
651 
652 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
653 		goto nla_put_failure;
654 
655 	attr = skb_put(msg, nla_total_size(payload_len));
656 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
657 	attr->nla_len = nla_attr_size(payload_len);
658 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
659 		goto nla_put_failure;
660 
661 out:
662 	genlmsg_end(msg, hdr);
663 
664 	return 0;
665 
666 nla_put_failure:
667 	genlmsg_cancel(msg, hdr);
668 	return -EMSGSIZE;
669 }
670 
671 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
672 
673 static void net_dm_packet_report(struct sk_buff *skb)
674 {
675 	struct sk_buff *msg;
676 	size_t payload_len;
677 	int rc;
678 
679 	/* Make sure we start copying the packet from the MAC header */
680 	if (skb->data > skb_mac_header(skb))
681 		skb_push(skb, skb->data - skb_mac_header(skb));
682 	else
683 		skb_pull(skb, skb_mac_header(skb) - skb->data);
684 
685 	/* Ensure packet fits inside a single netlink attribute */
686 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
687 	if (net_dm_trunc_len)
688 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
689 
690 	msg = nlmsg_new(net_dm_packet_report_size(payload_len,
691 						  NET_DM_SKB_CB(skb)->reason),
692 			GFP_KERNEL);
693 	if (!msg)
694 		goto out;
695 
696 	rc = net_dm_packet_report_fill(msg, skb, payload_len);
697 	if (rc) {
698 		nlmsg_free(msg);
699 		goto out;
700 	}
701 
702 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
703 
704 out:
705 	consume_skb(skb);
706 }
707 
708 static void net_dm_packet_work(struct work_struct *work)
709 {
710 	struct per_cpu_dm_data *data;
711 	struct sk_buff_head list;
712 	struct sk_buff *skb;
713 	unsigned long flags;
714 
715 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
716 
717 	__skb_queue_head_init(&list);
718 
719 	spin_lock_irqsave(&data->drop_queue.lock, flags);
720 	skb_queue_splice_tail_init(&data->drop_queue, &list);
721 	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
722 
723 	while ((skb = __skb_dequeue(&list)))
724 		net_dm_packet_report(skb);
725 }
726 
727 static size_t
728 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
729 {
730 	return hw_metadata->fa_cookie ?
731 	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
732 }
733 
734 static size_t
735 net_dm_hw_packet_report_size(size_t payload_len,
736 			     const struct devlink_trap_metadata *hw_metadata)
737 {
738 	size_t size;
739 
740 	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
741 
742 	return NLMSG_ALIGN(size) +
743 	       /* NET_DM_ATTR_ORIGIN */
744 	       nla_total_size(sizeof(u16)) +
745 	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
746 	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
747 	       /* NET_DM_ATTR_HW_TRAP_NAME */
748 	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
749 	       /* NET_DM_ATTR_IN_PORT */
750 	       net_dm_in_port_size() +
751 	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
752 	       net_dm_flow_action_cookie_size(hw_metadata) +
753 	       /* NET_DM_ATTR_TIMESTAMP */
754 	       nla_total_size(sizeof(u64)) +
755 	       /* NET_DM_ATTR_ORIG_LEN */
756 	       nla_total_size(sizeof(u32)) +
757 	       /* NET_DM_ATTR_PROTO */
758 	       nla_total_size(sizeof(u16)) +
759 	       /* NET_DM_ATTR_PAYLOAD */
760 	       nla_total_size(payload_len);
761 }
762 
763 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
764 					struct sk_buff *skb, size_t payload_len)
765 {
766 	struct devlink_trap_metadata *hw_metadata;
767 	struct nlattr *attr;
768 	void *hdr;
769 
770 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
771 
772 	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
773 			  NET_DM_CMD_PACKET_ALERT);
774 	if (!hdr)
775 		return -EMSGSIZE;
776 
777 	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
778 		goto nla_put_failure;
779 
780 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
781 			   hw_metadata->trap_group_name))
782 		goto nla_put_failure;
783 
784 	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
785 			   hw_metadata->trap_name))
786 		goto nla_put_failure;
787 
788 	if (hw_metadata->input_dev) {
789 		struct net_device *dev = hw_metadata->input_dev;
790 		int rc;
791 
792 		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
793 						      dev->name);
794 		if (rc)
795 			goto nla_put_failure;
796 	}
797 
798 	if (hw_metadata->fa_cookie &&
799 	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
800 		    hw_metadata->fa_cookie->cookie_len,
801 		    hw_metadata->fa_cookie->cookie))
802 		goto nla_put_failure;
803 
804 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
805 			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
806 		goto nla_put_failure;
807 
808 	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
809 		goto nla_put_failure;
810 
811 	if (!payload_len)
812 		goto out;
813 
814 	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
815 		goto nla_put_failure;
816 
817 	attr = skb_put(msg, nla_total_size(payload_len));
818 	attr->nla_type = NET_DM_ATTR_PAYLOAD;
819 	attr->nla_len = nla_attr_size(payload_len);
820 	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
821 		goto nla_put_failure;
822 
823 out:
824 	genlmsg_end(msg, hdr);
825 
826 	return 0;
827 
828 nla_put_failure:
829 	genlmsg_cancel(msg, hdr);
830 	return -EMSGSIZE;
831 }
832 
833 static struct devlink_trap_metadata *
834 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
835 {
836 	const struct flow_action_cookie *fa_cookie;
837 	struct devlink_trap_metadata *hw_metadata;
838 	const char *trap_group_name;
839 	const char *trap_name;
840 
841 	hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
842 	if (!hw_metadata)
843 		return NULL;
844 
845 	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
846 	if (!trap_group_name)
847 		goto free_hw_metadata;
848 	hw_metadata->trap_group_name = trap_group_name;
849 
850 	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
851 	if (!trap_name)
852 		goto free_trap_group;
853 	hw_metadata->trap_name = trap_name;
854 
855 	if (metadata->fa_cookie) {
856 		size_t cookie_size = sizeof(*fa_cookie) +
857 				     metadata->fa_cookie->cookie_len;
858 
859 		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
860 				    GFP_ATOMIC);
861 		if (!fa_cookie)
862 			goto free_trap_name;
863 		hw_metadata->fa_cookie = fa_cookie;
864 	}
865 
866 	hw_metadata->input_dev = metadata->input_dev;
867 	netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
868 		    GFP_ATOMIC);
869 
870 	return hw_metadata;
871 
872 free_trap_name:
873 	kfree(trap_name);
874 free_trap_group:
875 	kfree(trap_group_name);
876 free_hw_metadata:
877 	kfree(hw_metadata);
878 	return NULL;
879 }
880 
881 static void
882 net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
883 {
884 	netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
885 	kfree(hw_metadata->fa_cookie);
886 	kfree(hw_metadata->trap_name);
887 	kfree(hw_metadata->trap_group_name);
888 	kfree(hw_metadata);
889 }
890 
891 static void net_dm_hw_packet_report(struct sk_buff *skb)
892 {
893 	struct devlink_trap_metadata *hw_metadata;
894 	struct sk_buff *msg;
895 	size_t payload_len;
896 	int rc;
897 
898 	if (skb->data > skb_mac_header(skb))
899 		skb_push(skb, skb->data - skb_mac_header(skb));
900 	else
901 		skb_pull(skb, skb_mac_header(skb) - skb->data);
902 
903 	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
904 	if (net_dm_trunc_len)
905 		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
906 
907 	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
908 	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
909 			GFP_KERNEL);
910 	if (!msg)
911 		goto out;
912 
913 	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
914 	if (rc) {
915 		nlmsg_free(msg);
916 		goto out;
917 	}
918 
919 	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
920 
921 out:
922 	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
923 	consume_skb(skb);
924 }
925 
926 static void net_dm_hw_packet_work(struct work_struct *work)
927 {
928 	struct per_cpu_dm_data *hw_data;
929 	struct sk_buff_head list;
930 	struct sk_buff *skb;
931 	unsigned long flags;
932 
933 	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
934 
935 	__skb_queue_head_init(&list);
936 
937 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
938 	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
939 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
940 
941 	while ((skb = __skb_dequeue(&list)))
942 		net_dm_hw_packet_report(skb);
943 }
944 
945 static void
946 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
947 			    struct sk_buff *skb,
948 			    const struct devlink_trap_metadata *metadata)
949 {
950 	struct devlink_trap_metadata *n_hw_metadata;
951 	ktime_t tstamp = ktime_get_real();
952 	struct per_cpu_dm_data *hw_data;
953 	struct sk_buff *nskb;
954 	unsigned long flags;
955 
956 	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
957 		return;
958 
959 	if (!skb_mac_header_was_set(skb))
960 		return;
961 
962 	nskb = skb_clone(skb, GFP_ATOMIC);
963 	if (!nskb)
964 		return;
965 
966 	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
967 	if (!n_hw_metadata)
968 		goto free;
969 
970 	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
971 	nskb->tstamp = tstamp;
972 
973 	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
974 
975 	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
976 	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
977 		__skb_queue_tail(&hw_data->drop_queue, nskb);
978 	else
979 		goto unlock_free;
980 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
981 
982 	schedule_work(&hw_data->dm_alert_work);
983 
984 	return;
985 
986 unlock_free:
987 	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
988 	u64_stats_update_begin(&hw_data->stats.syncp);
989 	u64_stats_inc(&hw_data->stats.dropped);
990 	u64_stats_update_end(&hw_data->stats.syncp);
991 	net_dm_hw_metadata_free(n_hw_metadata);
992 free:
993 	consume_skb(nskb);
994 }
995 
996 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
997 	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
998 	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
999 	.work_item_func		= net_dm_packet_work,
1000 	.hw_work_item_func	= net_dm_hw_packet_work,
1001 	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
1002 };
1003 
1004 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1005 	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
1006 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
1007 };
1008 
1009 #if IS_ENABLED(CONFIG_NET_DEVLINK)
1010 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1011 {
1012 	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1013 }
1014 
1015 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1016 {
1017 	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1018 	tracepoint_synchronize_unregister();
1019 }
1020 #else
1021 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1022 {
1023 	return -EOPNOTSUPP;
1024 }
1025 
1026 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1027 {
1028 }
1029 #endif
1030 
1031 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1032 {
1033 	const struct net_dm_alert_ops *ops;
1034 	int cpu, rc;
1035 
1036 	if (monitor_hw) {
1037 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1038 		return -EAGAIN;
1039 	}
1040 
1041 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1042 
1043 	if (!try_module_get(THIS_MODULE)) {
1044 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1045 		return -ENODEV;
1046 	}
1047 
1048 	for_each_possible_cpu(cpu) {
1049 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1050 		struct net_dm_hw_entries *hw_entries;
1051 
1052 		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1053 		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1054 		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1055 		kfree(hw_entries);
1056 	}
1057 
1058 	rc = net_dm_hw_probe_register(ops);
1059 	if (rc) {
1060 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1061 		goto err_module_put;
1062 	}
1063 
1064 	monitor_hw = true;
1065 
1066 	return 0;
1067 
1068 err_module_put:
1069 	for_each_possible_cpu(cpu) {
1070 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1071 		struct sk_buff *skb;
1072 
1073 		del_timer_sync(&hw_data->send_timer);
1074 		cancel_work_sync(&hw_data->dm_alert_work);
1075 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1076 			struct devlink_trap_metadata *hw_metadata;
1077 
1078 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1079 			net_dm_hw_metadata_free(hw_metadata);
1080 			consume_skb(skb);
1081 		}
1082 	}
1083 	module_put(THIS_MODULE);
1084 	return rc;
1085 }
1086 
1087 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1088 {
1089 	const struct net_dm_alert_ops *ops;
1090 	int cpu;
1091 
1092 	if (!monitor_hw) {
1093 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1094 		return;
1095 	}
1096 
1097 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1098 
1099 	monitor_hw = false;
1100 
1101 	net_dm_hw_probe_unregister(ops);
1102 
1103 	for_each_possible_cpu(cpu) {
1104 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1105 		struct sk_buff *skb;
1106 
1107 		del_timer_sync(&hw_data->send_timer);
1108 		cancel_work_sync(&hw_data->dm_alert_work);
1109 		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1110 			struct devlink_trap_metadata *hw_metadata;
1111 
1112 			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1113 			net_dm_hw_metadata_free(hw_metadata);
1114 			consume_skb(skb);
1115 		}
1116 	}
1117 
1118 	module_put(THIS_MODULE);
1119 }
1120 
1121 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1122 {
1123 	const struct net_dm_alert_ops *ops;
1124 	int cpu, rc;
1125 
1126 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1127 
1128 	if (!try_module_get(THIS_MODULE)) {
1129 		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1130 		return -ENODEV;
1131 	}
1132 
1133 	for_each_possible_cpu(cpu) {
1134 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1135 		struct sk_buff *skb;
1136 
1137 		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1138 		timer_setup(&data->send_timer, sched_send_work, 0);
1139 		/* Allocate a new per-CPU skb for the summary alert message and
1140 		 * free the old one which might contain stale data from
1141 		 * previous tracing.
1142 		 */
1143 		skb = reset_per_cpu_data(data);
1144 		consume_skb(skb);
1145 	}
1146 
1147 	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1148 	if (rc) {
1149 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1150 		goto err_module_put;
1151 	}
1152 
1153 	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1154 	if (rc) {
1155 		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1156 		goto err_unregister_trace;
1157 	}
1158 
1159 	return 0;
1160 
1161 err_unregister_trace:
1162 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1163 err_module_put:
1164 	for_each_possible_cpu(cpu) {
1165 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1166 		struct sk_buff *skb;
1167 
1168 		del_timer_sync(&data->send_timer);
1169 		cancel_work_sync(&data->dm_alert_work);
1170 		while ((skb = __skb_dequeue(&data->drop_queue)))
1171 			consume_skb(skb);
1172 	}
1173 	module_put(THIS_MODULE);
1174 	return rc;
1175 }
1176 
1177 static void net_dm_trace_off_set(void)
1178 {
1179 	const struct net_dm_alert_ops *ops;
1180 	int cpu;
1181 
1182 	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1183 
1184 	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1185 	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1186 
1187 	tracepoint_synchronize_unregister();
1188 
1189 	/* Make sure we do not send notifications to user space after request
1190 	 * to stop tracing returns.
1191 	 */
1192 	for_each_possible_cpu(cpu) {
1193 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1194 		struct sk_buff *skb;
1195 
1196 		del_timer_sync(&data->send_timer);
1197 		cancel_work_sync(&data->dm_alert_work);
1198 		while ((skb = __skb_dequeue(&data->drop_queue)))
1199 			consume_skb(skb);
1200 	}
1201 
1202 	module_put(THIS_MODULE);
1203 }
1204 
1205 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1206 {
1207 	int rc = 0;
1208 
1209 	if (state == trace_state) {
1210 		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1211 		return -EAGAIN;
1212 	}
1213 
1214 	switch (state) {
1215 	case TRACE_ON:
1216 		rc = net_dm_trace_on_set(extack);
1217 		break;
1218 	case TRACE_OFF:
1219 		net_dm_trace_off_set();
1220 		break;
1221 	default:
1222 		rc = 1;
1223 		break;
1224 	}
1225 
1226 	if (!rc)
1227 		trace_state = state;
1228 	else
1229 		rc = -EINPROGRESS;
1230 
1231 	return rc;
1232 }
1233 
1234 static bool net_dm_is_monitoring(void)
1235 {
1236 	return trace_state == TRACE_ON || monitor_hw;
1237 }
1238 
1239 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1240 					   enum net_dm_alert_mode *p_alert_mode)
1241 {
1242 	u8 val;
1243 
1244 	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1245 
1246 	switch (val) {
1247 	case NET_DM_ALERT_MODE_SUMMARY:
1248 	case NET_DM_ALERT_MODE_PACKET:
1249 		*p_alert_mode = val;
1250 		break;
1251 	default:
1252 		return -EINVAL;
1253 	}
1254 
1255 	return 0;
1256 }
1257 
1258 static int net_dm_alert_mode_set(struct genl_info *info)
1259 {
1260 	struct netlink_ext_ack *extack = info->extack;
1261 	enum net_dm_alert_mode alert_mode;
1262 	int rc;
1263 
1264 	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1265 		return 0;
1266 
1267 	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1268 	if (rc) {
1269 		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1270 		return -EINVAL;
1271 	}
1272 
1273 	net_dm_alert_mode = alert_mode;
1274 
1275 	return 0;
1276 }
1277 
1278 static void net_dm_trunc_len_set(struct genl_info *info)
1279 {
1280 	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1281 		return;
1282 
1283 	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1284 }
1285 
1286 static void net_dm_queue_len_set(struct genl_info *info)
1287 {
1288 	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1289 		return;
1290 
1291 	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1292 }
1293 
1294 static int net_dm_cmd_config(struct sk_buff *skb,
1295 			struct genl_info *info)
1296 {
1297 	struct netlink_ext_ack *extack = info->extack;
1298 	int rc;
1299 
1300 	if (net_dm_is_monitoring()) {
1301 		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1302 		return -EBUSY;
1303 	}
1304 
1305 	rc = net_dm_alert_mode_set(info);
1306 	if (rc)
1307 		return rc;
1308 
1309 	net_dm_trunc_len_set(info);
1310 
1311 	net_dm_queue_len_set(info);
1312 
1313 	return 0;
1314 }
1315 
1316 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1317 				struct netlink_ext_ack *extack)
1318 {
1319 	bool sw_set = false;
1320 	int rc;
1321 
1322 	if (set_sw) {
1323 		rc = set_all_monitor_traces(TRACE_ON, extack);
1324 		if (rc)
1325 			return rc;
1326 		sw_set = true;
1327 	}
1328 
1329 	if (set_hw) {
1330 		rc = net_dm_hw_monitor_start(extack);
1331 		if (rc)
1332 			goto err_monitor_hw;
1333 	}
1334 
1335 	return 0;
1336 
1337 err_monitor_hw:
1338 	if (sw_set)
1339 		set_all_monitor_traces(TRACE_OFF, extack);
1340 	return rc;
1341 }
1342 
1343 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1344 				struct netlink_ext_ack *extack)
1345 {
1346 	if (set_hw)
1347 		net_dm_hw_monitor_stop(extack);
1348 	if (set_sw)
1349 		set_all_monitor_traces(TRACE_OFF, extack);
1350 }
1351 
1352 static int net_dm_cmd_trace(struct sk_buff *skb,
1353 			struct genl_info *info)
1354 {
1355 	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1356 	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1357 	struct netlink_ext_ack *extack = info->extack;
1358 
1359 	/* To maintain backward compatibility, we start / stop monitoring of
1360 	 * software drops if no flag is specified.
1361 	 */
1362 	if (!set_sw && !set_hw)
1363 		set_sw = true;
1364 
1365 	switch (info->genlhdr->cmd) {
1366 	case NET_DM_CMD_START:
1367 		return net_dm_monitor_start(set_sw, set_hw, extack);
1368 	case NET_DM_CMD_STOP:
1369 		net_dm_monitor_stop(set_sw, set_hw, extack);
1370 		return 0;
1371 	}
1372 
1373 	return -EOPNOTSUPP;
1374 }
1375 
1376 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1377 {
1378 	void *hdr;
1379 
1380 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1381 			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1382 	if (!hdr)
1383 		return -EMSGSIZE;
1384 
1385 	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1386 		goto nla_put_failure;
1387 
1388 	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1389 		goto nla_put_failure;
1390 
1391 	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1392 		goto nla_put_failure;
1393 
1394 	genlmsg_end(msg, hdr);
1395 
1396 	return 0;
1397 
1398 nla_put_failure:
1399 	genlmsg_cancel(msg, hdr);
1400 	return -EMSGSIZE;
1401 }
1402 
1403 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1404 {
1405 	struct sk_buff *msg;
1406 	int rc;
1407 
1408 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1409 	if (!msg)
1410 		return -ENOMEM;
1411 
1412 	rc = net_dm_config_fill(msg, info);
1413 	if (rc)
1414 		goto free_msg;
1415 
1416 	return genlmsg_reply(msg, info);
1417 
1418 free_msg:
1419 	nlmsg_free(msg);
1420 	return rc;
1421 }
1422 
1423 static void net_dm_stats_read(struct net_dm_stats *stats)
1424 {
1425 	int cpu;
1426 
1427 	memset(stats, 0, sizeof(*stats));
1428 	for_each_possible_cpu(cpu) {
1429 		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1430 		struct net_dm_stats *cpu_stats = &data->stats;
1431 		unsigned int start;
1432 		u64 dropped;
1433 
1434 		do {
1435 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1436 			dropped = u64_stats_read(&cpu_stats->dropped);
1437 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1438 
1439 		u64_stats_add(&stats->dropped, dropped);
1440 	}
1441 }
1442 
1443 static int net_dm_stats_put(struct sk_buff *msg)
1444 {
1445 	struct net_dm_stats stats;
1446 	struct nlattr *attr;
1447 
1448 	net_dm_stats_read(&stats);
1449 
1450 	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1451 	if (!attr)
1452 		return -EMSGSIZE;
1453 
1454 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1455 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1456 		goto nla_put_failure;
1457 
1458 	nla_nest_end(msg, attr);
1459 
1460 	return 0;
1461 
1462 nla_put_failure:
1463 	nla_nest_cancel(msg, attr);
1464 	return -EMSGSIZE;
1465 }
1466 
1467 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1468 {
1469 	int cpu;
1470 
1471 	memset(stats, 0, sizeof(*stats));
1472 	for_each_possible_cpu(cpu) {
1473 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1474 		struct net_dm_stats *cpu_stats = &hw_data->stats;
1475 		unsigned int start;
1476 		u64 dropped;
1477 
1478 		do {
1479 			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1480 			dropped = u64_stats_read(&cpu_stats->dropped);
1481 		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1482 
1483 		u64_stats_add(&stats->dropped, dropped);
1484 	}
1485 }
1486 
1487 static int net_dm_hw_stats_put(struct sk_buff *msg)
1488 {
1489 	struct net_dm_stats stats;
1490 	struct nlattr *attr;
1491 
1492 	net_dm_hw_stats_read(&stats);
1493 
1494 	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1495 	if (!attr)
1496 		return -EMSGSIZE;
1497 
1498 	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1499 			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1500 		goto nla_put_failure;
1501 
1502 	nla_nest_end(msg, attr);
1503 
1504 	return 0;
1505 
1506 nla_put_failure:
1507 	nla_nest_cancel(msg, attr);
1508 	return -EMSGSIZE;
1509 }
1510 
1511 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1512 {
1513 	void *hdr;
1514 	int rc;
1515 
1516 	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1517 			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1518 	if (!hdr)
1519 		return -EMSGSIZE;
1520 
1521 	rc = net_dm_stats_put(msg);
1522 	if (rc)
1523 		goto nla_put_failure;
1524 
1525 	rc = net_dm_hw_stats_put(msg);
1526 	if (rc)
1527 		goto nla_put_failure;
1528 
1529 	genlmsg_end(msg, hdr);
1530 
1531 	return 0;
1532 
1533 nla_put_failure:
1534 	genlmsg_cancel(msg, hdr);
1535 	return -EMSGSIZE;
1536 }
1537 
1538 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1539 {
1540 	struct sk_buff *msg;
1541 	int rc;
1542 
1543 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1544 	if (!msg)
1545 		return -ENOMEM;
1546 
1547 	rc = net_dm_stats_fill(msg, info);
1548 	if (rc)
1549 		goto free_msg;
1550 
1551 	return genlmsg_reply(msg, info);
1552 
1553 free_msg:
1554 	nlmsg_free(msg);
1555 	return rc;
1556 }
1557 
1558 static int dropmon_net_event(struct notifier_block *ev_block,
1559 			     unsigned long event, void *ptr)
1560 {
1561 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1562 	struct dm_hw_stat_delta *stat;
1563 
1564 	switch (event) {
1565 	case NETDEV_REGISTER:
1566 		if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1567 			break;
1568 		stat = kzalloc(sizeof(*stat), GFP_KERNEL);
1569 		if (!stat)
1570 			break;
1571 
1572 		stat->last_rx = jiffies;
1573 		rcu_assign_pointer(dev->dm_private, stat);
1574 
1575 		break;
1576 	case NETDEV_UNREGISTER:
1577 		stat = rtnl_dereference(dev->dm_private);
1578 		if (stat) {
1579 			rcu_assign_pointer(dev->dm_private, NULL);
1580 			kfree_rcu(stat, rcu);
1581 		}
1582 		break;
1583 	}
1584 	return NOTIFY_DONE;
1585 }
1586 
1587 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1588 	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1589 	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1590 	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1591 	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1592 	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1593 	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1594 };
1595 
1596 static const struct genl_small_ops dropmon_ops[] = {
1597 	{
1598 		.cmd = NET_DM_CMD_CONFIG,
1599 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1600 		.doit = net_dm_cmd_config,
1601 		.flags = GENL_ADMIN_PERM,
1602 	},
1603 	{
1604 		.cmd = NET_DM_CMD_START,
1605 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1606 		.doit = net_dm_cmd_trace,
1607 	},
1608 	{
1609 		.cmd = NET_DM_CMD_STOP,
1610 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1611 		.doit = net_dm_cmd_trace,
1612 	},
1613 	{
1614 		.cmd = NET_DM_CMD_CONFIG_GET,
1615 		.doit = net_dm_cmd_config_get,
1616 	},
1617 	{
1618 		.cmd = NET_DM_CMD_STATS_GET,
1619 		.doit = net_dm_cmd_stats_get,
1620 	},
1621 };
1622 
1623 static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1624 			      struct sk_buff *skb, struct genl_info *info)
1625 {
1626 	mutex_lock(&net_dm_mutex);
1627 
1628 	return 0;
1629 }
1630 
1631 static void net_dm_nl_post_doit(const struct genl_ops *ops,
1632 				struct sk_buff *skb, struct genl_info *info)
1633 {
1634 	mutex_unlock(&net_dm_mutex);
1635 }
1636 
1637 static struct genl_family net_drop_monitor_family __ro_after_init = {
1638 	.hdrsize        = 0,
1639 	.name           = "NET_DM",
1640 	.version        = 2,
1641 	.maxattr	= NET_DM_ATTR_MAX,
1642 	.policy		= net_dm_nl_policy,
1643 	.pre_doit	= net_dm_nl_pre_doit,
1644 	.post_doit	= net_dm_nl_post_doit,
1645 	.module		= THIS_MODULE,
1646 	.small_ops	= dropmon_ops,
1647 	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1648 	.mcgrps		= dropmon_mcgrps,
1649 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1650 };
1651 
1652 static struct notifier_block dropmon_net_notifier = {
1653 	.notifier_call = dropmon_net_event
1654 };
1655 
1656 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1657 {
1658 	spin_lock_init(&data->lock);
1659 	skb_queue_head_init(&data->drop_queue);
1660 	u64_stats_init(&data->stats.syncp);
1661 }
1662 
1663 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1664 {
1665 	WARN_ON(!skb_queue_empty(&data->drop_queue));
1666 }
1667 
1668 static void net_dm_cpu_data_init(int cpu)
1669 {
1670 	struct per_cpu_dm_data *data;
1671 
1672 	data = &per_cpu(dm_cpu_data, cpu);
1673 	__net_dm_cpu_data_init(data);
1674 }
1675 
1676 static void net_dm_cpu_data_fini(int cpu)
1677 {
1678 	struct per_cpu_dm_data *data;
1679 
1680 	data = &per_cpu(dm_cpu_data, cpu);
1681 	/* At this point, we should have exclusive access
1682 	 * to this struct and can free the skb inside it.
1683 	 */
1684 	consume_skb(data->skb);
1685 	__net_dm_cpu_data_fini(data);
1686 }
1687 
1688 static void net_dm_hw_cpu_data_init(int cpu)
1689 {
1690 	struct per_cpu_dm_data *hw_data;
1691 
1692 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1693 	__net_dm_cpu_data_init(hw_data);
1694 }
1695 
1696 static void net_dm_hw_cpu_data_fini(int cpu)
1697 {
1698 	struct per_cpu_dm_data *hw_data;
1699 
1700 	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1701 	kfree(hw_data->hw_entries);
1702 	__net_dm_cpu_data_fini(hw_data);
1703 }
1704 
1705 static int __init init_net_drop_monitor(void)
1706 {
1707 	int cpu, rc;
1708 
1709 	pr_info("Initializing network drop monitor service\n");
1710 
1711 	if (sizeof(void *) > 8) {
1712 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1713 		return -ENOSPC;
1714 	}
1715 
1716 	rc = genl_register_family(&net_drop_monitor_family);
1717 	if (rc) {
1718 		pr_err("Could not create drop monitor netlink family\n");
1719 		return rc;
1720 	}
1721 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1722 
1723 	rc = register_netdevice_notifier(&dropmon_net_notifier);
1724 	if (rc < 0) {
1725 		pr_crit("Failed to register netdevice notifier\n");
1726 		goto out_unreg;
1727 	}
1728 
1729 	rc = 0;
1730 
1731 	for_each_possible_cpu(cpu) {
1732 		net_dm_cpu_data_init(cpu);
1733 		net_dm_hw_cpu_data_init(cpu);
1734 	}
1735 
1736 	goto out;
1737 
1738 out_unreg:
1739 	genl_unregister_family(&net_drop_monitor_family);
1740 out:
1741 	return rc;
1742 }
1743 
1744 static void exit_net_drop_monitor(void)
1745 {
1746 	int cpu;
1747 
1748 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1749 
1750 	/*
1751 	 * Because of the module_get/put we do in the trace state change path
1752 	 * we are guaranteed not to have any current users when we get here
1753 	 */
1754 
1755 	for_each_possible_cpu(cpu) {
1756 		net_dm_hw_cpu_data_fini(cpu);
1757 		net_dm_cpu_data_fini(cpu);
1758 	}
1759 
1760 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1761 }
1762 
1763 module_init(init_net_drop_monitor);
1764 module_exit(exit_net_drop_monitor);
1765 
1766 MODULE_LICENSE("GPL v2");
1767 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1768 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1769 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1770