Loading...
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 */
50static int trace_state = TRACE_OFF;
51static bool monitor_hw;
52
53/* net_dm_mutex
54 *
55 * An overall lock guarding every operation coming from userspace.
56 */
57static DEFINE_MUTEX(net_dm_mutex);
58
59struct 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
66struct net_dm_hw_entry {
67 char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
68 u32 count;
69};
70
71struct net_dm_hw_entries {
72 u32 num_entries;
73 struct net_dm_hw_entry entries[];
74};
75
76struct 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
90struct dm_hw_stat_delta {
91 unsigned long last_rx;
92 unsigned long last_drop_val;
93 struct rcu_head rcu;
94};
95
96static struct genl_family net_drop_monitor_family;
97
98static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100
101static int dm_hit_limit = 64;
102static int dm_delay = 1;
103static unsigned long dm_hw_check_delta = 2*HZ;
104
105static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106static u32 net_dm_trunc_len;
107static u32 net_dm_queue_len = 1000;
108
109struct 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
123struct 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
133static 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
169err:
170 mod_timer(&data->send_timer, jiffies + HZ / 10);
171out:
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
186static const struct genl_multicast_group dropmon_mcgrps[] = {
187 { .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, },
188};
189
190static 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 */
209static void sched_send_work(struct timer_list *t)
210{
211 struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
212
213 schedule_work(&data->dm_alert_work);
214}
215
216static 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
262out:
263 raw_spin_unlock_irqrestore(&data->lock, flags);
264}
265
266static 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
274static 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
303static struct net_dm_hw_entries *
304net_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(struct_size(hw_entries, entries, dm_hit_limit),
310 GFP_KERNEL);
311 if (!hw_entries) {
312 /* If the memory allocation failed, we try to perform another
313 * allocation in 1/10 second. Otherwise, the probe function
314 * will constantly bail out.
315 */
316 mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
317 }
318
319 raw_spin_lock_irqsave(&hw_data->lock, flags);
320 swap(hw_data->hw_entries, hw_entries);
321 raw_spin_unlock_irqrestore(&hw_data->lock, flags);
322
323 return hw_entries;
324}
325
326static int net_dm_hw_entry_put(struct sk_buff *msg,
327 const struct net_dm_hw_entry *hw_entry)
328{
329 struct nlattr *attr;
330
331 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
332 if (!attr)
333 return -EMSGSIZE;
334
335 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
336 goto nla_put_failure;
337
338 if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
339 goto nla_put_failure;
340
341 nla_nest_end(msg, attr);
342
343 return 0;
344
345nla_put_failure:
346 nla_nest_cancel(msg, attr);
347 return -EMSGSIZE;
348}
349
350static int net_dm_hw_entries_put(struct sk_buff *msg,
351 const struct net_dm_hw_entries *hw_entries)
352{
353 struct nlattr *attr;
354 int i;
355
356 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
357 if (!attr)
358 return -EMSGSIZE;
359
360 for (i = 0; i < hw_entries->num_entries; i++) {
361 int rc;
362
363 rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
364 if (rc)
365 goto nla_put_failure;
366 }
367
368 nla_nest_end(msg, attr);
369
370 return 0;
371
372nla_put_failure:
373 nla_nest_cancel(msg, attr);
374 return -EMSGSIZE;
375}
376
377static int
378net_dm_hw_summary_report_fill(struct sk_buff *msg,
379 const struct net_dm_hw_entries *hw_entries)
380{
381 struct net_dm_alert_msg anc_hdr = { 0 };
382 void *hdr;
383 int rc;
384
385 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
386 NET_DM_CMD_ALERT);
387 if (!hdr)
388 return -EMSGSIZE;
389
390 /* We need to put the ancillary header in order not to break user
391 * space.
392 */
393 if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
394 goto nla_put_failure;
395
396 rc = net_dm_hw_entries_put(msg, hw_entries);
397 if (rc)
398 goto nla_put_failure;
399
400 genlmsg_end(msg, hdr);
401
402 return 0;
403
404nla_put_failure:
405 genlmsg_cancel(msg, hdr);
406 return -EMSGSIZE;
407}
408
409static void net_dm_hw_summary_work(struct work_struct *work)
410{
411 struct net_dm_hw_entries *hw_entries;
412 struct per_cpu_dm_data *hw_data;
413 struct sk_buff *msg;
414 int rc;
415
416 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
417
418 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
419 if (!hw_entries)
420 return;
421
422 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
423 if (!msg)
424 goto out;
425
426 rc = net_dm_hw_summary_report_fill(msg, hw_entries);
427 if (rc) {
428 nlmsg_free(msg);
429 goto out;
430 }
431
432 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
433
434out:
435 kfree(hw_entries);
436}
437
438static void
439net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
440 struct sk_buff *skb,
441 const struct devlink_trap_metadata *metadata)
442{
443 struct net_dm_hw_entries *hw_entries;
444 struct net_dm_hw_entry *hw_entry;
445 struct per_cpu_dm_data *hw_data;
446 unsigned long flags;
447 int i;
448
449 if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
450 return;
451
452 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
453 raw_spin_lock_irqsave(&hw_data->lock, flags);
454 hw_entries = hw_data->hw_entries;
455
456 if (!hw_entries)
457 goto out;
458
459 for (i = 0; i < hw_entries->num_entries; i++) {
460 hw_entry = &hw_entries->entries[i];
461 if (!strncmp(hw_entry->trap_name, metadata->trap_name,
462 NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
463 hw_entry->count++;
464 goto out;
465 }
466 }
467 if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
468 goto out;
469
470 hw_entry = &hw_entries->entries[hw_entries->num_entries];
471 strscpy(hw_entry->trap_name, metadata->trap_name,
472 NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
473 hw_entry->count = 1;
474 hw_entries->num_entries++;
475
476 if (!timer_pending(&hw_data->send_timer)) {
477 hw_data->send_timer.expires = jiffies + dm_delay * HZ;
478 add_timer(&hw_data->send_timer);
479 }
480
481out:
482 raw_spin_unlock_irqrestore(&hw_data->lock, flags);
483}
484
485static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
486 .kfree_skb_probe = trace_kfree_skb_hit,
487 .napi_poll_probe = trace_napi_poll_hit,
488 .work_item_func = send_dm_alert,
489 .hw_work_item_func = net_dm_hw_summary_work,
490 .hw_trap_probe = net_dm_hw_trap_summary_probe,
491};
492
493static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
494 struct sk_buff *skb,
495 void *location,
496 enum skb_drop_reason reason,
497 struct sock *rx_sk)
498{
499 ktime_t tstamp = ktime_get_real();
500 struct per_cpu_dm_data *data;
501 struct net_dm_skb_cb *cb;
502 struct sk_buff *nskb;
503 unsigned long flags;
504
505 if (!skb_mac_header_was_set(skb))
506 return;
507
508 nskb = skb_clone(skb, GFP_ATOMIC);
509 if (!nskb)
510 return;
511
512 cb = NET_DM_SKB_CB(nskb);
513 cb->reason = reason;
514 cb->pc = location;
515 /* Override the timestamp because we care about the time when the
516 * packet was dropped.
517 */
518 nskb->tstamp = tstamp;
519
520 data = this_cpu_ptr(&dm_cpu_data);
521
522 spin_lock_irqsave(&data->drop_queue.lock, flags);
523 if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
524 __skb_queue_tail(&data->drop_queue, nskb);
525 else
526 goto unlock_free;
527 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
528
529 schedule_work(&data->dm_alert_work);
530
531 return;
532
533unlock_free:
534 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
535 u64_stats_update_begin(&data->stats.syncp);
536 u64_stats_inc(&data->stats.dropped);
537 u64_stats_update_end(&data->stats.syncp);
538 consume_skb(nskb);
539}
540
541static void net_dm_packet_trace_napi_poll_hit(void *ignore,
542 struct napi_struct *napi,
543 int work, int budget)
544{
545}
546
547static size_t net_dm_in_port_size(void)
548{
549 /* NET_DM_ATTR_IN_PORT nest */
550 return nla_total_size(0) +
551 /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
552 nla_total_size(sizeof(u32)) +
553 /* NET_DM_ATTR_PORT_NETDEV_NAME */
554 nla_total_size(IFNAMSIZ + 1);
555}
556
557#define NET_DM_MAX_SYMBOL_LEN 40
558#define NET_DM_MAX_REASON_LEN 50
559
560static size_t net_dm_packet_report_size(size_t payload_len)
561{
562 size_t size;
563
564 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
565
566 return NLMSG_ALIGN(size) +
567 /* NET_DM_ATTR_ORIGIN */
568 nla_total_size(sizeof(u16)) +
569 /* NET_DM_ATTR_PC */
570 nla_total_size(sizeof(u64)) +
571 /* NET_DM_ATTR_SYMBOL */
572 nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
573 /* NET_DM_ATTR_IN_PORT */
574 net_dm_in_port_size() +
575 /* NET_DM_ATTR_TIMESTAMP */
576 nla_total_size(sizeof(u64)) +
577 /* NET_DM_ATTR_ORIG_LEN */
578 nla_total_size(sizeof(u32)) +
579 /* NET_DM_ATTR_PROTO */
580 nla_total_size(sizeof(u16)) +
581 /* NET_DM_ATTR_REASON */
582 nla_total_size(NET_DM_MAX_REASON_LEN + 1) +
583 /* NET_DM_ATTR_PAYLOAD */
584 nla_total_size(payload_len);
585}
586
587static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
588 const char *name)
589{
590 struct nlattr *attr;
591
592 attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
593 if (!attr)
594 return -EMSGSIZE;
595
596 if (ifindex &&
597 nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
598 goto nla_put_failure;
599
600 if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
601 goto nla_put_failure;
602
603 nla_nest_end(msg, attr);
604
605 return 0;
606
607nla_put_failure:
608 nla_nest_cancel(msg, attr);
609 return -EMSGSIZE;
610}
611
612static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
613 size_t payload_len)
614{
615 struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb);
616 const struct drop_reason_list *list = NULL;
617 unsigned int subsys, subsys_reason;
618 char buf[NET_DM_MAX_SYMBOL_LEN];
619 struct nlattr *attr;
620 void *hdr;
621 int rc;
622
623 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
624 NET_DM_CMD_PACKET_ALERT);
625 if (!hdr)
626 return -EMSGSIZE;
627
628 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
629 goto nla_put_failure;
630
631 if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
632 NET_DM_ATTR_PAD))
633 goto nla_put_failure;
634
635 rcu_read_lock();
636 subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK);
637 if (subsys < SKB_DROP_REASON_SUBSYS_NUM)
638 list = rcu_dereference(drop_reasons_by_subsys[subsys]);
639 subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK;
640 if (!list ||
641 subsys_reason >= list->n_reasons ||
642 !list->reasons[subsys_reason] ||
643 strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) {
644 list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]);
645 subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED;
646 }
647 if (nla_put_string(msg, NET_DM_ATTR_REASON,
648 list->reasons[subsys_reason])) {
649 rcu_read_unlock();
650 goto nla_put_failure;
651 }
652 rcu_read_unlock();
653
654 snprintf(buf, sizeof(buf), "%pS", cb->pc);
655 if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
656 goto nla_put_failure;
657
658 rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
659 if (rc)
660 goto nla_put_failure;
661
662 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
663 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
664 goto nla_put_failure;
665
666 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
667 goto nla_put_failure;
668
669 if (!payload_len)
670 goto out;
671
672 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
673 goto nla_put_failure;
674
675 attr = skb_put(msg, nla_total_size(payload_len));
676 attr->nla_type = NET_DM_ATTR_PAYLOAD;
677 attr->nla_len = nla_attr_size(payload_len);
678 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
679 goto nla_put_failure;
680
681out:
682 genlmsg_end(msg, hdr);
683
684 return 0;
685
686nla_put_failure:
687 genlmsg_cancel(msg, hdr);
688 return -EMSGSIZE;
689}
690
691#define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
692
693static void net_dm_packet_report(struct sk_buff *skb)
694{
695 struct sk_buff *msg;
696 size_t payload_len;
697 int rc;
698
699 /* Make sure we start copying the packet from the MAC header */
700 if (skb->data > skb_mac_header(skb))
701 skb_push(skb, skb->data - skb_mac_header(skb));
702 else
703 skb_pull(skb, skb_mac_header(skb) - skb->data);
704
705 /* Ensure packet fits inside a single netlink attribute */
706 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
707 if (net_dm_trunc_len)
708 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
709
710 msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
711 if (!msg)
712 goto out;
713
714 rc = net_dm_packet_report_fill(msg, skb, payload_len);
715 if (rc) {
716 nlmsg_free(msg);
717 goto out;
718 }
719
720 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
721
722out:
723 consume_skb(skb);
724}
725
726static void net_dm_packet_work(struct work_struct *work)
727{
728 struct per_cpu_dm_data *data;
729 struct sk_buff_head list;
730 struct sk_buff *skb;
731 unsigned long flags;
732
733 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
734
735 __skb_queue_head_init(&list);
736
737 spin_lock_irqsave(&data->drop_queue.lock, flags);
738 skb_queue_splice_tail_init(&data->drop_queue, &list);
739 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
740
741 while ((skb = __skb_dequeue(&list)))
742 net_dm_packet_report(skb);
743}
744
745static size_t
746net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata)
747{
748 return hw_metadata->fa_cookie ?
749 nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
750}
751
752static size_t
753net_dm_hw_packet_report_size(size_t payload_len,
754 const struct devlink_trap_metadata *hw_metadata)
755{
756 size_t size;
757
758 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
759
760 return NLMSG_ALIGN(size) +
761 /* NET_DM_ATTR_ORIGIN */
762 nla_total_size(sizeof(u16)) +
763 /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
764 nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
765 /* NET_DM_ATTR_HW_TRAP_NAME */
766 nla_total_size(strlen(hw_metadata->trap_name) + 1) +
767 /* NET_DM_ATTR_IN_PORT */
768 net_dm_in_port_size() +
769 /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
770 net_dm_flow_action_cookie_size(hw_metadata) +
771 /* NET_DM_ATTR_TIMESTAMP */
772 nla_total_size(sizeof(u64)) +
773 /* NET_DM_ATTR_ORIG_LEN */
774 nla_total_size(sizeof(u32)) +
775 /* NET_DM_ATTR_PROTO */
776 nla_total_size(sizeof(u16)) +
777 /* NET_DM_ATTR_PAYLOAD */
778 nla_total_size(payload_len);
779}
780
781static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
782 struct sk_buff *skb, size_t payload_len)
783{
784 struct devlink_trap_metadata *hw_metadata;
785 struct nlattr *attr;
786 void *hdr;
787
788 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
789
790 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
791 NET_DM_CMD_PACKET_ALERT);
792 if (!hdr)
793 return -EMSGSIZE;
794
795 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
796 goto nla_put_failure;
797
798 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
799 hw_metadata->trap_group_name))
800 goto nla_put_failure;
801
802 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
803 hw_metadata->trap_name))
804 goto nla_put_failure;
805
806 if (hw_metadata->input_dev) {
807 struct net_device *dev = hw_metadata->input_dev;
808 int rc;
809
810 rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
811 dev->name);
812 if (rc)
813 goto nla_put_failure;
814 }
815
816 if (hw_metadata->fa_cookie &&
817 nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
818 hw_metadata->fa_cookie->cookie_len,
819 hw_metadata->fa_cookie->cookie))
820 goto nla_put_failure;
821
822 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
823 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
824 goto nla_put_failure;
825
826 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
827 goto nla_put_failure;
828
829 if (!payload_len)
830 goto out;
831
832 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
833 goto nla_put_failure;
834
835 attr = skb_put(msg, nla_total_size(payload_len));
836 attr->nla_type = NET_DM_ATTR_PAYLOAD;
837 attr->nla_len = nla_attr_size(payload_len);
838 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
839 goto nla_put_failure;
840
841out:
842 genlmsg_end(msg, hdr);
843
844 return 0;
845
846nla_put_failure:
847 genlmsg_cancel(msg, hdr);
848 return -EMSGSIZE;
849}
850
851static struct devlink_trap_metadata *
852net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
853{
854 const struct flow_action_cookie *fa_cookie;
855 struct devlink_trap_metadata *hw_metadata;
856 const char *trap_group_name;
857 const char *trap_name;
858
859 hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
860 if (!hw_metadata)
861 return NULL;
862
863 trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
864 if (!trap_group_name)
865 goto free_hw_metadata;
866 hw_metadata->trap_group_name = trap_group_name;
867
868 trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
869 if (!trap_name)
870 goto free_trap_group;
871 hw_metadata->trap_name = trap_name;
872
873 if (metadata->fa_cookie) {
874 size_t cookie_size = sizeof(*fa_cookie) +
875 metadata->fa_cookie->cookie_len;
876
877 fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
878 GFP_ATOMIC);
879 if (!fa_cookie)
880 goto free_trap_name;
881 hw_metadata->fa_cookie = fa_cookie;
882 }
883
884 hw_metadata->input_dev = metadata->input_dev;
885 netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
886 GFP_ATOMIC);
887
888 return hw_metadata;
889
890free_trap_name:
891 kfree(trap_name);
892free_trap_group:
893 kfree(trap_group_name);
894free_hw_metadata:
895 kfree(hw_metadata);
896 return NULL;
897}
898
899static void
900net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
901{
902 netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
903 kfree(hw_metadata->fa_cookie);
904 kfree(hw_metadata->trap_name);
905 kfree(hw_metadata->trap_group_name);
906 kfree(hw_metadata);
907}
908
909static void net_dm_hw_packet_report(struct sk_buff *skb)
910{
911 struct devlink_trap_metadata *hw_metadata;
912 struct sk_buff *msg;
913 size_t payload_len;
914 int rc;
915
916 if (skb->data > skb_mac_header(skb))
917 skb_push(skb, skb->data - skb_mac_header(skb));
918 else
919 skb_pull(skb, skb_mac_header(skb) - skb->data);
920
921 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
922 if (net_dm_trunc_len)
923 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
924
925 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
926 msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
927 GFP_KERNEL);
928 if (!msg)
929 goto out;
930
931 rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
932 if (rc) {
933 nlmsg_free(msg);
934 goto out;
935 }
936
937 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
938
939out:
940 net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
941 consume_skb(skb);
942}
943
944static void net_dm_hw_packet_work(struct work_struct *work)
945{
946 struct per_cpu_dm_data *hw_data;
947 struct sk_buff_head list;
948 struct sk_buff *skb;
949 unsigned long flags;
950
951 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
952
953 __skb_queue_head_init(&list);
954
955 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
956 skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
957 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
958
959 while ((skb = __skb_dequeue(&list)))
960 net_dm_hw_packet_report(skb);
961}
962
963static void
964net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
965 struct sk_buff *skb,
966 const struct devlink_trap_metadata *metadata)
967{
968 struct devlink_trap_metadata *n_hw_metadata;
969 ktime_t tstamp = ktime_get_real();
970 struct per_cpu_dm_data *hw_data;
971 struct sk_buff *nskb;
972 unsigned long flags;
973
974 if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
975 return;
976
977 if (!skb_mac_header_was_set(skb))
978 return;
979
980 nskb = skb_clone(skb, GFP_ATOMIC);
981 if (!nskb)
982 return;
983
984 n_hw_metadata = net_dm_hw_metadata_copy(metadata);
985 if (!n_hw_metadata)
986 goto free;
987
988 NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
989 nskb->tstamp = tstamp;
990
991 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
992
993 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
994 if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
995 __skb_queue_tail(&hw_data->drop_queue, nskb);
996 else
997 goto unlock_free;
998 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
999
1000 schedule_work(&hw_data->dm_alert_work);
1001
1002 return;
1003
1004unlock_free:
1005 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
1006 u64_stats_update_begin(&hw_data->stats.syncp);
1007 u64_stats_inc(&hw_data->stats.dropped);
1008 u64_stats_update_end(&hw_data->stats.syncp);
1009 net_dm_hw_metadata_free(n_hw_metadata);
1010free:
1011 consume_skb(nskb);
1012}
1013
1014static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
1015 .kfree_skb_probe = net_dm_packet_trace_kfree_skb_hit,
1016 .napi_poll_probe = net_dm_packet_trace_napi_poll_hit,
1017 .work_item_func = net_dm_packet_work,
1018 .hw_work_item_func = net_dm_hw_packet_work,
1019 .hw_trap_probe = net_dm_hw_trap_packet_probe,
1020};
1021
1022static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1023 [NET_DM_ALERT_MODE_SUMMARY] = &net_dm_alert_summary_ops,
1024 [NET_DM_ALERT_MODE_PACKET] = &net_dm_alert_packet_ops,
1025};
1026
1027#if IS_ENABLED(CONFIG_NET_DEVLINK)
1028static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1029{
1030 return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1031}
1032
1033static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1034{
1035 unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1036 tracepoint_synchronize_unregister();
1037}
1038#else
1039static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1040{
1041 return -EOPNOTSUPP;
1042}
1043
1044static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1045{
1046}
1047#endif
1048
1049static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1050{
1051 const struct net_dm_alert_ops *ops;
1052 int cpu, rc;
1053
1054 if (monitor_hw) {
1055 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1056 return -EAGAIN;
1057 }
1058
1059 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1060
1061 if (!try_module_get(THIS_MODULE)) {
1062 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1063 return -ENODEV;
1064 }
1065
1066 for_each_possible_cpu(cpu) {
1067 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1068 struct net_dm_hw_entries *hw_entries;
1069
1070 INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1071 timer_setup(&hw_data->send_timer, sched_send_work, 0);
1072 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1073 kfree(hw_entries);
1074 }
1075
1076 rc = net_dm_hw_probe_register(ops);
1077 if (rc) {
1078 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1079 goto err_module_put;
1080 }
1081
1082 monitor_hw = true;
1083
1084 return 0;
1085
1086err_module_put:
1087 for_each_possible_cpu(cpu) {
1088 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1089 struct sk_buff *skb;
1090
1091 del_timer_sync(&hw_data->send_timer);
1092 cancel_work_sync(&hw_data->dm_alert_work);
1093 while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1094 struct devlink_trap_metadata *hw_metadata;
1095
1096 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1097 net_dm_hw_metadata_free(hw_metadata);
1098 consume_skb(skb);
1099 }
1100 }
1101 module_put(THIS_MODULE);
1102 return rc;
1103}
1104
1105static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1106{
1107 const struct net_dm_alert_ops *ops;
1108 int cpu;
1109
1110 if (!monitor_hw) {
1111 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1112 return;
1113 }
1114
1115 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1116
1117 monitor_hw = false;
1118
1119 net_dm_hw_probe_unregister(ops);
1120
1121 for_each_possible_cpu(cpu) {
1122 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1123 struct sk_buff *skb;
1124
1125 del_timer_sync(&hw_data->send_timer);
1126 cancel_work_sync(&hw_data->dm_alert_work);
1127 while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1128 struct devlink_trap_metadata *hw_metadata;
1129
1130 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1131 net_dm_hw_metadata_free(hw_metadata);
1132 consume_skb(skb);
1133 }
1134 }
1135
1136 module_put(THIS_MODULE);
1137}
1138
1139static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1140{
1141 const struct net_dm_alert_ops *ops;
1142 int cpu, rc;
1143
1144 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1145
1146 if (!try_module_get(THIS_MODULE)) {
1147 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1148 return -ENODEV;
1149 }
1150
1151 for_each_possible_cpu(cpu) {
1152 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1153 struct sk_buff *skb;
1154
1155 INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1156 timer_setup(&data->send_timer, sched_send_work, 0);
1157 /* Allocate a new per-CPU skb for the summary alert message and
1158 * free the old one which might contain stale data from
1159 * previous tracing.
1160 */
1161 skb = reset_per_cpu_data(data);
1162 consume_skb(skb);
1163 }
1164
1165 rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1166 if (rc) {
1167 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1168 goto err_module_put;
1169 }
1170
1171 rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1172 if (rc) {
1173 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1174 goto err_unregister_trace;
1175 }
1176
1177 return 0;
1178
1179err_unregister_trace:
1180 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1181err_module_put:
1182 for_each_possible_cpu(cpu) {
1183 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1184 struct sk_buff *skb;
1185
1186 del_timer_sync(&data->send_timer);
1187 cancel_work_sync(&data->dm_alert_work);
1188 while ((skb = __skb_dequeue(&data->drop_queue)))
1189 consume_skb(skb);
1190 }
1191 module_put(THIS_MODULE);
1192 return rc;
1193}
1194
1195static void net_dm_trace_off_set(void)
1196{
1197 const struct net_dm_alert_ops *ops;
1198 int cpu;
1199
1200 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1201
1202 unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1203 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1204
1205 tracepoint_synchronize_unregister();
1206
1207 /* Make sure we do not send notifications to user space after request
1208 * to stop tracing returns.
1209 */
1210 for_each_possible_cpu(cpu) {
1211 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1212 struct sk_buff *skb;
1213
1214 del_timer_sync(&data->send_timer);
1215 cancel_work_sync(&data->dm_alert_work);
1216 while ((skb = __skb_dequeue(&data->drop_queue)))
1217 consume_skb(skb);
1218 }
1219
1220 module_put(THIS_MODULE);
1221}
1222
1223static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1224{
1225 int rc = 0;
1226
1227 if (state == trace_state) {
1228 NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1229 return -EAGAIN;
1230 }
1231
1232 switch (state) {
1233 case TRACE_ON:
1234 rc = net_dm_trace_on_set(extack);
1235 break;
1236 case TRACE_OFF:
1237 net_dm_trace_off_set();
1238 break;
1239 default:
1240 rc = 1;
1241 break;
1242 }
1243
1244 if (!rc)
1245 trace_state = state;
1246 else
1247 rc = -EINPROGRESS;
1248
1249 return rc;
1250}
1251
1252static bool net_dm_is_monitoring(void)
1253{
1254 return trace_state == TRACE_ON || monitor_hw;
1255}
1256
1257static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1258 enum net_dm_alert_mode *p_alert_mode)
1259{
1260 u8 val;
1261
1262 val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1263
1264 switch (val) {
1265 case NET_DM_ALERT_MODE_SUMMARY:
1266 case NET_DM_ALERT_MODE_PACKET:
1267 *p_alert_mode = val;
1268 break;
1269 default:
1270 return -EINVAL;
1271 }
1272
1273 return 0;
1274}
1275
1276static int net_dm_alert_mode_set(struct genl_info *info)
1277{
1278 struct netlink_ext_ack *extack = info->extack;
1279 enum net_dm_alert_mode alert_mode;
1280 int rc;
1281
1282 if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1283 return 0;
1284
1285 rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1286 if (rc) {
1287 NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1288 return -EINVAL;
1289 }
1290
1291 net_dm_alert_mode = alert_mode;
1292
1293 return 0;
1294}
1295
1296static void net_dm_trunc_len_set(struct genl_info *info)
1297{
1298 if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1299 return;
1300
1301 net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1302}
1303
1304static void net_dm_queue_len_set(struct genl_info *info)
1305{
1306 if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1307 return;
1308
1309 net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1310}
1311
1312static int net_dm_cmd_config(struct sk_buff *skb,
1313 struct genl_info *info)
1314{
1315 struct netlink_ext_ack *extack = info->extack;
1316 int rc;
1317
1318 if (net_dm_is_monitoring()) {
1319 NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1320 return -EBUSY;
1321 }
1322
1323 rc = net_dm_alert_mode_set(info);
1324 if (rc)
1325 return rc;
1326
1327 net_dm_trunc_len_set(info);
1328
1329 net_dm_queue_len_set(info);
1330
1331 return 0;
1332}
1333
1334static int net_dm_monitor_start(bool set_sw, bool set_hw,
1335 struct netlink_ext_ack *extack)
1336{
1337 bool sw_set = false;
1338 int rc;
1339
1340 if (set_sw) {
1341 rc = set_all_monitor_traces(TRACE_ON, extack);
1342 if (rc)
1343 return rc;
1344 sw_set = true;
1345 }
1346
1347 if (set_hw) {
1348 rc = net_dm_hw_monitor_start(extack);
1349 if (rc)
1350 goto err_monitor_hw;
1351 }
1352
1353 return 0;
1354
1355err_monitor_hw:
1356 if (sw_set)
1357 set_all_monitor_traces(TRACE_OFF, extack);
1358 return rc;
1359}
1360
1361static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1362 struct netlink_ext_ack *extack)
1363{
1364 if (set_hw)
1365 net_dm_hw_monitor_stop(extack);
1366 if (set_sw)
1367 set_all_monitor_traces(TRACE_OFF, extack);
1368}
1369
1370static int net_dm_cmd_trace(struct sk_buff *skb,
1371 struct genl_info *info)
1372{
1373 bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1374 bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1375 struct netlink_ext_ack *extack = info->extack;
1376
1377 /* To maintain backward compatibility, we start / stop monitoring of
1378 * software drops if no flag is specified.
1379 */
1380 if (!set_sw && !set_hw)
1381 set_sw = true;
1382
1383 switch (info->genlhdr->cmd) {
1384 case NET_DM_CMD_START:
1385 return net_dm_monitor_start(set_sw, set_hw, extack);
1386 case NET_DM_CMD_STOP:
1387 net_dm_monitor_stop(set_sw, set_hw, extack);
1388 return 0;
1389 }
1390
1391 return -EOPNOTSUPP;
1392}
1393
1394static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1395{
1396 void *hdr;
1397
1398 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1399 &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1400 if (!hdr)
1401 return -EMSGSIZE;
1402
1403 if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1404 goto nla_put_failure;
1405
1406 if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1407 goto nla_put_failure;
1408
1409 if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1410 goto nla_put_failure;
1411
1412 genlmsg_end(msg, hdr);
1413
1414 return 0;
1415
1416nla_put_failure:
1417 genlmsg_cancel(msg, hdr);
1418 return -EMSGSIZE;
1419}
1420
1421static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1422{
1423 struct sk_buff *msg;
1424 int rc;
1425
1426 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1427 if (!msg)
1428 return -ENOMEM;
1429
1430 rc = net_dm_config_fill(msg, info);
1431 if (rc)
1432 goto free_msg;
1433
1434 return genlmsg_reply(msg, info);
1435
1436free_msg:
1437 nlmsg_free(msg);
1438 return rc;
1439}
1440
1441static void net_dm_stats_read(struct net_dm_stats *stats)
1442{
1443 int cpu;
1444
1445 memset(stats, 0, sizeof(*stats));
1446 for_each_possible_cpu(cpu) {
1447 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1448 struct net_dm_stats *cpu_stats = &data->stats;
1449 unsigned int start;
1450 u64 dropped;
1451
1452 do {
1453 start = u64_stats_fetch_begin(&cpu_stats->syncp);
1454 dropped = u64_stats_read(&cpu_stats->dropped);
1455 } while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1456
1457 u64_stats_add(&stats->dropped, dropped);
1458 }
1459}
1460
1461static int net_dm_stats_put(struct sk_buff *msg)
1462{
1463 struct net_dm_stats stats;
1464 struct nlattr *attr;
1465
1466 net_dm_stats_read(&stats);
1467
1468 attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1469 if (!attr)
1470 return -EMSGSIZE;
1471
1472 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1473 u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1474 goto nla_put_failure;
1475
1476 nla_nest_end(msg, attr);
1477
1478 return 0;
1479
1480nla_put_failure:
1481 nla_nest_cancel(msg, attr);
1482 return -EMSGSIZE;
1483}
1484
1485static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1486{
1487 int cpu;
1488
1489 memset(stats, 0, sizeof(*stats));
1490 for_each_possible_cpu(cpu) {
1491 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1492 struct net_dm_stats *cpu_stats = &hw_data->stats;
1493 unsigned int start;
1494 u64 dropped;
1495
1496 do {
1497 start = u64_stats_fetch_begin(&cpu_stats->syncp);
1498 dropped = u64_stats_read(&cpu_stats->dropped);
1499 } while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1500
1501 u64_stats_add(&stats->dropped, dropped);
1502 }
1503}
1504
1505static int net_dm_hw_stats_put(struct sk_buff *msg)
1506{
1507 struct net_dm_stats stats;
1508 struct nlattr *attr;
1509
1510 net_dm_hw_stats_read(&stats);
1511
1512 attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1513 if (!attr)
1514 return -EMSGSIZE;
1515
1516 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1517 u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1518 goto nla_put_failure;
1519
1520 nla_nest_end(msg, attr);
1521
1522 return 0;
1523
1524nla_put_failure:
1525 nla_nest_cancel(msg, attr);
1526 return -EMSGSIZE;
1527}
1528
1529static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1530{
1531 void *hdr;
1532 int rc;
1533
1534 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1535 &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1536 if (!hdr)
1537 return -EMSGSIZE;
1538
1539 rc = net_dm_stats_put(msg);
1540 if (rc)
1541 goto nla_put_failure;
1542
1543 rc = net_dm_hw_stats_put(msg);
1544 if (rc)
1545 goto nla_put_failure;
1546
1547 genlmsg_end(msg, hdr);
1548
1549 return 0;
1550
1551nla_put_failure:
1552 genlmsg_cancel(msg, hdr);
1553 return -EMSGSIZE;
1554}
1555
1556static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1557{
1558 struct sk_buff *msg;
1559 int rc;
1560
1561 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1562 if (!msg)
1563 return -ENOMEM;
1564
1565 rc = net_dm_stats_fill(msg, info);
1566 if (rc)
1567 goto free_msg;
1568
1569 return genlmsg_reply(msg, info);
1570
1571free_msg:
1572 nlmsg_free(msg);
1573 return rc;
1574}
1575
1576static int dropmon_net_event(struct notifier_block *ev_block,
1577 unsigned long event, void *ptr)
1578{
1579 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1580 struct dm_hw_stat_delta *stat;
1581
1582 switch (event) {
1583 case NETDEV_REGISTER:
1584 if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1585 break;
1586 stat = kzalloc(sizeof(*stat), GFP_KERNEL);
1587 if (!stat)
1588 break;
1589
1590 stat->last_rx = jiffies;
1591 rcu_assign_pointer(dev->dm_private, stat);
1592
1593 break;
1594 case NETDEV_UNREGISTER:
1595 stat = rtnl_dereference(dev->dm_private);
1596 if (stat) {
1597 rcu_assign_pointer(dev->dm_private, NULL);
1598 kfree_rcu(stat, rcu);
1599 }
1600 break;
1601 }
1602 return NOTIFY_DONE;
1603}
1604
1605static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1606 [NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1607 [NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1608 [NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1609 [NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1610 [NET_DM_ATTR_SW_DROPS] = {. type = NLA_FLAG },
1611 [NET_DM_ATTR_HW_DROPS] = {. type = NLA_FLAG },
1612};
1613
1614static const struct genl_small_ops dropmon_ops[] = {
1615 {
1616 .cmd = NET_DM_CMD_CONFIG,
1617 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1618 .doit = net_dm_cmd_config,
1619 .flags = GENL_ADMIN_PERM,
1620 },
1621 {
1622 .cmd = NET_DM_CMD_START,
1623 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1624 .doit = net_dm_cmd_trace,
1625 .flags = GENL_ADMIN_PERM,
1626 },
1627 {
1628 .cmd = NET_DM_CMD_STOP,
1629 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1630 .doit = net_dm_cmd_trace,
1631 .flags = GENL_ADMIN_PERM,
1632 },
1633 {
1634 .cmd = NET_DM_CMD_CONFIG_GET,
1635 .doit = net_dm_cmd_config_get,
1636 },
1637 {
1638 .cmd = NET_DM_CMD_STATS_GET,
1639 .doit = net_dm_cmd_stats_get,
1640 },
1641};
1642
1643static int net_dm_nl_pre_doit(const struct genl_split_ops *ops,
1644 struct sk_buff *skb, struct genl_info *info)
1645{
1646 mutex_lock(&net_dm_mutex);
1647
1648 return 0;
1649}
1650
1651static void net_dm_nl_post_doit(const struct genl_split_ops *ops,
1652 struct sk_buff *skb, struct genl_info *info)
1653{
1654 mutex_unlock(&net_dm_mutex);
1655}
1656
1657static struct genl_family net_drop_monitor_family __ro_after_init = {
1658 .hdrsize = 0,
1659 .name = "NET_DM",
1660 .version = 2,
1661 .maxattr = NET_DM_ATTR_MAX,
1662 .policy = net_dm_nl_policy,
1663 .pre_doit = net_dm_nl_pre_doit,
1664 .post_doit = net_dm_nl_post_doit,
1665 .module = THIS_MODULE,
1666 .small_ops = dropmon_ops,
1667 .n_small_ops = ARRAY_SIZE(dropmon_ops),
1668 .resv_start_op = NET_DM_CMD_STATS_GET + 1,
1669 .mcgrps = dropmon_mcgrps,
1670 .n_mcgrps = ARRAY_SIZE(dropmon_mcgrps),
1671};
1672
1673static struct notifier_block dropmon_net_notifier = {
1674 .notifier_call = dropmon_net_event
1675};
1676
1677static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1678{
1679 raw_spin_lock_init(&data->lock);
1680 skb_queue_head_init(&data->drop_queue);
1681 u64_stats_init(&data->stats.syncp);
1682}
1683
1684static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1685{
1686 WARN_ON(!skb_queue_empty(&data->drop_queue));
1687}
1688
1689static void net_dm_cpu_data_init(int cpu)
1690{
1691 struct per_cpu_dm_data *data;
1692
1693 data = &per_cpu(dm_cpu_data, cpu);
1694 __net_dm_cpu_data_init(data);
1695}
1696
1697static void net_dm_cpu_data_fini(int cpu)
1698{
1699 struct per_cpu_dm_data *data;
1700
1701 data = &per_cpu(dm_cpu_data, cpu);
1702 /* At this point, we should have exclusive access
1703 * to this struct and can free the skb inside it.
1704 */
1705 consume_skb(data->skb);
1706 __net_dm_cpu_data_fini(data);
1707}
1708
1709static void net_dm_hw_cpu_data_init(int cpu)
1710{
1711 struct per_cpu_dm_data *hw_data;
1712
1713 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1714 __net_dm_cpu_data_init(hw_data);
1715}
1716
1717static void net_dm_hw_cpu_data_fini(int cpu)
1718{
1719 struct per_cpu_dm_data *hw_data;
1720
1721 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1722 kfree(hw_data->hw_entries);
1723 __net_dm_cpu_data_fini(hw_data);
1724}
1725
1726static int __init init_net_drop_monitor(void)
1727{
1728 int cpu, rc;
1729
1730 pr_info("Initializing network drop monitor service\n");
1731
1732 if (sizeof(void *) > 8) {
1733 pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1734 return -ENOSPC;
1735 }
1736
1737 for_each_possible_cpu(cpu) {
1738 net_dm_cpu_data_init(cpu);
1739 net_dm_hw_cpu_data_init(cpu);
1740 }
1741
1742 rc = register_netdevice_notifier(&dropmon_net_notifier);
1743 if (rc < 0) {
1744 pr_crit("Failed to register netdevice notifier\n");
1745 return rc;
1746 }
1747
1748 rc = genl_register_family(&net_drop_monitor_family);
1749 if (rc) {
1750 pr_err("Could not create drop monitor netlink family\n");
1751 goto out_unreg;
1752 }
1753 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1754
1755 rc = 0;
1756
1757 goto out;
1758
1759out_unreg:
1760 WARN_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1761out:
1762 return rc;
1763}
1764
1765static void exit_net_drop_monitor(void)
1766{
1767 int cpu;
1768
1769 /*
1770 * Because of the module_get/put we do in the trace state change path
1771 * we are guaranteed not to have any current users when we get here
1772 */
1773 BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1774
1775 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1776
1777 for_each_possible_cpu(cpu) {
1778 net_dm_hw_cpu_data_fini(cpu);
1779 net_dm_cpu_data_fini(cpu);
1780 }
1781}
1782
1783module_init(init_net_drop_monitor);
1784module_exit(exit_net_drop_monitor);
1785
1786MODULE_LICENSE("GPL v2");
1787MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1788MODULE_ALIAS_GENL_FAMILY("NET_DM");
1789MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Monitoring code for network dropped packet alerts
4 *
5 * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/netdevice.h>
11#include <linux/etherdevice.h>
12#include <linux/string.h>
13#include <linux/if_arp.h>
14#include <linux/inetdevice.h>
15#include <linux/inet.h>
16#include <linux/interrupt.h>
17#include <linux/netpoll.h>
18#include <linux/sched.h>
19#include <linux/delay.h>
20#include <linux/types.h>
21#include <linux/workqueue.h>
22#include <linux/netlink.h>
23#include <linux/net_dropmon.h>
24#include <linux/percpu.h>
25#include <linux/timer.h>
26#include <linux/bitops.h>
27#include <linux/slab.h>
28#include <linux/module.h>
29#include <net/drop_monitor.h>
30#include <net/genetlink.h>
31#include <net/netevent.h>
32#include <net/flow_offload.h>
33
34#include <trace/events/skb.h>
35#include <trace/events/napi.h>
36
37#include <asm/unaligned.h>
38
39#define TRACE_ON 1
40#define TRACE_OFF 0
41
42/*
43 * Globals, our netlink socket pointer
44 * and the work handle that will send up
45 * netlink alerts
46 */
47static int trace_state = TRACE_OFF;
48static bool monitor_hw;
49
50/* net_dm_mutex
51 *
52 * An overall lock guarding every operation coming from userspace.
53 * It also guards the global 'hw_stats_list' list.
54 */
55static DEFINE_MUTEX(net_dm_mutex);
56
57struct net_dm_stats {
58 u64 dropped;
59 struct u64_stats_sync syncp;
60};
61
62#define NET_DM_MAX_HW_TRAP_NAME_LEN 40
63
64struct net_dm_hw_entry {
65 char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
66 u32 count;
67};
68
69struct net_dm_hw_entries {
70 u32 num_entries;
71 struct net_dm_hw_entry entries[];
72};
73
74struct 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
88struct dm_hw_stat_delta {
89 struct net_device *dev;
90 unsigned long last_rx;
91 struct list_head list;
92 struct rcu_head rcu;
93 unsigned long last_drop_val;
94};
95
96static struct genl_family net_drop_monitor_family;
97
98static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100
101static int dm_hit_limit = 64;
102static int dm_delay = 1;
103static unsigned long dm_hw_check_delta = 2*HZ;
104static LIST_HEAD(hw_stats_list);
105
106static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
107static u32 net_dm_trunc_len;
108static u32 net_dm_queue_len = 1000;
109
110struct net_dm_alert_ops {
111 void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
112 void *location);
113 void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
114 int work, int budget);
115 void (*work_item_func)(struct work_struct *work);
116 void (*hw_work_item_func)(struct work_struct *work);
117 void (*hw_probe)(struct sk_buff *skb,
118 const struct net_dm_hw_metadata *hw_metadata);
119};
120
121struct net_dm_skb_cb {
122 union {
123 struct net_dm_hw_metadata *hw_metadata;
124 void *pc;
125 };
126};
127
128#define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
129
130static 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
166err:
167 mod_timer(&data->send_timer, jiffies + HZ / 10);
168out:
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
183static const struct genl_multicast_group dropmon_mcgrps[] = {
184 { .name = "events", },
185};
186
187static 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 */
206static 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
213static 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
259out:
260 spin_unlock_irqrestore(&data->lock, flags);
261}
262
263static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
264{
265 trace_drop_common(skb, location);
266}
267
268static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
269 int work, int budget)
270{
271 struct dm_hw_stat_delta *new_stat;
272
273 /*
274 * Don't check napi structures with no associated device
275 */
276 if (!napi->dev)
277 return;
278
279 rcu_read_lock();
280 list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
281 /*
282 * only add a note to our monitor buffer if:
283 * 1) this is the dev we received on
284 * 2) its after the last_rx delta
285 * 3) our rx_dropped count has gone up
286 */
287 if ((new_stat->dev == napi->dev) &&
288 (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
289 (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
290 trace_drop_common(NULL, NULL);
291 new_stat->last_drop_val = napi->dev->stats.rx_dropped;
292 new_stat->last_rx = jiffies;
293 break;
294 }
295 }
296 rcu_read_unlock();
297}
298
299static struct net_dm_hw_entries *
300net_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
322static 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
341nla_put_failure:
342 nla_nest_cancel(msg, attr);
343 return -EMSGSIZE;
344}
345
346static 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
368nla_put_failure:
369 nla_nest_cancel(msg, attr);
370 return -EMSGSIZE;
371}
372
373static int
374net_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
400nla_put_failure:
401 genlmsg_cancel(msg, hdr);
402 return -EMSGSIZE;
403}
404
405static 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
430out:
431 kfree(hw_entries);
432}
433
434static void
435net_dm_hw_summary_probe(struct sk_buff *skb,
436 const struct net_dm_hw_metadata *hw_metadata)
437{
438 struct net_dm_hw_entries *hw_entries;
439 struct net_dm_hw_entry *hw_entry;
440 struct per_cpu_dm_data *hw_data;
441 unsigned long flags;
442 int i;
443
444 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
445 spin_lock_irqsave(&hw_data->lock, flags);
446 hw_entries = hw_data->hw_entries;
447
448 if (!hw_entries)
449 goto out;
450
451 for (i = 0; i < hw_entries->num_entries; i++) {
452 hw_entry = &hw_entries->entries[i];
453 if (!strncmp(hw_entry->trap_name, hw_metadata->trap_name,
454 NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
455 hw_entry->count++;
456 goto out;
457 }
458 }
459 if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
460 goto out;
461
462 hw_entry = &hw_entries->entries[hw_entries->num_entries];
463 strlcpy(hw_entry->trap_name, hw_metadata->trap_name,
464 NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
465 hw_entry->count = 1;
466 hw_entries->num_entries++;
467
468 if (!timer_pending(&hw_data->send_timer)) {
469 hw_data->send_timer.expires = jiffies + dm_delay * HZ;
470 add_timer(&hw_data->send_timer);
471 }
472
473out:
474 spin_unlock_irqrestore(&hw_data->lock, flags);
475}
476
477static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
478 .kfree_skb_probe = trace_kfree_skb_hit,
479 .napi_poll_probe = trace_napi_poll_hit,
480 .work_item_func = send_dm_alert,
481 .hw_work_item_func = net_dm_hw_summary_work,
482 .hw_probe = net_dm_hw_summary_probe,
483};
484
485static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
486 struct sk_buff *skb,
487 void *location)
488{
489 ktime_t tstamp = ktime_get_real();
490 struct per_cpu_dm_data *data;
491 struct sk_buff *nskb;
492 unsigned long flags;
493
494 if (!skb_mac_header_was_set(skb))
495 return;
496
497 nskb = skb_clone(skb, GFP_ATOMIC);
498 if (!nskb)
499 return;
500
501 NET_DM_SKB_CB(nskb)->pc = location;
502 /* Override the timestamp because we care about the time when the
503 * packet was dropped.
504 */
505 nskb->tstamp = tstamp;
506
507 data = this_cpu_ptr(&dm_cpu_data);
508
509 spin_lock_irqsave(&data->drop_queue.lock, flags);
510 if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
511 __skb_queue_tail(&data->drop_queue, nskb);
512 else
513 goto unlock_free;
514 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
515
516 schedule_work(&data->dm_alert_work);
517
518 return;
519
520unlock_free:
521 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
522 u64_stats_update_begin(&data->stats.syncp);
523 data->stats.dropped++;
524 u64_stats_update_end(&data->stats.syncp);
525 consume_skb(nskb);
526}
527
528static void net_dm_packet_trace_napi_poll_hit(void *ignore,
529 struct napi_struct *napi,
530 int work, int budget)
531{
532}
533
534static size_t net_dm_in_port_size(void)
535{
536 /* NET_DM_ATTR_IN_PORT nest */
537 return nla_total_size(0) +
538 /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
539 nla_total_size(sizeof(u32)) +
540 /* NET_DM_ATTR_PORT_NETDEV_NAME */
541 nla_total_size(IFNAMSIZ + 1);
542}
543
544#define NET_DM_MAX_SYMBOL_LEN 40
545
546static size_t net_dm_packet_report_size(size_t payload_len)
547{
548 size_t size;
549
550 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
551
552 return NLMSG_ALIGN(size) +
553 /* NET_DM_ATTR_ORIGIN */
554 nla_total_size(sizeof(u16)) +
555 /* NET_DM_ATTR_PC */
556 nla_total_size(sizeof(u64)) +
557 /* NET_DM_ATTR_SYMBOL */
558 nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
559 /* NET_DM_ATTR_IN_PORT */
560 net_dm_in_port_size() +
561 /* NET_DM_ATTR_TIMESTAMP */
562 nla_total_size(sizeof(u64)) +
563 /* NET_DM_ATTR_ORIG_LEN */
564 nla_total_size(sizeof(u32)) +
565 /* NET_DM_ATTR_PROTO */
566 nla_total_size(sizeof(u16)) +
567 /* NET_DM_ATTR_PAYLOAD */
568 nla_total_size(payload_len);
569}
570
571static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
572 const char *name)
573{
574 struct nlattr *attr;
575
576 attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
577 if (!attr)
578 return -EMSGSIZE;
579
580 if (ifindex &&
581 nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
582 goto nla_put_failure;
583
584 if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
585 goto nla_put_failure;
586
587 nla_nest_end(msg, attr);
588
589 return 0;
590
591nla_put_failure:
592 nla_nest_cancel(msg, attr);
593 return -EMSGSIZE;
594}
595
596static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
597 size_t payload_len)
598{
599 u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
600 char buf[NET_DM_MAX_SYMBOL_LEN];
601 struct nlattr *attr;
602 void *hdr;
603 int rc;
604
605 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
606 NET_DM_CMD_PACKET_ALERT);
607 if (!hdr)
608 return -EMSGSIZE;
609
610 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
611 goto nla_put_failure;
612
613 if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
614 goto nla_put_failure;
615
616 snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
617 if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
618 goto nla_put_failure;
619
620 rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
621 if (rc)
622 goto nla_put_failure;
623
624 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
625 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
626 goto nla_put_failure;
627
628 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
629 goto nla_put_failure;
630
631 if (!payload_len)
632 goto out;
633
634 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
635 goto nla_put_failure;
636
637 attr = skb_put(msg, nla_total_size(payload_len));
638 attr->nla_type = NET_DM_ATTR_PAYLOAD;
639 attr->nla_len = nla_attr_size(payload_len);
640 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
641 goto nla_put_failure;
642
643out:
644 genlmsg_end(msg, hdr);
645
646 return 0;
647
648nla_put_failure:
649 genlmsg_cancel(msg, hdr);
650 return -EMSGSIZE;
651}
652
653#define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
654
655static void net_dm_packet_report(struct sk_buff *skb)
656{
657 struct sk_buff *msg;
658 size_t payload_len;
659 int rc;
660
661 /* Make sure we start copying the packet from the MAC header */
662 if (skb->data > skb_mac_header(skb))
663 skb_push(skb, skb->data - skb_mac_header(skb));
664 else
665 skb_pull(skb, skb_mac_header(skb) - skb->data);
666
667 /* Ensure packet fits inside a single netlink attribute */
668 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
669 if (net_dm_trunc_len)
670 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
671
672 msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
673 if (!msg)
674 goto out;
675
676 rc = net_dm_packet_report_fill(msg, skb, payload_len);
677 if (rc) {
678 nlmsg_free(msg);
679 goto out;
680 }
681
682 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
683
684out:
685 consume_skb(skb);
686}
687
688static void net_dm_packet_work(struct work_struct *work)
689{
690 struct per_cpu_dm_data *data;
691 struct sk_buff_head list;
692 struct sk_buff *skb;
693 unsigned long flags;
694
695 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
696
697 __skb_queue_head_init(&list);
698
699 spin_lock_irqsave(&data->drop_queue.lock, flags);
700 skb_queue_splice_tail_init(&data->drop_queue, &list);
701 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
702
703 while ((skb = __skb_dequeue(&list)))
704 net_dm_packet_report(skb);
705}
706
707static size_t
708net_dm_flow_action_cookie_size(const struct net_dm_hw_metadata *hw_metadata)
709{
710 return hw_metadata->fa_cookie ?
711 nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
712}
713
714static size_t
715net_dm_hw_packet_report_size(size_t payload_len,
716 const struct net_dm_hw_metadata *hw_metadata)
717{
718 size_t size;
719
720 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
721
722 return NLMSG_ALIGN(size) +
723 /* NET_DM_ATTR_ORIGIN */
724 nla_total_size(sizeof(u16)) +
725 /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
726 nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
727 /* NET_DM_ATTR_HW_TRAP_NAME */
728 nla_total_size(strlen(hw_metadata->trap_name) + 1) +
729 /* NET_DM_ATTR_IN_PORT */
730 net_dm_in_port_size() +
731 /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
732 net_dm_flow_action_cookie_size(hw_metadata) +
733 /* NET_DM_ATTR_TIMESTAMP */
734 nla_total_size(sizeof(u64)) +
735 /* NET_DM_ATTR_ORIG_LEN */
736 nla_total_size(sizeof(u32)) +
737 /* NET_DM_ATTR_PROTO */
738 nla_total_size(sizeof(u16)) +
739 /* NET_DM_ATTR_PAYLOAD */
740 nla_total_size(payload_len);
741}
742
743static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
744 struct sk_buff *skb, size_t payload_len)
745{
746 struct net_dm_hw_metadata *hw_metadata;
747 struct nlattr *attr;
748 void *hdr;
749
750 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
751
752 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
753 NET_DM_CMD_PACKET_ALERT);
754 if (!hdr)
755 return -EMSGSIZE;
756
757 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
758 goto nla_put_failure;
759
760 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
761 hw_metadata->trap_group_name))
762 goto nla_put_failure;
763
764 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
765 hw_metadata->trap_name))
766 goto nla_put_failure;
767
768 if (hw_metadata->input_dev) {
769 struct net_device *dev = hw_metadata->input_dev;
770 int rc;
771
772 rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
773 dev->name);
774 if (rc)
775 goto nla_put_failure;
776 }
777
778 if (hw_metadata->fa_cookie &&
779 nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
780 hw_metadata->fa_cookie->cookie_len,
781 hw_metadata->fa_cookie->cookie))
782 goto nla_put_failure;
783
784 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
785 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
786 goto nla_put_failure;
787
788 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
789 goto nla_put_failure;
790
791 if (!payload_len)
792 goto out;
793
794 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
795 goto nla_put_failure;
796
797 attr = skb_put(msg, nla_total_size(payload_len));
798 attr->nla_type = NET_DM_ATTR_PAYLOAD;
799 attr->nla_len = nla_attr_size(payload_len);
800 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
801 goto nla_put_failure;
802
803out:
804 genlmsg_end(msg, hdr);
805
806 return 0;
807
808nla_put_failure:
809 genlmsg_cancel(msg, hdr);
810 return -EMSGSIZE;
811}
812
813static struct net_dm_hw_metadata *
814net_dm_hw_metadata_clone(const struct net_dm_hw_metadata *hw_metadata)
815{
816 const struct flow_action_cookie *fa_cookie;
817 struct net_dm_hw_metadata *n_hw_metadata;
818 const char *trap_group_name;
819 const char *trap_name;
820
821 n_hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
822 if (!n_hw_metadata)
823 return NULL;
824
825 trap_group_name = kstrdup(hw_metadata->trap_group_name, GFP_ATOMIC);
826 if (!trap_group_name)
827 goto free_hw_metadata;
828 n_hw_metadata->trap_group_name = trap_group_name;
829
830 trap_name = kstrdup(hw_metadata->trap_name, GFP_ATOMIC);
831 if (!trap_name)
832 goto free_trap_group;
833 n_hw_metadata->trap_name = trap_name;
834
835 if (hw_metadata->fa_cookie) {
836 size_t cookie_size = sizeof(*fa_cookie) +
837 hw_metadata->fa_cookie->cookie_len;
838
839 fa_cookie = kmemdup(hw_metadata->fa_cookie, cookie_size,
840 GFP_ATOMIC);
841 if (!fa_cookie)
842 goto free_trap_name;
843 n_hw_metadata->fa_cookie = fa_cookie;
844 }
845
846 n_hw_metadata->input_dev = hw_metadata->input_dev;
847 if (n_hw_metadata->input_dev)
848 dev_hold(n_hw_metadata->input_dev);
849
850 return n_hw_metadata;
851
852free_trap_name:
853 kfree(trap_name);
854free_trap_group:
855 kfree(trap_group_name);
856free_hw_metadata:
857 kfree(n_hw_metadata);
858 return NULL;
859}
860
861static void
862net_dm_hw_metadata_free(const struct net_dm_hw_metadata *hw_metadata)
863{
864 if (hw_metadata->input_dev)
865 dev_put(hw_metadata->input_dev);
866 kfree(hw_metadata->fa_cookie);
867 kfree(hw_metadata->trap_name);
868 kfree(hw_metadata->trap_group_name);
869 kfree(hw_metadata);
870}
871
872static void net_dm_hw_packet_report(struct sk_buff *skb)
873{
874 struct net_dm_hw_metadata *hw_metadata;
875 struct sk_buff *msg;
876 size_t payload_len;
877 int rc;
878
879 if (skb->data > skb_mac_header(skb))
880 skb_push(skb, skb->data - skb_mac_header(skb));
881 else
882 skb_pull(skb, skb_mac_header(skb) - skb->data);
883
884 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
885 if (net_dm_trunc_len)
886 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
887
888 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
889 msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
890 GFP_KERNEL);
891 if (!msg)
892 goto out;
893
894 rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
895 if (rc) {
896 nlmsg_free(msg);
897 goto out;
898 }
899
900 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
901
902out:
903 net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
904 consume_skb(skb);
905}
906
907static void net_dm_hw_packet_work(struct work_struct *work)
908{
909 struct per_cpu_dm_data *hw_data;
910 struct sk_buff_head list;
911 struct sk_buff *skb;
912 unsigned long flags;
913
914 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
915
916 __skb_queue_head_init(&list);
917
918 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
919 skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
920 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
921
922 while ((skb = __skb_dequeue(&list)))
923 net_dm_hw_packet_report(skb);
924}
925
926static void
927net_dm_hw_packet_probe(struct sk_buff *skb,
928 const struct net_dm_hw_metadata *hw_metadata)
929{
930 struct net_dm_hw_metadata *n_hw_metadata;
931 ktime_t tstamp = ktime_get_real();
932 struct per_cpu_dm_data *hw_data;
933 struct sk_buff *nskb;
934 unsigned long flags;
935
936 if (!skb_mac_header_was_set(skb))
937 return;
938
939 nskb = skb_clone(skb, GFP_ATOMIC);
940 if (!nskb)
941 return;
942
943 n_hw_metadata = net_dm_hw_metadata_clone(hw_metadata);
944 if (!n_hw_metadata)
945 goto free;
946
947 NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
948 nskb->tstamp = tstamp;
949
950 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
951
952 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
953 if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
954 __skb_queue_tail(&hw_data->drop_queue, nskb);
955 else
956 goto unlock_free;
957 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
958
959 schedule_work(&hw_data->dm_alert_work);
960
961 return;
962
963unlock_free:
964 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
965 u64_stats_update_begin(&hw_data->stats.syncp);
966 hw_data->stats.dropped++;
967 u64_stats_update_end(&hw_data->stats.syncp);
968 net_dm_hw_metadata_free(n_hw_metadata);
969free:
970 consume_skb(nskb);
971}
972
973static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
974 .kfree_skb_probe = net_dm_packet_trace_kfree_skb_hit,
975 .napi_poll_probe = net_dm_packet_trace_napi_poll_hit,
976 .work_item_func = net_dm_packet_work,
977 .hw_work_item_func = net_dm_hw_packet_work,
978 .hw_probe = net_dm_hw_packet_probe,
979};
980
981static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
982 [NET_DM_ALERT_MODE_SUMMARY] = &net_dm_alert_summary_ops,
983 [NET_DM_ALERT_MODE_PACKET] = &net_dm_alert_packet_ops,
984};
985
986void net_dm_hw_report(struct sk_buff *skb,
987 const struct net_dm_hw_metadata *hw_metadata)
988{
989 rcu_read_lock();
990
991 if (!monitor_hw)
992 goto out;
993
994 net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
995
996out:
997 rcu_read_unlock();
998}
999EXPORT_SYMBOL_GPL(net_dm_hw_report);
1000
1001static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1002{
1003 const struct net_dm_alert_ops *ops;
1004 int cpu;
1005
1006 if (monitor_hw) {
1007 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1008 return -EAGAIN;
1009 }
1010
1011 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1012
1013 if (!try_module_get(THIS_MODULE)) {
1014 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1015 return -ENODEV;
1016 }
1017
1018 for_each_possible_cpu(cpu) {
1019 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1020 struct net_dm_hw_entries *hw_entries;
1021
1022 INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1023 timer_setup(&hw_data->send_timer, sched_send_work, 0);
1024 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1025 kfree(hw_entries);
1026 }
1027
1028 monitor_hw = true;
1029
1030 return 0;
1031}
1032
1033static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1034{
1035 int cpu;
1036
1037 if (!monitor_hw) {
1038 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1039 return;
1040 }
1041
1042 monitor_hw = false;
1043
1044 /* After this call returns we are guaranteed that no CPU is processing
1045 * any hardware drops.
1046 */
1047 synchronize_rcu();
1048
1049 for_each_possible_cpu(cpu) {
1050 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1051 struct sk_buff *skb;
1052
1053 del_timer_sync(&hw_data->send_timer);
1054 cancel_work_sync(&hw_data->dm_alert_work);
1055 while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1056 struct net_dm_hw_metadata *hw_metadata;
1057
1058 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1059 net_dm_hw_metadata_free(hw_metadata);
1060 consume_skb(skb);
1061 }
1062 }
1063
1064 module_put(THIS_MODULE);
1065}
1066
1067static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1068{
1069 const struct net_dm_alert_ops *ops;
1070 int cpu, rc;
1071
1072 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1073
1074 if (!try_module_get(THIS_MODULE)) {
1075 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1076 return -ENODEV;
1077 }
1078
1079 for_each_possible_cpu(cpu) {
1080 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1081 struct sk_buff *skb;
1082
1083 INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1084 timer_setup(&data->send_timer, sched_send_work, 0);
1085 /* Allocate a new per-CPU skb for the summary alert message and
1086 * free the old one which might contain stale data from
1087 * previous tracing.
1088 */
1089 skb = reset_per_cpu_data(data);
1090 consume_skb(skb);
1091 }
1092
1093 rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1094 if (rc) {
1095 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1096 goto err_module_put;
1097 }
1098
1099 rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1100 if (rc) {
1101 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1102 goto err_unregister_trace;
1103 }
1104
1105 return 0;
1106
1107err_unregister_trace:
1108 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1109err_module_put:
1110 module_put(THIS_MODULE);
1111 return rc;
1112}
1113
1114static void net_dm_trace_off_set(void)
1115{
1116 struct dm_hw_stat_delta *new_stat, *temp;
1117 const struct net_dm_alert_ops *ops;
1118 int cpu;
1119
1120 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1121
1122 unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1123 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1124
1125 tracepoint_synchronize_unregister();
1126
1127 /* Make sure we do not send notifications to user space after request
1128 * to stop tracing returns.
1129 */
1130 for_each_possible_cpu(cpu) {
1131 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1132 struct sk_buff *skb;
1133
1134 del_timer_sync(&data->send_timer);
1135 cancel_work_sync(&data->dm_alert_work);
1136 while ((skb = __skb_dequeue(&data->drop_queue)))
1137 consume_skb(skb);
1138 }
1139
1140 list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1141 if (new_stat->dev == NULL) {
1142 list_del_rcu(&new_stat->list);
1143 kfree_rcu(new_stat, rcu);
1144 }
1145 }
1146
1147 module_put(THIS_MODULE);
1148}
1149
1150static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1151{
1152 int rc = 0;
1153
1154 if (state == trace_state) {
1155 NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1156 return -EAGAIN;
1157 }
1158
1159 switch (state) {
1160 case TRACE_ON:
1161 rc = net_dm_trace_on_set(extack);
1162 break;
1163 case TRACE_OFF:
1164 net_dm_trace_off_set();
1165 break;
1166 default:
1167 rc = 1;
1168 break;
1169 }
1170
1171 if (!rc)
1172 trace_state = state;
1173 else
1174 rc = -EINPROGRESS;
1175
1176 return rc;
1177}
1178
1179static bool net_dm_is_monitoring(void)
1180{
1181 return trace_state == TRACE_ON || monitor_hw;
1182}
1183
1184static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1185 enum net_dm_alert_mode *p_alert_mode)
1186{
1187 u8 val;
1188
1189 val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1190
1191 switch (val) {
1192 case NET_DM_ALERT_MODE_SUMMARY:
1193 case NET_DM_ALERT_MODE_PACKET:
1194 *p_alert_mode = val;
1195 break;
1196 default:
1197 return -EINVAL;
1198 }
1199
1200 return 0;
1201}
1202
1203static int net_dm_alert_mode_set(struct genl_info *info)
1204{
1205 struct netlink_ext_ack *extack = info->extack;
1206 enum net_dm_alert_mode alert_mode;
1207 int rc;
1208
1209 if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1210 return 0;
1211
1212 rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1213 if (rc) {
1214 NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1215 return -EINVAL;
1216 }
1217
1218 net_dm_alert_mode = alert_mode;
1219
1220 return 0;
1221}
1222
1223static void net_dm_trunc_len_set(struct genl_info *info)
1224{
1225 if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1226 return;
1227
1228 net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1229}
1230
1231static void net_dm_queue_len_set(struct genl_info *info)
1232{
1233 if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1234 return;
1235
1236 net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1237}
1238
1239static int net_dm_cmd_config(struct sk_buff *skb,
1240 struct genl_info *info)
1241{
1242 struct netlink_ext_ack *extack = info->extack;
1243 int rc;
1244
1245 if (net_dm_is_monitoring()) {
1246 NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1247 return -EBUSY;
1248 }
1249
1250 rc = net_dm_alert_mode_set(info);
1251 if (rc)
1252 return rc;
1253
1254 net_dm_trunc_len_set(info);
1255
1256 net_dm_queue_len_set(info);
1257
1258 return 0;
1259}
1260
1261static int net_dm_monitor_start(bool set_sw, bool set_hw,
1262 struct netlink_ext_ack *extack)
1263{
1264 bool sw_set = false;
1265 int rc;
1266
1267 if (set_sw) {
1268 rc = set_all_monitor_traces(TRACE_ON, extack);
1269 if (rc)
1270 return rc;
1271 sw_set = true;
1272 }
1273
1274 if (set_hw) {
1275 rc = net_dm_hw_monitor_start(extack);
1276 if (rc)
1277 goto err_monitor_hw;
1278 }
1279
1280 return 0;
1281
1282err_monitor_hw:
1283 if (sw_set)
1284 set_all_monitor_traces(TRACE_OFF, extack);
1285 return rc;
1286}
1287
1288static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1289 struct netlink_ext_ack *extack)
1290{
1291 if (set_hw)
1292 net_dm_hw_monitor_stop(extack);
1293 if (set_sw)
1294 set_all_monitor_traces(TRACE_OFF, extack);
1295}
1296
1297static int net_dm_cmd_trace(struct sk_buff *skb,
1298 struct genl_info *info)
1299{
1300 bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1301 bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1302 struct netlink_ext_ack *extack = info->extack;
1303
1304 /* To maintain backward compatibility, we start / stop monitoring of
1305 * software drops if no flag is specified.
1306 */
1307 if (!set_sw && !set_hw)
1308 set_sw = true;
1309
1310 switch (info->genlhdr->cmd) {
1311 case NET_DM_CMD_START:
1312 return net_dm_monitor_start(set_sw, set_hw, extack);
1313 case NET_DM_CMD_STOP:
1314 net_dm_monitor_stop(set_sw, set_hw, extack);
1315 return 0;
1316 }
1317
1318 return -EOPNOTSUPP;
1319}
1320
1321static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1322{
1323 void *hdr;
1324
1325 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1326 &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1327 if (!hdr)
1328 return -EMSGSIZE;
1329
1330 if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1331 goto nla_put_failure;
1332
1333 if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1334 goto nla_put_failure;
1335
1336 if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1337 goto nla_put_failure;
1338
1339 genlmsg_end(msg, hdr);
1340
1341 return 0;
1342
1343nla_put_failure:
1344 genlmsg_cancel(msg, hdr);
1345 return -EMSGSIZE;
1346}
1347
1348static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1349{
1350 struct sk_buff *msg;
1351 int rc;
1352
1353 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1354 if (!msg)
1355 return -ENOMEM;
1356
1357 rc = net_dm_config_fill(msg, info);
1358 if (rc)
1359 goto free_msg;
1360
1361 return genlmsg_reply(msg, info);
1362
1363free_msg:
1364 nlmsg_free(msg);
1365 return rc;
1366}
1367
1368static void net_dm_stats_read(struct net_dm_stats *stats)
1369{
1370 int cpu;
1371
1372 memset(stats, 0, sizeof(*stats));
1373 for_each_possible_cpu(cpu) {
1374 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1375 struct net_dm_stats *cpu_stats = &data->stats;
1376 unsigned int start;
1377 u64 dropped;
1378
1379 do {
1380 start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1381 dropped = cpu_stats->dropped;
1382 } while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1383
1384 stats->dropped += dropped;
1385 }
1386}
1387
1388static int net_dm_stats_put(struct sk_buff *msg)
1389{
1390 struct net_dm_stats stats;
1391 struct nlattr *attr;
1392
1393 net_dm_stats_read(&stats);
1394
1395 attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1396 if (!attr)
1397 return -EMSGSIZE;
1398
1399 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1400 stats.dropped, NET_DM_ATTR_PAD))
1401 goto nla_put_failure;
1402
1403 nla_nest_end(msg, attr);
1404
1405 return 0;
1406
1407nla_put_failure:
1408 nla_nest_cancel(msg, attr);
1409 return -EMSGSIZE;
1410}
1411
1412static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1413{
1414 int cpu;
1415
1416 memset(stats, 0, sizeof(*stats));
1417 for_each_possible_cpu(cpu) {
1418 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1419 struct net_dm_stats *cpu_stats = &hw_data->stats;
1420 unsigned int start;
1421 u64 dropped;
1422
1423 do {
1424 start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1425 dropped = cpu_stats->dropped;
1426 } while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1427
1428 stats->dropped += dropped;
1429 }
1430}
1431
1432static int net_dm_hw_stats_put(struct sk_buff *msg)
1433{
1434 struct net_dm_stats stats;
1435 struct nlattr *attr;
1436
1437 net_dm_hw_stats_read(&stats);
1438
1439 attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1440 if (!attr)
1441 return -EMSGSIZE;
1442
1443 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1444 stats.dropped, NET_DM_ATTR_PAD))
1445 goto nla_put_failure;
1446
1447 nla_nest_end(msg, attr);
1448
1449 return 0;
1450
1451nla_put_failure:
1452 nla_nest_cancel(msg, attr);
1453 return -EMSGSIZE;
1454}
1455
1456static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1457{
1458 void *hdr;
1459 int rc;
1460
1461 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1462 &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1463 if (!hdr)
1464 return -EMSGSIZE;
1465
1466 rc = net_dm_stats_put(msg);
1467 if (rc)
1468 goto nla_put_failure;
1469
1470 rc = net_dm_hw_stats_put(msg);
1471 if (rc)
1472 goto nla_put_failure;
1473
1474 genlmsg_end(msg, hdr);
1475
1476 return 0;
1477
1478nla_put_failure:
1479 genlmsg_cancel(msg, hdr);
1480 return -EMSGSIZE;
1481}
1482
1483static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1484{
1485 struct sk_buff *msg;
1486 int rc;
1487
1488 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1489 if (!msg)
1490 return -ENOMEM;
1491
1492 rc = net_dm_stats_fill(msg, info);
1493 if (rc)
1494 goto free_msg;
1495
1496 return genlmsg_reply(msg, info);
1497
1498free_msg:
1499 nlmsg_free(msg);
1500 return rc;
1501}
1502
1503static int dropmon_net_event(struct notifier_block *ev_block,
1504 unsigned long event, void *ptr)
1505{
1506 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1507 struct dm_hw_stat_delta *new_stat = NULL;
1508 struct dm_hw_stat_delta *tmp;
1509
1510 switch (event) {
1511 case NETDEV_REGISTER:
1512 new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
1513
1514 if (!new_stat)
1515 goto out;
1516
1517 new_stat->dev = dev;
1518 new_stat->last_rx = jiffies;
1519 mutex_lock(&net_dm_mutex);
1520 list_add_rcu(&new_stat->list, &hw_stats_list);
1521 mutex_unlock(&net_dm_mutex);
1522 break;
1523 case NETDEV_UNREGISTER:
1524 mutex_lock(&net_dm_mutex);
1525 list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1526 if (new_stat->dev == dev) {
1527 new_stat->dev = NULL;
1528 if (trace_state == TRACE_OFF) {
1529 list_del_rcu(&new_stat->list);
1530 kfree_rcu(new_stat, rcu);
1531 break;
1532 }
1533 }
1534 }
1535 mutex_unlock(&net_dm_mutex);
1536 break;
1537 }
1538out:
1539 return NOTIFY_DONE;
1540}
1541
1542static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1543 [NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1544 [NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1545 [NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1546 [NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1547 [NET_DM_ATTR_SW_DROPS] = {. type = NLA_FLAG },
1548 [NET_DM_ATTR_HW_DROPS] = {. type = NLA_FLAG },
1549};
1550
1551static const struct genl_ops dropmon_ops[] = {
1552 {
1553 .cmd = NET_DM_CMD_CONFIG,
1554 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1555 .doit = net_dm_cmd_config,
1556 .flags = GENL_ADMIN_PERM,
1557 },
1558 {
1559 .cmd = NET_DM_CMD_START,
1560 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1561 .doit = net_dm_cmd_trace,
1562 },
1563 {
1564 .cmd = NET_DM_CMD_STOP,
1565 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1566 .doit = net_dm_cmd_trace,
1567 },
1568 {
1569 .cmd = NET_DM_CMD_CONFIG_GET,
1570 .doit = net_dm_cmd_config_get,
1571 },
1572 {
1573 .cmd = NET_DM_CMD_STATS_GET,
1574 .doit = net_dm_cmd_stats_get,
1575 },
1576};
1577
1578static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1579 struct sk_buff *skb, struct genl_info *info)
1580{
1581 mutex_lock(&net_dm_mutex);
1582
1583 return 0;
1584}
1585
1586static void net_dm_nl_post_doit(const struct genl_ops *ops,
1587 struct sk_buff *skb, struct genl_info *info)
1588{
1589 mutex_unlock(&net_dm_mutex);
1590}
1591
1592static struct genl_family net_drop_monitor_family __ro_after_init = {
1593 .hdrsize = 0,
1594 .name = "NET_DM",
1595 .version = 2,
1596 .maxattr = NET_DM_ATTR_MAX,
1597 .policy = net_dm_nl_policy,
1598 .pre_doit = net_dm_nl_pre_doit,
1599 .post_doit = net_dm_nl_post_doit,
1600 .module = THIS_MODULE,
1601 .ops = dropmon_ops,
1602 .n_ops = ARRAY_SIZE(dropmon_ops),
1603 .mcgrps = dropmon_mcgrps,
1604 .n_mcgrps = ARRAY_SIZE(dropmon_mcgrps),
1605};
1606
1607static struct notifier_block dropmon_net_notifier = {
1608 .notifier_call = dropmon_net_event
1609};
1610
1611static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1612{
1613 spin_lock_init(&data->lock);
1614 skb_queue_head_init(&data->drop_queue);
1615 u64_stats_init(&data->stats.syncp);
1616}
1617
1618static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1619{
1620 WARN_ON(!skb_queue_empty(&data->drop_queue));
1621}
1622
1623static void net_dm_cpu_data_init(int cpu)
1624{
1625 struct per_cpu_dm_data *data;
1626
1627 data = &per_cpu(dm_cpu_data, cpu);
1628 __net_dm_cpu_data_init(data);
1629}
1630
1631static void net_dm_cpu_data_fini(int cpu)
1632{
1633 struct per_cpu_dm_data *data;
1634
1635 data = &per_cpu(dm_cpu_data, cpu);
1636 /* At this point, we should have exclusive access
1637 * to this struct and can free the skb inside it.
1638 */
1639 consume_skb(data->skb);
1640 __net_dm_cpu_data_fini(data);
1641}
1642
1643static void net_dm_hw_cpu_data_init(int cpu)
1644{
1645 struct per_cpu_dm_data *hw_data;
1646
1647 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1648 __net_dm_cpu_data_init(hw_data);
1649}
1650
1651static void net_dm_hw_cpu_data_fini(int cpu)
1652{
1653 struct per_cpu_dm_data *hw_data;
1654
1655 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1656 kfree(hw_data->hw_entries);
1657 __net_dm_cpu_data_fini(hw_data);
1658}
1659
1660static int __init init_net_drop_monitor(void)
1661{
1662 int cpu, rc;
1663
1664 pr_info("Initializing network drop monitor service\n");
1665
1666 if (sizeof(void *) > 8) {
1667 pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1668 return -ENOSPC;
1669 }
1670
1671 rc = genl_register_family(&net_drop_monitor_family);
1672 if (rc) {
1673 pr_err("Could not create drop monitor netlink family\n");
1674 return rc;
1675 }
1676 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1677
1678 rc = register_netdevice_notifier(&dropmon_net_notifier);
1679 if (rc < 0) {
1680 pr_crit("Failed to register netdevice notifier\n");
1681 goto out_unreg;
1682 }
1683
1684 rc = 0;
1685
1686 for_each_possible_cpu(cpu) {
1687 net_dm_cpu_data_init(cpu);
1688 net_dm_hw_cpu_data_init(cpu);
1689 }
1690
1691 goto out;
1692
1693out_unreg:
1694 genl_unregister_family(&net_drop_monitor_family);
1695out:
1696 return rc;
1697}
1698
1699static void exit_net_drop_monitor(void)
1700{
1701 int cpu;
1702
1703 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1704
1705 /*
1706 * Because of the module_get/put we do in the trace state change path
1707 * we are guarnateed not to have any current users when we get here
1708 */
1709
1710 for_each_possible_cpu(cpu) {
1711 net_dm_hw_cpu_data_fini(cpu);
1712 net_dm_cpu_data_fini(cpu);
1713 }
1714
1715 BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1716}
1717
1718module_init(init_net_drop_monitor);
1719module_exit(exit_net_drop_monitor);
1720
1721MODULE_LICENSE("GPL v2");
1722MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1723MODULE_ALIAS_GENL_FAMILY("NET_DM");
1724MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");