This source file includes following definitions.
- reset_per_cpu_data
- send_dm_alert
- sched_send_work
- trace_drop_common
- trace_kfree_skb_hit
- trace_napi_poll_hit
- net_dm_hw_reset_per_cpu_data
- net_dm_hw_entry_put
- net_dm_hw_entries_put
- net_dm_hw_summary_report_fill
- net_dm_hw_summary_work
- net_dm_hw_summary_probe
- net_dm_packet_trace_kfree_skb_hit
- net_dm_packet_trace_napi_poll_hit
- net_dm_in_port_size
- net_dm_packet_report_size
- net_dm_packet_report_in_port_put
- net_dm_packet_report_fill
- net_dm_packet_report
- net_dm_packet_work
- net_dm_hw_packet_report_size
- net_dm_hw_packet_report_fill
- net_dm_hw_metadata_clone
- net_dm_hw_metadata_free
- net_dm_hw_packet_report
- net_dm_hw_packet_work
- net_dm_hw_packet_probe
- net_dm_hw_report
- net_dm_hw_monitor_start
- net_dm_hw_monitor_stop
- net_dm_trace_on_set
- net_dm_trace_off_set
- set_all_monitor_traces
- net_dm_is_monitoring
- net_dm_alert_mode_get_from_info
- net_dm_alert_mode_set
- net_dm_trunc_len_set
- net_dm_queue_len_set
- net_dm_cmd_config
- net_dm_monitor_start
- net_dm_monitor_stop
- net_dm_cmd_trace
- net_dm_config_fill
- net_dm_cmd_config_get
- net_dm_stats_read
- net_dm_stats_put
- net_dm_hw_stats_read
- net_dm_hw_stats_put
- net_dm_stats_fill
- net_dm_cmd_stats_get
- dropmon_net_event
- net_dm_nl_pre_doit
- net_dm_nl_post_doit
- __net_dm_cpu_data_init
- __net_dm_cpu_data_fini
- net_dm_cpu_data_init
- net_dm_cpu_data_fini
- net_dm_hw_cpu_data_init
- net_dm_hw_cpu_data_fini
- init_net_drop_monitor
- exit_net_drop_monitor
1
2
3
4
5
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
33 #include <trace/events/skb.h>
34 #include <trace/events/napi.h>
35
36 #include <asm/unaligned.h>
37
38 #define TRACE_ON 1
39 #define TRACE_OFF 0
40
41
42
43
44
45
46 static int trace_state = TRACE_OFF;
47 static bool monitor_hw;
48
49
50
51
52
53
54 static DEFINE_MUTEX(net_dm_mutex);
55
56 struct net_dm_stats {
57 u64 dropped;
58 struct u64_stats_sync syncp;
59 };
60
61 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
62
63 struct net_dm_hw_entry {
64 char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
65 u32 count;
66 };
67
68 struct net_dm_hw_entries {
69 u32 num_entries;
70 struct net_dm_hw_entry entries[0];
71 };
72
73 struct per_cpu_dm_data {
74 spinlock_t lock;
75
76
77 union {
78 struct sk_buff *skb;
79 struct net_dm_hw_entries *hw_entries;
80 };
81 struct sk_buff_head drop_queue;
82 struct work_struct dm_alert_work;
83 struct timer_list send_timer;
84 struct net_dm_stats stats;
85 };
86
87 struct dm_hw_stat_delta {
88 struct net_device *dev;
89 unsigned long last_rx;
90 struct list_head list;
91 struct rcu_head rcu;
92 unsigned long last_drop_val;
93 };
94
95 static struct genl_family net_drop_monitor_family;
96
97 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
99
100 static int dm_hit_limit = 64;
101 static int dm_delay = 1;
102 static unsigned long dm_hw_check_delta = 2*HZ;
103 static LIST_HEAD(hw_stats_list);
104
105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106 static u32 net_dm_trunc_len;
107 static u32 net_dm_queue_len = 1000;
108
109 struct net_dm_alert_ops {
110 void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
111 void *location);
112 void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
113 int work, int budget);
114 void (*work_item_func)(struct work_struct *work);
115 void (*hw_work_item_func)(struct work_struct *work);
116 void (*hw_probe)(struct sk_buff *skb,
117 const struct net_dm_hw_metadata *hw_metadata);
118 };
119
120 struct net_dm_skb_cb {
121 union {
122 struct net_dm_hw_metadata *hw_metadata;
123 void *pc;
124 };
125 };
126
127 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
128
129 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
130 {
131 size_t al;
132 struct net_dm_alert_msg *msg;
133 struct nlattr *nla;
134 struct sk_buff *skb;
135 unsigned long flags;
136 void *msg_header;
137
138 al = sizeof(struct net_dm_alert_msg);
139 al += dm_hit_limit * sizeof(struct net_dm_drop_point);
140 al += sizeof(struct nlattr);
141
142 skb = genlmsg_new(al, GFP_KERNEL);
143
144 if (!skb)
145 goto err;
146
147 msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
148 0, NET_DM_CMD_ALERT);
149 if (!msg_header) {
150 nlmsg_free(skb);
151 skb = NULL;
152 goto err;
153 }
154 nla = nla_reserve(skb, NLA_UNSPEC,
155 sizeof(struct net_dm_alert_msg));
156 if (!nla) {
157 nlmsg_free(skb);
158 skb = NULL;
159 goto err;
160 }
161 msg = nla_data(nla);
162 memset(msg, 0, al);
163 goto out;
164
165 err:
166 mod_timer(&data->send_timer, jiffies + HZ / 10);
167 out:
168 spin_lock_irqsave(&data->lock, flags);
169 swap(data->skb, skb);
170 spin_unlock_irqrestore(&data->lock, flags);
171
172 if (skb) {
173 struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
174 struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
175
176 genlmsg_end(skb, genlmsg_data(gnlh));
177 }
178
179 return skb;
180 }
181
182 static const struct genl_multicast_group dropmon_mcgrps[] = {
183 { .name = "events", },
184 };
185
186 static void send_dm_alert(struct work_struct *work)
187 {
188 struct sk_buff *skb;
189 struct per_cpu_dm_data *data;
190
191 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
192
193 skb = reset_per_cpu_data(data);
194
195 if (skb)
196 genlmsg_multicast(&net_drop_monitor_family, skb, 0,
197 0, GFP_KERNEL);
198 }
199
200
201
202
203
204
205 static void sched_send_work(struct timer_list *t)
206 {
207 struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
208
209 schedule_work(&data->dm_alert_work);
210 }
211
212 static void trace_drop_common(struct sk_buff *skb, void *location)
213 {
214 struct net_dm_alert_msg *msg;
215 struct net_dm_drop_point *point;
216 struct nlmsghdr *nlh;
217 struct nlattr *nla;
218 int i;
219 struct sk_buff *dskb;
220 struct per_cpu_dm_data *data;
221 unsigned long flags;
222
223 local_irq_save(flags);
224 data = this_cpu_ptr(&dm_cpu_data);
225 spin_lock(&data->lock);
226 dskb = data->skb;
227
228 if (!dskb)
229 goto out;
230
231 nlh = (struct nlmsghdr *)dskb->data;
232 nla = genlmsg_data(nlmsg_data(nlh));
233 msg = nla_data(nla);
234 point = msg->points;
235 for (i = 0; i < msg->entries; i++) {
236 if (!memcmp(&location, &point->pc, sizeof(void *))) {
237 point->count++;
238 goto out;
239 }
240 point++;
241 }
242 if (msg->entries == dm_hit_limit)
243 goto out;
244
245
246
247 __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
248 nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
249 memcpy(point->pc, &location, sizeof(void *));
250 point->count = 1;
251 msg->entries++;
252
253 if (!timer_pending(&data->send_timer)) {
254 data->send_timer.expires = jiffies + dm_delay * HZ;
255 add_timer(&data->send_timer);
256 }
257
258 out:
259 spin_unlock_irqrestore(&data->lock, flags);
260 }
261
262 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
263 {
264 trace_drop_common(skb, location);
265 }
266
267 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
268 int work, int budget)
269 {
270 struct dm_hw_stat_delta *new_stat;
271
272
273
274
275 if (!napi->dev)
276 return;
277
278 rcu_read_lock();
279 list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
280
281
282
283
284
285
286 if ((new_stat->dev == napi->dev) &&
287 (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
288 (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
289 trace_drop_common(NULL, NULL);
290 new_stat->last_drop_val = napi->dev->stats.rx_dropped;
291 new_stat->last_rx = jiffies;
292 break;
293 }
294 }
295 rcu_read_unlock();
296 }
297
298 static struct net_dm_hw_entries *
299 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
300 {
301 struct net_dm_hw_entries *hw_entries;
302 unsigned long flags;
303
304 hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
305 GFP_KERNEL);
306 if (!hw_entries) {
307
308
309
310
311 mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
312 }
313
314 spin_lock_irqsave(&hw_data->lock, flags);
315 swap(hw_data->hw_entries, hw_entries);
316 spin_unlock_irqrestore(&hw_data->lock, flags);
317
318 return hw_entries;
319 }
320
321 static int net_dm_hw_entry_put(struct sk_buff *msg,
322 const struct net_dm_hw_entry *hw_entry)
323 {
324 struct nlattr *attr;
325
326 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
327 if (!attr)
328 return -EMSGSIZE;
329
330 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
331 goto nla_put_failure;
332
333 if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
334 goto nla_put_failure;
335
336 nla_nest_end(msg, attr);
337
338 return 0;
339
340 nla_put_failure:
341 nla_nest_cancel(msg, attr);
342 return -EMSGSIZE;
343 }
344
345 static int net_dm_hw_entries_put(struct sk_buff *msg,
346 const struct net_dm_hw_entries *hw_entries)
347 {
348 struct nlattr *attr;
349 int i;
350
351 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
352 if (!attr)
353 return -EMSGSIZE;
354
355 for (i = 0; i < hw_entries->num_entries; i++) {
356 int rc;
357
358 rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
359 if (rc)
360 goto nla_put_failure;
361 }
362
363 nla_nest_end(msg, attr);
364
365 return 0;
366
367 nla_put_failure:
368 nla_nest_cancel(msg, attr);
369 return -EMSGSIZE;
370 }
371
372 static int
373 net_dm_hw_summary_report_fill(struct sk_buff *msg,
374 const struct net_dm_hw_entries *hw_entries)
375 {
376 struct net_dm_alert_msg anc_hdr = { 0 };
377 void *hdr;
378 int rc;
379
380 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
381 NET_DM_CMD_ALERT);
382 if (!hdr)
383 return -EMSGSIZE;
384
385
386
387
388 if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
389 goto nla_put_failure;
390
391 rc = net_dm_hw_entries_put(msg, hw_entries);
392 if (rc)
393 goto nla_put_failure;
394
395 genlmsg_end(msg, hdr);
396
397 return 0;
398
399 nla_put_failure:
400 genlmsg_cancel(msg, hdr);
401 return -EMSGSIZE;
402 }
403
404 static void net_dm_hw_summary_work(struct work_struct *work)
405 {
406 struct net_dm_hw_entries *hw_entries;
407 struct per_cpu_dm_data *hw_data;
408 struct sk_buff *msg;
409 int rc;
410
411 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
412
413 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
414 if (!hw_entries)
415 return;
416
417 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
418 if (!msg)
419 goto out;
420
421 rc = net_dm_hw_summary_report_fill(msg, hw_entries);
422 if (rc) {
423 nlmsg_free(msg);
424 goto out;
425 }
426
427 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
428
429 out:
430 kfree(hw_entries);
431 }
432
433 static void
434 net_dm_hw_summary_probe(struct sk_buff *skb,
435 const struct net_dm_hw_metadata *hw_metadata)
436 {
437 struct net_dm_hw_entries *hw_entries;
438 struct net_dm_hw_entry *hw_entry;
439 struct per_cpu_dm_data *hw_data;
440 unsigned long flags;
441 int i;
442
443 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
444 spin_lock_irqsave(&hw_data->lock, flags);
445 hw_entries = hw_data->hw_entries;
446
447 if (!hw_entries)
448 goto out;
449
450 for (i = 0; i < hw_entries->num_entries; i++) {
451 hw_entry = &hw_entries->entries[i];
452 if (!strncmp(hw_entry->trap_name, hw_metadata->trap_name,
453 NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
454 hw_entry->count++;
455 goto out;
456 }
457 }
458 if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
459 goto out;
460
461 hw_entry = &hw_entries->entries[hw_entries->num_entries];
462 strlcpy(hw_entry->trap_name, hw_metadata->trap_name,
463 NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
464 hw_entry->count = 1;
465 hw_entries->num_entries++;
466
467 if (!timer_pending(&hw_data->send_timer)) {
468 hw_data->send_timer.expires = jiffies + dm_delay * HZ;
469 add_timer(&hw_data->send_timer);
470 }
471
472 out:
473 spin_unlock_irqrestore(&hw_data->lock, flags);
474 }
475
476 static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
477 .kfree_skb_probe = trace_kfree_skb_hit,
478 .napi_poll_probe = trace_napi_poll_hit,
479 .work_item_func = send_dm_alert,
480 .hw_work_item_func = net_dm_hw_summary_work,
481 .hw_probe = net_dm_hw_summary_probe,
482 };
483
484 static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
485 struct sk_buff *skb,
486 void *location)
487 {
488 ktime_t tstamp = ktime_get_real();
489 struct per_cpu_dm_data *data;
490 struct sk_buff *nskb;
491 unsigned long flags;
492
493 if (!skb_mac_header_was_set(skb))
494 return;
495
496 nskb = skb_clone(skb, GFP_ATOMIC);
497 if (!nskb)
498 return;
499
500 NET_DM_SKB_CB(nskb)->pc = location;
501
502
503
504 nskb->tstamp = tstamp;
505
506 data = this_cpu_ptr(&dm_cpu_data);
507
508 spin_lock_irqsave(&data->drop_queue.lock, flags);
509 if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
510 __skb_queue_tail(&data->drop_queue, nskb);
511 else
512 goto unlock_free;
513 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
514
515 schedule_work(&data->dm_alert_work);
516
517 return;
518
519 unlock_free:
520 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
521 u64_stats_update_begin(&data->stats.syncp);
522 data->stats.dropped++;
523 u64_stats_update_end(&data->stats.syncp);
524 consume_skb(nskb);
525 }
526
527 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
528 struct napi_struct *napi,
529 int work, int budget)
530 {
531 }
532
533 static size_t net_dm_in_port_size(void)
534 {
535
536 return nla_total_size(0) +
537
538 nla_total_size(sizeof(u32)) +
539
540 nla_total_size(IFNAMSIZ + 1);
541 }
542
543 #define NET_DM_MAX_SYMBOL_LEN 40
544
545 static size_t net_dm_packet_report_size(size_t payload_len)
546 {
547 size_t size;
548
549 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
550
551 return NLMSG_ALIGN(size) +
552
553 nla_total_size(sizeof(u16)) +
554
555 nla_total_size(sizeof(u64)) +
556
557 nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
558
559 net_dm_in_port_size() +
560
561 nla_total_size(sizeof(u64)) +
562
563 nla_total_size(sizeof(u32)) +
564
565 nla_total_size(sizeof(u16)) +
566
567 nla_total_size(payload_len);
568 }
569
570 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
571 const char *name)
572 {
573 struct nlattr *attr;
574
575 attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
576 if (!attr)
577 return -EMSGSIZE;
578
579 if (ifindex &&
580 nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
581 goto nla_put_failure;
582
583 if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
584 goto nla_put_failure;
585
586 nla_nest_end(msg, attr);
587
588 return 0;
589
590 nla_put_failure:
591 nla_nest_cancel(msg, attr);
592 return -EMSGSIZE;
593 }
594
595 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
596 size_t payload_len)
597 {
598 u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
599 char buf[NET_DM_MAX_SYMBOL_LEN];
600 struct nlattr *attr;
601 void *hdr;
602 int rc;
603
604 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
605 NET_DM_CMD_PACKET_ALERT);
606 if (!hdr)
607 return -EMSGSIZE;
608
609 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
610 goto nla_put_failure;
611
612 if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
613 goto nla_put_failure;
614
615 snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
616 if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
617 goto nla_put_failure;
618
619 rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
620 if (rc)
621 goto nla_put_failure;
622
623 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
624 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
625 goto nla_put_failure;
626
627 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
628 goto nla_put_failure;
629
630 if (!payload_len)
631 goto out;
632
633 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
634 goto nla_put_failure;
635
636 attr = skb_put(msg, nla_total_size(payload_len));
637 attr->nla_type = NET_DM_ATTR_PAYLOAD;
638 attr->nla_len = nla_attr_size(payload_len);
639 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
640 goto nla_put_failure;
641
642 out:
643 genlmsg_end(msg, hdr);
644
645 return 0;
646
647 nla_put_failure:
648 genlmsg_cancel(msg, hdr);
649 return -EMSGSIZE;
650 }
651
652 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
653
654 static void net_dm_packet_report(struct sk_buff *skb)
655 {
656 struct sk_buff *msg;
657 size_t payload_len;
658 int rc;
659
660
661 if (skb->data > skb_mac_header(skb))
662 skb_push(skb, skb->data - skb_mac_header(skb));
663 else
664 skb_pull(skb, skb_mac_header(skb) - skb->data);
665
666
667 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
668 if (net_dm_trunc_len)
669 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
670
671 msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
672 if (!msg)
673 goto out;
674
675 rc = net_dm_packet_report_fill(msg, skb, payload_len);
676 if (rc) {
677 nlmsg_free(msg);
678 goto out;
679 }
680
681 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
682
683 out:
684 consume_skb(skb);
685 }
686
687 static void net_dm_packet_work(struct work_struct *work)
688 {
689 struct per_cpu_dm_data *data;
690 struct sk_buff_head list;
691 struct sk_buff *skb;
692 unsigned long flags;
693
694 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
695
696 __skb_queue_head_init(&list);
697
698 spin_lock_irqsave(&data->drop_queue.lock, flags);
699 skb_queue_splice_tail_init(&data->drop_queue, &list);
700 spin_unlock_irqrestore(&data->drop_queue.lock, flags);
701
702 while ((skb = __skb_dequeue(&list)))
703 net_dm_packet_report(skb);
704 }
705
706 static size_t
707 net_dm_hw_packet_report_size(size_t payload_len,
708 const struct net_dm_hw_metadata *hw_metadata)
709 {
710 size_t size;
711
712 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
713
714 return NLMSG_ALIGN(size) +
715
716 nla_total_size(sizeof(u16)) +
717
718 nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
719
720 nla_total_size(strlen(hw_metadata->trap_name) + 1) +
721
722 net_dm_in_port_size() +
723
724 nla_total_size(sizeof(u64)) +
725
726 nla_total_size(sizeof(u32)) +
727
728 nla_total_size(sizeof(u16)) +
729
730 nla_total_size(payload_len);
731 }
732
733 static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
734 struct sk_buff *skb, size_t payload_len)
735 {
736 struct net_dm_hw_metadata *hw_metadata;
737 struct nlattr *attr;
738 void *hdr;
739
740 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
741
742 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
743 NET_DM_CMD_PACKET_ALERT);
744 if (!hdr)
745 return -EMSGSIZE;
746
747 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
748 goto nla_put_failure;
749
750 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
751 hw_metadata->trap_group_name))
752 goto nla_put_failure;
753
754 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
755 hw_metadata->trap_name))
756 goto nla_put_failure;
757
758 if (hw_metadata->input_dev) {
759 struct net_device *dev = hw_metadata->input_dev;
760 int rc;
761
762 rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
763 dev->name);
764 if (rc)
765 goto nla_put_failure;
766 }
767
768 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
769 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
770 goto nla_put_failure;
771
772 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
773 goto nla_put_failure;
774
775 if (!payload_len)
776 goto out;
777
778 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
779 goto nla_put_failure;
780
781 attr = skb_put(msg, nla_total_size(payload_len));
782 attr->nla_type = NET_DM_ATTR_PAYLOAD;
783 attr->nla_len = nla_attr_size(payload_len);
784 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
785 goto nla_put_failure;
786
787 out:
788 genlmsg_end(msg, hdr);
789
790 return 0;
791
792 nla_put_failure:
793 genlmsg_cancel(msg, hdr);
794 return -EMSGSIZE;
795 }
796
797 static struct net_dm_hw_metadata *
798 net_dm_hw_metadata_clone(const struct net_dm_hw_metadata *hw_metadata)
799 {
800 struct net_dm_hw_metadata *n_hw_metadata;
801 const char *trap_group_name;
802 const char *trap_name;
803
804 n_hw_metadata = kmalloc(sizeof(*hw_metadata), GFP_ATOMIC);
805 if (!n_hw_metadata)
806 return NULL;
807
808 trap_group_name = kmemdup(hw_metadata->trap_group_name,
809 strlen(hw_metadata->trap_group_name) + 1,
810 GFP_ATOMIC | __GFP_ZERO);
811 if (!trap_group_name)
812 goto free_hw_metadata;
813 n_hw_metadata->trap_group_name = trap_group_name;
814
815 trap_name = kmemdup(hw_metadata->trap_name,
816 strlen(hw_metadata->trap_name) + 1,
817 GFP_ATOMIC | __GFP_ZERO);
818 if (!trap_name)
819 goto free_trap_group;
820 n_hw_metadata->trap_name = trap_name;
821
822 n_hw_metadata->input_dev = hw_metadata->input_dev;
823 if (n_hw_metadata->input_dev)
824 dev_hold(n_hw_metadata->input_dev);
825
826 return n_hw_metadata;
827
828 free_trap_group:
829 kfree(trap_group_name);
830 free_hw_metadata:
831 kfree(n_hw_metadata);
832 return NULL;
833 }
834
835 static void
836 net_dm_hw_metadata_free(const struct net_dm_hw_metadata *hw_metadata)
837 {
838 if (hw_metadata->input_dev)
839 dev_put(hw_metadata->input_dev);
840 kfree(hw_metadata->trap_name);
841 kfree(hw_metadata->trap_group_name);
842 kfree(hw_metadata);
843 }
844
845 static void net_dm_hw_packet_report(struct sk_buff *skb)
846 {
847 struct net_dm_hw_metadata *hw_metadata;
848 struct sk_buff *msg;
849 size_t payload_len;
850 int rc;
851
852 if (skb->data > skb_mac_header(skb))
853 skb_push(skb, skb->data - skb_mac_header(skb));
854 else
855 skb_pull(skb, skb_mac_header(skb) - skb->data);
856
857 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
858 if (net_dm_trunc_len)
859 payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
860
861 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
862 msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
863 GFP_KERNEL);
864 if (!msg)
865 goto out;
866
867 rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
868 if (rc) {
869 nlmsg_free(msg);
870 goto out;
871 }
872
873 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
874
875 out:
876 net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
877 consume_skb(skb);
878 }
879
880 static void net_dm_hw_packet_work(struct work_struct *work)
881 {
882 struct per_cpu_dm_data *hw_data;
883 struct sk_buff_head list;
884 struct sk_buff *skb;
885 unsigned long flags;
886
887 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
888
889 __skb_queue_head_init(&list);
890
891 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
892 skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
893 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
894
895 while ((skb = __skb_dequeue(&list)))
896 net_dm_hw_packet_report(skb);
897 }
898
899 static void
900 net_dm_hw_packet_probe(struct sk_buff *skb,
901 const struct net_dm_hw_metadata *hw_metadata)
902 {
903 struct net_dm_hw_metadata *n_hw_metadata;
904 ktime_t tstamp = ktime_get_real();
905 struct per_cpu_dm_data *hw_data;
906 struct sk_buff *nskb;
907 unsigned long flags;
908
909 if (!skb_mac_header_was_set(skb))
910 return;
911
912 nskb = skb_clone(skb, GFP_ATOMIC);
913 if (!nskb)
914 return;
915
916 n_hw_metadata = net_dm_hw_metadata_clone(hw_metadata);
917 if (!n_hw_metadata)
918 goto free;
919
920 NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
921 nskb->tstamp = tstamp;
922
923 hw_data = this_cpu_ptr(&dm_hw_cpu_data);
924
925 spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
926 if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
927 __skb_queue_tail(&hw_data->drop_queue, nskb);
928 else
929 goto unlock_free;
930 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
931
932 schedule_work(&hw_data->dm_alert_work);
933
934 return;
935
936 unlock_free:
937 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
938 u64_stats_update_begin(&hw_data->stats.syncp);
939 hw_data->stats.dropped++;
940 u64_stats_update_end(&hw_data->stats.syncp);
941 net_dm_hw_metadata_free(n_hw_metadata);
942 free:
943 consume_skb(nskb);
944 }
945
946 static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
947 .kfree_skb_probe = net_dm_packet_trace_kfree_skb_hit,
948 .napi_poll_probe = net_dm_packet_trace_napi_poll_hit,
949 .work_item_func = net_dm_packet_work,
950 .hw_work_item_func = net_dm_hw_packet_work,
951 .hw_probe = net_dm_hw_packet_probe,
952 };
953
954 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
955 [NET_DM_ALERT_MODE_SUMMARY] = &net_dm_alert_summary_ops,
956 [NET_DM_ALERT_MODE_PACKET] = &net_dm_alert_packet_ops,
957 };
958
959 void net_dm_hw_report(struct sk_buff *skb,
960 const struct net_dm_hw_metadata *hw_metadata)
961 {
962 rcu_read_lock();
963
964 if (!monitor_hw)
965 goto out;
966
967 net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
968
969 out:
970 rcu_read_unlock();
971 }
972 EXPORT_SYMBOL_GPL(net_dm_hw_report);
973
974 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
975 {
976 const struct net_dm_alert_ops *ops;
977 int cpu;
978
979 if (monitor_hw) {
980 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
981 return -EAGAIN;
982 }
983
984 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
985
986 if (!try_module_get(THIS_MODULE)) {
987 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
988 return -ENODEV;
989 }
990
991 for_each_possible_cpu(cpu) {
992 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
993 struct net_dm_hw_entries *hw_entries;
994
995 INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
996 timer_setup(&hw_data->send_timer, sched_send_work, 0);
997 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
998 kfree(hw_entries);
999 }
1000
1001 monitor_hw = true;
1002
1003 return 0;
1004 }
1005
1006 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1007 {
1008 int cpu;
1009
1010 if (!monitor_hw) {
1011 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1012 return;
1013 }
1014
1015 monitor_hw = false;
1016
1017
1018
1019
1020 synchronize_rcu();
1021
1022 for_each_possible_cpu(cpu) {
1023 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1024 struct sk_buff *skb;
1025
1026 del_timer_sync(&hw_data->send_timer);
1027 cancel_work_sync(&hw_data->dm_alert_work);
1028 while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1029 struct net_dm_hw_metadata *hw_metadata;
1030
1031 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1032 net_dm_hw_metadata_free(hw_metadata);
1033 consume_skb(skb);
1034 }
1035 }
1036
1037 module_put(THIS_MODULE);
1038 }
1039
1040 static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1041 {
1042 const struct net_dm_alert_ops *ops;
1043 int cpu, rc;
1044
1045 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1046
1047 if (!try_module_get(THIS_MODULE)) {
1048 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1049 return -ENODEV;
1050 }
1051
1052 for_each_possible_cpu(cpu) {
1053 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1054 struct sk_buff *skb;
1055
1056 INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1057 timer_setup(&data->send_timer, sched_send_work, 0);
1058
1059
1060
1061
1062 skb = reset_per_cpu_data(data);
1063 consume_skb(skb);
1064 }
1065
1066 rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1067 if (rc) {
1068 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1069 goto err_module_put;
1070 }
1071
1072 rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1073 if (rc) {
1074 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1075 goto err_unregister_trace;
1076 }
1077
1078 return 0;
1079
1080 err_unregister_trace:
1081 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1082 err_module_put:
1083 module_put(THIS_MODULE);
1084 return rc;
1085 }
1086
1087 static void net_dm_trace_off_set(void)
1088 {
1089 struct dm_hw_stat_delta *new_stat, *temp;
1090 const struct net_dm_alert_ops *ops;
1091 int cpu;
1092
1093 ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1094
1095 unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1096 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1097
1098 tracepoint_synchronize_unregister();
1099
1100
1101
1102
1103 for_each_possible_cpu(cpu) {
1104 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1105 struct sk_buff *skb;
1106
1107 del_timer_sync(&data->send_timer);
1108 cancel_work_sync(&data->dm_alert_work);
1109 while ((skb = __skb_dequeue(&data->drop_queue)))
1110 consume_skb(skb);
1111 }
1112
1113 list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1114 if (new_stat->dev == NULL) {
1115 list_del_rcu(&new_stat->list);
1116 kfree_rcu(new_stat, rcu);
1117 }
1118 }
1119
1120 module_put(THIS_MODULE);
1121 }
1122
1123 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1124 {
1125 int rc = 0;
1126
1127 if (state == trace_state) {
1128 NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1129 return -EAGAIN;
1130 }
1131
1132 switch (state) {
1133 case TRACE_ON:
1134 rc = net_dm_trace_on_set(extack);
1135 break;
1136 case TRACE_OFF:
1137 net_dm_trace_off_set();
1138 break;
1139 default:
1140 rc = 1;
1141 break;
1142 }
1143
1144 if (!rc)
1145 trace_state = state;
1146 else
1147 rc = -EINPROGRESS;
1148
1149 return rc;
1150 }
1151
1152 static bool net_dm_is_monitoring(void)
1153 {
1154 return trace_state == TRACE_ON || monitor_hw;
1155 }
1156
1157 static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1158 enum net_dm_alert_mode *p_alert_mode)
1159 {
1160 u8 val;
1161
1162 val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1163
1164 switch (val) {
1165 case NET_DM_ALERT_MODE_SUMMARY:
1166 case NET_DM_ALERT_MODE_PACKET:
1167 *p_alert_mode = val;
1168 break;
1169 default:
1170 return -EINVAL;
1171 }
1172
1173 return 0;
1174 }
1175
1176 static int net_dm_alert_mode_set(struct genl_info *info)
1177 {
1178 struct netlink_ext_ack *extack = info->extack;
1179 enum net_dm_alert_mode alert_mode;
1180 int rc;
1181
1182 if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1183 return 0;
1184
1185 rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1186 if (rc) {
1187 NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1188 return -EINVAL;
1189 }
1190
1191 net_dm_alert_mode = alert_mode;
1192
1193 return 0;
1194 }
1195
1196 static void net_dm_trunc_len_set(struct genl_info *info)
1197 {
1198 if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1199 return;
1200
1201 net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1202 }
1203
1204 static void net_dm_queue_len_set(struct genl_info *info)
1205 {
1206 if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1207 return;
1208
1209 net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1210 }
1211
1212 static int net_dm_cmd_config(struct sk_buff *skb,
1213 struct genl_info *info)
1214 {
1215 struct netlink_ext_ack *extack = info->extack;
1216 int rc;
1217
1218 if (net_dm_is_monitoring()) {
1219 NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1220 return -EBUSY;
1221 }
1222
1223 rc = net_dm_alert_mode_set(info);
1224 if (rc)
1225 return rc;
1226
1227 net_dm_trunc_len_set(info);
1228
1229 net_dm_queue_len_set(info);
1230
1231 return 0;
1232 }
1233
1234 static int net_dm_monitor_start(bool set_sw, bool set_hw,
1235 struct netlink_ext_ack *extack)
1236 {
1237 bool sw_set = false;
1238 int rc;
1239
1240 if (set_sw) {
1241 rc = set_all_monitor_traces(TRACE_ON, extack);
1242 if (rc)
1243 return rc;
1244 sw_set = true;
1245 }
1246
1247 if (set_hw) {
1248 rc = net_dm_hw_monitor_start(extack);
1249 if (rc)
1250 goto err_monitor_hw;
1251 }
1252
1253 return 0;
1254
1255 err_monitor_hw:
1256 if (sw_set)
1257 set_all_monitor_traces(TRACE_OFF, extack);
1258 return rc;
1259 }
1260
1261 static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1262 struct netlink_ext_ack *extack)
1263 {
1264 if (set_hw)
1265 net_dm_hw_monitor_stop(extack);
1266 if (set_sw)
1267 set_all_monitor_traces(TRACE_OFF, extack);
1268 }
1269
1270 static int net_dm_cmd_trace(struct sk_buff *skb,
1271 struct genl_info *info)
1272 {
1273 bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1274 bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1275 struct netlink_ext_ack *extack = info->extack;
1276
1277
1278
1279
1280 if (!set_sw && !set_hw)
1281 set_sw = true;
1282
1283 switch (info->genlhdr->cmd) {
1284 case NET_DM_CMD_START:
1285 return net_dm_monitor_start(set_sw, set_hw, extack);
1286 case NET_DM_CMD_STOP:
1287 net_dm_monitor_stop(set_sw, set_hw, extack);
1288 return 0;
1289 }
1290
1291 return -EOPNOTSUPP;
1292 }
1293
1294 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1295 {
1296 void *hdr;
1297
1298 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1299 &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1300 if (!hdr)
1301 return -EMSGSIZE;
1302
1303 if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1304 goto nla_put_failure;
1305
1306 if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1307 goto nla_put_failure;
1308
1309 if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1310 goto nla_put_failure;
1311
1312 genlmsg_end(msg, hdr);
1313
1314 return 0;
1315
1316 nla_put_failure:
1317 genlmsg_cancel(msg, hdr);
1318 return -EMSGSIZE;
1319 }
1320
1321 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1322 {
1323 struct sk_buff *msg;
1324 int rc;
1325
1326 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1327 if (!msg)
1328 return -ENOMEM;
1329
1330 rc = net_dm_config_fill(msg, info);
1331 if (rc)
1332 goto free_msg;
1333
1334 return genlmsg_reply(msg, info);
1335
1336 free_msg:
1337 nlmsg_free(msg);
1338 return rc;
1339 }
1340
1341 static void net_dm_stats_read(struct net_dm_stats *stats)
1342 {
1343 int cpu;
1344
1345 memset(stats, 0, sizeof(*stats));
1346 for_each_possible_cpu(cpu) {
1347 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1348 struct net_dm_stats *cpu_stats = &data->stats;
1349 unsigned int start;
1350 u64 dropped;
1351
1352 do {
1353 start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1354 dropped = cpu_stats->dropped;
1355 } while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1356
1357 stats->dropped += dropped;
1358 }
1359 }
1360
1361 static int net_dm_stats_put(struct sk_buff *msg)
1362 {
1363 struct net_dm_stats stats;
1364 struct nlattr *attr;
1365
1366 net_dm_stats_read(&stats);
1367
1368 attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1369 if (!attr)
1370 return -EMSGSIZE;
1371
1372 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1373 stats.dropped, NET_DM_ATTR_PAD))
1374 goto nla_put_failure;
1375
1376 nla_nest_end(msg, attr);
1377
1378 return 0;
1379
1380 nla_put_failure:
1381 nla_nest_cancel(msg, attr);
1382 return -EMSGSIZE;
1383 }
1384
1385 static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1386 {
1387 int cpu;
1388
1389 memset(stats, 0, sizeof(*stats));
1390 for_each_possible_cpu(cpu) {
1391 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1392 struct net_dm_stats *cpu_stats = &hw_data->stats;
1393 unsigned int start;
1394 u64 dropped;
1395
1396 do {
1397 start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1398 dropped = cpu_stats->dropped;
1399 } while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1400
1401 stats->dropped += dropped;
1402 }
1403 }
1404
1405 static int net_dm_hw_stats_put(struct sk_buff *msg)
1406 {
1407 struct net_dm_stats stats;
1408 struct nlattr *attr;
1409
1410 net_dm_hw_stats_read(&stats);
1411
1412 attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1413 if (!attr)
1414 return -EMSGSIZE;
1415
1416 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1417 stats.dropped, NET_DM_ATTR_PAD))
1418 goto nla_put_failure;
1419
1420 nla_nest_end(msg, attr);
1421
1422 return 0;
1423
1424 nla_put_failure:
1425 nla_nest_cancel(msg, attr);
1426 return -EMSGSIZE;
1427 }
1428
1429 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1430 {
1431 void *hdr;
1432 int rc;
1433
1434 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1435 &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1436 if (!hdr)
1437 return -EMSGSIZE;
1438
1439 rc = net_dm_stats_put(msg);
1440 if (rc)
1441 goto nla_put_failure;
1442
1443 rc = net_dm_hw_stats_put(msg);
1444 if (rc)
1445 goto nla_put_failure;
1446
1447 genlmsg_end(msg, hdr);
1448
1449 return 0;
1450
1451 nla_put_failure:
1452 genlmsg_cancel(msg, hdr);
1453 return -EMSGSIZE;
1454 }
1455
1456 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1457 {
1458 struct sk_buff *msg;
1459 int rc;
1460
1461 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1462 if (!msg)
1463 return -ENOMEM;
1464
1465 rc = net_dm_stats_fill(msg, info);
1466 if (rc)
1467 goto free_msg;
1468
1469 return genlmsg_reply(msg, info);
1470
1471 free_msg:
1472 nlmsg_free(msg);
1473 return rc;
1474 }
1475
1476 static int dropmon_net_event(struct notifier_block *ev_block,
1477 unsigned long event, void *ptr)
1478 {
1479 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1480 struct dm_hw_stat_delta *new_stat = NULL;
1481 struct dm_hw_stat_delta *tmp;
1482
1483 switch (event) {
1484 case NETDEV_REGISTER:
1485 new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
1486
1487 if (!new_stat)
1488 goto out;
1489
1490 new_stat->dev = dev;
1491 new_stat->last_rx = jiffies;
1492 mutex_lock(&net_dm_mutex);
1493 list_add_rcu(&new_stat->list, &hw_stats_list);
1494 mutex_unlock(&net_dm_mutex);
1495 break;
1496 case NETDEV_UNREGISTER:
1497 mutex_lock(&net_dm_mutex);
1498 list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1499 if (new_stat->dev == dev) {
1500 new_stat->dev = NULL;
1501 if (trace_state == TRACE_OFF) {
1502 list_del_rcu(&new_stat->list);
1503 kfree_rcu(new_stat, rcu);
1504 break;
1505 }
1506 }
1507 }
1508 mutex_unlock(&net_dm_mutex);
1509 break;
1510 }
1511 out:
1512 return NOTIFY_DONE;
1513 }
1514
1515 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1516 [NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1517 [NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1518 [NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1519 [NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1520 [NET_DM_ATTR_SW_DROPS] = {. type = NLA_FLAG },
1521 [NET_DM_ATTR_HW_DROPS] = {. type = NLA_FLAG },
1522 };
1523
1524 static const struct genl_ops dropmon_ops[] = {
1525 {
1526 .cmd = NET_DM_CMD_CONFIG,
1527 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1528 .doit = net_dm_cmd_config,
1529 .flags = GENL_ADMIN_PERM,
1530 },
1531 {
1532 .cmd = NET_DM_CMD_START,
1533 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1534 .doit = net_dm_cmd_trace,
1535 },
1536 {
1537 .cmd = NET_DM_CMD_STOP,
1538 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1539 .doit = net_dm_cmd_trace,
1540 },
1541 {
1542 .cmd = NET_DM_CMD_CONFIG_GET,
1543 .doit = net_dm_cmd_config_get,
1544 },
1545 {
1546 .cmd = NET_DM_CMD_STATS_GET,
1547 .doit = net_dm_cmd_stats_get,
1548 },
1549 };
1550
1551 static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1552 struct sk_buff *skb, struct genl_info *info)
1553 {
1554 mutex_lock(&net_dm_mutex);
1555
1556 return 0;
1557 }
1558
1559 static void net_dm_nl_post_doit(const struct genl_ops *ops,
1560 struct sk_buff *skb, struct genl_info *info)
1561 {
1562 mutex_unlock(&net_dm_mutex);
1563 }
1564
1565 static struct genl_family net_drop_monitor_family __ro_after_init = {
1566 .hdrsize = 0,
1567 .name = "NET_DM",
1568 .version = 2,
1569 .maxattr = NET_DM_ATTR_MAX,
1570 .policy = net_dm_nl_policy,
1571 .pre_doit = net_dm_nl_pre_doit,
1572 .post_doit = net_dm_nl_post_doit,
1573 .module = THIS_MODULE,
1574 .ops = dropmon_ops,
1575 .n_ops = ARRAY_SIZE(dropmon_ops),
1576 .mcgrps = dropmon_mcgrps,
1577 .n_mcgrps = ARRAY_SIZE(dropmon_mcgrps),
1578 };
1579
1580 static struct notifier_block dropmon_net_notifier = {
1581 .notifier_call = dropmon_net_event
1582 };
1583
1584 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1585 {
1586 spin_lock_init(&data->lock);
1587 skb_queue_head_init(&data->drop_queue);
1588 u64_stats_init(&data->stats.syncp);
1589 }
1590
1591 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1592 {
1593 WARN_ON(!skb_queue_empty(&data->drop_queue));
1594 }
1595
1596 static void net_dm_cpu_data_init(int cpu)
1597 {
1598 struct per_cpu_dm_data *data;
1599
1600 data = &per_cpu(dm_cpu_data, cpu);
1601 __net_dm_cpu_data_init(data);
1602 }
1603
1604 static void net_dm_cpu_data_fini(int cpu)
1605 {
1606 struct per_cpu_dm_data *data;
1607
1608 data = &per_cpu(dm_cpu_data, cpu);
1609
1610
1611
1612 consume_skb(data->skb);
1613 __net_dm_cpu_data_fini(data);
1614 }
1615
1616 static void net_dm_hw_cpu_data_init(int cpu)
1617 {
1618 struct per_cpu_dm_data *hw_data;
1619
1620 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1621 __net_dm_cpu_data_init(hw_data);
1622 }
1623
1624 static void net_dm_hw_cpu_data_fini(int cpu)
1625 {
1626 struct per_cpu_dm_data *hw_data;
1627
1628 hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1629 kfree(hw_data->hw_entries);
1630 __net_dm_cpu_data_fini(hw_data);
1631 }
1632
1633 static int __init init_net_drop_monitor(void)
1634 {
1635 int cpu, rc;
1636
1637 pr_info("Initializing network drop monitor service\n");
1638
1639 if (sizeof(void *) > 8) {
1640 pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1641 return -ENOSPC;
1642 }
1643
1644 rc = genl_register_family(&net_drop_monitor_family);
1645 if (rc) {
1646 pr_err("Could not create drop monitor netlink family\n");
1647 return rc;
1648 }
1649 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1650
1651 rc = register_netdevice_notifier(&dropmon_net_notifier);
1652 if (rc < 0) {
1653 pr_crit("Failed to register netdevice notifier\n");
1654 goto out_unreg;
1655 }
1656
1657 rc = 0;
1658
1659 for_each_possible_cpu(cpu) {
1660 net_dm_cpu_data_init(cpu);
1661 net_dm_hw_cpu_data_init(cpu);
1662 }
1663
1664 goto out;
1665
1666 out_unreg:
1667 genl_unregister_family(&net_drop_monitor_family);
1668 out:
1669 return rc;
1670 }
1671
1672 static void exit_net_drop_monitor(void)
1673 {
1674 int cpu;
1675
1676 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1677
1678
1679
1680
1681
1682
1683 for_each_possible_cpu(cpu) {
1684 net_dm_hw_cpu_data_fini(cpu);
1685 net_dm_cpu_data_fini(cpu);
1686 }
1687
1688 BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1689 }
1690
1691 module_init(init_net_drop_monitor);
1692 module_exit(exit_net_drop_monitor);
1693
1694 MODULE_LICENSE("GPL v2");
1695 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1696 MODULE_ALIAS_GENL_FAMILY("NET_DM");