1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Monitoring code for network dropped packet alerts
4 *
5 * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
6 */
7
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/string.h>
13 #include <linux/if_arp.h>
14 #include <linux/inetdevice.h>
15 #include <linux/inet.h>
16 #include <linux/interrupt.h>
17 #include <linux/netpoll.h>
18 #include <linux/sched.h>
19 #include <linux/delay.h>
20 #include <linux/types.h>
21 #include <linux/workqueue.h>
22 #include <linux/netlink.h>
23 #include <linux/net_dropmon.h>
24 #include <linux/bitfield.h>
25 #include <linux/percpu.h>
26 #include <linux/timer.h>
27 #include <linux/bitops.h>
28 #include <linux/slab.h>
29 #include <linux/module.h>
30 #include <net/genetlink.h>
31 #include <net/netevent.h>
32 #include <net/flow_offload.h>
33 #include <net/dropreason.h>
34 #include <net/devlink.h>
35
36 #include <trace/events/skb.h>
37 #include <trace/events/napi.h>
38 #include <trace/events/devlink.h>
39
40 #include <asm/unaligned.h>
41
42 #define TRACE_ON 1
43 #define TRACE_OFF 0
44
45 /*
46 * Globals, our netlink socket pointer
47 * and the work handle that will send up
48 * netlink alerts
49 */
50 static int trace_state = TRACE_OFF;
51 static bool monitor_hw;
52
53 /* net_dm_mutex
54 *
55 * An overall lock guarding every operation coming from userspace.
56 */
57 static DEFINE_MUTEX(net_dm_mutex);
58
59 struct net_dm_stats {
60 u64_stats_t dropped;
61 struct u64_stats_sync syncp;
62 };
63
64 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40
65
66 struct net_dm_hw_entry {
67 char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
68 u32 count;
69 };
70
71 struct net_dm_hw_entries {
72 u32 num_entries;
73 struct net_dm_hw_entry entries[];
74 };
75
76 struct per_cpu_dm_data {
77 raw_spinlock_t lock; /* Protects 'skb', 'hw_entries' and
78 * 'send_timer'
79 */
80 union {
81 struct sk_buff *skb;
82 struct net_dm_hw_entries *hw_entries;
83 };
84 struct sk_buff_head drop_queue;
85 struct work_struct dm_alert_work;
86 struct timer_list send_timer;
87 struct net_dm_stats stats;
88 };
89
90 struct dm_hw_stat_delta {
91 unsigned long last_rx;
92 unsigned long last_drop_val;
93 struct rcu_head rcu;
94 };
95
96 static struct genl_family net_drop_monitor_family;
97
98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
100
101 static int dm_hit_limit = 64;
102 static int dm_delay = 1;
103 static unsigned long dm_hw_check_delta = 2*HZ;
104
105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
106 static u32 net_dm_trunc_len;
107 static u32 net_dm_queue_len = 1000;
108
109 struct net_dm_alert_ops {
110 void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
111 void *location,
112 enum skb_drop_reason reason,
113 struct sock *rx_sk);
114 void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
115 int work, int budget);
116 void (*work_item_func)(struct work_struct *work);
117 void (*hw_work_item_func)(struct work_struct *work);
118 void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
119 struct sk_buff *skb,
120 const struct devlink_trap_metadata *metadata);
121 };
122
123 struct net_dm_skb_cb {
124 union {
125 struct devlink_trap_metadata *hw_metadata;
126 void *pc;
127 };
128 enum skb_drop_reason reason;
129 };
130
131 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
132
133 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
134 {
135 size_t al;
136 struct net_dm_alert_msg *msg;
137 struct nlattr *nla;
138 struct sk_buff *skb;
139 unsigned long flags;
140 void *msg_header;
141
142 al = sizeof(struct net_dm_alert_msg);
143 al += dm_hit_limit * sizeof(struct net_dm_drop_point);
144 al += sizeof(struct nlattr);
145
146 skb = genlmsg_new(al, GFP_KERNEL);
147
148 if (!skb)
149 goto err;
150
151 msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
152 0, NET_DM_CMD_ALERT);
153 if (!msg_header) {
154 nlmsg_free(skb);
155 skb = NULL;
156 goto err;
157 }
158 nla = nla_reserve(skb, NLA_UNSPEC,
159 sizeof(struct net_dm_alert_msg));
160 if (!nla) {
161 nlmsg_free(skb);
162 skb = NULL;
163 goto err;
164 }
165 msg = nla_data(nla);
166 memset(msg, 0, al);
167 goto out;
168
169 err:
170 mod_timer(&data->send_timer, jiffies + HZ / 10);
171 out:
172 raw_spin_lock_irqsave(&data->lock, flags);
173 swap(data->skb, skb);
174 raw_spin_unlock_irqrestore(&data->lock, flags);
175
176 if (skb) {
177 struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
178 struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
179
180 genlmsg_end(skb, genlmsg_data(gnlh));
181 }
182
183 return skb;
184 }
185
186 static const struct genl_multicast_group dropmon_mcgrps[] = {
187 { .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, },
188 };
189
190 static void send_dm_alert(struct work_struct *work)
191 {
192 struct sk_buff *skb;
193 struct per_cpu_dm_data *data;
194
195 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
196
197 skb = reset_per_cpu_data(data);
198
199 if (skb)
200 genlmsg_multicast(&net_drop_monitor_family, skb, 0,
201 0, GFP_KERNEL);
202 }
203
204 /*
205 * This is the timer function to delay the sending of an alert
206 * in the event that more drops will arrive during the
207 * hysteresis period.
208 */
209 static void sched_send_work(struct timer_list *t)
210 {
211 struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
212
213 schedule_work(&data->dm_alert_work);
214 }
215
216 static void trace_drop_common(struct sk_buff *skb, void *location)
217 {
218 struct net_dm_alert_msg *msg;
219 struct net_dm_drop_point *point;
220 struct nlmsghdr *nlh;
221 struct nlattr *nla;
222 int i;
223 struct sk_buff *dskb;
224 struct per_cpu_dm_data *data;
225 unsigned long flags;
226
227 local_irq_save(flags);
228 data = this_cpu_ptr(&dm_cpu_data);
229 raw_spin_lock(&data->lock);
230 dskb = data->skb;
231
232 if (!dskb)
233 goto out;
234
235 nlh = (struct nlmsghdr *)dskb->data;
236 nla = genlmsg_data(nlmsg_data(nlh));
237 msg = nla_data(nla);
238 point = msg->points;
239 for (i = 0; i < msg->entries; i++) {
240 if (!memcmp(&location, &point->pc, sizeof(void *))) {
241 point->count++;
242 goto out;
243 }
244 point++;
245 }
246 if (msg->entries == dm_hit_limit)
247 goto out;
248 /*
249 * We need to create a new entry
250 */
251 __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
252 nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
253 memcpy(point->pc, &location, sizeof(void *));
254 point->count = 1;
255 msg->entries++;
256
257 if (!timer_pending(&data->send_timer)) {
258 data->send_timer.expires = jiffies + dm_delay * HZ;
259 add_timer(&data->send_timer);
260 }
261
262 out:
263 raw_spin_unlock_irqrestore(&data->lock, flags);
264 }
265
266 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb,
267 void *location,
268 enum skb_drop_reason reason,
269 struct sock *rx_sk)
270 {
271 trace_drop_common(skb, location);
272 }
273
274 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
275 int work, int budget)
276 {
277 struct net_device *dev = napi->dev;
278 struct dm_hw_stat_delta *stat;
279 /*
280 * Don't check napi structures with no associated device
281 */
282 if (!dev)
283 return;
284
285 rcu_read_lock();
286 stat = rcu_dereference(dev->dm_private);
287 if (stat) {
288 /*
289 * only add a note to our monitor buffer if:
290 * 1) its after the last_rx delta
291 * 2) our rx_dropped count has gone up
292 */
293 if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
294 (dev->stats.rx_dropped != stat->last_drop_val)) {
295 trace_drop_common(NULL, NULL);
296 stat->last_drop_val = dev->stats.rx_dropped;
297 stat->last_rx = jiffies;
298 }
299 }
300 rcu_read_unlock();
301 }
302
303 static struct net_dm_hw_entries *
304 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
305 {
306 struct net_dm_hw_entries *hw_entries;
307 unsigned long flags;
308
309 hw_entries = kzalloc(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
326 static 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
345 nla_put_failure:
346 nla_nest_cancel(msg, attr);
347 return -EMSGSIZE;
348 }
349
350 static 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
372 nla_put_failure:
373 nla_nest_cancel(msg, attr);
374 return -EMSGSIZE;
375 }
376
377 static int
378 net_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
404 nla_put_failure:
405 genlmsg_cancel(msg, hdr);
406 return -EMSGSIZE;
407 }
408
409 static 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
434 out:
435 kfree(hw_entries);
436 }
437
438 static void
439 net_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
481 out:
482 raw_spin_unlock_irqrestore(&hw_data->lock, flags);
483 }
484
485 static 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
493 static 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
533 unlock_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
541 static void net_dm_packet_trace_napi_poll_hit(void *ignore,
542 struct napi_struct *napi,
543 int work, int budget)
544 {
545 }
546
547 static 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
560 static 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
587 static 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
607 nla_put_failure:
608 nla_nest_cancel(msg, attr);
609 return -EMSGSIZE;
610 }
611
612 static 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
681 out:
682 genlmsg_end(msg, hdr);
683
684 return 0;
685
686 nla_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
693 static 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
722 out:
723 consume_skb(skb);
724 }
725
726 static 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
745 static size_t
746 net_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
752 static size_t
753 net_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
781 static 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
841 out:
842 genlmsg_end(msg, hdr);
843
844 return 0;
845
846 nla_put_failure:
847 genlmsg_cancel(msg, hdr);
848 return -EMSGSIZE;
849 }
850
851 static struct devlink_trap_metadata *
852 net_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
890 free_trap_name:
891 kfree(trap_name);
892 free_trap_group:
893 kfree(trap_group_name);
894 free_hw_metadata:
895 kfree(hw_metadata);
896 return NULL;
897 }
898
899 static void
900 net_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
909 static 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
939 out:
940 net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
941 consume_skb(skb);
942 }
943
944 static 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
963 static void
964 net_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
1004 unlock_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);
1010 free:
1011 consume_skb(nskb);
1012 }
1013
1014 static 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
1022 static 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)
1028 static 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
1033 static 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
1039 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1040 {
1041 return -EOPNOTSUPP;
1042 }
1043
1044 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1045 {
1046 }
1047 #endif
1048
1049 static 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
1086 err_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
1105 static 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
1139 static 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
1179 err_unregister_trace:
1180 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1181 err_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
1195 static 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
1223 static 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
1252 static bool net_dm_is_monitoring(void)
1253 {
1254 return trace_state == TRACE_ON || monitor_hw;
1255 }
1256
1257 static 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
1276 static 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
1296 static 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
1304 static 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
1312 static 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
1334 static 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
1355 err_monitor_hw:
1356 if (sw_set)
1357 set_all_monitor_traces(TRACE_OFF, extack);
1358 return rc;
1359 }
1360
1361 static 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
1370 static 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
1394 static 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
1416 nla_put_failure:
1417 genlmsg_cancel(msg, hdr);
1418 return -EMSGSIZE;
1419 }
1420
1421 static 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
1436 free_msg:
1437 nlmsg_free(msg);
1438 return rc;
1439 }
1440
1441 static 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
1461 static 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
1480 nla_put_failure:
1481 nla_nest_cancel(msg, attr);
1482 return -EMSGSIZE;
1483 }
1484
1485 static 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
1505 static 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
1524 nla_put_failure:
1525 nla_nest_cancel(msg, attr);
1526 return -EMSGSIZE;
1527 }
1528
1529 static 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
1551 nla_put_failure:
1552 genlmsg_cancel(msg, hdr);
1553 return -EMSGSIZE;
1554 }
1555
1556 static 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
1571 free_msg:
1572 nlmsg_free(msg);
1573 return rc;
1574 }
1575
1576 static 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
1605 static 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
1614 static 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
1643 static 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
1651 static 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
1657 static 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
1673 static struct notifier_block dropmon_net_notifier = {
1674 .notifier_call = dropmon_net_event
1675 };
1676
1677 static 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
1684 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1685 {
1686 WARN_ON(!skb_queue_empty(&data->drop_queue));
1687 }
1688
1689 static 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
1697 static 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
1709 static 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
1717 static 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
1726 static 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 rc = genl_register_family(&net_drop_monitor_family);
1738 if (rc) {
1739 pr_err("Could not create drop monitor netlink family\n");
1740 return rc;
1741 }
1742 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1743
1744 rc = register_netdevice_notifier(&dropmon_net_notifier);
1745 if (rc < 0) {
1746 pr_crit("Failed to register netdevice notifier\n");
1747 goto out_unreg;
1748 }
1749
1750 rc = 0;
1751
1752 for_each_possible_cpu(cpu) {
1753 net_dm_cpu_data_init(cpu);
1754 net_dm_hw_cpu_data_init(cpu);
1755 }
1756
1757 goto out;
1758
1759 out_unreg:
1760 genl_unregister_family(&net_drop_monitor_family);
1761 out:
1762 return rc;
1763 }
1764
1765 static void exit_net_drop_monitor(void)
1766 {
1767 int cpu;
1768
1769 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1770
1771 /*
1772 * Because of the module_get/put we do in the trace state change path
1773 * we are guaranteed not to have any current users when we get here
1774 */
1775
1776 for_each_possible_cpu(cpu) {
1777 net_dm_hw_cpu_data_fini(cpu);
1778 net_dm_cpu_data_fini(cpu);
1779 }
1780
1781 BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1782 }
1783
1784 module_init(init_net_drop_monitor);
1785 module_exit(exit_net_drop_monitor);
1786
1787 MODULE_LICENSE("GPL v2");
1788 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1789 MODULE_ALIAS_GENL_FAMILY("NET_DM");
1790 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
1791
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.