Commit | Line | Data |
---|---|---|
9a8afc8d NH |
1 | /* |
2 | * Monitoring code for network dropped packet alerts | |
3 | * | |
4 | * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com> | |
5 | */ | |
6 | ||
e005d193 JP |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | ||
9a8afc8d NH |
9 | #include <linux/netdevice.h> |
10 | #include <linux/etherdevice.h> | |
11 | #include <linux/string.h> | |
12 | #include <linux/if_arp.h> | |
13 | #include <linux/inetdevice.h> | |
14 | #include <linux/inet.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/netpoll.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/delay.h> | |
19 | #include <linux/types.h> | |
20 | #include <linux/workqueue.h> | |
21 | #include <linux/netlink.h> | |
22 | #include <linux/net_dropmon.h> | |
23 | #include <linux/percpu.h> | |
24 | #include <linux/timer.h> | |
25 | #include <linux/bitops.h> | |
5a0e3ad6 | 26 | #include <linux/slab.h> |
cad456d5 | 27 | #include <linux/module.h> |
9a8afc8d | 28 | #include <net/genetlink.h> |
4ea7e386 | 29 | #include <net/netevent.h> |
9a8afc8d | 30 | |
ad8d75ff | 31 | #include <trace/events/skb.h> |
9cbc1cb8 | 32 | #include <trace/events/napi.h> |
9a8afc8d NH |
33 | |
34 | #include <asm/unaligned.h> | |
35 | ||
36 | #define TRACE_ON 1 | |
37 | #define TRACE_OFF 0 | |
38 | ||
9a8afc8d NH |
39 | /* |
40 | * Globals, our netlink socket pointer | |
41 | * and the work handle that will send up | |
42 | * netlink alerts | |
43 | */ | |
4ea7e386 | 44 | static int trace_state = TRACE_OFF; |
cde2e9a6 | 45 | static DEFINE_MUTEX(trace_state_mutex); |
9a8afc8d NH |
46 | |
47 | struct per_cpu_dm_data { | |
bec4596b ED |
48 | spinlock_t lock; |
49 | struct sk_buff *skb; | |
50 | struct work_struct dm_alert_work; | |
51 | struct timer_list send_timer; | |
9a8afc8d NH |
52 | }; |
53 | ||
4ea7e386 NH |
54 | struct dm_hw_stat_delta { |
55 | struct net_device *dev; | |
5848cc09 | 56 | unsigned long last_rx; |
4ea7e386 NH |
57 | struct list_head list; |
58 | struct rcu_head rcu; | |
59 | unsigned long last_drop_val; | |
60 | }; | |
61 | ||
9a8afc8d NH |
62 | static struct genl_family net_drop_monitor_family = { |
63 | .id = GENL_ID_GENERATE, | |
64 | .hdrsize = 0, | |
65 | .name = "NET_DM", | |
683703a2 | 66 | .version = 2, |
9a8afc8d NH |
67 | .maxattr = NET_DM_CMD_MAX, |
68 | }; | |
69 | ||
70 | static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data); | |
71 | ||
72 | static int dm_hit_limit = 64; | |
73 | static int dm_delay = 1; | |
4ea7e386 NH |
74 | static unsigned long dm_hw_check_delta = 2*HZ; |
75 | static LIST_HEAD(hw_stats_list); | |
9a8afc8d | 76 | |
bec4596b | 77 | static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data) |
9a8afc8d NH |
78 | { |
79 | size_t al; | |
80 | struct net_dm_alert_msg *msg; | |
683703a2 | 81 | struct nlattr *nla; |
3885ca78 | 82 | struct sk_buff *skb; |
bec4596b | 83 | unsigned long flags; |
9a8afc8d NH |
84 | |
85 | al = sizeof(struct net_dm_alert_msg); | |
86 | al += dm_hit_limit * sizeof(struct net_dm_drop_point); | |
683703a2 NH |
87 | al += sizeof(struct nlattr); |
88 | ||
3885ca78 NH |
89 | skb = genlmsg_new(al, GFP_KERNEL); |
90 | ||
91 | if (skb) { | |
92 | genlmsg_put(skb, 0, 0, &net_drop_monitor_family, | |
93 | 0, NET_DM_CMD_ALERT); | |
94 | nla = nla_reserve(skb, NLA_UNSPEC, | |
95 | sizeof(struct net_dm_alert_msg)); | |
96 | msg = nla_data(nla); | |
97 | memset(msg, 0, al); | |
bec4596b ED |
98 | } else { |
99 | mod_timer(&data->send_timer, jiffies + HZ / 10); | |
3885ca78 NH |
100 | } |
101 | ||
bec4596b ED |
102 | spin_lock_irqsave(&data->lock, flags); |
103 | swap(data->skb, skb); | |
104 | spin_unlock_irqrestore(&data->lock, flags); | |
105 | ||
106 | return skb; | |
9a8afc8d NH |
107 | } |
108 | ||
2a94fe48 JB |
109 | static struct genl_multicast_group dropmon_mcgrps[] = { |
110 | { .name = "events", }, | |
e5dcecba JB |
111 | }; |
112 | ||
bec4596b | 113 | static void send_dm_alert(struct work_struct *work) |
9a8afc8d NH |
114 | { |
115 | struct sk_buff *skb; | |
bec4596b | 116 | struct per_cpu_dm_data *data; |
9a8afc8d | 117 | |
bec4596b | 118 | data = container_of(work, struct per_cpu_dm_data, dm_alert_work); |
4fdcfa12 | 119 | |
bec4596b | 120 | skb = reset_per_cpu_data(data); |
9a8afc8d | 121 | |
3885ca78 | 122 | if (skb) |
68eb5503 | 123 | genlmsg_multicast(&net_drop_monitor_family, skb, 0, |
2a94fe48 | 124 | 0, GFP_KERNEL); |
9a8afc8d NH |
125 | } |
126 | ||
127 | /* | |
128 | * This is the timer function to delay the sending of an alert | |
129 | * in the event that more drops will arrive during the | |
bec4596b | 130 | * hysteresis period. |
9a8afc8d | 131 | */ |
bec4596b | 132 | static void sched_send_work(unsigned long _data) |
9a8afc8d | 133 | { |
bec4596b | 134 | struct per_cpu_dm_data *data = (struct per_cpu_dm_data *)_data; |
3885ca78 | 135 | |
bec4596b | 136 | schedule_work(&data->dm_alert_work); |
9a8afc8d NH |
137 | } |
138 | ||
4ea7e386 | 139 | static void trace_drop_common(struct sk_buff *skb, void *location) |
9a8afc8d NH |
140 | { |
141 | struct net_dm_alert_msg *msg; | |
142 | struct nlmsghdr *nlh; | |
683703a2 | 143 | struct nlattr *nla; |
9a8afc8d | 144 | int i; |
3885ca78 | 145 | struct sk_buff *dskb; |
bec4596b ED |
146 | struct per_cpu_dm_data *data; |
147 | unsigned long flags; | |
9a8afc8d | 148 | |
bec4596b ED |
149 | local_irq_save(flags); |
150 | data = &__get_cpu_var(dm_cpu_data); | |
151 | spin_lock(&data->lock); | |
152 | dskb = data->skb; | |
3885ca78 NH |
153 | |
154 | if (!dskb) | |
155 | goto out; | |
156 | ||
3885ca78 | 157 | nlh = (struct nlmsghdr *)dskb->data; |
683703a2 NH |
158 | nla = genlmsg_data(nlmsg_data(nlh)); |
159 | msg = nla_data(nla); | |
9a8afc8d NH |
160 | for (i = 0; i < msg->entries; i++) { |
161 | if (!memcmp(&location, msg->points[i].pc, sizeof(void *))) { | |
162 | msg->points[i].count++; | |
163 | goto out; | |
164 | } | |
165 | } | |
bec4596b ED |
166 | if (msg->entries == dm_hit_limit) |
167 | goto out; | |
9a8afc8d NH |
168 | /* |
169 | * We need to create a new entry | |
170 | */ | |
3885ca78 | 171 | __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point)); |
683703a2 | 172 | nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point)); |
9a8afc8d NH |
173 | memcpy(msg->points[msg->entries].pc, &location, sizeof(void *)); |
174 | msg->points[msg->entries].count = 1; | |
175 | msg->entries++; | |
176 | ||
177 | if (!timer_pending(&data->send_timer)) { | |
178 | data->send_timer.expires = jiffies + dm_delay * HZ; | |
bec4596b | 179 | add_timer(&data->send_timer); |
9a8afc8d NH |
180 | } |
181 | ||
182 | out: | |
bec4596b | 183 | spin_unlock_irqrestore(&data->lock, flags); |
9a8afc8d NH |
184 | } |
185 | ||
38516ab5 | 186 | static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location) |
4ea7e386 NH |
187 | { |
188 | trace_drop_common(skb, location); | |
189 | } | |
190 | ||
38516ab5 | 191 | static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi) |
4ea7e386 NH |
192 | { |
193 | struct dm_hw_stat_delta *new_stat; | |
194 | ||
195 | /* | |
5848cc09 | 196 | * Don't check napi structures with no associated device |
4ea7e386 | 197 | */ |
5848cc09 | 198 | if (!napi->dev) |
4ea7e386 NH |
199 | return; |
200 | ||
201 | rcu_read_lock(); | |
202 | list_for_each_entry_rcu(new_stat, &hw_stats_list, list) { | |
5848cc09 NH |
203 | /* |
204 | * only add a note to our monitor buffer if: | |
205 | * 1) this is the dev we received on | |
206 | * 2) its after the last_rx delta | |
207 | * 3) our rx_dropped count has gone up | |
208 | */ | |
4ea7e386 | 209 | if ((new_stat->dev == napi->dev) && |
5848cc09 | 210 | (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) && |
4ea7e386 NH |
211 | (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) { |
212 | trace_drop_common(NULL, NULL); | |
213 | new_stat->last_drop_val = napi->dev->stats.rx_dropped; | |
5848cc09 | 214 | new_stat->last_rx = jiffies; |
4ea7e386 NH |
215 | break; |
216 | } | |
217 | } | |
218 | rcu_read_unlock(); | |
219 | } | |
220 | ||
9a8afc8d NH |
221 | static int set_all_monitor_traces(int state) |
222 | { | |
223 | int rc = 0; | |
4ea7e386 NH |
224 | struct dm_hw_stat_delta *new_stat = NULL; |
225 | struct dm_hw_stat_delta *temp; | |
226 | ||
cde2e9a6 | 227 | mutex_lock(&trace_state_mutex); |
9a8afc8d | 228 | |
4b706372 NH |
229 | if (state == trace_state) { |
230 | rc = -EAGAIN; | |
231 | goto out_unlock; | |
232 | } | |
233 | ||
9a8afc8d NH |
234 | switch (state) { |
235 | case TRACE_ON: | |
cad456d5 NH |
236 | if (!try_module_get(THIS_MODULE)) { |
237 | rc = -ENODEV; | |
238 | break; | |
239 | } | |
240 | ||
38516ab5 SR |
241 | rc |= register_trace_kfree_skb(trace_kfree_skb_hit, NULL); |
242 | rc |= register_trace_napi_poll(trace_napi_poll_hit, NULL); | |
9a8afc8d | 243 | break; |
cad456d5 | 244 | |
9a8afc8d | 245 | case TRACE_OFF: |
38516ab5 SR |
246 | rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit, NULL); |
247 | rc |= unregister_trace_napi_poll(trace_napi_poll_hit, NULL); | |
9a8afc8d NH |
248 | |
249 | tracepoint_synchronize_unregister(); | |
4ea7e386 NH |
250 | |
251 | /* | |
252 | * Clean the device list | |
253 | */ | |
254 | list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) { | |
255 | if (new_stat->dev == NULL) { | |
256 | list_del_rcu(&new_stat->list); | |
fa81c0e1 | 257 | kfree_rcu(new_stat, rcu); |
4ea7e386 NH |
258 | } |
259 | } | |
cad456d5 NH |
260 | |
261 | module_put(THIS_MODULE); | |
262 | ||
9a8afc8d NH |
263 | break; |
264 | default: | |
265 | rc = 1; | |
266 | break; | |
267 | } | |
268 | ||
4ea7e386 NH |
269 | if (!rc) |
270 | trace_state = state; | |
4b706372 NH |
271 | else |
272 | rc = -EINPROGRESS; | |
4ea7e386 | 273 | |
4b706372 | 274 | out_unlock: |
cde2e9a6 | 275 | mutex_unlock(&trace_state_mutex); |
4ea7e386 | 276 | |
9a8afc8d NH |
277 | return rc; |
278 | } | |
279 | ||
280 | ||
281 | static int net_dm_cmd_config(struct sk_buff *skb, | |
282 | struct genl_info *info) | |
283 | { | |
284 | return -ENOTSUPP; | |
285 | } | |
286 | ||
287 | static int net_dm_cmd_trace(struct sk_buff *skb, | |
288 | struct genl_info *info) | |
289 | { | |
290 | switch (info->genlhdr->cmd) { | |
291 | case NET_DM_CMD_START: | |
292 | return set_all_monitor_traces(TRACE_ON); | |
293 | break; | |
294 | case NET_DM_CMD_STOP: | |
295 | return set_all_monitor_traces(TRACE_OFF); | |
296 | break; | |
297 | } | |
298 | ||
299 | return -ENOTSUPP; | |
300 | } | |
301 | ||
4ea7e386 | 302 | static int dropmon_net_event(struct notifier_block *ev_block, |
351638e7 | 303 | unsigned long event, void *ptr) |
4ea7e386 | 304 | { |
351638e7 | 305 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
4ea7e386 NH |
306 | struct dm_hw_stat_delta *new_stat = NULL; |
307 | struct dm_hw_stat_delta *tmp; | |
308 | ||
309 | switch (event) { | |
310 | case NETDEV_REGISTER: | |
311 | new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL); | |
312 | ||
313 | if (!new_stat) | |
314 | goto out; | |
315 | ||
316 | new_stat->dev = dev; | |
5848cc09 | 317 | new_stat->last_rx = jiffies; |
cde2e9a6 | 318 | mutex_lock(&trace_state_mutex); |
4ea7e386 | 319 | list_add_rcu(&new_stat->list, &hw_stats_list); |
cde2e9a6 | 320 | mutex_unlock(&trace_state_mutex); |
4ea7e386 NH |
321 | break; |
322 | case NETDEV_UNREGISTER: | |
cde2e9a6 | 323 | mutex_lock(&trace_state_mutex); |
4ea7e386 NH |
324 | list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) { |
325 | if (new_stat->dev == dev) { | |
326 | new_stat->dev = NULL; | |
327 | if (trace_state == TRACE_OFF) { | |
328 | list_del_rcu(&new_stat->list); | |
fa81c0e1 | 329 | kfree_rcu(new_stat, rcu); |
4ea7e386 NH |
330 | break; |
331 | } | |
332 | } | |
333 | } | |
cde2e9a6 | 334 | mutex_unlock(&trace_state_mutex); |
4ea7e386 NH |
335 | break; |
336 | } | |
337 | out: | |
338 | return NOTIFY_DONE; | |
339 | } | |
9a8afc8d | 340 | |
4534de83 | 341 | static const struct genl_ops dropmon_ops[] = { |
9a8afc8d NH |
342 | { |
343 | .cmd = NET_DM_CMD_CONFIG, | |
344 | .doit = net_dm_cmd_config, | |
345 | }, | |
346 | { | |
347 | .cmd = NET_DM_CMD_START, | |
348 | .doit = net_dm_cmd_trace, | |
349 | }, | |
350 | { | |
351 | .cmd = NET_DM_CMD_STOP, | |
352 | .doit = net_dm_cmd_trace, | |
353 | }, | |
354 | }; | |
355 | ||
4ea7e386 NH |
356 | static struct notifier_block dropmon_net_notifier = { |
357 | .notifier_call = dropmon_net_event | |
358 | }; | |
359 | ||
9a8afc8d NH |
360 | static int __init init_net_drop_monitor(void) |
361 | { | |
9a8afc8d | 362 | struct per_cpu_dm_data *data; |
a256be70 CG |
363 | int cpu, rc; |
364 | ||
e005d193 | 365 | pr_info("Initializing network drop monitor service\n"); |
9a8afc8d NH |
366 | |
367 | if (sizeof(void *) > 8) { | |
e005d193 | 368 | pr_err("Unable to store program counters on this arch, Drop monitor failed\n"); |
9a8afc8d NH |
369 | return -ENOSPC; |
370 | } | |
371 | ||
2a94fe48 JB |
372 | rc = genl_register_family_with_ops_groups(&net_drop_monitor_family, |
373 | dropmon_ops, dropmon_mcgrps); | |
a256be70 | 374 | if (rc) { |
e005d193 | 375 | pr_err("Could not create drop monitor netlink family\n"); |
a256be70 | 376 | return rc; |
9a8afc8d | 377 | } |
2a94fe48 | 378 | WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT); |
e5dcecba | 379 | |
4ea7e386 NH |
380 | rc = register_netdevice_notifier(&dropmon_net_notifier); |
381 | if (rc < 0) { | |
e005d193 | 382 | pr_crit("Failed to register netdevice notifier\n"); |
4ea7e386 NH |
383 | goto out_unreg; |
384 | } | |
385 | ||
9a8afc8d NH |
386 | rc = 0; |
387 | ||
cad456d5 | 388 | for_each_possible_cpu(cpu) { |
9a8afc8d | 389 | data = &per_cpu(dm_cpu_data, cpu); |
9a8afc8d NH |
390 | INIT_WORK(&data->dm_alert_work, send_dm_alert); |
391 | init_timer(&data->send_timer); | |
bec4596b | 392 | data->send_timer.data = (unsigned long)data; |
9a8afc8d | 393 | data->send_timer.function = sched_send_work; |
bec4596b | 394 | spin_lock_init(&data->lock); |
4fdcfa12 | 395 | reset_per_cpu_data(data); |
9a8afc8d | 396 | } |
4ea7e386 | 397 | |
3885ca78 | 398 | |
9a8afc8d NH |
399 | goto out; |
400 | ||
401 | out_unreg: | |
402 | genl_unregister_family(&net_drop_monitor_family); | |
403 | out: | |
404 | return rc; | |
405 | } | |
406 | ||
cad456d5 NH |
407 | static void exit_net_drop_monitor(void) |
408 | { | |
409 | struct per_cpu_dm_data *data; | |
410 | int cpu; | |
411 | ||
412 | BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier)); | |
413 | ||
414 | /* | |
415 | * Because of the module_get/put we do in the trace state change path | |
416 | * we are guarnateed not to have any current users when we get here | |
417 | * all we need to do is make sure that we don't have any running timers | |
418 | * or pending schedule calls | |
419 | */ | |
420 | ||
421 | for_each_possible_cpu(cpu) { | |
422 | data = &per_cpu(dm_cpu_data, cpu); | |
423 | del_timer_sync(&data->send_timer); | |
424 | cancel_work_sync(&data->dm_alert_work); | |
425 | /* | |
426 | * At this point, we should have exclusive access | |
427 | * to this struct and can free the skb inside it | |
428 | */ | |
429 | kfree_skb(data->skb); | |
430 | } | |
431 | ||
432 | BUG_ON(genl_unregister_family(&net_drop_monitor_family)); | |
433 | } | |
434 | ||
435 | module_init(init_net_drop_monitor); | |
436 | module_exit(exit_net_drop_monitor); | |
437 | ||
438 | MODULE_LICENSE("GPL v2"); | |
439 | MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>"); | |
3fdcbd45 | 440 | MODULE_ALIAS_GENL_FAMILY("NET_DM"); |