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 | ||
bec4596b | 109 | static void send_dm_alert(struct work_struct *work) |
9a8afc8d NH |
110 | { |
111 | struct sk_buff *skb; | |
bec4596b | 112 | struct per_cpu_dm_data *data; |
9a8afc8d | 113 | |
bec4596b | 114 | data = container_of(work, struct per_cpu_dm_data, dm_alert_work); |
4fdcfa12 | 115 | |
bec4596b | 116 | skb = reset_per_cpu_data(data); |
9a8afc8d | 117 | |
3885ca78 NH |
118 | if (skb) |
119 | genlmsg_multicast(skb, 0, NET_DM_GRP_ALERT, GFP_KERNEL); | |
9a8afc8d NH |
120 | } |
121 | ||
122 | /* | |
123 | * This is the timer function to delay the sending of an alert | |
124 | * in the event that more drops will arrive during the | |
bec4596b | 125 | * hysteresis period. |
9a8afc8d | 126 | */ |
bec4596b | 127 | static void sched_send_work(unsigned long _data) |
9a8afc8d | 128 | { |
bec4596b | 129 | struct per_cpu_dm_data *data = (struct per_cpu_dm_data *)_data; |
3885ca78 | 130 | |
bec4596b | 131 | schedule_work(&data->dm_alert_work); |
9a8afc8d NH |
132 | } |
133 | ||
4ea7e386 | 134 | static void trace_drop_common(struct sk_buff *skb, void *location) |
9a8afc8d NH |
135 | { |
136 | struct net_dm_alert_msg *msg; | |
137 | struct nlmsghdr *nlh; | |
683703a2 | 138 | struct nlattr *nla; |
9a8afc8d | 139 | int i; |
3885ca78 | 140 | struct sk_buff *dskb; |
bec4596b ED |
141 | struct per_cpu_dm_data *data; |
142 | unsigned long flags; | |
9a8afc8d | 143 | |
bec4596b ED |
144 | local_irq_save(flags); |
145 | data = &__get_cpu_var(dm_cpu_data); | |
146 | spin_lock(&data->lock); | |
147 | dskb = data->skb; | |
3885ca78 NH |
148 | |
149 | if (!dskb) | |
150 | goto out; | |
151 | ||
3885ca78 | 152 | nlh = (struct nlmsghdr *)dskb->data; |
683703a2 NH |
153 | nla = genlmsg_data(nlmsg_data(nlh)); |
154 | msg = nla_data(nla); | |
9a8afc8d NH |
155 | for (i = 0; i < msg->entries; i++) { |
156 | if (!memcmp(&location, msg->points[i].pc, sizeof(void *))) { | |
157 | msg->points[i].count++; | |
158 | goto out; | |
159 | } | |
160 | } | |
bec4596b ED |
161 | if (msg->entries == dm_hit_limit) |
162 | goto out; | |
9a8afc8d NH |
163 | /* |
164 | * We need to create a new entry | |
165 | */ | |
3885ca78 | 166 | __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point)); |
683703a2 | 167 | nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point)); |
9a8afc8d NH |
168 | memcpy(msg->points[msg->entries].pc, &location, sizeof(void *)); |
169 | msg->points[msg->entries].count = 1; | |
170 | msg->entries++; | |
171 | ||
172 | if (!timer_pending(&data->send_timer)) { | |
173 | data->send_timer.expires = jiffies + dm_delay * HZ; | |
bec4596b | 174 | add_timer(&data->send_timer); |
9a8afc8d NH |
175 | } |
176 | ||
177 | out: | |
bec4596b | 178 | spin_unlock_irqrestore(&data->lock, flags); |
9a8afc8d NH |
179 | } |
180 | ||
38516ab5 | 181 | static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location) |
4ea7e386 NH |
182 | { |
183 | trace_drop_common(skb, location); | |
184 | } | |
185 | ||
38516ab5 | 186 | static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi) |
4ea7e386 NH |
187 | { |
188 | struct dm_hw_stat_delta *new_stat; | |
189 | ||
190 | /* | |
5848cc09 | 191 | * Don't check napi structures with no associated device |
4ea7e386 | 192 | */ |
5848cc09 | 193 | if (!napi->dev) |
4ea7e386 NH |
194 | return; |
195 | ||
196 | rcu_read_lock(); | |
197 | list_for_each_entry_rcu(new_stat, &hw_stats_list, list) { | |
5848cc09 NH |
198 | /* |
199 | * only add a note to our monitor buffer if: | |
200 | * 1) this is the dev we received on | |
201 | * 2) its after the last_rx delta | |
202 | * 3) our rx_dropped count has gone up | |
203 | */ | |
4ea7e386 | 204 | if ((new_stat->dev == napi->dev) && |
5848cc09 | 205 | (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) && |
4ea7e386 NH |
206 | (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) { |
207 | trace_drop_common(NULL, NULL); | |
208 | new_stat->last_drop_val = napi->dev->stats.rx_dropped; | |
5848cc09 | 209 | new_stat->last_rx = jiffies; |
4ea7e386 NH |
210 | break; |
211 | } | |
212 | } | |
213 | rcu_read_unlock(); | |
214 | } | |
215 | ||
9a8afc8d NH |
216 | static int set_all_monitor_traces(int state) |
217 | { | |
218 | int rc = 0; | |
4ea7e386 NH |
219 | struct dm_hw_stat_delta *new_stat = NULL; |
220 | struct dm_hw_stat_delta *temp; | |
221 | ||
cde2e9a6 | 222 | mutex_lock(&trace_state_mutex); |
9a8afc8d | 223 | |
4b706372 NH |
224 | if (state == trace_state) { |
225 | rc = -EAGAIN; | |
226 | goto out_unlock; | |
227 | } | |
228 | ||
9a8afc8d NH |
229 | switch (state) { |
230 | case TRACE_ON: | |
cad456d5 NH |
231 | if (!try_module_get(THIS_MODULE)) { |
232 | rc = -ENODEV; | |
233 | break; | |
234 | } | |
235 | ||
38516ab5 SR |
236 | rc |= register_trace_kfree_skb(trace_kfree_skb_hit, NULL); |
237 | rc |= register_trace_napi_poll(trace_napi_poll_hit, NULL); | |
9a8afc8d | 238 | break; |
cad456d5 | 239 | |
9a8afc8d | 240 | case TRACE_OFF: |
38516ab5 SR |
241 | rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit, NULL); |
242 | rc |= unregister_trace_napi_poll(trace_napi_poll_hit, NULL); | |
9a8afc8d NH |
243 | |
244 | tracepoint_synchronize_unregister(); | |
4ea7e386 NH |
245 | |
246 | /* | |
247 | * Clean the device list | |
248 | */ | |
249 | list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) { | |
250 | if (new_stat->dev == NULL) { | |
251 | list_del_rcu(&new_stat->list); | |
fa81c0e1 | 252 | kfree_rcu(new_stat, rcu); |
4ea7e386 NH |
253 | } |
254 | } | |
cad456d5 NH |
255 | |
256 | module_put(THIS_MODULE); | |
257 | ||
9a8afc8d NH |
258 | break; |
259 | default: | |
260 | rc = 1; | |
261 | break; | |
262 | } | |
263 | ||
4ea7e386 NH |
264 | if (!rc) |
265 | trace_state = state; | |
4b706372 NH |
266 | else |
267 | rc = -EINPROGRESS; | |
4ea7e386 | 268 | |
4b706372 | 269 | out_unlock: |
cde2e9a6 | 270 | mutex_unlock(&trace_state_mutex); |
4ea7e386 | 271 | |
9a8afc8d NH |
272 | return rc; |
273 | } | |
274 | ||
275 | ||
276 | static int net_dm_cmd_config(struct sk_buff *skb, | |
277 | struct genl_info *info) | |
278 | { | |
279 | return -ENOTSUPP; | |
280 | } | |
281 | ||
282 | static int net_dm_cmd_trace(struct sk_buff *skb, | |
283 | struct genl_info *info) | |
284 | { | |
285 | switch (info->genlhdr->cmd) { | |
286 | case NET_DM_CMD_START: | |
287 | return set_all_monitor_traces(TRACE_ON); | |
288 | break; | |
289 | case NET_DM_CMD_STOP: | |
290 | return set_all_monitor_traces(TRACE_OFF); | |
291 | break; | |
292 | } | |
293 | ||
294 | return -ENOTSUPP; | |
295 | } | |
296 | ||
4ea7e386 NH |
297 | static int dropmon_net_event(struct notifier_block *ev_block, |
298 | unsigned long event, void *ptr) | |
299 | { | |
300 | struct net_device *dev = ptr; | |
301 | struct dm_hw_stat_delta *new_stat = NULL; | |
302 | struct dm_hw_stat_delta *tmp; | |
303 | ||
304 | switch (event) { | |
305 | case NETDEV_REGISTER: | |
306 | new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL); | |
307 | ||
308 | if (!new_stat) | |
309 | goto out; | |
310 | ||
311 | new_stat->dev = dev; | |
5848cc09 | 312 | new_stat->last_rx = jiffies; |
cde2e9a6 | 313 | mutex_lock(&trace_state_mutex); |
4ea7e386 | 314 | list_add_rcu(&new_stat->list, &hw_stats_list); |
cde2e9a6 | 315 | mutex_unlock(&trace_state_mutex); |
4ea7e386 NH |
316 | break; |
317 | case NETDEV_UNREGISTER: | |
cde2e9a6 | 318 | mutex_lock(&trace_state_mutex); |
4ea7e386 NH |
319 | list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) { |
320 | if (new_stat->dev == dev) { | |
321 | new_stat->dev = NULL; | |
322 | if (trace_state == TRACE_OFF) { | |
323 | list_del_rcu(&new_stat->list); | |
fa81c0e1 | 324 | kfree_rcu(new_stat, rcu); |
4ea7e386 NH |
325 | break; |
326 | } | |
327 | } | |
328 | } | |
cde2e9a6 | 329 | mutex_unlock(&trace_state_mutex); |
4ea7e386 NH |
330 | break; |
331 | } | |
332 | out: | |
333 | return NOTIFY_DONE; | |
334 | } | |
9a8afc8d NH |
335 | |
336 | static struct genl_ops dropmon_ops[] = { | |
337 | { | |
338 | .cmd = NET_DM_CMD_CONFIG, | |
339 | .doit = net_dm_cmd_config, | |
340 | }, | |
341 | { | |
342 | .cmd = NET_DM_CMD_START, | |
343 | .doit = net_dm_cmd_trace, | |
344 | }, | |
345 | { | |
346 | .cmd = NET_DM_CMD_STOP, | |
347 | .doit = net_dm_cmd_trace, | |
348 | }, | |
349 | }; | |
350 | ||
4ea7e386 NH |
351 | static struct notifier_block dropmon_net_notifier = { |
352 | .notifier_call = dropmon_net_event | |
353 | }; | |
354 | ||
9a8afc8d NH |
355 | static int __init init_net_drop_monitor(void) |
356 | { | |
9a8afc8d | 357 | struct per_cpu_dm_data *data; |
a256be70 CG |
358 | int cpu, rc; |
359 | ||
e005d193 | 360 | pr_info("Initializing network drop monitor service\n"); |
9a8afc8d NH |
361 | |
362 | if (sizeof(void *) > 8) { | |
e005d193 | 363 | pr_err("Unable to store program counters on this arch, Drop monitor failed\n"); |
9a8afc8d NH |
364 | return -ENOSPC; |
365 | } | |
366 | ||
a256be70 CG |
367 | rc = genl_register_family_with_ops(&net_drop_monitor_family, |
368 | dropmon_ops, | |
369 | ARRAY_SIZE(dropmon_ops)); | |
370 | if (rc) { | |
e005d193 | 371 | pr_err("Could not create drop monitor netlink family\n"); |
a256be70 | 372 | return rc; |
9a8afc8d NH |
373 | } |
374 | ||
4ea7e386 NH |
375 | rc = register_netdevice_notifier(&dropmon_net_notifier); |
376 | if (rc < 0) { | |
e005d193 | 377 | pr_crit("Failed to register netdevice notifier\n"); |
4ea7e386 NH |
378 | goto out_unreg; |
379 | } | |
380 | ||
9a8afc8d NH |
381 | rc = 0; |
382 | ||
cad456d5 | 383 | for_each_possible_cpu(cpu) { |
9a8afc8d | 384 | data = &per_cpu(dm_cpu_data, cpu); |
9a8afc8d NH |
385 | INIT_WORK(&data->dm_alert_work, send_dm_alert); |
386 | init_timer(&data->send_timer); | |
bec4596b | 387 | data->send_timer.data = (unsigned long)data; |
9a8afc8d | 388 | data->send_timer.function = sched_send_work; |
bec4596b | 389 | spin_lock_init(&data->lock); |
4fdcfa12 | 390 | reset_per_cpu_data(data); |
9a8afc8d | 391 | } |
4ea7e386 | 392 | |
3885ca78 | 393 | |
9a8afc8d NH |
394 | goto out; |
395 | ||
396 | out_unreg: | |
397 | genl_unregister_family(&net_drop_monitor_family); | |
398 | out: | |
399 | return rc; | |
400 | } | |
401 | ||
cad456d5 NH |
402 | static void exit_net_drop_monitor(void) |
403 | { | |
404 | struct per_cpu_dm_data *data; | |
405 | int cpu; | |
406 | ||
407 | BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier)); | |
408 | ||
409 | /* | |
410 | * Because of the module_get/put we do in the trace state change path | |
411 | * we are guarnateed not to have any current users when we get here | |
412 | * all we need to do is make sure that we don't have any running timers | |
413 | * or pending schedule calls | |
414 | */ | |
415 | ||
416 | for_each_possible_cpu(cpu) { | |
417 | data = &per_cpu(dm_cpu_data, cpu); | |
418 | del_timer_sync(&data->send_timer); | |
419 | cancel_work_sync(&data->dm_alert_work); | |
420 | /* | |
421 | * At this point, we should have exclusive access | |
422 | * to this struct and can free the skb inside it | |
423 | */ | |
424 | kfree_skb(data->skb); | |
425 | } | |
426 | ||
427 | BUG_ON(genl_unregister_family(&net_drop_monitor_family)); | |
428 | } | |
429 | ||
430 | module_init(init_net_drop_monitor); | |
431 | module_exit(exit_net_drop_monitor); | |
432 | ||
433 | MODULE_LICENSE("GPL v2"); | |
434 | MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>"); | |
3fdcbd45 | 435 | MODULE_ALIAS_GENL_FAMILY("NET_DM"); |