1 /*
2 * Monitoring code for network dropped packet alerts
3 *
4 * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com>
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
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>
26 #include <linux/slab.h>
27 #include <linux/module.h>
28 #include <net/genetlink.h>
29 #include <net/netevent.h>
30
31 #include <trace/events/skb.h>
32 #include <trace/events/napi.h>
33
34 #include <asm/unaligned.h>
35
36 #define TRACE_ON 1
37 #define TRACE_OFF 0
38
39 /*
40 * Globals, our netlink socket pointer
41 * and the work handle that will send up
42 * netlink alerts
43 */
44 static int trace_state = TRACE_OFF;
45 static DEFINE_MUTEX(trace_state_mutex);
46
47 struct per_cpu_dm_data {
48 spinlock_t lock;
49 struct sk_buff *skb;
50 struct work_struct dm_alert_work;
51 struct timer_list send_timer;
52 };
53
54 struct dm_hw_stat_delta {
55 struct net_device *dev;
56 unsigned long last_rx;
57 struct list_head list;
58 struct rcu_head rcu;
59 unsigned long last_drop_val;
60 };
61
62 static struct genl_family net_drop_monitor_family;
63
64 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
65
66 static int dm_hit_limit = 64;
67 static int dm_delay = 1;
68 static unsigned long dm_hw_check_delta = 2*HZ;
69 static LIST_HEAD(hw_stats_list);
70
reset_per_cpu_data(struct per_cpu_dm_data * data)71 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
72 {
73 size_t al;
74 struct net_dm_alert_msg *msg;
75 struct nlattr *nla;
76 struct sk_buff *skb;
77 unsigned long flags;
78 void *msg_header;
79
80 al = sizeof(struct net_dm_alert_msg);
81 al += dm_hit_limit * sizeof(struct net_dm_drop_point);
82 al += sizeof(struct nlattr);
83
84 skb = genlmsg_new(al, GFP_KERNEL);
85
86 if (!skb)
87 goto err;
88
89 msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
90 0, NET_DM_CMD_ALERT);
91 if (!msg_header) {
92 nlmsg_free(skb);
93 skb = NULL;
94 goto err;
95 }
96 nla = nla_reserve(skb, NLA_UNSPEC,
97 sizeof(struct net_dm_alert_msg));
98 if (!nla) {
99 nlmsg_free(skb);
100 skb = NULL;
101 goto err;
102 }
103 msg = nla_data(nla);
104 memset(msg, 0, al);
105 goto out;
106
107 err:
108 mod_timer(&data->send_timer, jiffies + HZ / 10);
109 out:
110 spin_lock_irqsave(&data->lock, flags);
111 swap(data->skb, skb);
112 spin_unlock_irqrestore(&data->lock, flags);
113
114 if (skb) {
115 struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
116 struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
117
118 genlmsg_end(skb, genlmsg_data(gnlh));
119 }
120
121 return skb;
122 }
123
124 static const struct genl_multicast_group dropmon_mcgrps[] = {
125 { .name = "events", },
126 };
127
send_dm_alert(struct work_struct * work)128 static void send_dm_alert(struct work_struct *work)
129 {
130 struct sk_buff *skb;
131 struct per_cpu_dm_data *data;
132
133 data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
134
135 skb = reset_per_cpu_data(data);
136
137 if (skb)
138 genlmsg_multicast(&net_drop_monitor_family, skb, 0,
139 0, GFP_KERNEL);
140 }
141
142 /*
143 * This is the timer function to delay the sending of an alert
144 * in the event that more drops will arrive during the
145 * hysteresis period.
146 */
sched_send_work(struct timer_list * t)147 static void sched_send_work(struct timer_list *t)
148 {
149 struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
150
151 schedule_work(&data->dm_alert_work);
152 }
153
trace_drop_common(struct sk_buff * skb,void * location)154 static void trace_drop_common(struct sk_buff *skb, void *location)
155 {
156 struct net_dm_alert_msg *msg;
157 struct net_dm_drop_point *point;
158 struct nlmsghdr *nlh;
159 struct nlattr *nla;
160 int i;
161 struct sk_buff *dskb;
162 struct per_cpu_dm_data *data;
163 unsigned long flags;
164
165 local_irq_save(flags);
166 data = this_cpu_ptr(&dm_cpu_data);
167 spin_lock(&data->lock);
168 dskb = data->skb;
169
170 if (!dskb)
171 goto out;
172
173 nlh = (struct nlmsghdr *)dskb->data;
174 nla = genlmsg_data(nlmsg_data(nlh));
175 msg = nla_data(nla);
176 point = msg->points;
177 for (i = 0; i < msg->entries; i++) {
178 if (!memcmp(&location, &point->pc, sizeof(void *))) {
179 point->count++;
180 goto out;
181 }
182 point++;
183 }
184 if (msg->entries == dm_hit_limit)
185 goto out;
186 /*
187 * We need to create a new entry
188 */
189 __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
190 nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
191 memcpy(point->pc, &location, sizeof(void *));
192 point->count = 1;
193 msg->entries++;
194
195 if (!timer_pending(&data->send_timer)) {
196 data->send_timer.expires = jiffies + dm_delay * HZ;
197 add_timer(&data->send_timer);
198 }
199
200 out:
201 spin_unlock_irqrestore(&data->lock, flags);
202 }
203
trace_kfree_skb_hit(void * ignore,struct sk_buff * skb,void * location)204 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
205 {
206 trace_drop_common(skb, location);
207 }
208
trace_napi_poll_hit(void * ignore,struct napi_struct * napi,int work,int budget)209 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
210 int work, int budget)
211 {
212 struct dm_hw_stat_delta *new_stat;
213
214 /*
215 * Don't check napi structures with no associated device
216 */
217 if (!napi->dev)
218 return;
219
220 rcu_read_lock();
221 list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
222 /*
223 * only add a note to our monitor buffer if:
224 * 1) this is the dev we received on
225 * 2) its after the last_rx delta
226 * 3) our rx_dropped count has gone up
227 */
228 if ((new_stat->dev == napi->dev) &&
229 (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
230 (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
231 trace_drop_common(NULL, NULL);
232 new_stat->last_drop_val = napi->dev->stats.rx_dropped;
233 new_stat->last_rx = jiffies;
234 break;
235 }
236 }
237 rcu_read_unlock();
238 }
239
set_all_monitor_traces(int state)240 static int set_all_monitor_traces(int state)
241 {
242 int rc = 0;
243 struct dm_hw_stat_delta *new_stat = NULL;
244 struct dm_hw_stat_delta *temp;
245
246 mutex_lock(&trace_state_mutex);
247
248 if (state == trace_state) {
249 rc = -EAGAIN;
250 goto out_unlock;
251 }
252
253 switch (state) {
254 case TRACE_ON:
255 if (!try_module_get(THIS_MODULE)) {
256 rc = -ENODEV;
257 break;
258 }
259
260 rc |= register_trace_kfree_skb(trace_kfree_skb_hit, NULL);
261 rc |= register_trace_napi_poll(trace_napi_poll_hit, NULL);
262 break;
263
264 case TRACE_OFF:
265 rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit, NULL);
266 rc |= unregister_trace_napi_poll(trace_napi_poll_hit, NULL);
267
268 tracepoint_synchronize_unregister();
269
270 /*
271 * Clean the device list
272 */
273 list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
274 if (new_stat->dev == NULL) {
275 list_del_rcu(&new_stat->list);
276 kfree_rcu(new_stat, rcu);
277 }
278 }
279
280 module_put(THIS_MODULE);
281
282 break;
283 default:
284 rc = 1;
285 break;
286 }
287
288 if (!rc)
289 trace_state = state;
290 else
291 rc = -EINPROGRESS;
292
293 out_unlock:
294 mutex_unlock(&trace_state_mutex);
295
296 return rc;
297 }
298
299
net_dm_cmd_config(struct sk_buff * skb,struct genl_info * info)300 static int net_dm_cmd_config(struct sk_buff *skb,
301 struct genl_info *info)
302 {
303 return -ENOTSUPP;
304 }
305
net_dm_cmd_trace(struct sk_buff * skb,struct genl_info * info)306 static int net_dm_cmd_trace(struct sk_buff *skb,
307 struct genl_info *info)
308 {
309 switch (info->genlhdr->cmd) {
310 case NET_DM_CMD_START:
311 return set_all_monitor_traces(TRACE_ON);
312 case NET_DM_CMD_STOP:
313 return set_all_monitor_traces(TRACE_OFF);
314 }
315
316 return -ENOTSUPP;
317 }
318
dropmon_net_event(struct notifier_block * ev_block,unsigned long event,void * ptr)319 static int dropmon_net_event(struct notifier_block *ev_block,
320 unsigned long event, void *ptr)
321 {
322 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
323 struct dm_hw_stat_delta *new_stat = NULL;
324 struct dm_hw_stat_delta *tmp;
325
326 switch (event) {
327 case NETDEV_REGISTER:
328 new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
329
330 if (!new_stat)
331 goto out;
332
333 new_stat->dev = dev;
334 new_stat->last_rx = jiffies;
335 mutex_lock(&trace_state_mutex);
336 list_add_rcu(&new_stat->list, &hw_stats_list);
337 mutex_unlock(&trace_state_mutex);
338 break;
339 case NETDEV_UNREGISTER:
340 mutex_lock(&trace_state_mutex);
341 list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
342 if (new_stat->dev == dev) {
343 new_stat->dev = NULL;
344 if (trace_state == TRACE_OFF) {
345 list_del_rcu(&new_stat->list);
346 kfree_rcu(new_stat, rcu);
347 break;
348 }
349 }
350 }
351 mutex_unlock(&trace_state_mutex);
352 break;
353 }
354 out:
355 return NOTIFY_DONE;
356 }
357
358 static const struct genl_ops dropmon_ops[] = {
359 {
360 .cmd = NET_DM_CMD_CONFIG,
361 .doit = net_dm_cmd_config,
362 },
363 {
364 .cmd = NET_DM_CMD_START,
365 .doit = net_dm_cmd_trace,
366 },
367 {
368 .cmd = NET_DM_CMD_STOP,
369 .doit = net_dm_cmd_trace,
370 },
371 };
372
373 static struct genl_family net_drop_monitor_family __ro_after_init = {
374 .hdrsize = 0,
375 .name = "NET_DM",
376 .version = 2,
377 .module = THIS_MODULE,
378 .ops = dropmon_ops,
379 .n_ops = ARRAY_SIZE(dropmon_ops),
380 .mcgrps = dropmon_mcgrps,
381 .n_mcgrps = ARRAY_SIZE(dropmon_mcgrps),
382 };
383
384 static struct notifier_block dropmon_net_notifier = {
385 .notifier_call = dropmon_net_event
386 };
387
init_net_drop_monitor(void)388 static int __init init_net_drop_monitor(void)
389 {
390 struct per_cpu_dm_data *data;
391 int cpu, rc;
392
393 pr_info("Initializing network drop monitor service\n");
394
395 if (sizeof(void *) > 8) {
396 pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
397 return -ENOSPC;
398 }
399
400 rc = genl_register_family(&net_drop_monitor_family);
401 if (rc) {
402 pr_err("Could not create drop monitor netlink family\n");
403 return rc;
404 }
405 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
406
407 rc = register_netdevice_notifier(&dropmon_net_notifier);
408 if (rc < 0) {
409 pr_crit("Failed to register netdevice notifier\n");
410 goto out_unreg;
411 }
412
413 rc = 0;
414
415 for_each_possible_cpu(cpu) {
416 data = &per_cpu(dm_cpu_data, cpu);
417 INIT_WORK(&data->dm_alert_work, send_dm_alert);
418 timer_setup(&data->send_timer, sched_send_work, 0);
419 spin_lock_init(&data->lock);
420 reset_per_cpu_data(data);
421 }
422
423
424 goto out;
425
426 out_unreg:
427 genl_unregister_family(&net_drop_monitor_family);
428 out:
429 return rc;
430 }
431
exit_net_drop_monitor(void)432 static void exit_net_drop_monitor(void)
433 {
434 struct per_cpu_dm_data *data;
435 int cpu;
436
437 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
438
439 /*
440 * Because of the module_get/put we do in the trace state change path
441 * we are guarnateed not to have any current users when we get here
442 * all we need to do is make sure that we don't have any running timers
443 * or pending schedule calls
444 */
445
446 for_each_possible_cpu(cpu) {
447 data = &per_cpu(dm_cpu_data, cpu);
448 del_timer_sync(&data->send_timer);
449 cancel_work_sync(&data->dm_alert_work);
450 /*
451 * At this point, we should have exclusive access
452 * to this struct and can free the skb inside it
453 */
454 kfree_skb(data->skb);
455 }
456
457 BUG_ON(genl_unregister_family(&net_drop_monitor_family));
458 }
459
460 module_init(init_net_drop_monitor);
461 module_exit(exit_net_drop_monitor);
462
463 MODULE_LICENSE("GPL v2");
464 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
465 MODULE_ALIAS_GENL_FAMILY("NET_DM");
466