xref: /openbmc/linux/net/core/drop_monitor.c (revision 5e012745)
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/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/genetlink.h>
30 #include <net/netevent.h>
31 
32 #include <trace/events/skb.h>
33 #include <trace/events/napi.h>
34 
35 #include <asm/unaligned.h>
36 
37 #define TRACE_ON 1
38 #define TRACE_OFF 0
39 
40 /*
41  * Globals, our netlink socket pointer
42  * and the work handle that will send up
43  * netlink alerts
44  */
45 static int trace_state = TRACE_OFF;
46 static DEFINE_MUTEX(trace_state_mutex);
47 
48 struct per_cpu_dm_data {
49 	spinlock_t		lock;
50 	struct sk_buff		*skb;
51 	struct work_struct	dm_alert_work;
52 	struct timer_list	send_timer;
53 };
54 
55 struct dm_hw_stat_delta {
56 	struct net_device *dev;
57 	unsigned long last_rx;
58 	struct list_head list;
59 	struct rcu_head rcu;
60 	unsigned long last_drop_val;
61 };
62 
63 static struct genl_family net_drop_monitor_family;
64 
65 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
66 
67 static int dm_hit_limit = 64;
68 static int dm_delay = 1;
69 static unsigned long dm_hw_check_delta = 2*HZ;
70 static LIST_HEAD(hw_stats_list);
71 
72 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
73 {
74 	size_t al;
75 	struct net_dm_alert_msg *msg;
76 	struct nlattr *nla;
77 	struct sk_buff *skb;
78 	unsigned long flags;
79 	void *msg_header;
80 
81 	al = sizeof(struct net_dm_alert_msg);
82 	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
83 	al += sizeof(struct nlattr);
84 
85 	skb = genlmsg_new(al, GFP_KERNEL);
86 
87 	if (!skb)
88 		goto err;
89 
90 	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
91 				 0, NET_DM_CMD_ALERT);
92 	if (!msg_header) {
93 		nlmsg_free(skb);
94 		skb = NULL;
95 		goto err;
96 	}
97 	nla = nla_reserve(skb, NLA_UNSPEC,
98 			  sizeof(struct net_dm_alert_msg));
99 	if (!nla) {
100 		nlmsg_free(skb);
101 		skb = NULL;
102 		goto err;
103 	}
104 	msg = nla_data(nla);
105 	memset(msg, 0, al);
106 	goto out;
107 
108 err:
109 	mod_timer(&data->send_timer, jiffies + HZ / 10);
110 out:
111 	spin_lock_irqsave(&data->lock, flags);
112 	swap(data->skb, skb);
113 	spin_unlock_irqrestore(&data->lock, flags);
114 
115 	if (skb) {
116 		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
117 		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
118 
119 		genlmsg_end(skb, genlmsg_data(gnlh));
120 	}
121 
122 	return skb;
123 }
124 
125 static const struct genl_multicast_group dropmon_mcgrps[] = {
126 	{ .name = "events", },
127 };
128 
129 static void send_dm_alert(struct work_struct *work)
130 {
131 	struct sk_buff *skb;
132 	struct per_cpu_dm_data *data;
133 
134 	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
135 
136 	skb = reset_per_cpu_data(data);
137 
138 	if (skb)
139 		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
140 				  0, GFP_KERNEL);
141 }
142 
143 /*
144  * This is the timer function to delay the sending of an alert
145  * in the event that more drops will arrive during the
146  * hysteresis period.
147  */
148 static void sched_send_work(struct timer_list *t)
149 {
150 	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
151 
152 	schedule_work(&data->dm_alert_work);
153 }
154 
155 static void trace_drop_common(struct sk_buff *skb, void *location)
156 {
157 	struct net_dm_alert_msg *msg;
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 	for (i = 0; i < msg->entries; i++) {
177 		if (!memcmp(&location, msg->points[i].pc, sizeof(void *))) {
178 			msg->points[i].count++;
179 			goto out;
180 		}
181 	}
182 	if (msg->entries == dm_hit_limit)
183 		goto out;
184 	/*
185 	 * We need to create a new entry
186 	 */
187 	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
188 	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
189 	memcpy(msg->points[msg->entries].pc, &location, sizeof(void *));
190 	msg->points[msg->entries].count = 1;
191 	msg->entries++;
192 
193 	if (!timer_pending(&data->send_timer)) {
194 		data->send_timer.expires = jiffies + dm_delay * HZ;
195 		add_timer(&data->send_timer);
196 	}
197 
198 out:
199 	spin_unlock_irqrestore(&data->lock, flags);
200 }
201 
202 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
203 {
204 	trace_drop_common(skb, location);
205 }
206 
207 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
208 				int work, int budget)
209 {
210 	struct dm_hw_stat_delta *new_stat;
211 
212 	/*
213 	 * Don't check napi structures with no associated device
214 	 */
215 	if (!napi->dev)
216 		return;
217 
218 	rcu_read_lock();
219 	list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
220 		/*
221 		 * only add a note to our monitor buffer if:
222 		 * 1) this is the dev we received on
223 		 * 2) its after the last_rx delta
224 		 * 3) our rx_dropped count has gone up
225 		 */
226 		if ((new_stat->dev == napi->dev)  &&
227 		    (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
228 		    (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
229 			trace_drop_common(NULL, NULL);
230 			new_stat->last_drop_val = napi->dev->stats.rx_dropped;
231 			new_stat->last_rx = jiffies;
232 			break;
233 		}
234 	}
235 	rcu_read_unlock();
236 }
237 
238 static int set_all_monitor_traces(int state)
239 {
240 	int rc = 0;
241 	struct dm_hw_stat_delta *new_stat = NULL;
242 	struct dm_hw_stat_delta *temp;
243 
244 	mutex_lock(&trace_state_mutex);
245 
246 	if (state == trace_state) {
247 		rc = -EAGAIN;
248 		goto out_unlock;
249 	}
250 
251 	switch (state) {
252 	case TRACE_ON:
253 		if (!try_module_get(THIS_MODULE)) {
254 			rc = -ENODEV;
255 			break;
256 		}
257 
258 		rc |= register_trace_kfree_skb(trace_kfree_skb_hit, NULL);
259 		rc |= register_trace_napi_poll(trace_napi_poll_hit, NULL);
260 		break;
261 
262 	case TRACE_OFF:
263 		rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit, NULL);
264 		rc |= unregister_trace_napi_poll(trace_napi_poll_hit, NULL);
265 
266 		tracepoint_synchronize_unregister();
267 
268 		/*
269 		 * Clean the device list
270 		 */
271 		list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
272 			if (new_stat->dev == NULL) {
273 				list_del_rcu(&new_stat->list);
274 				kfree_rcu(new_stat, rcu);
275 			}
276 		}
277 
278 		module_put(THIS_MODULE);
279 
280 		break;
281 	default:
282 		rc = 1;
283 		break;
284 	}
285 
286 	if (!rc)
287 		trace_state = state;
288 	else
289 		rc = -EINPROGRESS;
290 
291 out_unlock:
292 	mutex_unlock(&trace_state_mutex);
293 
294 	return rc;
295 }
296 
297 
298 static int net_dm_cmd_config(struct sk_buff *skb,
299 			struct genl_info *info)
300 {
301 	return -ENOTSUPP;
302 }
303 
304 static int net_dm_cmd_trace(struct sk_buff *skb,
305 			struct genl_info *info)
306 {
307 	switch (info->genlhdr->cmd) {
308 	case NET_DM_CMD_START:
309 		return set_all_monitor_traces(TRACE_ON);
310 	case NET_DM_CMD_STOP:
311 		return set_all_monitor_traces(TRACE_OFF);
312 	}
313 
314 	return -ENOTSUPP;
315 }
316 
317 static int dropmon_net_event(struct notifier_block *ev_block,
318 			     unsigned long event, void *ptr)
319 {
320 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
321 	struct dm_hw_stat_delta *new_stat = NULL;
322 	struct dm_hw_stat_delta *tmp;
323 
324 	switch (event) {
325 	case NETDEV_REGISTER:
326 		new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
327 
328 		if (!new_stat)
329 			goto out;
330 
331 		new_stat->dev = dev;
332 		new_stat->last_rx = jiffies;
333 		mutex_lock(&trace_state_mutex);
334 		list_add_rcu(&new_stat->list, &hw_stats_list);
335 		mutex_unlock(&trace_state_mutex);
336 		break;
337 	case NETDEV_UNREGISTER:
338 		mutex_lock(&trace_state_mutex);
339 		list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
340 			if (new_stat->dev == dev) {
341 				new_stat->dev = NULL;
342 				if (trace_state == TRACE_OFF) {
343 					list_del_rcu(&new_stat->list);
344 					kfree_rcu(new_stat, rcu);
345 					break;
346 				}
347 			}
348 		}
349 		mutex_unlock(&trace_state_mutex);
350 		break;
351 	}
352 out:
353 	return NOTIFY_DONE;
354 }
355 
356 static const struct genl_ops dropmon_ops[] = {
357 	{
358 		.cmd = NET_DM_CMD_CONFIG,
359 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
360 		.doit = net_dm_cmd_config,
361 	},
362 	{
363 		.cmd = NET_DM_CMD_START,
364 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
365 		.doit = net_dm_cmd_trace,
366 	},
367 	{
368 		.cmd = NET_DM_CMD_STOP,
369 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
370 		.doit = net_dm_cmd_trace,
371 	},
372 };
373 
374 static struct genl_family net_drop_monitor_family __ro_after_init = {
375 	.hdrsize        = 0,
376 	.name           = "NET_DM",
377 	.version        = 2,
378 	.module		= THIS_MODULE,
379 	.ops		= dropmon_ops,
380 	.n_ops		= ARRAY_SIZE(dropmon_ops),
381 	.mcgrps		= dropmon_mcgrps,
382 	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
383 };
384 
385 static struct notifier_block dropmon_net_notifier = {
386 	.notifier_call = dropmon_net_event
387 };
388 
389 static int __init init_net_drop_monitor(void)
390 {
391 	struct per_cpu_dm_data *data;
392 	int cpu, rc;
393 
394 	pr_info("Initializing network drop monitor service\n");
395 
396 	if (sizeof(void *) > 8) {
397 		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
398 		return -ENOSPC;
399 	}
400 
401 	rc = genl_register_family(&net_drop_monitor_family);
402 	if (rc) {
403 		pr_err("Could not create drop monitor netlink family\n");
404 		return rc;
405 	}
406 	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
407 
408 	rc = register_netdevice_notifier(&dropmon_net_notifier);
409 	if (rc < 0) {
410 		pr_crit("Failed to register netdevice notifier\n");
411 		goto out_unreg;
412 	}
413 
414 	rc = 0;
415 
416 	for_each_possible_cpu(cpu) {
417 		data = &per_cpu(dm_cpu_data, cpu);
418 		INIT_WORK(&data->dm_alert_work, send_dm_alert);
419 		timer_setup(&data->send_timer, sched_send_work, 0);
420 		spin_lock_init(&data->lock);
421 		reset_per_cpu_data(data);
422 	}
423 
424 
425 	goto out;
426 
427 out_unreg:
428 	genl_unregister_family(&net_drop_monitor_family);
429 out:
430 	return rc;
431 }
432 
433 static void exit_net_drop_monitor(void)
434 {
435 	struct per_cpu_dm_data *data;
436 	int cpu;
437 
438 	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
439 
440 	/*
441 	 * Because of the module_get/put we do in the trace state change path
442 	 * we are guarnateed not to have any current users when we get here
443 	 * all we need to do is make sure that we don't have any running timers
444 	 * or pending schedule calls
445 	 */
446 
447 	for_each_possible_cpu(cpu) {
448 		data = &per_cpu(dm_cpu_data, cpu);
449 		del_timer_sync(&data->send_timer);
450 		cancel_work_sync(&data->dm_alert_work);
451 		/*
452 		 * At this point, we should have exclusive access
453 		 * to this struct and can free the skb inside it
454 		 */
455 		kfree_skb(data->skb);
456 	}
457 
458 	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
459 }
460 
461 module_init(init_net_drop_monitor);
462 module_exit(exit_net_drop_monitor);
463 
464 MODULE_LICENSE("GPL v2");
465 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
466 MODULE_ALIAS_GENL_FAMILY("NET_DM");
467