xref: /linux/net/core/net-sysfs.c (revision f8324e20f8289dffc646d64366332e05eaacab25)
1 /*
2  * net-sysfs.c - network device class and attributes
3  *
4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5  *
6  *	This program is free software; you can redistribute it and/or
7  *	modify it under the terms of the GNU General Public License
8  *	as published by the Free Software Foundation; either version
9  *	2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/capability.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/if_arp.h>
16 #include <linux/slab.h>
17 #include <linux/nsproxy.h>
18 #include <net/sock.h>
19 #include <net/net_namespace.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/wireless.h>
22 #include <linux/vmalloc.h>
23 #include <net/wext.h>
24 
25 #include "net-sysfs.h"
26 
27 #ifdef CONFIG_SYSFS
28 static const char fmt_hex[] = "%#x\n";
29 static const char fmt_long_hex[] = "%#lx\n";
30 static const char fmt_dec[] = "%d\n";
31 static const char fmt_ulong[] = "%lu\n";
32 
33 static inline int dev_isalive(const struct net_device *dev)
34 {
35 	return dev->reg_state <= NETREG_REGISTERED;
36 }
37 
38 /* use same locking rules as GIF* ioctl's */
39 static ssize_t netdev_show(const struct device *dev,
40 			   struct device_attribute *attr, char *buf,
41 			   ssize_t (*format)(const struct net_device *, char *))
42 {
43 	struct net_device *net = to_net_dev(dev);
44 	ssize_t ret = -EINVAL;
45 
46 	read_lock(&dev_base_lock);
47 	if (dev_isalive(net))
48 		ret = (*format)(net, buf);
49 	read_unlock(&dev_base_lock);
50 
51 	return ret;
52 }
53 
54 /* generate a show function for simple field */
55 #define NETDEVICE_SHOW(field, format_string)				\
56 static ssize_t format_##field(const struct net_device *net, char *buf)	\
57 {									\
58 	return sprintf(buf, format_string, net->field);			\
59 }									\
60 static ssize_t show_##field(struct device *dev,				\
61 			    struct device_attribute *attr, char *buf)	\
62 {									\
63 	return netdev_show(dev, attr, buf, format_##field);		\
64 }
65 
66 
67 /* use same locking and permission rules as SIF* ioctl's */
68 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
69 			    const char *buf, size_t len,
70 			    int (*set)(struct net_device *, unsigned long))
71 {
72 	struct net_device *net = to_net_dev(dev);
73 	char *endp;
74 	unsigned long new;
75 	int ret = -EINVAL;
76 
77 	if (!capable(CAP_NET_ADMIN))
78 		return -EPERM;
79 
80 	new = simple_strtoul(buf, &endp, 0);
81 	if (endp == buf)
82 		goto err;
83 
84 	if (!rtnl_trylock())
85 		return restart_syscall();
86 
87 	if (dev_isalive(net)) {
88 		if ((ret = (*set)(net, new)) == 0)
89 			ret = len;
90 	}
91 	rtnl_unlock();
92  err:
93 	return ret;
94 }
95 
96 NETDEVICE_SHOW(dev_id, fmt_hex);
97 NETDEVICE_SHOW(addr_len, fmt_dec);
98 NETDEVICE_SHOW(iflink, fmt_dec);
99 NETDEVICE_SHOW(ifindex, fmt_dec);
100 NETDEVICE_SHOW(features, fmt_long_hex);
101 NETDEVICE_SHOW(type, fmt_dec);
102 NETDEVICE_SHOW(link_mode, fmt_dec);
103 
104 /* use same locking rules as GIFHWADDR ioctl's */
105 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
106 			    char *buf)
107 {
108 	struct net_device *net = to_net_dev(dev);
109 	ssize_t ret = -EINVAL;
110 
111 	read_lock(&dev_base_lock);
112 	if (dev_isalive(net))
113 		ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
114 	read_unlock(&dev_base_lock);
115 	return ret;
116 }
117 
118 static ssize_t show_broadcast(struct device *dev,
119 			    struct device_attribute *attr, char *buf)
120 {
121 	struct net_device *net = to_net_dev(dev);
122 	if (dev_isalive(net))
123 		return sysfs_format_mac(buf, net->broadcast, net->addr_len);
124 	return -EINVAL;
125 }
126 
127 static ssize_t show_carrier(struct device *dev,
128 			    struct device_attribute *attr, char *buf)
129 {
130 	struct net_device *netdev = to_net_dev(dev);
131 	if (netif_running(netdev)) {
132 		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
133 	}
134 	return -EINVAL;
135 }
136 
137 static ssize_t show_speed(struct device *dev,
138 			  struct device_attribute *attr, char *buf)
139 {
140 	struct net_device *netdev = to_net_dev(dev);
141 	int ret = -EINVAL;
142 
143 	if (!rtnl_trylock())
144 		return restart_syscall();
145 
146 	if (netif_running(netdev) &&
147 	    netdev->ethtool_ops &&
148 	    netdev->ethtool_ops->get_settings) {
149 		struct ethtool_cmd cmd = { ETHTOOL_GSET };
150 
151 		if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
152 			ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
153 	}
154 	rtnl_unlock();
155 	return ret;
156 }
157 
158 static ssize_t show_duplex(struct device *dev,
159 			   struct device_attribute *attr, char *buf)
160 {
161 	struct net_device *netdev = to_net_dev(dev);
162 	int ret = -EINVAL;
163 
164 	if (!rtnl_trylock())
165 		return restart_syscall();
166 
167 	if (netif_running(netdev) &&
168 	    netdev->ethtool_ops &&
169 	    netdev->ethtool_ops->get_settings) {
170 		struct ethtool_cmd cmd = { ETHTOOL_GSET };
171 
172 		if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
173 			ret = sprintf(buf, "%s\n", cmd.duplex ? "full" : "half");
174 	}
175 	rtnl_unlock();
176 	return ret;
177 }
178 
179 static ssize_t show_dormant(struct device *dev,
180 			    struct device_attribute *attr, char *buf)
181 {
182 	struct net_device *netdev = to_net_dev(dev);
183 
184 	if (netif_running(netdev))
185 		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
186 
187 	return -EINVAL;
188 }
189 
190 static const char *const operstates[] = {
191 	"unknown",
192 	"notpresent", /* currently unused */
193 	"down",
194 	"lowerlayerdown",
195 	"testing", /* currently unused */
196 	"dormant",
197 	"up"
198 };
199 
200 static ssize_t show_operstate(struct device *dev,
201 			      struct device_attribute *attr, char *buf)
202 {
203 	const struct net_device *netdev = to_net_dev(dev);
204 	unsigned char operstate;
205 
206 	read_lock(&dev_base_lock);
207 	operstate = netdev->operstate;
208 	if (!netif_running(netdev))
209 		operstate = IF_OPER_DOWN;
210 	read_unlock(&dev_base_lock);
211 
212 	if (operstate >= ARRAY_SIZE(operstates))
213 		return -EINVAL; /* should not happen */
214 
215 	return sprintf(buf, "%s\n", operstates[operstate]);
216 }
217 
218 /* read-write attributes */
219 NETDEVICE_SHOW(mtu, fmt_dec);
220 
221 static int change_mtu(struct net_device *net, unsigned long new_mtu)
222 {
223 	return dev_set_mtu(net, (int) new_mtu);
224 }
225 
226 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
227 			 const char *buf, size_t len)
228 {
229 	return netdev_store(dev, attr, buf, len, change_mtu);
230 }
231 
232 NETDEVICE_SHOW(flags, fmt_hex);
233 
234 static int change_flags(struct net_device *net, unsigned long new_flags)
235 {
236 	return dev_change_flags(net, (unsigned) new_flags);
237 }
238 
239 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
240 			   const char *buf, size_t len)
241 {
242 	return netdev_store(dev, attr, buf, len, change_flags);
243 }
244 
245 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
246 
247 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
248 {
249 	net->tx_queue_len = new_len;
250 	return 0;
251 }
252 
253 static ssize_t store_tx_queue_len(struct device *dev,
254 				  struct device_attribute *attr,
255 				  const char *buf, size_t len)
256 {
257 	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
258 }
259 
260 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
261 			     const char *buf, size_t len)
262 {
263 	struct net_device *netdev = to_net_dev(dev);
264 	size_t count = len;
265 	ssize_t ret;
266 
267 	if (!capable(CAP_NET_ADMIN))
268 		return -EPERM;
269 
270 	/* ignore trailing newline */
271 	if (len >  0 && buf[len - 1] == '\n')
272 		--count;
273 
274 	if (!rtnl_trylock())
275 		return restart_syscall();
276 	ret = dev_set_alias(netdev, buf, count);
277 	rtnl_unlock();
278 
279 	return ret < 0 ? ret : len;
280 }
281 
282 static ssize_t show_ifalias(struct device *dev,
283 			    struct device_attribute *attr, char *buf)
284 {
285 	const struct net_device *netdev = to_net_dev(dev);
286 	ssize_t ret = 0;
287 
288 	if (!rtnl_trylock())
289 		return restart_syscall();
290 	if (netdev->ifalias)
291 		ret = sprintf(buf, "%s\n", netdev->ifalias);
292 	rtnl_unlock();
293 	return ret;
294 }
295 
296 static struct device_attribute net_class_attributes[] = {
297 	__ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
298 	__ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
299 	__ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
300 	__ATTR(iflink, S_IRUGO, show_iflink, NULL),
301 	__ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
302 	__ATTR(features, S_IRUGO, show_features, NULL),
303 	__ATTR(type, S_IRUGO, show_type, NULL),
304 	__ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
305 	__ATTR(address, S_IRUGO, show_address, NULL),
306 	__ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
307 	__ATTR(carrier, S_IRUGO, show_carrier, NULL),
308 	__ATTR(speed, S_IRUGO, show_speed, NULL),
309 	__ATTR(duplex, S_IRUGO, show_duplex, NULL),
310 	__ATTR(dormant, S_IRUGO, show_dormant, NULL),
311 	__ATTR(operstate, S_IRUGO, show_operstate, NULL),
312 	__ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
313 	__ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
314 	__ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
315 	       store_tx_queue_len),
316 	{}
317 };
318 
319 /* Show a given an attribute in the statistics group */
320 static ssize_t netstat_show(const struct device *d,
321 			    struct device_attribute *attr, char *buf,
322 			    unsigned long offset)
323 {
324 	struct net_device *dev = to_net_dev(d);
325 	ssize_t ret = -EINVAL;
326 
327 	WARN_ON(offset > sizeof(struct net_device_stats) ||
328 			offset % sizeof(unsigned long) != 0);
329 
330 	read_lock(&dev_base_lock);
331 	if (dev_isalive(dev)) {
332 		const struct net_device_stats *stats = dev_get_stats(dev);
333 		ret = sprintf(buf, fmt_ulong,
334 			      *(unsigned long *)(((u8 *) stats) + offset));
335 	}
336 	read_unlock(&dev_base_lock);
337 	return ret;
338 }
339 
340 /* generate a read-only statistics attribute */
341 #define NETSTAT_ENTRY(name)						\
342 static ssize_t show_##name(struct device *d,				\
343 			   struct device_attribute *attr, char *buf) 	\
344 {									\
345 	return netstat_show(d, attr, buf,				\
346 			    offsetof(struct net_device_stats, name));	\
347 }									\
348 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
349 
350 NETSTAT_ENTRY(rx_packets);
351 NETSTAT_ENTRY(tx_packets);
352 NETSTAT_ENTRY(rx_bytes);
353 NETSTAT_ENTRY(tx_bytes);
354 NETSTAT_ENTRY(rx_errors);
355 NETSTAT_ENTRY(tx_errors);
356 NETSTAT_ENTRY(rx_dropped);
357 NETSTAT_ENTRY(tx_dropped);
358 NETSTAT_ENTRY(multicast);
359 NETSTAT_ENTRY(collisions);
360 NETSTAT_ENTRY(rx_length_errors);
361 NETSTAT_ENTRY(rx_over_errors);
362 NETSTAT_ENTRY(rx_crc_errors);
363 NETSTAT_ENTRY(rx_frame_errors);
364 NETSTAT_ENTRY(rx_fifo_errors);
365 NETSTAT_ENTRY(rx_missed_errors);
366 NETSTAT_ENTRY(tx_aborted_errors);
367 NETSTAT_ENTRY(tx_carrier_errors);
368 NETSTAT_ENTRY(tx_fifo_errors);
369 NETSTAT_ENTRY(tx_heartbeat_errors);
370 NETSTAT_ENTRY(tx_window_errors);
371 NETSTAT_ENTRY(rx_compressed);
372 NETSTAT_ENTRY(tx_compressed);
373 
374 static struct attribute *netstat_attrs[] = {
375 	&dev_attr_rx_packets.attr,
376 	&dev_attr_tx_packets.attr,
377 	&dev_attr_rx_bytes.attr,
378 	&dev_attr_tx_bytes.attr,
379 	&dev_attr_rx_errors.attr,
380 	&dev_attr_tx_errors.attr,
381 	&dev_attr_rx_dropped.attr,
382 	&dev_attr_tx_dropped.attr,
383 	&dev_attr_multicast.attr,
384 	&dev_attr_collisions.attr,
385 	&dev_attr_rx_length_errors.attr,
386 	&dev_attr_rx_over_errors.attr,
387 	&dev_attr_rx_crc_errors.attr,
388 	&dev_attr_rx_frame_errors.attr,
389 	&dev_attr_rx_fifo_errors.attr,
390 	&dev_attr_rx_missed_errors.attr,
391 	&dev_attr_tx_aborted_errors.attr,
392 	&dev_attr_tx_carrier_errors.attr,
393 	&dev_attr_tx_fifo_errors.attr,
394 	&dev_attr_tx_heartbeat_errors.attr,
395 	&dev_attr_tx_window_errors.attr,
396 	&dev_attr_rx_compressed.attr,
397 	&dev_attr_tx_compressed.attr,
398 	NULL
399 };
400 
401 
402 static struct attribute_group netstat_group = {
403 	.name  = "statistics",
404 	.attrs  = netstat_attrs,
405 };
406 
407 #ifdef CONFIG_WIRELESS_EXT_SYSFS
408 /* helper function that does all the locking etc for wireless stats */
409 static ssize_t wireless_show(struct device *d, char *buf,
410 			     ssize_t (*format)(const struct iw_statistics *,
411 					       char *))
412 {
413 	struct net_device *dev = to_net_dev(d);
414 	const struct iw_statistics *iw;
415 	ssize_t ret = -EINVAL;
416 
417 	if (!rtnl_trylock())
418 		return restart_syscall();
419 	if (dev_isalive(dev)) {
420 		iw = get_wireless_stats(dev);
421 		if (iw)
422 			ret = (*format)(iw, buf);
423 	}
424 	rtnl_unlock();
425 
426 	return ret;
427 }
428 
429 /* show function template for wireless fields */
430 #define WIRELESS_SHOW(name, field, format_string)			\
431 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
432 {									\
433 	return sprintf(buf, format_string, iw->field);			\
434 }									\
435 static ssize_t show_iw_##name(struct device *d,				\
436 			      struct device_attribute *attr, char *buf)	\
437 {									\
438 	return wireless_show(d, buf, format_iw_##name);			\
439 }									\
440 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
441 
442 WIRELESS_SHOW(status, status, fmt_hex);
443 WIRELESS_SHOW(link, qual.qual, fmt_dec);
444 WIRELESS_SHOW(level, qual.level, fmt_dec);
445 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
446 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
447 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
448 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
449 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
450 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
451 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
452 
453 static struct attribute *wireless_attrs[] = {
454 	&dev_attr_status.attr,
455 	&dev_attr_link.attr,
456 	&dev_attr_level.attr,
457 	&dev_attr_noise.attr,
458 	&dev_attr_nwid.attr,
459 	&dev_attr_crypt.attr,
460 	&dev_attr_fragment.attr,
461 	&dev_attr_retries.attr,
462 	&dev_attr_misc.attr,
463 	&dev_attr_beacon.attr,
464 	NULL
465 };
466 
467 static struct attribute_group wireless_group = {
468 	.name = "wireless",
469 	.attrs = wireless_attrs,
470 };
471 #endif
472 #endif /* CONFIG_SYSFS */
473 
474 #ifdef CONFIG_RPS
475 /*
476  * RX queue sysfs structures and functions.
477  */
478 struct rx_queue_attribute {
479 	struct attribute attr;
480 	ssize_t (*show)(struct netdev_rx_queue *queue,
481 	    struct rx_queue_attribute *attr, char *buf);
482 	ssize_t (*store)(struct netdev_rx_queue *queue,
483 	    struct rx_queue_attribute *attr, const char *buf, size_t len);
484 };
485 #define to_rx_queue_attr(_attr) container_of(_attr,		\
486     struct rx_queue_attribute, attr)
487 
488 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
489 
490 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
491 				  char *buf)
492 {
493 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
494 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
495 
496 	if (!attribute->show)
497 		return -EIO;
498 
499 	return attribute->show(queue, attribute, buf);
500 }
501 
502 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
503 				   const char *buf, size_t count)
504 {
505 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
506 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
507 
508 	if (!attribute->store)
509 		return -EIO;
510 
511 	return attribute->store(queue, attribute, buf, count);
512 }
513 
514 static struct sysfs_ops rx_queue_sysfs_ops = {
515 	.show = rx_queue_attr_show,
516 	.store = rx_queue_attr_store,
517 };
518 
519 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
520 			    struct rx_queue_attribute *attribute, char *buf)
521 {
522 	struct rps_map *map;
523 	cpumask_var_t mask;
524 	size_t len = 0;
525 	int i;
526 
527 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
528 		return -ENOMEM;
529 
530 	rcu_read_lock();
531 	map = rcu_dereference(queue->rps_map);
532 	if (map)
533 		for (i = 0; i < map->len; i++)
534 			cpumask_set_cpu(map->cpus[i], mask);
535 
536 	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
537 	if (PAGE_SIZE - len < 3) {
538 		rcu_read_unlock();
539 		free_cpumask_var(mask);
540 		return -EINVAL;
541 	}
542 	rcu_read_unlock();
543 
544 	free_cpumask_var(mask);
545 	len += sprintf(buf + len, "\n");
546 	return len;
547 }
548 
549 static void rps_map_release(struct rcu_head *rcu)
550 {
551 	struct rps_map *map = container_of(rcu, struct rps_map, rcu);
552 
553 	kfree(map);
554 }
555 
556 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
557 		      struct rx_queue_attribute *attribute,
558 		      const char *buf, size_t len)
559 {
560 	struct rps_map *old_map, *map;
561 	cpumask_var_t mask;
562 	int err, cpu, i;
563 	static DEFINE_SPINLOCK(rps_map_lock);
564 
565 	if (!capable(CAP_NET_ADMIN))
566 		return -EPERM;
567 
568 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
569 		return -ENOMEM;
570 
571 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
572 	if (err) {
573 		free_cpumask_var(mask);
574 		return err;
575 	}
576 
577 	map = kzalloc(max_t(unsigned,
578 	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
579 	    GFP_KERNEL);
580 	if (!map) {
581 		free_cpumask_var(mask);
582 		return -ENOMEM;
583 	}
584 
585 	i = 0;
586 	for_each_cpu_and(cpu, mask, cpu_online_mask)
587 		map->cpus[i++] = cpu;
588 
589 	if (i)
590 		map->len = i;
591 	else {
592 		kfree(map);
593 		map = NULL;
594 	}
595 
596 	spin_lock(&rps_map_lock);
597 	old_map = queue->rps_map;
598 	rcu_assign_pointer(queue->rps_map, map);
599 	spin_unlock(&rps_map_lock);
600 
601 	if (old_map)
602 		call_rcu(&old_map->rcu, rps_map_release);
603 
604 	free_cpumask_var(mask);
605 	return len;
606 }
607 
608 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
609 					   struct rx_queue_attribute *attr,
610 					   char *buf)
611 {
612 	struct rps_dev_flow_table *flow_table;
613 	unsigned int val = 0;
614 
615 	rcu_read_lock();
616 	flow_table = rcu_dereference(queue->rps_flow_table);
617 	if (flow_table)
618 		val = flow_table->mask + 1;
619 	rcu_read_unlock();
620 
621 	return sprintf(buf, "%u\n", val);
622 }
623 
624 static void rps_dev_flow_table_release_work(struct work_struct *work)
625 {
626 	struct rps_dev_flow_table *table = container_of(work,
627 	    struct rps_dev_flow_table, free_work);
628 
629 	vfree(table);
630 }
631 
632 static void rps_dev_flow_table_release(struct rcu_head *rcu)
633 {
634 	struct rps_dev_flow_table *table = container_of(rcu,
635 	    struct rps_dev_flow_table, rcu);
636 
637 	INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
638 	schedule_work(&table->free_work);
639 }
640 
641 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
642 				     struct rx_queue_attribute *attr,
643 				     const char *buf, size_t len)
644 {
645 	unsigned int count;
646 	char *endp;
647 	struct rps_dev_flow_table *table, *old_table;
648 	static DEFINE_SPINLOCK(rps_dev_flow_lock);
649 
650 	if (!capable(CAP_NET_ADMIN))
651 		return -EPERM;
652 
653 	count = simple_strtoul(buf, &endp, 0);
654 	if (endp == buf)
655 		return -EINVAL;
656 
657 	if (count) {
658 		int i;
659 
660 		if (count > 1<<30) {
661 			/* Enforce a limit to prevent overflow */
662 			return -EINVAL;
663 		}
664 		count = roundup_pow_of_two(count);
665 		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
666 		if (!table)
667 			return -ENOMEM;
668 
669 		table->mask = count - 1;
670 		for (i = 0; i < count; i++)
671 			table->flows[i].cpu = RPS_NO_CPU;
672 	} else
673 		table = NULL;
674 
675 	spin_lock(&rps_dev_flow_lock);
676 	old_table = queue->rps_flow_table;
677 	rcu_assign_pointer(queue->rps_flow_table, table);
678 	spin_unlock(&rps_dev_flow_lock);
679 
680 	if (old_table)
681 		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
682 
683 	return len;
684 }
685 
686 static struct rx_queue_attribute rps_cpus_attribute =
687 	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
688 
689 
690 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
691 	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
692 	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
693 
694 static struct attribute *rx_queue_default_attrs[] = {
695 	&rps_cpus_attribute.attr,
696 	&rps_dev_flow_table_cnt_attribute.attr,
697 	NULL
698 };
699 
700 static void rx_queue_release(struct kobject *kobj)
701 {
702 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
703 	struct netdev_rx_queue *first = queue->first;
704 
705 	if (queue->rps_map)
706 		call_rcu(&queue->rps_map->rcu, rps_map_release);
707 
708 	if (queue->rps_flow_table)
709 		call_rcu(&queue->rps_flow_table->rcu,
710 		    rps_dev_flow_table_release);
711 
712 	if (atomic_dec_and_test(&first->count))
713 		kfree(first);
714 }
715 
716 static struct kobj_type rx_queue_ktype = {
717 	.sysfs_ops = &rx_queue_sysfs_ops,
718 	.release = rx_queue_release,
719 	.default_attrs = rx_queue_default_attrs,
720 };
721 
722 static int rx_queue_add_kobject(struct net_device *net, int index)
723 {
724 	struct netdev_rx_queue *queue = net->_rx + index;
725 	struct kobject *kobj = &queue->kobj;
726 	int error = 0;
727 
728 	kobj->kset = net->queues_kset;
729 	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
730 	    "rx-%u", index);
731 	if (error) {
732 		kobject_put(kobj);
733 		return error;
734 	}
735 
736 	kobject_uevent(kobj, KOBJ_ADD);
737 
738 	return error;
739 }
740 
741 static int rx_queue_register_kobjects(struct net_device *net)
742 {
743 	int i;
744 	int error = 0;
745 
746 	net->queues_kset = kset_create_and_add("queues",
747 	    NULL, &net->dev.kobj);
748 	if (!net->queues_kset)
749 		return -ENOMEM;
750 	for (i = 0; i < net->num_rx_queues; i++) {
751 		error = rx_queue_add_kobject(net, i);
752 		if (error)
753 			break;
754 	}
755 
756 	if (error)
757 		while (--i >= 0)
758 			kobject_put(&net->_rx[i].kobj);
759 
760 	return error;
761 }
762 
763 static void rx_queue_remove_kobjects(struct net_device *net)
764 {
765 	int i;
766 
767 	for (i = 0; i < net->num_rx_queues; i++)
768 		kobject_put(&net->_rx[i].kobj);
769 	kset_unregister(net->queues_kset);
770 }
771 #endif /* CONFIG_RPS */
772 
773 static const void *net_current_ns(void)
774 {
775 	return current->nsproxy->net_ns;
776 }
777 
778 static const void *net_initial_ns(void)
779 {
780 	return &init_net;
781 }
782 
783 static const void *net_netlink_ns(struct sock *sk)
784 {
785 	return sock_net(sk);
786 }
787 
788 static struct kobj_ns_type_operations net_ns_type_operations = {
789 	.type = KOBJ_NS_TYPE_NET,
790 	.current_ns = net_current_ns,
791 	.netlink_ns = net_netlink_ns,
792 	.initial_ns = net_initial_ns,
793 };
794 
795 static void net_kobj_ns_exit(struct net *net)
796 {
797 	kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
798 }
799 
800 static struct pernet_operations kobj_net_ops = {
801 	.exit = net_kobj_ns_exit,
802 };
803 
804 
805 #ifdef CONFIG_HOTPLUG
806 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
807 {
808 	struct net_device *dev = to_net_dev(d);
809 	int retval;
810 
811 	/* pass interface to uevent. */
812 	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
813 	if (retval)
814 		goto exit;
815 
816 	/* pass ifindex to uevent.
817 	 * ifindex is useful as it won't change (interface name may change)
818 	 * and is what RtNetlink uses natively. */
819 	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
820 
821 exit:
822 	return retval;
823 }
824 #endif
825 
826 /*
827  *	netdev_release -- destroy and free a dead device.
828  *	Called when last reference to device kobject is gone.
829  */
830 static void netdev_release(struct device *d)
831 {
832 	struct net_device *dev = to_net_dev(d);
833 
834 	BUG_ON(dev->reg_state != NETREG_RELEASED);
835 
836 	kfree(dev->ifalias);
837 	kfree((char *)dev - dev->padded);
838 }
839 
840 static const void *net_namespace(struct device *d)
841 {
842 	struct net_device *dev;
843 	dev = container_of(d, struct net_device, dev);
844 	return dev_net(dev);
845 }
846 
847 static struct class net_class = {
848 	.name = "net",
849 	.dev_release = netdev_release,
850 #ifdef CONFIG_SYSFS
851 	.dev_attrs = net_class_attributes,
852 #endif /* CONFIG_SYSFS */
853 #ifdef CONFIG_HOTPLUG
854 	.dev_uevent = netdev_uevent,
855 #endif
856 	.ns_type = &net_ns_type_operations,
857 	.namespace = net_namespace,
858 };
859 
860 /* Delete sysfs entries but hold kobject reference until after all
861  * netdev references are gone.
862  */
863 void netdev_unregister_kobject(struct net_device * net)
864 {
865 	struct device *dev = &(net->dev);
866 
867 	kobject_get(&dev->kobj);
868 
869 #ifdef CONFIG_RPS
870 	rx_queue_remove_kobjects(net);
871 #endif
872 
873 	device_del(dev);
874 }
875 
876 /* Create sysfs entries for network device. */
877 int netdev_register_kobject(struct net_device *net)
878 {
879 	struct device *dev = &(net->dev);
880 	const struct attribute_group **groups = net->sysfs_groups;
881 	int error = 0;
882 
883 	device_initialize(dev);
884 	dev->class = &net_class;
885 	dev->platform_data = net;
886 	dev->groups = groups;
887 
888 	dev_set_name(dev, "%s", net->name);
889 
890 #ifdef CONFIG_SYSFS
891 	/* Allow for a device specific group */
892 	if (*groups)
893 		groups++;
894 
895 	*groups++ = &netstat_group;
896 #ifdef CONFIG_WIRELESS_EXT_SYSFS
897 	if (net->ieee80211_ptr)
898 		*groups++ = &wireless_group;
899 #ifdef CONFIG_WIRELESS_EXT
900 	else if (net->wireless_handlers)
901 		*groups++ = &wireless_group;
902 #endif
903 #endif
904 #endif /* CONFIG_SYSFS */
905 
906 	error = device_add(dev);
907 	if (error)
908 		return error;
909 
910 #ifdef CONFIG_RPS
911 	error = rx_queue_register_kobjects(net);
912 	if (error) {
913 		device_del(dev);
914 		return error;
915 	}
916 #endif
917 
918 	return error;
919 }
920 
921 int netdev_class_create_file(struct class_attribute *class_attr)
922 {
923 	return class_create_file(&net_class, class_attr);
924 }
925 
926 void netdev_class_remove_file(struct class_attribute *class_attr)
927 {
928 	class_remove_file(&net_class, class_attr);
929 }
930 
931 EXPORT_SYMBOL(netdev_class_create_file);
932 EXPORT_SYMBOL(netdev_class_remove_file);
933 
934 int netdev_kobject_init(void)
935 {
936 	kobj_ns_type_register(&net_ns_type_operations);
937 	register_pernet_subsys(&kobj_net_ops);
938 	return class_register(&net_class);
939 }
940