xref: /openbmc/linux/net/core/net-sysfs.c (revision 81d67439)
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_udec[] = "%u\n";
32 static const char fmt_ulong[] = "%lu\n";
33 static const char fmt_u64[] = "%llu\n";
34 
35 static inline int dev_isalive(const struct net_device *dev)
36 {
37 	return dev->reg_state <= NETREG_REGISTERED;
38 }
39 
40 /* use same locking rules as GIF* ioctl's */
41 static ssize_t netdev_show(const struct device *dev,
42 			   struct device_attribute *attr, char *buf,
43 			   ssize_t (*format)(const struct net_device *, char *))
44 {
45 	struct net_device *net = to_net_dev(dev);
46 	ssize_t ret = -EINVAL;
47 
48 	read_lock(&dev_base_lock);
49 	if (dev_isalive(net))
50 		ret = (*format)(net, buf);
51 	read_unlock(&dev_base_lock);
52 
53 	return ret;
54 }
55 
56 /* generate a show function for simple field */
57 #define NETDEVICE_SHOW(field, format_string)				\
58 static ssize_t format_##field(const struct net_device *net, char *buf)	\
59 {									\
60 	return sprintf(buf, format_string, net->field);			\
61 }									\
62 static ssize_t show_##field(struct device *dev,				\
63 			    struct device_attribute *attr, char *buf)	\
64 {									\
65 	return netdev_show(dev, attr, buf, format_##field);		\
66 }
67 
68 
69 /* use same locking and permission rules as SIF* ioctl's */
70 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
71 			    const char *buf, size_t len,
72 			    int (*set)(struct net_device *, unsigned long))
73 {
74 	struct net_device *net = to_net_dev(dev);
75 	char *endp;
76 	unsigned long new;
77 	int ret = -EINVAL;
78 
79 	if (!capable(CAP_NET_ADMIN))
80 		return -EPERM;
81 
82 	new = simple_strtoul(buf, &endp, 0);
83 	if (endp == buf)
84 		goto err;
85 
86 	if (!rtnl_trylock())
87 		return restart_syscall();
88 
89 	if (dev_isalive(net)) {
90 		if ((ret = (*set)(net, new)) == 0)
91 			ret = len;
92 	}
93 	rtnl_unlock();
94  err:
95 	return ret;
96 }
97 
98 NETDEVICE_SHOW(dev_id, fmt_hex);
99 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
100 NETDEVICE_SHOW(addr_len, fmt_dec);
101 NETDEVICE_SHOW(iflink, fmt_dec);
102 NETDEVICE_SHOW(ifindex, fmt_dec);
103 NETDEVICE_SHOW(type, fmt_dec);
104 NETDEVICE_SHOW(link_mode, fmt_dec);
105 
106 /* use same locking rules as GIFHWADDR ioctl's */
107 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
108 			    char *buf)
109 {
110 	struct net_device *net = to_net_dev(dev);
111 	ssize_t ret = -EINVAL;
112 
113 	read_lock(&dev_base_lock);
114 	if (dev_isalive(net))
115 		ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
116 	read_unlock(&dev_base_lock);
117 	return ret;
118 }
119 
120 static ssize_t show_broadcast(struct device *dev,
121 			    struct device_attribute *attr, char *buf)
122 {
123 	struct net_device *net = to_net_dev(dev);
124 	if (dev_isalive(net))
125 		return sysfs_format_mac(buf, net->broadcast, net->addr_len);
126 	return -EINVAL;
127 }
128 
129 static ssize_t show_carrier(struct device *dev,
130 			    struct device_attribute *attr, char *buf)
131 {
132 	struct net_device *netdev = to_net_dev(dev);
133 	if (netif_running(netdev)) {
134 		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
135 	}
136 	return -EINVAL;
137 }
138 
139 static ssize_t show_speed(struct device *dev,
140 			  struct device_attribute *attr, char *buf)
141 {
142 	struct net_device *netdev = to_net_dev(dev);
143 	int ret = -EINVAL;
144 
145 	if (!rtnl_trylock())
146 		return restart_syscall();
147 
148 	if (netif_running(netdev)) {
149 		struct ethtool_cmd cmd;
150 		if (!dev_ethtool_get_settings(netdev, &cmd))
151 			ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
152 	}
153 	rtnl_unlock();
154 	return ret;
155 }
156 
157 static ssize_t show_duplex(struct device *dev,
158 			   struct device_attribute *attr, char *buf)
159 {
160 	struct net_device *netdev = to_net_dev(dev);
161 	int ret = -EINVAL;
162 
163 	if (!rtnl_trylock())
164 		return restart_syscall();
165 
166 	if (netif_running(netdev)) {
167 		struct ethtool_cmd cmd;
168 		if (!dev_ethtool_get_settings(netdev, &cmd))
169 			ret = sprintf(buf, "%s\n",
170 				      cmd.duplex ? "full" : "half");
171 	}
172 	rtnl_unlock();
173 	return ret;
174 }
175 
176 static ssize_t show_dormant(struct device *dev,
177 			    struct device_attribute *attr, char *buf)
178 {
179 	struct net_device *netdev = to_net_dev(dev);
180 
181 	if (netif_running(netdev))
182 		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
183 
184 	return -EINVAL;
185 }
186 
187 static const char *const operstates[] = {
188 	"unknown",
189 	"notpresent", /* currently unused */
190 	"down",
191 	"lowerlayerdown",
192 	"testing", /* currently unused */
193 	"dormant",
194 	"up"
195 };
196 
197 static ssize_t show_operstate(struct device *dev,
198 			      struct device_attribute *attr, char *buf)
199 {
200 	const struct net_device *netdev = to_net_dev(dev);
201 	unsigned char operstate;
202 
203 	read_lock(&dev_base_lock);
204 	operstate = netdev->operstate;
205 	if (!netif_running(netdev))
206 		operstate = IF_OPER_DOWN;
207 	read_unlock(&dev_base_lock);
208 
209 	if (operstate >= ARRAY_SIZE(operstates))
210 		return -EINVAL; /* should not happen */
211 
212 	return sprintf(buf, "%s\n", operstates[operstate]);
213 }
214 
215 /* read-write attributes */
216 NETDEVICE_SHOW(mtu, fmt_dec);
217 
218 static int change_mtu(struct net_device *net, unsigned long new_mtu)
219 {
220 	return dev_set_mtu(net, (int) new_mtu);
221 }
222 
223 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
224 			 const char *buf, size_t len)
225 {
226 	return netdev_store(dev, attr, buf, len, change_mtu);
227 }
228 
229 NETDEVICE_SHOW(flags, fmt_hex);
230 
231 static int change_flags(struct net_device *net, unsigned long new_flags)
232 {
233 	return dev_change_flags(net, (unsigned) new_flags);
234 }
235 
236 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
237 			   const char *buf, size_t len)
238 {
239 	return netdev_store(dev, attr, buf, len, change_flags);
240 }
241 
242 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
243 
244 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
245 {
246 	net->tx_queue_len = new_len;
247 	return 0;
248 }
249 
250 static ssize_t store_tx_queue_len(struct device *dev,
251 				  struct device_attribute *attr,
252 				  const char *buf, size_t len)
253 {
254 	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
255 }
256 
257 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
258 			     const char *buf, size_t len)
259 {
260 	struct net_device *netdev = to_net_dev(dev);
261 	size_t count = len;
262 	ssize_t ret;
263 
264 	if (!capable(CAP_NET_ADMIN))
265 		return -EPERM;
266 
267 	/* ignore trailing newline */
268 	if (len >  0 && buf[len - 1] == '\n')
269 		--count;
270 
271 	if (!rtnl_trylock())
272 		return restart_syscall();
273 	ret = dev_set_alias(netdev, buf, count);
274 	rtnl_unlock();
275 
276 	return ret < 0 ? ret : len;
277 }
278 
279 static ssize_t show_ifalias(struct device *dev,
280 			    struct device_attribute *attr, char *buf)
281 {
282 	const struct net_device *netdev = to_net_dev(dev);
283 	ssize_t ret = 0;
284 
285 	if (!rtnl_trylock())
286 		return restart_syscall();
287 	if (netdev->ifalias)
288 		ret = sprintf(buf, "%s\n", netdev->ifalias);
289 	rtnl_unlock();
290 	return ret;
291 }
292 
293 NETDEVICE_SHOW(group, fmt_dec);
294 
295 static int change_group(struct net_device *net, unsigned long new_group)
296 {
297 	dev_set_group(net, (int) new_group);
298 	return 0;
299 }
300 
301 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
302 			 const char *buf, size_t len)
303 {
304 	return netdev_store(dev, attr, buf, len, change_group);
305 }
306 
307 static struct device_attribute net_class_attributes[] = {
308 	__ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
309 	__ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
310 	__ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
311 	__ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
312 	__ATTR(iflink, S_IRUGO, show_iflink, NULL),
313 	__ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
314 	__ATTR(type, S_IRUGO, show_type, NULL),
315 	__ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
316 	__ATTR(address, S_IRUGO, show_address, NULL),
317 	__ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
318 	__ATTR(carrier, S_IRUGO, show_carrier, NULL),
319 	__ATTR(speed, S_IRUGO, show_speed, NULL),
320 	__ATTR(duplex, S_IRUGO, show_duplex, NULL),
321 	__ATTR(dormant, S_IRUGO, show_dormant, NULL),
322 	__ATTR(operstate, S_IRUGO, show_operstate, NULL),
323 	__ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
324 	__ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
325 	__ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
326 	       store_tx_queue_len),
327 	__ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
328 	{}
329 };
330 
331 /* Show a given an attribute in the statistics group */
332 static ssize_t netstat_show(const struct device *d,
333 			    struct device_attribute *attr, char *buf,
334 			    unsigned long offset)
335 {
336 	struct net_device *dev = to_net_dev(d);
337 	ssize_t ret = -EINVAL;
338 
339 	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
340 			offset % sizeof(u64) != 0);
341 
342 	read_lock(&dev_base_lock);
343 	if (dev_isalive(dev)) {
344 		struct rtnl_link_stats64 temp;
345 		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
346 
347 		ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
348 	}
349 	read_unlock(&dev_base_lock);
350 	return ret;
351 }
352 
353 /* generate a read-only statistics attribute */
354 #define NETSTAT_ENTRY(name)						\
355 static ssize_t show_##name(struct device *d,				\
356 			   struct device_attribute *attr, char *buf) 	\
357 {									\
358 	return netstat_show(d, attr, buf,				\
359 			    offsetof(struct rtnl_link_stats64, name));	\
360 }									\
361 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
362 
363 NETSTAT_ENTRY(rx_packets);
364 NETSTAT_ENTRY(tx_packets);
365 NETSTAT_ENTRY(rx_bytes);
366 NETSTAT_ENTRY(tx_bytes);
367 NETSTAT_ENTRY(rx_errors);
368 NETSTAT_ENTRY(tx_errors);
369 NETSTAT_ENTRY(rx_dropped);
370 NETSTAT_ENTRY(tx_dropped);
371 NETSTAT_ENTRY(multicast);
372 NETSTAT_ENTRY(collisions);
373 NETSTAT_ENTRY(rx_length_errors);
374 NETSTAT_ENTRY(rx_over_errors);
375 NETSTAT_ENTRY(rx_crc_errors);
376 NETSTAT_ENTRY(rx_frame_errors);
377 NETSTAT_ENTRY(rx_fifo_errors);
378 NETSTAT_ENTRY(rx_missed_errors);
379 NETSTAT_ENTRY(tx_aborted_errors);
380 NETSTAT_ENTRY(tx_carrier_errors);
381 NETSTAT_ENTRY(tx_fifo_errors);
382 NETSTAT_ENTRY(tx_heartbeat_errors);
383 NETSTAT_ENTRY(tx_window_errors);
384 NETSTAT_ENTRY(rx_compressed);
385 NETSTAT_ENTRY(tx_compressed);
386 
387 static struct attribute *netstat_attrs[] = {
388 	&dev_attr_rx_packets.attr,
389 	&dev_attr_tx_packets.attr,
390 	&dev_attr_rx_bytes.attr,
391 	&dev_attr_tx_bytes.attr,
392 	&dev_attr_rx_errors.attr,
393 	&dev_attr_tx_errors.attr,
394 	&dev_attr_rx_dropped.attr,
395 	&dev_attr_tx_dropped.attr,
396 	&dev_attr_multicast.attr,
397 	&dev_attr_collisions.attr,
398 	&dev_attr_rx_length_errors.attr,
399 	&dev_attr_rx_over_errors.attr,
400 	&dev_attr_rx_crc_errors.attr,
401 	&dev_attr_rx_frame_errors.attr,
402 	&dev_attr_rx_fifo_errors.attr,
403 	&dev_attr_rx_missed_errors.attr,
404 	&dev_attr_tx_aborted_errors.attr,
405 	&dev_attr_tx_carrier_errors.attr,
406 	&dev_attr_tx_fifo_errors.attr,
407 	&dev_attr_tx_heartbeat_errors.attr,
408 	&dev_attr_tx_window_errors.attr,
409 	&dev_attr_rx_compressed.attr,
410 	&dev_attr_tx_compressed.attr,
411 	NULL
412 };
413 
414 
415 static struct attribute_group netstat_group = {
416 	.name  = "statistics",
417 	.attrs  = netstat_attrs,
418 };
419 
420 #ifdef CONFIG_WIRELESS_EXT_SYSFS
421 /* helper function that does all the locking etc for wireless stats */
422 static ssize_t wireless_show(struct device *d, char *buf,
423 			     ssize_t (*format)(const struct iw_statistics *,
424 					       char *))
425 {
426 	struct net_device *dev = to_net_dev(d);
427 	const struct iw_statistics *iw;
428 	ssize_t ret = -EINVAL;
429 
430 	if (!rtnl_trylock())
431 		return restart_syscall();
432 	if (dev_isalive(dev)) {
433 		iw = get_wireless_stats(dev);
434 		if (iw)
435 			ret = (*format)(iw, buf);
436 	}
437 	rtnl_unlock();
438 
439 	return ret;
440 }
441 
442 /* show function template for wireless fields */
443 #define WIRELESS_SHOW(name, field, format_string)			\
444 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
445 {									\
446 	return sprintf(buf, format_string, iw->field);			\
447 }									\
448 static ssize_t show_iw_##name(struct device *d,				\
449 			      struct device_attribute *attr, char *buf)	\
450 {									\
451 	return wireless_show(d, buf, format_iw_##name);			\
452 }									\
453 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
454 
455 WIRELESS_SHOW(status, status, fmt_hex);
456 WIRELESS_SHOW(link, qual.qual, fmt_dec);
457 WIRELESS_SHOW(level, qual.level, fmt_dec);
458 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
459 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
460 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
461 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
462 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
463 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
464 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
465 
466 static struct attribute *wireless_attrs[] = {
467 	&dev_attr_status.attr,
468 	&dev_attr_link.attr,
469 	&dev_attr_level.attr,
470 	&dev_attr_noise.attr,
471 	&dev_attr_nwid.attr,
472 	&dev_attr_crypt.attr,
473 	&dev_attr_fragment.attr,
474 	&dev_attr_retries.attr,
475 	&dev_attr_misc.attr,
476 	&dev_attr_beacon.attr,
477 	NULL
478 };
479 
480 static struct attribute_group wireless_group = {
481 	.name = "wireless",
482 	.attrs = wireless_attrs,
483 };
484 #endif
485 #endif /* CONFIG_SYSFS */
486 
487 #ifdef CONFIG_RPS
488 /*
489  * RX queue sysfs structures and functions.
490  */
491 struct rx_queue_attribute {
492 	struct attribute attr;
493 	ssize_t (*show)(struct netdev_rx_queue *queue,
494 	    struct rx_queue_attribute *attr, char *buf);
495 	ssize_t (*store)(struct netdev_rx_queue *queue,
496 	    struct rx_queue_attribute *attr, const char *buf, size_t len);
497 };
498 #define to_rx_queue_attr(_attr) container_of(_attr,		\
499     struct rx_queue_attribute, attr)
500 
501 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
502 
503 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
504 				  char *buf)
505 {
506 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
507 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
508 
509 	if (!attribute->show)
510 		return -EIO;
511 
512 	return attribute->show(queue, attribute, buf);
513 }
514 
515 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
516 				   const char *buf, size_t count)
517 {
518 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
519 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
520 
521 	if (!attribute->store)
522 		return -EIO;
523 
524 	return attribute->store(queue, attribute, buf, count);
525 }
526 
527 static const struct sysfs_ops rx_queue_sysfs_ops = {
528 	.show = rx_queue_attr_show,
529 	.store = rx_queue_attr_store,
530 };
531 
532 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
533 			    struct rx_queue_attribute *attribute, char *buf)
534 {
535 	struct rps_map *map;
536 	cpumask_var_t mask;
537 	size_t len = 0;
538 	int i;
539 
540 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
541 		return -ENOMEM;
542 
543 	rcu_read_lock();
544 	map = rcu_dereference(queue->rps_map);
545 	if (map)
546 		for (i = 0; i < map->len; i++)
547 			cpumask_set_cpu(map->cpus[i], mask);
548 
549 	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
550 	if (PAGE_SIZE - len < 3) {
551 		rcu_read_unlock();
552 		free_cpumask_var(mask);
553 		return -EINVAL;
554 	}
555 	rcu_read_unlock();
556 
557 	free_cpumask_var(mask);
558 	len += sprintf(buf + len, "\n");
559 	return len;
560 }
561 
562 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
563 		      struct rx_queue_attribute *attribute,
564 		      const char *buf, size_t len)
565 {
566 	struct rps_map *old_map, *map;
567 	cpumask_var_t mask;
568 	int err, cpu, i;
569 	static DEFINE_SPINLOCK(rps_map_lock);
570 
571 	if (!capable(CAP_NET_ADMIN))
572 		return -EPERM;
573 
574 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
575 		return -ENOMEM;
576 
577 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
578 	if (err) {
579 		free_cpumask_var(mask);
580 		return err;
581 	}
582 
583 	map = kzalloc(max_t(unsigned,
584 	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
585 	    GFP_KERNEL);
586 	if (!map) {
587 		free_cpumask_var(mask);
588 		return -ENOMEM;
589 	}
590 
591 	i = 0;
592 	for_each_cpu_and(cpu, mask, cpu_online_mask)
593 		map->cpus[i++] = cpu;
594 
595 	if (i)
596 		map->len = i;
597 	else {
598 		kfree(map);
599 		map = NULL;
600 	}
601 
602 	spin_lock(&rps_map_lock);
603 	old_map = rcu_dereference_protected(queue->rps_map,
604 					    lockdep_is_held(&rps_map_lock));
605 	rcu_assign_pointer(queue->rps_map, map);
606 	spin_unlock(&rps_map_lock);
607 
608 	if (old_map)
609 		kfree_rcu(old_map, rcu);
610 
611 	free_cpumask_var(mask);
612 	return len;
613 }
614 
615 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
616 					   struct rx_queue_attribute *attr,
617 					   char *buf)
618 {
619 	struct rps_dev_flow_table *flow_table;
620 	unsigned int val = 0;
621 
622 	rcu_read_lock();
623 	flow_table = rcu_dereference(queue->rps_flow_table);
624 	if (flow_table)
625 		val = flow_table->mask + 1;
626 	rcu_read_unlock();
627 
628 	return sprintf(buf, "%u\n", val);
629 }
630 
631 static void rps_dev_flow_table_release_work(struct work_struct *work)
632 {
633 	struct rps_dev_flow_table *table = container_of(work,
634 	    struct rps_dev_flow_table, free_work);
635 
636 	vfree(table);
637 }
638 
639 static void rps_dev_flow_table_release(struct rcu_head *rcu)
640 {
641 	struct rps_dev_flow_table *table = container_of(rcu,
642 	    struct rps_dev_flow_table, rcu);
643 
644 	INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
645 	schedule_work(&table->free_work);
646 }
647 
648 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
649 				     struct rx_queue_attribute *attr,
650 				     const char *buf, size_t len)
651 {
652 	unsigned int count;
653 	char *endp;
654 	struct rps_dev_flow_table *table, *old_table;
655 	static DEFINE_SPINLOCK(rps_dev_flow_lock);
656 
657 	if (!capable(CAP_NET_ADMIN))
658 		return -EPERM;
659 
660 	count = simple_strtoul(buf, &endp, 0);
661 	if (endp == buf)
662 		return -EINVAL;
663 
664 	if (count) {
665 		int i;
666 
667 		if (count > 1<<30) {
668 			/* Enforce a limit to prevent overflow */
669 			return -EINVAL;
670 		}
671 		count = roundup_pow_of_two(count);
672 		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
673 		if (!table)
674 			return -ENOMEM;
675 
676 		table->mask = count - 1;
677 		for (i = 0; i < count; i++)
678 			table->flows[i].cpu = RPS_NO_CPU;
679 	} else
680 		table = NULL;
681 
682 	spin_lock(&rps_dev_flow_lock);
683 	old_table = rcu_dereference_protected(queue->rps_flow_table,
684 					      lockdep_is_held(&rps_dev_flow_lock));
685 	rcu_assign_pointer(queue->rps_flow_table, table);
686 	spin_unlock(&rps_dev_flow_lock);
687 
688 	if (old_table)
689 		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
690 
691 	return len;
692 }
693 
694 static struct rx_queue_attribute rps_cpus_attribute =
695 	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
696 
697 
698 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
699 	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
700 	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
701 
702 static struct attribute *rx_queue_default_attrs[] = {
703 	&rps_cpus_attribute.attr,
704 	&rps_dev_flow_table_cnt_attribute.attr,
705 	NULL
706 };
707 
708 static void rx_queue_release(struct kobject *kobj)
709 {
710 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
711 	struct rps_map *map;
712 	struct rps_dev_flow_table *flow_table;
713 
714 
715 	map = rcu_dereference_raw(queue->rps_map);
716 	if (map) {
717 		RCU_INIT_POINTER(queue->rps_map, NULL);
718 		kfree_rcu(map, rcu);
719 	}
720 
721 	flow_table = rcu_dereference_raw(queue->rps_flow_table);
722 	if (flow_table) {
723 		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
724 		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
725 	}
726 
727 	memset(kobj, 0, sizeof(*kobj));
728 	dev_put(queue->dev);
729 }
730 
731 static struct kobj_type rx_queue_ktype = {
732 	.sysfs_ops = &rx_queue_sysfs_ops,
733 	.release = rx_queue_release,
734 	.default_attrs = rx_queue_default_attrs,
735 };
736 
737 static int rx_queue_add_kobject(struct net_device *net, int index)
738 {
739 	struct netdev_rx_queue *queue = net->_rx + index;
740 	struct kobject *kobj = &queue->kobj;
741 	int error = 0;
742 
743 	kobj->kset = net->queues_kset;
744 	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
745 	    "rx-%u", index);
746 	if (error) {
747 		kobject_put(kobj);
748 		return error;
749 	}
750 
751 	kobject_uevent(kobj, KOBJ_ADD);
752 	dev_hold(queue->dev);
753 
754 	return error;
755 }
756 #endif /* CONFIG_RPS */
757 
758 int
759 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
760 {
761 #ifdef CONFIG_RPS
762 	int i;
763 	int error = 0;
764 
765 	for (i = old_num; i < new_num; i++) {
766 		error = rx_queue_add_kobject(net, i);
767 		if (error) {
768 			new_num = old_num;
769 			break;
770 		}
771 	}
772 
773 	while (--i >= new_num)
774 		kobject_put(&net->_rx[i].kobj);
775 
776 	return error;
777 #else
778 	return 0;
779 #endif
780 }
781 
782 #ifdef CONFIG_XPS
783 /*
784  * netdev_queue sysfs structures and functions.
785  */
786 struct netdev_queue_attribute {
787 	struct attribute attr;
788 	ssize_t (*show)(struct netdev_queue *queue,
789 	    struct netdev_queue_attribute *attr, char *buf);
790 	ssize_t (*store)(struct netdev_queue *queue,
791 	    struct netdev_queue_attribute *attr, const char *buf, size_t len);
792 };
793 #define to_netdev_queue_attr(_attr) container_of(_attr,		\
794     struct netdev_queue_attribute, attr)
795 
796 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
797 
798 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
799 				      struct attribute *attr, char *buf)
800 {
801 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
802 	struct netdev_queue *queue = to_netdev_queue(kobj);
803 
804 	if (!attribute->show)
805 		return -EIO;
806 
807 	return attribute->show(queue, attribute, buf);
808 }
809 
810 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
811 				       struct attribute *attr,
812 				       const char *buf, size_t count)
813 {
814 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
815 	struct netdev_queue *queue = to_netdev_queue(kobj);
816 
817 	if (!attribute->store)
818 		return -EIO;
819 
820 	return attribute->store(queue, attribute, buf, count);
821 }
822 
823 static const struct sysfs_ops netdev_queue_sysfs_ops = {
824 	.show = netdev_queue_attr_show,
825 	.store = netdev_queue_attr_store,
826 };
827 
828 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
829 {
830 	struct net_device *dev = queue->dev;
831 	int i;
832 
833 	for (i = 0; i < dev->num_tx_queues; i++)
834 		if (queue == &dev->_tx[i])
835 			break;
836 
837 	BUG_ON(i >= dev->num_tx_queues);
838 
839 	return i;
840 }
841 
842 
843 static ssize_t show_xps_map(struct netdev_queue *queue,
844 			    struct netdev_queue_attribute *attribute, char *buf)
845 {
846 	struct net_device *dev = queue->dev;
847 	struct xps_dev_maps *dev_maps;
848 	cpumask_var_t mask;
849 	unsigned long index;
850 	size_t len = 0;
851 	int i;
852 
853 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
854 		return -ENOMEM;
855 
856 	index = get_netdev_queue_index(queue);
857 
858 	rcu_read_lock();
859 	dev_maps = rcu_dereference(dev->xps_maps);
860 	if (dev_maps) {
861 		for_each_possible_cpu(i) {
862 			struct xps_map *map =
863 			    rcu_dereference(dev_maps->cpu_map[i]);
864 			if (map) {
865 				int j;
866 				for (j = 0; j < map->len; j++) {
867 					if (map->queues[j] == index) {
868 						cpumask_set_cpu(i, mask);
869 						break;
870 					}
871 				}
872 			}
873 		}
874 	}
875 	rcu_read_unlock();
876 
877 	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
878 	if (PAGE_SIZE - len < 3) {
879 		free_cpumask_var(mask);
880 		return -EINVAL;
881 	}
882 
883 	free_cpumask_var(mask);
884 	len += sprintf(buf + len, "\n");
885 	return len;
886 }
887 
888 static DEFINE_MUTEX(xps_map_mutex);
889 #define xmap_dereference(P)		\
890 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
891 
892 static ssize_t store_xps_map(struct netdev_queue *queue,
893 		      struct netdev_queue_attribute *attribute,
894 		      const char *buf, size_t len)
895 {
896 	struct net_device *dev = queue->dev;
897 	cpumask_var_t mask;
898 	int err, i, cpu, pos, map_len, alloc_len, need_set;
899 	unsigned long index;
900 	struct xps_map *map, *new_map;
901 	struct xps_dev_maps *dev_maps, *new_dev_maps;
902 	int nonempty = 0;
903 	int numa_node = -2;
904 
905 	if (!capable(CAP_NET_ADMIN))
906 		return -EPERM;
907 
908 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
909 		return -ENOMEM;
910 
911 	index = get_netdev_queue_index(queue);
912 
913 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
914 	if (err) {
915 		free_cpumask_var(mask);
916 		return err;
917 	}
918 
919 	new_dev_maps = kzalloc(max_t(unsigned,
920 	    XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
921 	if (!new_dev_maps) {
922 		free_cpumask_var(mask);
923 		return -ENOMEM;
924 	}
925 
926 	mutex_lock(&xps_map_mutex);
927 
928 	dev_maps = xmap_dereference(dev->xps_maps);
929 
930 	for_each_possible_cpu(cpu) {
931 		map = dev_maps ?
932 			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
933 		new_map = map;
934 		if (map) {
935 			for (pos = 0; pos < map->len; pos++)
936 				if (map->queues[pos] == index)
937 					break;
938 			map_len = map->len;
939 			alloc_len = map->alloc_len;
940 		} else
941 			pos = map_len = alloc_len = 0;
942 
943 		need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
944 #ifdef CONFIG_NUMA
945 		if (need_set) {
946 			if (numa_node == -2)
947 				numa_node = cpu_to_node(cpu);
948 			else if (numa_node != cpu_to_node(cpu))
949 				numa_node = -1;
950 		}
951 #endif
952 		if (need_set && pos >= map_len) {
953 			/* Need to add queue to this CPU's map */
954 			if (map_len >= alloc_len) {
955 				alloc_len = alloc_len ?
956 				    2 * alloc_len : XPS_MIN_MAP_ALLOC;
957 				new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
958 						       GFP_KERNEL,
959 						       cpu_to_node(cpu));
960 				if (!new_map)
961 					goto error;
962 				new_map->alloc_len = alloc_len;
963 				for (i = 0; i < map_len; i++)
964 					new_map->queues[i] = map->queues[i];
965 				new_map->len = map_len;
966 			}
967 			new_map->queues[new_map->len++] = index;
968 		} else if (!need_set && pos < map_len) {
969 			/* Need to remove queue from this CPU's map */
970 			if (map_len > 1)
971 				new_map->queues[pos] =
972 				    new_map->queues[--new_map->len];
973 			else
974 				new_map = NULL;
975 		}
976 		RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
977 	}
978 
979 	/* Cleanup old maps */
980 	for_each_possible_cpu(cpu) {
981 		map = dev_maps ?
982 			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
983 		if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
984 			kfree_rcu(map, rcu);
985 		if (new_dev_maps->cpu_map[cpu])
986 			nonempty = 1;
987 	}
988 
989 	if (nonempty)
990 		rcu_assign_pointer(dev->xps_maps, new_dev_maps);
991 	else {
992 		kfree(new_dev_maps);
993 		rcu_assign_pointer(dev->xps_maps, NULL);
994 	}
995 
996 	if (dev_maps)
997 		kfree_rcu(dev_maps, rcu);
998 
999 	netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1000 					    NUMA_NO_NODE);
1001 
1002 	mutex_unlock(&xps_map_mutex);
1003 
1004 	free_cpumask_var(mask);
1005 	return len;
1006 
1007 error:
1008 	mutex_unlock(&xps_map_mutex);
1009 
1010 	if (new_dev_maps)
1011 		for_each_possible_cpu(i)
1012 			kfree(rcu_dereference_protected(
1013 				new_dev_maps->cpu_map[i],
1014 				1));
1015 	kfree(new_dev_maps);
1016 	free_cpumask_var(mask);
1017 	return -ENOMEM;
1018 }
1019 
1020 static struct netdev_queue_attribute xps_cpus_attribute =
1021     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1022 
1023 static struct attribute *netdev_queue_default_attrs[] = {
1024 	&xps_cpus_attribute.attr,
1025 	NULL
1026 };
1027 
1028 static void netdev_queue_release(struct kobject *kobj)
1029 {
1030 	struct netdev_queue *queue = to_netdev_queue(kobj);
1031 	struct net_device *dev = queue->dev;
1032 	struct xps_dev_maps *dev_maps;
1033 	struct xps_map *map;
1034 	unsigned long index;
1035 	int i, pos, nonempty = 0;
1036 
1037 	index = get_netdev_queue_index(queue);
1038 
1039 	mutex_lock(&xps_map_mutex);
1040 	dev_maps = xmap_dereference(dev->xps_maps);
1041 
1042 	if (dev_maps) {
1043 		for_each_possible_cpu(i) {
1044 			map = xmap_dereference(dev_maps->cpu_map[i]);
1045 			if (!map)
1046 				continue;
1047 
1048 			for (pos = 0; pos < map->len; pos++)
1049 				if (map->queues[pos] == index)
1050 					break;
1051 
1052 			if (pos < map->len) {
1053 				if (map->len > 1)
1054 					map->queues[pos] =
1055 					    map->queues[--map->len];
1056 				else {
1057 					RCU_INIT_POINTER(dev_maps->cpu_map[i],
1058 					    NULL);
1059 					kfree_rcu(map, rcu);
1060 					map = NULL;
1061 				}
1062 			}
1063 			if (map)
1064 				nonempty = 1;
1065 		}
1066 
1067 		if (!nonempty) {
1068 			RCU_INIT_POINTER(dev->xps_maps, NULL);
1069 			kfree_rcu(dev_maps, rcu);
1070 		}
1071 	}
1072 
1073 	mutex_unlock(&xps_map_mutex);
1074 
1075 	memset(kobj, 0, sizeof(*kobj));
1076 	dev_put(queue->dev);
1077 }
1078 
1079 static struct kobj_type netdev_queue_ktype = {
1080 	.sysfs_ops = &netdev_queue_sysfs_ops,
1081 	.release = netdev_queue_release,
1082 	.default_attrs = netdev_queue_default_attrs,
1083 };
1084 
1085 static int netdev_queue_add_kobject(struct net_device *net, int index)
1086 {
1087 	struct netdev_queue *queue = net->_tx + index;
1088 	struct kobject *kobj = &queue->kobj;
1089 	int error = 0;
1090 
1091 	kobj->kset = net->queues_kset;
1092 	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1093 	    "tx-%u", index);
1094 	if (error) {
1095 		kobject_put(kobj);
1096 		return error;
1097 	}
1098 
1099 	kobject_uevent(kobj, KOBJ_ADD);
1100 	dev_hold(queue->dev);
1101 
1102 	return error;
1103 }
1104 #endif /* CONFIG_XPS */
1105 
1106 int
1107 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1108 {
1109 #ifdef CONFIG_XPS
1110 	int i;
1111 	int error = 0;
1112 
1113 	for (i = old_num; i < new_num; i++) {
1114 		error = netdev_queue_add_kobject(net, i);
1115 		if (error) {
1116 			new_num = old_num;
1117 			break;
1118 		}
1119 	}
1120 
1121 	while (--i >= new_num)
1122 		kobject_put(&net->_tx[i].kobj);
1123 
1124 	return error;
1125 #else
1126 	return 0;
1127 #endif
1128 }
1129 
1130 static int register_queue_kobjects(struct net_device *net)
1131 {
1132 	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1133 
1134 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1135 	net->queues_kset = kset_create_and_add("queues",
1136 	    NULL, &net->dev.kobj);
1137 	if (!net->queues_kset)
1138 		return -ENOMEM;
1139 #endif
1140 
1141 #ifdef CONFIG_RPS
1142 	real_rx = net->real_num_rx_queues;
1143 #endif
1144 	real_tx = net->real_num_tx_queues;
1145 
1146 	error = net_rx_queue_update_kobjects(net, 0, real_rx);
1147 	if (error)
1148 		goto error;
1149 	rxq = real_rx;
1150 
1151 	error = netdev_queue_update_kobjects(net, 0, real_tx);
1152 	if (error)
1153 		goto error;
1154 	txq = real_tx;
1155 
1156 	return 0;
1157 
1158 error:
1159 	netdev_queue_update_kobjects(net, txq, 0);
1160 	net_rx_queue_update_kobjects(net, rxq, 0);
1161 	return error;
1162 }
1163 
1164 static void remove_queue_kobjects(struct net_device *net)
1165 {
1166 	int real_rx = 0, real_tx = 0;
1167 
1168 #ifdef CONFIG_RPS
1169 	real_rx = net->real_num_rx_queues;
1170 #endif
1171 	real_tx = net->real_num_tx_queues;
1172 
1173 	net_rx_queue_update_kobjects(net, real_rx, 0);
1174 	netdev_queue_update_kobjects(net, real_tx, 0);
1175 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1176 	kset_unregister(net->queues_kset);
1177 #endif
1178 }
1179 
1180 static void *net_grab_current_ns(void)
1181 {
1182 	struct net *ns = current->nsproxy->net_ns;
1183 #ifdef CONFIG_NET_NS
1184 	if (ns)
1185 		atomic_inc(&ns->passive);
1186 #endif
1187 	return ns;
1188 }
1189 
1190 static const void *net_initial_ns(void)
1191 {
1192 	return &init_net;
1193 }
1194 
1195 static const void *net_netlink_ns(struct sock *sk)
1196 {
1197 	return sock_net(sk);
1198 }
1199 
1200 struct kobj_ns_type_operations net_ns_type_operations = {
1201 	.type = KOBJ_NS_TYPE_NET,
1202 	.grab_current_ns = net_grab_current_ns,
1203 	.netlink_ns = net_netlink_ns,
1204 	.initial_ns = net_initial_ns,
1205 	.drop_ns = net_drop_ns,
1206 };
1207 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1208 
1209 #ifdef CONFIG_HOTPLUG
1210 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1211 {
1212 	struct net_device *dev = to_net_dev(d);
1213 	int retval;
1214 
1215 	/* pass interface to uevent. */
1216 	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1217 	if (retval)
1218 		goto exit;
1219 
1220 	/* pass ifindex to uevent.
1221 	 * ifindex is useful as it won't change (interface name may change)
1222 	 * and is what RtNetlink uses natively. */
1223 	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1224 
1225 exit:
1226 	return retval;
1227 }
1228 #endif
1229 
1230 /*
1231  *	netdev_release -- destroy and free a dead device.
1232  *	Called when last reference to device kobject is gone.
1233  */
1234 static void netdev_release(struct device *d)
1235 {
1236 	struct net_device *dev = to_net_dev(d);
1237 
1238 	BUG_ON(dev->reg_state != NETREG_RELEASED);
1239 
1240 	kfree(dev->ifalias);
1241 	kfree((char *)dev - dev->padded);
1242 }
1243 
1244 static const void *net_namespace(struct device *d)
1245 {
1246 	struct net_device *dev;
1247 	dev = container_of(d, struct net_device, dev);
1248 	return dev_net(dev);
1249 }
1250 
1251 static struct class net_class = {
1252 	.name = "net",
1253 	.dev_release = netdev_release,
1254 #ifdef CONFIG_SYSFS
1255 	.dev_attrs = net_class_attributes,
1256 #endif /* CONFIG_SYSFS */
1257 #ifdef CONFIG_HOTPLUG
1258 	.dev_uevent = netdev_uevent,
1259 #endif
1260 	.ns_type = &net_ns_type_operations,
1261 	.namespace = net_namespace,
1262 };
1263 
1264 /* Delete sysfs entries but hold kobject reference until after all
1265  * netdev references are gone.
1266  */
1267 void netdev_unregister_kobject(struct net_device * net)
1268 {
1269 	struct device *dev = &(net->dev);
1270 
1271 	kobject_get(&dev->kobj);
1272 
1273 	remove_queue_kobjects(net);
1274 
1275 	device_del(dev);
1276 }
1277 
1278 /* Create sysfs entries for network device. */
1279 int netdev_register_kobject(struct net_device *net)
1280 {
1281 	struct device *dev = &(net->dev);
1282 	const struct attribute_group **groups = net->sysfs_groups;
1283 	int error = 0;
1284 
1285 	device_initialize(dev);
1286 	dev->class = &net_class;
1287 	dev->platform_data = net;
1288 	dev->groups = groups;
1289 
1290 	dev_set_name(dev, "%s", net->name);
1291 
1292 #ifdef CONFIG_SYSFS
1293 	/* Allow for a device specific group */
1294 	if (*groups)
1295 		groups++;
1296 
1297 	*groups++ = &netstat_group;
1298 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1299 	if (net->ieee80211_ptr)
1300 		*groups++ = &wireless_group;
1301 #ifdef CONFIG_WIRELESS_EXT
1302 	else if (net->wireless_handlers)
1303 		*groups++ = &wireless_group;
1304 #endif
1305 #endif
1306 #endif /* CONFIG_SYSFS */
1307 
1308 	error = device_add(dev);
1309 	if (error)
1310 		return error;
1311 
1312 	error = register_queue_kobjects(net);
1313 	if (error) {
1314 		device_del(dev);
1315 		return error;
1316 	}
1317 
1318 	return error;
1319 }
1320 
1321 int netdev_class_create_file(struct class_attribute *class_attr)
1322 {
1323 	return class_create_file(&net_class, class_attr);
1324 }
1325 EXPORT_SYMBOL(netdev_class_create_file);
1326 
1327 void netdev_class_remove_file(struct class_attribute *class_attr)
1328 {
1329 	class_remove_file(&net_class, class_attr);
1330 }
1331 EXPORT_SYMBOL(netdev_class_remove_file);
1332 
1333 int netdev_kobject_init(void)
1334 {
1335 	kobj_ns_type_register(&net_ns_type_operations);
1336 	return class_register(&net_class);
1337 }
1338