xref: /openbmc/linux/net/core/net-sysfs.c (revision eb3fcf007fffe5830d815e713591f3e858f2a365)
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 <net/switchdev.h>
16 #include <linux/if_arp.h>
17 #include <linux/slab.h>
18 #include <linux/nsproxy.h>
19 #include <net/sock.h>
20 #include <net/net_namespace.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/vmalloc.h>
23 #include <linux/export.h>
24 #include <linux/jiffies.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/of.h>
27 
28 #include "net-sysfs.h"
29 
30 #ifdef CONFIG_SYSFS
31 static const char fmt_hex[] = "%#x\n";
32 static const char fmt_long_hex[] = "%#lx\n";
33 static const char fmt_dec[] = "%d\n";
34 static const char fmt_udec[] = "%u\n";
35 static const char fmt_ulong[] = "%lu\n";
36 static const char fmt_u64[] = "%llu\n";
37 
38 static inline int dev_isalive(const struct net_device *dev)
39 {
40 	return dev->reg_state <= NETREG_REGISTERED;
41 }
42 
43 /* use same locking rules as GIF* ioctl's */
44 static ssize_t netdev_show(const struct device *dev,
45 			   struct device_attribute *attr, char *buf,
46 			   ssize_t (*format)(const struct net_device *, char *))
47 {
48 	struct net_device *ndev = to_net_dev(dev);
49 	ssize_t ret = -EINVAL;
50 
51 	read_lock(&dev_base_lock);
52 	if (dev_isalive(ndev))
53 		ret = (*format)(ndev, buf);
54 	read_unlock(&dev_base_lock);
55 
56 	return ret;
57 }
58 
59 /* generate a show function for simple field */
60 #define NETDEVICE_SHOW(field, format_string)				\
61 static ssize_t format_##field(const struct net_device *dev, char *buf)	\
62 {									\
63 	return sprintf(buf, format_string, dev->field);			\
64 }									\
65 static ssize_t field##_show(struct device *dev,				\
66 			    struct device_attribute *attr, char *buf)	\
67 {									\
68 	return netdev_show(dev, attr, buf, format_##field);		\
69 }									\
70 
71 #define NETDEVICE_SHOW_RO(field, format_string)				\
72 NETDEVICE_SHOW(field, format_string);					\
73 static DEVICE_ATTR_RO(field)
74 
75 #define NETDEVICE_SHOW_RW(field, format_string)				\
76 NETDEVICE_SHOW(field, format_string);					\
77 static DEVICE_ATTR_RW(field)
78 
79 /* use same locking and permission rules as SIF* ioctl's */
80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81 			    const char *buf, size_t len,
82 			    int (*set)(struct net_device *, unsigned long))
83 {
84 	struct net_device *netdev = to_net_dev(dev);
85 	struct net *net = dev_net(netdev);
86 	unsigned long new;
87 	int ret = -EINVAL;
88 
89 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90 		return -EPERM;
91 
92 	ret = kstrtoul(buf, 0, &new);
93 	if (ret)
94 		goto err;
95 
96 	if (!rtnl_trylock())
97 		return restart_syscall();
98 
99 	if (dev_isalive(netdev)) {
100 		if ((ret = (*set)(netdev, new)) == 0)
101 			ret = len;
102 	}
103 	rtnl_unlock();
104  err:
105 	return ret;
106 }
107 
108 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
109 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
110 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
111 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
112 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
113 NETDEVICE_SHOW_RO(type, fmt_dec);
114 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
115 
116 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
117 			   char *buf)
118 {
119 	struct net_device *ndev = to_net_dev(dev);
120 
121 	return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
122 }
123 static DEVICE_ATTR_RO(iflink);
124 
125 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
126 {
127 	return sprintf(buf, fmt_dec, dev->name_assign_type);
128 }
129 
130 static ssize_t name_assign_type_show(struct device *dev,
131 				     struct device_attribute *attr,
132 				     char *buf)
133 {
134 	struct net_device *ndev = to_net_dev(dev);
135 	ssize_t ret = -EINVAL;
136 
137 	if (ndev->name_assign_type != NET_NAME_UNKNOWN)
138 		ret = netdev_show(dev, attr, buf, format_name_assign_type);
139 
140 	return ret;
141 }
142 static DEVICE_ATTR_RO(name_assign_type);
143 
144 /* use same locking rules as GIFHWADDR ioctl's */
145 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
146 			    char *buf)
147 {
148 	struct net_device *ndev = to_net_dev(dev);
149 	ssize_t ret = -EINVAL;
150 
151 	read_lock(&dev_base_lock);
152 	if (dev_isalive(ndev))
153 		ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
154 	read_unlock(&dev_base_lock);
155 	return ret;
156 }
157 static DEVICE_ATTR_RO(address);
158 
159 static ssize_t broadcast_show(struct device *dev,
160 			      struct device_attribute *attr, char *buf)
161 {
162 	struct net_device *ndev = to_net_dev(dev);
163 	if (dev_isalive(ndev))
164 		return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
165 	return -EINVAL;
166 }
167 static DEVICE_ATTR_RO(broadcast);
168 
169 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
170 {
171 	if (!netif_running(dev))
172 		return -EINVAL;
173 	return dev_change_carrier(dev, (bool) new_carrier);
174 }
175 
176 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
177 			     const char *buf, size_t len)
178 {
179 	return netdev_store(dev, attr, buf, len, change_carrier);
180 }
181 
182 static ssize_t carrier_show(struct device *dev,
183 			    struct device_attribute *attr, char *buf)
184 {
185 	struct net_device *netdev = to_net_dev(dev);
186 	if (netif_running(netdev)) {
187 		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
188 	}
189 	return -EINVAL;
190 }
191 static DEVICE_ATTR_RW(carrier);
192 
193 static ssize_t speed_show(struct device *dev,
194 			  struct device_attribute *attr, char *buf)
195 {
196 	struct net_device *netdev = to_net_dev(dev);
197 	int ret = -EINVAL;
198 
199 	if (!rtnl_trylock())
200 		return restart_syscall();
201 
202 	if (netif_running(netdev)) {
203 		struct ethtool_cmd cmd;
204 		if (!__ethtool_get_settings(netdev, &cmd))
205 			ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
206 	}
207 	rtnl_unlock();
208 	return ret;
209 }
210 static DEVICE_ATTR_RO(speed);
211 
212 static ssize_t duplex_show(struct device *dev,
213 			   struct device_attribute *attr, char *buf)
214 {
215 	struct net_device *netdev = to_net_dev(dev);
216 	int ret = -EINVAL;
217 
218 	if (!rtnl_trylock())
219 		return restart_syscall();
220 
221 	if (netif_running(netdev)) {
222 		struct ethtool_cmd cmd;
223 		if (!__ethtool_get_settings(netdev, &cmd)) {
224 			const char *duplex;
225 			switch (cmd.duplex) {
226 			case DUPLEX_HALF:
227 				duplex = "half";
228 				break;
229 			case DUPLEX_FULL:
230 				duplex = "full";
231 				break;
232 			default:
233 				duplex = "unknown";
234 				break;
235 			}
236 			ret = sprintf(buf, "%s\n", duplex);
237 		}
238 	}
239 	rtnl_unlock();
240 	return ret;
241 }
242 static DEVICE_ATTR_RO(duplex);
243 
244 static ssize_t dormant_show(struct device *dev,
245 			    struct device_attribute *attr, char *buf)
246 {
247 	struct net_device *netdev = to_net_dev(dev);
248 
249 	if (netif_running(netdev))
250 		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
251 
252 	return -EINVAL;
253 }
254 static DEVICE_ATTR_RO(dormant);
255 
256 static const char *const operstates[] = {
257 	"unknown",
258 	"notpresent", /* currently unused */
259 	"down",
260 	"lowerlayerdown",
261 	"testing", /* currently unused */
262 	"dormant",
263 	"up"
264 };
265 
266 static ssize_t operstate_show(struct device *dev,
267 			      struct device_attribute *attr, char *buf)
268 {
269 	const struct net_device *netdev = to_net_dev(dev);
270 	unsigned char operstate;
271 
272 	read_lock(&dev_base_lock);
273 	operstate = netdev->operstate;
274 	if (!netif_running(netdev))
275 		operstate = IF_OPER_DOWN;
276 	read_unlock(&dev_base_lock);
277 
278 	if (operstate >= ARRAY_SIZE(operstates))
279 		return -EINVAL; /* should not happen */
280 
281 	return sprintf(buf, "%s\n", operstates[operstate]);
282 }
283 static DEVICE_ATTR_RO(operstate);
284 
285 static ssize_t carrier_changes_show(struct device *dev,
286 				    struct device_attribute *attr,
287 				    char *buf)
288 {
289 	struct net_device *netdev = to_net_dev(dev);
290 	return sprintf(buf, fmt_dec,
291 		       atomic_read(&netdev->carrier_changes));
292 }
293 static DEVICE_ATTR_RO(carrier_changes);
294 
295 /* read-write attributes */
296 
297 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
298 {
299 	return dev_set_mtu(dev, (int) new_mtu);
300 }
301 
302 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
303 			 const char *buf, size_t len)
304 {
305 	return netdev_store(dev, attr, buf, len, change_mtu);
306 }
307 NETDEVICE_SHOW_RW(mtu, fmt_dec);
308 
309 static int change_flags(struct net_device *dev, unsigned long new_flags)
310 {
311 	return dev_change_flags(dev, (unsigned int) new_flags);
312 }
313 
314 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
315 			   const char *buf, size_t len)
316 {
317 	return netdev_store(dev, attr, buf, len, change_flags);
318 }
319 NETDEVICE_SHOW_RW(flags, fmt_hex);
320 
321 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
322 {
323 	dev->tx_queue_len = new_len;
324 	return 0;
325 }
326 
327 static ssize_t tx_queue_len_store(struct device *dev,
328 				  struct device_attribute *attr,
329 				  const char *buf, size_t len)
330 {
331 	if (!capable(CAP_NET_ADMIN))
332 		return -EPERM;
333 
334 	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
335 }
336 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
337 
338 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
339 {
340 	dev->gro_flush_timeout = val;
341 	return 0;
342 }
343 
344 static ssize_t gro_flush_timeout_store(struct device *dev,
345 				  struct device_attribute *attr,
346 				  const char *buf, size_t len)
347 {
348 	if (!capable(CAP_NET_ADMIN))
349 		return -EPERM;
350 
351 	return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
352 }
353 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
354 
355 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
356 			     const char *buf, size_t len)
357 {
358 	struct net_device *netdev = to_net_dev(dev);
359 	struct net *net = dev_net(netdev);
360 	size_t count = len;
361 	ssize_t ret;
362 
363 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
364 		return -EPERM;
365 
366 	/* ignore trailing newline */
367 	if (len >  0 && buf[len - 1] == '\n')
368 		--count;
369 
370 	if (!rtnl_trylock())
371 		return restart_syscall();
372 	ret = dev_set_alias(netdev, buf, count);
373 	rtnl_unlock();
374 
375 	return ret < 0 ? ret : len;
376 }
377 
378 static ssize_t ifalias_show(struct device *dev,
379 			    struct device_attribute *attr, char *buf)
380 {
381 	const struct net_device *netdev = to_net_dev(dev);
382 	ssize_t ret = 0;
383 
384 	if (!rtnl_trylock())
385 		return restart_syscall();
386 	if (netdev->ifalias)
387 		ret = sprintf(buf, "%s\n", netdev->ifalias);
388 	rtnl_unlock();
389 	return ret;
390 }
391 static DEVICE_ATTR_RW(ifalias);
392 
393 static int change_group(struct net_device *dev, unsigned long new_group)
394 {
395 	dev_set_group(dev, (int) new_group);
396 	return 0;
397 }
398 
399 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
400 			   const char *buf, size_t len)
401 {
402 	return netdev_store(dev, attr, buf, len, change_group);
403 }
404 NETDEVICE_SHOW(group, fmt_dec);
405 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
406 
407 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
408 {
409 	return dev_change_proto_down(dev, (bool) proto_down);
410 }
411 
412 static ssize_t proto_down_store(struct device *dev,
413 				struct device_attribute *attr,
414 				const char *buf, size_t len)
415 {
416 	return netdev_store(dev, attr, buf, len, change_proto_down);
417 }
418 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
419 
420 static ssize_t phys_port_id_show(struct device *dev,
421 				 struct device_attribute *attr, char *buf)
422 {
423 	struct net_device *netdev = to_net_dev(dev);
424 	ssize_t ret = -EINVAL;
425 
426 	if (!rtnl_trylock())
427 		return restart_syscall();
428 
429 	if (dev_isalive(netdev)) {
430 		struct netdev_phys_item_id ppid;
431 
432 		ret = dev_get_phys_port_id(netdev, &ppid);
433 		if (!ret)
434 			ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
435 	}
436 	rtnl_unlock();
437 
438 	return ret;
439 }
440 static DEVICE_ATTR_RO(phys_port_id);
441 
442 static ssize_t phys_port_name_show(struct device *dev,
443 				   struct device_attribute *attr, char *buf)
444 {
445 	struct net_device *netdev = to_net_dev(dev);
446 	ssize_t ret = -EINVAL;
447 
448 	if (!rtnl_trylock())
449 		return restart_syscall();
450 
451 	if (dev_isalive(netdev)) {
452 		char name[IFNAMSIZ];
453 
454 		ret = dev_get_phys_port_name(netdev, name, sizeof(name));
455 		if (!ret)
456 			ret = sprintf(buf, "%s\n", name);
457 	}
458 	rtnl_unlock();
459 
460 	return ret;
461 }
462 static DEVICE_ATTR_RO(phys_port_name);
463 
464 static ssize_t phys_switch_id_show(struct device *dev,
465 				   struct device_attribute *attr, char *buf)
466 {
467 	struct net_device *netdev = to_net_dev(dev);
468 	ssize_t ret = -EINVAL;
469 
470 	if (!rtnl_trylock())
471 		return restart_syscall();
472 
473 	if (dev_isalive(netdev)) {
474 		struct switchdev_attr attr = {
475 			.id = SWITCHDEV_ATTR_PORT_PARENT_ID,
476 			.flags = SWITCHDEV_F_NO_RECURSE,
477 		};
478 
479 		ret = switchdev_port_attr_get(netdev, &attr);
480 		if (!ret)
481 			ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
482 				      attr.u.ppid.id);
483 	}
484 	rtnl_unlock();
485 
486 	return ret;
487 }
488 static DEVICE_ATTR_RO(phys_switch_id);
489 
490 static struct attribute *net_class_attrs[] = {
491 	&dev_attr_netdev_group.attr,
492 	&dev_attr_type.attr,
493 	&dev_attr_dev_id.attr,
494 	&dev_attr_dev_port.attr,
495 	&dev_attr_iflink.attr,
496 	&dev_attr_ifindex.attr,
497 	&dev_attr_name_assign_type.attr,
498 	&dev_attr_addr_assign_type.attr,
499 	&dev_attr_addr_len.attr,
500 	&dev_attr_link_mode.attr,
501 	&dev_attr_address.attr,
502 	&dev_attr_broadcast.attr,
503 	&dev_attr_speed.attr,
504 	&dev_attr_duplex.attr,
505 	&dev_attr_dormant.attr,
506 	&dev_attr_operstate.attr,
507 	&dev_attr_carrier_changes.attr,
508 	&dev_attr_ifalias.attr,
509 	&dev_attr_carrier.attr,
510 	&dev_attr_mtu.attr,
511 	&dev_attr_flags.attr,
512 	&dev_attr_tx_queue_len.attr,
513 	&dev_attr_gro_flush_timeout.attr,
514 	&dev_attr_phys_port_id.attr,
515 	&dev_attr_phys_port_name.attr,
516 	&dev_attr_phys_switch_id.attr,
517 	&dev_attr_proto_down.attr,
518 	NULL,
519 };
520 ATTRIBUTE_GROUPS(net_class);
521 
522 /* Show a given an attribute in the statistics group */
523 static ssize_t netstat_show(const struct device *d,
524 			    struct device_attribute *attr, char *buf,
525 			    unsigned long offset)
526 {
527 	struct net_device *dev = to_net_dev(d);
528 	ssize_t ret = -EINVAL;
529 
530 	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
531 			offset % sizeof(u64) != 0);
532 
533 	read_lock(&dev_base_lock);
534 	if (dev_isalive(dev)) {
535 		struct rtnl_link_stats64 temp;
536 		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
537 
538 		ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
539 	}
540 	read_unlock(&dev_base_lock);
541 	return ret;
542 }
543 
544 /* generate a read-only statistics attribute */
545 #define NETSTAT_ENTRY(name)						\
546 static ssize_t name##_show(struct device *d,				\
547 			   struct device_attribute *attr, char *buf) 	\
548 {									\
549 	return netstat_show(d, attr, buf,				\
550 			    offsetof(struct rtnl_link_stats64, name));	\
551 }									\
552 static DEVICE_ATTR_RO(name)
553 
554 NETSTAT_ENTRY(rx_packets);
555 NETSTAT_ENTRY(tx_packets);
556 NETSTAT_ENTRY(rx_bytes);
557 NETSTAT_ENTRY(tx_bytes);
558 NETSTAT_ENTRY(rx_errors);
559 NETSTAT_ENTRY(tx_errors);
560 NETSTAT_ENTRY(rx_dropped);
561 NETSTAT_ENTRY(tx_dropped);
562 NETSTAT_ENTRY(multicast);
563 NETSTAT_ENTRY(collisions);
564 NETSTAT_ENTRY(rx_length_errors);
565 NETSTAT_ENTRY(rx_over_errors);
566 NETSTAT_ENTRY(rx_crc_errors);
567 NETSTAT_ENTRY(rx_frame_errors);
568 NETSTAT_ENTRY(rx_fifo_errors);
569 NETSTAT_ENTRY(rx_missed_errors);
570 NETSTAT_ENTRY(tx_aborted_errors);
571 NETSTAT_ENTRY(tx_carrier_errors);
572 NETSTAT_ENTRY(tx_fifo_errors);
573 NETSTAT_ENTRY(tx_heartbeat_errors);
574 NETSTAT_ENTRY(tx_window_errors);
575 NETSTAT_ENTRY(rx_compressed);
576 NETSTAT_ENTRY(tx_compressed);
577 
578 static struct attribute *netstat_attrs[] = {
579 	&dev_attr_rx_packets.attr,
580 	&dev_attr_tx_packets.attr,
581 	&dev_attr_rx_bytes.attr,
582 	&dev_attr_tx_bytes.attr,
583 	&dev_attr_rx_errors.attr,
584 	&dev_attr_tx_errors.attr,
585 	&dev_attr_rx_dropped.attr,
586 	&dev_attr_tx_dropped.attr,
587 	&dev_attr_multicast.attr,
588 	&dev_attr_collisions.attr,
589 	&dev_attr_rx_length_errors.attr,
590 	&dev_attr_rx_over_errors.attr,
591 	&dev_attr_rx_crc_errors.attr,
592 	&dev_attr_rx_frame_errors.attr,
593 	&dev_attr_rx_fifo_errors.attr,
594 	&dev_attr_rx_missed_errors.attr,
595 	&dev_attr_tx_aborted_errors.attr,
596 	&dev_attr_tx_carrier_errors.attr,
597 	&dev_attr_tx_fifo_errors.attr,
598 	&dev_attr_tx_heartbeat_errors.attr,
599 	&dev_attr_tx_window_errors.attr,
600 	&dev_attr_rx_compressed.attr,
601 	&dev_attr_tx_compressed.attr,
602 	NULL
603 };
604 
605 
606 static struct attribute_group netstat_group = {
607 	.name  = "statistics",
608 	.attrs  = netstat_attrs,
609 };
610 
611 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
612 static struct attribute *wireless_attrs[] = {
613 	NULL
614 };
615 
616 static struct attribute_group wireless_group = {
617 	.name = "wireless",
618 	.attrs = wireless_attrs,
619 };
620 #endif
621 
622 #else /* CONFIG_SYSFS */
623 #define net_class_groups	NULL
624 #endif /* CONFIG_SYSFS */
625 
626 #ifdef CONFIG_SYSFS
627 #define to_rx_queue_attr(_attr) container_of(_attr,		\
628     struct rx_queue_attribute, attr)
629 
630 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
631 
632 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
633 				  char *buf)
634 {
635 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
636 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
637 
638 	if (!attribute->show)
639 		return -EIO;
640 
641 	return attribute->show(queue, attribute, buf);
642 }
643 
644 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
645 				   const char *buf, size_t count)
646 {
647 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
648 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
649 
650 	if (!attribute->store)
651 		return -EIO;
652 
653 	return attribute->store(queue, attribute, buf, count);
654 }
655 
656 static const struct sysfs_ops rx_queue_sysfs_ops = {
657 	.show = rx_queue_attr_show,
658 	.store = rx_queue_attr_store,
659 };
660 
661 #ifdef CONFIG_RPS
662 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
663 			    struct rx_queue_attribute *attribute, char *buf)
664 {
665 	struct rps_map *map;
666 	cpumask_var_t mask;
667 	int i, len;
668 
669 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
670 		return -ENOMEM;
671 
672 	rcu_read_lock();
673 	map = rcu_dereference(queue->rps_map);
674 	if (map)
675 		for (i = 0; i < map->len; i++)
676 			cpumask_set_cpu(map->cpus[i], mask);
677 
678 	len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
679 	rcu_read_unlock();
680 	free_cpumask_var(mask);
681 
682 	return len < PAGE_SIZE ? len : -EINVAL;
683 }
684 
685 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
686 		      struct rx_queue_attribute *attribute,
687 		      const char *buf, size_t len)
688 {
689 	struct rps_map *old_map, *map;
690 	cpumask_var_t mask;
691 	int err, cpu, i;
692 	static DEFINE_MUTEX(rps_map_mutex);
693 
694 	if (!capable(CAP_NET_ADMIN))
695 		return -EPERM;
696 
697 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
698 		return -ENOMEM;
699 
700 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
701 	if (err) {
702 		free_cpumask_var(mask);
703 		return err;
704 	}
705 
706 	map = kzalloc(max_t(unsigned int,
707 	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
708 	    GFP_KERNEL);
709 	if (!map) {
710 		free_cpumask_var(mask);
711 		return -ENOMEM;
712 	}
713 
714 	i = 0;
715 	for_each_cpu_and(cpu, mask, cpu_online_mask)
716 		map->cpus[i++] = cpu;
717 
718 	if (i)
719 		map->len = i;
720 	else {
721 		kfree(map);
722 		map = NULL;
723 	}
724 
725 	mutex_lock(&rps_map_mutex);
726 	old_map = rcu_dereference_protected(queue->rps_map,
727 					    mutex_is_locked(&rps_map_mutex));
728 	rcu_assign_pointer(queue->rps_map, map);
729 
730 	if (map)
731 		static_key_slow_inc(&rps_needed);
732 	if (old_map)
733 		static_key_slow_dec(&rps_needed);
734 
735 	mutex_unlock(&rps_map_mutex);
736 
737 	if (old_map)
738 		kfree_rcu(old_map, rcu);
739 
740 	free_cpumask_var(mask);
741 	return len;
742 }
743 
744 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
745 					   struct rx_queue_attribute *attr,
746 					   char *buf)
747 {
748 	struct rps_dev_flow_table *flow_table;
749 	unsigned long val = 0;
750 
751 	rcu_read_lock();
752 	flow_table = rcu_dereference(queue->rps_flow_table);
753 	if (flow_table)
754 		val = (unsigned long)flow_table->mask + 1;
755 	rcu_read_unlock();
756 
757 	return sprintf(buf, "%lu\n", val);
758 }
759 
760 static void rps_dev_flow_table_release(struct rcu_head *rcu)
761 {
762 	struct rps_dev_flow_table *table = container_of(rcu,
763 	    struct rps_dev_flow_table, rcu);
764 	vfree(table);
765 }
766 
767 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
768 				     struct rx_queue_attribute *attr,
769 				     const char *buf, size_t len)
770 {
771 	unsigned long mask, count;
772 	struct rps_dev_flow_table *table, *old_table;
773 	static DEFINE_SPINLOCK(rps_dev_flow_lock);
774 	int rc;
775 
776 	if (!capable(CAP_NET_ADMIN))
777 		return -EPERM;
778 
779 	rc = kstrtoul(buf, 0, &count);
780 	if (rc < 0)
781 		return rc;
782 
783 	if (count) {
784 		mask = count - 1;
785 		/* mask = roundup_pow_of_two(count) - 1;
786 		 * without overflows...
787 		 */
788 		while ((mask | (mask >> 1)) != mask)
789 			mask |= (mask >> 1);
790 		/* On 64 bit arches, must check mask fits in table->mask (u32),
791 		 * and on 32bit arches, must check
792 		 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
793 		 */
794 #if BITS_PER_LONG > 32
795 		if (mask > (unsigned long)(u32)mask)
796 			return -EINVAL;
797 #else
798 		if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
799 				/ sizeof(struct rps_dev_flow)) {
800 			/* Enforce a limit to prevent overflow */
801 			return -EINVAL;
802 		}
803 #endif
804 		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
805 		if (!table)
806 			return -ENOMEM;
807 
808 		table->mask = mask;
809 		for (count = 0; count <= mask; count++)
810 			table->flows[count].cpu = RPS_NO_CPU;
811 	} else
812 		table = NULL;
813 
814 	spin_lock(&rps_dev_flow_lock);
815 	old_table = rcu_dereference_protected(queue->rps_flow_table,
816 					      lockdep_is_held(&rps_dev_flow_lock));
817 	rcu_assign_pointer(queue->rps_flow_table, table);
818 	spin_unlock(&rps_dev_flow_lock);
819 
820 	if (old_table)
821 		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
822 
823 	return len;
824 }
825 
826 static struct rx_queue_attribute rps_cpus_attribute =
827 	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
828 
829 
830 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
831 	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
832 	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
833 #endif /* CONFIG_RPS */
834 
835 static struct attribute *rx_queue_default_attrs[] = {
836 #ifdef CONFIG_RPS
837 	&rps_cpus_attribute.attr,
838 	&rps_dev_flow_table_cnt_attribute.attr,
839 #endif
840 	NULL
841 };
842 
843 static void rx_queue_release(struct kobject *kobj)
844 {
845 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
846 #ifdef CONFIG_RPS
847 	struct rps_map *map;
848 	struct rps_dev_flow_table *flow_table;
849 
850 
851 	map = rcu_dereference_protected(queue->rps_map, 1);
852 	if (map) {
853 		RCU_INIT_POINTER(queue->rps_map, NULL);
854 		kfree_rcu(map, rcu);
855 	}
856 
857 	flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
858 	if (flow_table) {
859 		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
860 		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
861 	}
862 #endif
863 
864 	memset(kobj, 0, sizeof(*kobj));
865 	dev_put(queue->dev);
866 }
867 
868 static const void *rx_queue_namespace(struct kobject *kobj)
869 {
870 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
871 	struct device *dev = &queue->dev->dev;
872 	const void *ns = NULL;
873 
874 	if (dev->class && dev->class->ns_type)
875 		ns = dev->class->namespace(dev);
876 
877 	return ns;
878 }
879 
880 static struct kobj_type rx_queue_ktype = {
881 	.sysfs_ops = &rx_queue_sysfs_ops,
882 	.release = rx_queue_release,
883 	.default_attrs = rx_queue_default_attrs,
884 	.namespace = rx_queue_namespace
885 };
886 
887 static int rx_queue_add_kobject(struct net_device *dev, int index)
888 {
889 	struct netdev_rx_queue *queue = dev->_rx + index;
890 	struct kobject *kobj = &queue->kobj;
891 	int error = 0;
892 
893 	kobj->kset = dev->queues_kset;
894 	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
895 	    "rx-%u", index);
896 	if (error)
897 		goto exit;
898 
899 	if (dev->sysfs_rx_queue_group) {
900 		error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
901 		if (error)
902 			goto exit;
903 	}
904 
905 	kobject_uevent(kobj, KOBJ_ADD);
906 	dev_hold(queue->dev);
907 
908 	return error;
909 exit:
910 	kobject_put(kobj);
911 	return error;
912 }
913 #endif /* CONFIG_SYSFS */
914 
915 int
916 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
917 {
918 #ifdef CONFIG_SYSFS
919 	int i;
920 	int error = 0;
921 
922 #ifndef CONFIG_RPS
923 	if (!dev->sysfs_rx_queue_group)
924 		return 0;
925 #endif
926 	for (i = old_num; i < new_num; i++) {
927 		error = rx_queue_add_kobject(dev, i);
928 		if (error) {
929 			new_num = old_num;
930 			break;
931 		}
932 	}
933 
934 	while (--i >= new_num) {
935 		if (dev->sysfs_rx_queue_group)
936 			sysfs_remove_group(&dev->_rx[i].kobj,
937 					   dev->sysfs_rx_queue_group);
938 		kobject_put(&dev->_rx[i].kobj);
939 	}
940 
941 	return error;
942 #else
943 	return 0;
944 #endif
945 }
946 
947 #ifdef CONFIG_SYSFS
948 /*
949  * netdev_queue sysfs structures and functions.
950  */
951 struct netdev_queue_attribute {
952 	struct attribute attr;
953 	ssize_t (*show)(struct netdev_queue *queue,
954 	    struct netdev_queue_attribute *attr, char *buf);
955 	ssize_t (*store)(struct netdev_queue *queue,
956 	    struct netdev_queue_attribute *attr, const char *buf, size_t len);
957 };
958 #define to_netdev_queue_attr(_attr) container_of(_attr,		\
959     struct netdev_queue_attribute, attr)
960 
961 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
962 
963 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
964 				      struct attribute *attr, char *buf)
965 {
966 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
967 	struct netdev_queue *queue = to_netdev_queue(kobj);
968 
969 	if (!attribute->show)
970 		return -EIO;
971 
972 	return attribute->show(queue, attribute, buf);
973 }
974 
975 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
976 				       struct attribute *attr,
977 				       const char *buf, size_t count)
978 {
979 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
980 	struct netdev_queue *queue = to_netdev_queue(kobj);
981 
982 	if (!attribute->store)
983 		return -EIO;
984 
985 	return attribute->store(queue, attribute, buf, count);
986 }
987 
988 static const struct sysfs_ops netdev_queue_sysfs_ops = {
989 	.show = netdev_queue_attr_show,
990 	.store = netdev_queue_attr_store,
991 };
992 
993 static ssize_t show_trans_timeout(struct netdev_queue *queue,
994 				  struct netdev_queue_attribute *attribute,
995 				  char *buf)
996 {
997 	unsigned long trans_timeout;
998 
999 	spin_lock_irq(&queue->_xmit_lock);
1000 	trans_timeout = queue->trans_timeout;
1001 	spin_unlock_irq(&queue->_xmit_lock);
1002 
1003 	return sprintf(buf, "%lu", trans_timeout);
1004 }
1005 
1006 #ifdef CONFIG_XPS
1007 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1008 {
1009 	struct net_device *dev = queue->dev;
1010 	int i;
1011 
1012 	for (i = 0; i < dev->num_tx_queues; i++)
1013 		if (queue == &dev->_tx[i])
1014 			break;
1015 
1016 	BUG_ON(i >= dev->num_tx_queues);
1017 
1018 	return i;
1019 }
1020 
1021 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1022 			       struct netdev_queue_attribute *attribute,
1023 			       char *buf)
1024 {
1025 	return sprintf(buf, "%lu\n", queue->tx_maxrate);
1026 }
1027 
1028 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1029 			      struct netdev_queue_attribute *attribute,
1030 			      const char *buf, size_t len)
1031 {
1032 	struct net_device *dev = queue->dev;
1033 	int err, index = get_netdev_queue_index(queue);
1034 	u32 rate = 0;
1035 
1036 	err = kstrtou32(buf, 10, &rate);
1037 	if (err < 0)
1038 		return err;
1039 
1040 	if (!rtnl_trylock())
1041 		return restart_syscall();
1042 
1043 	err = -EOPNOTSUPP;
1044 	if (dev->netdev_ops->ndo_set_tx_maxrate)
1045 		err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1046 
1047 	rtnl_unlock();
1048 	if (!err) {
1049 		queue->tx_maxrate = rate;
1050 		return len;
1051 	}
1052 	return err;
1053 }
1054 
1055 static struct netdev_queue_attribute queue_tx_maxrate =
1056 	__ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1057 	       show_tx_maxrate, set_tx_maxrate);
1058 #endif
1059 
1060 static struct netdev_queue_attribute queue_trans_timeout =
1061 	__ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1062 
1063 #ifdef CONFIG_BQL
1064 /*
1065  * Byte queue limits sysfs structures and functions.
1066  */
1067 static ssize_t bql_show(char *buf, unsigned int value)
1068 {
1069 	return sprintf(buf, "%u\n", value);
1070 }
1071 
1072 static ssize_t bql_set(const char *buf, const size_t count,
1073 		       unsigned int *pvalue)
1074 {
1075 	unsigned int value;
1076 	int err;
1077 
1078 	if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1079 		value = DQL_MAX_LIMIT;
1080 	else {
1081 		err = kstrtouint(buf, 10, &value);
1082 		if (err < 0)
1083 			return err;
1084 		if (value > DQL_MAX_LIMIT)
1085 			return -EINVAL;
1086 	}
1087 
1088 	*pvalue = value;
1089 
1090 	return count;
1091 }
1092 
1093 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1094 				  struct netdev_queue_attribute *attr,
1095 				  char *buf)
1096 {
1097 	struct dql *dql = &queue->dql;
1098 
1099 	return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1100 }
1101 
1102 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1103 				 struct netdev_queue_attribute *attribute,
1104 				 const char *buf, size_t len)
1105 {
1106 	struct dql *dql = &queue->dql;
1107 	unsigned int value;
1108 	int err;
1109 
1110 	err = kstrtouint(buf, 10, &value);
1111 	if (err < 0)
1112 		return err;
1113 
1114 	dql->slack_hold_time = msecs_to_jiffies(value);
1115 
1116 	return len;
1117 }
1118 
1119 static struct netdev_queue_attribute bql_hold_time_attribute =
1120 	__ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1121 	    bql_set_hold_time);
1122 
1123 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1124 				 struct netdev_queue_attribute *attr,
1125 				 char *buf)
1126 {
1127 	struct dql *dql = &queue->dql;
1128 
1129 	return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1130 }
1131 
1132 static struct netdev_queue_attribute bql_inflight_attribute =
1133 	__ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1134 
1135 #define BQL_ATTR(NAME, FIELD)						\
1136 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,		\
1137 				 struct netdev_queue_attribute *attr,	\
1138 				 char *buf)				\
1139 {									\
1140 	return bql_show(buf, queue->dql.FIELD);				\
1141 }									\
1142 									\
1143 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,		\
1144 				struct netdev_queue_attribute *attr,	\
1145 				const char *buf, size_t len)		\
1146 {									\
1147 	return bql_set(buf, len, &queue->dql.FIELD);			\
1148 }									\
1149 									\
1150 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =	\
1151 	__ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,		\
1152 	    bql_set_ ## NAME);
1153 
1154 BQL_ATTR(limit, limit)
1155 BQL_ATTR(limit_max, max_limit)
1156 BQL_ATTR(limit_min, min_limit)
1157 
1158 static struct attribute *dql_attrs[] = {
1159 	&bql_limit_attribute.attr,
1160 	&bql_limit_max_attribute.attr,
1161 	&bql_limit_min_attribute.attr,
1162 	&bql_hold_time_attribute.attr,
1163 	&bql_inflight_attribute.attr,
1164 	NULL
1165 };
1166 
1167 static struct attribute_group dql_group = {
1168 	.name  = "byte_queue_limits",
1169 	.attrs  = dql_attrs,
1170 };
1171 #endif /* CONFIG_BQL */
1172 
1173 #ifdef CONFIG_XPS
1174 static ssize_t show_xps_map(struct netdev_queue *queue,
1175 			    struct netdev_queue_attribute *attribute, char *buf)
1176 {
1177 	struct net_device *dev = queue->dev;
1178 	struct xps_dev_maps *dev_maps;
1179 	cpumask_var_t mask;
1180 	unsigned long index;
1181 	int i, len;
1182 
1183 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1184 		return -ENOMEM;
1185 
1186 	index = get_netdev_queue_index(queue);
1187 
1188 	rcu_read_lock();
1189 	dev_maps = rcu_dereference(dev->xps_maps);
1190 	if (dev_maps) {
1191 		for_each_possible_cpu(i) {
1192 			struct xps_map *map =
1193 			    rcu_dereference(dev_maps->cpu_map[i]);
1194 			if (map) {
1195 				int j;
1196 				for (j = 0; j < map->len; j++) {
1197 					if (map->queues[j] == index) {
1198 						cpumask_set_cpu(i, mask);
1199 						break;
1200 					}
1201 				}
1202 			}
1203 		}
1204 	}
1205 	rcu_read_unlock();
1206 
1207 	len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1208 	free_cpumask_var(mask);
1209 	return len < PAGE_SIZE ? len : -EINVAL;
1210 }
1211 
1212 static ssize_t store_xps_map(struct netdev_queue *queue,
1213 		      struct netdev_queue_attribute *attribute,
1214 		      const char *buf, size_t len)
1215 {
1216 	struct net_device *dev = queue->dev;
1217 	unsigned long index;
1218 	cpumask_var_t mask;
1219 	int err;
1220 
1221 	if (!capable(CAP_NET_ADMIN))
1222 		return -EPERM;
1223 
1224 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1225 		return -ENOMEM;
1226 
1227 	index = get_netdev_queue_index(queue);
1228 
1229 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1230 	if (err) {
1231 		free_cpumask_var(mask);
1232 		return err;
1233 	}
1234 
1235 	err = netif_set_xps_queue(dev, mask, index);
1236 
1237 	free_cpumask_var(mask);
1238 
1239 	return err ? : len;
1240 }
1241 
1242 static struct netdev_queue_attribute xps_cpus_attribute =
1243     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1244 #endif /* CONFIG_XPS */
1245 
1246 static struct attribute *netdev_queue_default_attrs[] = {
1247 	&queue_trans_timeout.attr,
1248 #ifdef CONFIG_XPS
1249 	&xps_cpus_attribute.attr,
1250 	&queue_tx_maxrate.attr,
1251 #endif
1252 	NULL
1253 };
1254 
1255 static void netdev_queue_release(struct kobject *kobj)
1256 {
1257 	struct netdev_queue *queue = to_netdev_queue(kobj);
1258 
1259 	memset(kobj, 0, sizeof(*kobj));
1260 	dev_put(queue->dev);
1261 }
1262 
1263 static const void *netdev_queue_namespace(struct kobject *kobj)
1264 {
1265 	struct netdev_queue *queue = to_netdev_queue(kobj);
1266 	struct device *dev = &queue->dev->dev;
1267 	const void *ns = NULL;
1268 
1269 	if (dev->class && dev->class->ns_type)
1270 		ns = dev->class->namespace(dev);
1271 
1272 	return ns;
1273 }
1274 
1275 static struct kobj_type netdev_queue_ktype = {
1276 	.sysfs_ops = &netdev_queue_sysfs_ops,
1277 	.release = netdev_queue_release,
1278 	.default_attrs = netdev_queue_default_attrs,
1279 	.namespace = netdev_queue_namespace,
1280 };
1281 
1282 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1283 {
1284 	struct netdev_queue *queue = dev->_tx + index;
1285 	struct kobject *kobj = &queue->kobj;
1286 	int error = 0;
1287 
1288 	kobj->kset = dev->queues_kset;
1289 	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1290 	    "tx-%u", index);
1291 	if (error)
1292 		goto exit;
1293 
1294 #ifdef CONFIG_BQL
1295 	error = sysfs_create_group(kobj, &dql_group);
1296 	if (error)
1297 		goto exit;
1298 #endif
1299 
1300 	kobject_uevent(kobj, KOBJ_ADD);
1301 	dev_hold(queue->dev);
1302 
1303 	return 0;
1304 exit:
1305 	kobject_put(kobj);
1306 	return error;
1307 }
1308 #endif /* CONFIG_SYSFS */
1309 
1310 int
1311 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1312 {
1313 #ifdef CONFIG_SYSFS
1314 	int i;
1315 	int error = 0;
1316 
1317 	for (i = old_num; i < new_num; i++) {
1318 		error = netdev_queue_add_kobject(dev, i);
1319 		if (error) {
1320 			new_num = old_num;
1321 			break;
1322 		}
1323 	}
1324 
1325 	while (--i >= new_num) {
1326 		struct netdev_queue *queue = dev->_tx + i;
1327 
1328 #ifdef CONFIG_BQL
1329 		sysfs_remove_group(&queue->kobj, &dql_group);
1330 #endif
1331 		kobject_put(&queue->kobj);
1332 	}
1333 
1334 	return error;
1335 #else
1336 	return 0;
1337 #endif /* CONFIG_SYSFS */
1338 }
1339 
1340 static int register_queue_kobjects(struct net_device *dev)
1341 {
1342 	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1343 
1344 #ifdef CONFIG_SYSFS
1345 	dev->queues_kset = kset_create_and_add("queues",
1346 	    NULL, &dev->dev.kobj);
1347 	if (!dev->queues_kset)
1348 		return -ENOMEM;
1349 	real_rx = dev->real_num_rx_queues;
1350 #endif
1351 	real_tx = dev->real_num_tx_queues;
1352 
1353 	error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1354 	if (error)
1355 		goto error;
1356 	rxq = real_rx;
1357 
1358 	error = netdev_queue_update_kobjects(dev, 0, real_tx);
1359 	if (error)
1360 		goto error;
1361 	txq = real_tx;
1362 
1363 	return 0;
1364 
1365 error:
1366 	netdev_queue_update_kobjects(dev, txq, 0);
1367 	net_rx_queue_update_kobjects(dev, rxq, 0);
1368 	return error;
1369 }
1370 
1371 static void remove_queue_kobjects(struct net_device *dev)
1372 {
1373 	int real_rx = 0, real_tx = 0;
1374 
1375 #ifdef CONFIG_SYSFS
1376 	real_rx = dev->real_num_rx_queues;
1377 #endif
1378 	real_tx = dev->real_num_tx_queues;
1379 
1380 	net_rx_queue_update_kobjects(dev, real_rx, 0);
1381 	netdev_queue_update_kobjects(dev, real_tx, 0);
1382 #ifdef CONFIG_SYSFS
1383 	kset_unregister(dev->queues_kset);
1384 #endif
1385 }
1386 
1387 static bool net_current_may_mount(void)
1388 {
1389 	struct net *net = current->nsproxy->net_ns;
1390 
1391 	return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1392 }
1393 
1394 static void *net_grab_current_ns(void)
1395 {
1396 	struct net *ns = current->nsproxy->net_ns;
1397 #ifdef CONFIG_NET_NS
1398 	if (ns)
1399 		atomic_inc(&ns->passive);
1400 #endif
1401 	return ns;
1402 }
1403 
1404 static const void *net_initial_ns(void)
1405 {
1406 	return &init_net;
1407 }
1408 
1409 static const void *net_netlink_ns(struct sock *sk)
1410 {
1411 	return sock_net(sk);
1412 }
1413 
1414 struct kobj_ns_type_operations net_ns_type_operations = {
1415 	.type = KOBJ_NS_TYPE_NET,
1416 	.current_may_mount = net_current_may_mount,
1417 	.grab_current_ns = net_grab_current_ns,
1418 	.netlink_ns = net_netlink_ns,
1419 	.initial_ns = net_initial_ns,
1420 	.drop_ns = net_drop_ns,
1421 };
1422 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1423 
1424 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1425 {
1426 	struct net_device *dev = to_net_dev(d);
1427 	int retval;
1428 
1429 	/* pass interface to uevent. */
1430 	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1431 	if (retval)
1432 		goto exit;
1433 
1434 	/* pass ifindex to uevent.
1435 	 * ifindex is useful as it won't change (interface name may change)
1436 	 * and is what RtNetlink uses natively. */
1437 	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1438 
1439 exit:
1440 	return retval;
1441 }
1442 
1443 /*
1444  *	netdev_release -- destroy and free a dead device.
1445  *	Called when last reference to device kobject is gone.
1446  */
1447 static void netdev_release(struct device *d)
1448 {
1449 	struct net_device *dev = to_net_dev(d);
1450 
1451 	BUG_ON(dev->reg_state != NETREG_RELEASED);
1452 
1453 	kfree(dev->ifalias);
1454 	netdev_freemem(dev);
1455 }
1456 
1457 static const void *net_namespace(struct device *d)
1458 {
1459 	struct net_device *dev;
1460 	dev = container_of(d, struct net_device, dev);
1461 	return dev_net(dev);
1462 }
1463 
1464 static struct class net_class = {
1465 	.name = "net",
1466 	.dev_release = netdev_release,
1467 	.dev_groups = net_class_groups,
1468 	.dev_uevent = netdev_uevent,
1469 	.ns_type = &net_ns_type_operations,
1470 	.namespace = net_namespace,
1471 };
1472 
1473 #ifdef CONFIG_OF_NET
1474 static int of_dev_node_match(struct device *dev, const void *data)
1475 {
1476 	int ret = 0;
1477 
1478 	if (dev->parent)
1479 		ret = dev->parent->of_node == data;
1480 
1481 	return ret == 0 ? dev->of_node == data : ret;
1482 }
1483 
1484 /*
1485  * of_find_net_device_by_node - lookup the net device for the device node
1486  * @np: OF device node
1487  *
1488  * Looks up the net_device structure corresponding with the device node.
1489  * If successful, returns a pointer to the net_device with the embedded
1490  * struct device refcount incremented by one, or NULL on failure. The
1491  * refcount must be dropped when done with the net_device.
1492  */
1493 struct net_device *of_find_net_device_by_node(struct device_node *np)
1494 {
1495 	struct device *dev;
1496 
1497 	dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1498 	if (!dev)
1499 		return NULL;
1500 
1501 	return to_net_dev(dev);
1502 }
1503 EXPORT_SYMBOL(of_find_net_device_by_node);
1504 #endif
1505 
1506 /* Delete sysfs entries but hold kobject reference until after all
1507  * netdev references are gone.
1508  */
1509 void netdev_unregister_kobject(struct net_device *ndev)
1510 {
1511 	struct device *dev = &(ndev->dev);
1512 
1513 	kobject_get(&dev->kobj);
1514 
1515 	remove_queue_kobjects(ndev);
1516 
1517 	pm_runtime_set_memalloc_noio(dev, false);
1518 
1519 	device_del(dev);
1520 }
1521 
1522 /* Create sysfs entries for network device. */
1523 int netdev_register_kobject(struct net_device *ndev)
1524 {
1525 	struct device *dev = &(ndev->dev);
1526 	const struct attribute_group **groups = ndev->sysfs_groups;
1527 	int error = 0;
1528 
1529 	device_initialize(dev);
1530 	dev->class = &net_class;
1531 	dev->platform_data = ndev;
1532 	dev->groups = groups;
1533 
1534 	dev_set_name(dev, "%s", ndev->name);
1535 
1536 #ifdef CONFIG_SYSFS
1537 	/* Allow for a device specific group */
1538 	if (*groups)
1539 		groups++;
1540 
1541 	*groups++ = &netstat_group;
1542 
1543 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1544 	if (ndev->ieee80211_ptr)
1545 		*groups++ = &wireless_group;
1546 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1547 	else if (ndev->wireless_handlers)
1548 		*groups++ = &wireless_group;
1549 #endif
1550 #endif
1551 #endif /* CONFIG_SYSFS */
1552 
1553 	error = device_add(dev);
1554 	if (error)
1555 		return error;
1556 
1557 	error = register_queue_kobjects(ndev);
1558 	if (error) {
1559 		device_del(dev);
1560 		return error;
1561 	}
1562 
1563 	pm_runtime_set_memalloc_noio(dev, true);
1564 
1565 	return error;
1566 }
1567 
1568 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1569 				const void *ns)
1570 {
1571 	return class_create_file_ns(&net_class, class_attr, ns);
1572 }
1573 EXPORT_SYMBOL(netdev_class_create_file_ns);
1574 
1575 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1576 				 const void *ns)
1577 {
1578 	class_remove_file_ns(&net_class, class_attr, ns);
1579 }
1580 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1581 
1582 int __init netdev_kobject_init(void)
1583 {
1584 	kobj_ns_type_register(&net_ns_type_operations);
1585 	return class_register(&net_class);
1586 }
1587