xref: /openbmc/linux/net/core/net-sysfs.c (revision 5bd8e16d)
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/vmalloc.h>
22 #include <linux/export.h>
23 #include <linux/jiffies.h>
24 #include <linux/pm_runtime.h>
25 
26 #include "net-sysfs.h"
27 
28 #ifdef CONFIG_SYSFS
29 static const char fmt_hex[] = "%#x\n";
30 static const char fmt_long_hex[] = "%#lx\n";
31 static const char fmt_dec[] = "%d\n";
32 static const char fmt_udec[] = "%u\n";
33 static const char fmt_ulong[] = "%lu\n";
34 static const char fmt_u64[] = "%llu\n";
35 
36 static inline int dev_isalive(const struct net_device *dev)
37 {
38 	return dev->reg_state <= NETREG_REGISTERED;
39 }
40 
41 /* use same locking rules as GIF* ioctl's */
42 static ssize_t netdev_show(const struct device *dev,
43 			   struct device_attribute *attr, char *buf,
44 			   ssize_t (*format)(const struct net_device *, char *))
45 {
46 	struct net_device *net = to_net_dev(dev);
47 	ssize_t ret = -EINVAL;
48 
49 	read_lock(&dev_base_lock);
50 	if (dev_isalive(net))
51 		ret = (*format)(net, buf);
52 	read_unlock(&dev_base_lock);
53 
54 	return ret;
55 }
56 
57 /* generate a show function for simple field */
58 #define NETDEVICE_SHOW(field, format_string)				\
59 static ssize_t format_##field(const struct net_device *net, char *buf)	\
60 {									\
61 	return sprintf(buf, format_string, net->field);			\
62 }									\
63 static ssize_t field##_show(struct device *dev,				\
64 			    struct device_attribute *attr, char *buf)	\
65 {									\
66 	return netdev_show(dev, attr, buf, format_##field);		\
67 }									\
68 
69 #define NETDEVICE_SHOW_RO(field, format_string)				\
70 NETDEVICE_SHOW(field, format_string);					\
71 static DEVICE_ATTR_RO(field)
72 
73 #define NETDEVICE_SHOW_RW(field, format_string)				\
74 NETDEVICE_SHOW(field, format_string);					\
75 static DEVICE_ATTR_RW(field)
76 
77 /* use same locking and permission rules as SIF* ioctl's */
78 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
79 			    const char *buf, size_t len,
80 			    int (*set)(struct net_device *, unsigned long))
81 {
82 	struct net_device *netdev = to_net_dev(dev);
83 	struct net *net = dev_net(netdev);
84 	unsigned long new;
85 	int ret = -EINVAL;
86 
87 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
88 		return -EPERM;
89 
90 	ret = kstrtoul(buf, 0, &new);
91 	if (ret)
92 		goto err;
93 
94 	if (!rtnl_trylock())
95 		return restart_syscall();
96 
97 	if (dev_isalive(netdev)) {
98 		if ((ret = (*set)(netdev, new)) == 0)
99 			ret = len;
100 	}
101 	rtnl_unlock();
102  err:
103 	return ret;
104 }
105 
106 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
108 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
109 NETDEVICE_SHOW_RO(iflink, fmt_dec);
110 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
111 NETDEVICE_SHOW_RO(type, fmt_dec);
112 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
113 
114 /* use same locking rules as GIFHWADDR ioctl's */
115 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
116 			    char *buf)
117 {
118 	struct net_device *net = to_net_dev(dev);
119 	ssize_t ret = -EINVAL;
120 
121 	read_lock(&dev_base_lock);
122 	if (dev_isalive(net))
123 		ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
124 	read_unlock(&dev_base_lock);
125 	return ret;
126 }
127 static DEVICE_ATTR_RO(address);
128 
129 static ssize_t broadcast_show(struct device *dev,
130 			      struct device_attribute *attr, char *buf)
131 {
132 	struct net_device *net = to_net_dev(dev);
133 	if (dev_isalive(net))
134 		return sysfs_format_mac(buf, net->broadcast, net->addr_len);
135 	return -EINVAL;
136 }
137 static DEVICE_ATTR_RO(broadcast);
138 
139 static int change_carrier(struct net_device *net, unsigned long new_carrier)
140 {
141 	if (!netif_running(net))
142 		return -EINVAL;
143 	return dev_change_carrier(net, (bool) new_carrier);
144 }
145 
146 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
147 			     const char *buf, size_t len)
148 {
149 	return netdev_store(dev, attr, buf, len, change_carrier);
150 }
151 
152 static ssize_t carrier_show(struct device *dev,
153 			    struct device_attribute *attr, char *buf)
154 {
155 	struct net_device *netdev = to_net_dev(dev);
156 	if (netif_running(netdev)) {
157 		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
158 	}
159 	return -EINVAL;
160 }
161 static DEVICE_ATTR_RW(carrier);
162 
163 static ssize_t speed_show(struct device *dev,
164 			  struct device_attribute *attr, char *buf)
165 {
166 	struct net_device *netdev = to_net_dev(dev);
167 	int ret = -EINVAL;
168 
169 	if (!rtnl_trylock())
170 		return restart_syscall();
171 
172 	if (netif_running(netdev)) {
173 		struct ethtool_cmd cmd;
174 		if (!__ethtool_get_settings(netdev, &cmd))
175 			ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
176 	}
177 	rtnl_unlock();
178 	return ret;
179 }
180 static DEVICE_ATTR_RO(speed);
181 
182 static ssize_t duplex_show(struct device *dev,
183 			   struct device_attribute *attr, char *buf)
184 {
185 	struct net_device *netdev = to_net_dev(dev);
186 	int ret = -EINVAL;
187 
188 	if (!rtnl_trylock())
189 		return restart_syscall();
190 
191 	if (netif_running(netdev)) {
192 		struct ethtool_cmd cmd;
193 		if (!__ethtool_get_settings(netdev, &cmd)) {
194 			const char *duplex;
195 			switch (cmd.duplex) {
196 			case DUPLEX_HALF:
197 				duplex = "half";
198 				break;
199 			case DUPLEX_FULL:
200 				duplex = "full";
201 				break;
202 			default:
203 				duplex = "unknown";
204 				break;
205 			}
206 			ret = sprintf(buf, "%s\n", duplex);
207 		}
208 	}
209 	rtnl_unlock();
210 	return ret;
211 }
212 static DEVICE_ATTR_RO(duplex);
213 
214 static ssize_t dormant_show(struct device *dev,
215 			    struct device_attribute *attr, char *buf)
216 {
217 	struct net_device *netdev = to_net_dev(dev);
218 
219 	if (netif_running(netdev))
220 		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
221 
222 	return -EINVAL;
223 }
224 static DEVICE_ATTR_RO(dormant);
225 
226 static const char *const operstates[] = {
227 	"unknown",
228 	"notpresent", /* currently unused */
229 	"down",
230 	"lowerlayerdown",
231 	"testing", /* currently unused */
232 	"dormant",
233 	"up"
234 };
235 
236 static ssize_t operstate_show(struct device *dev,
237 			      struct device_attribute *attr, char *buf)
238 {
239 	const struct net_device *netdev = to_net_dev(dev);
240 	unsigned char operstate;
241 
242 	read_lock(&dev_base_lock);
243 	operstate = netdev->operstate;
244 	if (!netif_running(netdev))
245 		operstate = IF_OPER_DOWN;
246 	read_unlock(&dev_base_lock);
247 
248 	if (operstate >= ARRAY_SIZE(operstates))
249 		return -EINVAL; /* should not happen */
250 
251 	return sprintf(buf, "%s\n", operstates[operstate]);
252 }
253 static DEVICE_ATTR_RO(operstate);
254 
255 /* read-write attributes */
256 
257 static int change_mtu(struct net_device *net, unsigned long new_mtu)
258 {
259 	return dev_set_mtu(net, (int) new_mtu);
260 }
261 
262 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
263 			 const char *buf, size_t len)
264 {
265 	return netdev_store(dev, attr, buf, len, change_mtu);
266 }
267 NETDEVICE_SHOW_RW(mtu, fmt_dec);
268 
269 static int change_flags(struct net_device *net, unsigned long new_flags)
270 {
271 	return dev_change_flags(net, (unsigned int) new_flags);
272 }
273 
274 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
275 			   const char *buf, size_t len)
276 {
277 	return netdev_store(dev, attr, buf, len, change_flags);
278 }
279 NETDEVICE_SHOW_RW(flags, fmt_hex);
280 
281 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
282 {
283 	net->tx_queue_len = new_len;
284 	return 0;
285 }
286 
287 static ssize_t tx_queue_len_store(struct device *dev,
288 				  struct device_attribute *attr,
289 				  const char *buf, size_t len)
290 {
291 	if (!capable(CAP_NET_ADMIN))
292 		return -EPERM;
293 
294 	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
295 }
296 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
297 
298 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
299 			     const char *buf, size_t len)
300 {
301 	struct net_device *netdev = to_net_dev(dev);
302 	struct net *net = dev_net(netdev);
303 	size_t count = len;
304 	ssize_t ret;
305 
306 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
307 		return -EPERM;
308 
309 	/* ignore trailing newline */
310 	if (len >  0 && buf[len - 1] == '\n')
311 		--count;
312 
313 	if (!rtnl_trylock())
314 		return restart_syscall();
315 	ret = dev_set_alias(netdev, buf, count);
316 	rtnl_unlock();
317 
318 	return ret < 0 ? ret : len;
319 }
320 
321 static ssize_t ifalias_show(struct device *dev,
322 			    struct device_attribute *attr, char *buf)
323 {
324 	const struct net_device *netdev = to_net_dev(dev);
325 	ssize_t ret = 0;
326 
327 	if (!rtnl_trylock())
328 		return restart_syscall();
329 	if (netdev->ifalias)
330 		ret = sprintf(buf, "%s\n", netdev->ifalias);
331 	rtnl_unlock();
332 	return ret;
333 }
334 static DEVICE_ATTR_RW(ifalias);
335 
336 static int change_group(struct net_device *net, unsigned long new_group)
337 {
338 	dev_set_group(net, (int) new_group);
339 	return 0;
340 }
341 
342 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
343 			   const char *buf, size_t len)
344 {
345 	return netdev_store(dev, attr, buf, len, change_group);
346 }
347 NETDEVICE_SHOW(group, fmt_dec);
348 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
349 
350 static ssize_t phys_port_id_show(struct device *dev,
351 				 struct device_attribute *attr, char *buf)
352 {
353 	struct net_device *netdev = to_net_dev(dev);
354 	ssize_t ret = -EINVAL;
355 
356 	if (!rtnl_trylock())
357 		return restart_syscall();
358 
359 	if (dev_isalive(netdev)) {
360 		struct netdev_phys_port_id ppid;
361 
362 		ret = dev_get_phys_port_id(netdev, &ppid);
363 		if (!ret)
364 			ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
365 	}
366 	rtnl_unlock();
367 
368 	return ret;
369 }
370 static DEVICE_ATTR_RO(phys_port_id);
371 
372 static struct attribute *net_class_attrs[] = {
373 	&dev_attr_netdev_group.attr,
374 	&dev_attr_type.attr,
375 	&dev_attr_dev_id.attr,
376 	&dev_attr_iflink.attr,
377 	&dev_attr_ifindex.attr,
378 	&dev_attr_addr_assign_type.attr,
379 	&dev_attr_addr_len.attr,
380 	&dev_attr_link_mode.attr,
381 	&dev_attr_address.attr,
382 	&dev_attr_broadcast.attr,
383 	&dev_attr_speed.attr,
384 	&dev_attr_duplex.attr,
385 	&dev_attr_dormant.attr,
386 	&dev_attr_operstate.attr,
387 	&dev_attr_ifalias.attr,
388 	&dev_attr_carrier.attr,
389 	&dev_attr_mtu.attr,
390 	&dev_attr_flags.attr,
391 	&dev_attr_tx_queue_len.attr,
392 	&dev_attr_phys_port_id.attr,
393 	NULL,
394 };
395 ATTRIBUTE_GROUPS(net_class);
396 
397 /* Show a given an attribute in the statistics group */
398 static ssize_t netstat_show(const struct device *d,
399 			    struct device_attribute *attr, char *buf,
400 			    unsigned long offset)
401 {
402 	struct net_device *dev = to_net_dev(d);
403 	ssize_t ret = -EINVAL;
404 
405 	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
406 			offset % sizeof(u64) != 0);
407 
408 	read_lock(&dev_base_lock);
409 	if (dev_isalive(dev)) {
410 		struct rtnl_link_stats64 temp;
411 		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
412 
413 		ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
414 	}
415 	read_unlock(&dev_base_lock);
416 	return ret;
417 }
418 
419 /* generate a read-only statistics attribute */
420 #define NETSTAT_ENTRY(name)						\
421 static ssize_t name##_show(struct device *d,				\
422 			   struct device_attribute *attr, char *buf) 	\
423 {									\
424 	return netstat_show(d, attr, buf,				\
425 			    offsetof(struct rtnl_link_stats64, name));	\
426 }									\
427 static DEVICE_ATTR_RO(name)
428 
429 NETSTAT_ENTRY(rx_packets);
430 NETSTAT_ENTRY(tx_packets);
431 NETSTAT_ENTRY(rx_bytes);
432 NETSTAT_ENTRY(tx_bytes);
433 NETSTAT_ENTRY(rx_errors);
434 NETSTAT_ENTRY(tx_errors);
435 NETSTAT_ENTRY(rx_dropped);
436 NETSTAT_ENTRY(tx_dropped);
437 NETSTAT_ENTRY(multicast);
438 NETSTAT_ENTRY(collisions);
439 NETSTAT_ENTRY(rx_length_errors);
440 NETSTAT_ENTRY(rx_over_errors);
441 NETSTAT_ENTRY(rx_crc_errors);
442 NETSTAT_ENTRY(rx_frame_errors);
443 NETSTAT_ENTRY(rx_fifo_errors);
444 NETSTAT_ENTRY(rx_missed_errors);
445 NETSTAT_ENTRY(tx_aborted_errors);
446 NETSTAT_ENTRY(tx_carrier_errors);
447 NETSTAT_ENTRY(tx_fifo_errors);
448 NETSTAT_ENTRY(tx_heartbeat_errors);
449 NETSTAT_ENTRY(tx_window_errors);
450 NETSTAT_ENTRY(rx_compressed);
451 NETSTAT_ENTRY(tx_compressed);
452 
453 static struct attribute *netstat_attrs[] = {
454 	&dev_attr_rx_packets.attr,
455 	&dev_attr_tx_packets.attr,
456 	&dev_attr_rx_bytes.attr,
457 	&dev_attr_tx_bytes.attr,
458 	&dev_attr_rx_errors.attr,
459 	&dev_attr_tx_errors.attr,
460 	&dev_attr_rx_dropped.attr,
461 	&dev_attr_tx_dropped.attr,
462 	&dev_attr_multicast.attr,
463 	&dev_attr_collisions.attr,
464 	&dev_attr_rx_length_errors.attr,
465 	&dev_attr_rx_over_errors.attr,
466 	&dev_attr_rx_crc_errors.attr,
467 	&dev_attr_rx_frame_errors.attr,
468 	&dev_attr_rx_fifo_errors.attr,
469 	&dev_attr_rx_missed_errors.attr,
470 	&dev_attr_tx_aborted_errors.attr,
471 	&dev_attr_tx_carrier_errors.attr,
472 	&dev_attr_tx_fifo_errors.attr,
473 	&dev_attr_tx_heartbeat_errors.attr,
474 	&dev_attr_tx_window_errors.attr,
475 	&dev_attr_rx_compressed.attr,
476 	&dev_attr_tx_compressed.attr,
477 	NULL
478 };
479 
480 
481 static struct attribute_group netstat_group = {
482 	.name  = "statistics",
483 	.attrs  = netstat_attrs,
484 };
485 
486 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
487 static struct attribute *wireless_attrs[] = {
488 	NULL
489 };
490 
491 static struct attribute_group wireless_group = {
492 	.name = "wireless",
493 	.attrs = wireless_attrs,
494 };
495 #endif
496 
497 #else /* CONFIG_SYSFS */
498 #define net_class_groups	NULL
499 #endif /* CONFIG_SYSFS */
500 
501 #ifdef CONFIG_RPS
502 /*
503  * RX queue sysfs structures and functions.
504  */
505 struct rx_queue_attribute {
506 	struct attribute attr;
507 	ssize_t (*show)(struct netdev_rx_queue *queue,
508 	    struct rx_queue_attribute *attr, char *buf);
509 	ssize_t (*store)(struct netdev_rx_queue *queue,
510 	    struct rx_queue_attribute *attr, const char *buf, size_t len);
511 };
512 #define to_rx_queue_attr(_attr) container_of(_attr,		\
513     struct rx_queue_attribute, attr)
514 
515 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
516 
517 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
518 				  char *buf)
519 {
520 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
521 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
522 
523 	if (!attribute->show)
524 		return -EIO;
525 
526 	return attribute->show(queue, attribute, buf);
527 }
528 
529 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
530 				   const char *buf, size_t count)
531 {
532 	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
533 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
534 
535 	if (!attribute->store)
536 		return -EIO;
537 
538 	return attribute->store(queue, attribute, buf, count);
539 }
540 
541 static const struct sysfs_ops rx_queue_sysfs_ops = {
542 	.show = rx_queue_attr_show,
543 	.store = rx_queue_attr_store,
544 };
545 
546 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
547 			    struct rx_queue_attribute *attribute, char *buf)
548 {
549 	struct rps_map *map;
550 	cpumask_var_t mask;
551 	size_t len = 0;
552 	int i;
553 
554 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
555 		return -ENOMEM;
556 
557 	rcu_read_lock();
558 	map = rcu_dereference(queue->rps_map);
559 	if (map)
560 		for (i = 0; i < map->len; i++)
561 			cpumask_set_cpu(map->cpus[i], mask);
562 
563 	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
564 	if (PAGE_SIZE - len < 3) {
565 		rcu_read_unlock();
566 		free_cpumask_var(mask);
567 		return -EINVAL;
568 	}
569 	rcu_read_unlock();
570 
571 	free_cpumask_var(mask);
572 	len += sprintf(buf + len, "\n");
573 	return len;
574 }
575 
576 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
577 		      struct rx_queue_attribute *attribute,
578 		      const char *buf, size_t len)
579 {
580 	struct rps_map *old_map, *map;
581 	cpumask_var_t mask;
582 	int err, cpu, i;
583 	static DEFINE_SPINLOCK(rps_map_lock);
584 
585 	if (!capable(CAP_NET_ADMIN))
586 		return -EPERM;
587 
588 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
589 		return -ENOMEM;
590 
591 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
592 	if (err) {
593 		free_cpumask_var(mask);
594 		return err;
595 	}
596 
597 	map = kzalloc(max_t(unsigned int,
598 	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
599 	    GFP_KERNEL);
600 	if (!map) {
601 		free_cpumask_var(mask);
602 		return -ENOMEM;
603 	}
604 
605 	i = 0;
606 	for_each_cpu_and(cpu, mask, cpu_online_mask)
607 		map->cpus[i++] = cpu;
608 
609 	if (i)
610 		map->len = i;
611 	else {
612 		kfree(map);
613 		map = NULL;
614 	}
615 
616 	spin_lock(&rps_map_lock);
617 	old_map = rcu_dereference_protected(queue->rps_map,
618 					    lockdep_is_held(&rps_map_lock));
619 	rcu_assign_pointer(queue->rps_map, map);
620 	spin_unlock(&rps_map_lock);
621 
622 	if (map)
623 		static_key_slow_inc(&rps_needed);
624 	if (old_map) {
625 		kfree_rcu(old_map, rcu);
626 		static_key_slow_dec(&rps_needed);
627 	}
628 	free_cpumask_var(mask);
629 	return len;
630 }
631 
632 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
633 					   struct rx_queue_attribute *attr,
634 					   char *buf)
635 {
636 	struct rps_dev_flow_table *flow_table;
637 	unsigned long val = 0;
638 
639 	rcu_read_lock();
640 	flow_table = rcu_dereference(queue->rps_flow_table);
641 	if (flow_table)
642 		val = (unsigned long)flow_table->mask + 1;
643 	rcu_read_unlock();
644 
645 	return sprintf(buf, "%lu\n", val);
646 }
647 
648 static void rps_dev_flow_table_release(struct rcu_head *rcu)
649 {
650 	struct rps_dev_flow_table *table = container_of(rcu,
651 	    struct rps_dev_flow_table, rcu);
652 	vfree(table);
653 }
654 
655 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
656 				     struct rx_queue_attribute *attr,
657 				     const char *buf, size_t len)
658 {
659 	unsigned long mask, count;
660 	struct rps_dev_flow_table *table, *old_table;
661 	static DEFINE_SPINLOCK(rps_dev_flow_lock);
662 	int rc;
663 
664 	if (!capable(CAP_NET_ADMIN))
665 		return -EPERM;
666 
667 	rc = kstrtoul(buf, 0, &count);
668 	if (rc < 0)
669 		return rc;
670 
671 	if (count) {
672 		mask = count - 1;
673 		/* mask = roundup_pow_of_two(count) - 1;
674 		 * without overflows...
675 		 */
676 		while ((mask | (mask >> 1)) != mask)
677 			mask |= (mask >> 1);
678 		/* On 64 bit arches, must check mask fits in table->mask (u32),
679 		 * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
680 		 * doesnt overflow.
681 		 */
682 #if BITS_PER_LONG > 32
683 		if (mask > (unsigned long)(u32)mask)
684 			return -EINVAL;
685 #else
686 		if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
687 				/ sizeof(struct rps_dev_flow)) {
688 			/* Enforce a limit to prevent overflow */
689 			return -EINVAL;
690 		}
691 #endif
692 		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
693 		if (!table)
694 			return -ENOMEM;
695 
696 		table->mask = mask;
697 		for (count = 0; count <= mask; count++)
698 			table->flows[count].cpu = RPS_NO_CPU;
699 	} else
700 		table = NULL;
701 
702 	spin_lock(&rps_dev_flow_lock);
703 	old_table = rcu_dereference_protected(queue->rps_flow_table,
704 					      lockdep_is_held(&rps_dev_flow_lock));
705 	rcu_assign_pointer(queue->rps_flow_table, table);
706 	spin_unlock(&rps_dev_flow_lock);
707 
708 	if (old_table)
709 		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
710 
711 	return len;
712 }
713 
714 static struct rx_queue_attribute rps_cpus_attribute =
715 	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
716 
717 
718 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
719 	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
720 	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
721 
722 static struct attribute *rx_queue_default_attrs[] = {
723 	&rps_cpus_attribute.attr,
724 	&rps_dev_flow_table_cnt_attribute.attr,
725 	NULL
726 };
727 
728 static void rx_queue_release(struct kobject *kobj)
729 {
730 	struct netdev_rx_queue *queue = to_rx_queue(kobj);
731 	struct rps_map *map;
732 	struct rps_dev_flow_table *flow_table;
733 
734 
735 	map = rcu_dereference_protected(queue->rps_map, 1);
736 	if (map) {
737 		RCU_INIT_POINTER(queue->rps_map, NULL);
738 		kfree_rcu(map, rcu);
739 	}
740 
741 	flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
742 	if (flow_table) {
743 		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
744 		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
745 	}
746 
747 	memset(kobj, 0, sizeof(*kobj));
748 	dev_put(queue->dev);
749 }
750 
751 static struct kobj_type rx_queue_ktype = {
752 	.sysfs_ops = &rx_queue_sysfs_ops,
753 	.release = rx_queue_release,
754 	.default_attrs = rx_queue_default_attrs,
755 };
756 
757 static int rx_queue_add_kobject(struct net_device *net, int index)
758 {
759 	struct netdev_rx_queue *queue = net->_rx + index;
760 	struct kobject *kobj = &queue->kobj;
761 	int error = 0;
762 
763 	kobj->kset = net->queues_kset;
764 	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
765 	    "rx-%u", index);
766 	if (error) {
767 		kobject_put(kobj);
768 		return error;
769 	}
770 
771 	kobject_uevent(kobj, KOBJ_ADD);
772 	dev_hold(queue->dev);
773 
774 	return error;
775 }
776 #endif /* CONFIG_RPS */
777 
778 int
779 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
780 {
781 #ifdef CONFIG_RPS
782 	int i;
783 	int error = 0;
784 
785 	for (i = old_num; i < new_num; i++) {
786 		error = rx_queue_add_kobject(net, i);
787 		if (error) {
788 			new_num = old_num;
789 			break;
790 		}
791 	}
792 
793 	while (--i >= new_num)
794 		kobject_put(&net->_rx[i].kobj);
795 
796 	return error;
797 #else
798 	return 0;
799 #endif
800 }
801 
802 #ifdef CONFIG_SYSFS
803 /*
804  * netdev_queue sysfs structures and functions.
805  */
806 struct netdev_queue_attribute {
807 	struct attribute attr;
808 	ssize_t (*show)(struct netdev_queue *queue,
809 	    struct netdev_queue_attribute *attr, char *buf);
810 	ssize_t (*store)(struct netdev_queue *queue,
811 	    struct netdev_queue_attribute *attr, const char *buf, size_t len);
812 };
813 #define to_netdev_queue_attr(_attr) container_of(_attr,		\
814     struct netdev_queue_attribute, attr)
815 
816 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
817 
818 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
819 				      struct attribute *attr, char *buf)
820 {
821 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
822 	struct netdev_queue *queue = to_netdev_queue(kobj);
823 
824 	if (!attribute->show)
825 		return -EIO;
826 
827 	return attribute->show(queue, attribute, buf);
828 }
829 
830 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
831 				       struct attribute *attr,
832 				       const char *buf, size_t count)
833 {
834 	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
835 	struct netdev_queue *queue = to_netdev_queue(kobj);
836 
837 	if (!attribute->store)
838 		return -EIO;
839 
840 	return attribute->store(queue, attribute, buf, count);
841 }
842 
843 static const struct sysfs_ops netdev_queue_sysfs_ops = {
844 	.show = netdev_queue_attr_show,
845 	.store = netdev_queue_attr_store,
846 };
847 
848 static ssize_t show_trans_timeout(struct netdev_queue *queue,
849 				  struct netdev_queue_attribute *attribute,
850 				  char *buf)
851 {
852 	unsigned long trans_timeout;
853 
854 	spin_lock_irq(&queue->_xmit_lock);
855 	trans_timeout = queue->trans_timeout;
856 	spin_unlock_irq(&queue->_xmit_lock);
857 
858 	return sprintf(buf, "%lu", trans_timeout);
859 }
860 
861 static struct netdev_queue_attribute queue_trans_timeout =
862 	__ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
863 
864 #ifdef CONFIG_BQL
865 /*
866  * Byte queue limits sysfs structures and functions.
867  */
868 static ssize_t bql_show(char *buf, unsigned int value)
869 {
870 	return sprintf(buf, "%u\n", value);
871 }
872 
873 static ssize_t bql_set(const char *buf, const size_t count,
874 		       unsigned int *pvalue)
875 {
876 	unsigned int value;
877 	int err;
878 
879 	if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
880 		value = DQL_MAX_LIMIT;
881 	else {
882 		err = kstrtouint(buf, 10, &value);
883 		if (err < 0)
884 			return err;
885 		if (value > DQL_MAX_LIMIT)
886 			return -EINVAL;
887 	}
888 
889 	*pvalue = value;
890 
891 	return count;
892 }
893 
894 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
895 				  struct netdev_queue_attribute *attr,
896 				  char *buf)
897 {
898 	struct dql *dql = &queue->dql;
899 
900 	return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
901 }
902 
903 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
904 				 struct netdev_queue_attribute *attribute,
905 				 const char *buf, size_t len)
906 {
907 	struct dql *dql = &queue->dql;
908 	unsigned int value;
909 	int err;
910 
911 	err = kstrtouint(buf, 10, &value);
912 	if (err < 0)
913 		return err;
914 
915 	dql->slack_hold_time = msecs_to_jiffies(value);
916 
917 	return len;
918 }
919 
920 static struct netdev_queue_attribute bql_hold_time_attribute =
921 	__ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
922 	    bql_set_hold_time);
923 
924 static ssize_t bql_show_inflight(struct netdev_queue *queue,
925 				 struct netdev_queue_attribute *attr,
926 				 char *buf)
927 {
928 	struct dql *dql = &queue->dql;
929 
930 	return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
931 }
932 
933 static struct netdev_queue_attribute bql_inflight_attribute =
934 	__ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
935 
936 #define BQL_ATTR(NAME, FIELD)						\
937 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,		\
938 				 struct netdev_queue_attribute *attr,	\
939 				 char *buf)				\
940 {									\
941 	return bql_show(buf, queue->dql.FIELD);				\
942 }									\
943 									\
944 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,		\
945 				struct netdev_queue_attribute *attr,	\
946 				const char *buf, size_t len)		\
947 {									\
948 	return bql_set(buf, len, &queue->dql.FIELD);			\
949 }									\
950 									\
951 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =	\
952 	__ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,		\
953 	    bql_set_ ## NAME);
954 
955 BQL_ATTR(limit, limit)
956 BQL_ATTR(limit_max, max_limit)
957 BQL_ATTR(limit_min, min_limit)
958 
959 static struct attribute *dql_attrs[] = {
960 	&bql_limit_attribute.attr,
961 	&bql_limit_max_attribute.attr,
962 	&bql_limit_min_attribute.attr,
963 	&bql_hold_time_attribute.attr,
964 	&bql_inflight_attribute.attr,
965 	NULL
966 };
967 
968 static struct attribute_group dql_group = {
969 	.name  = "byte_queue_limits",
970 	.attrs  = dql_attrs,
971 };
972 #endif /* CONFIG_BQL */
973 
974 #ifdef CONFIG_XPS
975 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
976 {
977 	struct net_device *dev = queue->dev;
978 	int i;
979 
980 	for (i = 0; i < dev->num_tx_queues; i++)
981 		if (queue == &dev->_tx[i])
982 			break;
983 
984 	BUG_ON(i >= dev->num_tx_queues);
985 
986 	return i;
987 }
988 
989 
990 static ssize_t show_xps_map(struct netdev_queue *queue,
991 			    struct netdev_queue_attribute *attribute, char *buf)
992 {
993 	struct net_device *dev = queue->dev;
994 	struct xps_dev_maps *dev_maps;
995 	cpumask_var_t mask;
996 	unsigned long index;
997 	size_t len = 0;
998 	int i;
999 
1000 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1001 		return -ENOMEM;
1002 
1003 	index = get_netdev_queue_index(queue);
1004 
1005 	rcu_read_lock();
1006 	dev_maps = rcu_dereference(dev->xps_maps);
1007 	if (dev_maps) {
1008 		for_each_possible_cpu(i) {
1009 			struct xps_map *map =
1010 			    rcu_dereference(dev_maps->cpu_map[i]);
1011 			if (map) {
1012 				int j;
1013 				for (j = 0; j < map->len; j++) {
1014 					if (map->queues[j] == index) {
1015 						cpumask_set_cpu(i, mask);
1016 						break;
1017 					}
1018 				}
1019 			}
1020 		}
1021 	}
1022 	rcu_read_unlock();
1023 
1024 	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1025 	if (PAGE_SIZE - len < 3) {
1026 		free_cpumask_var(mask);
1027 		return -EINVAL;
1028 	}
1029 
1030 	free_cpumask_var(mask);
1031 	len += sprintf(buf + len, "\n");
1032 	return len;
1033 }
1034 
1035 static ssize_t store_xps_map(struct netdev_queue *queue,
1036 		      struct netdev_queue_attribute *attribute,
1037 		      const char *buf, size_t len)
1038 {
1039 	struct net_device *dev = queue->dev;
1040 	unsigned long index;
1041 	cpumask_var_t mask;
1042 	int err;
1043 
1044 	if (!capable(CAP_NET_ADMIN))
1045 		return -EPERM;
1046 
1047 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1048 		return -ENOMEM;
1049 
1050 	index = get_netdev_queue_index(queue);
1051 
1052 	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1053 	if (err) {
1054 		free_cpumask_var(mask);
1055 		return err;
1056 	}
1057 
1058 	err = netif_set_xps_queue(dev, mask, index);
1059 
1060 	free_cpumask_var(mask);
1061 
1062 	return err ? : len;
1063 }
1064 
1065 static struct netdev_queue_attribute xps_cpus_attribute =
1066     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1067 #endif /* CONFIG_XPS */
1068 
1069 static struct attribute *netdev_queue_default_attrs[] = {
1070 	&queue_trans_timeout.attr,
1071 #ifdef CONFIG_XPS
1072 	&xps_cpus_attribute.attr,
1073 #endif
1074 	NULL
1075 };
1076 
1077 static void netdev_queue_release(struct kobject *kobj)
1078 {
1079 	struct netdev_queue *queue = to_netdev_queue(kobj);
1080 
1081 	memset(kobj, 0, sizeof(*kobj));
1082 	dev_put(queue->dev);
1083 }
1084 
1085 static struct kobj_type netdev_queue_ktype = {
1086 	.sysfs_ops = &netdev_queue_sysfs_ops,
1087 	.release = netdev_queue_release,
1088 	.default_attrs = netdev_queue_default_attrs,
1089 };
1090 
1091 static int netdev_queue_add_kobject(struct net_device *net, int index)
1092 {
1093 	struct netdev_queue *queue = net->_tx + index;
1094 	struct kobject *kobj = &queue->kobj;
1095 	int error = 0;
1096 
1097 	kobj->kset = net->queues_kset;
1098 	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1099 	    "tx-%u", index);
1100 	if (error)
1101 		goto exit;
1102 
1103 #ifdef CONFIG_BQL
1104 	error = sysfs_create_group(kobj, &dql_group);
1105 	if (error)
1106 		goto exit;
1107 #endif
1108 
1109 	kobject_uevent(kobj, KOBJ_ADD);
1110 	dev_hold(queue->dev);
1111 
1112 	return 0;
1113 exit:
1114 	kobject_put(kobj);
1115 	return error;
1116 }
1117 #endif /* CONFIG_SYSFS */
1118 
1119 int
1120 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1121 {
1122 #ifdef CONFIG_SYSFS
1123 	int i;
1124 	int error = 0;
1125 
1126 	for (i = old_num; i < new_num; i++) {
1127 		error = netdev_queue_add_kobject(net, i);
1128 		if (error) {
1129 			new_num = old_num;
1130 			break;
1131 		}
1132 	}
1133 
1134 	while (--i >= new_num) {
1135 		struct netdev_queue *queue = net->_tx + i;
1136 
1137 #ifdef CONFIG_BQL
1138 		sysfs_remove_group(&queue->kobj, &dql_group);
1139 #endif
1140 		kobject_put(&queue->kobj);
1141 	}
1142 
1143 	return error;
1144 #else
1145 	return 0;
1146 #endif /* CONFIG_SYSFS */
1147 }
1148 
1149 static int register_queue_kobjects(struct net_device *net)
1150 {
1151 	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1152 
1153 #ifdef CONFIG_SYSFS
1154 	net->queues_kset = kset_create_and_add("queues",
1155 	    NULL, &net->dev.kobj);
1156 	if (!net->queues_kset)
1157 		return -ENOMEM;
1158 #endif
1159 
1160 #ifdef CONFIG_RPS
1161 	real_rx = net->real_num_rx_queues;
1162 #endif
1163 	real_tx = net->real_num_tx_queues;
1164 
1165 	error = net_rx_queue_update_kobjects(net, 0, real_rx);
1166 	if (error)
1167 		goto error;
1168 	rxq = real_rx;
1169 
1170 	error = netdev_queue_update_kobjects(net, 0, real_tx);
1171 	if (error)
1172 		goto error;
1173 	txq = real_tx;
1174 
1175 	return 0;
1176 
1177 error:
1178 	netdev_queue_update_kobjects(net, txq, 0);
1179 	net_rx_queue_update_kobjects(net, rxq, 0);
1180 	return error;
1181 }
1182 
1183 static void remove_queue_kobjects(struct net_device *net)
1184 {
1185 	int real_rx = 0, real_tx = 0;
1186 
1187 #ifdef CONFIG_RPS
1188 	real_rx = net->real_num_rx_queues;
1189 #endif
1190 	real_tx = net->real_num_tx_queues;
1191 
1192 	net_rx_queue_update_kobjects(net, real_rx, 0);
1193 	netdev_queue_update_kobjects(net, real_tx, 0);
1194 #ifdef CONFIG_SYSFS
1195 	kset_unregister(net->queues_kset);
1196 #endif
1197 }
1198 
1199 static bool net_current_may_mount(void)
1200 {
1201 	struct net *net = current->nsproxy->net_ns;
1202 
1203 	return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1204 }
1205 
1206 static void *net_grab_current_ns(void)
1207 {
1208 	struct net *ns = current->nsproxy->net_ns;
1209 #ifdef CONFIG_NET_NS
1210 	if (ns)
1211 		atomic_inc(&ns->passive);
1212 #endif
1213 	return ns;
1214 }
1215 
1216 static const void *net_initial_ns(void)
1217 {
1218 	return &init_net;
1219 }
1220 
1221 static const void *net_netlink_ns(struct sock *sk)
1222 {
1223 	return sock_net(sk);
1224 }
1225 
1226 struct kobj_ns_type_operations net_ns_type_operations = {
1227 	.type = KOBJ_NS_TYPE_NET,
1228 	.current_may_mount = net_current_may_mount,
1229 	.grab_current_ns = net_grab_current_ns,
1230 	.netlink_ns = net_netlink_ns,
1231 	.initial_ns = net_initial_ns,
1232 	.drop_ns = net_drop_ns,
1233 };
1234 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1235 
1236 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1237 {
1238 	struct net_device *dev = to_net_dev(d);
1239 	int retval;
1240 
1241 	/* pass interface to uevent. */
1242 	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1243 	if (retval)
1244 		goto exit;
1245 
1246 	/* pass ifindex to uevent.
1247 	 * ifindex is useful as it won't change (interface name may change)
1248 	 * and is what RtNetlink uses natively. */
1249 	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1250 
1251 exit:
1252 	return retval;
1253 }
1254 
1255 /*
1256  *	netdev_release -- destroy and free a dead device.
1257  *	Called when last reference to device kobject is gone.
1258  */
1259 static void netdev_release(struct device *d)
1260 {
1261 	struct net_device *dev = to_net_dev(d);
1262 
1263 	BUG_ON(dev->reg_state != NETREG_RELEASED);
1264 
1265 	kfree(dev->ifalias);
1266 	kfree((char *)dev - dev->padded);
1267 }
1268 
1269 static const void *net_namespace(struct device *d)
1270 {
1271 	struct net_device *dev;
1272 	dev = container_of(d, struct net_device, dev);
1273 	return dev_net(dev);
1274 }
1275 
1276 static struct class net_class = {
1277 	.name = "net",
1278 	.dev_release = netdev_release,
1279 	.dev_groups = net_class_groups,
1280 	.dev_uevent = netdev_uevent,
1281 	.ns_type = &net_ns_type_operations,
1282 	.namespace = net_namespace,
1283 };
1284 
1285 /* Delete sysfs entries but hold kobject reference until after all
1286  * netdev references are gone.
1287  */
1288 void netdev_unregister_kobject(struct net_device * net)
1289 {
1290 	struct device *dev = &(net->dev);
1291 
1292 	kobject_get(&dev->kobj);
1293 
1294 	remove_queue_kobjects(net);
1295 
1296 	pm_runtime_set_memalloc_noio(dev, false);
1297 
1298 	device_del(dev);
1299 }
1300 
1301 /* Create sysfs entries for network device. */
1302 int netdev_register_kobject(struct net_device *net)
1303 {
1304 	struct device *dev = &(net->dev);
1305 	const struct attribute_group **groups = net->sysfs_groups;
1306 	int error = 0;
1307 
1308 	device_initialize(dev);
1309 	dev->class = &net_class;
1310 	dev->platform_data = net;
1311 	dev->groups = groups;
1312 
1313 	dev_set_name(dev, "%s", net->name);
1314 
1315 #ifdef CONFIG_SYSFS
1316 	/* Allow for a device specific group */
1317 	if (*groups)
1318 		groups++;
1319 
1320 	*groups++ = &netstat_group;
1321 
1322 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1323 	if (net->ieee80211_ptr)
1324 		*groups++ = &wireless_group;
1325 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1326 	else if (net->wireless_handlers)
1327 		*groups++ = &wireless_group;
1328 #endif
1329 #endif
1330 #endif /* CONFIG_SYSFS */
1331 
1332 	error = device_add(dev);
1333 	if (error)
1334 		return error;
1335 
1336 	error = register_queue_kobjects(net);
1337 	if (error) {
1338 		device_del(dev);
1339 		return error;
1340 	}
1341 
1342 	pm_runtime_set_memalloc_noio(dev, true);
1343 
1344 	return error;
1345 }
1346 
1347 int netdev_class_create_file(struct class_attribute *class_attr)
1348 {
1349 	return class_create_file(&net_class, class_attr);
1350 }
1351 EXPORT_SYMBOL(netdev_class_create_file);
1352 
1353 void netdev_class_remove_file(struct class_attribute *class_attr)
1354 {
1355 	class_remove_file(&net_class, class_attr);
1356 }
1357 EXPORT_SYMBOL(netdev_class_remove_file);
1358 
1359 int netdev_kobject_init(void)
1360 {
1361 	kobj_ns_type_register(&net_ns_type_operations);
1362 	return class_register(&net_class);
1363 }
1364