Loading...
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 <linux/export.h>
24#include <linux/jiffies.h>
25#include <net/wext.h>
26
27#include "net-sysfs.h"
28
29#ifdef CONFIG_SYSFS
30static const char fmt_hex[] = "%#x\n";
31static const char fmt_long_hex[] = "%#lx\n";
32static const char fmt_dec[] = "%d\n";
33static const char fmt_udec[] = "%u\n";
34static const char fmt_ulong[] = "%lu\n";
35static const char fmt_u64[] = "%llu\n";
36
37static inline int dev_isalive(const struct net_device *dev)
38{
39 return dev->reg_state <= NETREG_REGISTERED;
40}
41
42/* use same locking rules as GIF* ioctl's */
43static ssize_t netdev_show(const struct device *dev,
44 struct device_attribute *attr, char *buf,
45 ssize_t (*format)(const struct net_device *, char *))
46{
47 struct net_device *net = to_net_dev(dev);
48 ssize_t ret = -EINVAL;
49
50 read_lock(&dev_base_lock);
51 if (dev_isalive(net))
52 ret = (*format)(net, buf);
53 read_unlock(&dev_base_lock);
54
55 return ret;
56}
57
58/* generate a show function for simple field */
59#define NETDEVICE_SHOW(field, format_string) \
60static ssize_t format_##field(const struct net_device *net, char *buf) \
61{ \
62 return sprintf(buf, format_string, net->field); \
63} \
64static ssize_t show_##field(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
66{ \
67 return netdev_show(dev, attr, buf, format_##field); \
68}
69
70
71/* use same locking and permission rules as SIF* ioctl's */
72static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
73 const char *buf, size_t len,
74 int (*set)(struct net_device *, unsigned long))
75{
76 struct net_device *net = to_net_dev(dev);
77 unsigned long new;
78 int ret = -EINVAL;
79
80 if (!capable(CAP_NET_ADMIN))
81 return -EPERM;
82
83 ret = kstrtoul(buf, 0, &new);
84 if (ret)
85 goto err;
86
87 if (!rtnl_trylock())
88 return restart_syscall();
89
90 if (dev_isalive(net)) {
91 if ((ret = (*set)(net, new)) == 0)
92 ret = len;
93 }
94 rtnl_unlock();
95 err:
96 return ret;
97}
98
99NETDEVICE_SHOW(dev_id, fmt_hex);
100NETDEVICE_SHOW(addr_assign_type, fmt_dec);
101NETDEVICE_SHOW(addr_len, fmt_dec);
102NETDEVICE_SHOW(iflink, fmt_dec);
103NETDEVICE_SHOW(ifindex, fmt_dec);
104NETDEVICE_SHOW(type, fmt_dec);
105NETDEVICE_SHOW(link_mode, fmt_dec);
106
107/* use same locking rules as GIFHWADDR ioctl's */
108static ssize_t show_address(struct device *dev, struct device_attribute *attr,
109 char *buf)
110{
111 struct net_device *net = to_net_dev(dev);
112 ssize_t ret = -EINVAL;
113
114 read_lock(&dev_base_lock);
115 if (dev_isalive(net))
116 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
117 read_unlock(&dev_base_lock);
118 return ret;
119}
120
121static ssize_t show_broadcast(struct device *dev,
122 struct device_attribute *attr, char *buf)
123{
124 struct net_device *net = to_net_dev(dev);
125 if (dev_isalive(net))
126 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
127 return -EINVAL;
128}
129
130static ssize_t show_carrier(struct device *dev,
131 struct device_attribute *attr, char *buf)
132{
133 struct net_device *netdev = to_net_dev(dev);
134 if (netif_running(netdev)) {
135 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
136 }
137 return -EINVAL;
138}
139
140static ssize_t show_speed(struct device *dev,
141 struct device_attribute *attr, char *buf)
142{
143 struct net_device *netdev = to_net_dev(dev);
144 int ret = -EINVAL;
145
146 if (!rtnl_trylock())
147 return restart_syscall();
148
149 if (netif_running(netdev)) {
150 struct ethtool_cmd cmd;
151 if (!__ethtool_get_settings(netdev, &cmd))
152 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
153 }
154 rtnl_unlock();
155 return ret;
156}
157
158static ssize_t show_duplex(struct device *dev,
159 struct device_attribute *attr, char *buf)
160{
161 struct net_device *netdev = to_net_dev(dev);
162 int ret = -EINVAL;
163
164 if (!rtnl_trylock())
165 return restart_syscall();
166
167 if (netif_running(netdev)) {
168 struct ethtool_cmd cmd;
169 if (!__ethtool_get_settings(netdev, &cmd))
170 ret = sprintf(buf, "%s\n",
171 cmd.duplex ? "full" : "half");
172 }
173 rtnl_unlock();
174 return ret;
175}
176
177static ssize_t show_dormant(struct device *dev,
178 struct device_attribute *attr, char *buf)
179{
180 struct net_device *netdev = to_net_dev(dev);
181
182 if (netif_running(netdev))
183 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
184
185 return -EINVAL;
186}
187
188static const char *const operstates[] = {
189 "unknown",
190 "notpresent", /* currently unused */
191 "down",
192 "lowerlayerdown",
193 "testing", /* currently unused */
194 "dormant",
195 "up"
196};
197
198static ssize_t show_operstate(struct device *dev,
199 struct device_attribute *attr, char *buf)
200{
201 const struct net_device *netdev = to_net_dev(dev);
202 unsigned char operstate;
203
204 read_lock(&dev_base_lock);
205 operstate = netdev->operstate;
206 if (!netif_running(netdev))
207 operstate = IF_OPER_DOWN;
208 read_unlock(&dev_base_lock);
209
210 if (operstate >= ARRAY_SIZE(operstates))
211 return -EINVAL; /* should not happen */
212
213 return sprintf(buf, "%s\n", operstates[operstate]);
214}
215
216/* read-write attributes */
217NETDEVICE_SHOW(mtu, fmt_dec);
218
219static int change_mtu(struct net_device *net, unsigned long new_mtu)
220{
221 return dev_set_mtu(net, (int) new_mtu);
222}
223
224static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
225 const char *buf, size_t len)
226{
227 return netdev_store(dev, attr, buf, len, change_mtu);
228}
229
230NETDEVICE_SHOW(flags, fmt_hex);
231
232static int change_flags(struct net_device *net, unsigned long new_flags)
233{
234 return dev_change_flags(net, (unsigned int) new_flags);
235}
236
237static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t len)
239{
240 return netdev_store(dev, attr, buf, len, change_flags);
241}
242
243NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
244
245static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
246{
247 net->tx_queue_len = new_len;
248 return 0;
249}
250
251static ssize_t store_tx_queue_len(struct device *dev,
252 struct device_attribute *attr,
253 const char *buf, size_t len)
254{
255 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
256}
257
258static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
259 const char *buf, size_t len)
260{
261 struct net_device *netdev = to_net_dev(dev);
262 size_t count = len;
263 ssize_t ret;
264
265 if (!capable(CAP_NET_ADMIN))
266 return -EPERM;
267
268 /* ignore trailing newline */
269 if (len > 0 && buf[len - 1] == '\n')
270 --count;
271
272 if (!rtnl_trylock())
273 return restart_syscall();
274 ret = dev_set_alias(netdev, buf, count);
275 rtnl_unlock();
276
277 return ret < 0 ? ret : len;
278}
279
280static ssize_t show_ifalias(struct device *dev,
281 struct device_attribute *attr, char *buf)
282{
283 const struct net_device *netdev = to_net_dev(dev);
284 ssize_t ret = 0;
285
286 if (!rtnl_trylock())
287 return restart_syscall();
288 if (netdev->ifalias)
289 ret = sprintf(buf, "%s\n", netdev->ifalias);
290 rtnl_unlock();
291 return ret;
292}
293
294NETDEVICE_SHOW(group, fmt_dec);
295
296static int change_group(struct net_device *net, unsigned long new_group)
297{
298 dev_set_group(net, (int) new_group);
299 return 0;
300}
301
302static ssize_t store_group(struct device *dev, struct device_attribute *attr,
303 const char *buf, size_t len)
304{
305 return netdev_store(dev, attr, buf, len, change_group);
306}
307
308static struct device_attribute net_class_attributes[] = {
309 __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
310 __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
311 __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
312 __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
313 __ATTR(iflink, S_IRUGO, show_iflink, NULL),
314 __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
315 __ATTR(type, S_IRUGO, show_type, NULL),
316 __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
317 __ATTR(address, S_IRUGO, show_address, NULL),
318 __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
319 __ATTR(carrier, S_IRUGO, show_carrier, NULL),
320 __ATTR(speed, S_IRUGO, show_speed, NULL),
321 __ATTR(duplex, S_IRUGO, show_duplex, NULL),
322 __ATTR(dormant, S_IRUGO, show_dormant, NULL),
323 __ATTR(operstate, S_IRUGO, show_operstate, NULL),
324 __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
325 __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
326 __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
327 store_tx_queue_len),
328 __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
329 {}
330};
331
332/* Show a given an attribute in the statistics group */
333static ssize_t netstat_show(const struct device *d,
334 struct device_attribute *attr, char *buf,
335 unsigned long offset)
336{
337 struct net_device *dev = to_net_dev(d);
338 ssize_t ret = -EINVAL;
339
340 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
341 offset % sizeof(u64) != 0);
342
343 read_lock(&dev_base_lock);
344 if (dev_isalive(dev)) {
345 struct rtnl_link_stats64 temp;
346 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
347
348 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
349 }
350 read_unlock(&dev_base_lock);
351 return ret;
352}
353
354/* generate a read-only statistics attribute */
355#define NETSTAT_ENTRY(name) \
356static ssize_t show_##name(struct device *d, \
357 struct device_attribute *attr, char *buf) \
358{ \
359 return netstat_show(d, attr, buf, \
360 offsetof(struct rtnl_link_stats64, name)); \
361} \
362static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
363
364NETSTAT_ENTRY(rx_packets);
365NETSTAT_ENTRY(tx_packets);
366NETSTAT_ENTRY(rx_bytes);
367NETSTAT_ENTRY(tx_bytes);
368NETSTAT_ENTRY(rx_errors);
369NETSTAT_ENTRY(tx_errors);
370NETSTAT_ENTRY(rx_dropped);
371NETSTAT_ENTRY(tx_dropped);
372NETSTAT_ENTRY(multicast);
373NETSTAT_ENTRY(collisions);
374NETSTAT_ENTRY(rx_length_errors);
375NETSTAT_ENTRY(rx_over_errors);
376NETSTAT_ENTRY(rx_crc_errors);
377NETSTAT_ENTRY(rx_frame_errors);
378NETSTAT_ENTRY(rx_fifo_errors);
379NETSTAT_ENTRY(rx_missed_errors);
380NETSTAT_ENTRY(tx_aborted_errors);
381NETSTAT_ENTRY(tx_carrier_errors);
382NETSTAT_ENTRY(tx_fifo_errors);
383NETSTAT_ENTRY(tx_heartbeat_errors);
384NETSTAT_ENTRY(tx_window_errors);
385NETSTAT_ENTRY(rx_compressed);
386NETSTAT_ENTRY(tx_compressed);
387
388static struct attribute *netstat_attrs[] = {
389 &dev_attr_rx_packets.attr,
390 &dev_attr_tx_packets.attr,
391 &dev_attr_rx_bytes.attr,
392 &dev_attr_tx_bytes.attr,
393 &dev_attr_rx_errors.attr,
394 &dev_attr_tx_errors.attr,
395 &dev_attr_rx_dropped.attr,
396 &dev_attr_tx_dropped.attr,
397 &dev_attr_multicast.attr,
398 &dev_attr_collisions.attr,
399 &dev_attr_rx_length_errors.attr,
400 &dev_attr_rx_over_errors.attr,
401 &dev_attr_rx_crc_errors.attr,
402 &dev_attr_rx_frame_errors.attr,
403 &dev_attr_rx_fifo_errors.attr,
404 &dev_attr_rx_missed_errors.attr,
405 &dev_attr_tx_aborted_errors.attr,
406 &dev_attr_tx_carrier_errors.attr,
407 &dev_attr_tx_fifo_errors.attr,
408 &dev_attr_tx_heartbeat_errors.attr,
409 &dev_attr_tx_window_errors.attr,
410 &dev_attr_rx_compressed.attr,
411 &dev_attr_tx_compressed.attr,
412 NULL
413};
414
415
416static struct attribute_group netstat_group = {
417 .name = "statistics",
418 .attrs = netstat_attrs,
419};
420
421#ifdef CONFIG_WIRELESS_EXT_SYSFS
422/* helper function that does all the locking etc for wireless stats */
423static ssize_t wireless_show(struct device *d, char *buf,
424 ssize_t (*format)(const struct iw_statistics *,
425 char *))
426{
427 struct net_device *dev = to_net_dev(d);
428 const struct iw_statistics *iw;
429 ssize_t ret = -EINVAL;
430
431 if (!rtnl_trylock())
432 return restart_syscall();
433 if (dev_isalive(dev)) {
434 iw = get_wireless_stats(dev);
435 if (iw)
436 ret = (*format)(iw, buf);
437 }
438 rtnl_unlock();
439
440 return ret;
441}
442
443/* show function template for wireless fields */
444#define WIRELESS_SHOW(name, field, format_string) \
445static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
446{ \
447 return sprintf(buf, format_string, iw->field); \
448} \
449static ssize_t show_iw_##name(struct device *d, \
450 struct device_attribute *attr, char *buf) \
451{ \
452 return wireless_show(d, buf, format_iw_##name); \
453} \
454static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
455
456WIRELESS_SHOW(status, status, fmt_hex);
457WIRELESS_SHOW(link, qual.qual, fmt_dec);
458WIRELESS_SHOW(level, qual.level, fmt_dec);
459WIRELESS_SHOW(noise, qual.noise, fmt_dec);
460WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
461WIRELESS_SHOW(crypt, discard.code, fmt_dec);
462WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
463WIRELESS_SHOW(misc, discard.misc, fmt_dec);
464WIRELESS_SHOW(retries, discard.retries, fmt_dec);
465WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
466
467static struct attribute *wireless_attrs[] = {
468 &dev_attr_status.attr,
469 &dev_attr_link.attr,
470 &dev_attr_level.attr,
471 &dev_attr_noise.attr,
472 &dev_attr_nwid.attr,
473 &dev_attr_crypt.attr,
474 &dev_attr_fragment.attr,
475 &dev_attr_retries.attr,
476 &dev_attr_misc.attr,
477 &dev_attr_beacon.attr,
478 NULL
479};
480
481static struct attribute_group wireless_group = {
482 .name = "wireless",
483 .attrs = wireless_attrs,
484};
485#endif
486#endif /* CONFIG_SYSFS */
487
488#ifdef CONFIG_RPS
489/*
490 * RX queue sysfs structures and functions.
491 */
492struct rx_queue_attribute {
493 struct attribute attr;
494 ssize_t (*show)(struct netdev_rx_queue *queue,
495 struct rx_queue_attribute *attr, char *buf);
496 ssize_t (*store)(struct netdev_rx_queue *queue,
497 struct rx_queue_attribute *attr, const char *buf, size_t len);
498};
499#define to_rx_queue_attr(_attr) container_of(_attr, \
500 struct rx_queue_attribute, attr)
501
502#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
503
504static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
505 char *buf)
506{
507 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
508 struct netdev_rx_queue *queue = to_rx_queue(kobj);
509
510 if (!attribute->show)
511 return -EIO;
512
513 return attribute->show(queue, attribute, buf);
514}
515
516static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
517 const char *buf, size_t count)
518{
519 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
520 struct netdev_rx_queue *queue = to_rx_queue(kobj);
521
522 if (!attribute->store)
523 return -EIO;
524
525 return attribute->store(queue, attribute, buf, count);
526}
527
528static const struct sysfs_ops rx_queue_sysfs_ops = {
529 .show = rx_queue_attr_show,
530 .store = rx_queue_attr_store,
531};
532
533static ssize_t show_rps_map(struct netdev_rx_queue *queue,
534 struct rx_queue_attribute *attribute, char *buf)
535{
536 struct rps_map *map;
537 cpumask_var_t mask;
538 size_t len = 0;
539 int i;
540
541 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
542 return -ENOMEM;
543
544 rcu_read_lock();
545 map = rcu_dereference(queue->rps_map);
546 if (map)
547 for (i = 0; i < map->len; i++)
548 cpumask_set_cpu(map->cpus[i], mask);
549
550 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
551 if (PAGE_SIZE - len < 3) {
552 rcu_read_unlock();
553 free_cpumask_var(mask);
554 return -EINVAL;
555 }
556 rcu_read_unlock();
557
558 free_cpumask_var(mask);
559 len += sprintf(buf + len, "\n");
560 return len;
561}
562
563static ssize_t store_rps_map(struct netdev_rx_queue *queue,
564 struct rx_queue_attribute *attribute,
565 const char *buf, size_t len)
566{
567 struct rps_map *old_map, *map;
568 cpumask_var_t mask;
569 int err, cpu, i;
570 static DEFINE_SPINLOCK(rps_map_lock);
571
572 if (!capable(CAP_NET_ADMIN))
573 return -EPERM;
574
575 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
576 return -ENOMEM;
577
578 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
579 if (err) {
580 free_cpumask_var(mask);
581 return err;
582 }
583
584 map = kzalloc(max_t(unsigned int,
585 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
586 GFP_KERNEL);
587 if (!map) {
588 free_cpumask_var(mask);
589 return -ENOMEM;
590 }
591
592 i = 0;
593 for_each_cpu_and(cpu, mask, cpu_online_mask)
594 map->cpus[i++] = cpu;
595
596 if (i)
597 map->len = i;
598 else {
599 kfree(map);
600 map = NULL;
601 }
602
603 spin_lock(&rps_map_lock);
604 old_map = rcu_dereference_protected(queue->rps_map,
605 lockdep_is_held(&rps_map_lock));
606 rcu_assign_pointer(queue->rps_map, map);
607 spin_unlock(&rps_map_lock);
608
609 if (map)
610 static_key_slow_inc(&rps_needed);
611 if (old_map) {
612 kfree_rcu(old_map, rcu);
613 static_key_slow_dec(&rps_needed);
614 }
615 free_cpumask_var(mask);
616 return len;
617}
618
619static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
620 struct rx_queue_attribute *attr,
621 char *buf)
622{
623 struct rps_dev_flow_table *flow_table;
624 unsigned long val = 0;
625
626 rcu_read_lock();
627 flow_table = rcu_dereference(queue->rps_flow_table);
628 if (flow_table)
629 val = (unsigned long)flow_table->mask + 1;
630 rcu_read_unlock();
631
632 return sprintf(buf, "%lu\n", val);
633}
634
635static void rps_dev_flow_table_release_work(struct work_struct *work)
636{
637 struct rps_dev_flow_table *table = container_of(work,
638 struct rps_dev_flow_table, free_work);
639
640 vfree(table);
641}
642
643static void rps_dev_flow_table_release(struct rcu_head *rcu)
644{
645 struct rps_dev_flow_table *table = container_of(rcu,
646 struct rps_dev_flow_table, rcu);
647
648 INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
649 schedule_work(&table->free_work);
650}
651
652static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
653 struct rx_queue_attribute *attr,
654 const char *buf, size_t len)
655{
656 unsigned long mask, count;
657 struct rps_dev_flow_table *table, *old_table;
658 static DEFINE_SPINLOCK(rps_dev_flow_lock);
659 int rc;
660
661 if (!capable(CAP_NET_ADMIN))
662 return -EPERM;
663
664 rc = kstrtoul(buf, 0, &count);
665 if (rc < 0)
666 return rc;
667
668 if (count) {
669 mask = count - 1;
670 /* mask = roundup_pow_of_two(count) - 1;
671 * without overflows...
672 */
673 while ((mask | (mask >> 1)) != mask)
674 mask |= (mask >> 1);
675 /* On 64 bit arches, must check mask fits in table->mask (u32),
676 * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
677 * doesnt overflow.
678 */
679#if BITS_PER_LONG > 32
680 if (mask > (unsigned long)(u32)mask)
681 return -EINVAL;
682#else
683 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
684 / sizeof(struct rps_dev_flow)) {
685 /* Enforce a limit to prevent overflow */
686 return -EINVAL;
687 }
688#endif
689 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
690 if (!table)
691 return -ENOMEM;
692
693 table->mask = mask;
694 for (count = 0; count <= mask; count++)
695 table->flows[count].cpu = RPS_NO_CPU;
696 } else
697 table = NULL;
698
699 spin_lock(&rps_dev_flow_lock);
700 old_table = rcu_dereference_protected(queue->rps_flow_table,
701 lockdep_is_held(&rps_dev_flow_lock));
702 rcu_assign_pointer(queue->rps_flow_table, table);
703 spin_unlock(&rps_dev_flow_lock);
704
705 if (old_table)
706 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
707
708 return len;
709}
710
711static struct rx_queue_attribute rps_cpus_attribute =
712 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
713
714
715static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
716 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
717 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
718
719static struct attribute *rx_queue_default_attrs[] = {
720 &rps_cpus_attribute.attr,
721 &rps_dev_flow_table_cnt_attribute.attr,
722 NULL
723};
724
725static void rx_queue_release(struct kobject *kobj)
726{
727 struct netdev_rx_queue *queue = to_rx_queue(kobj);
728 struct rps_map *map;
729 struct rps_dev_flow_table *flow_table;
730
731
732 map = rcu_dereference_protected(queue->rps_map, 1);
733 if (map) {
734 RCU_INIT_POINTER(queue->rps_map, NULL);
735 kfree_rcu(map, rcu);
736 }
737
738 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
739 if (flow_table) {
740 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
741 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
742 }
743
744 memset(kobj, 0, sizeof(*kobj));
745 dev_put(queue->dev);
746}
747
748static struct kobj_type rx_queue_ktype = {
749 .sysfs_ops = &rx_queue_sysfs_ops,
750 .release = rx_queue_release,
751 .default_attrs = rx_queue_default_attrs,
752};
753
754static int rx_queue_add_kobject(struct net_device *net, int index)
755{
756 struct netdev_rx_queue *queue = net->_rx + index;
757 struct kobject *kobj = &queue->kobj;
758 int error = 0;
759
760 kobj->kset = net->queues_kset;
761 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
762 "rx-%u", index);
763 if (error) {
764 kobject_put(kobj);
765 return error;
766 }
767
768 kobject_uevent(kobj, KOBJ_ADD);
769 dev_hold(queue->dev);
770
771 return error;
772}
773#endif /* CONFIG_RPS */
774
775int
776net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
777{
778#ifdef CONFIG_RPS
779 int i;
780 int error = 0;
781
782 for (i = old_num; i < new_num; i++) {
783 error = rx_queue_add_kobject(net, i);
784 if (error) {
785 new_num = old_num;
786 break;
787 }
788 }
789
790 while (--i >= new_num)
791 kobject_put(&net->_rx[i].kobj);
792
793 return error;
794#else
795 return 0;
796#endif
797}
798
799#ifdef CONFIG_SYSFS
800/*
801 * netdev_queue sysfs structures and functions.
802 */
803struct netdev_queue_attribute {
804 struct attribute attr;
805 ssize_t (*show)(struct netdev_queue *queue,
806 struct netdev_queue_attribute *attr, char *buf);
807 ssize_t (*store)(struct netdev_queue *queue,
808 struct netdev_queue_attribute *attr, const char *buf, size_t len);
809};
810#define to_netdev_queue_attr(_attr) container_of(_attr, \
811 struct netdev_queue_attribute, attr)
812
813#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
814
815static ssize_t netdev_queue_attr_show(struct kobject *kobj,
816 struct attribute *attr, char *buf)
817{
818 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
819 struct netdev_queue *queue = to_netdev_queue(kobj);
820
821 if (!attribute->show)
822 return -EIO;
823
824 return attribute->show(queue, attribute, buf);
825}
826
827static ssize_t netdev_queue_attr_store(struct kobject *kobj,
828 struct attribute *attr,
829 const char *buf, size_t count)
830{
831 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
832 struct netdev_queue *queue = to_netdev_queue(kobj);
833
834 if (!attribute->store)
835 return -EIO;
836
837 return attribute->store(queue, attribute, buf, count);
838}
839
840static const struct sysfs_ops netdev_queue_sysfs_ops = {
841 .show = netdev_queue_attr_show,
842 .store = netdev_queue_attr_store,
843};
844
845static ssize_t show_trans_timeout(struct netdev_queue *queue,
846 struct netdev_queue_attribute *attribute,
847 char *buf)
848{
849 unsigned long trans_timeout;
850
851 spin_lock_irq(&queue->_xmit_lock);
852 trans_timeout = queue->trans_timeout;
853 spin_unlock_irq(&queue->_xmit_lock);
854
855 return sprintf(buf, "%lu", trans_timeout);
856}
857
858static struct netdev_queue_attribute queue_trans_timeout =
859 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
860
861#ifdef CONFIG_BQL
862/*
863 * Byte queue limits sysfs structures and functions.
864 */
865static ssize_t bql_show(char *buf, unsigned int value)
866{
867 return sprintf(buf, "%u\n", value);
868}
869
870static ssize_t bql_set(const char *buf, const size_t count,
871 unsigned int *pvalue)
872{
873 unsigned int value;
874 int err;
875
876 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
877 value = DQL_MAX_LIMIT;
878 else {
879 err = kstrtouint(buf, 10, &value);
880 if (err < 0)
881 return err;
882 if (value > DQL_MAX_LIMIT)
883 return -EINVAL;
884 }
885
886 *pvalue = value;
887
888 return count;
889}
890
891static ssize_t bql_show_hold_time(struct netdev_queue *queue,
892 struct netdev_queue_attribute *attr,
893 char *buf)
894{
895 struct dql *dql = &queue->dql;
896
897 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
898}
899
900static ssize_t bql_set_hold_time(struct netdev_queue *queue,
901 struct netdev_queue_attribute *attribute,
902 const char *buf, size_t len)
903{
904 struct dql *dql = &queue->dql;
905 unsigned int value;
906 int err;
907
908 err = kstrtouint(buf, 10, &value);
909 if (err < 0)
910 return err;
911
912 dql->slack_hold_time = msecs_to_jiffies(value);
913
914 return len;
915}
916
917static struct netdev_queue_attribute bql_hold_time_attribute =
918 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
919 bql_set_hold_time);
920
921static ssize_t bql_show_inflight(struct netdev_queue *queue,
922 struct netdev_queue_attribute *attr,
923 char *buf)
924{
925 struct dql *dql = &queue->dql;
926
927 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
928}
929
930static struct netdev_queue_attribute bql_inflight_attribute =
931 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
932
933#define BQL_ATTR(NAME, FIELD) \
934static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
935 struct netdev_queue_attribute *attr, \
936 char *buf) \
937{ \
938 return bql_show(buf, queue->dql.FIELD); \
939} \
940 \
941static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
942 struct netdev_queue_attribute *attr, \
943 const char *buf, size_t len) \
944{ \
945 return bql_set(buf, len, &queue->dql.FIELD); \
946} \
947 \
948static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
949 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
950 bql_set_ ## NAME);
951
952BQL_ATTR(limit, limit)
953BQL_ATTR(limit_max, max_limit)
954BQL_ATTR(limit_min, min_limit)
955
956static struct attribute *dql_attrs[] = {
957 &bql_limit_attribute.attr,
958 &bql_limit_max_attribute.attr,
959 &bql_limit_min_attribute.attr,
960 &bql_hold_time_attribute.attr,
961 &bql_inflight_attribute.attr,
962 NULL
963};
964
965static struct attribute_group dql_group = {
966 .name = "byte_queue_limits",
967 .attrs = dql_attrs,
968};
969#endif /* CONFIG_BQL */
970
971#ifdef CONFIG_XPS
972static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
973{
974 struct net_device *dev = queue->dev;
975 int i;
976
977 for (i = 0; i < dev->num_tx_queues; i++)
978 if (queue == &dev->_tx[i])
979 break;
980
981 BUG_ON(i >= dev->num_tx_queues);
982
983 return i;
984}
985
986
987static ssize_t show_xps_map(struct netdev_queue *queue,
988 struct netdev_queue_attribute *attribute, char *buf)
989{
990 struct net_device *dev = queue->dev;
991 struct xps_dev_maps *dev_maps;
992 cpumask_var_t mask;
993 unsigned long index;
994 size_t len = 0;
995 int i;
996
997 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
998 return -ENOMEM;
999
1000 index = get_netdev_queue_index(queue);
1001
1002 rcu_read_lock();
1003 dev_maps = rcu_dereference(dev->xps_maps);
1004 if (dev_maps) {
1005 for_each_possible_cpu(i) {
1006 struct xps_map *map =
1007 rcu_dereference(dev_maps->cpu_map[i]);
1008 if (map) {
1009 int j;
1010 for (j = 0; j < map->len; j++) {
1011 if (map->queues[j] == index) {
1012 cpumask_set_cpu(i, mask);
1013 break;
1014 }
1015 }
1016 }
1017 }
1018 }
1019 rcu_read_unlock();
1020
1021 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1022 if (PAGE_SIZE - len < 3) {
1023 free_cpumask_var(mask);
1024 return -EINVAL;
1025 }
1026
1027 free_cpumask_var(mask);
1028 len += sprintf(buf + len, "\n");
1029 return len;
1030}
1031
1032static DEFINE_MUTEX(xps_map_mutex);
1033#define xmap_dereference(P) \
1034 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1035
1036static void xps_queue_release(struct netdev_queue *queue)
1037{
1038 struct net_device *dev = queue->dev;
1039 struct xps_dev_maps *dev_maps;
1040 struct xps_map *map;
1041 unsigned long index;
1042 int i, pos, nonempty = 0;
1043
1044 index = get_netdev_queue_index(queue);
1045
1046 mutex_lock(&xps_map_mutex);
1047 dev_maps = xmap_dereference(dev->xps_maps);
1048
1049 if (dev_maps) {
1050 for_each_possible_cpu(i) {
1051 map = xmap_dereference(dev_maps->cpu_map[i]);
1052 if (!map)
1053 continue;
1054
1055 for (pos = 0; pos < map->len; pos++)
1056 if (map->queues[pos] == index)
1057 break;
1058
1059 if (pos < map->len) {
1060 if (map->len > 1)
1061 map->queues[pos] =
1062 map->queues[--map->len];
1063 else {
1064 RCU_INIT_POINTER(dev_maps->cpu_map[i],
1065 NULL);
1066 kfree_rcu(map, rcu);
1067 map = NULL;
1068 }
1069 }
1070 if (map)
1071 nonempty = 1;
1072 }
1073
1074 if (!nonempty) {
1075 RCU_INIT_POINTER(dev->xps_maps, NULL);
1076 kfree_rcu(dev_maps, rcu);
1077 }
1078 }
1079 mutex_unlock(&xps_map_mutex);
1080}
1081
1082static ssize_t store_xps_map(struct netdev_queue *queue,
1083 struct netdev_queue_attribute *attribute,
1084 const char *buf, size_t len)
1085{
1086 struct net_device *dev = queue->dev;
1087 cpumask_var_t mask;
1088 int err, i, cpu, pos, map_len, alloc_len, need_set;
1089 unsigned long index;
1090 struct xps_map *map, *new_map;
1091 struct xps_dev_maps *dev_maps, *new_dev_maps;
1092 int nonempty = 0;
1093 int numa_node_id = -2;
1094
1095 if (!capable(CAP_NET_ADMIN))
1096 return -EPERM;
1097
1098 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1099 return -ENOMEM;
1100
1101 index = get_netdev_queue_index(queue);
1102
1103 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1104 if (err) {
1105 free_cpumask_var(mask);
1106 return err;
1107 }
1108
1109 new_dev_maps = kzalloc(max_t(unsigned int,
1110 XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1111 if (!new_dev_maps) {
1112 free_cpumask_var(mask);
1113 return -ENOMEM;
1114 }
1115
1116 mutex_lock(&xps_map_mutex);
1117
1118 dev_maps = xmap_dereference(dev->xps_maps);
1119
1120 for_each_possible_cpu(cpu) {
1121 map = dev_maps ?
1122 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1123 new_map = map;
1124 if (map) {
1125 for (pos = 0; pos < map->len; pos++)
1126 if (map->queues[pos] == index)
1127 break;
1128 map_len = map->len;
1129 alloc_len = map->alloc_len;
1130 } else
1131 pos = map_len = alloc_len = 0;
1132
1133 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1134#ifdef CONFIG_NUMA
1135 if (need_set) {
1136 if (numa_node_id == -2)
1137 numa_node_id = cpu_to_node(cpu);
1138 else if (numa_node_id != cpu_to_node(cpu))
1139 numa_node_id = -1;
1140 }
1141#endif
1142 if (need_set && pos >= map_len) {
1143 /* Need to add queue to this CPU's map */
1144 if (map_len >= alloc_len) {
1145 alloc_len = alloc_len ?
1146 2 * alloc_len : XPS_MIN_MAP_ALLOC;
1147 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1148 GFP_KERNEL,
1149 cpu_to_node(cpu));
1150 if (!new_map)
1151 goto error;
1152 new_map->alloc_len = alloc_len;
1153 for (i = 0; i < map_len; i++)
1154 new_map->queues[i] = map->queues[i];
1155 new_map->len = map_len;
1156 }
1157 new_map->queues[new_map->len++] = index;
1158 } else if (!need_set && pos < map_len) {
1159 /* Need to remove queue from this CPU's map */
1160 if (map_len > 1)
1161 new_map->queues[pos] =
1162 new_map->queues[--new_map->len];
1163 else
1164 new_map = NULL;
1165 }
1166 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1167 }
1168
1169 /* Cleanup old maps */
1170 for_each_possible_cpu(cpu) {
1171 map = dev_maps ?
1172 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1173 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1174 kfree_rcu(map, rcu);
1175 if (new_dev_maps->cpu_map[cpu])
1176 nonempty = 1;
1177 }
1178
1179 if (nonempty) {
1180 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1181 } else {
1182 kfree(new_dev_maps);
1183 RCU_INIT_POINTER(dev->xps_maps, NULL);
1184 }
1185
1186 if (dev_maps)
1187 kfree_rcu(dev_maps, rcu);
1188
1189 netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1190 NUMA_NO_NODE);
1191
1192 mutex_unlock(&xps_map_mutex);
1193
1194 free_cpumask_var(mask);
1195 return len;
1196
1197error:
1198 mutex_unlock(&xps_map_mutex);
1199
1200 if (new_dev_maps)
1201 for_each_possible_cpu(i)
1202 kfree(rcu_dereference_protected(
1203 new_dev_maps->cpu_map[i],
1204 1));
1205 kfree(new_dev_maps);
1206 free_cpumask_var(mask);
1207 return -ENOMEM;
1208}
1209
1210static struct netdev_queue_attribute xps_cpus_attribute =
1211 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1212#endif /* CONFIG_XPS */
1213
1214static struct attribute *netdev_queue_default_attrs[] = {
1215 &queue_trans_timeout.attr,
1216#ifdef CONFIG_XPS
1217 &xps_cpus_attribute.attr,
1218#endif
1219 NULL
1220};
1221
1222static void netdev_queue_release(struct kobject *kobj)
1223{
1224 struct netdev_queue *queue = to_netdev_queue(kobj);
1225
1226#ifdef CONFIG_XPS
1227 xps_queue_release(queue);
1228#endif
1229
1230 memset(kobj, 0, sizeof(*kobj));
1231 dev_put(queue->dev);
1232}
1233
1234static struct kobj_type netdev_queue_ktype = {
1235 .sysfs_ops = &netdev_queue_sysfs_ops,
1236 .release = netdev_queue_release,
1237 .default_attrs = netdev_queue_default_attrs,
1238};
1239
1240static int netdev_queue_add_kobject(struct net_device *net, int index)
1241{
1242 struct netdev_queue *queue = net->_tx + index;
1243 struct kobject *kobj = &queue->kobj;
1244 int error = 0;
1245
1246 kobj->kset = net->queues_kset;
1247 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1248 "tx-%u", index);
1249 if (error)
1250 goto exit;
1251
1252#ifdef CONFIG_BQL
1253 error = sysfs_create_group(kobj, &dql_group);
1254 if (error)
1255 goto exit;
1256#endif
1257
1258 kobject_uevent(kobj, KOBJ_ADD);
1259 dev_hold(queue->dev);
1260
1261 return 0;
1262exit:
1263 kobject_put(kobj);
1264 return error;
1265}
1266#endif /* CONFIG_SYSFS */
1267
1268int
1269netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1270{
1271#ifdef CONFIG_SYSFS
1272 int i;
1273 int error = 0;
1274
1275 for (i = old_num; i < new_num; i++) {
1276 error = netdev_queue_add_kobject(net, i);
1277 if (error) {
1278 new_num = old_num;
1279 break;
1280 }
1281 }
1282
1283 while (--i >= new_num) {
1284 struct netdev_queue *queue = net->_tx + i;
1285
1286#ifdef CONFIG_BQL
1287 sysfs_remove_group(&queue->kobj, &dql_group);
1288#endif
1289 kobject_put(&queue->kobj);
1290 }
1291
1292 return error;
1293#else
1294 return 0;
1295#endif /* CONFIG_SYSFS */
1296}
1297
1298static int register_queue_kobjects(struct net_device *net)
1299{
1300 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1301
1302#ifdef CONFIG_SYSFS
1303 net->queues_kset = kset_create_and_add("queues",
1304 NULL, &net->dev.kobj);
1305 if (!net->queues_kset)
1306 return -ENOMEM;
1307#endif
1308
1309#ifdef CONFIG_RPS
1310 real_rx = net->real_num_rx_queues;
1311#endif
1312 real_tx = net->real_num_tx_queues;
1313
1314 error = net_rx_queue_update_kobjects(net, 0, real_rx);
1315 if (error)
1316 goto error;
1317 rxq = real_rx;
1318
1319 error = netdev_queue_update_kobjects(net, 0, real_tx);
1320 if (error)
1321 goto error;
1322 txq = real_tx;
1323
1324 return 0;
1325
1326error:
1327 netdev_queue_update_kobjects(net, txq, 0);
1328 net_rx_queue_update_kobjects(net, rxq, 0);
1329 return error;
1330}
1331
1332static void remove_queue_kobjects(struct net_device *net)
1333{
1334 int real_rx = 0, real_tx = 0;
1335
1336#ifdef CONFIG_RPS
1337 real_rx = net->real_num_rx_queues;
1338#endif
1339 real_tx = net->real_num_tx_queues;
1340
1341 net_rx_queue_update_kobjects(net, real_rx, 0);
1342 netdev_queue_update_kobjects(net, real_tx, 0);
1343#ifdef CONFIG_SYSFS
1344 kset_unregister(net->queues_kset);
1345#endif
1346}
1347
1348static void *net_grab_current_ns(void)
1349{
1350 struct net *ns = current->nsproxy->net_ns;
1351#ifdef CONFIG_NET_NS
1352 if (ns)
1353 atomic_inc(&ns->passive);
1354#endif
1355 return ns;
1356}
1357
1358static const void *net_initial_ns(void)
1359{
1360 return &init_net;
1361}
1362
1363static const void *net_netlink_ns(struct sock *sk)
1364{
1365 return sock_net(sk);
1366}
1367
1368struct kobj_ns_type_operations net_ns_type_operations = {
1369 .type = KOBJ_NS_TYPE_NET,
1370 .grab_current_ns = net_grab_current_ns,
1371 .netlink_ns = net_netlink_ns,
1372 .initial_ns = net_initial_ns,
1373 .drop_ns = net_drop_ns,
1374};
1375EXPORT_SYMBOL_GPL(net_ns_type_operations);
1376
1377#ifdef CONFIG_HOTPLUG
1378static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1379{
1380 struct net_device *dev = to_net_dev(d);
1381 int retval;
1382
1383 /* pass interface to uevent. */
1384 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1385 if (retval)
1386 goto exit;
1387
1388 /* pass ifindex to uevent.
1389 * ifindex is useful as it won't change (interface name may change)
1390 * and is what RtNetlink uses natively. */
1391 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1392
1393exit:
1394 return retval;
1395}
1396#endif
1397
1398/*
1399 * netdev_release -- destroy and free a dead device.
1400 * Called when last reference to device kobject is gone.
1401 */
1402static void netdev_release(struct device *d)
1403{
1404 struct net_device *dev = to_net_dev(d);
1405
1406 BUG_ON(dev->reg_state != NETREG_RELEASED);
1407
1408 kfree(dev->ifalias);
1409 kfree((char *)dev - dev->padded);
1410}
1411
1412static const void *net_namespace(struct device *d)
1413{
1414 struct net_device *dev;
1415 dev = container_of(d, struct net_device, dev);
1416 return dev_net(dev);
1417}
1418
1419static struct class net_class = {
1420 .name = "net",
1421 .dev_release = netdev_release,
1422#ifdef CONFIG_SYSFS
1423 .dev_attrs = net_class_attributes,
1424#endif /* CONFIG_SYSFS */
1425#ifdef CONFIG_HOTPLUG
1426 .dev_uevent = netdev_uevent,
1427#endif
1428 .ns_type = &net_ns_type_operations,
1429 .namespace = net_namespace,
1430};
1431
1432/* Delete sysfs entries but hold kobject reference until after all
1433 * netdev references are gone.
1434 */
1435void netdev_unregister_kobject(struct net_device * net)
1436{
1437 struct device *dev = &(net->dev);
1438
1439 kobject_get(&dev->kobj);
1440
1441 remove_queue_kobjects(net);
1442
1443 device_del(dev);
1444}
1445
1446/* Create sysfs entries for network device. */
1447int netdev_register_kobject(struct net_device *net)
1448{
1449 struct device *dev = &(net->dev);
1450 const struct attribute_group **groups = net->sysfs_groups;
1451 int error = 0;
1452
1453 device_initialize(dev);
1454 dev->class = &net_class;
1455 dev->platform_data = net;
1456 dev->groups = groups;
1457
1458 dev_set_name(dev, "%s", net->name);
1459
1460#ifdef CONFIG_SYSFS
1461 /* Allow for a device specific group */
1462 if (*groups)
1463 groups++;
1464
1465 *groups++ = &netstat_group;
1466#ifdef CONFIG_WIRELESS_EXT_SYSFS
1467 if (net->ieee80211_ptr)
1468 *groups++ = &wireless_group;
1469#ifdef CONFIG_WIRELESS_EXT
1470 else if (net->wireless_handlers)
1471 *groups++ = &wireless_group;
1472#endif
1473#endif
1474#endif /* CONFIG_SYSFS */
1475
1476 error = device_add(dev);
1477 if (error)
1478 return error;
1479
1480 error = register_queue_kobjects(net);
1481 if (error) {
1482 device_del(dev);
1483 return error;
1484 }
1485
1486 return error;
1487}
1488
1489int netdev_class_create_file(struct class_attribute *class_attr)
1490{
1491 return class_create_file(&net_class, class_attr);
1492}
1493EXPORT_SYMBOL(netdev_class_create_file);
1494
1495void netdev_class_remove_file(struct class_attribute *class_attr)
1496{
1497 class_remove_file(&net_class, class_attr);
1498}
1499EXPORT_SYMBOL(netdev_class_remove_file);
1500
1501int netdev_kobject_init(void)
1502{
1503 kobj_ns_type_register(&net_ns_type_operations);
1504 return class_register(&net_class);
1505}
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/sched/signal.h>
19#include <linux/nsproxy.h>
20#include <net/sock.h>
21#include <net/net_namespace.h>
22#include <linux/rtnetlink.h>
23#include <linux/vmalloc.h>
24#include <linux/export.h>
25#include <linux/jiffies.h>
26#include <linux/pm_runtime.h>
27#include <linux/of.h>
28#include <linux/of_net.h>
29
30#include "net-sysfs.h"
31
32#ifdef CONFIG_SYSFS
33static const char fmt_hex[] = "%#x\n";
34static const char fmt_dec[] = "%d\n";
35static const char fmt_ulong[] = "%lu\n";
36static const char fmt_u64[] = "%llu\n";
37
38static 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 */
44static 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) \
61static ssize_t format_##field(const struct net_device *dev, char *buf) \
62{ \
63 return sprintf(buf, format_string, dev->field); \
64} \
65static 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) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RO(field)
74
75#define NETDEVICE_SHOW_RW(field, format_string) \
76NETDEVICE_SHOW(field, format_string); \
77static DEVICE_ATTR_RW(field)
78
79/* use same locking and permission rules as SIF* ioctl's */
80static 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 ret = (*set)(netdev, new);
101 if (ret == 0)
102 ret = len;
103 }
104 rtnl_unlock();
105 err:
106 return ret;
107}
108
109NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114NETDEVICE_SHOW_RO(type, fmt_dec);
115NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116
117static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 char *buf)
119{
120 struct net_device *ndev = to_net_dev(dev);
121
122 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
123}
124static DEVICE_ATTR_RO(iflink);
125
126static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127{
128 return sprintf(buf, fmt_dec, dev->name_assign_type);
129}
130
131static ssize_t name_assign_type_show(struct device *dev,
132 struct device_attribute *attr,
133 char *buf)
134{
135 struct net_device *ndev = to_net_dev(dev);
136 ssize_t ret = -EINVAL;
137
138 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140
141 return ret;
142}
143static DEVICE_ATTR_RO(name_assign_type);
144
145/* use same locking rules as GIFHWADDR ioctl's */
146static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 char *buf)
148{
149 struct net_device *ndev = to_net_dev(dev);
150 ssize_t ret = -EINVAL;
151
152 read_lock(&dev_base_lock);
153 if (dev_isalive(ndev))
154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155 read_unlock(&dev_base_lock);
156 return ret;
157}
158static DEVICE_ATTR_RO(address);
159
160static ssize_t broadcast_show(struct device *dev,
161 struct device_attribute *attr, char *buf)
162{
163 struct net_device *ndev = to_net_dev(dev);
164
165 if (dev_isalive(ndev))
166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167 return -EINVAL;
168}
169static DEVICE_ATTR_RO(broadcast);
170
171static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172{
173 if (!netif_running(dev))
174 return -EINVAL;
175 return dev_change_carrier(dev, (bool)new_carrier);
176}
177
178static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179 const char *buf, size_t len)
180{
181 return netdev_store(dev, attr, buf, len, change_carrier);
182}
183
184static ssize_t carrier_show(struct device *dev,
185 struct device_attribute *attr, char *buf)
186{
187 struct net_device *netdev = to_net_dev(dev);
188
189 if (netif_running(netdev))
190 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
191
192 return -EINVAL;
193}
194static DEVICE_ATTR_RW(carrier);
195
196static ssize_t speed_show(struct device *dev,
197 struct device_attribute *attr, char *buf)
198{
199 struct net_device *netdev = to_net_dev(dev);
200 int ret = -EINVAL;
201
202 if (!rtnl_trylock())
203 return restart_syscall();
204
205 if (netif_running(netdev)) {
206 struct ethtool_link_ksettings cmd;
207
208 if (!__ethtool_get_link_ksettings(netdev, &cmd))
209 ret = sprintf(buf, fmt_dec, cmd.base.speed);
210 }
211 rtnl_unlock();
212 return ret;
213}
214static DEVICE_ATTR_RO(speed);
215
216static ssize_t duplex_show(struct device *dev,
217 struct device_attribute *attr, char *buf)
218{
219 struct net_device *netdev = to_net_dev(dev);
220 int ret = -EINVAL;
221
222 if (!rtnl_trylock())
223 return restart_syscall();
224
225 if (netif_running(netdev)) {
226 struct ethtool_link_ksettings cmd;
227
228 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
229 const char *duplex;
230
231 switch (cmd.base.duplex) {
232 case DUPLEX_HALF:
233 duplex = "half";
234 break;
235 case DUPLEX_FULL:
236 duplex = "full";
237 break;
238 default:
239 duplex = "unknown";
240 break;
241 }
242 ret = sprintf(buf, "%s\n", duplex);
243 }
244 }
245 rtnl_unlock();
246 return ret;
247}
248static DEVICE_ATTR_RO(duplex);
249
250static ssize_t dormant_show(struct device *dev,
251 struct device_attribute *attr, char *buf)
252{
253 struct net_device *netdev = to_net_dev(dev);
254
255 if (netif_running(netdev))
256 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
257
258 return -EINVAL;
259}
260static DEVICE_ATTR_RO(dormant);
261
262static const char *const operstates[] = {
263 "unknown",
264 "notpresent", /* currently unused */
265 "down",
266 "lowerlayerdown",
267 "testing", /* currently unused */
268 "dormant",
269 "up"
270};
271
272static ssize_t operstate_show(struct device *dev,
273 struct device_attribute *attr, char *buf)
274{
275 const struct net_device *netdev = to_net_dev(dev);
276 unsigned char operstate;
277
278 read_lock(&dev_base_lock);
279 operstate = netdev->operstate;
280 if (!netif_running(netdev))
281 operstate = IF_OPER_DOWN;
282 read_unlock(&dev_base_lock);
283
284 if (operstate >= ARRAY_SIZE(operstates))
285 return -EINVAL; /* should not happen */
286
287 return sprintf(buf, "%s\n", operstates[operstate]);
288}
289static DEVICE_ATTR_RO(operstate);
290
291static ssize_t carrier_changes_show(struct device *dev,
292 struct device_attribute *attr,
293 char *buf)
294{
295 struct net_device *netdev = to_net_dev(dev);
296
297 return sprintf(buf, fmt_dec,
298 atomic_read(&netdev->carrier_up_count) +
299 atomic_read(&netdev->carrier_down_count));
300}
301static DEVICE_ATTR_RO(carrier_changes);
302
303static ssize_t carrier_up_count_show(struct device *dev,
304 struct device_attribute *attr,
305 char *buf)
306{
307 struct net_device *netdev = to_net_dev(dev);
308
309 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
310}
311static DEVICE_ATTR_RO(carrier_up_count);
312
313static ssize_t carrier_down_count_show(struct device *dev,
314 struct device_attribute *attr,
315 char *buf)
316{
317 struct net_device *netdev = to_net_dev(dev);
318
319 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
320}
321static DEVICE_ATTR_RO(carrier_down_count);
322
323/* read-write attributes */
324
325static int change_mtu(struct net_device *dev, unsigned long new_mtu)
326{
327 return dev_set_mtu(dev, (int)new_mtu);
328}
329
330static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
331 const char *buf, size_t len)
332{
333 return netdev_store(dev, attr, buf, len, change_mtu);
334}
335NETDEVICE_SHOW_RW(mtu, fmt_dec);
336
337static int change_flags(struct net_device *dev, unsigned long new_flags)
338{
339 return dev_change_flags(dev, (unsigned int)new_flags);
340}
341
342static ssize_t flags_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_flags);
346}
347NETDEVICE_SHOW_RW(flags, fmt_hex);
348
349static ssize_t tx_queue_len_store(struct device *dev,
350 struct device_attribute *attr,
351 const char *buf, size_t len)
352{
353 if (!capable(CAP_NET_ADMIN))
354 return -EPERM;
355
356 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
357}
358NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
359
360static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
361{
362 dev->gro_flush_timeout = val;
363 return 0;
364}
365
366static ssize_t gro_flush_timeout_store(struct device *dev,
367 struct device_attribute *attr,
368 const char *buf, size_t len)
369{
370 if (!capable(CAP_NET_ADMIN))
371 return -EPERM;
372
373 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
374}
375NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
376
377static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
378 const char *buf, size_t len)
379{
380 struct net_device *netdev = to_net_dev(dev);
381 struct net *net = dev_net(netdev);
382 size_t count = len;
383 ssize_t ret = 0;
384
385 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
386 return -EPERM;
387
388 /* ignore trailing newline */
389 if (len > 0 && buf[len - 1] == '\n')
390 --count;
391
392 if (!rtnl_trylock())
393 return restart_syscall();
394
395 if (dev_isalive(netdev)) {
396 ret = dev_set_alias(netdev, buf, count);
397 if (ret < 0)
398 goto err;
399 ret = len;
400 netdev_state_change(netdev);
401 }
402err:
403 rtnl_unlock();
404
405 return ret;
406}
407
408static ssize_t ifalias_show(struct device *dev,
409 struct device_attribute *attr, char *buf)
410{
411 const struct net_device *netdev = to_net_dev(dev);
412 char tmp[IFALIASZ];
413 ssize_t ret = 0;
414
415 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
416 if (ret > 0)
417 ret = sprintf(buf, "%s\n", tmp);
418 return ret;
419}
420static DEVICE_ATTR_RW(ifalias);
421
422static int change_group(struct net_device *dev, unsigned long new_group)
423{
424 dev_set_group(dev, (int)new_group);
425 return 0;
426}
427
428static ssize_t group_store(struct device *dev, struct device_attribute *attr,
429 const char *buf, size_t len)
430{
431 return netdev_store(dev, attr, buf, len, change_group);
432}
433NETDEVICE_SHOW(group, fmt_dec);
434static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
435
436static int change_proto_down(struct net_device *dev, unsigned long proto_down)
437{
438 return dev_change_proto_down(dev, (bool)proto_down);
439}
440
441static ssize_t proto_down_store(struct device *dev,
442 struct device_attribute *attr,
443 const char *buf, size_t len)
444{
445 return netdev_store(dev, attr, buf, len, change_proto_down);
446}
447NETDEVICE_SHOW_RW(proto_down, fmt_dec);
448
449static ssize_t phys_port_id_show(struct device *dev,
450 struct device_attribute *attr, char *buf)
451{
452 struct net_device *netdev = to_net_dev(dev);
453 ssize_t ret = -EINVAL;
454
455 if (!rtnl_trylock())
456 return restart_syscall();
457
458 if (dev_isalive(netdev)) {
459 struct netdev_phys_item_id ppid;
460
461 ret = dev_get_phys_port_id(netdev, &ppid);
462 if (!ret)
463 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
464 }
465 rtnl_unlock();
466
467 return ret;
468}
469static DEVICE_ATTR_RO(phys_port_id);
470
471static ssize_t phys_port_name_show(struct device *dev,
472 struct device_attribute *attr, char *buf)
473{
474 struct net_device *netdev = to_net_dev(dev);
475 ssize_t ret = -EINVAL;
476
477 if (!rtnl_trylock())
478 return restart_syscall();
479
480 if (dev_isalive(netdev)) {
481 char name[IFNAMSIZ];
482
483 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
484 if (!ret)
485 ret = sprintf(buf, "%s\n", name);
486 }
487 rtnl_unlock();
488
489 return ret;
490}
491static DEVICE_ATTR_RO(phys_port_name);
492
493static ssize_t phys_switch_id_show(struct device *dev,
494 struct device_attribute *attr, char *buf)
495{
496 struct net_device *netdev = to_net_dev(dev);
497 ssize_t ret = -EINVAL;
498
499 if (!rtnl_trylock())
500 return restart_syscall();
501
502 if (dev_isalive(netdev)) {
503 struct switchdev_attr attr = {
504 .orig_dev = netdev,
505 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
506 .flags = SWITCHDEV_F_NO_RECURSE,
507 };
508
509 ret = switchdev_port_attr_get(netdev, &attr);
510 if (!ret)
511 ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
512 attr.u.ppid.id);
513 }
514 rtnl_unlock();
515
516 return ret;
517}
518static DEVICE_ATTR_RO(phys_switch_id);
519
520static struct attribute *net_class_attrs[] __ro_after_init = {
521 &dev_attr_netdev_group.attr,
522 &dev_attr_type.attr,
523 &dev_attr_dev_id.attr,
524 &dev_attr_dev_port.attr,
525 &dev_attr_iflink.attr,
526 &dev_attr_ifindex.attr,
527 &dev_attr_name_assign_type.attr,
528 &dev_attr_addr_assign_type.attr,
529 &dev_attr_addr_len.attr,
530 &dev_attr_link_mode.attr,
531 &dev_attr_address.attr,
532 &dev_attr_broadcast.attr,
533 &dev_attr_speed.attr,
534 &dev_attr_duplex.attr,
535 &dev_attr_dormant.attr,
536 &dev_attr_operstate.attr,
537 &dev_attr_carrier_changes.attr,
538 &dev_attr_ifalias.attr,
539 &dev_attr_carrier.attr,
540 &dev_attr_mtu.attr,
541 &dev_attr_flags.attr,
542 &dev_attr_tx_queue_len.attr,
543 &dev_attr_gro_flush_timeout.attr,
544 &dev_attr_phys_port_id.attr,
545 &dev_attr_phys_port_name.attr,
546 &dev_attr_phys_switch_id.attr,
547 &dev_attr_proto_down.attr,
548 &dev_attr_carrier_up_count.attr,
549 &dev_attr_carrier_down_count.attr,
550 NULL,
551};
552ATTRIBUTE_GROUPS(net_class);
553
554/* Show a given an attribute in the statistics group */
555static ssize_t netstat_show(const struct device *d,
556 struct device_attribute *attr, char *buf,
557 unsigned long offset)
558{
559 struct net_device *dev = to_net_dev(d);
560 ssize_t ret = -EINVAL;
561
562 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
563 offset % sizeof(u64) != 0);
564
565 read_lock(&dev_base_lock);
566 if (dev_isalive(dev)) {
567 struct rtnl_link_stats64 temp;
568 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
569
570 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
571 }
572 read_unlock(&dev_base_lock);
573 return ret;
574}
575
576/* generate a read-only statistics attribute */
577#define NETSTAT_ENTRY(name) \
578static ssize_t name##_show(struct device *d, \
579 struct device_attribute *attr, char *buf) \
580{ \
581 return netstat_show(d, attr, buf, \
582 offsetof(struct rtnl_link_stats64, name)); \
583} \
584static DEVICE_ATTR_RO(name)
585
586NETSTAT_ENTRY(rx_packets);
587NETSTAT_ENTRY(tx_packets);
588NETSTAT_ENTRY(rx_bytes);
589NETSTAT_ENTRY(tx_bytes);
590NETSTAT_ENTRY(rx_errors);
591NETSTAT_ENTRY(tx_errors);
592NETSTAT_ENTRY(rx_dropped);
593NETSTAT_ENTRY(tx_dropped);
594NETSTAT_ENTRY(multicast);
595NETSTAT_ENTRY(collisions);
596NETSTAT_ENTRY(rx_length_errors);
597NETSTAT_ENTRY(rx_over_errors);
598NETSTAT_ENTRY(rx_crc_errors);
599NETSTAT_ENTRY(rx_frame_errors);
600NETSTAT_ENTRY(rx_fifo_errors);
601NETSTAT_ENTRY(rx_missed_errors);
602NETSTAT_ENTRY(tx_aborted_errors);
603NETSTAT_ENTRY(tx_carrier_errors);
604NETSTAT_ENTRY(tx_fifo_errors);
605NETSTAT_ENTRY(tx_heartbeat_errors);
606NETSTAT_ENTRY(tx_window_errors);
607NETSTAT_ENTRY(rx_compressed);
608NETSTAT_ENTRY(tx_compressed);
609NETSTAT_ENTRY(rx_nohandler);
610
611static struct attribute *netstat_attrs[] __ro_after_init = {
612 &dev_attr_rx_packets.attr,
613 &dev_attr_tx_packets.attr,
614 &dev_attr_rx_bytes.attr,
615 &dev_attr_tx_bytes.attr,
616 &dev_attr_rx_errors.attr,
617 &dev_attr_tx_errors.attr,
618 &dev_attr_rx_dropped.attr,
619 &dev_attr_tx_dropped.attr,
620 &dev_attr_multicast.attr,
621 &dev_attr_collisions.attr,
622 &dev_attr_rx_length_errors.attr,
623 &dev_attr_rx_over_errors.attr,
624 &dev_attr_rx_crc_errors.attr,
625 &dev_attr_rx_frame_errors.attr,
626 &dev_attr_rx_fifo_errors.attr,
627 &dev_attr_rx_missed_errors.attr,
628 &dev_attr_tx_aborted_errors.attr,
629 &dev_attr_tx_carrier_errors.attr,
630 &dev_attr_tx_fifo_errors.attr,
631 &dev_attr_tx_heartbeat_errors.attr,
632 &dev_attr_tx_window_errors.attr,
633 &dev_attr_rx_compressed.attr,
634 &dev_attr_tx_compressed.attr,
635 &dev_attr_rx_nohandler.attr,
636 NULL
637};
638
639static const struct attribute_group netstat_group = {
640 .name = "statistics",
641 .attrs = netstat_attrs,
642};
643
644#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
645static struct attribute *wireless_attrs[] = {
646 NULL
647};
648
649static const struct attribute_group wireless_group = {
650 .name = "wireless",
651 .attrs = wireless_attrs,
652};
653#endif
654
655#else /* CONFIG_SYSFS */
656#define net_class_groups NULL
657#endif /* CONFIG_SYSFS */
658
659#ifdef CONFIG_SYSFS
660#define to_rx_queue_attr(_attr) \
661 container_of(_attr, struct rx_queue_attribute, attr)
662
663#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
664
665static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
666 char *buf)
667{
668 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
669 struct netdev_rx_queue *queue = to_rx_queue(kobj);
670
671 if (!attribute->show)
672 return -EIO;
673
674 return attribute->show(queue, buf);
675}
676
677static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
678 const char *buf, size_t count)
679{
680 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
681 struct netdev_rx_queue *queue = to_rx_queue(kobj);
682
683 if (!attribute->store)
684 return -EIO;
685
686 return attribute->store(queue, buf, count);
687}
688
689static const struct sysfs_ops rx_queue_sysfs_ops = {
690 .show = rx_queue_attr_show,
691 .store = rx_queue_attr_store,
692};
693
694#ifdef CONFIG_RPS
695static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
696{
697 struct rps_map *map;
698 cpumask_var_t mask;
699 int i, len;
700
701 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
702 return -ENOMEM;
703
704 rcu_read_lock();
705 map = rcu_dereference(queue->rps_map);
706 if (map)
707 for (i = 0; i < map->len; i++)
708 cpumask_set_cpu(map->cpus[i], mask);
709
710 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
711 rcu_read_unlock();
712 free_cpumask_var(mask);
713
714 return len < PAGE_SIZE ? len : -EINVAL;
715}
716
717static ssize_t store_rps_map(struct netdev_rx_queue *queue,
718 const char *buf, size_t len)
719{
720 struct rps_map *old_map, *map;
721 cpumask_var_t mask;
722 int err, cpu, i;
723 static DEFINE_MUTEX(rps_map_mutex);
724
725 if (!capable(CAP_NET_ADMIN))
726 return -EPERM;
727
728 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
729 return -ENOMEM;
730
731 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
732 if (err) {
733 free_cpumask_var(mask);
734 return err;
735 }
736
737 map = kzalloc(max_t(unsigned int,
738 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
739 GFP_KERNEL);
740 if (!map) {
741 free_cpumask_var(mask);
742 return -ENOMEM;
743 }
744
745 i = 0;
746 for_each_cpu_and(cpu, mask, cpu_online_mask)
747 map->cpus[i++] = cpu;
748
749 if (i) {
750 map->len = i;
751 } else {
752 kfree(map);
753 map = NULL;
754 }
755
756 mutex_lock(&rps_map_mutex);
757 old_map = rcu_dereference_protected(queue->rps_map,
758 mutex_is_locked(&rps_map_mutex));
759 rcu_assign_pointer(queue->rps_map, map);
760
761 if (map)
762 static_key_slow_inc(&rps_needed);
763 if (old_map)
764 static_key_slow_dec(&rps_needed);
765
766 mutex_unlock(&rps_map_mutex);
767
768 if (old_map)
769 kfree_rcu(old_map, rcu);
770
771 free_cpumask_var(mask);
772 return len;
773}
774
775static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
776 char *buf)
777{
778 struct rps_dev_flow_table *flow_table;
779 unsigned long val = 0;
780
781 rcu_read_lock();
782 flow_table = rcu_dereference(queue->rps_flow_table);
783 if (flow_table)
784 val = (unsigned long)flow_table->mask + 1;
785 rcu_read_unlock();
786
787 return sprintf(buf, "%lu\n", val);
788}
789
790static void rps_dev_flow_table_release(struct rcu_head *rcu)
791{
792 struct rps_dev_flow_table *table = container_of(rcu,
793 struct rps_dev_flow_table, rcu);
794 vfree(table);
795}
796
797static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
798 const char *buf, size_t len)
799{
800 unsigned long mask, count;
801 struct rps_dev_flow_table *table, *old_table;
802 static DEFINE_SPINLOCK(rps_dev_flow_lock);
803 int rc;
804
805 if (!capable(CAP_NET_ADMIN))
806 return -EPERM;
807
808 rc = kstrtoul(buf, 0, &count);
809 if (rc < 0)
810 return rc;
811
812 if (count) {
813 mask = count - 1;
814 /* mask = roundup_pow_of_two(count) - 1;
815 * without overflows...
816 */
817 while ((mask | (mask >> 1)) != mask)
818 mask |= (mask >> 1);
819 /* On 64 bit arches, must check mask fits in table->mask (u32),
820 * and on 32bit arches, must check
821 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
822 */
823#if BITS_PER_LONG > 32
824 if (mask > (unsigned long)(u32)mask)
825 return -EINVAL;
826#else
827 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
828 / sizeof(struct rps_dev_flow)) {
829 /* Enforce a limit to prevent overflow */
830 return -EINVAL;
831 }
832#endif
833 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
834 if (!table)
835 return -ENOMEM;
836
837 table->mask = mask;
838 for (count = 0; count <= mask; count++)
839 table->flows[count].cpu = RPS_NO_CPU;
840 } else {
841 table = NULL;
842 }
843
844 spin_lock(&rps_dev_flow_lock);
845 old_table = rcu_dereference_protected(queue->rps_flow_table,
846 lockdep_is_held(&rps_dev_flow_lock));
847 rcu_assign_pointer(queue->rps_flow_table, table);
848 spin_unlock(&rps_dev_flow_lock);
849
850 if (old_table)
851 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
852
853 return len;
854}
855
856static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
857 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
858
859static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
860 = __ATTR(rps_flow_cnt, 0644,
861 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
862#endif /* CONFIG_RPS */
863
864static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
865#ifdef CONFIG_RPS
866 &rps_cpus_attribute.attr,
867 &rps_dev_flow_table_cnt_attribute.attr,
868#endif
869 NULL
870};
871
872static void rx_queue_release(struct kobject *kobj)
873{
874 struct netdev_rx_queue *queue = to_rx_queue(kobj);
875#ifdef CONFIG_RPS
876 struct rps_map *map;
877 struct rps_dev_flow_table *flow_table;
878
879 map = rcu_dereference_protected(queue->rps_map, 1);
880 if (map) {
881 RCU_INIT_POINTER(queue->rps_map, NULL);
882 kfree_rcu(map, rcu);
883 }
884
885 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
886 if (flow_table) {
887 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
888 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
889 }
890#endif
891
892 memset(kobj, 0, sizeof(*kobj));
893 dev_put(queue->dev);
894}
895
896static const void *rx_queue_namespace(struct kobject *kobj)
897{
898 struct netdev_rx_queue *queue = to_rx_queue(kobj);
899 struct device *dev = &queue->dev->dev;
900 const void *ns = NULL;
901
902 if (dev->class && dev->class->ns_type)
903 ns = dev->class->namespace(dev);
904
905 return ns;
906}
907
908static struct kobj_type rx_queue_ktype __ro_after_init = {
909 .sysfs_ops = &rx_queue_sysfs_ops,
910 .release = rx_queue_release,
911 .default_attrs = rx_queue_default_attrs,
912 .namespace = rx_queue_namespace
913};
914
915static int rx_queue_add_kobject(struct net_device *dev, int index)
916{
917 struct netdev_rx_queue *queue = dev->_rx + index;
918 struct kobject *kobj = &queue->kobj;
919 int error = 0;
920
921 kobj->kset = dev->queues_kset;
922 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
923 "rx-%u", index);
924 if (error)
925 return error;
926
927 if (dev->sysfs_rx_queue_group) {
928 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
929 if (error) {
930 kobject_put(kobj);
931 return error;
932 }
933 }
934
935 kobject_uevent(kobj, KOBJ_ADD);
936 dev_hold(queue->dev);
937
938 return error;
939}
940#endif /* CONFIG_SYSFS */
941
942int
943net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
944{
945#ifdef CONFIG_SYSFS
946 int i;
947 int error = 0;
948
949#ifndef CONFIG_RPS
950 if (!dev->sysfs_rx_queue_group)
951 return 0;
952#endif
953 for (i = old_num; i < new_num; i++) {
954 error = rx_queue_add_kobject(dev, i);
955 if (error) {
956 new_num = old_num;
957 break;
958 }
959 }
960
961 while (--i >= new_num) {
962 struct kobject *kobj = &dev->_rx[i].kobj;
963
964 if (!refcount_read(&dev_net(dev)->count))
965 kobj->uevent_suppress = 1;
966 if (dev->sysfs_rx_queue_group)
967 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
968 kobject_put(kobj);
969 }
970
971 return error;
972#else
973 return 0;
974#endif
975}
976
977#ifdef CONFIG_SYSFS
978/*
979 * netdev_queue sysfs structures and functions.
980 */
981struct netdev_queue_attribute {
982 struct attribute attr;
983 ssize_t (*show)(struct netdev_queue *queue, char *buf);
984 ssize_t (*store)(struct netdev_queue *queue,
985 const char *buf, size_t len);
986};
987#define to_netdev_queue_attr(_attr) \
988 container_of(_attr, struct netdev_queue_attribute, attr)
989
990#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
991
992static ssize_t netdev_queue_attr_show(struct kobject *kobj,
993 struct attribute *attr, char *buf)
994{
995 const struct netdev_queue_attribute *attribute
996 = to_netdev_queue_attr(attr);
997 struct netdev_queue *queue = to_netdev_queue(kobj);
998
999 if (!attribute->show)
1000 return -EIO;
1001
1002 return attribute->show(queue, buf);
1003}
1004
1005static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1006 struct attribute *attr,
1007 const char *buf, size_t count)
1008{
1009 const struct netdev_queue_attribute *attribute
1010 = to_netdev_queue_attr(attr);
1011 struct netdev_queue *queue = to_netdev_queue(kobj);
1012
1013 if (!attribute->store)
1014 return -EIO;
1015
1016 return attribute->store(queue, buf, count);
1017}
1018
1019static const struct sysfs_ops netdev_queue_sysfs_ops = {
1020 .show = netdev_queue_attr_show,
1021 .store = netdev_queue_attr_store,
1022};
1023
1024static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1025{
1026 unsigned long trans_timeout;
1027
1028 spin_lock_irq(&queue->_xmit_lock);
1029 trans_timeout = queue->trans_timeout;
1030 spin_unlock_irq(&queue->_xmit_lock);
1031
1032 return sprintf(buf, "%lu", trans_timeout);
1033}
1034
1035static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1036{
1037 struct net_device *dev = queue->dev;
1038 unsigned int i;
1039
1040 i = queue - dev->_tx;
1041 BUG_ON(i >= dev->num_tx_queues);
1042
1043 return i;
1044}
1045
1046static ssize_t traffic_class_show(struct netdev_queue *queue,
1047 char *buf)
1048{
1049 struct net_device *dev = queue->dev;
1050 int index = get_netdev_queue_index(queue);
1051 int tc = netdev_txq_to_tc(dev, index);
1052
1053 if (tc < 0)
1054 return -EINVAL;
1055
1056 return sprintf(buf, "%u\n", tc);
1057}
1058
1059#ifdef CONFIG_XPS
1060static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1061 char *buf)
1062{
1063 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1064}
1065
1066static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1067 const char *buf, size_t len)
1068{
1069 struct net_device *dev = queue->dev;
1070 int err, index = get_netdev_queue_index(queue);
1071 u32 rate = 0;
1072
1073 err = kstrtou32(buf, 10, &rate);
1074 if (err < 0)
1075 return err;
1076
1077 if (!rtnl_trylock())
1078 return restart_syscall();
1079
1080 err = -EOPNOTSUPP;
1081 if (dev->netdev_ops->ndo_set_tx_maxrate)
1082 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1083
1084 rtnl_unlock();
1085 if (!err) {
1086 queue->tx_maxrate = rate;
1087 return len;
1088 }
1089 return err;
1090}
1091
1092static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1093 = __ATTR_RW(tx_maxrate);
1094#endif
1095
1096static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1097 = __ATTR_RO(tx_timeout);
1098
1099static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1100 = __ATTR_RO(traffic_class);
1101
1102#ifdef CONFIG_BQL
1103/*
1104 * Byte queue limits sysfs structures and functions.
1105 */
1106static ssize_t bql_show(char *buf, unsigned int value)
1107{
1108 return sprintf(buf, "%u\n", value);
1109}
1110
1111static ssize_t bql_set(const char *buf, const size_t count,
1112 unsigned int *pvalue)
1113{
1114 unsigned int value;
1115 int err;
1116
1117 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1118 value = DQL_MAX_LIMIT;
1119 } else {
1120 err = kstrtouint(buf, 10, &value);
1121 if (err < 0)
1122 return err;
1123 if (value > DQL_MAX_LIMIT)
1124 return -EINVAL;
1125 }
1126
1127 *pvalue = value;
1128
1129 return count;
1130}
1131
1132static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1133 char *buf)
1134{
1135 struct dql *dql = &queue->dql;
1136
1137 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1138}
1139
1140static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1141 const char *buf, size_t len)
1142{
1143 struct dql *dql = &queue->dql;
1144 unsigned int value;
1145 int err;
1146
1147 err = kstrtouint(buf, 10, &value);
1148 if (err < 0)
1149 return err;
1150
1151 dql->slack_hold_time = msecs_to_jiffies(value);
1152
1153 return len;
1154}
1155
1156static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1157 = __ATTR(hold_time, 0644,
1158 bql_show_hold_time, bql_set_hold_time);
1159
1160static ssize_t bql_show_inflight(struct netdev_queue *queue,
1161 char *buf)
1162{
1163 struct dql *dql = &queue->dql;
1164
1165 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1166}
1167
1168static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1169 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1170
1171#define BQL_ATTR(NAME, FIELD) \
1172static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1173 char *buf) \
1174{ \
1175 return bql_show(buf, queue->dql.FIELD); \
1176} \
1177 \
1178static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1179 const char *buf, size_t len) \
1180{ \
1181 return bql_set(buf, len, &queue->dql.FIELD); \
1182} \
1183 \
1184static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1185 = __ATTR(NAME, 0644, \
1186 bql_show_ ## NAME, bql_set_ ## NAME)
1187
1188BQL_ATTR(limit, limit);
1189BQL_ATTR(limit_max, max_limit);
1190BQL_ATTR(limit_min, min_limit);
1191
1192static struct attribute *dql_attrs[] __ro_after_init = {
1193 &bql_limit_attribute.attr,
1194 &bql_limit_max_attribute.attr,
1195 &bql_limit_min_attribute.attr,
1196 &bql_hold_time_attribute.attr,
1197 &bql_inflight_attribute.attr,
1198 NULL
1199};
1200
1201static const struct attribute_group dql_group = {
1202 .name = "byte_queue_limits",
1203 .attrs = dql_attrs,
1204};
1205#endif /* CONFIG_BQL */
1206
1207#ifdef CONFIG_XPS
1208static ssize_t xps_cpus_show(struct netdev_queue *queue,
1209 char *buf)
1210{
1211 struct net_device *dev = queue->dev;
1212 int cpu, len, num_tc = 1, tc = 0;
1213 struct xps_dev_maps *dev_maps;
1214 cpumask_var_t mask;
1215 unsigned long index;
1216
1217 index = get_netdev_queue_index(queue);
1218
1219 if (dev->num_tc) {
1220 num_tc = dev->num_tc;
1221 tc = netdev_txq_to_tc(dev, index);
1222 if (tc < 0)
1223 return -EINVAL;
1224 }
1225
1226 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1227 return -ENOMEM;
1228
1229 rcu_read_lock();
1230 dev_maps = rcu_dereference(dev->xps_maps);
1231 if (dev_maps) {
1232 for_each_possible_cpu(cpu) {
1233 int i, tci = cpu * num_tc + tc;
1234 struct xps_map *map;
1235
1236 map = rcu_dereference(dev_maps->cpu_map[tci]);
1237 if (!map)
1238 continue;
1239
1240 for (i = map->len; i--;) {
1241 if (map->queues[i] == index) {
1242 cpumask_set_cpu(cpu, mask);
1243 break;
1244 }
1245 }
1246 }
1247 }
1248 rcu_read_unlock();
1249
1250 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1251 free_cpumask_var(mask);
1252 return len < PAGE_SIZE ? len : -EINVAL;
1253}
1254
1255static ssize_t xps_cpus_store(struct netdev_queue *queue,
1256 const char *buf, size_t len)
1257{
1258 struct net_device *dev = queue->dev;
1259 unsigned long index;
1260 cpumask_var_t mask;
1261 int err;
1262
1263 if (!capable(CAP_NET_ADMIN))
1264 return -EPERM;
1265
1266 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1267 return -ENOMEM;
1268
1269 index = get_netdev_queue_index(queue);
1270
1271 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1272 if (err) {
1273 free_cpumask_var(mask);
1274 return err;
1275 }
1276
1277 err = netif_set_xps_queue(dev, mask, index);
1278
1279 free_cpumask_var(mask);
1280
1281 return err ? : len;
1282}
1283
1284static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1285 = __ATTR_RW(xps_cpus);
1286#endif /* CONFIG_XPS */
1287
1288static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1289 &queue_trans_timeout.attr,
1290 &queue_traffic_class.attr,
1291#ifdef CONFIG_XPS
1292 &xps_cpus_attribute.attr,
1293 &queue_tx_maxrate.attr,
1294#endif
1295 NULL
1296};
1297
1298static void netdev_queue_release(struct kobject *kobj)
1299{
1300 struct netdev_queue *queue = to_netdev_queue(kobj);
1301
1302 memset(kobj, 0, sizeof(*kobj));
1303 dev_put(queue->dev);
1304}
1305
1306static const void *netdev_queue_namespace(struct kobject *kobj)
1307{
1308 struct netdev_queue *queue = to_netdev_queue(kobj);
1309 struct device *dev = &queue->dev->dev;
1310 const void *ns = NULL;
1311
1312 if (dev->class && dev->class->ns_type)
1313 ns = dev->class->namespace(dev);
1314
1315 return ns;
1316}
1317
1318static struct kobj_type netdev_queue_ktype __ro_after_init = {
1319 .sysfs_ops = &netdev_queue_sysfs_ops,
1320 .release = netdev_queue_release,
1321 .default_attrs = netdev_queue_default_attrs,
1322 .namespace = netdev_queue_namespace,
1323};
1324
1325static int netdev_queue_add_kobject(struct net_device *dev, int index)
1326{
1327 struct netdev_queue *queue = dev->_tx + index;
1328 struct kobject *kobj = &queue->kobj;
1329 int error = 0;
1330
1331 kobj->kset = dev->queues_kset;
1332 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1333 "tx-%u", index);
1334 if (error)
1335 return error;
1336
1337#ifdef CONFIG_BQL
1338 error = sysfs_create_group(kobj, &dql_group);
1339 if (error) {
1340 kobject_put(kobj);
1341 return error;
1342 }
1343#endif
1344
1345 kobject_uevent(kobj, KOBJ_ADD);
1346 dev_hold(queue->dev);
1347
1348 return 0;
1349}
1350#endif /* CONFIG_SYSFS */
1351
1352int
1353netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1354{
1355#ifdef CONFIG_SYSFS
1356 int i;
1357 int error = 0;
1358
1359 for (i = old_num; i < new_num; i++) {
1360 error = netdev_queue_add_kobject(dev, i);
1361 if (error) {
1362 new_num = old_num;
1363 break;
1364 }
1365 }
1366
1367 while (--i >= new_num) {
1368 struct netdev_queue *queue = dev->_tx + i;
1369
1370 if (!refcount_read(&dev_net(dev)->count))
1371 queue->kobj.uevent_suppress = 1;
1372#ifdef CONFIG_BQL
1373 sysfs_remove_group(&queue->kobj, &dql_group);
1374#endif
1375 kobject_put(&queue->kobj);
1376 }
1377
1378 return error;
1379#else
1380 return 0;
1381#endif /* CONFIG_SYSFS */
1382}
1383
1384static int register_queue_kobjects(struct net_device *dev)
1385{
1386 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1387
1388#ifdef CONFIG_SYSFS
1389 dev->queues_kset = kset_create_and_add("queues",
1390 NULL, &dev->dev.kobj);
1391 if (!dev->queues_kset)
1392 return -ENOMEM;
1393 real_rx = dev->real_num_rx_queues;
1394#endif
1395 real_tx = dev->real_num_tx_queues;
1396
1397 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1398 if (error)
1399 goto error;
1400 rxq = real_rx;
1401
1402 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1403 if (error)
1404 goto error;
1405 txq = real_tx;
1406
1407 return 0;
1408
1409error:
1410 netdev_queue_update_kobjects(dev, txq, 0);
1411 net_rx_queue_update_kobjects(dev, rxq, 0);
1412 return error;
1413}
1414
1415static void remove_queue_kobjects(struct net_device *dev)
1416{
1417 int real_rx = 0, real_tx = 0;
1418
1419#ifdef CONFIG_SYSFS
1420 real_rx = dev->real_num_rx_queues;
1421#endif
1422 real_tx = dev->real_num_tx_queues;
1423
1424 net_rx_queue_update_kobjects(dev, real_rx, 0);
1425 netdev_queue_update_kobjects(dev, real_tx, 0);
1426#ifdef CONFIG_SYSFS
1427 kset_unregister(dev->queues_kset);
1428#endif
1429}
1430
1431static bool net_current_may_mount(void)
1432{
1433 struct net *net = current->nsproxy->net_ns;
1434
1435 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1436}
1437
1438static void *net_grab_current_ns(void)
1439{
1440 struct net *ns = current->nsproxy->net_ns;
1441#ifdef CONFIG_NET_NS
1442 if (ns)
1443 refcount_inc(&ns->passive);
1444#endif
1445 return ns;
1446}
1447
1448static const void *net_initial_ns(void)
1449{
1450 return &init_net;
1451}
1452
1453static const void *net_netlink_ns(struct sock *sk)
1454{
1455 return sock_net(sk);
1456}
1457
1458const struct kobj_ns_type_operations net_ns_type_operations = {
1459 .type = KOBJ_NS_TYPE_NET,
1460 .current_may_mount = net_current_may_mount,
1461 .grab_current_ns = net_grab_current_ns,
1462 .netlink_ns = net_netlink_ns,
1463 .initial_ns = net_initial_ns,
1464 .drop_ns = net_drop_ns,
1465};
1466EXPORT_SYMBOL_GPL(net_ns_type_operations);
1467
1468static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1469{
1470 struct net_device *dev = to_net_dev(d);
1471 int retval;
1472
1473 /* pass interface to uevent. */
1474 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1475 if (retval)
1476 goto exit;
1477
1478 /* pass ifindex to uevent.
1479 * ifindex is useful as it won't change (interface name may change)
1480 * and is what RtNetlink uses natively.
1481 */
1482 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1483
1484exit:
1485 return retval;
1486}
1487
1488/*
1489 * netdev_release -- destroy and free a dead device.
1490 * Called when last reference to device kobject is gone.
1491 */
1492static void netdev_release(struct device *d)
1493{
1494 struct net_device *dev = to_net_dev(d);
1495
1496 BUG_ON(dev->reg_state != NETREG_RELEASED);
1497
1498 /* no need to wait for rcu grace period:
1499 * device is dead and about to be freed.
1500 */
1501 kfree(rcu_access_pointer(dev->ifalias));
1502 netdev_freemem(dev);
1503}
1504
1505static const void *net_namespace(struct device *d)
1506{
1507 struct net_device *dev = to_net_dev(d);
1508
1509 return dev_net(dev);
1510}
1511
1512static struct class net_class __ro_after_init = {
1513 .name = "net",
1514 .dev_release = netdev_release,
1515 .dev_groups = net_class_groups,
1516 .dev_uevent = netdev_uevent,
1517 .ns_type = &net_ns_type_operations,
1518 .namespace = net_namespace,
1519};
1520
1521#ifdef CONFIG_OF_NET
1522static int of_dev_node_match(struct device *dev, const void *data)
1523{
1524 int ret = 0;
1525
1526 if (dev->parent)
1527 ret = dev->parent->of_node == data;
1528
1529 return ret == 0 ? dev->of_node == data : ret;
1530}
1531
1532/*
1533 * of_find_net_device_by_node - lookup the net device for the device node
1534 * @np: OF device node
1535 *
1536 * Looks up the net_device structure corresponding with the device node.
1537 * If successful, returns a pointer to the net_device with the embedded
1538 * struct device refcount incremented by one, or NULL on failure. The
1539 * refcount must be dropped when done with the net_device.
1540 */
1541struct net_device *of_find_net_device_by_node(struct device_node *np)
1542{
1543 struct device *dev;
1544
1545 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1546 if (!dev)
1547 return NULL;
1548
1549 return to_net_dev(dev);
1550}
1551EXPORT_SYMBOL(of_find_net_device_by_node);
1552#endif
1553
1554/* Delete sysfs entries but hold kobject reference until after all
1555 * netdev references are gone.
1556 */
1557void netdev_unregister_kobject(struct net_device *ndev)
1558{
1559 struct device *dev = &ndev->dev;
1560
1561 if (!refcount_read(&dev_net(ndev)->count))
1562 dev_set_uevent_suppress(dev, 1);
1563
1564 kobject_get(&dev->kobj);
1565
1566 remove_queue_kobjects(ndev);
1567
1568 pm_runtime_set_memalloc_noio(dev, false);
1569
1570 device_del(dev);
1571}
1572
1573/* Create sysfs entries for network device. */
1574int netdev_register_kobject(struct net_device *ndev)
1575{
1576 struct device *dev = &ndev->dev;
1577 const struct attribute_group **groups = ndev->sysfs_groups;
1578 int error = 0;
1579
1580 device_initialize(dev);
1581 dev->class = &net_class;
1582 dev->platform_data = ndev;
1583 dev->groups = groups;
1584
1585 dev_set_name(dev, "%s", ndev->name);
1586
1587#ifdef CONFIG_SYSFS
1588 /* Allow for a device specific group */
1589 if (*groups)
1590 groups++;
1591
1592 *groups++ = &netstat_group;
1593
1594#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1595 if (ndev->ieee80211_ptr)
1596 *groups++ = &wireless_group;
1597#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1598 else if (ndev->wireless_handlers)
1599 *groups++ = &wireless_group;
1600#endif
1601#endif
1602#endif /* CONFIG_SYSFS */
1603
1604 error = device_add(dev);
1605 if (error)
1606 return error;
1607
1608 error = register_queue_kobjects(ndev);
1609 if (error) {
1610 device_del(dev);
1611 return error;
1612 }
1613
1614 pm_runtime_set_memalloc_noio(dev, true);
1615
1616 return error;
1617}
1618
1619int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1620 const void *ns)
1621{
1622 return class_create_file_ns(&net_class, class_attr, ns);
1623}
1624EXPORT_SYMBOL(netdev_class_create_file_ns);
1625
1626void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1627 const void *ns)
1628{
1629 class_remove_file_ns(&net_class, class_attr, ns);
1630}
1631EXPORT_SYMBOL(netdev_class_remove_file_ns);
1632
1633int __init netdev_kobject_init(void)
1634{
1635 kobj_ns_type_register(&net_ns_type_operations);
1636 return class_register(&net_class);
1637}