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