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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net-sysfs.c - network device class and attributes
4 *
5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
6 */
7
8#include <linux/capability.h>
9#include <linux/kernel.h>
10#include <linux/netdevice.h>
11#include <linux/if_arp.h>
12#include <linux/slab.h>
13#include <linux/sched/signal.h>
14#include <linux/nsproxy.h>
15#include <net/sock.h>
16#include <net/net_namespace.h>
17#include <linux/rtnetlink.h>
18#include <linux/vmalloc.h>
19#include <linux/export.h>
20#include <linux/jiffies.h>
21#include <linux/pm_runtime.h>
22#include <linux/of.h>
23#include <linux/of_net.h>
24#include <linux/cpu.h>
25
26#include "net-sysfs.h"
27
28#ifdef CONFIG_SYSFS
29static const char fmt_hex[] = "%#x\n";
30static const char fmt_dec[] = "%d\n";
31static const char fmt_ulong[] = "%lu\n";
32static const char fmt_u64[] = "%llu\n";
33
34static inline int dev_isalive(const struct net_device *dev)
35{
36 return dev->reg_state <= NETREG_REGISTERED;
37}
38
39/* use same locking rules as GIF* ioctl's */
40static ssize_t netdev_show(const struct device *dev,
41 struct device_attribute *attr, char *buf,
42 ssize_t (*format)(const struct net_device *, char *))
43{
44 struct net_device *ndev = to_net_dev(dev);
45 ssize_t ret = -EINVAL;
46
47 read_lock(&dev_base_lock);
48 if (dev_isalive(ndev))
49 ret = (*format)(ndev, buf);
50 read_unlock(&dev_base_lock);
51
52 return ret;
53}
54
55/* generate a show function for simple field */
56#define NETDEVICE_SHOW(field, format_string) \
57static ssize_t format_##field(const struct net_device *dev, char *buf) \
58{ \
59 return sprintf(buf, format_string, dev->field); \
60} \
61static ssize_t field##_show(struct device *dev, \
62 struct device_attribute *attr, char *buf) \
63{ \
64 return netdev_show(dev, attr, buf, format_##field); \
65} \
66
67#define NETDEVICE_SHOW_RO(field, format_string) \
68NETDEVICE_SHOW(field, format_string); \
69static DEVICE_ATTR_RO(field)
70
71#define NETDEVICE_SHOW_RW(field, format_string) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RW(field)
74
75/* use same locking and permission rules as SIF* ioctl's */
76static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
77 const char *buf, size_t len,
78 int (*set)(struct net_device *, unsigned long))
79{
80 struct net_device *netdev = to_net_dev(dev);
81 struct net *net = dev_net(netdev);
82 unsigned long new;
83 int ret = -EINVAL;
84
85 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
86 return -EPERM;
87
88 ret = kstrtoul(buf, 0, &new);
89 if (ret)
90 goto err;
91
92 if (!rtnl_trylock())
93 return restart_syscall();
94
95 if (dev_isalive(netdev)) {
96 ret = (*set)(netdev, new);
97 if (ret == 0)
98 ret = len;
99 }
100 rtnl_unlock();
101 err:
102 return ret;
103}
104
105NETDEVICE_SHOW_RO(dev_id, fmt_hex);
106NETDEVICE_SHOW_RO(dev_port, fmt_dec);
107NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
108NETDEVICE_SHOW_RO(addr_len, fmt_dec);
109NETDEVICE_SHOW_RO(ifindex, fmt_dec);
110NETDEVICE_SHOW_RO(type, fmt_dec);
111NETDEVICE_SHOW_RO(link_mode, fmt_dec);
112
113static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
114 char *buf)
115{
116 struct net_device *ndev = to_net_dev(dev);
117
118 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
119}
120static DEVICE_ATTR_RO(iflink);
121
122static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
123{
124 return sprintf(buf, fmt_dec, dev->name_assign_type);
125}
126
127static ssize_t name_assign_type_show(struct device *dev,
128 struct device_attribute *attr,
129 char *buf)
130{
131 struct net_device *ndev = to_net_dev(dev);
132 ssize_t ret = -EINVAL;
133
134 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
135 ret = netdev_show(dev, attr, buf, format_name_assign_type);
136
137 return ret;
138}
139static DEVICE_ATTR_RO(name_assign_type);
140
141/* use same locking rules as GIFHWADDR ioctl's */
142static ssize_t address_show(struct device *dev, struct device_attribute *attr,
143 char *buf)
144{
145 struct net_device *ndev = to_net_dev(dev);
146 ssize_t ret = -EINVAL;
147
148 read_lock(&dev_base_lock);
149 if (dev_isalive(ndev))
150 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
151 read_unlock(&dev_base_lock);
152 return ret;
153}
154static DEVICE_ATTR_RO(address);
155
156static ssize_t broadcast_show(struct device *dev,
157 struct device_attribute *attr, char *buf)
158{
159 struct net_device *ndev = to_net_dev(dev);
160
161 if (dev_isalive(ndev))
162 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
163 return -EINVAL;
164}
165static DEVICE_ATTR_RO(broadcast);
166
167static int change_carrier(struct net_device *dev, unsigned long new_carrier)
168{
169 if (!netif_running(dev))
170 return -EINVAL;
171 return dev_change_carrier(dev, (bool)new_carrier);
172}
173
174static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
175 const char *buf, size_t len)
176{
177 return netdev_store(dev, attr, buf, len, change_carrier);
178}
179
180static ssize_t carrier_show(struct device *dev,
181 struct device_attribute *attr, char *buf)
182{
183 struct net_device *netdev = to_net_dev(dev);
184
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
293 return sprintf(buf, fmt_dec,
294 atomic_read(&netdev->carrier_up_count) +
295 atomic_read(&netdev->carrier_down_count));
296}
297static DEVICE_ATTR_RO(carrier_changes);
298
299static ssize_t carrier_up_count_show(struct device *dev,
300 struct device_attribute *attr,
301 char *buf)
302{
303 struct net_device *netdev = to_net_dev(dev);
304
305 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
306}
307static DEVICE_ATTR_RO(carrier_up_count);
308
309static ssize_t carrier_down_count_show(struct device *dev,
310 struct device_attribute *attr,
311 char *buf)
312{
313 struct net_device *netdev = to_net_dev(dev);
314
315 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
316}
317static DEVICE_ATTR_RO(carrier_down_count);
318
319/* read-write attributes */
320
321static int change_mtu(struct net_device *dev, unsigned long new_mtu)
322{
323 return dev_set_mtu(dev, (int)new_mtu);
324}
325
326static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
327 const char *buf, size_t len)
328{
329 return netdev_store(dev, attr, buf, len, change_mtu);
330}
331NETDEVICE_SHOW_RW(mtu, fmt_dec);
332
333static int change_flags(struct net_device *dev, unsigned long new_flags)
334{
335 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
336}
337
338static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
339 const char *buf, size_t len)
340{
341 return netdev_store(dev, attr, buf, len, change_flags);
342}
343NETDEVICE_SHOW_RW(flags, fmt_hex);
344
345static ssize_t tx_queue_len_store(struct device *dev,
346 struct device_attribute *attr,
347 const char *buf, size_t len)
348{
349 if (!capable(CAP_NET_ADMIN))
350 return -EPERM;
351
352 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
353}
354NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
355
356static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
357{
358 dev->gro_flush_timeout = val;
359 return 0;
360}
361
362static ssize_t gro_flush_timeout_store(struct device *dev,
363 struct device_attribute *attr,
364 const char *buf, size_t len)
365{
366 if (!capable(CAP_NET_ADMIN))
367 return -EPERM;
368
369 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
370}
371NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
372
373static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
375{
376 struct net_device *netdev = to_net_dev(dev);
377 struct net *net = dev_net(netdev);
378 size_t count = len;
379 ssize_t ret = 0;
380
381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
382 return -EPERM;
383
384 /* ignore trailing newline */
385 if (len > 0 && buf[len - 1] == '\n')
386 --count;
387
388 if (!rtnl_trylock())
389 return restart_syscall();
390
391 if (dev_isalive(netdev)) {
392 ret = dev_set_alias(netdev, buf, count);
393 if (ret < 0)
394 goto err;
395 ret = len;
396 netdev_state_change(netdev);
397 }
398err:
399 rtnl_unlock();
400
401 return ret;
402}
403
404static ssize_t ifalias_show(struct device *dev,
405 struct device_attribute *attr, char *buf)
406{
407 const struct net_device *netdev = to_net_dev(dev);
408 char tmp[IFALIASZ];
409 ssize_t ret = 0;
410
411 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
412 if (ret > 0)
413 ret = sprintf(buf, "%s\n", tmp);
414 return ret;
415}
416static DEVICE_ATTR_RW(ifalias);
417
418static int change_group(struct net_device *dev, unsigned long new_group)
419{
420 dev_set_group(dev, (int)new_group);
421 return 0;
422}
423
424static ssize_t group_store(struct device *dev, struct device_attribute *attr,
425 const char *buf, size_t len)
426{
427 return netdev_store(dev, attr, buf, len, change_group);
428}
429NETDEVICE_SHOW(group, fmt_dec);
430static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
431
432static int change_proto_down(struct net_device *dev, unsigned long proto_down)
433{
434 return dev_change_proto_down(dev, (bool)proto_down);
435}
436
437static ssize_t proto_down_store(struct device *dev,
438 struct device_attribute *attr,
439 const char *buf, size_t len)
440{
441 return netdev_store(dev, attr, buf, len, change_proto_down);
442}
443NETDEVICE_SHOW_RW(proto_down, fmt_dec);
444
445static ssize_t phys_port_id_show(struct device *dev,
446 struct device_attribute *attr, char *buf)
447{
448 struct net_device *netdev = to_net_dev(dev);
449 ssize_t ret = -EINVAL;
450
451 if (!rtnl_trylock())
452 return restart_syscall();
453
454 if (dev_isalive(netdev)) {
455 struct netdev_phys_item_id ppid;
456
457 ret = dev_get_phys_port_id(netdev, &ppid);
458 if (!ret)
459 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
460 }
461 rtnl_unlock();
462
463 return ret;
464}
465static DEVICE_ATTR_RO(phys_port_id);
466
467static ssize_t phys_port_name_show(struct device *dev,
468 struct device_attribute *attr, char *buf)
469{
470 struct net_device *netdev = to_net_dev(dev);
471 ssize_t ret = -EINVAL;
472
473 if (!rtnl_trylock())
474 return restart_syscall();
475
476 if (dev_isalive(netdev)) {
477 char name[IFNAMSIZ];
478
479 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
480 if (!ret)
481 ret = sprintf(buf, "%s\n", name);
482 }
483 rtnl_unlock();
484
485 return ret;
486}
487static DEVICE_ATTR_RO(phys_port_name);
488
489static ssize_t phys_switch_id_show(struct device *dev,
490 struct device_attribute *attr, char *buf)
491{
492 struct net_device *netdev = to_net_dev(dev);
493 ssize_t ret = -EINVAL;
494
495 if (!rtnl_trylock())
496 return restart_syscall();
497
498 if (dev_isalive(netdev)) {
499 struct netdev_phys_item_id ppid = { };
500
501 ret = dev_get_port_parent_id(netdev, &ppid, false);
502 if (!ret)
503 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
504 }
505 rtnl_unlock();
506
507 return ret;
508}
509static DEVICE_ATTR_RO(phys_switch_id);
510
511static struct attribute *net_class_attrs[] __ro_after_init = {
512 &dev_attr_netdev_group.attr,
513 &dev_attr_type.attr,
514 &dev_attr_dev_id.attr,
515 &dev_attr_dev_port.attr,
516 &dev_attr_iflink.attr,
517 &dev_attr_ifindex.attr,
518 &dev_attr_name_assign_type.attr,
519 &dev_attr_addr_assign_type.attr,
520 &dev_attr_addr_len.attr,
521 &dev_attr_link_mode.attr,
522 &dev_attr_address.attr,
523 &dev_attr_broadcast.attr,
524 &dev_attr_speed.attr,
525 &dev_attr_duplex.attr,
526 &dev_attr_dormant.attr,
527 &dev_attr_operstate.attr,
528 &dev_attr_carrier_changes.attr,
529 &dev_attr_ifalias.attr,
530 &dev_attr_carrier.attr,
531 &dev_attr_mtu.attr,
532 &dev_attr_flags.attr,
533 &dev_attr_tx_queue_len.attr,
534 &dev_attr_gro_flush_timeout.attr,
535 &dev_attr_phys_port_id.attr,
536 &dev_attr_phys_port_name.attr,
537 &dev_attr_phys_switch_id.attr,
538 &dev_attr_proto_down.attr,
539 &dev_attr_carrier_up_count.attr,
540 &dev_attr_carrier_down_count.attr,
541 NULL,
542};
543ATTRIBUTE_GROUPS(net_class);
544
545/* Show a given an attribute in the statistics group */
546static ssize_t netstat_show(const struct device *d,
547 struct device_attribute *attr, char *buf,
548 unsigned long offset)
549{
550 struct net_device *dev = to_net_dev(d);
551 ssize_t ret = -EINVAL;
552
553 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
554 offset % sizeof(u64) != 0);
555
556 read_lock(&dev_base_lock);
557 if (dev_isalive(dev)) {
558 struct rtnl_link_stats64 temp;
559 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
560
561 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
562 }
563 read_unlock(&dev_base_lock);
564 return ret;
565}
566
567/* generate a read-only statistics attribute */
568#define NETSTAT_ENTRY(name) \
569static ssize_t name##_show(struct device *d, \
570 struct device_attribute *attr, char *buf) \
571{ \
572 return netstat_show(d, attr, buf, \
573 offsetof(struct rtnl_link_stats64, name)); \
574} \
575static DEVICE_ATTR_RO(name)
576
577NETSTAT_ENTRY(rx_packets);
578NETSTAT_ENTRY(tx_packets);
579NETSTAT_ENTRY(rx_bytes);
580NETSTAT_ENTRY(tx_bytes);
581NETSTAT_ENTRY(rx_errors);
582NETSTAT_ENTRY(tx_errors);
583NETSTAT_ENTRY(rx_dropped);
584NETSTAT_ENTRY(tx_dropped);
585NETSTAT_ENTRY(multicast);
586NETSTAT_ENTRY(collisions);
587NETSTAT_ENTRY(rx_length_errors);
588NETSTAT_ENTRY(rx_over_errors);
589NETSTAT_ENTRY(rx_crc_errors);
590NETSTAT_ENTRY(rx_frame_errors);
591NETSTAT_ENTRY(rx_fifo_errors);
592NETSTAT_ENTRY(rx_missed_errors);
593NETSTAT_ENTRY(tx_aborted_errors);
594NETSTAT_ENTRY(tx_carrier_errors);
595NETSTAT_ENTRY(tx_fifo_errors);
596NETSTAT_ENTRY(tx_heartbeat_errors);
597NETSTAT_ENTRY(tx_window_errors);
598NETSTAT_ENTRY(rx_compressed);
599NETSTAT_ENTRY(tx_compressed);
600NETSTAT_ENTRY(rx_nohandler);
601
602static struct attribute *netstat_attrs[] __ro_after_init = {
603 &dev_attr_rx_packets.attr,
604 &dev_attr_tx_packets.attr,
605 &dev_attr_rx_bytes.attr,
606 &dev_attr_tx_bytes.attr,
607 &dev_attr_rx_errors.attr,
608 &dev_attr_tx_errors.attr,
609 &dev_attr_rx_dropped.attr,
610 &dev_attr_tx_dropped.attr,
611 &dev_attr_multicast.attr,
612 &dev_attr_collisions.attr,
613 &dev_attr_rx_length_errors.attr,
614 &dev_attr_rx_over_errors.attr,
615 &dev_attr_rx_crc_errors.attr,
616 &dev_attr_rx_frame_errors.attr,
617 &dev_attr_rx_fifo_errors.attr,
618 &dev_attr_rx_missed_errors.attr,
619 &dev_attr_tx_aborted_errors.attr,
620 &dev_attr_tx_carrier_errors.attr,
621 &dev_attr_tx_fifo_errors.attr,
622 &dev_attr_tx_heartbeat_errors.attr,
623 &dev_attr_tx_window_errors.attr,
624 &dev_attr_rx_compressed.attr,
625 &dev_attr_tx_compressed.attr,
626 &dev_attr_rx_nohandler.attr,
627 NULL
628};
629
630static const struct attribute_group netstat_group = {
631 .name = "statistics",
632 .attrs = netstat_attrs,
633};
634
635#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
636static struct attribute *wireless_attrs[] = {
637 NULL
638};
639
640static const struct attribute_group wireless_group = {
641 .name = "wireless",
642 .attrs = wireless_attrs,
643};
644#endif
645
646#else /* CONFIG_SYSFS */
647#define net_class_groups NULL
648#endif /* CONFIG_SYSFS */
649
650#ifdef CONFIG_SYSFS
651#define to_rx_queue_attr(_attr) \
652 container_of(_attr, struct rx_queue_attribute, attr)
653
654#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
655
656static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
657 char *buf)
658{
659 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
660 struct netdev_rx_queue *queue = to_rx_queue(kobj);
661
662 if (!attribute->show)
663 return -EIO;
664
665 return attribute->show(queue, buf);
666}
667
668static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
669 const char *buf, size_t count)
670{
671 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
672 struct netdev_rx_queue *queue = to_rx_queue(kobj);
673
674 if (!attribute->store)
675 return -EIO;
676
677 return attribute->store(queue, buf, count);
678}
679
680static const struct sysfs_ops rx_queue_sysfs_ops = {
681 .show = rx_queue_attr_show,
682 .store = rx_queue_attr_store,
683};
684
685#ifdef CONFIG_RPS
686static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
687{
688 struct rps_map *map;
689 cpumask_var_t mask;
690 int i, len;
691
692 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
693 return -ENOMEM;
694
695 rcu_read_lock();
696 map = rcu_dereference(queue->rps_map);
697 if (map)
698 for (i = 0; i < map->len; i++)
699 cpumask_set_cpu(map->cpus[i], mask);
700
701 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
702 rcu_read_unlock();
703 free_cpumask_var(mask);
704
705 return len < PAGE_SIZE ? len : -EINVAL;
706}
707
708static ssize_t store_rps_map(struct netdev_rx_queue *queue,
709 const char *buf, size_t len)
710{
711 struct rps_map *old_map, *map;
712 cpumask_var_t mask;
713 int err, cpu, i;
714 static DEFINE_MUTEX(rps_map_mutex);
715
716 if (!capable(CAP_NET_ADMIN))
717 return -EPERM;
718
719 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
720 return -ENOMEM;
721
722 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
723 if (err) {
724 free_cpumask_var(mask);
725 return err;
726 }
727
728 map = kzalloc(max_t(unsigned int,
729 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
730 GFP_KERNEL);
731 if (!map) {
732 free_cpumask_var(mask);
733 return -ENOMEM;
734 }
735
736 i = 0;
737 for_each_cpu_and(cpu, mask, cpu_online_mask)
738 map->cpus[i++] = cpu;
739
740 if (i) {
741 map->len = i;
742 } else {
743 kfree(map);
744 map = NULL;
745 }
746
747 mutex_lock(&rps_map_mutex);
748 old_map = rcu_dereference_protected(queue->rps_map,
749 mutex_is_locked(&rps_map_mutex));
750 rcu_assign_pointer(queue->rps_map, map);
751
752 if (map)
753 static_branch_inc(&rps_needed);
754 if (old_map)
755 static_branch_dec(&rps_needed);
756
757 mutex_unlock(&rps_map_mutex);
758
759 if (old_map)
760 kfree_rcu(old_map, rcu);
761
762 free_cpumask_var(mask);
763 return len;
764}
765
766static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
767 char *buf)
768{
769 struct rps_dev_flow_table *flow_table;
770 unsigned long val = 0;
771
772 rcu_read_lock();
773 flow_table = rcu_dereference(queue->rps_flow_table);
774 if (flow_table)
775 val = (unsigned long)flow_table->mask + 1;
776 rcu_read_unlock();
777
778 return sprintf(buf, "%lu\n", val);
779}
780
781static void rps_dev_flow_table_release(struct rcu_head *rcu)
782{
783 struct rps_dev_flow_table *table = container_of(rcu,
784 struct rps_dev_flow_table, rcu);
785 vfree(table);
786}
787
788static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
789 const char *buf, size_t len)
790{
791 unsigned long mask, count;
792 struct rps_dev_flow_table *table, *old_table;
793 static DEFINE_SPINLOCK(rps_dev_flow_lock);
794 int rc;
795
796 if (!capable(CAP_NET_ADMIN))
797 return -EPERM;
798
799 rc = kstrtoul(buf, 0, &count);
800 if (rc < 0)
801 return rc;
802
803 if (count) {
804 mask = count - 1;
805 /* mask = roundup_pow_of_two(count) - 1;
806 * without overflows...
807 */
808 while ((mask | (mask >> 1)) != mask)
809 mask |= (mask >> 1);
810 /* On 64 bit arches, must check mask fits in table->mask (u32),
811 * and on 32bit arches, must check
812 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
813 */
814#if BITS_PER_LONG > 32
815 if (mask > (unsigned long)(u32)mask)
816 return -EINVAL;
817#else
818 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
819 / sizeof(struct rps_dev_flow)) {
820 /* Enforce a limit to prevent overflow */
821 return -EINVAL;
822 }
823#endif
824 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
825 if (!table)
826 return -ENOMEM;
827
828 table->mask = mask;
829 for (count = 0; count <= mask; count++)
830 table->flows[count].cpu = RPS_NO_CPU;
831 } else {
832 table = NULL;
833 }
834
835 spin_lock(&rps_dev_flow_lock);
836 old_table = rcu_dereference_protected(queue->rps_flow_table,
837 lockdep_is_held(&rps_dev_flow_lock));
838 rcu_assign_pointer(queue->rps_flow_table, table);
839 spin_unlock(&rps_dev_flow_lock);
840
841 if (old_table)
842 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
843
844 return len;
845}
846
847static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
848 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
849
850static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
851 = __ATTR(rps_flow_cnt, 0644,
852 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
853#endif /* CONFIG_RPS */
854
855static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
856#ifdef CONFIG_RPS
857 &rps_cpus_attribute.attr,
858 &rps_dev_flow_table_cnt_attribute.attr,
859#endif
860 NULL
861};
862ATTRIBUTE_GROUPS(rx_queue_default);
863
864static void rx_queue_release(struct kobject *kobj)
865{
866 struct netdev_rx_queue *queue = to_rx_queue(kobj);
867#ifdef CONFIG_RPS
868 struct rps_map *map;
869 struct rps_dev_flow_table *flow_table;
870
871 map = rcu_dereference_protected(queue->rps_map, 1);
872 if (map) {
873 RCU_INIT_POINTER(queue->rps_map, NULL);
874 kfree_rcu(map, rcu);
875 }
876
877 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
878 if (flow_table) {
879 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
880 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
881 }
882#endif
883
884 memset(kobj, 0, sizeof(*kobj));
885 dev_put(queue->dev);
886}
887
888static const void *rx_queue_namespace(struct kobject *kobj)
889{
890 struct netdev_rx_queue *queue = to_rx_queue(kobj);
891 struct device *dev = &queue->dev->dev;
892 const void *ns = NULL;
893
894 if (dev->class && dev->class->ns_type)
895 ns = dev->class->namespace(dev);
896
897 return ns;
898}
899
900static void rx_queue_get_ownership(struct kobject *kobj,
901 kuid_t *uid, kgid_t *gid)
902{
903 const struct net *net = rx_queue_namespace(kobj);
904
905 net_ns_get_ownership(net, uid, gid);
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_groups = rx_queue_default_groups,
912 .namespace = rx_queue_namespace,
913 .get_ownership = rx_queue_get_ownership,
914};
915
916static int rx_queue_add_kobject(struct net_device *dev, int index)
917{
918 struct netdev_rx_queue *queue = dev->_rx + index;
919 struct kobject *kobj = &queue->kobj;
920 int error = 0;
921
922 kobj->kset = dev->queues_kset;
923 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
924 "rx-%u", index);
925 if (error)
926 goto err;
927
928 dev_hold(queue->dev);
929
930 if (dev->sysfs_rx_queue_group) {
931 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
932 if (error)
933 goto err;
934 }
935
936 kobject_uevent(kobj, KOBJ_ADD);
937
938 return error;
939
940err:
941 kobject_put(kobj);
942 return error;
943}
944#endif /* CONFIG_SYSFS */
945
946int
947net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
948{
949#ifdef CONFIG_SYSFS
950 int i;
951 int error = 0;
952
953#ifndef CONFIG_RPS
954 if (!dev->sysfs_rx_queue_group)
955 return 0;
956#endif
957 for (i = old_num; i < new_num; i++) {
958 error = rx_queue_add_kobject(dev, i);
959 if (error) {
960 new_num = old_num;
961 break;
962 }
963 }
964
965 while (--i >= new_num) {
966 struct kobject *kobj = &dev->_rx[i].kobj;
967
968 if (!refcount_read(&dev_net(dev)->count))
969 kobj->uevent_suppress = 1;
970 if (dev->sysfs_rx_queue_group)
971 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
972 kobject_put(kobj);
973 }
974
975 return error;
976#else
977 return 0;
978#endif
979}
980
981#ifdef CONFIG_SYSFS
982/*
983 * netdev_queue sysfs structures and functions.
984 */
985struct netdev_queue_attribute {
986 struct attribute attr;
987 ssize_t (*show)(struct netdev_queue *queue, char *buf);
988 ssize_t (*store)(struct netdev_queue *queue,
989 const char *buf, size_t len);
990};
991#define to_netdev_queue_attr(_attr) \
992 container_of(_attr, struct netdev_queue_attribute, attr)
993
994#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
995
996static ssize_t netdev_queue_attr_show(struct kobject *kobj,
997 struct attribute *attr, char *buf)
998{
999 const struct netdev_queue_attribute *attribute
1000 = to_netdev_queue_attr(attr);
1001 struct netdev_queue *queue = to_netdev_queue(kobj);
1002
1003 if (!attribute->show)
1004 return -EIO;
1005
1006 return attribute->show(queue, buf);
1007}
1008
1009static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1010 struct attribute *attr,
1011 const char *buf, size_t count)
1012{
1013 const struct netdev_queue_attribute *attribute
1014 = to_netdev_queue_attr(attr);
1015 struct netdev_queue *queue = to_netdev_queue(kobj);
1016
1017 if (!attribute->store)
1018 return -EIO;
1019
1020 return attribute->store(queue, buf, count);
1021}
1022
1023static const struct sysfs_ops netdev_queue_sysfs_ops = {
1024 .show = netdev_queue_attr_show,
1025 .store = netdev_queue_attr_store,
1026};
1027
1028static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1029{
1030 unsigned long trans_timeout;
1031
1032 spin_lock_irq(&queue->_xmit_lock);
1033 trans_timeout = queue->trans_timeout;
1034 spin_unlock_irq(&queue->_xmit_lock);
1035
1036 return sprintf(buf, "%lu", trans_timeout);
1037}
1038
1039static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1040{
1041 struct net_device *dev = queue->dev;
1042 unsigned int i;
1043
1044 i = queue - dev->_tx;
1045 BUG_ON(i >= dev->num_tx_queues);
1046
1047 return i;
1048}
1049
1050static ssize_t traffic_class_show(struct netdev_queue *queue,
1051 char *buf)
1052{
1053 struct net_device *dev = queue->dev;
1054 int index;
1055 int tc;
1056
1057 if (!netif_is_multiqueue(dev))
1058 return -ENOENT;
1059
1060 index = get_netdev_queue_index(queue);
1061
1062 /* If queue belongs to subordinate dev use its TC mapping */
1063 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1064
1065 tc = netdev_txq_to_tc(dev, index);
1066 if (tc < 0)
1067 return -EINVAL;
1068
1069 /* We can report the traffic class one of two ways:
1070 * Subordinate device traffic classes are reported with the traffic
1071 * class first, and then the subordinate class so for example TC0 on
1072 * subordinate device 2 will be reported as "0-2". If the queue
1073 * belongs to the root device it will be reported with just the
1074 * traffic class, so just "0" for TC 0 for example.
1075 */
1076 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1077 sprintf(buf, "%u\n", tc);
1078}
1079
1080#ifdef CONFIG_XPS
1081static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1082 char *buf)
1083{
1084 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1085}
1086
1087static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1088 const char *buf, size_t len)
1089{
1090 struct net_device *dev = queue->dev;
1091 int err, index = get_netdev_queue_index(queue);
1092 u32 rate = 0;
1093
1094 if (!capable(CAP_NET_ADMIN))
1095 return -EPERM;
1096
1097 err = kstrtou32(buf, 10, &rate);
1098 if (err < 0)
1099 return err;
1100
1101 if (!rtnl_trylock())
1102 return restart_syscall();
1103
1104 err = -EOPNOTSUPP;
1105 if (dev->netdev_ops->ndo_set_tx_maxrate)
1106 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1107
1108 rtnl_unlock();
1109 if (!err) {
1110 queue->tx_maxrate = rate;
1111 return len;
1112 }
1113 return err;
1114}
1115
1116static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1117 = __ATTR_RW(tx_maxrate);
1118#endif
1119
1120static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1121 = __ATTR_RO(tx_timeout);
1122
1123static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1124 = __ATTR_RO(traffic_class);
1125
1126#ifdef CONFIG_BQL
1127/*
1128 * Byte queue limits sysfs structures and functions.
1129 */
1130static ssize_t bql_show(char *buf, unsigned int value)
1131{
1132 return sprintf(buf, "%u\n", value);
1133}
1134
1135static ssize_t bql_set(const char *buf, const size_t count,
1136 unsigned int *pvalue)
1137{
1138 unsigned int value;
1139 int err;
1140
1141 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1142 value = DQL_MAX_LIMIT;
1143 } else {
1144 err = kstrtouint(buf, 10, &value);
1145 if (err < 0)
1146 return err;
1147 if (value > DQL_MAX_LIMIT)
1148 return -EINVAL;
1149 }
1150
1151 *pvalue = value;
1152
1153 return count;
1154}
1155
1156static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1157 char *buf)
1158{
1159 struct dql *dql = &queue->dql;
1160
1161 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1162}
1163
1164static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1165 const char *buf, size_t len)
1166{
1167 struct dql *dql = &queue->dql;
1168 unsigned int value;
1169 int err;
1170
1171 err = kstrtouint(buf, 10, &value);
1172 if (err < 0)
1173 return err;
1174
1175 dql->slack_hold_time = msecs_to_jiffies(value);
1176
1177 return len;
1178}
1179
1180static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1181 = __ATTR(hold_time, 0644,
1182 bql_show_hold_time, bql_set_hold_time);
1183
1184static ssize_t bql_show_inflight(struct netdev_queue *queue,
1185 char *buf)
1186{
1187 struct dql *dql = &queue->dql;
1188
1189 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1190}
1191
1192static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1193 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1194
1195#define BQL_ATTR(NAME, FIELD) \
1196static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1197 char *buf) \
1198{ \
1199 return bql_show(buf, queue->dql.FIELD); \
1200} \
1201 \
1202static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1203 const char *buf, size_t len) \
1204{ \
1205 return bql_set(buf, len, &queue->dql.FIELD); \
1206} \
1207 \
1208static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1209 = __ATTR(NAME, 0644, \
1210 bql_show_ ## NAME, bql_set_ ## NAME)
1211
1212BQL_ATTR(limit, limit);
1213BQL_ATTR(limit_max, max_limit);
1214BQL_ATTR(limit_min, min_limit);
1215
1216static struct attribute *dql_attrs[] __ro_after_init = {
1217 &bql_limit_attribute.attr,
1218 &bql_limit_max_attribute.attr,
1219 &bql_limit_min_attribute.attr,
1220 &bql_hold_time_attribute.attr,
1221 &bql_inflight_attribute.attr,
1222 NULL
1223};
1224
1225static const struct attribute_group dql_group = {
1226 .name = "byte_queue_limits",
1227 .attrs = dql_attrs,
1228};
1229#endif /* CONFIG_BQL */
1230
1231#ifdef CONFIG_XPS
1232static ssize_t xps_cpus_show(struct netdev_queue *queue,
1233 char *buf)
1234{
1235 struct net_device *dev = queue->dev;
1236 int cpu, len, num_tc = 1, tc = 0;
1237 struct xps_dev_maps *dev_maps;
1238 cpumask_var_t mask;
1239 unsigned long index;
1240
1241 if (!netif_is_multiqueue(dev))
1242 return -ENOENT;
1243
1244 index = get_netdev_queue_index(queue);
1245
1246 if (dev->num_tc) {
1247 /* Do not allow XPS on subordinate device directly */
1248 num_tc = dev->num_tc;
1249 if (num_tc < 0)
1250 return -EINVAL;
1251
1252 /* If queue belongs to subordinate dev use its map */
1253 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1254
1255 tc = netdev_txq_to_tc(dev, index);
1256 if (tc < 0)
1257 return -EINVAL;
1258 }
1259
1260 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1261 return -ENOMEM;
1262
1263 rcu_read_lock();
1264 dev_maps = rcu_dereference(dev->xps_cpus_map);
1265 if (dev_maps) {
1266 for_each_possible_cpu(cpu) {
1267 int i, tci = cpu * num_tc + tc;
1268 struct xps_map *map;
1269
1270 map = rcu_dereference(dev_maps->attr_map[tci]);
1271 if (!map)
1272 continue;
1273
1274 for (i = map->len; i--;) {
1275 if (map->queues[i] == index) {
1276 cpumask_set_cpu(cpu, mask);
1277 break;
1278 }
1279 }
1280 }
1281 }
1282 rcu_read_unlock();
1283
1284 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1285 free_cpumask_var(mask);
1286 return len < PAGE_SIZE ? len : -EINVAL;
1287}
1288
1289static ssize_t xps_cpus_store(struct netdev_queue *queue,
1290 const char *buf, size_t len)
1291{
1292 struct net_device *dev = queue->dev;
1293 unsigned long index;
1294 cpumask_var_t mask;
1295 int err;
1296
1297 if (!netif_is_multiqueue(dev))
1298 return -ENOENT;
1299
1300 if (!capable(CAP_NET_ADMIN))
1301 return -EPERM;
1302
1303 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1304 return -ENOMEM;
1305
1306 index = get_netdev_queue_index(queue);
1307
1308 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1309 if (err) {
1310 free_cpumask_var(mask);
1311 return err;
1312 }
1313
1314 err = netif_set_xps_queue(dev, mask, index);
1315
1316 free_cpumask_var(mask);
1317
1318 return err ? : len;
1319}
1320
1321static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1322 = __ATTR_RW(xps_cpus);
1323
1324static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1325{
1326 struct net_device *dev = queue->dev;
1327 struct xps_dev_maps *dev_maps;
1328 unsigned long *mask, index;
1329 int j, len, num_tc = 1, tc = 0;
1330
1331 index = get_netdev_queue_index(queue);
1332
1333 if (dev->num_tc) {
1334 num_tc = dev->num_tc;
1335 tc = netdev_txq_to_tc(dev, index);
1336 if (tc < 0)
1337 return -EINVAL;
1338 }
1339 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1340 if (!mask)
1341 return -ENOMEM;
1342
1343 rcu_read_lock();
1344 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1345 if (!dev_maps)
1346 goto out_no_maps;
1347
1348 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1349 j < dev->num_rx_queues;) {
1350 int i, tci = j * num_tc + tc;
1351 struct xps_map *map;
1352
1353 map = rcu_dereference(dev_maps->attr_map[tci]);
1354 if (!map)
1355 continue;
1356
1357 for (i = map->len; i--;) {
1358 if (map->queues[i] == index) {
1359 set_bit(j, mask);
1360 break;
1361 }
1362 }
1363 }
1364out_no_maps:
1365 rcu_read_unlock();
1366
1367 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1368 bitmap_free(mask);
1369
1370 return len < PAGE_SIZE ? len : -EINVAL;
1371}
1372
1373static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1374 size_t len)
1375{
1376 struct net_device *dev = queue->dev;
1377 struct net *net = dev_net(dev);
1378 unsigned long *mask, index;
1379 int err;
1380
1381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1382 return -EPERM;
1383
1384 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1385 if (!mask)
1386 return -ENOMEM;
1387
1388 index = get_netdev_queue_index(queue);
1389
1390 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1391 if (err) {
1392 bitmap_free(mask);
1393 return err;
1394 }
1395
1396 cpus_read_lock();
1397 err = __netif_set_xps_queue(dev, mask, index, true);
1398 cpus_read_unlock();
1399
1400 bitmap_free(mask);
1401 return err ? : len;
1402}
1403
1404static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1405 = __ATTR_RW(xps_rxqs);
1406#endif /* CONFIG_XPS */
1407
1408static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1409 &queue_trans_timeout.attr,
1410 &queue_traffic_class.attr,
1411#ifdef CONFIG_XPS
1412 &xps_cpus_attribute.attr,
1413 &xps_rxqs_attribute.attr,
1414 &queue_tx_maxrate.attr,
1415#endif
1416 NULL
1417};
1418ATTRIBUTE_GROUPS(netdev_queue_default);
1419
1420static void netdev_queue_release(struct kobject *kobj)
1421{
1422 struct netdev_queue *queue = to_netdev_queue(kobj);
1423
1424 memset(kobj, 0, sizeof(*kobj));
1425 dev_put(queue->dev);
1426}
1427
1428static const void *netdev_queue_namespace(struct kobject *kobj)
1429{
1430 struct netdev_queue *queue = to_netdev_queue(kobj);
1431 struct device *dev = &queue->dev->dev;
1432 const void *ns = NULL;
1433
1434 if (dev->class && dev->class->ns_type)
1435 ns = dev->class->namespace(dev);
1436
1437 return ns;
1438}
1439
1440static void netdev_queue_get_ownership(struct kobject *kobj,
1441 kuid_t *uid, kgid_t *gid)
1442{
1443 const struct net *net = netdev_queue_namespace(kobj);
1444
1445 net_ns_get_ownership(net, uid, gid);
1446}
1447
1448static struct kobj_type netdev_queue_ktype __ro_after_init = {
1449 .sysfs_ops = &netdev_queue_sysfs_ops,
1450 .release = netdev_queue_release,
1451 .default_groups = netdev_queue_default_groups,
1452 .namespace = netdev_queue_namespace,
1453 .get_ownership = netdev_queue_get_ownership,
1454};
1455
1456static int netdev_queue_add_kobject(struct net_device *dev, int index)
1457{
1458 struct netdev_queue *queue = dev->_tx + index;
1459 struct kobject *kobj = &queue->kobj;
1460 int error = 0;
1461
1462 kobj->kset = dev->queues_kset;
1463 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1464 "tx-%u", index);
1465 if (error)
1466 goto err;
1467
1468 dev_hold(queue->dev);
1469
1470#ifdef CONFIG_BQL
1471 error = sysfs_create_group(kobj, &dql_group);
1472 if (error)
1473 goto err;
1474#endif
1475
1476 kobject_uevent(kobj, KOBJ_ADD);
1477 return 0;
1478
1479err:
1480 kobject_put(kobj);
1481 return error;
1482}
1483#endif /* CONFIG_SYSFS */
1484
1485int
1486netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1487{
1488#ifdef CONFIG_SYSFS
1489 int i;
1490 int error = 0;
1491
1492 for (i = old_num; i < new_num; i++) {
1493 error = netdev_queue_add_kobject(dev, i);
1494 if (error) {
1495 new_num = old_num;
1496 break;
1497 }
1498 }
1499
1500 while (--i >= new_num) {
1501 struct netdev_queue *queue = dev->_tx + i;
1502
1503 if (!refcount_read(&dev_net(dev)->count))
1504 queue->kobj.uevent_suppress = 1;
1505#ifdef CONFIG_BQL
1506 sysfs_remove_group(&queue->kobj, &dql_group);
1507#endif
1508 kobject_put(&queue->kobj);
1509 }
1510
1511 return error;
1512#else
1513 return 0;
1514#endif /* CONFIG_SYSFS */
1515}
1516
1517static int register_queue_kobjects(struct net_device *dev)
1518{
1519 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1520
1521#ifdef CONFIG_SYSFS
1522 dev->queues_kset = kset_create_and_add("queues",
1523 NULL, &dev->dev.kobj);
1524 if (!dev->queues_kset)
1525 return -ENOMEM;
1526 real_rx = dev->real_num_rx_queues;
1527#endif
1528 real_tx = dev->real_num_tx_queues;
1529
1530 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1531 if (error)
1532 goto error;
1533 rxq = real_rx;
1534
1535 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1536 if (error)
1537 goto error;
1538 txq = real_tx;
1539
1540 return 0;
1541
1542error:
1543 netdev_queue_update_kobjects(dev, txq, 0);
1544 net_rx_queue_update_kobjects(dev, rxq, 0);
1545#ifdef CONFIG_SYSFS
1546 kset_unregister(dev->queues_kset);
1547#endif
1548 return error;
1549}
1550
1551static void remove_queue_kobjects(struct net_device *dev)
1552{
1553 int real_rx = 0, real_tx = 0;
1554
1555#ifdef CONFIG_SYSFS
1556 real_rx = dev->real_num_rx_queues;
1557#endif
1558 real_tx = dev->real_num_tx_queues;
1559
1560 net_rx_queue_update_kobjects(dev, real_rx, 0);
1561 netdev_queue_update_kobjects(dev, real_tx, 0);
1562#ifdef CONFIG_SYSFS
1563 kset_unregister(dev->queues_kset);
1564#endif
1565}
1566
1567static bool net_current_may_mount(void)
1568{
1569 struct net *net = current->nsproxy->net_ns;
1570
1571 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1572}
1573
1574static void *net_grab_current_ns(void)
1575{
1576 struct net *ns = current->nsproxy->net_ns;
1577#ifdef CONFIG_NET_NS
1578 if (ns)
1579 refcount_inc(&ns->passive);
1580#endif
1581 return ns;
1582}
1583
1584static const void *net_initial_ns(void)
1585{
1586 return &init_net;
1587}
1588
1589static const void *net_netlink_ns(struct sock *sk)
1590{
1591 return sock_net(sk);
1592}
1593
1594const struct kobj_ns_type_operations net_ns_type_operations = {
1595 .type = KOBJ_NS_TYPE_NET,
1596 .current_may_mount = net_current_may_mount,
1597 .grab_current_ns = net_grab_current_ns,
1598 .netlink_ns = net_netlink_ns,
1599 .initial_ns = net_initial_ns,
1600 .drop_ns = net_drop_ns,
1601};
1602EXPORT_SYMBOL_GPL(net_ns_type_operations);
1603
1604static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1605{
1606 struct net_device *dev = to_net_dev(d);
1607 int retval;
1608
1609 /* pass interface to uevent. */
1610 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1611 if (retval)
1612 goto exit;
1613
1614 /* pass ifindex to uevent.
1615 * ifindex is useful as it won't change (interface name may change)
1616 * and is what RtNetlink uses natively.
1617 */
1618 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1619
1620exit:
1621 return retval;
1622}
1623
1624/*
1625 * netdev_release -- destroy and free a dead device.
1626 * Called when last reference to device kobject is gone.
1627 */
1628static void netdev_release(struct device *d)
1629{
1630 struct net_device *dev = to_net_dev(d);
1631
1632 BUG_ON(dev->reg_state != NETREG_RELEASED);
1633
1634 /* no need to wait for rcu grace period:
1635 * device is dead and about to be freed.
1636 */
1637 kfree(rcu_access_pointer(dev->ifalias));
1638 netdev_freemem(dev);
1639}
1640
1641static const void *net_namespace(struct device *d)
1642{
1643 struct net_device *dev = to_net_dev(d);
1644
1645 return dev_net(dev);
1646}
1647
1648static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1649{
1650 struct net_device *dev = to_net_dev(d);
1651 const struct net *net = dev_net(dev);
1652
1653 net_ns_get_ownership(net, uid, gid);
1654}
1655
1656static struct class net_class __ro_after_init = {
1657 .name = "net",
1658 .dev_release = netdev_release,
1659 .dev_groups = net_class_groups,
1660 .dev_uevent = netdev_uevent,
1661 .ns_type = &net_ns_type_operations,
1662 .namespace = net_namespace,
1663 .get_ownership = net_get_ownership,
1664};
1665
1666#ifdef CONFIG_OF_NET
1667static int of_dev_node_match(struct device *dev, const void *data)
1668{
1669 int ret = 0;
1670
1671 if (dev->parent)
1672 ret = dev->parent->of_node == data;
1673
1674 return ret == 0 ? dev->of_node == data : ret;
1675}
1676
1677/*
1678 * of_find_net_device_by_node - lookup the net device for the device node
1679 * @np: OF device node
1680 *
1681 * Looks up the net_device structure corresponding with the device node.
1682 * If successful, returns a pointer to the net_device with the embedded
1683 * struct device refcount incremented by one, or NULL on failure. The
1684 * refcount must be dropped when done with the net_device.
1685 */
1686struct net_device *of_find_net_device_by_node(struct device_node *np)
1687{
1688 struct device *dev;
1689
1690 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1691 if (!dev)
1692 return NULL;
1693
1694 return to_net_dev(dev);
1695}
1696EXPORT_SYMBOL(of_find_net_device_by_node);
1697#endif
1698
1699/* Delete sysfs entries but hold kobject reference until after all
1700 * netdev references are gone.
1701 */
1702void netdev_unregister_kobject(struct net_device *ndev)
1703{
1704 struct device *dev = &ndev->dev;
1705
1706 if (!refcount_read(&dev_net(ndev)->count))
1707 dev_set_uevent_suppress(dev, 1);
1708
1709 kobject_get(&dev->kobj);
1710
1711 remove_queue_kobjects(ndev);
1712
1713 pm_runtime_set_memalloc_noio(dev, false);
1714
1715 device_del(dev);
1716}
1717
1718/* Create sysfs entries for network device. */
1719int netdev_register_kobject(struct net_device *ndev)
1720{
1721 struct device *dev = &ndev->dev;
1722 const struct attribute_group **groups = ndev->sysfs_groups;
1723 int error = 0;
1724
1725 device_initialize(dev);
1726 dev->class = &net_class;
1727 dev->platform_data = ndev;
1728 dev->groups = groups;
1729
1730 dev_set_name(dev, "%s", ndev->name);
1731
1732#ifdef CONFIG_SYSFS
1733 /* Allow for a device specific group */
1734 if (*groups)
1735 groups++;
1736
1737 *groups++ = &netstat_group;
1738
1739#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1740 if (ndev->ieee80211_ptr)
1741 *groups++ = &wireless_group;
1742#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1743 else if (ndev->wireless_handlers)
1744 *groups++ = &wireless_group;
1745#endif
1746#endif
1747#endif /* CONFIG_SYSFS */
1748
1749 error = device_add(dev);
1750 if (error)
1751 return error;
1752
1753 error = register_queue_kobjects(ndev);
1754 if (error) {
1755 device_del(dev);
1756 return error;
1757 }
1758
1759 pm_runtime_set_memalloc_noio(dev, true);
1760
1761 return error;
1762}
1763
1764int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1765 const void *ns)
1766{
1767 return class_create_file_ns(&net_class, class_attr, ns);
1768}
1769EXPORT_SYMBOL(netdev_class_create_file_ns);
1770
1771void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1772 const void *ns)
1773{
1774 class_remove_file_ns(&net_class, class_attr, ns);
1775}
1776EXPORT_SYMBOL(netdev_class_remove_file_ns);
1777
1778int __init netdev_kobject_init(void)
1779{
1780 kobj_ns_type_register(&net_ns_type_operations);
1781 return class_register(&net_class);
1782}