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/sched/isolation.h>
15#include <linux/nsproxy.h>
16#include <net/sock.h>
17#include <net/net_namespace.h>
18#include <linux/rtnetlink.h>
19#include <linux/vmalloc.h>
20#include <linux/export.h>
21#include <linux/jiffies.h>
22#include <linux/pm_runtime.h>
23#include <linux/of.h>
24#include <linux/of_net.h>
25#include <linux/cpu.h>
26
27#include "dev.h"
28#include "net-sysfs.h"
29
30#ifdef CONFIG_SYSFS
31static const char fmt_hex[] = "%#x\n";
32static const char fmt_dec[] = "%d\n";
33static const char fmt_ulong[] = "%lu\n";
34static const char fmt_u64[] = "%llu\n";
35
36/* Caller holds RTNL or dev_base_lock */
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 sysfs_emit(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;
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 ret = (*set)(netdev, new);
100 if (ret == 0)
101 ret = len;
102 }
103 rtnl_unlock();
104 err:
105 return ret;
106}
107
108NETDEVICE_SHOW_RO(dev_id, fmt_hex);
109NETDEVICE_SHOW_RO(dev_port, fmt_dec);
110NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
111NETDEVICE_SHOW_RO(addr_len, fmt_dec);
112NETDEVICE_SHOW_RO(ifindex, fmt_dec);
113NETDEVICE_SHOW_RO(type, fmt_dec);
114NETDEVICE_SHOW_RO(link_mode, fmt_dec);
115
116static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
117 char *buf)
118{
119 struct net_device *ndev = to_net_dev(dev);
120
121 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
122}
123static DEVICE_ATTR_RO(iflink);
124
125static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
126{
127 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
128}
129
130static ssize_t name_assign_type_show(struct device *dev,
131 struct device_attribute *attr,
132 char *buf)
133{
134 struct net_device *ndev = to_net_dev(dev);
135 ssize_t ret = -EINVAL;
136
137 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
138 ret = netdev_show(dev, attr, buf, format_name_assign_type);
139
140 return ret;
141}
142static DEVICE_ATTR_RO(name_assign_type);
143
144/* use same locking rules as GIFHWADDR ioctl's */
145static ssize_t address_show(struct device *dev, struct device_attribute *attr,
146 char *buf)
147{
148 struct net_device *ndev = to_net_dev(dev);
149 ssize_t ret = -EINVAL;
150
151 read_lock(&dev_base_lock);
152 if (dev_isalive(ndev))
153 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
154 read_unlock(&dev_base_lock);
155 return ret;
156}
157static DEVICE_ATTR_RO(address);
158
159static ssize_t broadcast_show(struct device *dev,
160 struct device_attribute *attr, char *buf)
161{
162 struct net_device *ndev = to_net_dev(dev);
163
164 if (dev_isalive(ndev))
165 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
166 return -EINVAL;
167}
168static DEVICE_ATTR_RO(broadcast);
169
170static int change_carrier(struct net_device *dev, unsigned long new_carrier)
171{
172 if (!netif_running(dev))
173 return -EINVAL;
174 return dev_change_carrier(dev, (bool)new_carrier);
175}
176
177static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
178 const char *buf, size_t len)
179{
180 struct net_device *netdev = to_net_dev(dev);
181
182 /* The check is also done in change_carrier; this helps returning early
183 * without hitting the trylock/restart in netdev_store.
184 */
185 if (!netdev->netdev_ops->ndo_change_carrier)
186 return -EOPNOTSUPP;
187
188 return netdev_store(dev, attr, buf, len, change_carrier);
189}
190
191static ssize_t carrier_show(struct device *dev,
192 struct device_attribute *attr, char *buf)
193{
194 struct net_device *netdev = to_net_dev(dev);
195
196 if (netif_running(netdev))
197 return sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
198
199 return -EINVAL;
200}
201static DEVICE_ATTR_RW(carrier);
202
203static ssize_t speed_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
205{
206 struct net_device *netdev = to_net_dev(dev);
207 int ret = -EINVAL;
208
209 /* The check is also done in __ethtool_get_link_ksettings; this helps
210 * returning early without hitting the trylock/restart below.
211 */
212 if (!netdev->ethtool_ops->get_link_ksettings)
213 return ret;
214
215 if (!rtnl_trylock())
216 return restart_syscall();
217
218 if (netif_running(netdev) && netif_device_present(netdev)) {
219 struct ethtool_link_ksettings cmd;
220
221 if (!__ethtool_get_link_ksettings(netdev, &cmd))
222 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
223 }
224 rtnl_unlock();
225 return ret;
226}
227static DEVICE_ATTR_RO(speed);
228
229static ssize_t duplex_show(struct device *dev,
230 struct device_attribute *attr, char *buf)
231{
232 struct net_device *netdev = to_net_dev(dev);
233 int ret = -EINVAL;
234
235 /* The check is also done in __ethtool_get_link_ksettings; this helps
236 * returning early without hitting the trylock/restart below.
237 */
238 if (!netdev->ethtool_ops->get_link_ksettings)
239 return ret;
240
241 if (!rtnl_trylock())
242 return restart_syscall();
243
244 if (netif_running(netdev)) {
245 struct ethtool_link_ksettings cmd;
246
247 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
248 const char *duplex;
249
250 switch (cmd.base.duplex) {
251 case DUPLEX_HALF:
252 duplex = "half";
253 break;
254 case DUPLEX_FULL:
255 duplex = "full";
256 break;
257 default:
258 duplex = "unknown";
259 break;
260 }
261 ret = sysfs_emit(buf, "%s\n", duplex);
262 }
263 }
264 rtnl_unlock();
265 return ret;
266}
267static DEVICE_ATTR_RO(duplex);
268
269static ssize_t testing_show(struct device *dev,
270 struct device_attribute *attr, char *buf)
271{
272 struct net_device *netdev = to_net_dev(dev);
273
274 if (netif_running(netdev))
275 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
276
277 return -EINVAL;
278}
279static DEVICE_ATTR_RO(testing);
280
281static ssize_t dormant_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 struct net_device *netdev = to_net_dev(dev);
285
286 if (netif_running(netdev))
287 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
288
289 return -EINVAL;
290}
291static DEVICE_ATTR_RO(dormant);
292
293static const char *const operstates[] = {
294 "unknown",
295 "notpresent", /* currently unused */
296 "down",
297 "lowerlayerdown",
298 "testing",
299 "dormant",
300 "up"
301};
302
303static ssize_t operstate_show(struct device *dev,
304 struct device_attribute *attr, char *buf)
305{
306 const struct net_device *netdev = to_net_dev(dev);
307 unsigned char operstate;
308
309 read_lock(&dev_base_lock);
310 operstate = netdev->operstate;
311 if (!netif_running(netdev))
312 operstate = IF_OPER_DOWN;
313 read_unlock(&dev_base_lock);
314
315 if (operstate >= ARRAY_SIZE(operstates))
316 return -EINVAL; /* should not happen */
317
318 return sysfs_emit(buf, "%s\n", operstates[operstate]);
319}
320static DEVICE_ATTR_RO(operstate);
321
322static ssize_t carrier_changes_show(struct device *dev,
323 struct device_attribute *attr,
324 char *buf)
325{
326 struct net_device *netdev = to_net_dev(dev);
327
328 return sysfs_emit(buf, fmt_dec,
329 atomic_read(&netdev->carrier_up_count) +
330 atomic_read(&netdev->carrier_down_count));
331}
332static DEVICE_ATTR_RO(carrier_changes);
333
334static ssize_t carrier_up_count_show(struct device *dev,
335 struct device_attribute *attr,
336 char *buf)
337{
338 struct net_device *netdev = to_net_dev(dev);
339
340 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
341}
342static DEVICE_ATTR_RO(carrier_up_count);
343
344static ssize_t carrier_down_count_show(struct device *dev,
345 struct device_attribute *attr,
346 char *buf)
347{
348 struct net_device *netdev = to_net_dev(dev);
349
350 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
351}
352static DEVICE_ATTR_RO(carrier_down_count);
353
354/* read-write attributes */
355
356static int change_mtu(struct net_device *dev, unsigned long new_mtu)
357{
358 return dev_set_mtu(dev, (int)new_mtu);
359}
360
361static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
362 const char *buf, size_t len)
363{
364 return netdev_store(dev, attr, buf, len, change_mtu);
365}
366NETDEVICE_SHOW_RW(mtu, fmt_dec);
367
368static int change_flags(struct net_device *dev, unsigned long new_flags)
369{
370 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
371}
372
373static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
375{
376 return netdev_store(dev, attr, buf, len, change_flags);
377}
378NETDEVICE_SHOW_RW(flags, fmt_hex);
379
380static ssize_t tx_queue_len_store(struct device *dev,
381 struct device_attribute *attr,
382 const char *buf, size_t len)
383{
384 if (!capable(CAP_NET_ADMIN))
385 return -EPERM;
386
387 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
388}
389NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
390
391static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
392{
393 WRITE_ONCE(dev->gro_flush_timeout, val);
394 return 0;
395}
396
397static ssize_t gro_flush_timeout_store(struct device *dev,
398 struct device_attribute *attr,
399 const char *buf, size_t len)
400{
401 if (!capable(CAP_NET_ADMIN))
402 return -EPERM;
403
404 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
405}
406NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
407
408static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
409{
410 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
411 return 0;
412}
413
414static ssize_t napi_defer_hard_irqs_store(struct device *dev,
415 struct device_attribute *attr,
416 const char *buf, size_t len)
417{
418 if (!capable(CAP_NET_ADMIN))
419 return -EPERM;
420
421 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
422}
423NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
424
425static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
426 const char *buf, size_t len)
427{
428 struct net_device *netdev = to_net_dev(dev);
429 struct net *net = dev_net(netdev);
430 size_t count = len;
431 ssize_t ret = 0;
432
433 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
434 return -EPERM;
435
436 /* ignore trailing newline */
437 if (len > 0 && buf[len - 1] == '\n')
438 --count;
439
440 if (!rtnl_trylock())
441 return restart_syscall();
442
443 if (dev_isalive(netdev)) {
444 ret = dev_set_alias(netdev, buf, count);
445 if (ret < 0)
446 goto err;
447 ret = len;
448 netdev_state_change(netdev);
449 }
450err:
451 rtnl_unlock();
452
453 return ret;
454}
455
456static ssize_t ifalias_show(struct device *dev,
457 struct device_attribute *attr, char *buf)
458{
459 const struct net_device *netdev = to_net_dev(dev);
460 char tmp[IFALIASZ];
461 ssize_t ret = 0;
462
463 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
464 if (ret > 0)
465 ret = sysfs_emit(buf, "%s\n", tmp);
466 return ret;
467}
468static DEVICE_ATTR_RW(ifalias);
469
470static int change_group(struct net_device *dev, unsigned long new_group)
471{
472 dev_set_group(dev, (int)new_group);
473 return 0;
474}
475
476static ssize_t group_store(struct device *dev, struct device_attribute *attr,
477 const char *buf, size_t len)
478{
479 return netdev_store(dev, attr, buf, len, change_group);
480}
481NETDEVICE_SHOW(group, fmt_dec);
482static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
483
484static int change_proto_down(struct net_device *dev, unsigned long proto_down)
485{
486 return dev_change_proto_down(dev, (bool)proto_down);
487}
488
489static ssize_t proto_down_store(struct device *dev,
490 struct device_attribute *attr,
491 const char *buf, size_t len)
492{
493 return netdev_store(dev, attr, buf, len, change_proto_down);
494}
495NETDEVICE_SHOW_RW(proto_down, fmt_dec);
496
497static ssize_t phys_port_id_show(struct device *dev,
498 struct device_attribute *attr, char *buf)
499{
500 struct net_device *netdev = to_net_dev(dev);
501 ssize_t ret = -EINVAL;
502
503 /* The check is also done in dev_get_phys_port_id; this helps returning
504 * early without hitting the trylock/restart below.
505 */
506 if (!netdev->netdev_ops->ndo_get_phys_port_id)
507 return -EOPNOTSUPP;
508
509 if (!rtnl_trylock())
510 return restart_syscall();
511
512 if (dev_isalive(netdev)) {
513 struct netdev_phys_item_id ppid;
514
515 ret = dev_get_phys_port_id(netdev, &ppid);
516 if (!ret)
517 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
518 }
519 rtnl_unlock();
520
521 return ret;
522}
523static DEVICE_ATTR_RO(phys_port_id);
524
525static ssize_t phys_port_name_show(struct device *dev,
526 struct device_attribute *attr, char *buf)
527{
528 struct net_device *netdev = to_net_dev(dev);
529 ssize_t ret = -EINVAL;
530
531 /* The checks are also done in dev_get_phys_port_name; this helps
532 * returning early without hitting the trylock/restart below.
533 */
534 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
535 !netdev->devlink_port)
536 return -EOPNOTSUPP;
537
538 if (!rtnl_trylock())
539 return restart_syscall();
540
541 if (dev_isalive(netdev)) {
542 char name[IFNAMSIZ];
543
544 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
545 if (!ret)
546 ret = sysfs_emit(buf, "%s\n", name);
547 }
548 rtnl_unlock();
549
550 return ret;
551}
552static DEVICE_ATTR_RO(phys_port_name);
553
554static ssize_t phys_switch_id_show(struct device *dev,
555 struct device_attribute *attr, char *buf)
556{
557 struct net_device *netdev = to_net_dev(dev);
558 ssize_t ret = -EINVAL;
559
560 /* The checks are also done in dev_get_phys_port_name; this helps
561 * returning early without hitting the trylock/restart below. This works
562 * because recurse is false when calling dev_get_port_parent_id.
563 */
564 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
565 !netdev->devlink_port)
566 return -EOPNOTSUPP;
567
568 if (!rtnl_trylock())
569 return restart_syscall();
570
571 if (dev_isalive(netdev)) {
572 struct netdev_phys_item_id ppid = { };
573
574 ret = dev_get_port_parent_id(netdev, &ppid, false);
575 if (!ret)
576 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
577 }
578 rtnl_unlock();
579
580 return ret;
581}
582static DEVICE_ATTR_RO(phys_switch_id);
583
584static ssize_t threaded_show(struct device *dev,
585 struct device_attribute *attr, char *buf)
586{
587 struct net_device *netdev = to_net_dev(dev);
588 ssize_t ret = -EINVAL;
589
590 if (!rtnl_trylock())
591 return restart_syscall();
592
593 if (dev_isalive(netdev))
594 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
595
596 rtnl_unlock();
597 return ret;
598}
599
600static int modify_napi_threaded(struct net_device *dev, unsigned long val)
601{
602 int ret;
603
604 if (list_empty(&dev->napi_list))
605 return -EOPNOTSUPP;
606
607 if (val != 0 && val != 1)
608 return -EOPNOTSUPP;
609
610 ret = dev_set_threaded(dev, val);
611
612 return ret;
613}
614
615static ssize_t threaded_store(struct device *dev,
616 struct device_attribute *attr,
617 const char *buf, size_t len)
618{
619 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
620}
621static DEVICE_ATTR_RW(threaded);
622
623static struct attribute *net_class_attrs[] __ro_after_init = {
624 &dev_attr_netdev_group.attr,
625 &dev_attr_type.attr,
626 &dev_attr_dev_id.attr,
627 &dev_attr_dev_port.attr,
628 &dev_attr_iflink.attr,
629 &dev_attr_ifindex.attr,
630 &dev_attr_name_assign_type.attr,
631 &dev_attr_addr_assign_type.attr,
632 &dev_attr_addr_len.attr,
633 &dev_attr_link_mode.attr,
634 &dev_attr_address.attr,
635 &dev_attr_broadcast.attr,
636 &dev_attr_speed.attr,
637 &dev_attr_duplex.attr,
638 &dev_attr_dormant.attr,
639 &dev_attr_testing.attr,
640 &dev_attr_operstate.attr,
641 &dev_attr_carrier_changes.attr,
642 &dev_attr_ifalias.attr,
643 &dev_attr_carrier.attr,
644 &dev_attr_mtu.attr,
645 &dev_attr_flags.attr,
646 &dev_attr_tx_queue_len.attr,
647 &dev_attr_gro_flush_timeout.attr,
648 &dev_attr_napi_defer_hard_irqs.attr,
649 &dev_attr_phys_port_id.attr,
650 &dev_attr_phys_port_name.attr,
651 &dev_attr_phys_switch_id.attr,
652 &dev_attr_proto_down.attr,
653 &dev_attr_carrier_up_count.attr,
654 &dev_attr_carrier_down_count.attr,
655 &dev_attr_threaded.attr,
656 NULL,
657};
658ATTRIBUTE_GROUPS(net_class);
659
660/* Show a given an attribute in the statistics group */
661static ssize_t netstat_show(const struct device *d,
662 struct device_attribute *attr, char *buf,
663 unsigned long offset)
664{
665 struct net_device *dev = to_net_dev(d);
666 ssize_t ret = -EINVAL;
667
668 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
669 offset % sizeof(u64) != 0);
670
671 read_lock(&dev_base_lock);
672 if (dev_isalive(dev)) {
673 struct rtnl_link_stats64 temp;
674 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
675
676 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
677 }
678 read_unlock(&dev_base_lock);
679 return ret;
680}
681
682/* generate a read-only statistics attribute */
683#define NETSTAT_ENTRY(name) \
684static ssize_t name##_show(struct device *d, \
685 struct device_attribute *attr, char *buf) \
686{ \
687 return netstat_show(d, attr, buf, \
688 offsetof(struct rtnl_link_stats64, name)); \
689} \
690static DEVICE_ATTR_RO(name)
691
692NETSTAT_ENTRY(rx_packets);
693NETSTAT_ENTRY(tx_packets);
694NETSTAT_ENTRY(rx_bytes);
695NETSTAT_ENTRY(tx_bytes);
696NETSTAT_ENTRY(rx_errors);
697NETSTAT_ENTRY(tx_errors);
698NETSTAT_ENTRY(rx_dropped);
699NETSTAT_ENTRY(tx_dropped);
700NETSTAT_ENTRY(multicast);
701NETSTAT_ENTRY(collisions);
702NETSTAT_ENTRY(rx_length_errors);
703NETSTAT_ENTRY(rx_over_errors);
704NETSTAT_ENTRY(rx_crc_errors);
705NETSTAT_ENTRY(rx_frame_errors);
706NETSTAT_ENTRY(rx_fifo_errors);
707NETSTAT_ENTRY(rx_missed_errors);
708NETSTAT_ENTRY(tx_aborted_errors);
709NETSTAT_ENTRY(tx_carrier_errors);
710NETSTAT_ENTRY(tx_fifo_errors);
711NETSTAT_ENTRY(tx_heartbeat_errors);
712NETSTAT_ENTRY(tx_window_errors);
713NETSTAT_ENTRY(rx_compressed);
714NETSTAT_ENTRY(tx_compressed);
715NETSTAT_ENTRY(rx_nohandler);
716
717static struct attribute *netstat_attrs[] __ro_after_init = {
718 &dev_attr_rx_packets.attr,
719 &dev_attr_tx_packets.attr,
720 &dev_attr_rx_bytes.attr,
721 &dev_attr_tx_bytes.attr,
722 &dev_attr_rx_errors.attr,
723 &dev_attr_tx_errors.attr,
724 &dev_attr_rx_dropped.attr,
725 &dev_attr_tx_dropped.attr,
726 &dev_attr_multicast.attr,
727 &dev_attr_collisions.attr,
728 &dev_attr_rx_length_errors.attr,
729 &dev_attr_rx_over_errors.attr,
730 &dev_attr_rx_crc_errors.attr,
731 &dev_attr_rx_frame_errors.attr,
732 &dev_attr_rx_fifo_errors.attr,
733 &dev_attr_rx_missed_errors.attr,
734 &dev_attr_tx_aborted_errors.attr,
735 &dev_attr_tx_carrier_errors.attr,
736 &dev_attr_tx_fifo_errors.attr,
737 &dev_attr_tx_heartbeat_errors.attr,
738 &dev_attr_tx_window_errors.attr,
739 &dev_attr_rx_compressed.attr,
740 &dev_attr_tx_compressed.attr,
741 &dev_attr_rx_nohandler.attr,
742 NULL
743};
744
745static const struct attribute_group netstat_group = {
746 .name = "statistics",
747 .attrs = netstat_attrs,
748};
749
750static struct attribute *wireless_attrs[] = {
751 NULL
752};
753
754static const struct attribute_group wireless_group = {
755 .name = "wireless",
756 .attrs = wireless_attrs,
757};
758
759static bool wireless_group_needed(struct net_device *ndev)
760{
761#if IS_ENABLED(CONFIG_CFG80211)
762 if (ndev->ieee80211_ptr)
763 return true;
764#endif
765#if IS_ENABLED(CONFIG_WIRELESS_EXT)
766 if (ndev->wireless_handlers)
767 return true;
768#endif
769 return false;
770}
771
772#else /* CONFIG_SYSFS */
773#define net_class_groups NULL
774#endif /* CONFIG_SYSFS */
775
776#ifdef CONFIG_SYSFS
777#define to_rx_queue_attr(_attr) \
778 container_of(_attr, struct rx_queue_attribute, attr)
779
780#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
781
782static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
783 char *buf)
784{
785 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
786 struct netdev_rx_queue *queue = to_rx_queue(kobj);
787
788 if (!attribute->show)
789 return -EIO;
790
791 return attribute->show(queue, buf);
792}
793
794static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
795 const char *buf, size_t count)
796{
797 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
798 struct netdev_rx_queue *queue = to_rx_queue(kobj);
799
800 if (!attribute->store)
801 return -EIO;
802
803 return attribute->store(queue, buf, count);
804}
805
806static const struct sysfs_ops rx_queue_sysfs_ops = {
807 .show = rx_queue_attr_show,
808 .store = rx_queue_attr_store,
809};
810
811#ifdef CONFIG_RPS
812static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
813{
814 struct rps_map *map;
815 cpumask_var_t mask;
816 int i, len;
817
818 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
819 return -ENOMEM;
820
821 rcu_read_lock();
822 map = rcu_dereference(queue->rps_map);
823 if (map)
824 for (i = 0; i < map->len; i++)
825 cpumask_set_cpu(map->cpus[i], mask);
826
827 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
828 rcu_read_unlock();
829 free_cpumask_var(mask);
830
831 return len < PAGE_SIZE ? len : -EINVAL;
832}
833
834static ssize_t store_rps_map(struct netdev_rx_queue *queue,
835 const char *buf, size_t len)
836{
837 struct rps_map *old_map, *map;
838 cpumask_var_t mask;
839 int err, cpu, i;
840 static DEFINE_MUTEX(rps_map_mutex);
841
842 if (!capable(CAP_NET_ADMIN))
843 return -EPERM;
844
845 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
846 return -ENOMEM;
847
848 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
849 if (err) {
850 free_cpumask_var(mask);
851 return err;
852 }
853
854 if (!cpumask_empty(mask)) {
855 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
856 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
857 if (cpumask_empty(mask)) {
858 free_cpumask_var(mask);
859 return -EINVAL;
860 }
861 }
862
863 map = kzalloc(max_t(unsigned int,
864 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
865 GFP_KERNEL);
866 if (!map) {
867 free_cpumask_var(mask);
868 return -ENOMEM;
869 }
870
871 i = 0;
872 for_each_cpu_and(cpu, mask, cpu_online_mask)
873 map->cpus[i++] = cpu;
874
875 if (i) {
876 map->len = i;
877 } else {
878 kfree(map);
879 map = NULL;
880 }
881
882 mutex_lock(&rps_map_mutex);
883 old_map = rcu_dereference_protected(queue->rps_map,
884 mutex_is_locked(&rps_map_mutex));
885 rcu_assign_pointer(queue->rps_map, map);
886
887 if (map)
888 static_branch_inc(&rps_needed);
889 if (old_map)
890 static_branch_dec(&rps_needed);
891
892 mutex_unlock(&rps_map_mutex);
893
894 if (old_map)
895 kfree_rcu(old_map, rcu);
896
897 free_cpumask_var(mask);
898 return len;
899}
900
901static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
902 char *buf)
903{
904 struct rps_dev_flow_table *flow_table;
905 unsigned long val = 0;
906
907 rcu_read_lock();
908 flow_table = rcu_dereference(queue->rps_flow_table);
909 if (flow_table)
910 val = (unsigned long)flow_table->mask + 1;
911 rcu_read_unlock();
912
913 return sysfs_emit(buf, "%lu\n", val);
914}
915
916static void rps_dev_flow_table_release(struct rcu_head *rcu)
917{
918 struct rps_dev_flow_table *table = container_of(rcu,
919 struct rps_dev_flow_table, rcu);
920 vfree(table);
921}
922
923static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
924 const char *buf, size_t len)
925{
926 unsigned long mask, count;
927 struct rps_dev_flow_table *table, *old_table;
928 static DEFINE_SPINLOCK(rps_dev_flow_lock);
929 int rc;
930
931 if (!capable(CAP_NET_ADMIN))
932 return -EPERM;
933
934 rc = kstrtoul(buf, 0, &count);
935 if (rc < 0)
936 return rc;
937
938 if (count) {
939 mask = count - 1;
940 /* mask = roundup_pow_of_two(count) - 1;
941 * without overflows...
942 */
943 while ((mask | (mask >> 1)) != mask)
944 mask |= (mask >> 1);
945 /* On 64 bit arches, must check mask fits in table->mask (u32),
946 * and on 32bit arches, must check
947 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
948 */
949#if BITS_PER_LONG > 32
950 if (mask > (unsigned long)(u32)mask)
951 return -EINVAL;
952#else
953 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
954 / sizeof(struct rps_dev_flow)) {
955 /* Enforce a limit to prevent overflow */
956 return -EINVAL;
957 }
958#endif
959 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
960 if (!table)
961 return -ENOMEM;
962
963 table->mask = mask;
964 for (count = 0; count <= mask; count++)
965 table->flows[count].cpu = RPS_NO_CPU;
966 } else {
967 table = NULL;
968 }
969
970 spin_lock(&rps_dev_flow_lock);
971 old_table = rcu_dereference_protected(queue->rps_flow_table,
972 lockdep_is_held(&rps_dev_flow_lock));
973 rcu_assign_pointer(queue->rps_flow_table, table);
974 spin_unlock(&rps_dev_flow_lock);
975
976 if (old_table)
977 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
978
979 return len;
980}
981
982static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
983 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
984
985static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
986 = __ATTR(rps_flow_cnt, 0644,
987 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
988#endif /* CONFIG_RPS */
989
990static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
991#ifdef CONFIG_RPS
992 &rps_cpus_attribute.attr,
993 &rps_dev_flow_table_cnt_attribute.attr,
994#endif
995 NULL
996};
997ATTRIBUTE_GROUPS(rx_queue_default);
998
999static void rx_queue_release(struct kobject *kobj)
1000{
1001 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1002#ifdef CONFIG_RPS
1003 struct rps_map *map;
1004 struct rps_dev_flow_table *flow_table;
1005
1006 map = rcu_dereference_protected(queue->rps_map, 1);
1007 if (map) {
1008 RCU_INIT_POINTER(queue->rps_map, NULL);
1009 kfree_rcu(map, rcu);
1010 }
1011
1012 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1013 if (flow_table) {
1014 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1015 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1016 }
1017#endif
1018
1019 memset(kobj, 0, sizeof(*kobj));
1020 netdev_put(queue->dev, &queue->dev_tracker);
1021}
1022
1023static const void *rx_queue_namespace(const struct kobject *kobj)
1024{
1025 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1026 struct device *dev = &queue->dev->dev;
1027 const void *ns = NULL;
1028
1029 if (dev->class && dev->class->ns_type)
1030 ns = dev->class->namespace(dev);
1031
1032 return ns;
1033}
1034
1035static void rx_queue_get_ownership(const struct kobject *kobj,
1036 kuid_t *uid, kgid_t *gid)
1037{
1038 const struct net *net = rx_queue_namespace(kobj);
1039
1040 net_ns_get_ownership(net, uid, gid);
1041}
1042
1043static struct kobj_type rx_queue_ktype __ro_after_init = {
1044 .sysfs_ops = &rx_queue_sysfs_ops,
1045 .release = rx_queue_release,
1046 .default_groups = rx_queue_default_groups,
1047 .namespace = rx_queue_namespace,
1048 .get_ownership = rx_queue_get_ownership,
1049};
1050
1051static int rx_queue_add_kobject(struct net_device *dev, int index)
1052{
1053 struct netdev_rx_queue *queue = dev->_rx + index;
1054 struct kobject *kobj = &queue->kobj;
1055 int error = 0;
1056
1057 /* Kobject_put later will trigger rx_queue_release call which
1058 * decreases dev refcount: Take that reference here
1059 */
1060 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1061
1062 kobj->kset = dev->queues_kset;
1063 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1064 "rx-%u", index);
1065 if (error)
1066 goto err;
1067
1068 if (dev->sysfs_rx_queue_group) {
1069 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1070 if (error)
1071 goto err;
1072 }
1073
1074 kobject_uevent(kobj, KOBJ_ADD);
1075
1076 return error;
1077
1078err:
1079 kobject_put(kobj);
1080 return error;
1081}
1082
1083static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1084 kgid_t kgid)
1085{
1086 struct netdev_rx_queue *queue = dev->_rx + index;
1087 struct kobject *kobj = &queue->kobj;
1088 int error;
1089
1090 error = sysfs_change_owner(kobj, kuid, kgid);
1091 if (error)
1092 return error;
1093
1094 if (dev->sysfs_rx_queue_group)
1095 error = sysfs_group_change_owner(
1096 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1097
1098 return error;
1099}
1100#endif /* CONFIG_SYSFS */
1101
1102int
1103net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1104{
1105#ifdef CONFIG_SYSFS
1106 int i;
1107 int error = 0;
1108
1109#ifndef CONFIG_RPS
1110 if (!dev->sysfs_rx_queue_group)
1111 return 0;
1112#endif
1113 for (i = old_num; i < new_num; i++) {
1114 error = rx_queue_add_kobject(dev, i);
1115 if (error) {
1116 new_num = old_num;
1117 break;
1118 }
1119 }
1120
1121 while (--i >= new_num) {
1122 struct kobject *kobj = &dev->_rx[i].kobj;
1123
1124 if (!refcount_read(&dev_net(dev)->ns.count))
1125 kobj->uevent_suppress = 1;
1126 if (dev->sysfs_rx_queue_group)
1127 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1128 kobject_put(kobj);
1129 }
1130
1131 return error;
1132#else
1133 return 0;
1134#endif
1135}
1136
1137static int net_rx_queue_change_owner(struct net_device *dev, int num,
1138 kuid_t kuid, kgid_t kgid)
1139{
1140#ifdef CONFIG_SYSFS
1141 int error = 0;
1142 int i;
1143
1144#ifndef CONFIG_RPS
1145 if (!dev->sysfs_rx_queue_group)
1146 return 0;
1147#endif
1148 for (i = 0; i < num; i++) {
1149 error = rx_queue_change_owner(dev, i, kuid, kgid);
1150 if (error)
1151 break;
1152 }
1153
1154 return error;
1155#else
1156 return 0;
1157#endif
1158}
1159
1160#ifdef CONFIG_SYSFS
1161/*
1162 * netdev_queue sysfs structures and functions.
1163 */
1164struct netdev_queue_attribute {
1165 struct attribute attr;
1166 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1167 ssize_t (*store)(struct netdev_queue *queue,
1168 const char *buf, size_t len);
1169};
1170#define to_netdev_queue_attr(_attr) \
1171 container_of(_attr, struct netdev_queue_attribute, attr)
1172
1173#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1174
1175static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1176 struct attribute *attr, char *buf)
1177{
1178 const struct netdev_queue_attribute *attribute
1179 = to_netdev_queue_attr(attr);
1180 struct netdev_queue *queue = to_netdev_queue(kobj);
1181
1182 if (!attribute->show)
1183 return -EIO;
1184
1185 return attribute->show(queue, buf);
1186}
1187
1188static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1189 struct attribute *attr,
1190 const char *buf, size_t count)
1191{
1192 const struct netdev_queue_attribute *attribute
1193 = to_netdev_queue_attr(attr);
1194 struct netdev_queue *queue = to_netdev_queue(kobj);
1195
1196 if (!attribute->store)
1197 return -EIO;
1198
1199 return attribute->store(queue, buf, count);
1200}
1201
1202static const struct sysfs_ops netdev_queue_sysfs_ops = {
1203 .show = netdev_queue_attr_show,
1204 .store = netdev_queue_attr_store,
1205};
1206
1207static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1208{
1209 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1210
1211 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1212}
1213
1214static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1215{
1216 struct net_device *dev = queue->dev;
1217 unsigned int i;
1218
1219 i = queue - dev->_tx;
1220 BUG_ON(i >= dev->num_tx_queues);
1221
1222 return i;
1223}
1224
1225static ssize_t traffic_class_show(struct netdev_queue *queue,
1226 char *buf)
1227{
1228 struct net_device *dev = queue->dev;
1229 int num_tc, tc;
1230 int index;
1231
1232 if (!netif_is_multiqueue(dev))
1233 return -ENOENT;
1234
1235 if (!rtnl_trylock())
1236 return restart_syscall();
1237
1238 index = get_netdev_queue_index(queue);
1239
1240 /* If queue belongs to subordinate dev use its TC mapping */
1241 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1242
1243 num_tc = dev->num_tc;
1244 tc = netdev_txq_to_tc(dev, index);
1245
1246 rtnl_unlock();
1247
1248 if (tc < 0)
1249 return -EINVAL;
1250
1251 /* We can report the traffic class one of two ways:
1252 * Subordinate device traffic classes are reported with the traffic
1253 * class first, and then the subordinate class so for example TC0 on
1254 * subordinate device 2 will be reported as "0-2". If the queue
1255 * belongs to the root device it will be reported with just the
1256 * traffic class, so just "0" for TC 0 for example.
1257 */
1258 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1259 sysfs_emit(buf, "%d\n", tc);
1260}
1261
1262#ifdef CONFIG_XPS
1263static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1264 char *buf)
1265{
1266 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1267}
1268
1269static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1270 const char *buf, size_t len)
1271{
1272 struct net_device *dev = queue->dev;
1273 int err, index = get_netdev_queue_index(queue);
1274 u32 rate = 0;
1275
1276 if (!capable(CAP_NET_ADMIN))
1277 return -EPERM;
1278
1279 /* The check is also done later; this helps returning early without
1280 * hitting the trylock/restart below.
1281 */
1282 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1283 return -EOPNOTSUPP;
1284
1285 err = kstrtou32(buf, 10, &rate);
1286 if (err < 0)
1287 return err;
1288
1289 if (!rtnl_trylock())
1290 return restart_syscall();
1291
1292 err = -EOPNOTSUPP;
1293 if (dev->netdev_ops->ndo_set_tx_maxrate)
1294 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1295
1296 rtnl_unlock();
1297 if (!err) {
1298 queue->tx_maxrate = rate;
1299 return len;
1300 }
1301 return err;
1302}
1303
1304static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1305 = __ATTR_RW(tx_maxrate);
1306#endif
1307
1308static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1309 = __ATTR_RO(tx_timeout);
1310
1311static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1312 = __ATTR_RO(traffic_class);
1313
1314#ifdef CONFIG_BQL
1315/*
1316 * Byte queue limits sysfs structures and functions.
1317 */
1318static ssize_t bql_show(char *buf, unsigned int value)
1319{
1320 return sysfs_emit(buf, "%u\n", value);
1321}
1322
1323static ssize_t bql_set(const char *buf, const size_t count,
1324 unsigned int *pvalue)
1325{
1326 unsigned int value;
1327 int err;
1328
1329 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1330 value = DQL_MAX_LIMIT;
1331 } else {
1332 err = kstrtouint(buf, 10, &value);
1333 if (err < 0)
1334 return err;
1335 if (value > DQL_MAX_LIMIT)
1336 return -EINVAL;
1337 }
1338
1339 *pvalue = value;
1340
1341 return count;
1342}
1343
1344static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1345 char *buf)
1346{
1347 struct dql *dql = &queue->dql;
1348
1349 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1350}
1351
1352static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1353 const char *buf, size_t len)
1354{
1355 struct dql *dql = &queue->dql;
1356 unsigned int value;
1357 int err;
1358
1359 err = kstrtouint(buf, 10, &value);
1360 if (err < 0)
1361 return err;
1362
1363 dql->slack_hold_time = msecs_to_jiffies(value);
1364
1365 return len;
1366}
1367
1368static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1369 = __ATTR(hold_time, 0644,
1370 bql_show_hold_time, bql_set_hold_time);
1371
1372static ssize_t bql_show_inflight(struct netdev_queue *queue,
1373 char *buf)
1374{
1375 struct dql *dql = &queue->dql;
1376
1377 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1378}
1379
1380static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1381 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1382
1383#define BQL_ATTR(NAME, FIELD) \
1384static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1385 char *buf) \
1386{ \
1387 return bql_show(buf, queue->dql.FIELD); \
1388} \
1389 \
1390static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1391 const char *buf, size_t len) \
1392{ \
1393 return bql_set(buf, len, &queue->dql.FIELD); \
1394} \
1395 \
1396static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1397 = __ATTR(NAME, 0644, \
1398 bql_show_ ## NAME, bql_set_ ## NAME)
1399
1400BQL_ATTR(limit, limit);
1401BQL_ATTR(limit_max, max_limit);
1402BQL_ATTR(limit_min, min_limit);
1403
1404static struct attribute *dql_attrs[] __ro_after_init = {
1405 &bql_limit_attribute.attr,
1406 &bql_limit_max_attribute.attr,
1407 &bql_limit_min_attribute.attr,
1408 &bql_hold_time_attribute.attr,
1409 &bql_inflight_attribute.attr,
1410 NULL
1411};
1412
1413static const struct attribute_group dql_group = {
1414 .name = "byte_queue_limits",
1415 .attrs = dql_attrs,
1416};
1417#endif /* CONFIG_BQL */
1418
1419#ifdef CONFIG_XPS
1420static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1421 int tc, char *buf, enum xps_map_type type)
1422{
1423 struct xps_dev_maps *dev_maps;
1424 unsigned long *mask;
1425 unsigned int nr_ids;
1426 int j, len;
1427
1428 rcu_read_lock();
1429 dev_maps = rcu_dereference(dev->xps_maps[type]);
1430
1431 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1432 * when dev_maps hasn't been allocated yet, to be backward compatible.
1433 */
1434 nr_ids = dev_maps ? dev_maps->nr_ids :
1435 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1436
1437 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1438 if (!mask) {
1439 rcu_read_unlock();
1440 return -ENOMEM;
1441 }
1442
1443 if (!dev_maps || tc >= dev_maps->num_tc)
1444 goto out_no_maps;
1445
1446 for (j = 0; j < nr_ids; j++) {
1447 int i, tci = j * dev_maps->num_tc + tc;
1448 struct xps_map *map;
1449
1450 map = rcu_dereference(dev_maps->attr_map[tci]);
1451 if (!map)
1452 continue;
1453
1454 for (i = map->len; i--;) {
1455 if (map->queues[i] == index) {
1456 __set_bit(j, mask);
1457 break;
1458 }
1459 }
1460 }
1461out_no_maps:
1462 rcu_read_unlock();
1463
1464 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1465 bitmap_free(mask);
1466
1467 return len < PAGE_SIZE ? len : -EINVAL;
1468}
1469
1470static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1471{
1472 struct net_device *dev = queue->dev;
1473 unsigned int index;
1474 int len, tc;
1475
1476 if (!netif_is_multiqueue(dev))
1477 return -ENOENT;
1478
1479 index = get_netdev_queue_index(queue);
1480
1481 if (!rtnl_trylock())
1482 return restart_syscall();
1483
1484 /* If queue belongs to subordinate dev use its map */
1485 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1486
1487 tc = netdev_txq_to_tc(dev, index);
1488 if (tc < 0) {
1489 rtnl_unlock();
1490 return -EINVAL;
1491 }
1492
1493 /* Make sure the subordinate device can't be freed */
1494 get_device(&dev->dev);
1495 rtnl_unlock();
1496
1497 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1498
1499 put_device(&dev->dev);
1500 return len;
1501}
1502
1503static ssize_t xps_cpus_store(struct netdev_queue *queue,
1504 const char *buf, size_t len)
1505{
1506 struct net_device *dev = queue->dev;
1507 unsigned int index;
1508 cpumask_var_t mask;
1509 int err;
1510
1511 if (!netif_is_multiqueue(dev))
1512 return -ENOENT;
1513
1514 if (!capable(CAP_NET_ADMIN))
1515 return -EPERM;
1516
1517 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1518 return -ENOMEM;
1519
1520 index = get_netdev_queue_index(queue);
1521
1522 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1523 if (err) {
1524 free_cpumask_var(mask);
1525 return err;
1526 }
1527
1528 if (!rtnl_trylock()) {
1529 free_cpumask_var(mask);
1530 return restart_syscall();
1531 }
1532
1533 err = netif_set_xps_queue(dev, mask, index);
1534 rtnl_unlock();
1535
1536 free_cpumask_var(mask);
1537
1538 return err ? : len;
1539}
1540
1541static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1542 = __ATTR_RW(xps_cpus);
1543
1544static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1545{
1546 struct net_device *dev = queue->dev;
1547 unsigned int index;
1548 int tc;
1549
1550 index = get_netdev_queue_index(queue);
1551
1552 if (!rtnl_trylock())
1553 return restart_syscall();
1554
1555 tc = netdev_txq_to_tc(dev, index);
1556 rtnl_unlock();
1557 if (tc < 0)
1558 return -EINVAL;
1559
1560 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1561}
1562
1563static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1564 size_t len)
1565{
1566 struct net_device *dev = queue->dev;
1567 struct net *net = dev_net(dev);
1568 unsigned long *mask;
1569 unsigned int index;
1570 int err;
1571
1572 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1573 return -EPERM;
1574
1575 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1576 if (!mask)
1577 return -ENOMEM;
1578
1579 index = get_netdev_queue_index(queue);
1580
1581 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1582 if (err) {
1583 bitmap_free(mask);
1584 return err;
1585 }
1586
1587 if (!rtnl_trylock()) {
1588 bitmap_free(mask);
1589 return restart_syscall();
1590 }
1591
1592 cpus_read_lock();
1593 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1594 cpus_read_unlock();
1595
1596 rtnl_unlock();
1597
1598 bitmap_free(mask);
1599 return err ? : len;
1600}
1601
1602static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1603 = __ATTR_RW(xps_rxqs);
1604#endif /* CONFIG_XPS */
1605
1606static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1607 &queue_trans_timeout.attr,
1608 &queue_traffic_class.attr,
1609#ifdef CONFIG_XPS
1610 &xps_cpus_attribute.attr,
1611 &xps_rxqs_attribute.attr,
1612 &queue_tx_maxrate.attr,
1613#endif
1614 NULL
1615};
1616ATTRIBUTE_GROUPS(netdev_queue_default);
1617
1618static void netdev_queue_release(struct kobject *kobj)
1619{
1620 struct netdev_queue *queue = to_netdev_queue(kobj);
1621
1622 memset(kobj, 0, sizeof(*kobj));
1623 netdev_put(queue->dev, &queue->dev_tracker);
1624}
1625
1626static const void *netdev_queue_namespace(const struct kobject *kobj)
1627{
1628 struct netdev_queue *queue = to_netdev_queue(kobj);
1629 struct device *dev = &queue->dev->dev;
1630 const void *ns = NULL;
1631
1632 if (dev->class && dev->class->ns_type)
1633 ns = dev->class->namespace(dev);
1634
1635 return ns;
1636}
1637
1638static void netdev_queue_get_ownership(const struct kobject *kobj,
1639 kuid_t *uid, kgid_t *gid)
1640{
1641 const struct net *net = netdev_queue_namespace(kobj);
1642
1643 net_ns_get_ownership(net, uid, gid);
1644}
1645
1646static struct kobj_type netdev_queue_ktype __ro_after_init = {
1647 .sysfs_ops = &netdev_queue_sysfs_ops,
1648 .release = netdev_queue_release,
1649 .default_groups = netdev_queue_default_groups,
1650 .namespace = netdev_queue_namespace,
1651 .get_ownership = netdev_queue_get_ownership,
1652};
1653
1654static int netdev_queue_add_kobject(struct net_device *dev, int index)
1655{
1656 struct netdev_queue *queue = dev->_tx + index;
1657 struct kobject *kobj = &queue->kobj;
1658 int error = 0;
1659
1660 /* Kobject_put later will trigger netdev_queue_release call
1661 * which decreases dev refcount: Take that reference here
1662 */
1663 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1664
1665 kobj->kset = dev->queues_kset;
1666 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1667 "tx-%u", index);
1668 if (error)
1669 goto err;
1670
1671#ifdef CONFIG_BQL
1672 error = sysfs_create_group(kobj, &dql_group);
1673 if (error)
1674 goto err;
1675#endif
1676
1677 kobject_uevent(kobj, KOBJ_ADD);
1678 return 0;
1679
1680err:
1681 kobject_put(kobj);
1682 return error;
1683}
1684
1685static int tx_queue_change_owner(struct net_device *ndev, int index,
1686 kuid_t kuid, kgid_t kgid)
1687{
1688 struct netdev_queue *queue = ndev->_tx + index;
1689 struct kobject *kobj = &queue->kobj;
1690 int error;
1691
1692 error = sysfs_change_owner(kobj, kuid, kgid);
1693 if (error)
1694 return error;
1695
1696#ifdef CONFIG_BQL
1697 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1698#endif
1699 return error;
1700}
1701#endif /* CONFIG_SYSFS */
1702
1703int
1704netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1705{
1706#ifdef CONFIG_SYSFS
1707 int i;
1708 int error = 0;
1709
1710 /* Tx queue kobjects are allowed to be updated when a device is being
1711 * unregistered, but solely to remove queues from qdiscs. Any path
1712 * adding queues should be fixed.
1713 */
1714 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1715 "New queues can't be registered after device unregistration.");
1716
1717 for (i = old_num; i < new_num; i++) {
1718 error = netdev_queue_add_kobject(dev, i);
1719 if (error) {
1720 new_num = old_num;
1721 break;
1722 }
1723 }
1724
1725 while (--i >= new_num) {
1726 struct netdev_queue *queue = dev->_tx + i;
1727
1728 if (!refcount_read(&dev_net(dev)->ns.count))
1729 queue->kobj.uevent_suppress = 1;
1730#ifdef CONFIG_BQL
1731 sysfs_remove_group(&queue->kobj, &dql_group);
1732#endif
1733 kobject_put(&queue->kobj);
1734 }
1735
1736 return error;
1737#else
1738 return 0;
1739#endif /* CONFIG_SYSFS */
1740}
1741
1742static int net_tx_queue_change_owner(struct net_device *dev, int num,
1743 kuid_t kuid, kgid_t kgid)
1744{
1745#ifdef CONFIG_SYSFS
1746 int error = 0;
1747 int i;
1748
1749 for (i = 0; i < num; i++) {
1750 error = tx_queue_change_owner(dev, i, kuid, kgid);
1751 if (error)
1752 break;
1753 }
1754
1755 return error;
1756#else
1757 return 0;
1758#endif /* CONFIG_SYSFS */
1759}
1760
1761static int register_queue_kobjects(struct net_device *dev)
1762{
1763 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1764
1765#ifdef CONFIG_SYSFS
1766 dev->queues_kset = kset_create_and_add("queues",
1767 NULL, &dev->dev.kobj);
1768 if (!dev->queues_kset)
1769 return -ENOMEM;
1770 real_rx = dev->real_num_rx_queues;
1771#endif
1772 real_tx = dev->real_num_tx_queues;
1773
1774 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1775 if (error)
1776 goto error;
1777 rxq = real_rx;
1778
1779 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1780 if (error)
1781 goto error;
1782 txq = real_tx;
1783
1784 return 0;
1785
1786error:
1787 netdev_queue_update_kobjects(dev, txq, 0);
1788 net_rx_queue_update_kobjects(dev, rxq, 0);
1789#ifdef CONFIG_SYSFS
1790 kset_unregister(dev->queues_kset);
1791#endif
1792 return error;
1793}
1794
1795static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1796{
1797 int error = 0, real_rx = 0, real_tx = 0;
1798
1799#ifdef CONFIG_SYSFS
1800 if (ndev->queues_kset) {
1801 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1802 if (error)
1803 return error;
1804 }
1805 real_rx = ndev->real_num_rx_queues;
1806#endif
1807 real_tx = ndev->real_num_tx_queues;
1808
1809 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1810 if (error)
1811 return error;
1812
1813 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1814 if (error)
1815 return error;
1816
1817 return 0;
1818}
1819
1820static void remove_queue_kobjects(struct net_device *dev)
1821{
1822 int real_rx = 0, real_tx = 0;
1823
1824#ifdef CONFIG_SYSFS
1825 real_rx = dev->real_num_rx_queues;
1826#endif
1827 real_tx = dev->real_num_tx_queues;
1828
1829 net_rx_queue_update_kobjects(dev, real_rx, 0);
1830 netdev_queue_update_kobjects(dev, real_tx, 0);
1831
1832 dev->real_num_rx_queues = 0;
1833 dev->real_num_tx_queues = 0;
1834#ifdef CONFIG_SYSFS
1835 kset_unregister(dev->queues_kset);
1836#endif
1837}
1838
1839static bool net_current_may_mount(void)
1840{
1841 struct net *net = current->nsproxy->net_ns;
1842
1843 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1844}
1845
1846static void *net_grab_current_ns(void)
1847{
1848 struct net *ns = current->nsproxy->net_ns;
1849#ifdef CONFIG_NET_NS
1850 if (ns)
1851 refcount_inc(&ns->passive);
1852#endif
1853 return ns;
1854}
1855
1856static const void *net_initial_ns(void)
1857{
1858 return &init_net;
1859}
1860
1861static const void *net_netlink_ns(struct sock *sk)
1862{
1863 return sock_net(sk);
1864}
1865
1866const struct kobj_ns_type_operations net_ns_type_operations = {
1867 .type = KOBJ_NS_TYPE_NET,
1868 .current_may_mount = net_current_may_mount,
1869 .grab_current_ns = net_grab_current_ns,
1870 .netlink_ns = net_netlink_ns,
1871 .initial_ns = net_initial_ns,
1872 .drop_ns = net_drop_ns,
1873};
1874EXPORT_SYMBOL_GPL(net_ns_type_operations);
1875
1876static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1877{
1878 const struct net_device *dev = to_net_dev(d);
1879 int retval;
1880
1881 /* pass interface to uevent. */
1882 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1883 if (retval)
1884 goto exit;
1885
1886 /* pass ifindex to uevent.
1887 * ifindex is useful as it won't change (interface name may change)
1888 * and is what RtNetlink uses natively.
1889 */
1890 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1891
1892exit:
1893 return retval;
1894}
1895
1896/*
1897 * netdev_release -- destroy and free a dead device.
1898 * Called when last reference to device kobject is gone.
1899 */
1900static void netdev_release(struct device *d)
1901{
1902 struct net_device *dev = to_net_dev(d);
1903
1904 BUG_ON(dev->reg_state != NETREG_RELEASED);
1905
1906 /* no need to wait for rcu grace period:
1907 * device is dead and about to be freed.
1908 */
1909 kfree(rcu_access_pointer(dev->ifalias));
1910 netdev_freemem(dev);
1911}
1912
1913static const void *net_namespace(const struct device *d)
1914{
1915 const struct net_device *dev = to_net_dev(d);
1916
1917 return dev_net(dev);
1918}
1919
1920static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1921{
1922 const struct net_device *dev = to_net_dev(d);
1923 const struct net *net = dev_net(dev);
1924
1925 net_ns_get_ownership(net, uid, gid);
1926}
1927
1928static struct class net_class __ro_after_init = {
1929 .name = "net",
1930 .dev_release = netdev_release,
1931 .dev_groups = net_class_groups,
1932 .dev_uevent = netdev_uevent,
1933 .ns_type = &net_ns_type_operations,
1934 .namespace = net_namespace,
1935 .get_ownership = net_get_ownership,
1936};
1937
1938#ifdef CONFIG_OF
1939static int of_dev_node_match(struct device *dev, const void *data)
1940{
1941 for (; dev; dev = dev->parent) {
1942 if (dev->of_node == data)
1943 return 1;
1944 }
1945
1946 return 0;
1947}
1948
1949/*
1950 * of_find_net_device_by_node - lookup the net device for the device node
1951 * @np: OF device node
1952 *
1953 * Looks up the net_device structure corresponding with the device node.
1954 * If successful, returns a pointer to the net_device with the embedded
1955 * struct device refcount incremented by one, or NULL on failure. The
1956 * refcount must be dropped when done with the net_device.
1957 */
1958struct net_device *of_find_net_device_by_node(struct device_node *np)
1959{
1960 struct device *dev;
1961
1962 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1963 if (!dev)
1964 return NULL;
1965
1966 return to_net_dev(dev);
1967}
1968EXPORT_SYMBOL(of_find_net_device_by_node);
1969#endif
1970
1971/* Delete sysfs entries but hold kobject reference until after all
1972 * netdev references are gone.
1973 */
1974void netdev_unregister_kobject(struct net_device *ndev)
1975{
1976 struct device *dev = &ndev->dev;
1977
1978 if (!refcount_read(&dev_net(ndev)->ns.count))
1979 dev_set_uevent_suppress(dev, 1);
1980
1981 kobject_get(&dev->kobj);
1982
1983 remove_queue_kobjects(ndev);
1984
1985 pm_runtime_set_memalloc_noio(dev, false);
1986
1987 device_del(dev);
1988}
1989
1990/* Create sysfs entries for network device. */
1991int netdev_register_kobject(struct net_device *ndev)
1992{
1993 struct device *dev = &ndev->dev;
1994 const struct attribute_group **groups = ndev->sysfs_groups;
1995 int error = 0;
1996
1997 device_initialize(dev);
1998 dev->class = &net_class;
1999 dev->platform_data = ndev;
2000 dev->groups = groups;
2001
2002 dev_set_name(dev, "%s", ndev->name);
2003
2004#ifdef CONFIG_SYSFS
2005 /* Allow for a device specific group */
2006 if (*groups)
2007 groups++;
2008
2009 *groups++ = &netstat_group;
2010
2011 if (wireless_group_needed(ndev))
2012 *groups++ = &wireless_group;
2013#endif /* CONFIG_SYSFS */
2014
2015 error = device_add(dev);
2016 if (error)
2017 return error;
2018
2019 error = register_queue_kobjects(ndev);
2020 if (error) {
2021 device_del(dev);
2022 return error;
2023 }
2024
2025 pm_runtime_set_memalloc_noio(dev, true);
2026
2027 return error;
2028}
2029
2030/* Change owner for sysfs entries when moving network devices across network
2031 * namespaces owned by different user namespaces.
2032 */
2033int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2034 const struct net *net_new)
2035{
2036 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2037 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2038 struct device *dev = &ndev->dev;
2039 int error;
2040
2041 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2042 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2043
2044 /* The network namespace was changed but the owning user namespace is
2045 * identical so there's no need to change the owner of sysfs entries.
2046 */
2047 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2048 return 0;
2049
2050 error = device_change_owner(dev, new_uid, new_gid);
2051 if (error)
2052 return error;
2053
2054 error = queue_change_owner(ndev, new_uid, new_gid);
2055 if (error)
2056 return error;
2057
2058 return 0;
2059}
2060
2061int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2062 const void *ns)
2063{
2064 return class_create_file_ns(&net_class, class_attr, ns);
2065}
2066EXPORT_SYMBOL(netdev_class_create_file_ns);
2067
2068void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2069 const void *ns)
2070{
2071 class_remove_file_ns(&net_class, class_attr, ns);
2072}
2073EXPORT_SYMBOL(netdev_class_remove_file_ns);
2074
2075int __init netdev_kobject_init(void)
2076{
2077 kobj_ns_type_register(&net_ns_type_operations);
2078 return class_register(&net_class);
2079}