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 <net/wext.h>
24
25#include "net-sysfs.h"
26
27#ifdef CONFIG_SYSFS
28static const char fmt_hex[] = "%#x\n";
29static const char fmt_long_hex[] = "%#lx\n";
30static const char fmt_dec[] = "%d\n";
31static const char fmt_udec[] = "%u\n";
32static const char fmt_ulong[] = "%lu\n";
33static const char fmt_u64[] = "%llu\n";
34
35static inline int dev_isalive(const struct net_device *dev)
36{
37 return dev->reg_state <= NETREG_REGISTERED;
38}
39
40/* use same locking rules as GIF* ioctl's */
41static ssize_t netdev_show(const struct device *dev,
42 struct device_attribute *attr, char *buf,
43 ssize_t (*format)(const struct net_device *, char *))
44{
45 struct net_device *net = to_net_dev(dev);
46 ssize_t ret = -EINVAL;
47
48 read_lock(&dev_base_lock);
49 if (dev_isalive(net))
50 ret = (*format)(net, buf);
51 read_unlock(&dev_base_lock);
52
53 return ret;
54}
55
56/* generate a show function for simple field */
57#define NETDEVICE_SHOW(field, format_string) \
58static ssize_t format_##field(const struct net_device *net, char *buf) \
59{ \
60 return sprintf(buf, format_string, net->field); \
61} \
62static ssize_t show_##field(struct device *dev, \
63 struct device_attribute *attr, char *buf) \
64{ \
65 return netdev_show(dev, attr, buf, format_##field); \
66}
67
68
69/* use same locking and permission rules as SIF* ioctl's */
70static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
71 const char *buf, size_t len,
72 int (*set)(struct net_device *, unsigned long))
73{
74 struct net_device *net = to_net_dev(dev);
75 char *endp;
76 unsigned long new;
77 int ret = -EINVAL;
78
79 if (!capable(CAP_NET_ADMIN))
80 return -EPERM;
81
82 new = simple_strtoul(buf, &endp, 0);
83 if (endp == buf)
84 goto err;
85
86 if (!rtnl_trylock())
87 return restart_syscall();
88
89 if (dev_isalive(net)) {
90 if ((ret = (*set)(net, new)) == 0)
91 ret = len;
92 }
93 rtnl_unlock();
94 err:
95 return ret;
96}
97
98NETDEVICE_SHOW(dev_id, fmt_hex);
99NETDEVICE_SHOW(addr_assign_type, fmt_dec);
100NETDEVICE_SHOW(addr_len, fmt_dec);
101NETDEVICE_SHOW(iflink, fmt_dec);
102NETDEVICE_SHOW(ifindex, fmt_dec);
103NETDEVICE_SHOW(type, fmt_dec);
104NETDEVICE_SHOW(link_mode, fmt_dec);
105
106/* use same locking rules as GIFHWADDR ioctl's */
107static ssize_t show_address(struct device *dev, struct device_attribute *attr,
108 char *buf)
109{
110 struct net_device *net = to_net_dev(dev);
111 ssize_t ret = -EINVAL;
112
113 read_lock(&dev_base_lock);
114 if (dev_isalive(net))
115 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
116 read_unlock(&dev_base_lock);
117 return ret;
118}
119
120static ssize_t show_broadcast(struct device *dev,
121 struct device_attribute *attr, char *buf)
122{
123 struct net_device *net = to_net_dev(dev);
124 if (dev_isalive(net))
125 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
126 return -EINVAL;
127}
128
129static ssize_t show_carrier(struct device *dev,
130 struct device_attribute *attr, char *buf)
131{
132 struct net_device *netdev = to_net_dev(dev);
133 if (netif_running(netdev)) {
134 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
135 }
136 return -EINVAL;
137}
138
139static ssize_t show_speed(struct device *dev,
140 struct device_attribute *attr, char *buf)
141{
142 struct net_device *netdev = to_net_dev(dev);
143 int ret = -EINVAL;
144
145 if (!rtnl_trylock())
146 return restart_syscall();
147
148 if (netif_running(netdev)) {
149 struct ethtool_cmd cmd;
150 if (!dev_ethtool_get_settings(netdev, &cmd))
151 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
152 }
153 rtnl_unlock();
154 return ret;
155}
156
157static ssize_t show_duplex(struct device *dev,
158 struct device_attribute *attr, char *buf)
159{
160 struct net_device *netdev = to_net_dev(dev);
161 int ret = -EINVAL;
162
163 if (!rtnl_trylock())
164 return restart_syscall();
165
166 if (netif_running(netdev)) {
167 struct ethtool_cmd cmd;
168 if (!dev_ethtool_get_settings(netdev, &cmd))
169 ret = sprintf(buf, "%s\n",
170 cmd.duplex ? "full" : "half");
171 }
172 rtnl_unlock();
173 return ret;
174}
175
176static ssize_t show_dormant(struct device *dev,
177 struct device_attribute *attr, char *buf)
178{
179 struct net_device *netdev = to_net_dev(dev);
180
181 if (netif_running(netdev))
182 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
183
184 return -EINVAL;
185}
186
187static const char *const operstates[] = {
188 "unknown",
189 "notpresent", /* currently unused */
190 "down",
191 "lowerlayerdown",
192 "testing", /* currently unused */
193 "dormant",
194 "up"
195};
196
197static ssize_t show_operstate(struct device *dev,
198 struct device_attribute *attr, char *buf)
199{
200 const struct net_device *netdev = to_net_dev(dev);
201 unsigned char operstate;
202
203 read_lock(&dev_base_lock);
204 operstate = netdev->operstate;
205 if (!netif_running(netdev))
206 operstate = IF_OPER_DOWN;
207 read_unlock(&dev_base_lock);
208
209 if (operstate >= ARRAY_SIZE(operstates))
210 return -EINVAL; /* should not happen */
211
212 return sprintf(buf, "%s\n", operstates[operstate]);
213}
214
215/* read-write attributes */
216NETDEVICE_SHOW(mtu, fmt_dec);
217
218static int change_mtu(struct net_device *net, unsigned long new_mtu)
219{
220 return dev_set_mtu(net, (int) new_mtu);
221}
222
223static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
224 const char *buf, size_t len)
225{
226 return netdev_store(dev, attr, buf, len, change_mtu);
227}
228
229NETDEVICE_SHOW(flags, fmt_hex);
230
231static int change_flags(struct net_device *net, unsigned long new_flags)
232{
233 return dev_change_flags(net, (unsigned) new_flags);
234}
235
236static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
237 const char *buf, size_t len)
238{
239 return netdev_store(dev, attr, buf, len, change_flags);
240}
241
242NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
243
244static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
245{
246 net->tx_queue_len = new_len;
247 return 0;
248}
249
250static ssize_t store_tx_queue_len(struct device *dev,
251 struct device_attribute *attr,
252 const char *buf, size_t len)
253{
254 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
255}
256
257static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t len)
259{
260 struct net_device *netdev = to_net_dev(dev);
261 size_t count = len;
262 ssize_t ret;
263
264 if (!capable(CAP_NET_ADMIN))
265 return -EPERM;
266
267 /* ignore trailing newline */
268 if (len > 0 && buf[len - 1] == '\n')
269 --count;
270
271 if (!rtnl_trylock())
272 return restart_syscall();
273 ret = dev_set_alias(netdev, buf, count);
274 rtnl_unlock();
275
276 return ret < 0 ? ret : len;
277}
278
279static ssize_t show_ifalias(struct device *dev,
280 struct device_attribute *attr, char *buf)
281{
282 const struct net_device *netdev = to_net_dev(dev);
283 ssize_t ret = 0;
284
285 if (!rtnl_trylock())
286 return restart_syscall();
287 if (netdev->ifalias)
288 ret = sprintf(buf, "%s\n", netdev->ifalias);
289 rtnl_unlock();
290 return ret;
291}
292
293NETDEVICE_SHOW(group, fmt_dec);
294
295static int change_group(struct net_device *net, unsigned long new_group)
296{
297 dev_set_group(net, (int) new_group);
298 return 0;
299}
300
301static ssize_t store_group(struct device *dev, struct device_attribute *attr,
302 const char *buf, size_t len)
303{
304 return netdev_store(dev, attr, buf, len, change_group);
305}
306
307static struct device_attribute net_class_attributes[] = {
308 __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
309 __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
310 __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
311 __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
312 __ATTR(iflink, S_IRUGO, show_iflink, NULL),
313 __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
314 __ATTR(type, S_IRUGO, show_type, NULL),
315 __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
316 __ATTR(address, S_IRUGO, show_address, NULL),
317 __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
318 __ATTR(carrier, S_IRUGO, show_carrier, NULL),
319 __ATTR(speed, S_IRUGO, show_speed, NULL),
320 __ATTR(duplex, S_IRUGO, show_duplex, NULL),
321 __ATTR(dormant, S_IRUGO, show_dormant, NULL),
322 __ATTR(operstate, S_IRUGO, show_operstate, NULL),
323 __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
324 __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
325 __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
326 store_tx_queue_len),
327 __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
328 {}
329};
330
331/* Show a given an attribute in the statistics group */
332static ssize_t netstat_show(const struct device *d,
333 struct device_attribute *attr, char *buf,
334 unsigned long offset)
335{
336 struct net_device *dev = to_net_dev(d);
337 ssize_t ret = -EINVAL;
338
339 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
340 offset % sizeof(u64) != 0);
341
342 read_lock(&dev_base_lock);
343 if (dev_isalive(dev)) {
344 struct rtnl_link_stats64 temp;
345 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
346
347 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
348 }
349 read_unlock(&dev_base_lock);
350 return ret;
351}
352
353/* generate a read-only statistics attribute */
354#define NETSTAT_ENTRY(name) \
355static ssize_t show_##name(struct device *d, \
356 struct device_attribute *attr, char *buf) \
357{ \
358 return netstat_show(d, attr, buf, \
359 offsetof(struct rtnl_link_stats64, name)); \
360} \
361static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
362
363NETSTAT_ENTRY(rx_packets);
364NETSTAT_ENTRY(tx_packets);
365NETSTAT_ENTRY(rx_bytes);
366NETSTAT_ENTRY(tx_bytes);
367NETSTAT_ENTRY(rx_errors);
368NETSTAT_ENTRY(tx_errors);
369NETSTAT_ENTRY(rx_dropped);
370NETSTAT_ENTRY(tx_dropped);
371NETSTAT_ENTRY(multicast);
372NETSTAT_ENTRY(collisions);
373NETSTAT_ENTRY(rx_length_errors);
374NETSTAT_ENTRY(rx_over_errors);
375NETSTAT_ENTRY(rx_crc_errors);
376NETSTAT_ENTRY(rx_frame_errors);
377NETSTAT_ENTRY(rx_fifo_errors);
378NETSTAT_ENTRY(rx_missed_errors);
379NETSTAT_ENTRY(tx_aborted_errors);
380NETSTAT_ENTRY(tx_carrier_errors);
381NETSTAT_ENTRY(tx_fifo_errors);
382NETSTAT_ENTRY(tx_heartbeat_errors);
383NETSTAT_ENTRY(tx_window_errors);
384NETSTAT_ENTRY(rx_compressed);
385NETSTAT_ENTRY(tx_compressed);
386
387static struct attribute *netstat_attrs[] = {
388 &dev_attr_rx_packets.attr,
389 &dev_attr_tx_packets.attr,
390 &dev_attr_rx_bytes.attr,
391 &dev_attr_tx_bytes.attr,
392 &dev_attr_rx_errors.attr,
393 &dev_attr_tx_errors.attr,
394 &dev_attr_rx_dropped.attr,
395 &dev_attr_tx_dropped.attr,
396 &dev_attr_multicast.attr,
397 &dev_attr_collisions.attr,
398 &dev_attr_rx_length_errors.attr,
399 &dev_attr_rx_over_errors.attr,
400 &dev_attr_rx_crc_errors.attr,
401 &dev_attr_rx_frame_errors.attr,
402 &dev_attr_rx_fifo_errors.attr,
403 &dev_attr_rx_missed_errors.attr,
404 &dev_attr_tx_aborted_errors.attr,
405 &dev_attr_tx_carrier_errors.attr,
406 &dev_attr_tx_fifo_errors.attr,
407 &dev_attr_tx_heartbeat_errors.attr,
408 &dev_attr_tx_window_errors.attr,
409 &dev_attr_rx_compressed.attr,
410 &dev_attr_tx_compressed.attr,
411 NULL
412};
413
414
415static struct attribute_group netstat_group = {
416 .name = "statistics",
417 .attrs = netstat_attrs,
418};
419
420#ifdef CONFIG_WIRELESS_EXT_SYSFS
421/* helper function that does all the locking etc for wireless stats */
422static ssize_t wireless_show(struct device *d, char *buf,
423 ssize_t (*format)(const struct iw_statistics *,
424 char *))
425{
426 struct net_device *dev = to_net_dev(d);
427 const struct iw_statistics *iw;
428 ssize_t ret = -EINVAL;
429
430 if (!rtnl_trylock())
431 return restart_syscall();
432 if (dev_isalive(dev)) {
433 iw = get_wireless_stats(dev);
434 if (iw)
435 ret = (*format)(iw, buf);
436 }
437 rtnl_unlock();
438
439 return ret;
440}
441
442/* show function template for wireless fields */
443#define WIRELESS_SHOW(name, field, format_string) \
444static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
445{ \
446 return sprintf(buf, format_string, iw->field); \
447} \
448static ssize_t show_iw_##name(struct device *d, \
449 struct device_attribute *attr, char *buf) \
450{ \
451 return wireless_show(d, buf, format_iw_##name); \
452} \
453static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
454
455WIRELESS_SHOW(status, status, fmt_hex);
456WIRELESS_SHOW(link, qual.qual, fmt_dec);
457WIRELESS_SHOW(level, qual.level, fmt_dec);
458WIRELESS_SHOW(noise, qual.noise, fmt_dec);
459WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
460WIRELESS_SHOW(crypt, discard.code, fmt_dec);
461WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
462WIRELESS_SHOW(misc, discard.misc, fmt_dec);
463WIRELESS_SHOW(retries, discard.retries, fmt_dec);
464WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
465
466static struct attribute *wireless_attrs[] = {
467 &dev_attr_status.attr,
468 &dev_attr_link.attr,
469 &dev_attr_level.attr,
470 &dev_attr_noise.attr,
471 &dev_attr_nwid.attr,
472 &dev_attr_crypt.attr,
473 &dev_attr_fragment.attr,
474 &dev_attr_retries.attr,
475 &dev_attr_misc.attr,
476 &dev_attr_beacon.attr,
477 NULL
478};
479
480static struct attribute_group wireless_group = {
481 .name = "wireless",
482 .attrs = wireless_attrs,
483};
484#endif
485#endif /* CONFIG_SYSFS */
486
487#ifdef CONFIG_RPS
488/*
489 * RX queue sysfs structures and functions.
490 */
491struct rx_queue_attribute {
492 struct attribute attr;
493 ssize_t (*show)(struct netdev_rx_queue *queue,
494 struct rx_queue_attribute *attr, char *buf);
495 ssize_t (*store)(struct netdev_rx_queue *queue,
496 struct rx_queue_attribute *attr, const char *buf, size_t len);
497};
498#define to_rx_queue_attr(_attr) container_of(_attr, \
499 struct rx_queue_attribute, attr)
500
501#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
502
503static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
504 char *buf)
505{
506 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
507 struct netdev_rx_queue *queue = to_rx_queue(kobj);
508
509 if (!attribute->show)
510 return -EIO;
511
512 return attribute->show(queue, attribute, buf);
513}
514
515static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
516 const char *buf, size_t count)
517{
518 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
519 struct netdev_rx_queue *queue = to_rx_queue(kobj);
520
521 if (!attribute->store)
522 return -EIO;
523
524 return attribute->store(queue, attribute, buf, count);
525}
526
527static const struct sysfs_ops rx_queue_sysfs_ops = {
528 .show = rx_queue_attr_show,
529 .store = rx_queue_attr_store,
530};
531
532static ssize_t show_rps_map(struct netdev_rx_queue *queue,
533 struct rx_queue_attribute *attribute, char *buf)
534{
535 struct rps_map *map;
536 cpumask_var_t mask;
537 size_t len = 0;
538 int i;
539
540 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
541 return -ENOMEM;
542
543 rcu_read_lock();
544 map = rcu_dereference(queue->rps_map);
545 if (map)
546 for (i = 0; i < map->len; i++)
547 cpumask_set_cpu(map->cpus[i], mask);
548
549 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
550 if (PAGE_SIZE - len < 3) {
551 rcu_read_unlock();
552 free_cpumask_var(mask);
553 return -EINVAL;
554 }
555 rcu_read_unlock();
556
557 free_cpumask_var(mask);
558 len += sprintf(buf + len, "\n");
559 return len;
560}
561
562static ssize_t store_rps_map(struct netdev_rx_queue *queue,
563 struct rx_queue_attribute *attribute,
564 const char *buf, size_t len)
565{
566 struct rps_map *old_map, *map;
567 cpumask_var_t mask;
568 int err, cpu, i;
569 static DEFINE_SPINLOCK(rps_map_lock);
570
571 if (!capable(CAP_NET_ADMIN))
572 return -EPERM;
573
574 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
575 return -ENOMEM;
576
577 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
578 if (err) {
579 free_cpumask_var(mask);
580 return err;
581 }
582
583 map = kzalloc(max_t(unsigned,
584 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
585 GFP_KERNEL);
586 if (!map) {
587 free_cpumask_var(mask);
588 return -ENOMEM;
589 }
590
591 i = 0;
592 for_each_cpu_and(cpu, mask, cpu_online_mask)
593 map->cpus[i++] = cpu;
594
595 if (i)
596 map->len = i;
597 else {
598 kfree(map);
599 map = NULL;
600 }
601
602 spin_lock(&rps_map_lock);
603 old_map = rcu_dereference_protected(queue->rps_map,
604 lockdep_is_held(&rps_map_lock));
605 rcu_assign_pointer(queue->rps_map, map);
606 spin_unlock(&rps_map_lock);
607
608 if (old_map)
609 kfree_rcu(old_map, rcu);
610
611 free_cpumask_var(mask);
612 return len;
613}
614
615static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
616 struct rx_queue_attribute *attr,
617 char *buf)
618{
619 struct rps_dev_flow_table *flow_table;
620 unsigned int val = 0;
621
622 rcu_read_lock();
623 flow_table = rcu_dereference(queue->rps_flow_table);
624 if (flow_table)
625 val = flow_table->mask + 1;
626 rcu_read_unlock();
627
628 return sprintf(buf, "%u\n", val);
629}
630
631static void rps_dev_flow_table_release_work(struct work_struct *work)
632{
633 struct rps_dev_flow_table *table = container_of(work,
634 struct rps_dev_flow_table, free_work);
635
636 vfree(table);
637}
638
639static void rps_dev_flow_table_release(struct rcu_head *rcu)
640{
641 struct rps_dev_flow_table *table = container_of(rcu,
642 struct rps_dev_flow_table, rcu);
643
644 INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
645 schedule_work(&table->free_work);
646}
647
648static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
649 struct rx_queue_attribute *attr,
650 const char *buf, size_t len)
651{
652 unsigned int count;
653 char *endp;
654 struct rps_dev_flow_table *table, *old_table;
655 static DEFINE_SPINLOCK(rps_dev_flow_lock);
656
657 if (!capable(CAP_NET_ADMIN))
658 return -EPERM;
659
660 count = simple_strtoul(buf, &endp, 0);
661 if (endp == buf)
662 return -EINVAL;
663
664 if (count) {
665 int i;
666
667 if (count > 1<<30) {
668 /* Enforce a limit to prevent overflow */
669 return -EINVAL;
670 }
671 count = roundup_pow_of_two(count);
672 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
673 if (!table)
674 return -ENOMEM;
675
676 table->mask = count - 1;
677 for (i = 0; i < count; i++)
678 table->flows[i].cpu = RPS_NO_CPU;
679 } else
680 table = NULL;
681
682 spin_lock(&rps_dev_flow_lock);
683 old_table = rcu_dereference_protected(queue->rps_flow_table,
684 lockdep_is_held(&rps_dev_flow_lock));
685 rcu_assign_pointer(queue->rps_flow_table, table);
686 spin_unlock(&rps_dev_flow_lock);
687
688 if (old_table)
689 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
690
691 return len;
692}
693
694static struct rx_queue_attribute rps_cpus_attribute =
695 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
696
697
698static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
699 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
700 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
701
702static struct attribute *rx_queue_default_attrs[] = {
703 &rps_cpus_attribute.attr,
704 &rps_dev_flow_table_cnt_attribute.attr,
705 NULL
706};
707
708static void rx_queue_release(struct kobject *kobj)
709{
710 struct netdev_rx_queue *queue = to_rx_queue(kobj);
711 struct rps_map *map;
712 struct rps_dev_flow_table *flow_table;
713
714
715 map = rcu_dereference_raw(queue->rps_map);
716 if (map) {
717 RCU_INIT_POINTER(queue->rps_map, NULL);
718 kfree_rcu(map, rcu);
719 }
720
721 flow_table = rcu_dereference_raw(queue->rps_flow_table);
722 if (flow_table) {
723 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
724 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
725 }
726
727 memset(kobj, 0, sizeof(*kobj));
728 dev_put(queue->dev);
729}
730
731static struct kobj_type rx_queue_ktype = {
732 .sysfs_ops = &rx_queue_sysfs_ops,
733 .release = rx_queue_release,
734 .default_attrs = rx_queue_default_attrs,
735};
736
737static int rx_queue_add_kobject(struct net_device *net, int index)
738{
739 struct netdev_rx_queue *queue = net->_rx + index;
740 struct kobject *kobj = &queue->kobj;
741 int error = 0;
742
743 kobj->kset = net->queues_kset;
744 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
745 "rx-%u", index);
746 if (error) {
747 kobject_put(kobj);
748 return error;
749 }
750
751 kobject_uevent(kobj, KOBJ_ADD);
752 dev_hold(queue->dev);
753
754 return error;
755}
756#endif /* CONFIG_RPS */
757
758int
759net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
760{
761#ifdef CONFIG_RPS
762 int i;
763 int error = 0;
764
765 for (i = old_num; i < new_num; i++) {
766 error = rx_queue_add_kobject(net, i);
767 if (error) {
768 new_num = old_num;
769 break;
770 }
771 }
772
773 while (--i >= new_num)
774 kobject_put(&net->_rx[i].kobj);
775
776 return error;
777#else
778 return 0;
779#endif
780}
781
782#ifdef CONFIG_XPS
783/*
784 * netdev_queue sysfs structures and functions.
785 */
786struct netdev_queue_attribute {
787 struct attribute attr;
788 ssize_t (*show)(struct netdev_queue *queue,
789 struct netdev_queue_attribute *attr, char *buf);
790 ssize_t (*store)(struct netdev_queue *queue,
791 struct netdev_queue_attribute *attr, const char *buf, size_t len);
792};
793#define to_netdev_queue_attr(_attr) container_of(_attr, \
794 struct netdev_queue_attribute, attr)
795
796#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
797
798static ssize_t netdev_queue_attr_show(struct kobject *kobj,
799 struct attribute *attr, char *buf)
800{
801 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
802 struct netdev_queue *queue = to_netdev_queue(kobj);
803
804 if (!attribute->show)
805 return -EIO;
806
807 return attribute->show(queue, attribute, buf);
808}
809
810static ssize_t netdev_queue_attr_store(struct kobject *kobj,
811 struct attribute *attr,
812 const char *buf, size_t count)
813{
814 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
815 struct netdev_queue *queue = to_netdev_queue(kobj);
816
817 if (!attribute->store)
818 return -EIO;
819
820 return attribute->store(queue, attribute, buf, count);
821}
822
823static const struct sysfs_ops netdev_queue_sysfs_ops = {
824 .show = netdev_queue_attr_show,
825 .store = netdev_queue_attr_store,
826};
827
828static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
829{
830 struct net_device *dev = queue->dev;
831 int i;
832
833 for (i = 0; i < dev->num_tx_queues; i++)
834 if (queue == &dev->_tx[i])
835 break;
836
837 BUG_ON(i >= dev->num_tx_queues);
838
839 return i;
840}
841
842
843static ssize_t show_xps_map(struct netdev_queue *queue,
844 struct netdev_queue_attribute *attribute, char *buf)
845{
846 struct net_device *dev = queue->dev;
847 struct xps_dev_maps *dev_maps;
848 cpumask_var_t mask;
849 unsigned long index;
850 size_t len = 0;
851 int i;
852
853 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
854 return -ENOMEM;
855
856 index = get_netdev_queue_index(queue);
857
858 rcu_read_lock();
859 dev_maps = rcu_dereference(dev->xps_maps);
860 if (dev_maps) {
861 for_each_possible_cpu(i) {
862 struct xps_map *map =
863 rcu_dereference(dev_maps->cpu_map[i]);
864 if (map) {
865 int j;
866 for (j = 0; j < map->len; j++) {
867 if (map->queues[j] == index) {
868 cpumask_set_cpu(i, mask);
869 break;
870 }
871 }
872 }
873 }
874 }
875 rcu_read_unlock();
876
877 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
878 if (PAGE_SIZE - len < 3) {
879 free_cpumask_var(mask);
880 return -EINVAL;
881 }
882
883 free_cpumask_var(mask);
884 len += sprintf(buf + len, "\n");
885 return len;
886}
887
888static DEFINE_MUTEX(xps_map_mutex);
889#define xmap_dereference(P) \
890 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
891
892static ssize_t store_xps_map(struct netdev_queue *queue,
893 struct netdev_queue_attribute *attribute,
894 const char *buf, size_t len)
895{
896 struct net_device *dev = queue->dev;
897 cpumask_var_t mask;
898 int err, i, cpu, pos, map_len, alloc_len, need_set;
899 unsigned long index;
900 struct xps_map *map, *new_map;
901 struct xps_dev_maps *dev_maps, *new_dev_maps;
902 int nonempty = 0;
903 int numa_node = -2;
904
905 if (!capable(CAP_NET_ADMIN))
906 return -EPERM;
907
908 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
909 return -ENOMEM;
910
911 index = get_netdev_queue_index(queue);
912
913 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
914 if (err) {
915 free_cpumask_var(mask);
916 return err;
917 }
918
919 new_dev_maps = kzalloc(max_t(unsigned,
920 XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
921 if (!new_dev_maps) {
922 free_cpumask_var(mask);
923 return -ENOMEM;
924 }
925
926 mutex_lock(&xps_map_mutex);
927
928 dev_maps = xmap_dereference(dev->xps_maps);
929
930 for_each_possible_cpu(cpu) {
931 map = dev_maps ?
932 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
933 new_map = map;
934 if (map) {
935 for (pos = 0; pos < map->len; pos++)
936 if (map->queues[pos] == index)
937 break;
938 map_len = map->len;
939 alloc_len = map->alloc_len;
940 } else
941 pos = map_len = alloc_len = 0;
942
943 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
944#ifdef CONFIG_NUMA
945 if (need_set) {
946 if (numa_node == -2)
947 numa_node = cpu_to_node(cpu);
948 else if (numa_node != cpu_to_node(cpu))
949 numa_node = -1;
950 }
951#endif
952 if (need_set && pos >= map_len) {
953 /* Need to add queue to this CPU's map */
954 if (map_len >= alloc_len) {
955 alloc_len = alloc_len ?
956 2 * alloc_len : XPS_MIN_MAP_ALLOC;
957 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
958 GFP_KERNEL,
959 cpu_to_node(cpu));
960 if (!new_map)
961 goto error;
962 new_map->alloc_len = alloc_len;
963 for (i = 0; i < map_len; i++)
964 new_map->queues[i] = map->queues[i];
965 new_map->len = map_len;
966 }
967 new_map->queues[new_map->len++] = index;
968 } else if (!need_set && pos < map_len) {
969 /* Need to remove queue from this CPU's map */
970 if (map_len > 1)
971 new_map->queues[pos] =
972 new_map->queues[--new_map->len];
973 else
974 new_map = NULL;
975 }
976 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
977 }
978
979 /* Cleanup old maps */
980 for_each_possible_cpu(cpu) {
981 map = dev_maps ?
982 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
983 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
984 kfree_rcu(map, rcu);
985 if (new_dev_maps->cpu_map[cpu])
986 nonempty = 1;
987 }
988
989 if (nonempty)
990 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
991 else {
992 kfree(new_dev_maps);
993 rcu_assign_pointer(dev->xps_maps, NULL);
994 }
995
996 if (dev_maps)
997 kfree_rcu(dev_maps, rcu);
998
999 netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1000 NUMA_NO_NODE);
1001
1002 mutex_unlock(&xps_map_mutex);
1003
1004 free_cpumask_var(mask);
1005 return len;
1006
1007error:
1008 mutex_unlock(&xps_map_mutex);
1009
1010 if (new_dev_maps)
1011 for_each_possible_cpu(i)
1012 kfree(rcu_dereference_protected(
1013 new_dev_maps->cpu_map[i],
1014 1));
1015 kfree(new_dev_maps);
1016 free_cpumask_var(mask);
1017 return -ENOMEM;
1018}
1019
1020static struct netdev_queue_attribute xps_cpus_attribute =
1021 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1022
1023static struct attribute *netdev_queue_default_attrs[] = {
1024 &xps_cpus_attribute.attr,
1025 NULL
1026};
1027
1028static void netdev_queue_release(struct kobject *kobj)
1029{
1030 struct netdev_queue *queue = to_netdev_queue(kobj);
1031 struct net_device *dev = queue->dev;
1032 struct xps_dev_maps *dev_maps;
1033 struct xps_map *map;
1034 unsigned long index;
1035 int i, pos, nonempty = 0;
1036
1037 index = get_netdev_queue_index(queue);
1038
1039 mutex_lock(&xps_map_mutex);
1040 dev_maps = xmap_dereference(dev->xps_maps);
1041
1042 if (dev_maps) {
1043 for_each_possible_cpu(i) {
1044 map = xmap_dereference(dev_maps->cpu_map[i]);
1045 if (!map)
1046 continue;
1047
1048 for (pos = 0; pos < map->len; pos++)
1049 if (map->queues[pos] == index)
1050 break;
1051
1052 if (pos < map->len) {
1053 if (map->len > 1)
1054 map->queues[pos] =
1055 map->queues[--map->len];
1056 else {
1057 RCU_INIT_POINTER(dev_maps->cpu_map[i],
1058 NULL);
1059 kfree_rcu(map, rcu);
1060 map = NULL;
1061 }
1062 }
1063 if (map)
1064 nonempty = 1;
1065 }
1066
1067 if (!nonempty) {
1068 RCU_INIT_POINTER(dev->xps_maps, NULL);
1069 kfree_rcu(dev_maps, rcu);
1070 }
1071 }
1072
1073 mutex_unlock(&xps_map_mutex);
1074
1075 memset(kobj, 0, sizeof(*kobj));
1076 dev_put(queue->dev);
1077}
1078
1079static struct kobj_type netdev_queue_ktype = {
1080 .sysfs_ops = &netdev_queue_sysfs_ops,
1081 .release = netdev_queue_release,
1082 .default_attrs = netdev_queue_default_attrs,
1083};
1084
1085static int netdev_queue_add_kobject(struct net_device *net, int index)
1086{
1087 struct netdev_queue *queue = net->_tx + index;
1088 struct kobject *kobj = &queue->kobj;
1089 int error = 0;
1090
1091 kobj->kset = net->queues_kset;
1092 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1093 "tx-%u", index);
1094 if (error) {
1095 kobject_put(kobj);
1096 return error;
1097 }
1098
1099 kobject_uevent(kobj, KOBJ_ADD);
1100 dev_hold(queue->dev);
1101
1102 return error;
1103}
1104#endif /* CONFIG_XPS */
1105
1106int
1107netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1108{
1109#ifdef CONFIG_XPS
1110 int i;
1111 int error = 0;
1112
1113 for (i = old_num; i < new_num; i++) {
1114 error = netdev_queue_add_kobject(net, i);
1115 if (error) {
1116 new_num = old_num;
1117 break;
1118 }
1119 }
1120
1121 while (--i >= new_num)
1122 kobject_put(&net->_tx[i].kobj);
1123
1124 return error;
1125#else
1126 return 0;
1127#endif
1128}
1129
1130static int register_queue_kobjects(struct net_device *net)
1131{
1132 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1133
1134#if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1135 net->queues_kset = kset_create_and_add("queues",
1136 NULL, &net->dev.kobj);
1137 if (!net->queues_kset)
1138 return -ENOMEM;
1139#endif
1140
1141#ifdef CONFIG_RPS
1142 real_rx = net->real_num_rx_queues;
1143#endif
1144 real_tx = net->real_num_tx_queues;
1145
1146 error = net_rx_queue_update_kobjects(net, 0, real_rx);
1147 if (error)
1148 goto error;
1149 rxq = real_rx;
1150
1151 error = netdev_queue_update_kobjects(net, 0, real_tx);
1152 if (error)
1153 goto error;
1154 txq = real_tx;
1155
1156 return 0;
1157
1158error:
1159 netdev_queue_update_kobjects(net, txq, 0);
1160 net_rx_queue_update_kobjects(net, rxq, 0);
1161 return error;
1162}
1163
1164static void remove_queue_kobjects(struct net_device *net)
1165{
1166 int real_rx = 0, real_tx = 0;
1167
1168#ifdef CONFIG_RPS
1169 real_rx = net->real_num_rx_queues;
1170#endif
1171 real_tx = net->real_num_tx_queues;
1172
1173 net_rx_queue_update_kobjects(net, real_rx, 0);
1174 netdev_queue_update_kobjects(net, real_tx, 0);
1175#if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1176 kset_unregister(net->queues_kset);
1177#endif
1178}
1179
1180static void *net_grab_current_ns(void)
1181{
1182 struct net *ns = current->nsproxy->net_ns;
1183#ifdef CONFIG_NET_NS
1184 if (ns)
1185 atomic_inc(&ns->passive);
1186#endif
1187 return ns;
1188}
1189
1190static const void *net_initial_ns(void)
1191{
1192 return &init_net;
1193}
1194
1195static const void *net_netlink_ns(struct sock *sk)
1196{
1197 return sock_net(sk);
1198}
1199
1200struct kobj_ns_type_operations net_ns_type_operations = {
1201 .type = KOBJ_NS_TYPE_NET,
1202 .grab_current_ns = net_grab_current_ns,
1203 .netlink_ns = net_netlink_ns,
1204 .initial_ns = net_initial_ns,
1205 .drop_ns = net_drop_ns,
1206};
1207EXPORT_SYMBOL_GPL(net_ns_type_operations);
1208
1209#ifdef CONFIG_HOTPLUG
1210static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1211{
1212 struct net_device *dev = to_net_dev(d);
1213 int retval;
1214
1215 /* pass interface to uevent. */
1216 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1217 if (retval)
1218 goto exit;
1219
1220 /* pass ifindex to uevent.
1221 * ifindex is useful as it won't change (interface name may change)
1222 * and is what RtNetlink uses natively. */
1223 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1224
1225exit:
1226 return retval;
1227}
1228#endif
1229
1230/*
1231 * netdev_release -- destroy and free a dead device.
1232 * Called when last reference to device kobject is gone.
1233 */
1234static void netdev_release(struct device *d)
1235{
1236 struct net_device *dev = to_net_dev(d);
1237
1238 BUG_ON(dev->reg_state != NETREG_RELEASED);
1239
1240 kfree(dev->ifalias);
1241 kfree((char *)dev - dev->padded);
1242}
1243
1244static const void *net_namespace(struct device *d)
1245{
1246 struct net_device *dev;
1247 dev = container_of(d, struct net_device, dev);
1248 return dev_net(dev);
1249}
1250
1251static struct class net_class = {
1252 .name = "net",
1253 .dev_release = netdev_release,
1254#ifdef CONFIG_SYSFS
1255 .dev_attrs = net_class_attributes,
1256#endif /* CONFIG_SYSFS */
1257#ifdef CONFIG_HOTPLUG
1258 .dev_uevent = netdev_uevent,
1259#endif
1260 .ns_type = &net_ns_type_operations,
1261 .namespace = net_namespace,
1262};
1263
1264/* Delete sysfs entries but hold kobject reference until after all
1265 * netdev references are gone.
1266 */
1267void netdev_unregister_kobject(struct net_device * net)
1268{
1269 struct device *dev = &(net->dev);
1270
1271 kobject_get(&dev->kobj);
1272
1273 remove_queue_kobjects(net);
1274
1275 device_del(dev);
1276}
1277
1278/* Create sysfs entries for network device. */
1279int netdev_register_kobject(struct net_device *net)
1280{
1281 struct device *dev = &(net->dev);
1282 const struct attribute_group **groups = net->sysfs_groups;
1283 int error = 0;
1284
1285 device_initialize(dev);
1286 dev->class = &net_class;
1287 dev->platform_data = net;
1288 dev->groups = groups;
1289
1290 dev_set_name(dev, "%s", net->name);
1291
1292#ifdef CONFIG_SYSFS
1293 /* Allow for a device specific group */
1294 if (*groups)
1295 groups++;
1296
1297 *groups++ = &netstat_group;
1298#ifdef CONFIG_WIRELESS_EXT_SYSFS
1299 if (net->ieee80211_ptr)
1300 *groups++ = &wireless_group;
1301#ifdef CONFIG_WIRELESS_EXT
1302 else if (net->wireless_handlers)
1303 *groups++ = &wireless_group;
1304#endif
1305#endif
1306#endif /* CONFIG_SYSFS */
1307
1308 error = device_add(dev);
1309 if (error)
1310 return error;
1311
1312 error = register_queue_kobjects(net);
1313 if (error) {
1314 device_del(dev);
1315 return error;
1316 }
1317
1318 return error;
1319}
1320
1321int netdev_class_create_file(struct class_attribute *class_attr)
1322{
1323 return class_create_file(&net_class, class_attr);
1324}
1325EXPORT_SYMBOL(netdev_class_create_file);
1326
1327void netdev_class_remove_file(struct class_attribute *class_attr)
1328{
1329 class_remove_file(&net_class, class_attr);
1330}
1331EXPORT_SYMBOL(netdev_class_remove_file);
1332
1333int netdev_kobject_init(void)
1334{
1335 kobj_ns_type_register(&net_ns_type_operations);
1336 return class_register(&net_class);
1337}
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 "net-sysfs.h"
28
29#ifdef CONFIG_SYSFS
30static const char fmt_hex[] = "%#x\n";
31static const char fmt_dec[] = "%d\n";
32static const char fmt_ulong[] = "%lu\n";
33static const char fmt_u64[] = "%llu\n";
34
35static inline int dev_isalive(const struct net_device *dev)
36{
37 return dev->reg_state <= NETREG_REGISTERED;
38}
39
40/* use same locking rules as GIF* ioctl's */
41static ssize_t netdev_show(const struct device *dev,
42 struct device_attribute *attr, char *buf,
43 ssize_t (*format)(const struct net_device *, char *))
44{
45 struct net_device *ndev = to_net_dev(dev);
46 ssize_t ret = -EINVAL;
47
48 read_lock(&dev_base_lock);
49 if (dev_isalive(ndev))
50 ret = (*format)(ndev, buf);
51 read_unlock(&dev_base_lock);
52
53 return ret;
54}
55
56/* generate a show function for simple field */
57#define NETDEVICE_SHOW(field, format_string) \
58static ssize_t format_##field(const struct net_device *dev, char *buf) \
59{ \
60 return sprintf(buf, format_string, dev->field); \
61} \
62static ssize_t field##_show(struct device *dev, \
63 struct device_attribute *attr, char *buf) \
64{ \
65 return netdev_show(dev, attr, buf, format_##field); \
66} \
67
68#define NETDEVICE_SHOW_RO(field, format_string) \
69NETDEVICE_SHOW(field, format_string); \
70static DEVICE_ATTR_RO(field)
71
72#define NETDEVICE_SHOW_RW(field, format_string) \
73NETDEVICE_SHOW(field, format_string); \
74static DEVICE_ATTR_RW(field)
75
76/* use same locking and permission rules as SIF* ioctl's */
77static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
78 const char *buf, size_t len,
79 int (*set)(struct net_device *, unsigned long))
80{
81 struct net_device *netdev = to_net_dev(dev);
82 struct net *net = dev_net(netdev);
83 unsigned long new;
84 int ret;
85
86 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
87 return -EPERM;
88
89 ret = kstrtoul(buf, 0, &new);
90 if (ret)
91 goto err;
92
93 if (!rtnl_trylock())
94 return restart_syscall();
95
96 if (dev_isalive(netdev)) {
97 ret = (*set)(netdev, new);
98 if (ret == 0)
99 ret = len;
100 }
101 rtnl_unlock();
102 err:
103 return ret;
104}
105
106NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107NETDEVICE_SHOW_RO(dev_port, fmt_dec);
108NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
109NETDEVICE_SHOW_RO(addr_len, fmt_dec);
110NETDEVICE_SHOW_RO(ifindex, fmt_dec);
111NETDEVICE_SHOW_RO(type, fmt_dec);
112NETDEVICE_SHOW_RO(link_mode, fmt_dec);
113
114static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
115 char *buf)
116{
117 struct net_device *ndev = to_net_dev(dev);
118
119 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
120}
121static DEVICE_ATTR_RO(iflink);
122
123static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
124{
125 return sprintf(buf, fmt_dec, dev->name_assign_type);
126}
127
128static ssize_t name_assign_type_show(struct device *dev,
129 struct device_attribute *attr,
130 char *buf)
131{
132 struct net_device *ndev = to_net_dev(dev);
133 ssize_t ret = -EINVAL;
134
135 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
136 ret = netdev_show(dev, attr, buf, format_name_assign_type);
137
138 return ret;
139}
140static DEVICE_ATTR_RO(name_assign_type);
141
142/* use same locking rules as GIFHWADDR ioctl's */
143static ssize_t address_show(struct device *dev, struct device_attribute *attr,
144 char *buf)
145{
146 struct net_device *ndev = to_net_dev(dev);
147 ssize_t ret = -EINVAL;
148
149 read_lock(&dev_base_lock);
150 if (dev_isalive(ndev))
151 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
152 read_unlock(&dev_base_lock);
153 return ret;
154}
155static DEVICE_ATTR_RO(address);
156
157static ssize_t broadcast_show(struct device *dev,
158 struct device_attribute *attr, char *buf)
159{
160 struct net_device *ndev = to_net_dev(dev);
161
162 if (dev_isalive(ndev))
163 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
164 return -EINVAL;
165}
166static DEVICE_ATTR_RO(broadcast);
167
168static int change_carrier(struct net_device *dev, unsigned long new_carrier)
169{
170 if (!netif_running(dev))
171 return -EINVAL;
172 return dev_change_carrier(dev, (bool)new_carrier);
173}
174
175static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
176 const char *buf, size_t len)
177{
178 return netdev_store(dev, attr, buf, len, change_carrier);
179}
180
181static ssize_t carrier_show(struct device *dev,
182 struct device_attribute *attr, char *buf)
183{
184 struct net_device *netdev = to_net_dev(dev);
185
186 if (netif_running(netdev))
187 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
188
189 return -EINVAL;
190}
191static DEVICE_ATTR_RW(carrier);
192
193static ssize_t speed_show(struct device *dev,
194 struct device_attribute *attr, char *buf)
195{
196 struct net_device *netdev = to_net_dev(dev);
197 int ret = -EINVAL;
198
199 if (!rtnl_trylock())
200 return restart_syscall();
201
202 if (netif_running(netdev)) {
203 struct ethtool_link_ksettings cmd;
204
205 if (!__ethtool_get_link_ksettings(netdev, &cmd))
206 ret = sprintf(buf, fmt_dec, cmd.base.speed);
207 }
208 rtnl_unlock();
209 return ret;
210}
211static DEVICE_ATTR_RO(speed);
212
213static ssize_t duplex_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
215{
216 struct net_device *netdev = to_net_dev(dev);
217 int ret = -EINVAL;
218
219 if (!rtnl_trylock())
220 return restart_syscall();
221
222 if (netif_running(netdev)) {
223 struct ethtool_link_ksettings cmd;
224
225 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
226 const char *duplex;
227
228 switch (cmd.base.duplex) {
229 case DUPLEX_HALF:
230 duplex = "half";
231 break;
232 case DUPLEX_FULL:
233 duplex = "full";
234 break;
235 default:
236 duplex = "unknown";
237 break;
238 }
239 ret = sprintf(buf, "%s\n", duplex);
240 }
241 }
242 rtnl_unlock();
243 return ret;
244}
245static DEVICE_ATTR_RO(duplex);
246
247static ssize_t testing_show(struct device *dev,
248 struct device_attribute *attr, char *buf)
249{
250 struct net_device *netdev = to_net_dev(dev);
251
252 if (netif_running(netdev))
253 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
254
255 return -EINVAL;
256}
257static DEVICE_ATTR_RO(testing);
258
259static ssize_t dormant_show(struct device *dev,
260 struct device_attribute *attr, char *buf)
261{
262 struct net_device *netdev = to_net_dev(dev);
263
264 if (netif_running(netdev))
265 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
266
267 return -EINVAL;
268}
269static DEVICE_ATTR_RO(dormant);
270
271static const char *const operstates[] = {
272 "unknown",
273 "notpresent", /* currently unused */
274 "down",
275 "lowerlayerdown",
276 "testing",
277 "dormant",
278 "up"
279};
280
281static ssize_t operstate_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 const struct net_device *netdev = to_net_dev(dev);
285 unsigned char operstate;
286
287 read_lock(&dev_base_lock);
288 operstate = netdev->operstate;
289 if (!netif_running(netdev))
290 operstate = IF_OPER_DOWN;
291 read_unlock(&dev_base_lock);
292
293 if (operstate >= ARRAY_SIZE(operstates))
294 return -EINVAL; /* should not happen */
295
296 return sprintf(buf, "%s\n", operstates[operstate]);
297}
298static DEVICE_ATTR_RO(operstate);
299
300static ssize_t carrier_changes_show(struct device *dev,
301 struct device_attribute *attr,
302 char *buf)
303{
304 struct net_device *netdev = to_net_dev(dev);
305
306 return sprintf(buf, fmt_dec,
307 atomic_read(&netdev->carrier_up_count) +
308 atomic_read(&netdev->carrier_down_count));
309}
310static DEVICE_ATTR_RO(carrier_changes);
311
312static ssize_t carrier_up_count_show(struct device *dev,
313 struct device_attribute *attr,
314 char *buf)
315{
316 struct net_device *netdev = to_net_dev(dev);
317
318 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
319}
320static DEVICE_ATTR_RO(carrier_up_count);
321
322static ssize_t carrier_down_count_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 sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
329}
330static DEVICE_ATTR_RO(carrier_down_count);
331
332/* read-write attributes */
333
334static int change_mtu(struct net_device *dev, unsigned long new_mtu)
335{
336 return dev_set_mtu(dev, (int)new_mtu);
337}
338
339static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
340 const char *buf, size_t len)
341{
342 return netdev_store(dev, attr, buf, len, change_mtu);
343}
344NETDEVICE_SHOW_RW(mtu, fmt_dec);
345
346static int change_flags(struct net_device *dev, unsigned long new_flags)
347{
348 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
349}
350
351static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
352 const char *buf, size_t len)
353{
354 return netdev_store(dev, attr, buf, len, change_flags);
355}
356NETDEVICE_SHOW_RW(flags, fmt_hex);
357
358static ssize_t tx_queue_len_store(struct device *dev,
359 struct device_attribute *attr,
360 const char *buf, size_t len)
361{
362 if (!capable(CAP_NET_ADMIN))
363 return -EPERM;
364
365 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
366}
367NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
368
369static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
370{
371 WRITE_ONCE(dev->gro_flush_timeout, val);
372 return 0;
373}
374
375static ssize_t gro_flush_timeout_store(struct device *dev,
376 struct device_attribute *attr,
377 const char *buf, size_t len)
378{
379 if (!capable(CAP_NET_ADMIN))
380 return -EPERM;
381
382 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
383}
384NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
385
386static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
387{
388 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
389 return 0;
390}
391
392static ssize_t napi_defer_hard_irqs_store(struct device *dev,
393 struct device_attribute *attr,
394 const char *buf, size_t len)
395{
396 if (!capable(CAP_NET_ADMIN))
397 return -EPERM;
398
399 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
400}
401NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
402
403static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
404 const char *buf, size_t len)
405{
406 struct net_device *netdev = to_net_dev(dev);
407 struct net *net = dev_net(netdev);
408 size_t count = len;
409 ssize_t ret = 0;
410
411 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
412 return -EPERM;
413
414 /* ignore trailing newline */
415 if (len > 0 && buf[len - 1] == '\n')
416 --count;
417
418 if (!rtnl_trylock())
419 return restart_syscall();
420
421 if (dev_isalive(netdev)) {
422 ret = dev_set_alias(netdev, buf, count);
423 if (ret < 0)
424 goto err;
425 ret = len;
426 netdev_state_change(netdev);
427 }
428err:
429 rtnl_unlock();
430
431 return ret;
432}
433
434static ssize_t ifalias_show(struct device *dev,
435 struct device_attribute *attr, char *buf)
436{
437 const struct net_device *netdev = to_net_dev(dev);
438 char tmp[IFALIASZ];
439 ssize_t ret = 0;
440
441 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
442 if (ret > 0)
443 ret = sprintf(buf, "%s\n", tmp);
444 return ret;
445}
446static DEVICE_ATTR_RW(ifalias);
447
448static int change_group(struct net_device *dev, unsigned long new_group)
449{
450 dev_set_group(dev, (int)new_group);
451 return 0;
452}
453
454static ssize_t group_store(struct device *dev, struct device_attribute *attr,
455 const char *buf, size_t len)
456{
457 return netdev_store(dev, attr, buf, len, change_group);
458}
459NETDEVICE_SHOW(group, fmt_dec);
460static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
461
462static int change_proto_down(struct net_device *dev, unsigned long proto_down)
463{
464 return dev_change_proto_down(dev, (bool)proto_down);
465}
466
467static ssize_t proto_down_store(struct device *dev,
468 struct device_attribute *attr,
469 const char *buf, size_t len)
470{
471 return netdev_store(dev, attr, buf, len, change_proto_down);
472}
473NETDEVICE_SHOW_RW(proto_down, fmt_dec);
474
475static ssize_t phys_port_id_show(struct device *dev,
476 struct device_attribute *attr, char *buf)
477{
478 struct net_device *netdev = to_net_dev(dev);
479 ssize_t ret = -EINVAL;
480
481 if (!rtnl_trylock())
482 return restart_syscall();
483
484 if (dev_isalive(netdev)) {
485 struct netdev_phys_item_id ppid;
486
487 ret = dev_get_phys_port_id(netdev, &ppid);
488 if (!ret)
489 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
490 }
491 rtnl_unlock();
492
493 return ret;
494}
495static DEVICE_ATTR_RO(phys_port_id);
496
497static ssize_t phys_port_name_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 if (!rtnl_trylock())
504 return restart_syscall();
505
506 if (dev_isalive(netdev)) {
507 char name[IFNAMSIZ];
508
509 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
510 if (!ret)
511 ret = sprintf(buf, "%s\n", name);
512 }
513 rtnl_unlock();
514
515 return ret;
516}
517static DEVICE_ATTR_RO(phys_port_name);
518
519static ssize_t phys_switch_id_show(struct device *dev,
520 struct device_attribute *attr, char *buf)
521{
522 struct net_device *netdev = to_net_dev(dev);
523 ssize_t ret = -EINVAL;
524
525 if (!rtnl_trylock())
526 return restart_syscall();
527
528 if (dev_isalive(netdev)) {
529 struct netdev_phys_item_id ppid = { };
530
531 ret = dev_get_port_parent_id(netdev, &ppid, false);
532 if (!ret)
533 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
534 }
535 rtnl_unlock();
536
537 return ret;
538}
539static DEVICE_ATTR_RO(phys_switch_id);
540
541static struct attribute *net_class_attrs[] __ro_after_init = {
542 &dev_attr_netdev_group.attr,
543 &dev_attr_type.attr,
544 &dev_attr_dev_id.attr,
545 &dev_attr_dev_port.attr,
546 &dev_attr_iflink.attr,
547 &dev_attr_ifindex.attr,
548 &dev_attr_name_assign_type.attr,
549 &dev_attr_addr_assign_type.attr,
550 &dev_attr_addr_len.attr,
551 &dev_attr_link_mode.attr,
552 &dev_attr_address.attr,
553 &dev_attr_broadcast.attr,
554 &dev_attr_speed.attr,
555 &dev_attr_duplex.attr,
556 &dev_attr_dormant.attr,
557 &dev_attr_testing.attr,
558 &dev_attr_operstate.attr,
559 &dev_attr_carrier_changes.attr,
560 &dev_attr_ifalias.attr,
561 &dev_attr_carrier.attr,
562 &dev_attr_mtu.attr,
563 &dev_attr_flags.attr,
564 &dev_attr_tx_queue_len.attr,
565 &dev_attr_gro_flush_timeout.attr,
566 &dev_attr_napi_defer_hard_irqs.attr,
567 &dev_attr_phys_port_id.attr,
568 &dev_attr_phys_port_name.attr,
569 &dev_attr_phys_switch_id.attr,
570 &dev_attr_proto_down.attr,
571 &dev_attr_carrier_up_count.attr,
572 &dev_attr_carrier_down_count.attr,
573 NULL,
574};
575ATTRIBUTE_GROUPS(net_class);
576
577/* Show a given an attribute in the statistics group */
578static ssize_t netstat_show(const struct device *d,
579 struct device_attribute *attr, char *buf,
580 unsigned long offset)
581{
582 struct net_device *dev = to_net_dev(d);
583 ssize_t ret = -EINVAL;
584
585 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
586 offset % sizeof(u64) != 0);
587
588 read_lock(&dev_base_lock);
589 if (dev_isalive(dev)) {
590 struct rtnl_link_stats64 temp;
591 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
592
593 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
594 }
595 read_unlock(&dev_base_lock);
596 return ret;
597}
598
599/* generate a read-only statistics attribute */
600#define NETSTAT_ENTRY(name) \
601static ssize_t name##_show(struct device *d, \
602 struct device_attribute *attr, char *buf) \
603{ \
604 return netstat_show(d, attr, buf, \
605 offsetof(struct rtnl_link_stats64, name)); \
606} \
607static DEVICE_ATTR_RO(name)
608
609NETSTAT_ENTRY(rx_packets);
610NETSTAT_ENTRY(tx_packets);
611NETSTAT_ENTRY(rx_bytes);
612NETSTAT_ENTRY(tx_bytes);
613NETSTAT_ENTRY(rx_errors);
614NETSTAT_ENTRY(tx_errors);
615NETSTAT_ENTRY(rx_dropped);
616NETSTAT_ENTRY(tx_dropped);
617NETSTAT_ENTRY(multicast);
618NETSTAT_ENTRY(collisions);
619NETSTAT_ENTRY(rx_length_errors);
620NETSTAT_ENTRY(rx_over_errors);
621NETSTAT_ENTRY(rx_crc_errors);
622NETSTAT_ENTRY(rx_frame_errors);
623NETSTAT_ENTRY(rx_fifo_errors);
624NETSTAT_ENTRY(rx_missed_errors);
625NETSTAT_ENTRY(tx_aborted_errors);
626NETSTAT_ENTRY(tx_carrier_errors);
627NETSTAT_ENTRY(tx_fifo_errors);
628NETSTAT_ENTRY(tx_heartbeat_errors);
629NETSTAT_ENTRY(tx_window_errors);
630NETSTAT_ENTRY(rx_compressed);
631NETSTAT_ENTRY(tx_compressed);
632NETSTAT_ENTRY(rx_nohandler);
633
634static struct attribute *netstat_attrs[] __ro_after_init = {
635 &dev_attr_rx_packets.attr,
636 &dev_attr_tx_packets.attr,
637 &dev_attr_rx_bytes.attr,
638 &dev_attr_tx_bytes.attr,
639 &dev_attr_rx_errors.attr,
640 &dev_attr_tx_errors.attr,
641 &dev_attr_rx_dropped.attr,
642 &dev_attr_tx_dropped.attr,
643 &dev_attr_multicast.attr,
644 &dev_attr_collisions.attr,
645 &dev_attr_rx_length_errors.attr,
646 &dev_attr_rx_over_errors.attr,
647 &dev_attr_rx_crc_errors.attr,
648 &dev_attr_rx_frame_errors.attr,
649 &dev_attr_rx_fifo_errors.attr,
650 &dev_attr_rx_missed_errors.attr,
651 &dev_attr_tx_aborted_errors.attr,
652 &dev_attr_tx_carrier_errors.attr,
653 &dev_attr_tx_fifo_errors.attr,
654 &dev_attr_tx_heartbeat_errors.attr,
655 &dev_attr_tx_window_errors.attr,
656 &dev_attr_rx_compressed.attr,
657 &dev_attr_tx_compressed.attr,
658 &dev_attr_rx_nohandler.attr,
659 NULL
660};
661
662static const struct attribute_group netstat_group = {
663 .name = "statistics",
664 .attrs = netstat_attrs,
665};
666
667#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
668static struct attribute *wireless_attrs[] = {
669 NULL
670};
671
672static const struct attribute_group wireless_group = {
673 .name = "wireless",
674 .attrs = wireless_attrs,
675};
676#endif
677
678#else /* CONFIG_SYSFS */
679#define net_class_groups NULL
680#endif /* CONFIG_SYSFS */
681
682#ifdef CONFIG_SYSFS
683#define to_rx_queue_attr(_attr) \
684 container_of(_attr, struct rx_queue_attribute, attr)
685
686#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
687
688static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
689 char *buf)
690{
691 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
692 struct netdev_rx_queue *queue = to_rx_queue(kobj);
693
694 if (!attribute->show)
695 return -EIO;
696
697 return attribute->show(queue, buf);
698}
699
700static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
701 const char *buf, size_t count)
702{
703 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
704 struct netdev_rx_queue *queue = to_rx_queue(kobj);
705
706 if (!attribute->store)
707 return -EIO;
708
709 return attribute->store(queue, buf, count);
710}
711
712static const struct sysfs_ops rx_queue_sysfs_ops = {
713 .show = rx_queue_attr_show,
714 .store = rx_queue_attr_store,
715};
716
717#ifdef CONFIG_RPS
718static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
719{
720 struct rps_map *map;
721 cpumask_var_t mask;
722 int i, len;
723
724 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
725 return -ENOMEM;
726
727 rcu_read_lock();
728 map = rcu_dereference(queue->rps_map);
729 if (map)
730 for (i = 0; i < map->len; i++)
731 cpumask_set_cpu(map->cpus[i], mask);
732
733 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
734 rcu_read_unlock();
735 free_cpumask_var(mask);
736
737 return len < PAGE_SIZE ? len : -EINVAL;
738}
739
740static ssize_t store_rps_map(struct netdev_rx_queue *queue,
741 const char *buf, size_t len)
742{
743 struct rps_map *old_map, *map;
744 cpumask_var_t mask;
745 int err, cpu, i, hk_flags;
746 static DEFINE_MUTEX(rps_map_mutex);
747
748 if (!capable(CAP_NET_ADMIN))
749 return -EPERM;
750
751 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
752 return -ENOMEM;
753
754 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
755 if (err) {
756 free_cpumask_var(mask);
757 return err;
758 }
759
760 if (!cpumask_empty(mask)) {
761 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
762 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
763 if (cpumask_empty(mask)) {
764 free_cpumask_var(mask);
765 return -EINVAL;
766 }
767 }
768
769 map = kzalloc(max_t(unsigned int,
770 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
771 GFP_KERNEL);
772 if (!map) {
773 free_cpumask_var(mask);
774 return -ENOMEM;
775 }
776
777 i = 0;
778 for_each_cpu_and(cpu, mask, cpu_online_mask)
779 map->cpus[i++] = cpu;
780
781 if (i) {
782 map->len = i;
783 } else {
784 kfree(map);
785 map = NULL;
786 }
787
788 mutex_lock(&rps_map_mutex);
789 old_map = rcu_dereference_protected(queue->rps_map,
790 mutex_is_locked(&rps_map_mutex));
791 rcu_assign_pointer(queue->rps_map, map);
792
793 if (map)
794 static_branch_inc(&rps_needed);
795 if (old_map)
796 static_branch_dec(&rps_needed);
797
798 mutex_unlock(&rps_map_mutex);
799
800 if (old_map)
801 kfree_rcu(old_map, rcu);
802
803 free_cpumask_var(mask);
804 return len;
805}
806
807static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
808 char *buf)
809{
810 struct rps_dev_flow_table *flow_table;
811 unsigned long val = 0;
812
813 rcu_read_lock();
814 flow_table = rcu_dereference(queue->rps_flow_table);
815 if (flow_table)
816 val = (unsigned long)flow_table->mask + 1;
817 rcu_read_unlock();
818
819 return sprintf(buf, "%lu\n", val);
820}
821
822static void rps_dev_flow_table_release(struct rcu_head *rcu)
823{
824 struct rps_dev_flow_table *table = container_of(rcu,
825 struct rps_dev_flow_table, rcu);
826 vfree(table);
827}
828
829static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
830 const char *buf, size_t len)
831{
832 unsigned long mask, count;
833 struct rps_dev_flow_table *table, *old_table;
834 static DEFINE_SPINLOCK(rps_dev_flow_lock);
835 int rc;
836
837 if (!capable(CAP_NET_ADMIN))
838 return -EPERM;
839
840 rc = kstrtoul(buf, 0, &count);
841 if (rc < 0)
842 return rc;
843
844 if (count) {
845 mask = count - 1;
846 /* mask = roundup_pow_of_two(count) - 1;
847 * without overflows...
848 */
849 while ((mask | (mask >> 1)) != mask)
850 mask |= (mask >> 1);
851 /* On 64 bit arches, must check mask fits in table->mask (u32),
852 * and on 32bit arches, must check
853 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
854 */
855#if BITS_PER_LONG > 32
856 if (mask > (unsigned long)(u32)mask)
857 return -EINVAL;
858#else
859 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
860 / sizeof(struct rps_dev_flow)) {
861 /* Enforce a limit to prevent overflow */
862 return -EINVAL;
863 }
864#endif
865 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
866 if (!table)
867 return -ENOMEM;
868
869 table->mask = mask;
870 for (count = 0; count <= mask; count++)
871 table->flows[count].cpu = RPS_NO_CPU;
872 } else {
873 table = NULL;
874 }
875
876 spin_lock(&rps_dev_flow_lock);
877 old_table = rcu_dereference_protected(queue->rps_flow_table,
878 lockdep_is_held(&rps_dev_flow_lock));
879 rcu_assign_pointer(queue->rps_flow_table, table);
880 spin_unlock(&rps_dev_flow_lock);
881
882 if (old_table)
883 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
884
885 return len;
886}
887
888static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
889 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
890
891static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
892 = __ATTR(rps_flow_cnt, 0644,
893 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
894#endif /* CONFIG_RPS */
895
896static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
897#ifdef CONFIG_RPS
898 &rps_cpus_attribute.attr,
899 &rps_dev_flow_table_cnt_attribute.attr,
900#endif
901 NULL
902};
903ATTRIBUTE_GROUPS(rx_queue_default);
904
905static void rx_queue_release(struct kobject *kobj)
906{
907 struct netdev_rx_queue *queue = to_rx_queue(kobj);
908#ifdef CONFIG_RPS
909 struct rps_map *map;
910 struct rps_dev_flow_table *flow_table;
911
912 map = rcu_dereference_protected(queue->rps_map, 1);
913 if (map) {
914 RCU_INIT_POINTER(queue->rps_map, NULL);
915 kfree_rcu(map, rcu);
916 }
917
918 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
919 if (flow_table) {
920 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
921 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
922 }
923#endif
924
925 memset(kobj, 0, sizeof(*kobj));
926 dev_put(queue->dev);
927}
928
929static const void *rx_queue_namespace(struct kobject *kobj)
930{
931 struct netdev_rx_queue *queue = to_rx_queue(kobj);
932 struct device *dev = &queue->dev->dev;
933 const void *ns = NULL;
934
935 if (dev->class && dev->class->ns_type)
936 ns = dev->class->namespace(dev);
937
938 return ns;
939}
940
941static void rx_queue_get_ownership(struct kobject *kobj,
942 kuid_t *uid, kgid_t *gid)
943{
944 const struct net *net = rx_queue_namespace(kobj);
945
946 net_ns_get_ownership(net, uid, gid);
947}
948
949static struct kobj_type rx_queue_ktype __ro_after_init = {
950 .sysfs_ops = &rx_queue_sysfs_ops,
951 .release = rx_queue_release,
952 .default_groups = rx_queue_default_groups,
953 .namespace = rx_queue_namespace,
954 .get_ownership = rx_queue_get_ownership,
955};
956
957static int rx_queue_add_kobject(struct net_device *dev, int index)
958{
959 struct netdev_rx_queue *queue = dev->_rx + index;
960 struct kobject *kobj = &queue->kobj;
961 int error = 0;
962
963 /* Kobject_put later will trigger rx_queue_release call which
964 * decreases dev refcount: Take that reference here
965 */
966 dev_hold(queue->dev);
967
968 kobj->kset = dev->queues_kset;
969 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
970 "rx-%u", index);
971 if (error)
972 goto err;
973
974 if (dev->sysfs_rx_queue_group) {
975 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
976 if (error)
977 goto err;
978 }
979
980 kobject_uevent(kobj, KOBJ_ADD);
981
982 return error;
983
984err:
985 kobject_put(kobj);
986 return error;
987}
988
989static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
990 kgid_t kgid)
991{
992 struct netdev_rx_queue *queue = dev->_rx + index;
993 struct kobject *kobj = &queue->kobj;
994 int error;
995
996 error = sysfs_change_owner(kobj, kuid, kgid);
997 if (error)
998 return error;
999
1000 if (dev->sysfs_rx_queue_group)
1001 error = sysfs_group_change_owner(
1002 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1003
1004 return error;
1005}
1006#endif /* CONFIG_SYSFS */
1007
1008int
1009net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1010{
1011#ifdef CONFIG_SYSFS
1012 int i;
1013 int error = 0;
1014
1015#ifndef CONFIG_RPS
1016 if (!dev->sysfs_rx_queue_group)
1017 return 0;
1018#endif
1019 for (i = old_num; i < new_num; i++) {
1020 error = rx_queue_add_kobject(dev, i);
1021 if (error) {
1022 new_num = old_num;
1023 break;
1024 }
1025 }
1026
1027 while (--i >= new_num) {
1028 struct kobject *kobj = &dev->_rx[i].kobj;
1029
1030 if (!refcount_read(&dev_net(dev)->count))
1031 kobj->uevent_suppress = 1;
1032 if (dev->sysfs_rx_queue_group)
1033 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1034 kobject_put(kobj);
1035 }
1036
1037 return error;
1038#else
1039 return 0;
1040#endif
1041}
1042
1043static int net_rx_queue_change_owner(struct net_device *dev, int num,
1044 kuid_t kuid, kgid_t kgid)
1045{
1046#ifdef CONFIG_SYSFS
1047 int error = 0;
1048 int i;
1049
1050#ifndef CONFIG_RPS
1051 if (!dev->sysfs_rx_queue_group)
1052 return 0;
1053#endif
1054 for (i = 0; i < num; i++) {
1055 error = rx_queue_change_owner(dev, i, kuid, kgid);
1056 if (error)
1057 break;
1058 }
1059
1060 return error;
1061#else
1062 return 0;
1063#endif
1064}
1065
1066#ifdef CONFIG_SYSFS
1067/*
1068 * netdev_queue sysfs structures and functions.
1069 */
1070struct netdev_queue_attribute {
1071 struct attribute attr;
1072 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1073 ssize_t (*store)(struct netdev_queue *queue,
1074 const char *buf, size_t len);
1075};
1076#define to_netdev_queue_attr(_attr) \
1077 container_of(_attr, struct netdev_queue_attribute, attr)
1078
1079#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1080
1081static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1082 struct attribute *attr, char *buf)
1083{
1084 const struct netdev_queue_attribute *attribute
1085 = to_netdev_queue_attr(attr);
1086 struct netdev_queue *queue = to_netdev_queue(kobj);
1087
1088 if (!attribute->show)
1089 return -EIO;
1090
1091 return attribute->show(queue, buf);
1092}
1093
1094static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1095 struct attribute *attr,
1096 const char *buf, size_t count)
1097{
1098 const struct netdev_queue_attribute *attribute
1099 = to_netdev_queue_attr(attr);
1100 struct netdev_queue *queue = to_netdev_queue(kobj);
1101
1102 if (!attribute->store)
1103 return -EIO;
1104
1105 return attribute->store(queue, buf, count);
1106}
1107
1108static const struct sysfs_ops netdev_queue_sysfs_ops = {
1109 .show = netdev_queue_attr_show,
1110 .store = netdev_queue_attr_store,
1111};
1112
1113static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1114{
1115 unsigned long trans_timeout;
1116
1117 spin_lock_irq(&queue->_xmit_lock);
1118 trans_timeout = queue->trans_timeout;
1119 spin_unlock_irq(&queue->_xmit_lock);
1120
1121 return sprintf(buf, fmt_ulong, trans_timeout);
1122}
1123
1124static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1125{
1126 struct net_device *dev = queue->dev;
1127 unsigned int i;
1128
1129 i = queue - dev->_tx;
1130 BUG_ON(i >= dev->num_tx_queues);
1131
1132 return i;
1133}
1134
1135static ssize_t traffic_class_show(struct netdev_queue *queue,
1136 char *buf)
1137{
1138 struct net_device *dev = queue->dev;
1139 int index;
1140 int tc;
1141
1142 if (!netif_is_multiqueue(dev))
1143 return -ENOENT;
1144
1145 index = get_netdev_queue_index(queue);
1146
1147 /* If queue belongs to subordinate dev use its TC mapping */
1148 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1149
1150 tc = netdev_txq_to_tc(dev, index);
1151 if (tc < 0)
1152 return -EINVAL;
1153
1154 /* We can report the traffic class one of two ways:
1155 * Subordinate device traffic classes are reported with the traffic
1156 * class first, and then the subordinate class so for example TC0 on
1157 * subordinate device 2 will be reported as "0-2". If the queue
1158 * belongs to the root device it will be reported with just the
1159 * traffic class, so just "0" for TC 0 for example.
1160 */
1161 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1162 sprintf(buf, "%u\n", tc);
1163}
1164
1165#ifdef CONFIG_XPS
1166static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1167 char *buf)
1168{
1169 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1170}
1171
1172static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1173 const char *buf, size_t len)
1174{
1175 struct net_device *dev = queue->dev;
1176 int err, index = get_netdev_queue_index(queue);
1177 u32 rate = 0;
1178
1179 if (!capable(CAP_NET_ADMIN))
1180 return -EPERM;
1181
1182 err = kstrtou32(buf, 10, &rate);
1183 if (err < 0)
1184 return err;
1185
1186 if (!rtnl_trylock())
1187 return restart_syscall();
1188
1189 err = -EOPNOTSUPP;
1190 if (dev->netdev_ops->ndo_set_tx_maxrate)
1191 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1192
1193 rtnl_unlock();
1194 if (!err) {
1195 queue->tx_maxrate = rate;
1196 return len;
1197 }
1198 return err;
1199}
1200
1201static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1202 = __ATTR_RW(tx_maxrate);
1203#endif
1204
1205static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1206 = __ATTR_RO(tx_timeout);
1207
1208static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1209 = __ATTR_RO(traffic_class);
1210
1211#ifdef CONFIG_BQL
1212/*
1213 * Byte queue limits sysfs structures and functions.
1214 */
1215static ssize_t bql_show(char *buf, unsigned int value)
1216{
1217 return sprintf(buf, "%u\n", value);
1218}
1219
1220static ssize_t bql_set(const char *buf, const size_t count,
1221 unsigned int *pvalue)
1222{
1223 unsigned int value;
1224 int err;
1225
1226 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1227 value = DQL_MAX_LIMIT;
1228 } else {
1229 err = kstrtouint(buf, 10, &value);
1230 if (err < 0)
1231 return err;
1232 if (value > DQL_MAX_LIMIT)
1233 return -EINVAL;
1234 }
1235
1236 *pvalue = value;
1237
1238 return count;
1239}
1240
1241static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1242 char *buf)
1243{
1244 struct dql *dql = &queue->dql;
1245
1246 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1247}
1248
1249static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1250 const char *buf, size_t len)
1251{
1252 struct dql *dql = &queue->dql;
1253 unsigned int value;
1254 int err;
1255
1256 err = kstrtouint(buf, 10, &value);
1257 if (err < 0)
1258 return err;
1259
1260 dql->slack_hold_time = msecs_to_jiffies(value);
1261
1262 return len;
1263}
1264
1265static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1266 = __ATTR(hold_time, 0644,
1267 bql_show_hold_time, bql_set_hold_time);
1268
1269static ssize_t bql_show_inflight(struct netdev_queue *queue,
1270 char *buf)
1271{
1272 struct dql *dql = &queue->dql;
1273
1274 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1275}
1276
1277static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1278 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1279
1280#define BQL_ATTR(NAME, FIELD) \
1281static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1282 char *buf) \
1283{ \
1284 return bql_show(buf, queue->dql.FIELD); \
1285} \
1286 \
1287static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1288 const char *buf, size_t len) \
1289{ \
1290 return bql_set(buf, len, &queue->dql.FIELD); \
1291} \
1292 \
1293static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1294 = __ATTR(NAME, 0644, \
1295 bql_show_ ## NAME, bql_set_ ## NAME)
1296
1297BQL_ATTR(limit, limit);
1298BQL_ATTR(limit_max, max_limit);
1299BQL_ATTR(limit_min, min_limit);
1300
1301static struct attribute *dql_attrs[] __ro_after_init = {
1302 &bql_limit_attribute.attr,
1303 &bql_limit_max_attribute.attr,
1304 &bql_limit_min_attribute.attr,
1305 &bql_hold_time_attribute.attr,
1306 &bql_inflight_attribute.attr,
1307 NULL
1308};
1309
1310static const struct attribute_group dql_group = {
1311 .name = "byte_queue_limits",
1312 .attrs = dql_attrs,
1313};
1314#endif /* CONFIG_BQL */
1315
1316#ifdef CONFIG_XPS
1317static ssize_t xps_cpus_show(struct netdev_queue *queue,
1318 char *buf)
1319{
1320 struct net_device *dev = queue->dev;
1321 int cpu, len, num_tc = 1, tc = 0;
1322 struct xps_dev_maps *dev_maps;
1323 cpumask_var_t mask;
1324 unsigned long index;
1325
1326 if (!netif_is_multiqueue(dev))
1327 return -ENOENT;
1328
1329 index = get_netdev_queue_index(queue);
1330
1331 if (dev->num_tc) {
1332 /* Do not allow XPS on subordinate device directly */
1333 num_tc = dev->num_tc;
1334 if (num_tc < 0)
1335 return -EINVAL;
1336
1337 /* If queue belongs to subordinate dev use its map */
1338 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1339
1340 tc = netdev_txq_to_tc(dev, index);
1341 if (tc < 0)
1342 return -EINVAL;
1343 }
1344
1345 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1346 return -ENOMEM;
1347
1348 rcu_read_lock();
1349 dev_maps = rcu_dereference(dev->xps_cpus_map);
1350 if (dev_maps) {
1351 for_each_possible_cpu(cpu) {
1352 int i, tci = cpu * num_tc + tc;
1353 struct xps_map *map;
1354
1355 map = rcu_dereference(dev_maps->attr_map[tci]);
1356 if (!map)
1357 continue;
1358
1359 for (i = map->len; i--;) {
1360 if (map->queues[i] == index) {
1361 cpumask_set_cpu(cpu, mask);
1362 break;
1363 }
1364 }
1365 }
1366 }
1367 rcu_read_unlock();
1368
1369 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1370 free_cpumask_var(mask);
1371 return len < PAGE_SIZE ? len : -EINVAL;
1372}
1373
1374static ssize_t xps_cpus_store(struct netdev_queue *queue,
1375 const char *buf, size_t len)
1376{
1377 struct net_device *dev = queue->dev;
1378 unsigned long index;
1379 cpumask_var_t mask;
1380 int err;
1381
1382 if (!netif_is_multiqueue(dev))
1383 return -ENOENT;
1384
1385 if (!capable(CAP_NET_ADMIN))
1386 return -EPERM;
1387
1388 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1389 return -ENOMEM;
1390
1391 index = get_netdev_queue_index(queue);
1392
1393 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1394 if (err) {
1395 free_cpumask_var(mask);
1396 return err;
1397 }
1398
1399 err = netif_set_xps_queue(dev, mask, index);
1400
1401 free_cpumask_var(mask);
1402
1403 return err ? : len;
1404}
1405
1406static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1407 = __ATTR_RW(xps_cpus);
1408
1409static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1410{
1411 struct net_device *dev = queue->dev;
1412 struct xps_dev_maps *dev_maps;
1413 unsigned long *mask, index;
1414 int j, len, num_tc = 1, tc = 0;
1415
1416 index = get_netdev_queue_index(queue);
1417
1418 if (dev->num_tc) {
1419 num_tc = dev->num_tc;
1420 tc = netdev_txq_to_tc(dev, index);
1421 if (tc < 0)
1422 return -EINVAL;
1423 }
1424 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1425 if (!mask)
1426 return -ENOMEM;
1427
1428 rcu_read_lock();
1429 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1430 if (!dev_maps)
1431 goto out_no_maps;
1432
1433 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1434 j < dev->num_rx_queues;) {
1435 int i, tci = j * num_tc + tc;
1436 struct xps_map *map;
1437
1438 map = rcu_dereference(dev_maps->attr_map[tci]);
1439 if (!map)
1440 continue;
1441
1442 for (i = map->len; i--;) {
1443 if (map->queues[i] == index) {
1444 set_bit(j, mask);
1445 break;
1446 }
1447 }
1448 }
1449out_no_maps:
1450 rcu_read_unlock();
1451
1452 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1453 bitmap_free(mask);
1454
1455 return len < PAGE_SIZE ? len : -EINVAL;
1456}
1457
1458static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1459 size_t len)
1460{
1461 struct net_device *dev = queue->dev;
1462 struct net *net = dev_net(dev);
1463 unsigned long *mask, index;
1464 int err;
1465
1466 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1467 return -EPERM;
1468
1469 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1470 if (!mask)
1471 return -ENOMEM;
1472
1473 index = get_netdev_queue_index(queue);
1474
1475 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1476 if (err) {
1477 bitmap_free(mask);
1478 return err;
1479 }
1480
1481 cpus_read_lock();
1482 err = __netif_set_xps_queue(dev, mask, index, true);
1483 cpus_read_unlock();
1484
1485 bitmap_free(mask);
1486 return err ? : len;
1487}
1488
1489static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1490 = __ATTR_RW(xps_rxqs);
1491#endif /* CONFIG_XPS */
1492
1493static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1494 &queue_trans_timeout.attr,
1495 &queue_traffic_class.attr,
1496#ifdef CONFIG_XPS
1497 &xps_cpus_attribute.attr,
1498 &xps_rxqs_attribute.attr,
1499 &queue_tx_maxrate.attr,
1500#endif
1501 NULL
1502};
1503ATTRIBUTE_GROUPS(netdev_queue_default);
1504
1505static void netdev_queue_release(struct kobject *kobj)
1506{
1507 struct netdev_queue *queue = to_netdev_queue(kobj);
1508
1509 memset(kobj, 0, sizeof(*kobj));
1510 dev_put(queue->dev);
1511}
1512
1513static const void *netdev_queue_namespace(struct kobject *kobj)
1514{
1515 struct netdev_queue *queue = to_netdev_queue(kobj);
1516 struct device *dev = &queue->dev->dev;
1517 const void *ns = NULL;
1518
1519 if (dev->class && dev->class->ns_type)
1520 ns = dev->class->namespace(dev);
1521
1522 return ns;
1523}
1524
1525static void netdev_queue_get_ownership(struct kobject *kobj,
1526 kuid_t *uid, kgid_t *gid)
1527{
1528 const struct net *net = netdev_queue_namespace(kobj);
1529
1530 net_ns_get_ownership(net, uid, gid);
1531}
1532
1533static struct kobj_type netdev_queue_ktype __ro_after_init = {
1534 .sysfs_ops = &netdev_queue_sysfs_ops,
1535 .release = netdev_queue_release,
1536 .default_groups = netdev_queue_default_groups,
1537 .namespace = netdev_queue_namespace,
1538 .get_ownership = netdev_queue_get_ownership,
1539};
1540
1541static int netdev_queue_add_kobject(struct net_device *dev, int index)
1542{
1543 struct netdev_queue *queue = dev->_tx + index;
1544 struct kobject *kobj = &queue->kobj;
1545 int error = 0;
1546
1547 /* Kobject_put later will trigger netdev_queue_release call
1548 * which decreases dev refcount: Take that reference here
1549 */
1550 dev_hold(queue->dev);
1551
1552 kobj->kset = dev->queues_kset;
1553 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1554 "tx-%u", index);
1555 if (error)
1556 goto err;
1557
1558#ifdef CONFIG_BQL
1559 error = sysfs_create_group(kobj, &dql_group);
1560 if (error)
1561 goto err;
1562#endif
1563
1564 kobject_uevent(kobj, KOBJ_ADD);
1565 return 0;
1566
1567err:
1568 kobject_put(kobj);
1569 return error;
1570}
1571
1572static int tx_queue_change_owner(struct net_device *ndev, int index,
1573 kuid_t kuid, kgid_t kgid)
1574{
1575 struct netdev_queue *queue = ndev->_tx + index;
1576 struct kobject *kobj = &queue->kobj;
1577 int error;
1578
1579 error = sysfs_change_owner(kobj, kuid, kgid);
1580 if (error)
1581 return error;
1582
1583#ifdef CONFIG_BQL
1584 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1585#endif
1586 return error;
1587}
1588#endif /* CONFIG_SYSFS */
1589
1590int
1591netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1592{
1593#ifdef CONFIG_SYSFS
1594 int i;
1595 int error = 0;
1596
1597 for (i = old_num; i < new_num; i++) {
1598 error = netdev_queue_add_kobject(dev, i);
1599 if (error) {
1600 new_num = old_num;
1601 break;
1602 }
1603 }
1604
1605 while (--i >= new_num) {
1606 struct netdev_queue *queue = dev->_tx + i;
1607
1608 if (!refcount_read(&dev_net(dev)->count))
1609 queue->kobj.uevent_suppress = 1;
1610#ifdef CONFIG_BQL
1611 sysfs_remove_group(&queue->kobj, &dql_group);
1612#endif
1613 kobject_put(&queue->kobj);
1614 }
1615
1616 return error;
1617#else
1618 return 0;
1619#endif /* CONFIG_SYSFS */
1620}
1621
1622static int net_tx_queue_change_owner(struct net_device *dev, int num,
1623 kuid_t kuid, kgid_t kgid)
1624{
1625#ifdef CONFIG_SYSFS
1626 int error = 0;
1627 int i;
1628
1629 for (i = 0; i < num; i++) {
1630 error = tx_queue_change_owner(dev, i, kuid, kgid);
1631 if (error)
1632 break;
1633 }
1634
1635 return error;
1636#else
1637 return 0;
1638#endif /* CONFIG_SYSFS */
1639}
1640
1641static int register_queue_kobjects(struct net_device *dev)
1642{
1643 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1644
1645#ifdef CONFIG_SYSFS
1646 dev->queues_kset = kset_create_and_add("queues",
1647 NULL, &dev->dev.kobj);
1648 if (!dev->queues_kset)
1649 return -ENOMEM;
1650 real_rx = dev->real_num_rx_queues;
1651#endif
1652 real_tx = dev->real_num_tx_queues;
1653
1654 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1655 if (error)
1656 goto error;
1657 rxq = real_rx;
1658
1659 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1660 if (error)
1661 goto error;
1662 txq = real_tx;
1663
1664 return 0;
1665
1666error:
1667 netdev_queue_update_kobjects(dev, txq, 0);
1668 net_rx_queue_update_kobjects(dev, rxq, 0);
1669#ifdef CONFIG_SYSFS
1670 kset_unregister(dev->queues_kset);
1671#endif
1672 return error;
1673}
1674
1675static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1676{
1677 int error = 0, real_rx = 0, real_tx = 0;
1678
1679#ifdef CONFIG_SYSFS
1680 if (ndev->queues_kset) {
1681 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1682 if (error)
1683 return error;
1684 }
1685 real_rx = ndev->real_num_rx_queues;
1686#endif
1687 real_tx = ndev->real_num_tx_queues;
1688
1689 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1690 if (error)
1691 return error;
1692
1693 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1694 if (error)
1695 return error;
1696
1697 return 0;
1698}
1699
1700static void remove_queue_kobjects(struct net_device *dev)
1701{
1702 int real_rx = 0, real_tx = 0;
1703
1704#ifdef CONFIG_SYSFS
1705 real_rx = dev->real_num_rx_queues;
1706#endif
1707 real_tx = dev->real_num_tx_queues;
1708
1709 net_rx_queue_update_kobjects(dev, real_rx, 0);
1710 netdev_queue_update_kobjects(dev, real_tx, 0);
1711#ifdef CONFIG_SYSFS
1712 kset_unregister(dev->queues_kset);
1713#endif
1714}
1715
1716static bool net_current_may_mount(void)
1717{
1718 struct net *net = current->nsproxy->net_ns;
1719
1720 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1721}
1722
1723static void *net_grab_current_ns(void)
1724{
1725 struct net *ns = current->nsproxy->net_ns;
1726#ifdef CONFIG_NET_NS
1727 if (ns)
1728 refcount_inc(&ns->passive);
1729#endif
1730 return ns;
1731}
1732
1733static const void *net_initial_ns(void)
1734{
1735 return &init_net;
1736}
1737
1738static const void *net_netlink_ns(struct sock *sk)
1739{
1740 return sock_net(sk);
1741}
1742
1743const struct kobj_ns_type_operations net_ns_type_operations = {
1744 .type = KOBJ_NS_TYPE_NET,
1745 .current_may_mount = net_current_may_mount,
1746 .grab_current_ns = net_grab_current_ns,
1747 .netlink_ns = net_netlink_ns,
1748 .initial_ns = net_initial_ns,
1749 .drop_ns = net_drop_ns,
1750};
1751EXPORT_SYMBOL_GPL(net_ns_type_operations);
1752
1753static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1754{
1755 struct net_device *dev = to_net_dev(d);
1756 int retval;
1757
1758 /* pass interface to uevent. */
1759 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1760 if (retval)
1761 goto exit;
1762
1763 /* pass ifindex to uevent.
1764 * ifindex is useful as it won't change (interface name may change)
1765 * and is what RtNetlink uses natively.
1766 */
1767 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1768
1769exit:
1770 return retval;
1771}
1772
1773/*
1774 * netdev_release -- destroy and free a dead device.
1775 * Called when last reference to device kobject is gone.
1776 */
1777static void netdev_release(struct device *d)
1778{
1779 struct net_device *dev = to_net_dev(d);
1780
1781 BUG_ON(dev->reg_state != NETREG_RELEASED);
1782
1783 /* no need to wait for rcu grace period:
1784 * device is dead and about to be freed.
1785 */
1786 kfree(rcu_access_pointer(dev->ifalias));
1787 netdev_freemem(dev);
1788}
1789
1790static const void *net_namespace(struct device *d)
1791{
1792 struct net_device *dev = to_net_dev(d);
1793
1794 return dev_net(dev);
1795}
1796
1797static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1798{
1799 struct net_device *dev = to_net_dev(d);
1800 const struct net *net = dev_net(dev);
1801
1802 net_ns_get_ownership(net, uid, gid);
1803}
1804
1805static struct class net_class __ro_after_init = {
1806 .name = "net",
1807 .dev_release = netdev_release,
1808 .dev_groups = net_class_groups,
1809 .dev_uevent = netdev_uevent,
1810 .ns_type = &net_ns_type_operations,
1811 .namespace = net_namespace,
1812 .get_ownership = net_get_ownership,
1813};
1814
1815#ifdef CONFIG_OF_NET
1816static int of_dev_node_match(struct device *dev, const void *data)
1817{
1818 for (; dev; dev = dev->parent) {
1819 if (dev->of_node == data)
1820 return 1;
1821 }
1822
1823 return 0;
1824}
1825
1826/*
1827 * of_find_net_device_by_node - lookup the net device for the device node
1828 * @np: OF device node
1829 *
1830 * Looks up the net_device structure corresponding with the device node.
1831 * If successful, returns a pointer to the net_device with the embedded
1832 * struct device refcount incremented by one, or NULL on failure. The
1833 * refcount must be dropped when done with the net_device.
1834 */
1835struct net_device *of_find_net_device_by_node(struct device_node *np)
1836{
1837 struct device *dev;
1838
1839 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1840 if (!dev)
1841 return NULL;
1842
1843 return to_net_dev(dev);
1844}
1845EXPORT_SYMBOL(of_find_net_device_by_node);
1846#endif
1847
1848/* Delete sysfs entries but hold kobject reference until after all
1849 * netdev references are gone.
1850 */
1851void netdev_unregister_kobject(struct net_device *ndev)
1852{
1853 struct device *dev = &ndev->dev;
1854
1855 if (!refcount_read(&dev_net(ndev)->count))
1856 dev_set_uevent_suppress(dev, 1);
1857
1858 kobject_get(&dev->kobj);
1859
1860 remove_queue_kobjects(ndev);
1861
1862 pm_runtime_set_memalloc_noio(dev, false);
1863
1864 device_del(dev);
1865}
1866
1867/* Create sysfs entries for network device. */
1868int netdev_register_kobject(struct net_device *ndev)
1869{
1870 struct device *dev = &ndev->dev;
1871 const struct attribute_group **groups = ndev->sysfs_groups;
1872 int error = 0;
1873
1874 device_initialize(dev);
1875 dev->class = &net_class;
1876 dev->platform_data = ndev;
1877 dev->groups = groups;
1878
1879 dev_set_name(dev, "%s", ndev->name);
1880
1881#ifdef CONFIG_SYSFS
1882 /* Allow for a device specific group */
1883 if (*groups)
1884 groups++;
1885
1886 *groups++ = &netstat_group;
1887
1888#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1889 if (ndev->ieee80211_ptr)
1890 *groups++ = &wireless_group;
1891#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1892 else if (ndev->wireless_handlers)
1893 *groups++ = &wireless_group;
1894#endif
1895#endif
1896#endif /* CONFIG_SYSFS */
1897
1898 error = device_add(dev);
1899 if (error)
1900 return error;
1901
1902 error = register_queue_kobjects(ndev);
1903 if (error) {
1904 device_del(dev);
1905 return error;
1906 }
1907
1908 pm_runtime_set_memalloc_noio(dev, true);
1909
1910 return error;
1911}
1912
1913/* Change owner for sysfs entries when moving network devices across network
1914 * namespaces owned by different user namespaces.
1915 */
1916int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
1917 const struct net *net_new)
1918{
1919 struct device *dev = &ndev->dev;
1920 kuid_t old_uid, new_uid;
1921 kgid_t old_gid, new_gid;
1922 int error;
1923
1924 net_ns_get_ownership(net_old, &old_uid, &old_gid);
1925 net_ns_get_ownership(net_new, &new_uid, &new_gid);
1926
1927 /* The network namespace was changed but the owning user namespace is
1928 * identical so there's no need to change the owner of sysfs entries.
1929 */
1930 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
1931 return 0;
1932
1933 error = device_change_owner(dev, new_uid, new_gid);
1934 if (error)
1935 return error;
1936
1937 error = queue_change_owner(ndev, new_uid, new_gid);
1938 if (error)
1939 return error;
1940
1941 return 0;
1942}
1943
1944int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1945 const void *ns)
1946{
1947 return class_create_file_ns(&net_class, class_attr, ns);
1948}
1949EXPORT_SYMBOL(netdev_class_create_file_ns);
1950
1951void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1952 const void *ns)
1953{
1954 class_remove_file_ns(&net_class, class_attr, ns);
1955}
1956EXPORT_SYMBOL(netdev_class_remove_file_ns);
1957
1958int __init netdev_kobject_init(void)
1959{
1960 kobj_ns_type_register(&net_ns_type_operations);
1961 return class_register(&net_class);
1962}