Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net-sysfs.c - network device class and attributes
4 *
5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
6 */
7
8#include <linux/capability.h>
9#include <linux/kernel.h>
10#include <linux/netdevice.h>
11#include <linux/if_arp.h>
12#include <linux/slab.h>
13#include <linux/sched/signal.h>
14#include <linux/nsproxy.h>
15#include <net/sock.h>
16#include <net/net_namespace.h>
17#include <linux/rtnetlink.h>
18#include <linux/vmalloc.h>
19#include <linux/export.h>
20#include <linux/jiffies.h>
21#include <linux/pm_runtime.h>
22#include <linux/of.h>
23#include <linux/of_net.h>
24#include <linux/cpu.h>
25
26#include "net-sysfs.h"
27
28#ifdef CONFIG_SYSFS
29static const char fmt_hex[] = "%#x\n";
30static const char fmt_dec[] = "%d\n";
31static const char fmt_ulong[] = "%lu\n";
32static const char fmt_u64[] = "%llu\n";
33
34static inline int dev_isalive(const struct net_device *dev)
35{
36 return dev->reg_state <= NETREG_REGISTERED;
37}
38
39/* use same locking rules as GIF* ioctl's */
40static ssize_t netdev_show(const struct device *dev,
41 struct device_attribute *attr, char *buf,
42 ssize_t (*format)(const struct net_device *, char *))
43{
44 struct net_device *ndev = to_net_dev(dev);
45 ssize_t ret = -EINVAL;
46
47 read_lock(&dev_base_lock);
48 if (dev_isalive(ndev))
49 ret = (*format)(ndev, buf);
50 read_unlock(&dev_base_lock);
51
52 return ret;
53}
54
55/* generate a show function for simple field */
56#define NETDEVICE_SHOW(field, format_string) \
57static ssize_t format_##field(const struct net_device *dev, char *buf) \
58{ \
59 return sprintf(buf, format_string, dev->field); \
60} \
61static ssize_t field##_show(struct device *dev, \
62 struct device_attribute *attr, char *buf) \
63{ \
64 return netdev_show(dev, attr, buf, format_##field); \
65} \
66
67#define NETDEVICE_SHOW_RO(field, format_string) \
68NETDEVICE_SHOW(field, format_string); \
69static DEVICE_ATTR_RO(field)
70
71#define NETDEVICE_SHOW_RW(field, format_string) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RW(field)
74
75/* use same locking and permission rules as SIF* ioctl's */
76static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
77 const char *buf, size_t len,
78 int (*set)(struct net_device *, unsigned long))
79{
80 struct net_device *netdev = to_net_dev(dev);
81 struct net *net = dev_net(netdev);
82 unsigned long new;
83 int ret = -EINVAL;
84
85 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
86 return -EPERM;
87
88 ret = kstrtoul(buf, 0, &new);
89 if (ret)
90 goto err;
91
92 if (!rtnl_trylock())
93 return restart_syscall();
94
95 if (dev_isalive(netdev)) {
96 ret = (*set)(netdev, new);
97 if (ret == 0)
98 ret = len;
99 }
100 rtnl_unlock();
101 err:
102 return ret;
103}
104
105NETDEVICE_SHOW_RO(dev_id, fmt_hex);
106NETDEVICE_SHOW_RO(dev_port, fmt_dec);
107NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
108NETDEVICE_SHOW_RO(addr_len, fmt_dec);
109NETDEVICE_SHOW_RO(ifindex, fmt_dec);
110NETDEVICE_SHOW_RO(type, fmt_dec);
111NETDEVICE_SHOW_RO(link_mode, fmt_dec);
112
113static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
114 char *buf)
115{
116 struct net_device *ndev = to_net_dev(dev);
117
118 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
119}
120static DEVICE_ATTR_RO(iflink);
121
122static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
123{
124 return sprintf(buf, fmt_dec, dev->name_assign_type);
125}
126
127static ssize_t name_assign_type_show(struct device *dev,
128 struct device_attribute *attr,
129 char *buf)
130{
131 struct net_device *ndev = to_net_dev(dev);
132 ssize_t ret = -EINVAL;
133
134 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
135 ret = netdev_show(dev, attr, buf, format_name_assign_type);
136
137 return ret;
138}
139static DEVICE_ATTR_RO(name_assign_type);
140
141/* use same locking rules as GIFHWADDR ioctl's */
142static ssize_t address_show(struct device *dev, struct device_attribute *attr,
143 char *buf)
144{
145 struct net_device *ndev = to_net_dev(dev);
146 ssize_t ret = -EINVAL;
147
148 read_lock(&dev_base_lock);
149 if (dev_isalive(ndev))
150 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
151 read_unlock(&dev_base_lock);
152 return ret;
153}
154static DEVICE_ATTR_RO(address);
155
156static ssize_t broadcast_show(struct device *dev,
157 struct device_attribute *attr, char *buf)
158{
159 struct net_device *ndev = to_net_dev(dev);
160
161 if (dev_isalive(ndev))
162 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
163 return -EINVAL;
164}
165static DEVICE_ATTR_RO(broadcast);
166
167static int change_carrier(struct net_device *dev, unsigned long new_carrier)
168{
169 if (!netif_running(dev))
170 return -EINVAL;
171 return dev_change_carrier(dev, (bool)new_carrier);
172}
173
174static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
175 const char *buf, size_t len)
176{
177 return netdev_store(dev, attr, buf, len, change_carrier);
178}
179
180static ssize_t carrier_show(struct device *dev,
181 struct device_attribute *attr, char *buf)
182{
183 struct net_device *netdev = to_net_dev(dev);
184
185 if (netif_running(netdev))
186 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
187
188 return -EINVAL;
189}
190static DEVICE_ATTR_RW(carrier);
191
192static ssize_t speed_show(struct device *dev,
193 struct device_attribute *attr, char *buf)
194{
195 struct net_device *netdev = to_net_dev(dev);
196 int ret = -EINVAL;
197
198 if (!rtnl_trylock())
199 return restart_syscall();
200
201 if (netif_running(netdev)) {
202 struct ethtool_link_ksettings cmd;
203
204 if (!__ethtool_get_link_ksettings(netdev, &cmd))
205 ret = sprintf(buf, fmt_dec, cmd.base.speed);
206 }
207 rtnl_unlock();
208 return ret;
209}
210static DEVICE_ATTR_RO(speed);
211
212static ssize_t duplex_show(struct device *dev,
213 struct device_attribute *attr, char *buf)
214{
215 struct net_device *netdev = to_net_dev(dev);
216 int ret = -EINVAL;
217
218 if (!rtnl_trylock())
219 return restart_syscall();
220
221 if (netif_running(netdev)) {
222 struct ethtool_link_ksettings cmd;
223
224 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
225 const char *duplex;
226
227 switch (cmd.base.duplex) {
228 case DUPLEX_HALF:
229 duplex = "half";
230 break;
231 case DUPLEX_FULL:
232 duplex = "full";
233 break;
234 default:
235 duplex = "unknown";
236 break;
237 }
238 ret = sprintf(buf, "%s\n", duplex);
239 }
240 }
241 rtnl_unlock();
242 return ret;
243}
244static DEVICE_ATTR_RO(duplex);
245
246static ssize_t dormant_show(struct device *dev,
247 struct device_attribute *attr, char *buf)
248{
249 struct net_device *netdev = to_net_dev(dev);
250
251 if (netif_running(netdev))
252 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
253
254 return -EINVAL;
255}
256static DEVICE_ATTR_RO(dormant);
257
258static const char *const operstates[] = {
259 "unknown",
260 "notpresent", /* currently unused */
261 "down",
262 "lowerlayerdown",
263 "testing", /* currently unused */
264 "dormant",
265 "up"
266};
267
268static ssize_t operstate_show(struct device *dev,
269 struct device_attribute *attr, char *buf)
270{
271 const struct net_device *netdev = to_net_dev(dev);
272 unsigned char operstate;
273
274 read_lock(&dev_base_lock);
275 operstate = netdev->operstate;
276 if (!netif_running(netdev))
277 operstate = IF_OPER_DOWN;
278 read_unlock(&dev_base_lock);
279
280 if (operstate >= ARRAY_SIZE(operstates))
281 return -EINVAL; /* should not happen */
282
283 return sprintf(buf, "%s\n", operstates[operstate]);
284}
285static DEVICE_ATTR_RO(operstate);
286
287static ssize_t carrier_changes_show(struct device *dev,
288 struct device_attribute *attr,
289 char *buf)
290{
291 struct net_device *netdev = to_net_dev(dev);
292
293 return sprintf(buf, fmt_dec,
294 atomic_read(&netdev->carrier_up_count) +
295 atomic_read(&netdev->carrier_down_count));
296}
297static DEVICE_ATTR_RO(carrier_changes);
298
299static ssize_t carrier_up_count_show(struct device *dev,
300 struct device_attribute *attr,
301 char *buf)
302{
303 struct net_device *netdev = to_net_dev(dev);
304
305 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
306}
307static DEVICE_ATTR_RO(carrier_up_count);
308
309static ssize_t carrier_down_count_show(struct device *dev,
310 struct device_attribute *attr,
311 char *buf)
312{
313 struct net_device *netdev = to_net_dev(dev);
314
315 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
316}
317static DEVICE_ATTR_RO(carrier_down_count);
318
319/* read-write attributes */
320
321static int change_mtu(struct net_device *dev, unsigned long new_mtu)
322{
323 return dev_set_mtu(dev, (int)new_mtu);
324}
325
326static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
327 const char *buf, size_t len)
328{
329 return netdev_store(dev, attr, buf, len, change_mtu);
330}
331NETDEVICE_SHOW_RW(mtu, fmt_dec);
332
333static int change_flags(struct net_device *dev, unsigned long new_flags)
334{
335 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
336}
337
338static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
339 const char *buf, size_t len)
340{
341 return netdev_store(dev, attr, buf, len, change_flags);
342}
343NETDEVICE_SHOW_RW(flags, fmt_hex);
344
345static ssize_t tx_queue_len_store(struct device *dev,
346 struct device_attribute *attr,
347 const char *buf, size_t len)
348{
349 if (!capable(CAP_NET_ADMIN))
350 return -EPERM;
351
352 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
353}
354NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
355
356static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
357{
358 dev->gro_flush_timeout = val;
359 return 0;
360}
361
362static ssize_t gro_flush_timeout_store(struct device *dev,
363 struct device_attribute *attr,
364 const char *buf, size_t len)
365{
366 if (!capable(CAP_NET_ADMIN))
367 return -EPERM;
368
369 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
370}
371NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
372
373static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
375{
376 struct net_device *netdev = to_net_dev(dev);
377 struct net *net = dev_net(netdev);
378 size_t count = len;
379 ssize_t ret = 0;
380
381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
382 return -EPERM;
383
384 /* ignore trailing newline */
385 if (len > 0 && buf[len - 1] == '\n')
386 --count;
387
388 if (!rtnl_trylock())
389 return restart_syscall();
390
391 if (dev_isalive(netdev)) {
392 ret = dev_set_alias(netdev, buf, count);
393 if (ret < 0)
394 goto err;
395 ret = len;
396 netdev_state_change(netdev);
397 }
398err:
399 rtnl_unlock();
400
401 return ret;
402}
403
404static ssize_t ifalias_show(struct device *dev,
405 struct device_attribute *attr, char *buf)
406{
407 const struct net_device *netdev = to_net_dev(dev);
408 char tmp[IFALIASZ];
409 ssize_t ret = 0;
410
411 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
412 if (ret > 0)
413 ret = sprintf(buf, "%s\n", tmp);
414 return ret;
415}
416static DEVICE_ATTR_RW(ifalias);
417
418static int change_group(struct net_device *dev, unsigned long new_group)
419{
420 dev_set_group(dev, (int)new_group);
421 return 0;
422}
423
424static ssize_t group_store(struct device *dev, struct device_attribute *attr,
425 const char *buf, size_t len)
426{
427 return netdev_store(dev, attr, buf, len, change_group);
428}
429NETDEVICE_SHOW(group, fmt_dec);
430static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
431
432static int change_proto_down(struct net_device *dev, unsigned long proto_down)
433{
434 return dev_change_proto_down(dev, (bool)proto_down);
435}
436
437static ssize_t proto_down_store(struct device *dev,
438 struct device_attribute *attr,
439 const char *buf, size_t len)
440{
441 return netdev_store(dev, attr, buf, len, change_proto_down);
442}
443NETDEVICE_SHOW_RW(proto_down, fmt_dec);
444
445static ssize_t phys_port_id_show(struct device *dev,
446 struct device_attribute *attr, char *buf)
447{
448 struct net_device *netdev = to_net_dev(dev);
449 ssize_t ret = -EINVAL;
450
451 if (!rtnl_trylock())
452 return restart_syscall();
453
454 if (dev_isalive(netdev)) {
455 struct netdev_phys_item_id ppid;
456
457 ret = dev_get_phys_port_id(netdev, &ppid);
458 if (!ret)
459 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
460 }
461 rtnl_unlock();
462
463 return ret;
464}
465static DEVICE_ATTR_RO(phys_port_id);
466
467static ssize_t phys_port_name_show(struct device *dev,
468 struct device_attribute *attr, char *buf)
469{
470 struct net_device *netdev = to_net_dev(dev);
471 ssize_t ret = -EINVAL;
472
473 if (!rtnl_trylock())
474 return restart_syscall();
475
476 if (dev_isalive(netdev)) {
477 char name[IFNAMSIZ];
478
479 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
480 if (!ret)
481 ret = sprintf(buf, "%s\n", name);
482 }
483 rtnl_unlock();
484
485 return ret;
486}
487static DEVICE_ATTR_RO(phys_port_name);
488
489static ssize_t phys_switch_id_show(struct device *dev,
490 struct device_attribute *attr, char *buf)
491{
492 struct net_device *netdev = to_net_dev(dev);
493 ssize_t ret = -EINVAL;
494
495 if (!rtnl_trylock())
496 return restart_syscall();
497
498 if (dev_isalive(netdev)) {
499 struct netdev_phys_item_id ppid = { };
500
501 ret = dev_get_port_parent_id(netdev, &ppid, false);
502 if (!ret)
503 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
504 }
505 rtnl_unlock();
506
507 return ret;
508}
509static DEVICE_ATTR_RO(phys_switch_id);
510
511static struct attribute *net_class_attrs[] __ro_after_init = {
512 &dev_attr_netdev_group.attr,
513 &dev_attr_type.attr,
514 &dev_attr_dev_id.attr,
515 &dev_attr_dev_port.attr,
516 &dev_attr_iflink.attr,
517 &dev_attr_ifindex.attr,
518 &dev_attr_name_assign_type.attr,
519 &dev_attr_addr_assign_type.attr,
520 &dev_attr_addr_len.attr,
521 &dev_attr_link_mode.attr,
522 &dev_attr_address.attr,
523 &dev_attr_broadcast.attr,
524 &dev_attr_speed.attr,
525 &dev_attr_duplex.attr,
526 &dev_attr_dormant.attr,
527 &dev_attr_operstate.attr,
528 &dev_attr_carrier_changes.attr,
529 &dev_attr_ifalias.attr,
530 &dev_attr_carrier.attr,
531 &dev_attr_mtu.attr,
532 &dev_attr_flags.attr,
533 &dev_attr_tx_queue_len.attr,
534 &dev_attr_gro_flush_timeout.attr,
535 &dev_attr_phys_port_id.attr,
536 &dev_attr_phys_port_name.attr,
537 &dev_attr_phys_switch_id.attr,
538 &dev_attr_proto_down.attr,
539 &dev_attr_carrier_up_count.attr,
540 &dev_attr_carrier_down_count.attr,
541 NULL,
542};
543ATTRIBUTE_GROUPS(net_class);
544
545/* Show a given an attribute in the statistics group */
546static ssize_t netstat_show(const struct device *d,
547 struct device_attribute *attr, char *buf,
548 unsigned long offset)
549{
550 struct net_device *dev = to_net_dev(d);
551 ssize_t ret = -EINVAL;
552
553 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
554 offset % sizeof(u64) != 0);
555
556 read_lock(&dev_base_lock);
557 if (dev_isalive(dev)) {
558 struct rtnl_link_stats64 temp;
559 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
560
561 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
562 }
563 read_unlock(&dev_base_lock);
564 return ret;
565}
566
567/* generate a read-only statistics attribute */
568#define NETSTAT_ENTRY(name) \
569static ssize_t name##_show(struct device *d, \
570 struct device_attribute *attr, char *buf) \
571{ \
572 return netstat_show(d, attr, buf, \
573 offsetof(struct rtnl_link_stats64, name)); \
574} \
575static DEVICE_ATTR_RO(name)
576
577NETSTAT_ENTRY(rx_packets);
578NETSTAT_ENTRY(tx_packets);
579NETSTAT_ENTRY(rx_bytes);
580NETSTAT_ENTRY(tx_bytes);
581NETSTAT_ENTRY(rx_errors);
582NETSTAT_ENTRY(tx_errors);
583NETSTAT_ENTRY(rx_dropped);
584NETSTAT_ENTRY(tx_dropped);
585NETSTAT_ENTRY(multicast);
586NETSTAT_ENTRY(collisions);
587NETSTAT_ENTRY(rx_length_errors);
588NETSTAT_ENTRY(rx_over_errors);
589NETSTAT_ENTRY(rx_crc_errors);
590NETSTAT_ENTRY(rx_frame_errors);
591NETSTAT_ENTRY(rx_fifo_errors);
592NETSTAT_ENTRY(rx_missed_errors);
593NETSTAT_ENTRY(tx_aborted_errors);
594NETSTAT_ENTRY(tx_carrier_errors);
595NETSTAT_ENTRY(tx_fifo_errors);
596NETSTAT_ENTRY(tx_heartbeat_errors);
597NETSTAT_ENTRY(tx_window_errors);
598NETSTAT_ENTRY(rx_compressed);
599NETSTAT_ENTRY(tx_compressed);
600NETSTAT_ENTRY(rx_nohandler);
601
602static struct attribute *netstat_attrs[] __ro_after_init = {
603 &dev_attr_rx_packets.attr,
604 &dev_attr_tx_packets.attr,
605 &dev_attr_rx_bytes.attr,
606 &dev_attr_tx_bytes.attr,
607 &dev_attr_rx_errors.attr,
608 &dev_attr_tx_errors.attr,
609 &dev_attr_rx_dropped.attr,
610 &dev_attr_tx_dropped.attr,
611 &dev_attr_multicast.attr,
612 &dev_attr_collisions.attr,
613 &dev_attr_rx_length_errors.attr,
614 &dev_attr_rx_over_errors.attr,
615 &dev_attr_rx_crc_errors.attr,
616 &dev_attr_rx_frame_errors.attr,
617 &dev_attr_rx_fifo_errors.attr,
618 &dev_attr_rx_missed_errors.attr,
619 &dev_attr_tx_aborted_errors.attr,
620 &dev_attr_tx_carrier_errors.attr,
621 &dev_attr_tx_fifo_errors.attr,
622 &dev_attr_tx_heartbeat_errors.attr,
623 &dev_attr_tx_window_errors.attr,
624 &dev_attr_rx_compressed.attr,
625 &dev_attr_tx_compressed.attr,
626 &dev_attr_rx_nohandler.attr,
627 NULL
628};
629
630static const struct attribute_group netstat_group = {
631 .name = "statistics",
632 .attrs = netstat_attrs,
633};
634
635#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
636static struct attribute *wireless_attrs[] = {
637 NULL
638};
639
640static const struct attribute_group wireless_group = {
641 .name = "wireless",
642 .attrs = wireless_attrs,
643};
644#endif
645
646#else /* CONFIG_SYSFS */
647#define net_class_groups NULL
648#endif /* CONFIG_SYSFS */
649
650#ifdef CONFIG_SYSFS
651#define to_rx_queue_attr(_attr) \
652 container_of(_attr, struct rx_queue_attribute, attr)
653
654#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
655
656static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
657 char *buf)
658{
659 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
660 struct netdev_rx_queue *queue = to_rx_queue(kobj);
661
662 if (!attribute->show)
663 return -EIO;
664
665 return attribute->show(queue, buf);
666}
667
668static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
669 const char *buf, size_t count)
670{
671 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
672 struct netdev_rx_queue *queue = to_rx_queue(kobj);
673
674 if (!attribute->store)
675 return -EIO;
676
677 return attribute->store(queue, buf, count);
678}
679
680static const struct sysfs_ops rx_queue_sysfs_ops = {
681 .show = rx_queue_attr_show,
682 .store = rx_queue_attr_store,
683};
684
685#ifdef CONFIG_RPS
686static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
687{
688 struct rps_map *map;
689 cpumask_var_t mask;
690 int i, len;
691
692 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
693 return -ENOMEM;
694
695 rcu_read_lock();
696 map = rcu_dereference(queue->rps_map);
697 if (map)
698 for (i = 0; i < map->len; i++)
699 cpumask_set_cpu(map->cpus[i], mask);
700
701 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
702 rcu_read_unlock();
703 free_cpumask_var(mask);
704
705 return len < PAGE_SIZE ? len : -EINVAL;
706}
707
708static ssize_t store_rps_map(struct netdev_rx_queue *queue,
709 const char *buf, size_t len)
710{
711 struct rps_map *old_map, *map;
712 cpumask_var_t mask;
713 int err, cpu, i;
714 static DEFINE_MUTEX(rps_map_mutex);
715
716 if (!capable(CAP_NET_ADMIN))
717 return -EPERM;
718
719 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
720 return -ENOMEM;
721
722 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
723 if (err) {
724 free_cpumask_var(mask);
725 return err;
726 }
727
728 map = kzalloc(max_t(unsigned int,
729 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
730 GFP_KERNEL);
731 if (!map) {
732 free_cpumask_var(mask);
733 return -ENOMEM;
734 }
735
736 i = 0;
737 for_each_cpu_and(cpu, mask, cpu_online_mask)
738 map->cpus[i++] = cpu;
739
740 if (i) {
741 map->len = i;
742 } else {
743 kfree(map);
744 map = NULL;
745 }
746
747 mutex_lock(&rps_map_mutex);
748 old_map = rcu_dereference_protected(queue->rps_map,
749 mutex_is_locked(&rps_map_mutex));
750 rcu_assign_pointer(queue->rps_map, map);
751
752 if (map)
753 static_branch_inc(&rps_needed);
754 if (old_map)
755 static_branch_dec(&rps_needed);
756
757 mutex_unlock(&rps_map_mutex);
758
759 if (old_map)
760 kfree_rcu(old_map, rcu);
761
762 free_cpumask_var(mask);
763 return len;
764}
765
766static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
767 char *buf)
768{
769 struct rps_dev_flow_table *flow_table;
770 unsigned long val = 0;
771
772 rcu_read_lock();
773 flow_table = rcu_dereference(queue->rps_flow_table);
774 if (flow_table)
775 val = (unsigned long)flow_table->mask + 1;
776 rcu_read_unlock();
777
778 return sprintf(buf, "%lu\n", val);
779}
780
781static void rps_dev_flow_table_release(struct rcu_head *rcu)
782{
783 struct rps_dev_flow_table *table = container_of(rcu,
784 struct rps_dev_flow_table, rcu);
785 vfree(table);
786}
787
788static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
789 const char *buf, size_t len)
790{
791 unsigned long mask, count;
792 struct rps_dev_flow_table *table, *old_table;
793 static DEFINE_SPINLOCK(rps_dev_flow_lock);
794 int rc;
795
796 if (!capable(CAP_NET_ADMIN))
797 return -EPERM;
798
799 rc = kstrtoul(buf, 0, &count);
800 if (rc < 0)
801 return rc;
802
803 if (count) {
804 mask = count - 1;
805 /* mask = roundup_pow_of_two(count) - 1;
806 * without overflows...
807 */
808 while ((mask | (mask >> 1)) != mask)
809 mask |= (mask >> 1);
810 /* On 64 bit arches, must check mask fits in table->mask (u32),
811 * and on 32bit arches, must check
812 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
813 */
814#if BITS_PER_LONG > 32
815 if (mask > (unsigned long)(u32)mask)
816 return -EINVAL;
817#else
818 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
819 / sizeof(struct rps_dev_flow)) {
820 /* Enforce a limit to prevent overflow */
821 return -EINVAL;
822 }
823#endif
824 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
825 if (!table)
826 return -ENOMEM;
827
828 table->mask = mask;
829 for (count = 0; count <= mask; count++)
830 table->flows[count].cpu = RPS_NO_CPU;
831 } else {
832 table = NULL;
833 }
834
835 spin_lock(&rps_dev_flow_lock);
836 old_table = rcu_dereference_protected(queue->rps_flow_table,
837 lockdep_is_held(&rps_dev_flow_lock));
838 rcu_assign_pointer(queue->rps_flow_table, table);
839 spin_unlock(&rps_dev_flow_lock);
840
841 if (old_table)
842 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
843
844 return len;
845}
846
847static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
848 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
849
850static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
851 = __ATTR(rps_flow_cnt, 0644,
852 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
853#endif /* CONFIG_RPS */
854
855static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
856#ifdef CONFIG_RPS
857 &rps_cpus_attribute.attr,
858 &rps_dev_flow_table_cnt_attribute.attr,
859#endif
860 NULL
861};
862ATTRIBUTE_GROUPS(rx_queue_default);
863
864static void rx_queue_release(struct kobject *kobj)
865{
866 struct netdev_rx_queue *queue = to_rx_queue(kobj);
867#ifdef CONFIG_RPS
868 struct rps_map *map;
869 struct rps_dev_flow_table *flow_table;
870
871 map = rcu_dereference_protected(queue->rps_map, 1);
872 if (map) {
873 RCU_INIT_POINTER(queue->rps_map, NULL);
874 kfree_rcu(map, rcu);
875 }
876
877 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
878 if (flow_table) {
879 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
880 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
881 }
882#endif
883
884 memset(kobj, 0, sizeof(*kobj));
885 dev_put(queue->dev);
886}
887
888static const void *rx_queue_namespace(struct kobject *kobj)
889{
890 struct netdev_rx_queue *queue = to_rx_queue(kobj);
891 struct device *dev = &queue->dev->dev;
892 const void *ns = NULL;
893
894 if (dev->class && dev->class->ns_type)
895 ns = dev->class->namespace(dev);
896
897 return ns;
898}
899
900static void rx_queue_get_ownership(struct kobject *kobj,
901 kuid_t *uid, kgid_t *gid)
902{
903 const struct net *net = rx_queue_namespace(kobj);
904
905 net_ns_get_ownership(net, uid, gid);
906}
907
908static struct kobj_type rx_queue_ktype __ro_after_init = {
909 .sysfs_ops = &rx_queue_sysfs_ops,
910 .release = rx_queue_release,
911 .default_groups = rx_queue_default_groups,
912 .namespace = rx_queue_namespace,
913 .get_ownership = rx_queue_get_ownership,
914};
915
916static int rx_queue_add_kobject(struct net_device *dev, int index)
917{
918 struct netdev_rx_queue *queue = dev->_rx + index;
919 struct kobject *kobj = &queue->kobj;
920 int error = 0;
921
922 kobj->kset = dev->queues_kset;
923 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
924 "rx-%u", index);
925 if (error)
926 goto err;
927
928 dev_hold(queue->dev);
929
930 if (dev->sysfs_rx_queue_group) {
931 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
932 if (error)
933 goto err;
934 }
935
936 kobject_uevent(kobj, KOBJ_ADD);
937
938 return error;
939
940err:
941 kobject_put(kobj);
942 return error;
943}
944#endif /* CONFIG_SYSFS */
945
946int
947net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
948{
949#ifdef CONFIG_SYSFS
950 int i;
951 int error = 0;
952
953#ifndef CONFIG_RPS
954 if (!dev->sysfs_rx_queue_group)
955 return 0;
956#endif
957 for (i = old_num; i < new_num; i++) {
958 error = rx_queue_add_kobject(dev, i);
959 if (error) {
960 new_num = old_num;
961 break;
962 }
963 }
964
965 while (--i >= new_num) {
966 struct kobject *kobj = &dev->_rx[i].kobj;
967
968 if (!refcount_read(&dev_net(dev)->count))
969 kobj->uevent_suppress = 1;
970 if (dev->sysfs_rx_queue_group)
971 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
972 kobject_put(kobj);
973 }
974
975 return error;
976#else
977 return 0;
978#endif
979}
980
981#ifdef CONFIG_SYSFS
982/*
983 * netdev_queue sysfs structures and functions.
984 */
985struct netdev_queue_attribute {
986 struct attribute attr;
987 ssize_t (*show)(struct netdev_queue *queue, char *buf);
988 ssize_t (*store)(struct netdev_queue *queue,
989 const char *buf, size_t len);
990};
991#define to_netdev_queue_attr(_attr) \
992 container_of(_attr, struct netdev_queue_attribute, attr)
993
994#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
995
996static ssize_t netdev_queue_attr_show(struct kobject *kobj,
997 struct attribute *attr, char *buf)
998{
999 const struct netdev_queue_attribute *attribute
1000 = to_netdev_queue_attr(attr);
1001 struct netdev_queue *queue = to_netdev_queue(kobj);
1002
1003 if (!attribute->show)
1004 return -EIO;
1005
1006 return attribute->show(queue, buf);
1007}
1008
1009static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1010 struct attribute *attr,
1011 const char *buf, size_t count)
1012{
1013 const struct netdev_queue_attribute *attribute
1014 = to_netdev_queue_attr(attr);
1015 struct netdev_queue *queue = to_netdev_queue(kobj);
1016
1017 if (!attribute->store)
1018 return -EIO;
1019
1020 return attribute->store(queue, buf, count);
1021}
1022
1023static const struct sysfs_ops netdev_queue_sysfs_ops = {
1024 .show = netdev_queue_attr_show,
1025 .store = netdev_queue_attr_store,
1026};
1027
1028static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1029{
1030 unsigned long trans_timeout;
1031
1032 spin_lock_irq(&queue->_xmit_lock);
1033 trans_timeout = queue->trans_timeout;
1034 spin_unlock_irq(&queue->_xmit_lock);
1035
1036 return sprintf(buf, "%lu", trans_timeout);
1037}
1038
1039static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1040{
1041 struct net_device *dev = queue->dev;
1042 unsigned int i;
1043
1044 i = queue - dev->_tx;
1045 BUG_ON(i >= dev->num_tx_queues);
1046
1047 return i;
1048}
1049
1050static ssize_t traffic_class_show(struct netdev_queue *queue,
1051 char *buf)
1052{
1053 struct net_device *dev = queue->dev;
1054 int index;
1055 int tc;
1056
1057 if (!netif_is_multiqueue(dev))
1058 return -ENOENT;
1059
1060 index = get_netdev_queue_index(queue);
1061
1062 /* If queue belongs to subordinate dev use its TC mapping */
1063 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1064
1065 tc = netdev_txq_to_tc(dev, index);
1066 if (tc < 0)
1067 return -EINVAL;
1068
1069 /* We can report the traffic class one of two ways:
1070 * Subordinate device traffic classes are reported with the traffic
1071 * class first, and then the subordinate class so for example TC0 on
1072 * subordinate device 2 will be reported as "0-2". If the queue
1073 * belongs to the root device it will be reported with just the
1074 * traffic class, so just "0" for TC 0 for example.
1075 */
1076 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1077 sprintf(buf, "%u\n", tc);
1078}
1079
1080#ifdef CONFIG_XPS
1081static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1082 char *buf)
1083{
1084 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1085}
1086
1087static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1088 const char *buf, size_t len)
1089{
1090 struct net_device *dev = queue->dev;
1091 int err, index = get_netdev_queue_index(queue);
1092 u32 rate = 0;
1093
1094 if (!capable(CAP_NET_ADMIN))
1095 return -EPERM;
1096
1097 err = kstrtou32(buf, 10, &rate);
1098 if (err < 0)
1099 return err;
1100
1101 if (!rtnl_trylock())
1102 return restart_syscall();
1103
1104 err = -EOPNOTSUPP;
1105 if (dev->netdev_ops->ndo_set_tx_maxrate)
1106 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1107
1108 rtnl_unlock();
1109 if (!err) {
1110 queue->tx_maxrate = rate;
1111 return len;
1112 }
1113 return err;
1114}
1115
1116static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1117 = __ATTR_RW(tx_maxrate);
1118#endif
1119
1120static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1121 = __ATTR_RO(tx_timeout);
1122
1123static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1124 = __ATTR_RO(traffic_class);
1125
1126#ifdef CONFIG_BQL
1127/*
1128 * Byte queue limits sysfs structures and functions.
1129 */
1130static ssize_t bql_show(char *buf, unsigned int value)
1131{
1132 return sprintf(buf, "%u\n", value);
1133}
1134
1135static ssize_t bql_set(const char *buf, const size_t count,
1136 unsigned int *pvalue)
1137{
1138 unsigned int value;
1139 int err;
1140
1141 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1142 value = DQL_MAX_LIMIT;
1143 } else {
1144 err = kstrtouint(buf, 10, &value);
1145 if (err < 0)
1146 return err;
1147 if (value > DQL_MAX_LIMIT)
1148 return -EINVAL;
1149 }
1150
1151 *pvalue = value;
1152
1153 return count;
1154}
1155
1156static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1157 char *buf)
1158{
1159 struct dql *dql = &queue->dql;
1160
1161 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1162}
1163
1164static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1165 const char *buf, size_t len)
1166{
1167 struct dql *dql = &queue->dql;
1168 unsigned int value;
1169 int err;
1170
1171 err = kstrtouint(buf, 10, &value);
1172 if (err < 0)
1173 return err;
1174
1175 dql->slack_hold_time = msecs_to_jiffies(value);
1176
1177 return len;
1178}
1179
1180static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1181 = __ATTR(hold_time, 0644,
1182 bql_show_hold_time, bql_set_hold_time);
1183
1184static ssize_t bql_show_inflight(struct netdev_queue *queue,
1185 char *buf)
1186{
1187 struct dql *dql = &queue->dql;
1188
1189 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1190}
1191
1192static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1193 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1194
1195#define BQL_ATTR(NAME, FIELD) \
1196static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1197 char *buf) \
1198{ \
1199 return bql_show(buf, queue->dql.FIELD); \
1200} \
1201 \
1202static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1203 const char *buf, size_t len) \
1204{ \
1205 return bql_set(buf, len, &queue->dql.FIELD); \
1206} \
1207 \
1208static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1209 = __ATTR(NAME, 0644, \
1210 bql_show_ ## NAME, bql_set_ ## NAME)
1211
1212BQL_ATTR(limit, limit);
1213BQL_ATTR(limit_max, max_limit);
1214BQL_ATTR(limit_min, min_limit);
1215
1216static struct attribute *dql_attrs[] __ro_after_init = {
1217 &bql_limit_attribute.attr,
1218 &bql_limit_max_attribute.attr,
1219 &bql_limit_min_attribute.attr,
1220 &bql_hold_time_attribute.attr,
1221 &bql_inflight_attribute.attr,
1222 NULL
1223};
1224
1225static const struct attribute_group dql_group = {
1226 .name = "byte_queue_limits",
1227 .attrs = dql_attrs,
1228};
1229#endif /* CONFIG_BQL */
1230
1231#ifdef CONFIG_XPS
1232static ssize_t xps_cpus_show(struct netdev_queue *queue,
1233 char *buf)
1234{
1235 struct net_device *dev = queue->dev;
1236 int cpu, len, num_tc = 1, tc = 0;
1237 struct xps_dev_maps *dev_maps;
1238 cpumask_var_t mask;
1239 unsigned long index;
1240
1241 if (!netif_is_multiqueue(dev))
1242 return -ENOENT;
1243
1244 index = get_netdev_queue_index(queue);
1245
1246 if (dev->num_tc) {
1247 /* Do not allow XPS on subordinate device directly */
1248 num_tc = dev->num_tc;
1249 if (num_tc < 0)
1250 return -EINVAL;
1251
1252 /* If queue belongs to subordinate dev use its map */
1253 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1254
1255 tc = netdev_txq_to_tc(dev, index);
1256 if (tc < 0)
1257 return -EINVAL;
1258 }
1259
1260 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1261 return -ENOMEM;
1262
1263 rcu_read_lock();
1264 dev_maps = rcu_dereference(dev->xps_cpus_map);
1265 if (dev_maps) {
1266 for_each_possible_cpu(cpu) {
1267 int i, tci = cpu * num_tc + tc;
1268 struct xps_map *map;
1269
1270 map = rcu_dereference(dev_maps->attr_map[tci]);
1271 if (!map)
1272 continue;
1273
1274 for (i = map->len; i--;) {
1275 if (map->queues[i] == index) {
1276 cpumask_set_cpu(cpu, mask);
1277 break;
1278 }
1279 }
1280 }
1281 }
1282 rcu_read_unlock();
1283
1284 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1285 free_cpumask_var(mask);
1286 return len < PAGE_SIZE ? len : -EINVAL;
1287}
1288
1289static ssize_t xps_cpus_store(struct netdev_queue *queue,
1290 const char *buf, size_t len)
1291{
1292 struct net_device *dev = queue->dev;
1293 unsigned long index;
1294 cpumask_var_t mask;
1295 int err;
1296
1297 if (!netif_is_multiqueue(dev))
1298 return -ENOENT;
1299
1300 if (!capable(CAP_NET_ADMIN))
1301 return -EPERM;
1302
1303 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1304 return -ENOMEM;
1305
1306 index = get_netdev_queue_index(queue);
1307
1308 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1309 if (err) {
1310 free_cpumask_var(mask);
1311 return err;
1312 }
1313
1314 err = netif_set_xps_queue(dev, mask, index);
1315
1316 free_cpumask_var(mask);
1317
1318 return err ? : len;
1319}
1320
1321static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1322 = __ATTR_RW(xps_cpus);
1323
1324static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1325{
1326 struct net_device *dev = queue->dev;
1327 struct xps_dev_maps *dev_maps;
1328 unsigned long *mask, index;
1329 int j, len, num_tc = 1, tc = 0;
1330
1331 index = get_netdev_queue_index(queue);
1332
1333 if (dev->num_tc) {
1334 num_tc = dev->num_tc;
1335 tc = netdev_txq_to_tc(dev, index);
1336 if (tc < 0)
1337 return -EINVAL;
1338 }
1339 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1340 if (!mask)
1341 return -ENOMEM;
1342
1343 rcu_read_lock();
1344 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1345 if (!dev_maps)
1346 goto out_no_maps;
1347
1348 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1349 j < dev->num_rx_queues;) {
1350 int i, tci = j * num_tc + tc;
1351 struct xps_map *map;
1352
1353 map = rcu_dereference(dev_maps->attr_map[tci]);
1354 if (!map)
1355 continue;
1356
1357 for (i = map->len; i--;) {
1358 if (map->queues[i] == index) {
1359 set_bit(j, mask);
1360 break;
1361 }
1362 }
1363 }
1364out_no_maps:
1365 rcu_read_unlock();
1366
1367 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1368 bitmap_free(mask);
1369
1370 return len < PAGE_SIZE ? len : -EINVAL;
1371}
1372
1373static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1374 size_t len)
1375{
1376 struct net_device *dev = queue->dev;
1377 struct net *net = dev_net(dev);
1378 unsigned long *mask, index;
1379 int err;
1380
1381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1382 return -EPERM;
1383
1384 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1385 if (!mask)
1386 return -ENOMEM;
1387
1388 index = get_netdev_queue_index(queue);
1389
1390 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1391 if (err) {
1392 bitmap_free(mask);
1393 return err;
1394 }
1395
1396 cpus_read_lock();
1397 err = __netif_set_xps_queue(dev, mask, index, true);
1398 cpus_read_unlock();
1399
1400 bitmap_free(mask);
1401 return err ? : len;
1402}
1403
1404static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1405 = __ATTR_RW(xps_rxqs);
1406#endif /* CONFIG_XPS */
1407
1408static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1409 &queue_trans_timeout.attr,
1410 &queue_traffic_class.attr,
1411#ifdef CONFIG_XPS
1412 &xps_cpus_attribute.attr,
1413 &xps_rxqs_attribute.attr,
1414 &queue_tx_maxrate.attr,
1415#endif
1416 NULL
1417};
1418ATTRIBUTE_GROUPS(netdev_queue_default);
1419
1420static void netdev_queue_release(struct kobject *kobj)
1421{
1422 struct netdev_queue *queue = to_netdev_queue(kobj);
1423
1424 memset(kobj, 0, sizeof(*kobj));
1425 dev_put(queue->dev);
1426}
1427
1428static const void *netdev_queue_namespace(struct kobject *kobj)
1429{
1430 struct netdev_queue *queue = to_netdev_queue(kobj);
1431 struct device *dev = &queue->dev->dev;
1432 const void *ns = NULL;
1433
1434 if (dev->class && dev->class->ns_type)
1435 ns = dev->class->namespace(dev);
1436
1437 return ns;
1438}
1439
1440static void netdev_queue_get_ownership(struct kobject *kobj,
1441 kuid_t *uid, kgid_t *gid)
1442{
1443 const struct net *net = netdev_queue_namespace(kobj);
1444
1445 net_ns_get_ownership(net, uid, gid);
1446}
1447
1448static struct kobj_type netdev_queue_ktype __ro_after_init = {
1449 .sysfs_ops = &netdev_queue_sysfs_ops,
1450 .release = netdev_queue_release,
1451 .default_groups = netdev_queue_default_groups,
1452 .namespace = netdev_queue_namespace,
1453 .get_ownership = netdev_queue_get_ownership,
1454};
1455
1456static int netdev_queue_add_kobject(struct net_device *dev, int index)
1457{
1458 struct netdev_queue *queue = dev->_tx + index;
1459 struct kobject *kobj = &queue->kobj;
1460 int error = 0;
1461
1462 kobj->kset = dev->queues_kset;
1463 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1464 "tx-%u", index);
1465 if (error)
1466 goto err;
1467
1468 dev_hold(queue->dev);
1469
1470#ifdef CONFIG_BQL
1471 error = sysfs_create_group(kobj, &dql_group);
1472 if (error)
1473 goto err;
1474#endif
1475
1476 kobject_uevent(kobj, KOBJ_ADD);
1477 return 0;
1478
1479err:
1480 kobject_put(kobj);
1481 return error;
1482}
1483#endif /* CONFIG_SYSFS */
1484
1485int
1486netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1487{
1488#ifdef CONFIG_SYSFS
1489 int i;
1490 int error = 0;
1491
1492 for (i = old_num; i < new_num; i++) {
1493 error = netdev_queue_add_kobject(dev, i);
1494 if (error) {
1495 new_num = old_num;
1496 break;
1497 }
1498 }
1499
1500 while (--i >= new_num) {
1501 struct netdev_queue *queue = dev->_tx + i;
1502
1503 if (!refcount_read(&dev_net(dev)->count))
1504 queue->kobj.uevent_suppress = 1;
1505#ifdef CONFIG_BQL
1506 sysfs_remove_group(&queue->kobj, &dql_group);
1507#endif
1508 kobject_put(&queue->kobj);
1509 }
1510
1511 return error;
1512#else
1513 return 0;
1514#endif /* CONFIG_SYSFS */
1515}
1516
1517static int register_queue_kobjects(struct net_device *dev)
1518{
1519 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1520
1521#ifdef CONFIG_SYSFS
1522 dev->queues_kset = kset_create_and_add("queues",
1523 NULL, &dev->dev.kobj);
1524 if (!dev->queues_kset)
1525 return -ENOMEM;
1526 real_rx = dev->real_num_rx_queues;
1527#endif
1528 real_tx = dev->real_num_tx_queues;
1529
1530 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1531 if (error)
1532 goto error;
1533 rxq = real_rx;
1534
1535 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1536 if (error)
1537 goto error;
1538 txq = real_tx;
1539
1540 return 0;
1541
1542error:
1543 netdev_queue_update_kobjects(dev, txq, 0);
1544 net_rx_queue_update_kobjects(dev, rxq, 0);
1545#ifdef CONFIG_SYSFS
1546 kset_unregister(dev->queues_kset);
1547#endif
1548 return error;
1549}
1550
1551static void remove_queue_kobjects(struct net_device *dev)
1552{
1553 int real_rx = 0, real_tx = 0;
1554
1555#ifdef CONFIG_SYSFS
1556 real_rx = dev->real_num_rx_queues;
1557#endif
1558 real_tx = dev->real_num_tx_queues;
1559
1560 net_rx_queue_update_kobjects(dev, real_rx, 0);
1561 netdev_queue_update_kobjects(dev, real_tx, 0);
1562#ifdef CONFIG_SYSFS
1563 kset_unregister(dev->queues_kset);
1564#endif
1565}
1566
1567static bool net_current_may_mount(void)
1568{
1569 struct net *net = current->nsproxy->net_ns;
1570
1571 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1572}
1573
1574static void *net_grab_current_ns(void)
1575{
1576 struct net *ns = current->nsproxy->net_ns;
1577#ifdef CONFIG_NET_NS
1578 if (ns)
1579 refcount_inc(&ns->passive);
1580#endif
1581 return ns;
1582}
1583
1584static const void *net_initial_ns(void)
1585{
1586 return &init_net;
1587}
1588
1589static const void *net_netlink_ns(struct sock *sk)
1590{
1591 return sock_net(sk);
1592}
1593
1594const struct kobj_ns_type_operations net_ns_type_operations = {
1595 .type = KOBJ_NS_TYPE_NET,
1596 .current_may_mount = net_current_may_mount,
1597 .grab_current_ns = net_grab_current_ns,
1598 .netlink_ns = net_netlink_ns,
1599 .initial_ns = net_initial_ns,
1600 .drop_ns = net_drop_ns,
1601};
1602EXPORT_SYMBOL_GPL(net_ns_type_operations);
1603
1604static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1605{
1606 struct net_device *dev = to_net_dev(d);
1607 int retval;
1608
1609 /* pass interface to uevent. */
1610 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1611 if (retval)
1612 goto exit;
1613
1614 /* pass ifindex to uevent.
1615 * ifindex is useful as it won't change (interface name may change)
1616 * and is what RtNetlink uses natively.
1617 */
1618 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1619
1620exit:
1621 return retval;
1622}
1623
1624/*
1625 * netdev_release -- destroy and free a dead device.
1626 * Called when last reference to device kobject is gone.
1627 */
1628static void netdev_release(struct device *d)
1629{
1630 struct net_device *dev = to_net_dev(d);
1631
1632 BUG_ON(dev->reg_state != NETREG_RELEASED);
1633
1634 /* no need to wait for rcu grace period:
1635 * device is dead and about to be freed.
1636 */
1637 kfree(rcu_access_pointer(dev->ifalias));
1638 netdev_freemem(dev);
1639}
1640
1641static const void *net_namespace(struct device *d)
1642{
1643 struct net_device *dev = to_net_dev(d);
1644
1645 return dev_net(dev);
1646}
1647
1648static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1649{
1650 struct net_device *dev = to_net_dev(d);
1651 const struct net *net = dev_net(dev);
1652
1653 net_ns_get_ownership(net, uid, gid);
1654}
1655
1656static struct class net_class __ro_after_init = {
1657 .name = "net",
1658 .dev_release = netdev_release,
1659 .dev_groups = net_class_groups,
1660 .dev_uevent = netdev_uevent,
1661 .ns_type = &net_ns_type_operations,
1662 .namespace = net_namespace,
1663 .get_ownership = net_get_ownership,
1664};
1665
1666#ifdef CONFIG_OF_NET
1667static int of_dev_node_match(struct device *dev, const void *data)
1668{
1669 int ret = 0;
1670
1671 if (dev->parent)
1672 ret = dev->parent->of_node == data;
1673
1674 return ret == 0 ? dev->of_node == data : ret;
1675}
1676
1677/*
1678 * of_find_net_device_by_node - lookup the net device for the device node
1679 * @np: OF device node
1680 *
1681 * Looks up the net_device structure corresponding with the device node.
1682 * If successful, returns a pointer to the net_device with the embedded
1683 * struct device refcount incremented by one, or NULL on failure. The
1684 * refcount must be dropped when done with the net_device.
1685 */
1686struct net_device *of_find_net_device_by_node(struct device_node *np)
1687{
1688 struct device *dev;
1689
1690 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1691 if (!dev)
1692 return NULL;
1693
1694 return to_net_dev(dev);
1695}
1696EXPORT_SYMBOL(of_find_net_device_by_node);
1697#endif
1698
1699/* Delete sysfs entries but hold kobject reference until after all
1700 * netdev references are gone.
1701 */
1702void netdev_unregister_kobject(struct net_device *ndev)
1703{
1704 struct device *dev = &ndev->dev;
1705
1706 if (!refcount_read(&dev_net(ndev)->count))
1707 dev_set_uevent_suppress(dev, 1);
1708
1709 kobject_get(&dev->kobj);
1710
1711 remove_queue_kobjects(ndev);
1712
1713 pm_runtime_set_memalloc_noio(dev, false);
1714
1715 device_del(dev);
1716}
1717
1718/* Create sysfs entries for network device. */
1719int netdev_register_kobject(struct net_device *ndev)
1720{
1721 struct device *dev = &ndev->dev;
1722 const struct attribute_group **groups = ndev->sysfs_groups;
1723 int error = 0;
1724
1725 device_initialize(dev);
1726 dev->class = &net_class;
1727 dev->platform_data = ndev;
1728 dev->groups = groups;
1729
1730 dev_set_name(dev, "%s", ndev->name);
1731
1732#ifdef CONFIG_SYSFS
1733 /* Allow for a device specific group */
1734 if (*groups)
1735 groups++;
1736
1737 *groups++ = &netstat_group;
1738
1739#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1740 if (ndev->ieee80211_ptr)
1741 *groups++ = &wireless_group;
1742#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1743 else if (ndev->wireless_handlers)
1744 *groups++ = &wireless_group;
1745#endif
1746#endif
1747#endif /* CONFIG_SYSFS */
1748
1749 error = device_add(dev);
1750 if (error)
1751 return error;
1752
1753 error = register_queue_kobjects(ndev);
1754 if (error) {
1755 device_del(dev);
1756 return error;
1757 }
1758
1759 pm_runtime_set_memalloc_noio(dev, true);
1760
1761 return error;
1762}
1763
1764int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1765 const void *ns)
1766{
1767 return class_create_file_ns(&net_class, class_attr, ns);
1768}
1769EXPORT_SYMBOL(netdev_class_create_file_ns);
1770
1771void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1772 const void *ns)
1773{
1774 class_remove_file_ns(&net_class, class_attr, ns);
1775}
1776EXPORT_SYMBOL(netdev_class_remove_file_ns);
1777
1778int __init netdev_kobject_init(void)
1779{
1780 kobj_ns_type_register(&net_ns_type_operations);
1781 return class_register(&net_class);
1782}
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#include <net/netdev_rx_queue.h>
27
28#include "dev.h"
29#include "net-sysfs.h"
30
31#ifdef CONFIG_SYSFS
32static const char fmt_hex[] = "%#x\n";
33static const char fmt_dec[] = "%d\n";
34static const char fmt_ulong[] = "%lu\n";
35static const char fmt_u64[] = "%llu\n";
36
37/* Caller holds RTNL or dev_base_lock */
38static inline int dev_isalive(const struct net_device *dev)
39{
40 return dev->reg_state <= NETREG_REGISTERED;
41}
42
43/* use same locking rules as GIF* ioctl's */
44static ssize_t netdev_show(const struct device *dev,
45 struct device_attribute *attr, char *buf,
46 ssize_t (*format)(const struct net_device *, char *))
47{
48 struct net_device *ndev = to_net_dev(dev);
49 ssize_t ret = -EINVAL;
50
51 read_lock(&dev_base_lock);
52 if (dev_isalive(ndev))
53 ret = (*format)(ndev, buf);
54 read_unlock(&dev_base_lock);
55
56 return ret;
57}
58
59/* generate a show function for simple field */
60#define NETDEVICE_SHOW(field, format_string) \
61static ssize_t format_##field(const struct net_device *dev, char *buf) \
62{ \
63 return sysfs_emit(buf, format_string, dev->field); \
64} \
65static ssize_t field##_show(struct device *dev, \
66 struct device_attribute *attr, char *buf) \
67{ \
68 return netdev_show(dev, attr, buf, format_##field); \
69} \
70
71#define NETDEVICE_SHOW_RO(field, format_string) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RO(field)
74
75#define NETDEVICE_SHOW_RW(field, format_string) \
76NETDEVICE_SHOW(field, format_string); \
77static DEVICE_ATTR_RW(field)
78
79/* use same locking and permission rules as SIF* ioctl's */
80static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81 const char *buf, size_t len,
82 int (*set)(struct net_device *, unsigned long))
83{
84 struct net_device *netdev = to_net_dev(dev);
85 struct net *net = dev_net(netdev);
86 unsigned long new;
87 int ret;
88
89 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90 return -EPERM;
91
92 ret = kstrtoul(buf, 0, &new);
93 if (ret)
94 goto err;
95
96 if (!rtnl_trylock())
97 return restart_syscall();
98
99 if (dev_isalive(netdev)) {
100 ret = (*set)(netdev, new);
101 if (ret == 0)
102 ret = len;
103 }
104 rtnl_unlock();
105 err:
106 return ret;
107}
108
109NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114NETDEVICE_SHOW_RO(type, fmt_dec);
115NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116
117static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 char *buf)
119{
120 struct net_device *ndev = to_net_dev(dev);
121
122 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
123}
124static DEVICE_ATTR_RO(iflink);
125
126static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127{
128 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
129}
130
131static ssize_t name_assign_type_show(struct device *dev,
132 struct device_attribute *attr,
133 char *buf)
134{
135 struct net_device *ndev = to_net_dev(dev);
136 ssize_t ret = -EINVAL;
137
138 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140
141 return ret;
142}
143static DEVICE_ATTR_RO(name_assign_type);
144
145/* use same locking rules as GIFHWADDR ioctl's */
146static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 char *buf)
148{
149 struct net_device *ndev = to_net_dev(dev);
150 ssize_t ret = -EINVAL;
151
152 read_lock(&dev_base_lock);
153 if (dev_isalive(ndev))
154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155 read_unlock(&dev_base_lock);
156 return ret;
157}
158static DEVICE_ATTR_RO(address);
159
160static ssize_t broadcast_show(struct device *dev,
161 struct device_attribute *attr, char *buf)
162{
163 struct net_device *ndev = to_net_dev(dev);
164
165 if (dev_isalive(ndev))
166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167 return -EINVAL;
168}
169static DEVICE_ATTR_RO(broadcast);
170
171static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172{
173 if (!netif_running(dev))
174 return -EINVAL;
175 return dev_change_carrier(dev, (bool)new_carrier);
176}
177
178static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179 const char *buf, size_t len)
180{
181 struct net_device *netdev = to_net_dev(dev);
182
183 /* The check is also done in change_carrier; this helps returning early
184 * without hitting the trylock/restart in netdev_store.
185 */
186 if (!netdev->netdev_ops->ndo_change_carrier)
187 return -EOPNOTSUPP;
188
189 return netdev_store(dev, attr, buf, len, change_carrier);
190}
191
192static ssize_t carrier_show(struct device *dev,
193 struct device_attribute *attr, char *buf)
194{
195 struct net_device *netdev = to_net_dev(dev);
196 int ret = -EINVAL;
197
198 if (!rtnl_trylock())
199 return restart_syscall();
200
201 if (netif_running(netdev)) {
202 /* Synchronize carrier state with link watch,
203 * see also rtnl_getlink().
204 */
205 linkwatch_sync_dev(netdev);
206
207 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
208 }
209 rtnl_unlock();
210
211 return ret;
212}
213static DEVICE_ATTR_RW(carrier);
214
215static ssize_t speed_show(struct device *dev,
216 struct device_attribute *attr, char *buf)
217{
218 struct net_device *netdev = to_net_dev(dev);
219 int ret = -EINVAL;
220
221 /* The check is also done in __ethtool_get_link_ksettings; this helps
222 * returning early without hitting the trylock/restart below.
223 */
224 if (!netdev->ethtool_ops->get_link_ksettings)
225 return ret;
226
227 if (!rtnl_trylock())
228 return restart_syscall();
229
230 if (netif_running(netdev) && netif_device_present(netdev)) {
231 struct ethtool_link_ksettings cmd;
232
233 if (!__ethtool_get_link_ksettings(netdev, &cmd))
234 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
235 }
236 rtnl_unlock();
237 return ret;
238}
239static DEVICE_ATTR_RO(speed);
240
241static ssize_t duplex_show(struct device *dev,
242 struct device_attribute *attr, char *buf)
243{
244 struct net_device *netdev = to_net_dev(dev);
245 int ret = -EINVAL;
246
247 /* The check is also done in __ethtool_get_link_ksettings; this helps
248 * returning early without hitting the trylock/restart below.
249 */
250 if (!netdev->ethtool_ops->get_link_ksettings)
251 return ret;
252
253 if (!rtnl_trylock())
254 return restart_syscall();
255
256 if (netif_running(netdev)) {
257 struct ethtool_link_ksettings cmd;
258
259 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
260 const char *duplex;
261
262 switch (cmd.base.duplex) {
263 case DUPLEX_HALF:
264 duplex = "half";
265 break;
266 case DUPLEX_FULL:
267 duplex = "full";
268 break;
269 default:
270 duplex = "unknown";
271 break;
272 }
273 ret = sysfs_emit(buf, "%s\n", duplex);
274 }
275 }
276 rtnl_unlock();
277 return ret;
278}
279static DEVICE_ATTR_RO(duplex);
280
281static ssize_t testing_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_testing(netdev));
288
289 return -EINVAL;
290}
291static DEVICE_ATTR_RO(testing);
292
293static ssize_t dormant_show(struct device *dev,
294 struct device_attribute *attr, char *buf)
295{
296 struct net_device *netdev = to_net_dev(dev);
297
298 if (netif_running(netdev))
299 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
300
301 return -EINVAL;
302}
303static DEVICE_ATTR_RO(dormant);
304
305static const char *const operstates[] = {
306 "unknown",
307 "notpresent", /* currently unused */
308 "down",
309 "lowerlayerdown",
310 "testing",
311 "dormant",
312 "up"
313};
314
315static ssize_t operstate_show(struct device *dev,
316 struct device_attribute *attr, char *buf)
317{
318 const struct net_device *netdev = to_net_dev(dev);
319 unsigned char operstate;
320
321 read_lock(&dev_base_lock);
322 operstate = netdev->operstate;
323 if (!netif_running(netdev))
324 operstate = IF_OPER_DOWN;
325 read_unlock(&dev_base_lock);
326
327 if (operstate >= ARRAY_SIZE(operstates))
328 return -EINVAL; /* should not happen */
329
330 return sysfs_emit(buf, "%s\n", operstates[operstate]);
331}
332static DEVICE_ATTR_RO(operstate);
333
334static ssize_t carrier_changes_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,
341 atomic_read(&netdev->carrier_up_count) +
342 atomic_read(&netdev->carrier_down_count));
343}
344static DEVICE_ATTR_RO(carrier_changes);
345
346static ssize_t carrier_up_count_show(struct device *dev,
347 struct device_attribute *attr,
348 char *buf)
349{
350 struct net_device *netdev = to_net_dev(dev);
351
352 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
353}
354static DEVICE_ATTR_RO(carrier_up_count);
355
356static ssize_t carrier_down_count_show(struct device *dev,
357 struct device_attribute *attr,
358 char *buf)
359{
360 struct net_device *netdev = to_net_dev(dev);
361
362 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
363}
364static DEVICE_ATTR_RO(carrier_down_count);
365
366/* read-write attributes */
367
368static int change_mtu(struct net_device *dev, unsigned long new_mtu)
369{
370 return dev_set_mtu(dev, (int)new_mtu);
371}
372
373static ssize_t mtu_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_mtu);
377}
378NETDEVICE_SHOW_RW(mtu, fmt_dec);
379
380static int change_flags(struct net_device *dev, unsigned long new_flags)
381{
382 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
383}
384
385static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
386 const char *buf, size_t len)
387{
388 return netdev_store(dev, attr, buf, len, change_flags);
389}
390NETDEVICE_SHOW_RW(flags, fmt_hex);
391
392static ssize_t tx_queue_len_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, dev_change_tx_queue_len);
400}
401NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
402
403static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
404{
405 WRITE_ONCE(dev->gro_flush_timeout, val);
406 return 0;
407}
408
409static ssize_t gro_flush_timeout_store(struct device *dev,
410 struct device_attribute *attr,
411 const char *buf, size_t len)
412{
413 if (!capable(CAP_NET_ADMIN))
414 return -EPERM;
415
416 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
417}
418NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
419
420static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
421{
422 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
423 return 0;
424}
425
426static ssize_t napi_defer_hard_irqs_store(struct device *dev,
427 struct device_attribute *attr,
428 const char *buf, size_t len)
429{
430 if (!capable(CAP_NET_ADMIN))
431 return -EPERM;
432
433 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
434}
435NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
436
437static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t len)
439{
440 struct net_device *netdev = to_net_dev(dev);
441 struct net *net = dev_net(netdev);
442 size_t count = len;
443 ssize_t ret = 0;
444
445 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
446 return -EPERM;
447
448 /* ignore trailing newline */
449 if (len > 0 && buf[len - 1] == '\n')
450 --count;
451
452 if (!rtnl_trylock())
453 return restart_syscall();
454
455 if (dev_isalive(netdev)) {
456 ret = dev_set_alias(netdev, buf, count);
457 if (ret < 0)
458 goto err;
459 ret = len;
460 netdev_state_change(netdev);
461 }
462err:
463 rtnl_unlock();
464
465 return ret;
466}
467
468static ssize_t ifalias_show(struct device *dev,
469 struct device_attribute *attr, char *buf)
470{
471 const struct net_device *netdev = to_net_dev(dev);
472 char tmp[IFALIASZ];
473 ssize_t ret = 0;
474
475 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
476 if (ret > 0)
477 ret = sysfs_emit(buf, "%s\n", tmp);
478 return ret;
479}
480static DEVICE_ATTR_RW(ifalias);
481
482static int change_group(struct net_device *dev, unsigned long new_group)
483{
484 dev_set_group(dev, (int)new_group);
485 return 0;
486}
487
488static ssize_t group_store(struct device *dev, struct device_attribute *attr,
489 const char *buf, size_t len)
490{
491 return netdev_store(dev, attr, buf, len, change_group);
492}
493NETDEVICE_SHOW(group, fmt_dec);
494static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
495
496static int change_proto_down(struct net_device *dev, unsigned long proto_down)
497{
498 return dev_change_proto_down(dev, (bool)proto_down);
499}
500
501static ssize_t proto_down_store(struct device *dev,
502 struct device_attribute *attr,
503 const char *buf, size_t len)
504{
505 return netdev_store(dev, attr, buf, len, change_proto_down);
506}
507NETDEVICE_SHOW_RW(proto_down, fmt_dec);
508
509static ssize_t phys_port_id_show(struct device *dev,
510 struct device_attribute *attr, char *buf)
511{
512 struct net_device *netdev = to_net_dev(dev);
513 ssize_t ret = -EINVAL;
514
515 /* The check is also done in dev_get_phys_port_id; this helps returning
516 * early without hitting the trylock/restart below.
517 */
518 if (!netdev->netdev_ops->ndo_get_phys_port_id)
519 return -EOPNOTSUPP;
520
521 if (!rtnl_trylock())
522 return restart_syscall();
523
524 if (dev_isalive(netdev)) {
525 struct netdev_phys_item_id ppid;
526
527 ret = dev_get_phys_port_id(netdev, &ppid);
528 if (!ret)
529 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
530 }
531 rtnl_unlock();
532
533 return ret;
534}
535static DEVICE_ATTR_RO(phys_port_id);
536
537static ssize_t phys_port_name_show(struct device *dev,
538 struct device_attribute *attr, char *buf)
539{
540 struct net_device *netdev = to_net_dev(dev);
541 ssize_t ret = -EINVAL;
542
543 /* The checks are also done in dev_get_phys_port_name; this helps
544 * returning early without hitting the trylock/restart below.
545 */
546 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
547 !netdev->devlink_port)
548 return -EOPNOTSUPP;
549
550 if (!rtnl_trylock())
551 return restart_syscall();
552
553 if (dev_isalive(netdev)) {
554 char name[IFNAMSIZ];
555
556 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
557 if (!ret)
558 ret = sysfs_emit(buf, "%s\n", name);
559 }
560 rtnl_unlock();
561
562 return ret;
563}
564static DEVICE_ATTR_RO(phys_port_name);
565
566static ssize_t phys_switch_id_show(struct device *dev,
567 struct device_attribute *attr, char *buf)
568{
569 struct net_device *netdev = to_net_dev(dev);
570 ssize_t ret = -EINVAL;
571
572 /* The checks are also done in dev_get_phys_port_name; this helps
573 * returning early without hitting the trylock/restart below. This works
574 * because recurse is false when calling dev_get_port_parent_id.
575 */
576 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
577 !netdev->devlink_port)
578 return -EOPNOTSUPP;
579
580 if (!rtnl_trylock())
581 return restart_syscall();
582
583 if (dev_isalive(netdev)) {
584 struct netdev_phys_item_id ppid = { };
585
586 ret = dev_get_port_parent_id(netdev, &ppid, false);
587 if (!ret)
588 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
589 }
590 rtnl_unlock();
591
592 return ret;
593}
594static DEVICE_ATTR_RO(phys_switch_id);
595
596static ssize_t threaded_show(struct device *dev,
597 struct device_attribute *attr, char *buf)
598{
599 struct net_device *netdev = to_net_dev(dev);
600 ssize_t ret = -EINVAL;
601
602 if (!rtnl_trylock())
603 return restart_syscall();
604
605 if (dev_isalive(netdev))
606 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
607
608 rtnl_unlock();
609 return ret;
610}
611
612static int modify_napi_threaded(struct net_device *dev, unsigned long val)
613{
614 int ret;
615
616 if (list_empty(&dev->napi_list))
617 return -EOPNOTSUPP;
618
619 if (val != 0 && val != 1)
620 return -EOPNOTSUPP;
621
622 ret = dev_set_threaded(dev, val);
623
624 return ret;
625}
626
627static ssize_t threaded_store(struct device *dev,
628 struct device_attribute *attr,
629 const char *buf, size_t len)
630{
631 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
632}
633static DEVICE_ATTR_RW(threaded);
634
635static struct attribute *net_class_attrs[] __ro_after_init = {
636 &dev_attr_netdev_group.attr,
637 &dev_attr_type.attr,
638 &dev_attr_dev_id.attr,
639 &dev_attr_dev_port.attr,
640 &dev_attr_iflink.attr,
641 &dev_attr_ifindex.attr,
642 &dev_attr_name_assign_type.attr,
643 &dev_attr_addr_assign_type.attr,
644 &dev_attr_addr_len.attr,
645 &dev_attr_link_mode.attr,
646 &dev_attr_address.attr,
647 &dev_attr_broadcast.attr,
648 &dev_attr_speed.attr,
649 &dev_attr_duplex.attr,
650 &dev_attr_dormant.attr,
651 &dev_attr_testing.attr,
652 &dev_attr_operstate.attr,
653 &dev_attr_carrier_changes.attr,
654 &dev_attr_ifalias.attr,
655 &dev_attr_carrier.attr,
656 &dev_attr_mtu.attr,
657 &dev_attr_flags.attr,
658 &dev_attr_tx_queue_len.attr,
659 &dev_attr_gro_flush_timeout.attr,
660 &dev_attr_napi_defer_hard_irqs.attr,
661 &dev_attr_phys_port_id.attr,
662 &dev_attr_phys_port_name.attr,
663 &dev_attr_phys_switch_id.attr,
664 &dev_attr_proto_down.attr,
665 &dev_attr_carrier_up_count.attr,
666 &dev_attr_carrier_down_count.attr,
667 &dev_attr_threaded.attr,
668 NULL,
669};
670ATTRIBUTE_GROUPS(net_class);
671
672/* Show a given an attribute in the statistics group */
673static ssize_t netstat_show(const struct device *d,
674 struct device_attribute *attr, char *buf,
675 unsigned long offset)
676{
677 struct net_device *dev = to_net_dev(d);
678 ssize_t ret = -EINVAL;
679
680 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
681 offset % sizeof(u64) != 0);
682
683 read_lock(&dev_base_lock);
684 if (dev_isalive(dev)) {
685 struct rtnl_link_stats64 temp;
686 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
687
688 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
689 }
690 read_unlock(&dev_base_lock);
691 return ret;
692}
693
694/* generate a read-only statistics attribute */
695#define NETSTAT_ENTRY(name) \
696static ssize_t name##_show(struct device *d, \
697 struct device_attribute *attr, char *buf) \
698{ \
699 return netstat_show(d, attr, buf, \
700 offsetof(struct rtnl_link_stats64, name)); \
701} \
702static DEVICE_ATTR_RO(name)
703
704NETSTAT_ENTRY(rx_packets);
705NETSTAT_ENTRY(tx_packets);
706NETSTAT_ENTRY(rx_bytes);
707NETSTAT_ENTRY(tx_bytes);
708NETSTAT_ENTRY(rx_errors);
709NETSTAT_ENTRY(tx_errors);
710NETSTAT_ENTRY(rx_dropped);
711NETSTAT_ENTRY(tx_dropped);
712NETSTAT_ENTRY(multicast);
713NETSTAT_ENTRY(collisions);
714NETSTAT_ENTRY(rx_length_errors);
715NETSTAT_ENTRY(rx_over_errors);
716NETSTAT_ENTRY(rx_crc_errors);
717NETSTAT_ENTRY(rx_frame_errors);
718NETSTAT_ENTRY(rx_fifo_errors);
719NETSTAT_ENTRY(rx_missed_errors);
720NETSTAT_ENTRY(tx_aborted_errors);
721NETSTAT_ENTRY(tx_carrier_errors);
722NETSTAT_ENTRY(tx_fifo_errors);
723NETSTAT_ENTRY(tx_heartbeat_errors);
724NETSTAT_ENTRY(tx_window_errors);
725NETSTAT_ENTRY(rx_compressed);
726NETSTAT_ENTRY(tx_compressed);
727NETSTAT_ENTRY(rx_nohandler);
728
729static struct attribute *netstat_attrs[] __ro_after_init = {
730 &dev_attr_rx_packets.attr,
731 &dev_attr_tx_packets.attr,
732 &dev_attr_rx_bytes.attr,
733 &dev_attr_tx_bytes.attr,
734 &dev_attr_rx_errors.attr,
735 &dev_attr_tx_errors.attr,
736 &dev_attr_rx_dropped.attr,
737 &dev_attr_tx_dropped.attr,
738 &dev_attr_multicast.attr,
739 &dev_attr_collisions.attr,
740 &dev_attr_rx_length_errors.attr,
741 &dev_attr_rx_over_errors.attr,
742 &dev_attr_rx_crc_errors.attr,
743 &dev_attr_rx_frame_errors.attr,
744 &dev_attr_rx_fifo_errors.attr,
745 &dev_attr_rx_missed_errors.attr,
746 &dev_attr_tx_aborted_errors.attr,
747 &dev_attr_tx_carrier_errors.attr,
748 &dev_attr_tx_fifo_errors.attr,
749 &dev_attr_tx_heartbeat_errors.attr,
750 &dev_attr_tx_window_errors.attr,
751 &dev_attr_rx_compressed.attr,
752 &dev_attr_tx_compressed.attr,
753 &dev_attr_rx_nohandler.attr,
754 NULL
755};
756
757static const struct attribute_group netstat_group = {
758 .name = "statistics",
759 .attrs = netstat_attrs,
760};
761
762static struct attribute *wireless_attrs[] = {
763 NULL
764};
765
766static const struct attribute_group wireless_group = {
767 .name = "wireless",
768 .attrs = wireless_attrs,
769};
770
771static bool wireless_group_needed(struct net_device *ndev)
772{
773#if IS_ENABLED(CONFIG_CFG80211)
774 if (ndev->ieee80211_ptr)
775 return true;
776#endif
777#if IS_ENABLED(CONFIG_WIRELESS_EXT)
778 if (ndev->wireless_handlers)
779 return true;
780#endif
781 return false;
782}
783
784#else /* CONFIG_SYSFS */
785#define net_class_groups NULL
786#endif /* CONFIG_SYSFS */
787
788#ifdef CONFIG_SYSFS
789#define to_rx_queue_attr(_attr) \
790 container_of(_attr, struct rx_queue_attribute, attr)
791
792#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
793
794static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
795 char *buf)
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->show)
801 return -EIO;
802
803 return attribute->show(queue, buf);
804}
805
806static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
807 const char *buf, size_t count)
808{
809 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
810 struct netdev_rx_queue *queue = to_rx_queue(kobj);
811
812 if (!attribute->store)
813 return -EIO;
814
815 return attribute->store(queue, buf, count);
816}
817
818static const struct sysfs_ops rx_queue_sysfs_ops = {
819 .show = rx_queue_attr_show,
820 .store = rx_queue_attr_store,
821};
822
823#ifdef CONFIG_RPS
824static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
825{
826 struct rps_map *map;
827 cpumask_var_t mask;
828 int i, len;
829
830 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
831 return -ENOMEM;
832
833 rcu_read_lock();
834 map = rcu_dereference(queue->rps_map);
835 if (map)
836 for (i = 0; i < map->len; i++)
837 cpumask_set_cpu(map->cpus[i], mask);
838
839 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
840 rcu_read_unlock();
841 free_cpumask_var(mask);
842
843 return len < PAGE_SIZE ? len : -EINVAL;
844}
845
846static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
847 cpumask_var_t mask)
848{
849 static DEFINE_MUTEX(rps_map_mutex);
850 struct rps_map *old_map, *map;
851 int cpu, i;
852
853 map = kzalloc(max_t(unsigned int,
854 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
855 GFP_KERNEL);
856 if (!map)
857 return -ENOMEM;
858
859 i = 0;
860 for_each_cpu_and(cpu, mask, cpu_online_mask)
861 map->cpus[i++] = cpu;
862
863 if (i) {
864 map->len = i;
865 } else {
866 kfree(map);
867 map = NULL;
868 }
869
870 mutex_lock(&rps_map_mutex);
871 old_map = rcu_dereference_protected(queue->rps_map,
872 mutex_is_locked(&rps_map_mutex));
873 rcu_assign_pointer(queue->rps_map, map);
874
875 if (map)
876 static_branch_inc(&rps_needed);
877 if (old_map)
878 static_branch_dec(&rps_needed);
879
880 mutex_unlock(&rps_map_mutex);
881
882 if (old_map)
883 kfree_rcu(old_map, rcu);
884 return 0;
885}
886
887int rps_cpumask_housekeeping(struct cpumask *mask)
888{
889 if (!cpumask_empty(mask)) {
890 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
891 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
892 if (cpumask_empty(mask))
893 return -EINVAL;
894 }
895 return 0;
896}
897
898static ssize_t store_rps_map(struct netdev_rx_queue *queue,
899 const char *buf, size_t len)
900{
901 cpumask_var_t mask;
902 int err;
903
904 if (!capable(CAP_NET_ADMIN))
905 return -EPERM;
906
907 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
908 return -ENOMEM;
909
910 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
911 if (err)
912 goto out;
913
914 err = rps_cpumask_housekeeping(mask);
915 if (err)
916 goto out;
917
918 err = netdev_rx_queue_set_rps_mask(queue, mask);
919
920out:
921 free_cpumask_var(mask);
922 return err ? : len;
923}
924
925static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
926 char *buf)
927{
928 struct rps_dev_flow_table *flow_table;
929 unsigned long val = 0;
930
931 rcu_read_lock();
932 flow_table = rcu_dereference(queue->rps_flow_table);
933 if (flow_table)
934 val = (unsigned long)flow_table->mask + 1;
935 rcu_read_unlock();
936
937 return sysfs_emit(buf, "%lu\n", val);
938}
939
940static void rps_dev_flow_table_release(struct rcu_head *rcu)
941{
942 struct rps_dev_flow_table *table = container_of(rcu,
943 struct rps_dev_flow_table, rcu);
944 vfree(table);
945}
946
947static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
948 const char *buf, size_t len)
949{
950 unsigned long mask, count;
951 struct rps_dev_flow_table *table, *old_table;
952 static DEFINE_SPINLOCK(rps_dev_flow_lock);
953 int rc;
954
955 if (!capable(CAP_NET_ADMIN))
956 return -EPERM;
957
958 rc = kstrtoul(buf, 0, &count);
959 if (rc < 0)
960 return rc;
961
962 if (count) {
963 mask = count - 1;
964 /* mask = roundup_pow_of_two(count) - 1;
965 * without overflows...
966 */
967 while ((mask | (mask >> 1)) != mask)
968 mask |= (mask >> 1);
969 /* On 64 bit arches, must check mask fits in table->mask (u32),
970 * and on 32bit arches, must check
971 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
972 */
973#if BITS_PER_LONG > 32
974 if (mask > (unsigned long)(u32)mask)
975 return -EINVAL;
976#else
977 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
978 / sizeof(struct rps_dev_flow)) {
979 /* Enforce a limit to prevent overflow */
980 return -EINVAL;
981 }
982#endif
983 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
984 if (!table)
985 return -ENOMEM;
986
987 table->mask = mask;
988 for (count = 0; count <= mask; count++)
989 table->flows[count].cpu = RPS_NO_CPU;
990 } else {
991 table = NULL;
992 }
993
994 spin_lock(&rps_dev_flow_lock);
995 old_table = rcu_dereference_protected(queue->rps_flow_table,
996 lockdep_is_held(&rps_dev_flow_lock));
997 rcu_assign_pointer(queue->rps_flow_table, table);
998 spin_unlock(&rps_dev_flow_lock);
999
1000 if (old_table)
1001 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1002
1003 return len;
1004}
1005
1006static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1007 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1008
1009static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1010 = __ATTR(rps_flow_cnt, 0644,
1011 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1012#endif /* CONFIG_RPS */
1013
1014static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1015#ifdef CONFIG_RPS
1016 &rps_cpus_attribute.attr,
1017 &rps_dev_flow_table_cnt_attribute.attr,
1018#endif
1019 NULL
1020};
1021ATTRIBUTE_GROUPS(rx_queue_default);
1022
1023static void rx_queue_release(struct kobject *kobj)
1024{
1025 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1026#ifdef CONFIG_RPS
1027 struct rps_map *map;
1028 struct rps_dev_flow_table *flow_table;
1029
1030 map = rcu_dereference_protected(queue->rps_map, 1);
1031 if (map) {
1032 RCU_INIT_POINTER(queue->rps_map, NULL);
1033 kfree_rcu(map, rcu);
1034 }
1035
1036 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1037 if (flow_table) {
1038 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1039 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1040 }
1041#endif
1042
1043 memset(kobj, 0, sizeof(*kobj));
1044 netdev_put(queue->dev, &queue->dev_tracker);
1045}
1046
1047static const void *rx_queue_namespace(const struct kobject *kobj)
1048{
1049 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1050 struct device *dev = &queue->dev->dev;
1051 const void *ns = NULL;
1052
1053 if (dev->class && dev->class->ns_type)
1054 ns = dev->class->namespace(dev);
1055
1056 return ns;
1057}
1058
1059static void rx_queue_get_ownership(const struct kobject *kobj,
1060 kuid_t *uid, kgid_t *gid)
1061{
1062 const struct net *net = rx_queue_namespace(kobj);
1063
1064 net_ns_get_ownership(net, uid, gid);
1065}
1066
1067static const struct kobj_type rx_queue_ktype = {
1068 .sysfs_ops = &rx_queue_sysfs_ops,
1069 .release = rx_queue_release,
1070 .default_groups = rx_queue_default_groups,
1071 .namespace = rx_queue_namespace,
1072 .get_ownership = rx_queue_get_ownership,
1073};
1074
1075static int rx_queue_default_mask(struct net_device *dev,
1076 struct netdev_rx_queue *queue)
1077{
1078#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1079 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1080
1081 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1082 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1083#endif
1084 return 0;
1085}
1086
1087static int rx_queue_add_kobject(struct net_device *dev, int index)
1088{
1089 struct netdev_rx_queue *queue = dev->_rx + index;
1090 struct kobject *kobj = &queue->kobj;
1091 int error = 0;
1092
1093 /* Kobject_put later will trigger rx_queue_release call which
1094 * decreases dev refcount: Take that reference here
1095 */
1096 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1097
1098 kobj->kset = dev->queues_kset;
1099 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1100 "rx-%u", index);
1101 if (error)
1102 goto err;
1103
1104 if (dev->sysfs_rx_queue_group) {
1105 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1106 if (error)
1107 goto err;
1108 }
1109
1110 error = rx_queue_default_mask(dev, queue);
1111 if (error)
1112 goto err;
1113
1114 kobject_uevent(kobj, KOBJ_ADD);
1115
1116 return error;
1117
1118err:
1119 kobject_put(kobj);
1120 return error;
1121}
1122
1123static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1124 kgid_t kgid)
1125{
1126 struct netdev_rx_queue *queue = dev->_rx + index;
1127 struct kobject *kobj = &queue->kobj;
1128 int error;
1129
1130 error = sysfs_change_owner(kobj, kuid, kgid);
1131 if (error)
1132 return error;
1133
1134 if (dev->sysfs_rx_queue_group)
1135 error = sysfs_group_change_owner(
1136 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1137
1138 return error;
1139}
1140#endif /* CONFIG_SYSFS */
1141
1142int
1143net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1144{
1145#ifdef CONFIG_SYSFS
1146 int i;
1147 int error = 0;
1148
1149#ifndef CONFIG_RPS
1150 if (!dev->sysfs_rx_queue_group)
1151 return 0;
1152#endif
1153 for (i = old_num; i < new_num; i++) {
1154 error = rx_queue_add_kobject(dev, i);
1155 if (error) {
1156 new_num = old_num;
1157 break;
1158 }
1159 }
1160
1161 while (--i >= new_num) {
1162 struct kobject *kobj = &dev->_rx[i].kobj;
1163
1164 if (!refcount_read(&dev_net(dev)->ns.count))
1165 kobj->uevent_suppress = 1;
1166 if (dev->sysfs_rx_queue_group)
1167 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1168 kobject_put(kobj);
1169 }
1170
1171 return error;
1172#else
1173 return 0;
1174#endif
1175}
1176
1177static int net_rx_queue_change_owner(struct net_device *dev, int num,
1178 kuid_t kuid, kgid_t kgid)
1179{
1180#ifdef CONFIG_SYSFS
1181 int error = 0;
1182 int i;
1183
1184#ifndef CONFIG_RPS
1185 if (!dev->sysfs_rx_queue_group)
1186 return 0;
1187#endif
1188 for (i = 0; i < num; i++) {
1189 error = rx_queue_change_owner(dev, i, kuid, kgid);
1190 if (error)
1191 break;
1192 }
1193
1194 return error;
1195#else
1196 return 0;
1197#endif
1198}
1199
1200#ifdef CONFIG_SYSFS
1201/*
1202 * netdev_queue sysfs structures and functions.
1203 */
1204struct netdev_queue_attribute {
1205 struct attribute attr;
1206 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1207 ssize_t (*store)(struct netdev_queue *queue,
1208 const char *buf, size_t len);
1209};
1210#define to_netdev_queue_attr(_attr) \
1211 container_of(_attr, struct netdev_queue_attribute, attr)
1212
1213#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1214
1215static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1216 struct attribute *attr, char *buf)
1217{
1218 const struct netdev_queue_attribute *attribute
1219 = to_netdev_queue_attr(attr);
1220 struct netdev_queue *queue = to_netdev_queue(kobj);
1221
1222 if (!attribute->show)
1223 return -EIO;
1224
1225 return attribute->show(queue, buf);
1226}
1227
1228static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1229 struct attribute *attr,
1230 const char *buf, size_t count)
1231{
1232 const struct netdev_queue_attribute *attribute
1233 = to_netdev_queue_attr(attr);
1234 struct netdev_queue *queue = to_netdev_queue(kobj);
1235
1236 if (!attribute->store)
1237 return -EIO;
1238
1239 return attribute->store(queue, buf, count);
1240}
1241
1242static const struct sysfs_ops netdev_queue_sysfs_ops = {
1243 .show = netdev_queue_attr_show,
1244 .store = netdev_queue_attr_store,
1245};
1246
1247static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1248{
1249 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1250
1251 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1252}
1253
1254static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1255{
1256 struct net_device *dev = queue->dev;
1257 unsigned int i;
1258
1259 i = queue - dev->_tx;
1260 BUG_ON(i >= dev->num_tx_queues);
1261
1262 return i;
1263}
1264
1265static ssize_t traffic_class_show(struct netdev_queue *queue,
1266 char *buf)
1267{
1268 struct net_device *dev = queue->dev;
1269 int num_tc, tc;
1270 int index;
1271
1272 if (!netif_is_multiqueue(dev))
1273 return -ENOENT;
1274
1275 if (!rtnl_trylock())
1276 return restart_syscall();
1277
1278 index = get_netdev_queue_index(queue);
1279
1280 /* If queue belongs to subordinate dev use its TC mapping */
1281 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1282
1283 num_tc = dev->num_tc;
1284 tc = netdev_txq_to_tc(dev, index);
1285
1286 rtnl_unlock();
1287
1288 if (tc < 0)
1289 return -EINVAL;
1290
1291 /* We can report the traffic class one of two ways:
1292 * Subordinate device traffic classes are reported with the traffic
1293 * class first, and then the subordinate class so for example TC0 on
1294 * subordinate device 2 will be reported as "0-2". If the queue
1295 * belongs to the root device it will be reported with just the
1296 * traffic class, so just "0" for TC 0 for example.
1297 */
1298 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1299 sysfs_emit(buf, "%d\n", tc);
1300}
1301
1302#ifdef CONFIG_XPS
1303static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1304 char *buf)
1305{
1306 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1307}
1308
1309static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1310 const char *buf, size_t len)
1311{
1312 struct net_device *dev = queue->dev;
1313 int err, index = get_netdev_queue_index(queue);
1314 u32 rate = 0;
1315
1316 if (!capable(CAP_NET_ADMIN))
1317 return -EPERM;
1318
1319 /* The check is also done later; this helps returning early without
1320 * hitting the trylock/restart below.
1321 */
1322 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1323 return -EOPNOTSUPP;
1324
1325 err = kstrtou32(buf, 10, &rate);
1326 if (err < 0)
1327 return err;
1328
1329 if (!rtnl_trylock())
1330 return restart_syscall();
1331
1332 err = -EOPNOTSUPP;
1333 if (dev->netdev_ops->ndo_set_tx_maxrate)
1334 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1335
1336 rtnl_unlock();
1337 if (!err) {
1338 queue->tx_maxrate = rate;
1339 return len;
1340 }
1341 return err;
1342}
1343
1344static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1345 = __ATTR_RW(tx_maxrate);
1346#endif
1347
1348static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1349 = __ATTR_RO(tx_timeout);
1350
1351static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1352 = __ATTR_RO(traffic_class);
1353
1354#ifdef CONFIG_BQL
1355/*
1356 * Byte queue limits sysfs structures and functions.
1357 */
1358static ssize_t bql_show(char *buf, unsigned int value)
1359{
1360 return sysfs_emit(buf, "%u\n", value);
1361}
1362
1363static ssize_t bql_set(const char *buf, const size_t count,
1364 unsigned int *pvalue)
1365{
1366 unsigned int value;
1367 int err;
1368
1369 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1370 value = DQL_MAX_LIMIT;
1371 } else {
1372 err = kstrtouint(buf, 10, &value);
1373 if (err < 0)
1374 return err;
1375 if (value > DQL_MAX_LIMIT)
1376 return -EINVAL;
1377 }
1378
1379 *pvalue = value;
1380
1381 return count;
1382}
1383
1384static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1385 char *buf)
1386{
1387 struct dql *dql = &queue->dql;
1388
1389 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1390}
1391
1392static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1393 const char *buf, size_t len)
1394{
1395 struct dql *dql = &queue->dql;
1396 unsigned int value;
1397 int err;
1398
1399 err = kstrtouint(buf, 10, &value);
1400 if (err < 0)
1401 return err;
1402
1403 dql->slack_hold_time = msecs_to_jiffies(value);
1404
1405 return len;
1406}
1407
1408static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1409 = __ATTR(hold_time, 0644,
1410 bql_show_hold_time, bql_set_hold_time);
1411
1412static ssize_t bql_show_inflight(struct netdev_queue *queue,
1413 char *buf)
1414{
1415 struct dql *dql = &queue->dql;
1416
1417 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1418}
1419
1420static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1421 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1422
1423#define BQL_ATTR(NAME, FIELD) \
1424static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1425 char *buf) \
1426{ \
1427 return bql_show(buf, queue->dql.FIELD); \
1428} \
1429 \
1430static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1431 const char *buf, size_t len) \
1432{ \
1433 return bql_set(buf, len, &queue->dql.FIELD); \
1434} \
1435 \
1436static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1437 = __ATTR(NAME, 0644, \
1438 bql_show_ ## NAME, bql_set_ ## NAME)
1439
1440BQL_ATTR(limit, limit);
1441BQL_ATTR(limit_max, max_limit);
1442BQL_ATTR(limit_min, min_limit);
1443
1444static struct attribute *dql_attrs[] __ro_after_init = {
1445 &bql_limit_attribute.attr,
1446 &bql_limit_max_attribute.attr,
1447 &bql_limit_min_attribute.attr,
1448 &bql_hold_time_attribute.attr,
1449 &bql_inflight_attribute.attr,
1450 NULL
1451};
1452
1453static const struct attribute_group dql_group = {
1454 .name = "byte_queue_limits",
1455 .attrs = dql_attrs,
1456};
1457#endif /* CONFIG_BQL */
1458
1459#ifdef CONFIG_XPS
1460static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1461 int tc, char *buf, enum xps_map_type type)
1462{
1463 struct xps_dev_maps *dev_maps;
1464 unsigned long *mask;
1465 unsigned int nr_ids;
1466 int j, len;
1467
1468 rcu_read_lock();
1469 dev_maps = rcu_dereference(dev->xps_maps[type]);
1470
1471 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1472 * when dev_maps hasn't been allocated yet, to be backward compatible.
1473 */
1474 nr_ids = dev_maps ? dev_maps->nr_ids :
1475 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1476
1477 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1478 if (!mask) {
1479 rcu_read_unlock();
1480 return -ENOMEM;
1481 }
1482
1483 if (!dev_maps || tc >= dev_maps->num_tc)
1484 goto out_no_maps;
1485
1486 for (j = 0; j < nr_ids; j++) {
1487 int i, tci = j * dev_maps->num_tc + tc;
1488 struct xps_map *map;
1489
1490 map = rcu_dereference(dev_maps->attr_map[tci]);
1491 if (!map)
1492 continue;
1493
1494 for (i = map->len; i--;) {
1495 if (map->queues[i] == index) {
1496 __set_bit(j, mask);
1497 break;
1498 }
1499 }
1500 }
1501out_no_maps:
1502 rcu_read_unlock();
1503
1504 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1505 bitmap_free(mask);
1506
1507 return len < PAGE_SIZE ? len : -EINVAL;
1508}
1509
1510static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1511{
1512 struct net_device *dev = queue->dev;
1513 unsigned int index;
1514 int len, tc;
1515
1516 if (!netif_is_multiqueue(dev))
1517 return -ENOENT;
1518
1519 index = get_netdev_queue_index(queue);
1520
1521 if (!rtnl_trylock())
1522 return restart_syscall();
1523
1524 /* If queue belongs to subordinate dev use its map */
1525 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1526
1527 tc = netdev_txq_to_tc(dev, index);
1528 if (tc < 0) {
1529 rtnl_unlock();
1530 return -EINVAL;
1531 }
1532
1533 /* Make sure the subordinate device can't be freed */
1534 get_device(&dev->dev);
1535 rtnl_unlock();
1536
1537 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1538
1539 put_device(&dev->dev);
1540 return len;
1541}
1542
1543static ssize_t xps_cpus_store(struct netdev_queue *queue,
1544 const char *buf, size_t len)
1545{
1546 struct net_device *dev = queue->dev;
1547 unsigned int index;
1548 cpumask_var_t mask;
1549 int err;
1550
1551 if (!netif_is_multiqueue(dev))
1552 return -ENOENT;
1553
1554 if (!capable(CAP_NET_ADMIN))
1555 return -EPERM;
1556
1557 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1558 return -ENOMEM;
1559
1560 index = get_netdev_queue_index(queue);
1561
1562 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1563 if (err) {
1564 free_cpumask_var(mask);
1565 return err;
1566 }
1567
1568 if (!rtnl_trylock()) {
1569 free_cpumask_var(mask);
1570 return restart_syscall();
1571 }
1572
1573 err = netif_set_xps_queue(dev, mask, index);
1574 rtnl_unlock();
1575
1576 free_cpumask_var(mask);
1577
1578 return err ? : len;
1579}
1580
1581static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1582 = __ATTR_RW(xps_cpus);
1583
1584static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1585{
1586 struct net_device *dev = queue->dev;
1587 unsigned int index;
1588 int tc;
1589
1590 index = get_netdev_queue_index(queue);
1591
1592 if (!rtnl_trylock())
1593 return restart_syscall();
1594
1595 tc = netdev_txq_to_tc(dev, index);
1596 rtnl_unlock();
1597 if (tc < 0)
1598 return -EINVAL;
1599
1600 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1601}
1602
1603static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1604 size_t len)
1605{
1606 struct net_device *dev = queue->dev;
1607 struct net *net = dev_net(dev);
1608 unsigned long *mask;
1609 unsigned int index;
1610 int err;
1611
1612 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1613 return -EPERM;
1614
1615 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1616 if (!mask)
1617 return -ENOMEM;
1618
1619 index = get_netdev_queue_index(queue);
1620
1621 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1622 if (err) {
1623 bitmap_free(mask);
1624 return err;
1625 }
1626
1627 if (!rtnl_trylock()) {
1628 bitmap_free(mask);
1629 return restart_syscall();
1630 }
1631
1632 cpus_read_lock();
1633 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1634 cpus_read_unlock();
1635
1636 rtnl_unlock();
1637
1638 bitmap_free(mask);
1639 return err ? : len;
1640}
1641
1642static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1643 = __ATTR_RW(xps_rxqs);
1644#endif /* CONFIG_XPS */
1645
1646static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1647 &queue_trans_timeout.attr,
1648 &queue_traffic_class.attr,
1649#ifdef CONFIG_XPS
1650 &xps_cpus_attribute.attr,
1651 &xps_rxqs_attribute.attr,
1652 &queue_tx_maxrate.attr,
1653#endif
1654 NULL
1655};
1656ATTRIBUTE_GROUPS(netdev_queue_default);
1657
1658static void netdev_queue_release(struct kobject *kobj)
1659{
1660 struct netdev_queue *queue = to_netdev_queue(kobj);
1661
1662 memset(kobj, 0, sizeof(*kobj));
1663 netdev_put(queue->dev, &queue->dev_tracker);
1664}
1665
1666static const void *netdev_queue_namespace(const struct kobject *kobj)
1667{
1668 struct netdev_queue *queue = to_netdev_queue(kobj);
1669 struct device *dev = &queue->dev->dev;
1670 const void *ns = NULL;
1671
1672 if (dev->class && dev->class->ns_type)
1673 ns = dev->class->namespace(dev);
1674
1675 return ns;
1676}
1677
1678static void netdev_queue_get_ownership(const struct kobject *kobj,
1679 kuid_t *uid, kgid_t *gid)
1680{
1681 const struct net *net = netdev_queue_namespace(kobj);
1682
1683 net_ns_get_ownership(net, uid, gid);
1684}
1685
1686static const struct kobj_type netdev_queue_ktype = {
1687 .sysfs_ops = &netdev_queue_sysfs_ops,
1688 .release = netdev_queue_release,
1689 .default_groups = netdev_queue_default_groups,
1690 .namespace = netdev_queue_namespace,
1691 .get_ownership = netdev_queue_get_ownership,
1692};
1693
1694static int netdev_queue_add_kobject(struct net_device *dev, int index)
1695{
1696 struct netdev_queue *queue = dev->_tx + index;
1697 struct kobject *kobj = &queue->kobj;
1698 int error = 0;
1699
1700 /* Kobject_put later will trigger netdev_queue_release call
1701 * which decreases dev refcount: Take that reference here
1702 */
1703 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1704
1705 kobj->kset = dev->queues_kset;
1706 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1707 "tx-%u", index);
1708 if (error)
1709 goto err;
1710
1711#ifdef CONFIG_BQL
1712 error = sysfs_create_group(kobj, &dql_group);
1713 if (error)
1714 goto err;
1715#endif
1716
1717 kobject_uevent(kobj, KOBJ_ADD);
1718 return 0;
1719
1720err:
1721 kobject_put(kobj);
1722 return error;
1723}
1724
1725static int tx_queue_change_owner(struct net_device *ndev, int index,
1726 kuid_t kuid, kgid_t kgid)
1727{
1728 struct netdev_queue *queue = ndev->_tx + index;
1729 struct kobject *kobj = &queue->kobj;
1730 int error;
1731
1732 error = sysfs_change_owner(kobj, kuid, kgid);
1733 if (error)
1734 return error;
1735
1736#ifdef CONFIG_BQL
1737 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1738#endif
1739 return error;
1740}
1741#endif /* CONFIG_SYSFS */
1742
1743int
1744netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1745{
1746#ifdef CONFIG_SYSFS
1747 int i;
1748 int error = 0;
1749
1750 /* Tx queue kobjects are allowed to be updated when a device is being
1751 * unregistered, but solely to remove queues from qdiscs. Any path
1752 * adding queues should be fixed.
1753 */
1754 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1755 "New queues can't be registered after device unregistration.");
1756
1757 for (i = old_num; i < new_num; i++) {
1758 error = netdev_queue_add_kobject(dev, i);
1759 if (error) {
1760 new_num = old_num;
1761 break;
1762 }
1763 }
1764
1765 while (--i >= new_num) {
1766 struct netdev_queue *queue = dev->_tx + i;
1767
1768 if (!refcount_read(&dev_net(dev)->ns.count))
1769 queue->kobj.uevent_suppress = 1;
1770#ifdef CONFIG_BQL
1771 sysfs_remove_group(&queue->kobj, &dql_group);
1772#endif
1773 kobject_put(&queue->kobj);
1774 }
1775
1776 return error;
1777#else
1778 return 0;
1779#endif /* CONFIG_SYSFS */
1780}
1781
1782static int net_tx_queue_change_owner(struct net_device *dev, int num,
1783 kuid_t kuid, kgid_t kgid)
1784{
1785#ifdef CONFIG_SYSFS
1786 int error = 0;
1787 int i;
1788
1789 for (i = 0; i < num; i++) {
1790 error = tx_queue_change_owner(dev, i, kuid, kgid);
1791 if (error)
1792 break;
1793 }
1794
1795 return error;
1796#else
1797 return 0;
1798#endif /* CONFIG_SYSFS */
1799}
1800
1801static int register_queue_kobjects(struct net_device *dev)
1802{
1803 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1804
1805#ifdef CONFIG_SYSFS
1806 dev->queues_kset = kset_create_and_add("queues",
1807 NULL, &dev->dev.kobj);
1808 if (!dev->queues_kset)
1809 return -ENOMEM;
1810 real_rx = dev->real_num_rx_queues;
1811#endif
1812 real_tx = dev->real_num_tx_queues;
1813
1814 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1815 if (error)
1816 goto error;
1817 rxq = real_rx;
1818
1819 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1820 if (error)
1821 goto error;
1822 txq = real_tx;
1823
1824 return 0;
1825
1826error:
1827 netdev_queue_update_kobjects(dev, txq, 0);
1828 net_rx_queue_update_kobjects(dev, rxq, 0);
1829#ifdef CONFIG_SYSFS
1830 kset_unregister(dev->queues_kset);
1831#endif
1832 return error;
1833}
1834
1835static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1836{
1837 int error = 0, real_rx = 0, real_tx = 0;
1838
1839#ifdef CONFIG_SYSFS
1840 if (ndev->queues_kset) {
1841 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1842 if (error)
1843 return error;
1844 }
1845 real_rx = ndev->real_num_rx_queues;
1846#endif
1847 real_tx = ndev->real_num_tx_queues;
1848
1849 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1850 if (error)
1851 return error;
1852
1853 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1854 if (error)
1855 return error;
1856
1857 return 0;
1858}
1859
1860static void remove_queue_kobjects(struct net_device *dev)
1861{
1862 int real_rx = 0, real_tx = 0;
1863
1864#ifdef CONFIG_SYSFS
1865 real_rx = dev->real_num_rx_queues;
1866#endif
1867 real_tx = dev->real_num_tx_queues;
1868
1869 net_rx_queue_update_kobjects(dev, real_rx, 0);
1870 netdev_queue_update_kobjects(dev, real_tx, 0);
1871
1872 dev->real_num_rx_queues = 0;
1873 dev->real_num_tx_queues = 0;
1874#ifdef CONFIG_SYSFS
1875 kset_unregister(dev->queues_kset);
1876#endif
1877}
1878
1879static bool net_current_may_mount(void)
1880{
1881 struct net *net = current->nsproxy->net_ns;
1882
1883 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1884}
1885
1886static void *net_grab_current_ns(void)
1887{
1888 struct net *ns = current->nsproxy->net_ns;
1889#ifdef CONFIG_NET_NS
1890 if (ns)
1891 refcount_inc(&ns->passive);
1892#endif
1893 return ns;
1894}
1895
1896static const void *net_initial_ns(void)
1897{
1898 return &init_net;
1899}
1900
1901static const void *net_netlink_ns(struct sock *sk)
1902{
1903 return sock_net(sk);
1904}
1905
1906const struct kobj_ns_type_operations net_ns_type_operations = {
1907 .type = KOBJ_NS_TYPE_NET,
1908 .current_may_mount = net_current_may_mount,
1909 .grab_current_ns = net_grab_current_ns,
1910 .netlink_ns = net_netlink_ns,
1911 .initial_ns = net_initial_ns,
1912 .drop_ns = net_drop_ns,
1913};
1914EXPORT_SYMBOL_GPL(net_ns_type_operations);
1915
1916static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1917{
1918 const struct net_device *dev = to_net_dev(d);
1919 int retval;
1920
1921 /* pass interface to uevent. */
1922 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1923 if (retval)
1924 goto exit;
1925
1926 /* pass ifindex to uevent.
1927 * ifindex is useful as it won't change (interface name may change)
1928 * and is what RtNetlink uses natively.
1929 */
1930 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1931
1932exit:
1933 return retval;
1934}
1935
1936/*
1937 * netdev_release -- destroy and free a dead device.
1938 * Called when last reference to device kobject is gone.
1939 */
1940static void netdev_release(struct device *d)
1941{
1942 struct net_device *dev = to_net_dev(d);
1943
1944 BUG_ON(dev->reg_state != NETREG_RELEASED);
1945
1946 /* no need to wait for rcu grace period:
1947 * device is dead and about to be freed.
1948 */
1949 kfree(rcu_access_pointer(dev->ifalias));
1950 netdev_freemem(dev);
1951}
1952
1953static const void *net_namespace(const struct device *d)
1954{
1955 const struct net_device *dev = to_net_dev(d);
1956
1957 return dev_net(dev);
1958}
1959
1960static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1961{
1962 const struct net_device *dev = to_net_dev(d);
1963 const struct net *net = dev_net(dev);
1964
1965 net_ns_get_ownership(net, uid, gid);
1966}
1967
1968static struct class net_class __ro_after_init = {
1969 .name = "net",
1970 .dev_release = netdev_release,
1971 .dev_groups = net_class_groups,
1972 .dev_uevent = netdev_uevent,
1973 .ns_type = &net_ns_type_operations,
1974 .namespace = net_namespace,
1975 .get_ownership = net_get_ownership,
1976};
1977
1978#ifdef CONFIG_OF
1979static int of_dev_node_match(struct device *dev, const void *data)
1980{
1981 for (; dev; dev = dev->parent) {
1982 if (dev->of_node == data)
1983 return 1;
1984 }
1985
1986 return 0;
1987}
1988
1989/*
1990 * of_find_net_device_by_node - lookup the net device for the device node
1991 * @np: OF device node
1992 *
1993 * Looks up the net_device structure corresponding with the device node.
1994 * If successful, returns a pointer to the net_device with the embedded
1995 * struct device refcount incremented by one, or NULL on failure. The
1996 * refcount must be dropped when done with the net_device.
1997 */
1998struct net_device *of_find_net_device_by_node(struct device_node *np)
1999{
2000 struct device *dev;
2001
2002 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2003 if (!dev)
2004 return NULL;
2005
2006 return to_net_dev(dev);
2007}
2008EXPORT_SYMBOL(of_find_net_device_by_node);
2009#endif
2010
2011/* Delete sysfs entries but hold kobject reference until after all
2012 * netdev references are gone.
2013 */
2014void netdev_unregister_kobject(struct net_device *ndev)
2015{
2016 struct device *dev = &ndev->dev;
2017
2018 if (!refcount_read(&dev_net(ndev)->ns.count))
2019 dev_set_uevent_suppress(dev, 1);
2020
2021 kobject_get(&dev->kobj);
2022
2023 remove_queue_kobjects(ndev);
2024
2025 pm_runtime_set_memalloc_noio(dev, false);
2026
2027 device_del(dev);
2028}
2029
2030/* Create sysfs entries for network device. */
2031int netdev_register_kobject(struct net_device *ndev)
2032{
2033 struct device *dev = &ndev->dev;
2034 const struct attribute_group **groups = ndev->sysfs_groups;
2035 int error = 0;
2036
2037 device_initialize(dev);
2038 dev->class = &net_class;
2039 dev->platform_data = ndev;
2040 dev->groups = groups;
2041
2042 dev_set_name(dev, "%s", ndev->name);
2043
2044#ifdef CONFIG_SYSFS
2045 /* Allow for a device specific group */
2046 if (*groups)
2047 groups++;
2048
2049 *groups++ = &netstat_group;
2050
2051 if (wireless_group_needed(ndev))
2052 *groups++ = &wireless_group;
2053#endif /* CONFIG_SYSFS */
2054
2055 error = device_add(dev);
2056 if (error)
2057 return error;
2058
2059 error = register_queue_kobjects(ndev);
2060 if (error) {
2061 device_del(dev);
2062 return error;
2063 }
2064
2065 pm_runtime_set_memalloc_noio(dev, true);
2066
2067 return error;
2068}
2069
2070/* Change owner for sysfs entries when moving network devices across network
2071 * namespaces owned by different user namespaces.
2072 */
2073int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2074 const struct net *net_new)
2075{
2076 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2077 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2078 struct device *dev = &ndev->dev;
2079 int error;
2080
2081 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2082 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2083
2084 /* The network namespace was changed but the owning user namespace is
2085 * identical so there's no need to change the owner of sysfs entries.
2086 */
2087 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2088 return 0;
2089
2090 error = device_change_owner(dev, new_uid, new_gid);
2091 if (error)
2092 return error;
2093
2094 error = queue_change_owner(ndev, new_uid, new_gid);
2095 if (error)
2096 return error;
2097
2098 return 0;
2099}
2100
2101int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2102 const void *ns)
2103{
2104 return class_create_file_ns(&net_class, class_attr, ns);
2105}
2106EXPORT_SYMBOL(netdev_class_create_file_ns);
2107
2108void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2109 const void *ns)
2110{
2111 class_remove_file_ns(&net_class, class_attr, ns);
2112}
2113EXPORT_SYMBOL(netdev_class_remove_file_ns);
2114
2115int __init netdev_kobject_init(void)
2116{
2117 kobj_ns_type_register(&net_ns_type_operations);
2118 return class_register(&net_class);
2119}