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