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