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