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/*
2 * net-sysfs.c - network device class and attributes
3 *
4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/capability.h>
13#include <linux/kernel.h>
14#include <linux/netdevice.h>
15#include <linux/if_arp.h>
16#include <linux/slab.h>
17#include <linux/nsproxy.h>
18#include <net/sock.h>
19#include <net/net_namespace.h>
20#include <linux/rtnetlink.h>
21#include <linux/wireless.h>
22#include <linux/vmalloc.h>
23#include <linux/export.h>
24#include <linux/jiffies.h>
25#include <net/wext.h>
26
27#include "net-sysfs.h"
28
29#ifdef CONFIG_SYSFS
30static const char fmt_hex[] = "%#x\n";
31static const char fmt_long_hex[] = "%#lx\n";
32static const char fmt_dec[] = "%d\n";
33static const char fmt_udec[] = "%u\n";
34static const char fmt_ulong[] = "%lu\n";
35static const char fmt_u64[] = "%llu\n";
36
37static inline int dev_isalive(const struct net_device *dev)
38{
39 return dev->reg_state <= NETREG_REGISTERED;
40}
41
42/* use same locking rules as GIF* ioctl's */
43static ssize_t netdev_show(const struct device *dev,
44 struct device_attribute *attr, char *buf,
45 ssize_t (*format)(const struct net_device *, char *))
46{
47 struct net_device *net = to_net_dev(dev);
48 ssize_t ret = -EINVAL;
49
50 read_lock(&dev_base_lock);
51 if (dev_isalive(net))
52 ret = (*format)(net, buf);
53 read_unlock(&dev_base_lock);
54
55 return ret;
56}
57
58/* generate a show function for simple field */
59#define NETDEVICE_SHOW(field, format_string) \
60static ssize_t format_##field(const struct net_device *net, char *buf) \
61{ \
62 return sprintf(buf, format_string, net->field); \
63} \
64static ssize_t show_##field(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
66{ \
67 return netdev_show(dev, attr, buf, format_##field); \
68}
69
70
71/* use same locking and permission rules as SIF* ioctl's */
72static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
73 const char *buf, size_t len,
74 int (*set)(struct net_device *, unsigned long))
75{
76 struct net_device *net = to_net_dev(dev);
77 unsigned long new;
78 int ret = -EINVAL;
79
80 if (!capable(CAP_NET_ADMIN))
81 return -EPERM;
82
83 ret = kstrtoul(buf, 0, &new);
84 if (ret)
85 goto err;
86
87 if (!rtnl_trylock())
88 return restart_syscall();
89
90 if (dev_isalive(net)) {
91 if ((ret = (*set)(net, new)) == 0)
92 ret = len;
93 }
94 rtnl_unlock();
95 err:
96 return ret;
97}
98
99NETDEVICE_SHOW(dev_id, fmt_hex);
100NETDEVICE_SHOW(addr_assign_type, fmt_dec);
101NETDEVICE_SHOW(addr_len, fmt_dec);
102NETDEVICE_SHOW(iflink, fmt_dec);
103NETDEVICE_SHOW(ifindex, fmt_dec);
104NETDEVICE_SHOW(type, fmt_dec);
105NETDEVICE_SHOW(link_mode, fmt_dec);
106
107/* use same locking rules as GIFHWADDR ioctl's */
108static ssize_t show_address(struct device *dev, struct device_attribute *attr,
109 char *buf)
110{
111 struct net_device *net = to_net_dev(dev);
112 ssize_t ret = -EINVAL;
113
114 read_lock(&dev_base_lock);
115 if (dev_isalive(net))
116 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
117 read_unlock(&dev_base_lock);
118 return ret;
119}
120
121static ssize_t show_broadcast(struct device *dev,
122 struct device_attribute *attr, char *buf)
123{
124 struct net_device *net = to_net_dev(dev);
125 if (dev_isalive(net))
126 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
127 return -EINVAL;
128}
129
130static ssize_t show_carrier(struct device *dev,
131 struct device_attribute *attr, char *buf)
132{
133 struct net_device *netdev = to_net_dev(dev);
134 if (netif_running(netdev)) {
135 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
136 }
137 return -EINVAL;
138}
139
140static ssize_t show_speed(struct device *dev,
141 struct device_attribute *attr, char *buf)
142{
143 struct net_device *netdev = to_net_dev(dev);
144 int ret = -EINVAL;
145
146 if (!rtnl_trylock())
147 return restart_syscall();
148
149 if (netif_running(netdev)) {
150 struct ethtool_cmd cmd;
151 if (!__ethtool_get_settings(netdev, &cmd))
152 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
153 }
154 rtnl_unlock();
155 return ret;
156}
157
158static ssize_t show_duplex(struct device *dev,
159 struct device_attribute *attr, char *buf)
160{
161 struct net_device *netdev = to_net_dev(dev);
162 int ret = -EINVAL;
163
164 if (!rtnl_trylock())
165 return restart_syscall();
166
167 if (netif_running(netdev)) {
168 struct ethtool_cmd cmd;
169 if (!__ethtool_get_settings(netdev, &cmd))
170 ret = sprintf(buf, "%s\n",
171 cmd.duplex ? "full" : "half");
172 }
173 rtnl_unlock();
174 return ret;
175}
176
177static ssize_t show_dormant(struct device *dev,
178 struct device_attribute *attr, char *buf)
179{
180 struct net_device *netdev = to_net_dev(dev);
181
182 if (netif_running(netdev))
183 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
184
185 return -EINVAL;
186}
187
188static const char *const operstates[] = {
189 "unknown",
190 "notpresent", /* currently unused */
191 "down",
192 "lowerlayerdown",
193 "testing", /* currently unused */
194 "dormant",
195 "up"
196};
197
198static ssize_t show_operstate(struct device *dev,
199 struct device_attribute *attr, char *buf)
200{
201 const struct net_device *netdev = to_net_dev(dev);
202 unsigned char operstate;
203
204 read_lock(&dev_base_lock);
205 operstate = netdev->operstate;
206 if (!netif_running(netdev))
207 operstate = IF_OPER_DOWN;
208 read_unlock(&dev_base_lock);
209
210 if (operstate >= ARRAY_SIZE(operstates))
211 return -EINVAL; /* should not happen */
212
213 return sprintf(buf, "%s\n", operstates[operstate]);
214}
215
216/* read-write attributes */
217NETDEVICE_SHOW(mtu, fmt_dec);
218
219static int change_mtu(struct net_device *net, unsigned long new_mtu)
220{
221 return dev_set_mtu(net, (int) new_mtu);
222}
223
224static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
225 const char *buf, size_t len)
226{
227 return netdev_store(dev, attr, buf, len, change_mtu);
228}
229
230NETDEVICE_SHOW(flags, fmt_hex);
231
232static int change_flags(struct net_device *net, unsigned long new_flags)
233{
234 return dev_change_flags(net, (unsigned int) new_flags);
235}
236
237static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t len)
239{
240 return netdev_store(dev, attr, buf, len, change_flags);
241}
242
243NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
244
245static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
246{
247 net->tx_queue_len = new_len;
248 return 0;
249}
250
251static ssize_t store_tx_queue_len(struct device *dev,
252 struct device_attribute *attr,
253 const char *buf, size_t len)
254{
255 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
256}
257
258static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
259 const char *buf, size_t len)
260{
261 struct net_device *netdev = to_net_dev(dev);
262 size_t count = len;
263 ssize_t ret;
264
265 if (!capable(CAP_NET_ADMIN))
266 return -EPERM;
267
268 /* ignore trailing newline */
269 if (len > 0 && buf[len - 1] == '\n')
270 --count;
271
272 if (!rtnl_trylock())
273 return restart_syscall();
274 ret = dev_set_alias(netdev, buf, count);
275 rtnl_unlock();
276
277 return ret < 0 ? ret : len;
278}
279
280static ssize_t show_ifalias(struct device *dev,
281 struct device_attribute *attr, char *buf)
282{
283 const struct net_device *netdev = to_net_dev(dev);
284 ssize_t ret = 0;
285
286 if (!rtnl_trylock())
287 return restart_syscall();
288 if (netdev->ifalias)
289 ret = sprintf(buf, "%s\n", netdev->ifalias);
290 rtnl_unlock();
291 return ret;
292}
293
294NETDEVICE_SHOW(group, fmt_dec);
295
296static int change_group(struct net_device *net, unsigned long new_group)
297{
298 dev_set_group(net, (int) new_group);
299 return 0;
300}
301
302static ssize_t store_group(struct device *dev, struct device_attribute *attr,
303 const char *buf, size_t len)
304{
305 return netdev_store(dev, attr, buf, len, change_group);
306}
307
308static struct device_attribute net_class_attributes[] = {
309 __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
310 __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
311 __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
312 __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
313 __ATTR(iflink, S_IRUGO, show_iflink, NULL),
314 __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
315 __ATTR(type, S_IRUGO, show_type, NULL),
316 __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
317 __ATTR(address, S_IRUGO, show_address, NULL),
318 __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
319 __ATTR(carrier, S_IRUGO, show_carrier, NULL),
320 __ATTR(speed, S_IRUGO, show_speed, NULL),
321 __ATTR(duplex, S_IRUGO, show_duplex, NULL),
322 __ATTR(dormant, S_IRUGO, show_dormant, NULL),
323 __ATTR(operstate, S_IRUGO, show_operstate, NULL),
324 __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
325 __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
326 __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
327 store_tx_queue_len),
328 __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
329 {}
330};
331
332/* Show a given an attribute in the statistics group */
333static ssize_t netstat_show(const struct device *d,
334 struct device_attribute *attr, char *buf,
335 unsigned long offset)
336{
337 struct net_device *dev = to_net_dev(d);
338 ssize_t ret = -EINVAL;
339
340 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
341 offset % sizeof(u64) != 0);
342
343 read_lock(&dev_base_lock);
344 if (dev_isalive(dev)) {
345 struct rtnl_link_stats64 temp;
346 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
347
348 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
349 }
350 read_unlock(&dev_base_lock);
351 return ret;
352}
353
354/* generate a read-only statistics attribute */
355#define NETSTAT_ENTRY(name) \
356static ssize_t show_##name(struct device *d, \
357 struct device_attribute *attr, char *buf) \
358{ \
359 return netstat_show(d, attr, buf, \
360 offsetof(struct rtnl_link_stats64, name)); \
361} \
362static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
363
364NETSTAT_ENTRY(rx_packets);
365NETSTAT_ENTRY(tx_packets);
366NETSTAT_ENTRY(rx_bytes);
367NETSTAT_ENTRY(tx_bytes);
368NETSTAT_ENTRY(rx_errors);
369NETSTAT_ENTRY(tx_errors);
370NETSTAT_ENTRY(rx_dropped);
371NETSTAT_ENTRY(tx_dropped);
372NETSTAT_ENTRY(multicast);
373NETSTAT_ENTRY(collisions);
374NETSTAT_ENTRY(rx_length_errors);
375NETSTAT_ENTRY(rx_over_errors);
376NETSTAT_ENTRY(rx_crc_errors);
377NETSTAT_ENTRY(rx_frame_errors);
378NETSTAT_ENTRY(rx_fifo_errors);
379NETSTAT_ENTRY(rx_missed_errors);
380NETSTAT_ENTRY(tx_aborted_errors);
381NETSTAT_ENTRY(tx_carrier_errors);
382NETSTAT_ENTRY(tx_fifo_errors);
383NETSTAT_ENTRY(tx_heartbeat_errors);
384NETSTAT_ENTRY(tx_window_errors);
385NETSTAT_ENTRY(rx_compressed);
386NETSTAT_ENTRY(tx_compressed);
387
388static struct attribute *netstat_attrs[] = {
389 &dev_attr_rx_packets.attr,
390 &dev_attr_tx_packets.attr,
391 &dev_attr_rx_bytes.attr,
392 &dev_attr_tx_bytes.attr,
393 &dev_attr_rx_errors.attr,
394 &dev_attr_tx_errors.attr,
395 &dev_attr_rx_dropped.attr,
396 &dev_attr_tx_dropped.attr,
397 &dev_attr_multicast.attr,
398 &dev_attr_collisions.attr,
399 &dev_attr_rx_length_errors.attr,
400 &dev_attr_rx_over_errors.attr,
401 &dev_attr_rx_crc_errors.attr,
402 &dev_attr_rx_frame_errors.attr,
403 &dev_attr_rx_fifo_errors.attr,
404 &dev_attr_rx_missed_errors.attr,
405 &dev_attr_tx_aborted_errors.attr,
406 &dev_attr_tx_carrier_errors.attr,
407 &dev_attr_tx_fifo_errors.attr,
408 &dev_attr_tx_heartbeat_errors.attr,
409 &dev_attr_tx_window_errors.attr,
410 &dev_attr_rx_compressed.attr,
411 &dev_attr_tx_compressed.attr,
412 NULL
413};
414
415
416static struct attribute_group netstat_group = {
417 .name = "statistics",
418 .attrs = netstat_attrs,
419};
420
421#ifdef CONFIG_WIRELESS_EXT_SYSFS
422/* helper function that does all the locking etc for wireless stats */
423static ssize_t wireless_show(struct device *d, char *buf,
424 ssize_t (*format)(const struct iw_statistics *,
425 char *))
426{
427 struct net_device *dev = to_net_dev(d);
428 const struct iw_statistics *iw;
429 ssize_t ret = -EINVAL;
430
431 if (!rtnl_trylock())
432 return restart_syscall();
433 if (dev_isalive(dev)) {
434 iw = get_wireless_stats(dev);
435 if (iw)
436 ret = (*format)(iw, buf);
437 }
438 rtnl_unlock();
439
440 return ret;
441}
442
443/* show function template for wireless fields */
444#define WIRELESS_SHOW(name, field, format_string) \
445static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
446{ \
447 return sprintf(buf, format_string, iw->field); \
448} \
449static ssize_t show_iw_##name(struct device *d, \
450 struct device_attribute *attr, char *buf) \
451{ \
452 return wireless_show(d, buf, format_iw_##name); \
453} \
454static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
455
456WIRELESS_SHOW(status, status, fmt_hex);
457WIRELESS_SHOW(link, qual.qual, fmt_dec);
458WIRELESS_SHOW(level, qual.level, fmt_dec);
459WIRELESS_SHOW(noise, qual.noise, fmt_dec);
460WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
461WIRELESS_SHOW(crypt, discard.code, fmt_dec);
462WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
463WIRELESS_SHOW(misc, discard.misc, fmt_dec);
464WIRELESS_SHOW(retries, discard.retries, fmt_dec);
465WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
466
467static struct attribute *wireless_attrs[] = {
468 &dev_attr_status.attr,
469 &dev_attr_link.attr,
470 &dev_attr_level.attr,
471 &dev_attr_noise.attr,
472 &dev_attr_nwid.attr,
473 &dev_attr_crypt.attr,
474 &dev_attr_fragment.attr,
475 &dev_attr_retries.attr,
476 &dev_attr_misc.attr,
477 &dev_attr_beacon.attr,
478 NULL
479};
480
481static struct attribute_group wireless_group = {
482 .name = "wireless",
483 .attrs = wireless_attrs,
484};
485#endif
486#endif /* CONFIG_SYSFS */
487
488#ifdef CONFIG_RPS
489/*
490 * RX queue sysfs structures and functions.
491 */
492struct rx_queue_attribute {
493 struct attribute attr;
494 ssize_t (*show)(struct netdev_rx_queue *queue,
495 struct rx_queue_attribute *attr, char *buf);
496 ssize_t (*store)(struct netdev_rx_queue *queue,
497 struct rx_queue_attribute *attr, const char *buf, size_t len);
498};
499#define to_rx_queue_attr(_attr) container_of(_attr, \
500 struct rx_queue_attribute, attr)
501
502#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
503
504static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
505 char *buf)
506{
507 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
508 struct netdev_rx_queue *queue = to_rx_queue(kobj);
509
510 if (!attribute->show)
511 return -EIO;
512
513 return attribute->show(queue, attribute, buf);
514}
515
516static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
517 const char *buf, size_t count)
518{
519 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
520 struct netdev_rx_queue *queue = to_rx_queue(kobj);
521
522 if (!attribute->store)
523 return -EIO;
524
525 return attribute->store(queue, attribute, buf, count);
526}
527
528static const struct sysfs_ops rx_queue_sysfs_ops = {
529 .show = rx_queue_attr_show,
530 .store = rx_queue_attr_store,
531};
532
533static ssize_t show_rps_map(struct netdev_rx_queue *queue,
534 struct rx_queue_attribute *attribute, char *buf)
535{
536 struct rps_map *map;
537 cpumask_var_t mask;
538 size_t len = 0;
539 int i;
540
541 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
542 return -ENOMEM;
543
544 rcu_read_lock();
545 map = rcu_dereference(queue->rps_map);
546 if (map)
547 for (i = 0; i < map->len; i++)
548 cpumask_set_cpu(map->cpus[i], mask);
549
550 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
551 if (PAGE_SIZE - len < 3) {
552 rcu_read_unlock();
553 free_cpumask_var(mask);
554 return -EINVAL;
555 }
556 rcu_read_unlock();
557
558 free_cpumask_var(mask);
559 len += sprintf(buf + len, "\n");
560 return len;
561}
562
563static ssize_t store_rps_map(struct netdev_rx_queue *queue,
564 struct rx_queue_attribute *attribute,
565 const char *buf, size_t len)
566{
567 struct rps_map *old_map, *map;
568 cpumask_var_t mask;
569 int err, cpu, i;
570 static DEFINE_SPINLOCK(rps_map_lock);
571
572 if (!capable(CAP_NET_ADMIN))
573 return -EPERM;
574
575 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
576 return -ENOMEM;
577
578 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
579 if (err) {
580 free_cpumask_var(mask);
581 return err;
582 }
583
584 map = kzalloc(max_t(unsigned int,
585 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
586 GFP_KERNEL);
587 if (!map) {
588 free_cpumask_var(mask);
589 return -ENOMEM;
590 }
591
592 i = 0;
593 for_each_cpu_and(cpu, mask, cpu_online_mask)
594 map->cpus[i++] = cpu;
595
596 if (i)
597 map->len = i;
598 else {
599 kfree(map);
600 map = NULL;
601 }
602
603 spin_lock(&rps_map_lock);
604 old_map = rcu_dereference_protected(queue->rps_map,
605 lockdep_is_held(&rps_map_lock));
606 rcu_assign_pointer(queue->rps_map, map);
607 spin_unlock(&rps_map_lock);
608
609 if (map)
610 static_key_slow_inc(&rps_needed);
611 if (old_map) {
612 kfree_rcu(old_map, rcu);
613 static_key_slow_dec(&rps_needed);
614 }
615 free_cpumask_var(mask);
616 return len;
617}
618
619static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
620 struct rx_queue_attribute *attr,
621 char *buf)
622{
623 struct rps_dev_flow_table *flow_table;
624 unsigned long val = 0;
625
626 rcu_read_lock();
627 flow_table = rcu_dereference(queue->rps_flow_table);
628 if (flow_table)
629 val = (unsigned long)flow_table->mask + 1;
630 rcu_read_unlock();
631
632 return sprintf(buf, "%lu\n", val);
633}
634
635static void rps_dev_flow_table_release_work(struct work_struct *work)
636{
637 struct rps_dev_flow_table *table = container_of(work,
638 struct rps_dev_flow_table, free_work);
639
640 vfree(table);
641}
642
643static void rps_dev_flow_table_release(struct rcu_head *rcu)
644{
645 struct rps_dev_flow_table *table = container_of(rcu,
646 struct rps_dev_flow_table, rcu);
647
648 INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
649 schedule_work(&table->free_work);
650}
651
652static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
653 struct rx_queue_attribute *attr,
654 const char *buf, size_t len)
655{
656 unsigned long mask, count;
657 struct rps_dev_flow_table *table, *old_table;
658 static DEFINE_SPINLOCK(rps_dev_flow_lock);
659 int rc;
660
661 if (!capable(CAP_NET_ADMIN))
662 return -EPERM;
663
664 rc = kstrtoul(buf, 0, &count);
665 if (rc < 0)
666 return rc;
667
668 if (count) {
669 mask = count - 1;
670 /* mask = roundup_pow_of_two(count) - 1;
671 * without overflows...
672 */
673 while ((mask | (mask >> 1)) != mask)
674 mask |= (mask >> 1);
675 /* On 64 bit arches, must check mask fits in table->mask (u32),
676 * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
677 * doesnt overflow.
678 */
679#if BITS_PER_LONG > 32
680 if (mask > (unsigned long)(u32)mask)
681 return -EINVAL;
682#else
683 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
684 / sizeof(struct rps_dev_flow)) {
685 /* Enforce a limit to prevent overflow */
686 return -EINVAL;
687 }
688#endif
689 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
690 if (!table)
691 return -ENOMEM;
692
693 table->mask = mask;
694 for (count = 0; count <= mask; count++)
695 table->flows[count].cpu = RPS_NO_CPU;
696 } else
697 table = NULL;
698
699 spin_lock(&rps_dev_flow_lock);
700 old_table = rcu_dereference_protected(queue->rps_flow_table,
701 lockdep_is_held(&rps_dev_flow_lock));
702 rcu_assign_pointer(queue->rps_flow_table, table);
703 spin_unlock(&rps_dev_flow_lock);
704
705 if (old_table)
706 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
707
708 return len;
709}
710
711static struct rx_queue_attribute rps_cpus_attribute =
712 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
713
714
715static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
716 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
717 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
718
719static struct attribute *rx_queue_default_attrs[] = {
720 &rps_cpus_attribute.attr,
721 &rps_dev_flow_table_cnt_attribute.attr,
722 NULL
723};
724
725static void rx_queue_release(struct kobject *kobj)
726{
727 struct netdev_rx_queue *queue = to_rx_queue(kobj);
728 struct rps_map *map;
729 struct rps_dev_flow_table *flow_table;
730
731
732 map = rcu_dereference_protected(queue->rps_map, 1);
733 if (map) {
734 RCU_INIT_POINTER(queue->rps_map, NULL);
735 kfree_rcu(map, rcu);
736 }
737
738 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
739 if (flow_table) {
740 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
741 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
742 }
743
744 memset(kobj, 0, sizeof(*kobj));
745 dev_put(queue->dev);
746}
747
748static struct kobj_type rx_queue_ktype = {
749 .sysfs_ops = &rx_queue_sysfs_ops,
750 .release = rx_queue_release,
751 .default_attrs = rx_queue_default_attrs,
752};
753
754static int rx_queue_add_kobject(struct net_device *net, int index)
755{
756 struct netdev_rx_queue *queue = net->_rx + index;
757 struct kobject *kobj = &queue->kobj;
758 int error = 0;
759
760 kobj->kset = net->queues_kset;
761 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
762 "rx-%u", index);
763 if (error) {
764 kobject_put(kobj);
765 return error;
766 }
767
768 kobject_uevent(kobj, KOBJ_ADD);
769 dev_hold(queue->dev);
770
771 return error;
772}
773#endif /* CONFIG_RPS */
774
775int
776net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
777{
778#ifdef CONFIG_RPS
779 int i;
780 int error = 0;
781
782 for (i = old_num; i < new_num; i++) {
783 error = rx_queue_add_kobject(net, i);
784 if (error) {
785 new_num = old_num;
786 break;
787 }
788 }
789
790 while (--i >= new_num)
791 kobject_put(&net->_rx[i].kobj);
792
793 return error;
794#else
795 return 0;
796#endif
797}
798
799#ifdef CONFIG_SYSFS
800/*
801 * netdev_queue sysfs structures and functions.
802 */
803struct netdev_queue_attribute {
804 struct attribute attr;
805 ssize_t (*show)(struct netdev_queue *queue,
806 struct netdev_queue_attribute *attr, char *buf);
807 ssize_t (*store)(struct netdev_queue *queue,
808 struct netdev_queue_attribute *attr, const char *buf, size_t len);
809};
810#define to_netdev_queue_attr(_attr) container_of(_attr, \
811 struct netdev_queue_attribute, attr)
812
813#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
814
815static ssize_t netdev_queue_attr_show(struct kobject *kobj,
816 struct attribute *attr, char *buf)
817{
818 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
819 struct netdev_queue *queue = to_netdev_queue(kobj);
820
821 if (!attribute->show)
822 return -EIO;
823
824 return attribute->show(queue, attribute, buf);
825}
826
827static ssize_t netdev_queue_attr_store(struct kobject *kobj,
828 struct attribute *attr,
829 const char *buf, size_t count)
830{
831 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
832 struct netdev_queue *queue = to_netdev_queue(kobj);
833
834 if (!attribute->store)
835 return -EIO;
836
837 return attribute->store(queue, attribute, buf, count);
838}
839
840static const struct sysfs_ops netdev_queue_sysfs_ops = {
841 .show = netdev_queue_attr_show,
842 .store = netdev_queue_attr_store,
843};
844
845static ssize_t show_trans_timeout(struct netdev_queue *queue,
846 struct netdev_queue_attribute *attribute,
847 char *buf)
848{
849 unsigned long trans_timeout;
850
851 spin_lock_irq(&queue->_xmit_lock);
852 trans_timeout = queue->trans_timeout;
853 spin_unlock_irq(&queue->_xmit_lock);
854
855 return sprintf(buf, "%lu", trans_timeout);
856}
857
858static struct netdev_queue_attribute queue_trans_timeout =
859 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
860
861#ifdef CONFIG_BQL
862/*
863 * Byte queue limits sysfs structures and functions.
864 */
865static ssize_t bql_show(char *buf, unsigned int value)
866{
867 return sprintf(buf, "%u\n", value);
868}
869
870static ssize_t bql_set(const char *buf, const size_t count,
871 unsigned int *pvalue)
872{
873 unsigned int value;
874 int err;
875
876 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
877 value = DQL_MAX_LIMIT;
878 else {
879 err = kstrtouint(buf, 10, &value);
880 if (err < 0)
881 return err;
882 if (value > DQL_MAX_LIMIT)
883 return -EINVAL;
884 }
885
886 *pvalue = value;
887
888 return count;
889}
890
891static ssize_t bql_show_hold_time(struct netdev_queue *queue,
892 struct netdev_queue_attribute *attr,
893 char *buf)
894{
895 struct dql *dql = &queue->dql;
896
897 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
898}
899
900static ssize_t bql_set_hold_time(struct netdev_queue *queue,
901 struct netdev_queue_attribute *attribute,
902 const char *buf, size_t len)
903{
904 struct dql *dql = &queue->dql;
905 unsigned int value;
906 int err;
907
908 err = kstrtouint(buf, 10, &value);
909 if (err < 0)
910 return err;
911
912 dql->slack_hold_time = msecs_to_jiffies(value);
913
914 return len;
915}
916
917static struct netdev_queue_attribute bql_hold_time_attribute =
918 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
919 bql_set_hold_time);
920
921static ssize_t bql_show_inflight(struct netdev_queue *queue,
922 struct netdev_queue_attribute *attr,
923 char *buf)
924{
925 struct dql *dql = &queue->dql;
926
927 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
928}
929
930static struct netdev_queue_attribute bql_inflight_attribute =
931 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
932
933#define BQL_ATTR(NAME, FIELD) \
934static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
935 struct netdev_queue_attribute *attr, \
936 char *buf) \
937{ \
938 return bql_show(buf, queue->dql.FIELD); \
939} \
940 \
941static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
942 struct netdev_queue_attribute *attr, \
943 const char *buf, size_t len) \
944{ \
945 return bql_set(buf, len, &queue->dql.FIELD); \
946} \
947 \
948static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
949 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
950 bql_set_ ## NAME);
951
952BQL_ATTR(limit, limit)
953BQL_ATTR(limit_max, max_limit)
954BQL_ATTR(limit_min, min_limit)
955
956static struct attribute *dql_attrs[] = {
957 &bql_limit_attribute.attr,
958 &bql_limit_max_attribute.attr,
959 &bql_limit_min_attribute.attr,
960 &bql_hold_time_attribute.attr,
961 &bql_inflight_attribute.attr,
962 NULL
963};
964
965static struct attribute_group dql_group = {
966 .name = "byte_queue_limits",
967 .attrs = dql_attrs,
968};
969#endif /* CONFIG_BQL */
970
971#ifdef CONFIG_XPS
972static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
973{
974 struct net_device *dev = queue->dev;
975 int i;
976
977 for (i = 0; i < dev->num_tx_queues; i++)
978 if (queue == &dev->_tx[i])
979 break;
980
981 BUG_ON(i >= dev->num_tx_queues);
982
983 return i;
984}
985
986
987static ssize_t show_xps_map(struct netdev_queue *queue,
988 struct netdev_queue_attribute *attribute, char *buf)
989{
990 struct net_device *dev = queue->dev;
991 struct xps_dev_maps *dev_maps;
992 cpumask_var_t mask;
993 unsigned long index;
994 size_t len = 0;
995 int i;
996
997 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
998 return -ENOMEM;
999
1000 index = get_netdev_queue_index(queue);
1001
1002 rcu_read_lock();
1003 dev_maps = rcu_dereference(dev->xps_maps);
1004 if (dev_maps) {
1005 for_each_possible_cpu(i) {
1006 struct xps_map *map =
1007 rcu_dereference(dev_maps->cpu_map[i]);
1008 if (map) {
1009 int j;
1010 for (j = 0; j < map->len; j++) {
1011 if (map->queues[j] == index) {
1012 cpumask_set_cpu(i, mask);
1013 break;
1014 }
1015 }
1016 }
1017 }
1018 }
1019 rcu_read_unlock();
1020
1021 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1022 if (PAGE_SIZE - len < 3) {
1023 free_cpumask_var(mask);
1024 return -EINVAL;
1025 }
1026
1027 free_cpumask_var(mask);
1028 len += sprintf(buf + len, "\n");
1029 return len;
1030}
1031
1032static DEFINE_MUTEX(xps_map_mutex);
1033#define xmap_dereference(P) \
1034 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1035
1036static void xps_queue_release(struct netdev_queue *queue)
1037{
1038 struct net_device *dev = queue->dev;
1039 struct xps_dev_maps *dev_maps;
1040 struct xps_map *map;
1041 unsigned long index;
1042 int i, pos, nonempty = 0;
1043
1044 index = get_netdev_queue_index(queue);
1045
1046 mutex_lock(&xps_map_mutex);
1047 dev_maps = xmap_dereference(dev->xps_maps);
1048
1049 if (dev_maps) {
1050 for_each_possible_cpu(i) {
1051 map = xmap_dereference(dev_maps->cpu_map[i]);
1052 if (!map)
1053 continue;
1054
1055 for (pos = 0; pos < map->len; pos++)
1056 if (map->queues[pos] == index)
1057 break;
1058
1059 if (pos < map->len) {
1060 if (map->len > 1)
1061 map->queues[pos] =
1062 map->queues[--map->len];
1063 else {
1064 RCU_INIT_POINTER(dev_maps->cpu_map[i],
1065 NULL);
1066 kfree_rcu(map, rcu);
1067 map = NULL;
1068 }
1069 }
1070 if (map)
1071 nonempty = 1;
1072 }
1073
1074 if (!nonempty) {
1075 RCU_INIT_POINTER(dev->xps_maps, NULL);
1076 kfree_rcu(dev_maps, rcu);
1077 }
1078 }
1079 mutex_unlock(&xps_map_mutex);
1080}
1081
1082static ssize_t store_xps_map(struct netdev_queue *queue,
1083 struct netdev_queue_attribute *attribute,
1084 const char *buf, size_t len)
1085{
1086 struct net_device *dev = queue->dev;
1087 cpumask_var_t mask;
1088 int err, i, cpu, pos, map_len, alloc_len, need_set;
1089 unsigned long index;
1090 struct xps_map *map, *new_map;
1091 struct xps_dev_maps *dev_maps, *new_dev_maps;
1092 int nonempty = 0;
1093 int numa_node_id = -2;
1094
1095 if (!capable(CAP_NET_ADMIN))
1096 return -EPERM;
1097
1098 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1099 return -ENOMEM;
1100
1101 index = get_netdev_queue_index(queue);
1102
1103 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1104 if (err) {
1105 free_cpumask_var(mask);
1106 return err;
1107 }
1108
1109 new_dev_maps = kzalloc(max_t(unsigned int,
1110 XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1111 if (!new_dev_maps) {
1112 free_cpumask_var(mask);
1113 return -ENOMEM;
1114 }
1115
1116 mutex_lock(&xps_map_mutex);
1117
1118 dev_maps = xmap_dereference(dev->xps_maps);
1119
1120 for_each_possible_cpu(cpu) {
1121 map = dev_maps ?
1122 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1123 new_map = map;
1124 if (map) {
1125 for (pos = 0; pos < map->len; pos++)
1126 if (map->queues[pos] == index)
1127 break;
1128 map_len = map->len;
1129 alloc_len = map->alloc_len;
1130 } else
1131 pos = map_len = alloc_len = 0;
1132
1133 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1134#ifdef CONFIG_NUMA
1135 if (need_set) {
1136 if (numa_node_id == -2)
1137 numa_node_id = cpu_to_node(cpu);
1138 else if (numa_node_id != cpu_to_node(cpu))
1139 numa_node_id = -1;
1140 }
1141#endif
1142 if (need_set && pos >= map_len) {
1143 /* Need to add queue to this CPU's map */
1144 if (map_len >= alloc_len) {
1145 alloc_len = alloc_len ?
1146 2 * alloc_len : XPS_MIN_MAP_ALLOC;
1147 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1148 GFP_KERNEL,
1149 cpu_to_node(cpu));
1150 if (!new_map)
1151 goto error;
1152 new_map->alloc_len = alloc_len;
1153 for (i = 0; i < map_len; i++)
1154 new_map->queues[i] = map->queues[i];
1155 new_map->len = map_len;
1156 }
1157 new_map->queues[new_map->len++] = index;
1158 } else if (!need_set && pos < map_len) {
1159 /* Need to remove queue from this CPU's map */
1160 if (map_len > 1)
1161 new_map->queues[pos] =
1162 new_map->queues[--new_map->len];
1163 else
1164 new_map = NULL;
1165 }
1166 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1167 }
1168
1169 /* Cleanup old maps */
1170 for_each_possible_cpu(cpu) {
1171 map = dev_maps ?
1172 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1173 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1174 kfree_rcu(map, rcu);
1175 if (new_dev_maps->cpu_map[cpu])
1176 nonempty = 1;
1177 }
1178
1179 if (nonempty) {
1180 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1181 } else {
1182 kfree(new_dev_maps);
1183 RCU_INIT_POINTER(dev->xps_maps, NULL);
1184 }
1185
1186 if (dev_maps)
1187 kfree_rcu(dev_maps, rcu);
1188
1189 netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1190 NUMA_NO_NODE);
1191
1192 mutex_unlock(&xps_map_mutex);
1193
1194 free_cpumask_var(mask);
1195 return len;
1196
1197error:
1198 mutex_unlock(&xps_map_mutex);
1199
1200 if (new_dev_maps)
1201 for_each_possible_cpu(i)
1202 kfree(rcu_dereference_protected(
1203 new_dev_maps->cpu_map[i],
1204 1));
1205 kfree(new_dev_maps);
1206 free_cpumask_var(mask);
1207 return -ENOMEM;
1208}
1209
1210static struct netdev_queue_attribute xps_cpus_attribute =
1211 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1212#endif /* CONFIG_XPS */
1213
1214static struct attribute *netdev_queue_default_attrs[] = {
1215 &queue_trans_timeout.attr,
1216#ifdef CONFIG_XPS
1217 &xps_cpus_attribute.attr,
1218#endif
1219 NULL
1220};
1221
1222static void netdev_queue_release(struct kobject *kobj)
1223{
1224 struct netdev_queue *queue = to_netdev_queue(kobj);
1225
1226#ifdef CONFIG_XPS
1227 xps_queue_release(queue);
1228#endif
1229
1230 memset(kobj, 0, sizeof(*kobj));
1231 dev_put(queue->dev);
1232}
1233
1234static struct kobj_type netdev_queue_ktype = {
1235 .sysfs_ops = &netdev_queue_sysfs_ops,
1236 .release = netdev_queue_release,
1237 .default_attrs = netdev_queue_default_attrs,
1238};
1239
1240static int netdev_queue_add_kobject(struct net_device *net, int index)
1241{
1242 struct netdev_queue *queue = net->_tx + index;
1243 struct kobject *kobj = &queue->kobj;
1244 int error = 0;
1245
1246 kobj->kset = net->queues_kset;
1247 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1248 "tx-%u", index);
1249 if (error)
1250 goto exit;
1251
1252#ifdef CONFIG_BQL
1253 error = sysfs_create_group(kobj, &dql_group);
1254 if (error)
1255 goto exit;
1256#endif
1257
1258 kobject_uevent(kobj, KOBJ_ADD);
1259 dev_hold(queue->dev);
1260
1261 return 0;
1262exit:
1263 kobject_put(kobj);
1264 return error;
1265}
1266#endif /* CONFIG_SYSFS */
1267
1268int
1269netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1270{
1271#ifdef CONFIG_SYSFS
1272 int i;
1273 int error = 0;
1274
1275 for (i = old_num; i < new_num; i++) {
1276 error = netdev_queue_add_kobject(net, i);
1277 if (error) {
1278 new_num = old_num;
1279 break;
1280 }
1281 }
1282
1283 while (--i >= new_num) {
1284 struct netdev_queue *queue = net->_tx + i;
1285
1286#ifdef CONFIG_BQL
1287 sysfs_remove_group(&queue->kobj, &dql_group);
1288#endif
1289 kobject_put(&queue->kobj);
1290 }
1291
1292 return error;
1293#else
1294 return 0;
1295#endif /* CONFIG_SYSFS */
1296}
1297
1298static int register_queue_kobjects(struct net_device *net)
1299{
1300 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1301
1302#ifdef CONFIG_SYSFS
1303 net->queues_kset = kset_create_and_add("queues",
1304 NULL, &net->dev.kobj);
1305 if (!net->queues_kset)
1306 return -ENOMEM;
1307#endif
1308
1309#ifdef CONFIG_RPS
1310 real_rx = net->real_num_rx_queues;
1311#endif
1312 real_tx = net->real_num_tx_queues;
1313
1314 error = net_rx_queue_update_kobjects(net, 0, real_rx);
1315 if (error)
1316 goto error;
1317 rxq = real_rx;
1318
1319 error = netdev_queue_update_kobjects(net, 0, real_tx);
1320 if (error)
1321 goto error;
1322 txq = real_tx;
1323
1324 return 0;
1325
1326error:
1327 netdev_queue_update_kobjects(net, txq, 0);
1328 net_rx_queue_update_kobjects(net, rxq, 0);
1329 return error;
1330}
1331
1332static void remove_queue_kobjects(struct net_device *net)
1333{
1334 int real_rx = 0, real_tx = 0;
1335
1336#ifdef CONFIG_RPS
1337 real_rx = net->real_num_rx_queues;
1338#endif
1339 real_tx = net->real_num_tx_queues;
1340
1341 net_rx_queue_update_kobjects(net, real_rx, 0);
1342 netdev_queue_update_kobjects(net, real_tx, 0);
1343#ifdef CONFIG_SYSFS
1344 kset_unregister(net->queues_kset);
1345#endif
1346}
1347
1348static void *net_grab_current_ns(void)
1349{
1350 struct net *ns = current->nsproxy->net_ns;
1351#ifdef CONFIG_NET_NS
1352 if (ns)
1353 atomic_inc(&ns->passive);
1354#endif
1355 return ns;
1356}
1357
1358static const void *net_initial_ns(void)
1359{
1360 return &init_net;
1361}
1362
1363static const void *net_netlink_ns(struct sock *sk)
1364{
1365 return sock_net(sk);
1366}
1367
1368struct kobj_ns_type_operations net_ns_type_operations = {
1369 .type = KOBJ_NS_TYPE_NET,
1370 .grab_current_ns = net_grab_current_ns,
1371 .netlink_ns = net_netlink_ns,
1372 .initial_ns = net_initial_ns,
1373 .drop_ns = net_drop_ns,
1374};
1375EXPORT_SYMBOL_GPL(net_ns_type_operations);
1376
1377#ifdef CONFIG_HOTPLUG
1378static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1379{
1380 struct net_device *dev = to_net_dev(d);
1381 int retval;
1382
1383 /* pass interface to uevent. */
1384 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1385 if (retval)
1386 goto exit;
1387
1388 /* pass ifindex to uevent.
1389 * ifindex is useful as it won't change (interface name may change)
1390 * and is what RtNetlink uses natively. */
1391 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1392
1393exit:
1394 return retval;
1395}
1396#endif
1397
1398/*
1399 * netdev_release -- destroy and free a dead device.
1400 * Called when last reference to device kobject is gone.
1401 */
1402static void netdev_release(struct device *d)
1403{
1404 struct net_device *dev = to_net_dev(d);
1405
1406 BUG_ON(dev->reg_state != NETREG_RELEASED);
1407
1408 kfree(dev->ifalias);
1409 kfree((char *)dev - dev->padded);
1410}
1411
1412static const void *net_namespace(struct device *d)
1413{
1414 struct net_device *dev;
1415 dev = container_of(d, struct net_device, dev);
1416 return dev_net(dev);
1417}
1418
1419static struct class net_class = {
1420 .name = "net",
1421 .dev_release = netdev_release,
1422#ifdef CONFIG_SYSFS
1423 .dev_attrs = net_class_attributes,
1424#endif /* CONFIG_SYSFS */
1425#ifdef CONFIG_HOTPLUG
1426 .dev_uevent = netdev_uevent,
1427#endif
1428 .ns_type = &net_ns_type_operations,
1429 .namespace = net_namespace,
1430};
1431
1432/* Delete sysfs entries but hold kobject reference until after all
1433 * netdev references are gone.
1434 */
1435void netdev_unregister_kobject(struct net_device * net)
1436{
1437 struct device *dev = &(net->dev);
1438
1439 kobject_get(&dev->kobj);
1440
1441 remove_queue_kobjects(net);
1442
1443 device_del(dev);
1444}
1445
1446/* Create sysfs entries for network device. */
1447int netdev_register_kobject(struct net_device *net)
1448{
1449 struct device *dev = &(net->dev);
1450 const struct attribute_group **groups = net->sysfs_groups;
1451 int error = 0;
1452
1453 device_initialize(dev);
1454 dev->class = &net_class;
1455 dev->platform_data = net;
1456 dev->groups = groups;
1457
1458 dev_set_name(dev, "%s", net->name);
1459
1460#ifdef CONFIG_SYSFS
1461 /* Allow for a device specific group */
1462 if (*groups)
1463 groups++;
1464
1465 *groups++ = &netstat_group;
1466#ifdef CONFIG_WIRELESS_EXT_SYSFS
1467 if (net->ieee80211_ptr)
1468 *groups++ = &wireless_group;
1469#ifdef CONFIG_WIRELESS_EXT
1470 else if (net->wireless_handlers)
1471 *groups++ = &wireless_group;
1472#endif
1473#endif
1474#endif /* CONFIG_SYSFS */
1475
1476 error = device_add(dev);
1477 if (error)
1478 return error;
1479
1480 error = register_queue_kobjects(net);
1481 if (error) {
1482 device_del(dev);
1483 return error;
1484 }
1485
1486 return error;
1487}
1488
1489int netdev_class_create_file(struct class_attribute *class_attr)
1490{
1491 return class_create_file(&net_class, class_attr);
1492}
1493EXPORT_SYMBOL(netdev_class_create_file);
1494
1495void netdev_class_remove_file(struct class_attribute *class_attr)
1496{
1497 class_remove_file(&net_class, class_attr);
1498}
1499EXPORT_SYMBOL(netdev_class_remove_file);
1500
1501int netdev_kobject_init(void)
1502{
1503 kobj_ns_type_register(&net_ns_type_operations);
1504 return class_register(&net_class);
1505}