GNU Linux-libre 4.14.266-gnu1
[releases.git] / net / core / net-sysfs.c
1 /*
2  * net-sysfs.c - network device class and attributes
3  *
4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5  *
6  *      This program is free software; you can redistribute it and/or
7  *      modify it under the terms of the GNU General Public License
8  *      as published by the Free Software Foundation; either version
9  *      2 of the License, or (at your option) any later version.
10  */
11
12 #include <linux/capability.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <net/switchdev.h>
16 #include <linux/if_arp.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/nsproxy.h>
20 #include <net/sock.h>
21 #include <net/net_namespace.h>
22 #include <linux/rtnetlink.h>
23 #include <linux/vmalloc.h>
24 #include <linux/export.h>
25 #include <linux/jiffies.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/of.h>
28 #include <linux/of_net.h>
29
30 #include "net-sysfs.h"
31
32 #ifdef CONFIG_SYSFS
33 static const char fmt_hex[] = "%#x\n";
34 static const char fmt_dec[] = "%d\n";
35 static const char fmt_ulong[] = "%lu\n";
36 static const char fmt_u64[] = "%llu\n";
37
38 static inline int dev_isalive(const struct net_device *dev)
39 {
40         return dev->reg_state <= NETREG_REGISTERED;
41 }
42
43 /* use same locking rules as GIF* ioctl's */
44 static ssize_t netdev_show(const struct device *dev,
45                            struct device_attribute *attr, char *buf,
46                            ssize_t (*format)(const struct net_device *, char *))
47 {
48         struct net_device *ndev = to_net_dev(dev);
49         ssize_t ret = -EINVAL;
50
51         read_lock(&dev_base_lock);
52         if (dev_isalive(ndev))
53                 ret = (*format)(ndev, buf);
54         read_unlock(&dev_base_lock);
55
56         return ret;
57 }
58
59 /* generate a show function for simple field */
60 #define NETDEVICE_SHOW(field, format_string)                            \
61 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
62 {                                                                       \
63         return sprintf(buf, format_string, dev->field);                 \
64 }                                                                       \
65 static ssize_t field##_show(struct device *dev,                         \
66                             struct device_attribute *attr, char *buf)   \
67 {                                                                       \
68         return netdev_show(dev, attr, buf, format_##field);             \
69 }                                                                       \
70
71 #define NETDEVICE_SHOW_RO(field, format_string)                         \
72 NETDEVICE_SHOW(field, format_string);                                   \
73 static DEVICE_ATTR_RO(field)
74
75 #define NETDEVICE_SHOW_RW(field, format_string)                         \
76 NETDEVICE_SHOW(field, format_string);                                   \
77 static DEVICE_ATTR_RW(field)
78
79 /* use same locking and permission rules as SIF* ioctl's */
80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81                             const char *buf, size_t len,
82                             int (*set)(struct net_device *, unsigned long))
83 {
84         struct net_device *netdev = to_net_dev(dev);
85         struct net *net = dev_net(netdev);
86         unsigned long new;
87         int ret = -EINVAL;
88
89         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90                 return -EPERM;
91
92         ret = kstrtoul(buf, 0, &new);
93         if (ret)
94                 goto err;
95
96         if (!rtnl_trylock())
97                 return restart_syscall();
98
99         if (dev_isalive(netdev)) {
100                 ret = (*set)(netdev, new);
101                 if (ret == 0)
102                         ret = len;
103         }
104         rtnl_unlock();
105  err:
106         return ret;
107 }
108
109 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114 NETDEVICE_SHOW_RO(type, fmt_dec);
115 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116
117 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118                            char *buf)
119 {
120         struct net_device *ndev = to_net_dev(dev);
121
122         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
123 }
124 static DEVICE_ATTR_RO(iflink);
125
126 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127 {
128         return sprintf(buf, fmt_dec, dev->name_assign_type);
129 }
130
131 static ssize_t name_assign_type_show(struct device *dev,
132                                      struct device_attribute *attr,
133                                      char *buf)
134 {
135         struct net_device *ndev = to_net_dev(dev);
136         ssize_t ret = -EINVAL;
137
138         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140
141         return ret;
142 }
143 static DEVICE_ATTR_RO(name_assign_type);
144
145 /* use same locking rules as GIFHWADDR ioctl's */
146 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147                             char *buf)
148 {
149         struct net_device *ndev = to_net_dev(dev);
150         ssize_t ret = -EINVAL;
151
152         read_lock(&dev_base_lock);
153         if (dev_isalive(ndev))
154                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155         read_unlock(&dev_base_lock);
156         return ret;
157 }
158 static DEVICE_ATTR_RO(address);
159
160 static ssize_t broadcast_show(struct device *dev,
161                               struct device_attribute *attr, char *buf)
162 {
163         struct net_device *ndev = to_net_dev(dev);
164
165         if (dev_isalive(ndev))
166                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167         return -EINVAL;
168 }
169 static DEVICE_ATTR_RO(broadcast);
170
171 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172 {
173         if (!netif_running(dev))
174                 return -EINVAL;
175         return dev_change_carrier(dev, (bool)new_carrier);
176 }
177
178 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179                              const char *buf, size_t len)
180 {
181         return netdev_store(dev, attr, buf, len, change_carrier);
182 }
183
184 static ssize_t carrier_show(struct device *dev,
185                             struct device_attribute *attr, char *buf)
186 {
187         struct net_device *netdev = to_net_dev(dev);
188
189         if (netif_running(netdev))
190                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
191
192         return -EINVAL;
193 }
194 static DEVICE_ATTR_RW(carrier);
195
196 static ssize_t speed_show(struct device *dev,
197                           struct device_attribute *attr, char *buf)
198 {
199         struct net_device *netdev = to_net_dev(dev);
200         int ret = -EINVAL;
201
202         if (!rtnl_trylock())
203                 return restart_syscall();
204
205         if (netif_running(netdev)) {
206                 struct ethtool_link_ksettings cmd;
207
208                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
209                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
210         }
211         rtnl_unlock();
212         return ret;
213 }
214 static DEVICE_ATTR_RO(speed);
215
216 static ssize_t duplex_show(struct device *dev,
217                            struct device_attribute *attr, char *buf)
218 {
219         struct net_device *netdev = to_net_dev(dev);
220         int ret = -EINVAL;
221
222         if (!rtnl_trylock())
223                 return restart_syscall();
224
225         if (netif_running(netdev)) {
226                 struct ethtool_link_ksettings cmd;
227
228                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
229                         const char *duplex;
230
231                         switch (cmd.base.duplex) {
232                         case DUPLEX_HALF:
233                                 duplex = "half";
234                                 break;
235                         case DUPLEX_FULL:
236                                 duplex = "full";
237                                 break;
238                         default:
239                                 duplex = "unknown";
240                                 break;
241                         }
242                         ret = sprintf(buf, "%s\n", duplex);
243                 }
244         }
245         rtnl_unlock();
246         return ret;
247 }
248 static DEVICE_ATTR_RO(duplex);
249
250 static ssize_t dormant_show(struct device *dev,
251                             struct device_attribute *attr, char *buf)
252 {
253         struct net_device *netdev = to_net_dev(dev);
254
255         if (netif_running(netdev))
256                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
257
258         return -EINVAL;
259 }
260 static DEVICE_ATTR_RO(dormant);
261
262 static const char *const operstates[] = {
263         "unknown",
264         "notpresent", /* currently unused */
265         "down",
266         "lowerlayerdown",
267         "testing", /* currently unused */
268         "dormant",
269         "up"
270 };
271
272 static ssize_t operstate_show(struct device *dev,
273                               struct device_attribute *attr, char *buf)
274 {
275         const struct net_device *netdev = to_net_dev(dev);
276         unsigned char operstate;
277
278         read_lock(&dev_base_lock);
279         operstate = netdev->operstate;
280         if (!netif_running(netdev))
281                 operstate = IF_OPER_DOWN;
282         read_unlock(&dev_base_lock);
283
284         if (operstate >= ARRAY_SIZE(operstates))
285                 return -EINVAL; /* should not happen */
286
287         return sprintf(buf, "%s\n", operstates[operstate]);
288 }
289 static DEVICE_ATTR_RO(operstate);
290
291 static ssize_t carrier_changes_show(struct device *dev,
292                                     struct device_attribute *attr,
293                                     char *buf)
294 {
295         struct net_device *netdev = to_net_dev(dev);
296
297         return sprintf(buf, fmt_dec,
298                        atomic_read(&netdev->carrier_changes));
299 }
300 static DEVICE_ATTR_RO(carrier_changes);
301
302 /* read-write attributes */
303
304 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
305 {
306         return dev_set_mtu(dev, (int)new_mtu);
307 }
308
309 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
310                          const char *buf, size_t len)
311 {
312         return netdev_store(dev, attr, buf, len, change_mtu);
313 }
314 NETDEVICE_SHOW_RW(mtu, fmt_dec);
315
316 static int change_flags(struct net_device *dev, unsigned long new_flags)
317 {
318         return dev_change_flags(dev, (unsigned int)new_flags);
319 }
320
321 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
322                            const char *buf, size_t len)
323 {
324         return netdev_store(dev, attr, buf, len, change_flags);
325 }
326 NETDEVICE_SHOW_RW(flags, fmt_hex);
327
328 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
329 {
330         unsigned int orig_len = dev->tx_queue_len;
331         int res;
332
333         if (new_len != (unsigned int)new_len)
334                 return -ERANGE;
335
336         if (new_len != orig_len) {
337                 dev->tx_queue_len = new_len;
338                 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
339                 res = notifier_to_errno(res);
340                 if (res) {
341                         netdev_err(dev,
342                                    "refused to change device tx_queue_len\n");
343                         dev->tx_queue_len = orig_len;
344                         return -EFAULT;
345                 }
346         }
347
348         return 0;
349 }
350
351 static ssize_t tx_queue_len_store(struct device *dev,
352                                   struct device_attribute *attr,
353                                   const char *buf, size_t len)
354 {
355         if (!capable(CAP_NET_ADMIN))
356                 return -EPERM;
357
358         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
359 }
360 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
361
362 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
363 {
364         dev->gro_flush_timeout = val;
365         return 0;
366 }
367
368 static ssize_t gro_flush_timeout_store(struct device *dev,
369                                        struct device_attribute *attr,
370                                        const char *buf, size_t len)
371 {
372         if (!capable(CAP_NET_ADMIN))
373                 return -EPERM;
374
375         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
376 }
377 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
378
379 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
380                              const char *buf, size_t len)
381 {
382         struct net_device *netdev = to_net_dev(dev);
383         struct net *net = dev_net(netdev);
384         size_t count = len;
385         ssize_t ret;
386
387         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
388                 return -EPERM;
389
390         /* ignore trailing newline */
391         if (len >  0 && buf[len - 1] == '\n')
392                 --count;
393
394         if (!rtnl_trylock())
395                 return restart_syscall();
396         ret = dev_set_alias(netdev, buf, count);
397         rtnl_unlock();
398
399         return ret < 0 ? ret : len;
400 }
401
402 static ssize_t ifalias_show(struct device *dev,
403                             struct device_attribute *attr, char *buf)
404 {
405         const struct net_device *netdev = to_net_dev(dev);
406         ssize_t ret = 0;
407
408         if (!rtnl_trylock())
409                 return restart_syscall();
410         if (netdev->ifalias)
411                 ret = sprintf(buf, "%s\n", netdev->ifalias);
412         rtnl_unlock();
413         return ret;
414 }
415 static DEVICE_ATTR_RW(ifalias);
416
417 static int change_group(struct net_device *dev, unsigned long new_group)
418 {
419         dev_set_group(dev, (int)new_group);
420         return 0;
421 }
422
423 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
424                            const char *buf, size_t len)
425 {
426         return netdev_store(dev, attr, buf, len, change_group);
427 }
428 NETDEVICE_SHOW(group, fmt_dec);
429 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
430
431 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
432 {
433         return dev_change_proto_down(dev, (bool)proto_down);
434 }
435
436 static ssize_t proto_down_store(struct device *dev,
437                                 struct device_attribute *attr,
438                                 const char *buf, size_t len)
439 {
440         return netdev_store(dev, attr, buf, len, change_proto_down);
441 }
442 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
443
444 static ssize_t phys_port_id_show(struct device *dev,
445                                  struct device_attribute *attr, char *buf)
446 {
447         struct net_device *netdev = to_net_dev(dev);
448         ssize_t ret = -EINVAL;
449
450         if (!rtnl_trylock())
451                 return restart_syscall();
452
453         if (dev_isalive(netdev)) {
454                 struct netdev_phys_item_id ppid;
455
456                 ret = dev_get_phys_port_id(netdev, &ppid);
457                 if (!ret)
458                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
459         }
460         rtnl_unlock();
461
462         return ret;
463 }
464 static DEVICE_ATTR_RO(phys_port_id);
465
466 static ssize_t phys_port_name_show(struct device *dev,
467                                    struct device_attribute *attr, char *buf)
468 {
469         struct net_device *netdev = to_net_dev(dev);
470         ssize_t ret = -EINVAL;
471
472         if (!rtnl_trylock())
473                 return restart_syscall();
474
475         if (dev_isalive(netdev)) {
476                 char name[IFNAMSIZ];
477
478                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
479                 if (!ret)
480                         ret = sprintf(buf, "%s\n", name);
481         }
482         rtnl_unlock();
483
484         return ret;
485 }
486 static DEVICE_ATTR_RO(phys_port_name);
487
488 static ssize_t phys_switch_id_show(struct device *dev,
489                                    struct device_attribute *attr, char *buf)
490 {
491         struct net_device *netdev = to_net_dev(dev);
492         ssize_t ret = -EINVAL;
493
494         if (!rtnl_trylock())
495                 return restart_syscall();
496
497         if (dev_isalive(netdev)) {
498                 struct switchdev_attr attr = {
499                         .orig_dev = netdev,
500                         .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
501                         .flags = SWITCHDEV_F_NO_RECURSE,
502                 };
503
504                 ret = switchdev_port_attr_get(netdev, &attr);
505                 if (!ret)
506                         ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
507                                       attr.u.ppid.id);
508         }
509         rtnl_unlock();
510
511         return ret;
512 }
513 static DEVICE_ATTR_RO(phys_switch_id);
514
515 static struct attribute *net_class_attrs[] __ro_after_init = {
516         &dev_attr_netdev_group.attr,
517         &dev_attr_type.attr,
518         &dev_attr_dev_id.attr,
519         &dev_attr_dev_port.attr,
520         &dev_attr_iflink.attr,
521         &dev_attr_ifindex.attr,
522         &dev_attr_name_assign_type.attr,
523         &dev_attr_addr_assign_type.attr,
524         &dev_attr_addr_len.attr,
525         &dev_attr_link_mode.attr,
526         &dev_attr_address.attr,
527         &dev_attr_broadcast.attr,
528         &dev_attr_speed.attr,
529         &dev_attr_duplex.attr,
530         &dev_attr_dormant.attr,
531         &dev_attr_operstate.attr,
532         &dev_attr_carrier_changes.attr,
533         &dev_attr_ifalias.attr,
534         &dev_attr_carrier.attr,
535         &dev_attr_mtu.attr,
536         &dev_attr_flags.attr,
537         &dev_attr_tx_queue_len.attr,
538         &dev_attr_gro_flush_timeout.attr,
539         &dev_attr_phys_port_id.attr,
540         &dev_attr_phys_port_name.attr,
541         &dev_attr_phys_switch_id.attr,
542         &dev_attr_proto_down.attr,
543         NULL,
544 };
545 ATTRIBUTE_GROUPS(net_class);
546
547 /* Show a given an attribute in the statistics group */
548 static ssize_t netstat_show(const struct device *d,
549                             struct device_attribute *attr, char *buf,
550                             unsigned long offset)
551 {
552         struct net_device *dev = to_net_dev(d);
553         ssize_t ret = -EINVAL;
554
555         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
556                 offset % sizeof(u64) != 0);
557
558         read_lock(&dev_base_lock);
559         if (dev_isalive(dev)) {
560                 struct rtnl_link_stats64 temp;
561                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
562
563                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
564         }
565         read_unlock(&dev_base_lock);
566         return ret;
567 }
568
569 /* generate a read-only statistics attribute */
570 #define NETSTAT_ENTRY(name)                                             \
571 static ssize_t name##_show(struct device *d,                            \
572                            struct device_attribute *attr, char *buf)    \
573 {                                                                       \
574         return netstat_show(d, attr, buf,                               \
575                             offsetof(struct rtnl_link_stats64, name));  \
576 }                                                                       \
577 static DEVICE_ATTR_RO(name)
578
579 NETSTAT_ENTRY(rx_packets);
580 NETSTAT_ENTRY(tx_packets);
581 NETSTAT_ENTRY(rx_bytes);
582 NETSTAT_ENTRY(tx_bytes);
583 NETSTAT_ENTRY(rx_errors);
584 NETSTAT_ENTRY(tx_errors);
585 NETSTAT_ENTRY(rx_dropped);
586 NETSTAT_ENTRY(tx_dropped);
587 NETSTAT_ENTRY(multicast);
588 NETSTAT_ENTRY(collisions);
589 NETSTAT_ENTRY(rx_length_errors);
590 NETSTAT_ENTRY(rx_over_errors);
591 NETSTAT_ENTRY(rx_crc_errors);
592 NETSTAT_ENTRY(rx_frame_errors);
593 NETSTAT_ENTRY(rx_fifo_errors);
594 NETSTAT_ENTRY(rx_missed_errors);
595 NETSTAT_ENTRY(tx_aborted_errors);
596 NETSTAT_ENTRY(tx_carrier_errors);
597 NETSTAT_ENTRY(tx_fifo_errors);
598 NETSTAT_ENTRY(tx_heartbeat_errors);
599 NETSTAT_ENTRY(tx_window_errors);
600 NETSTAT_ENTRY(rx_compressed);
601 NETSTAT_ENTRY(tx_compressed);
602 NETSTAT_ENTRY(rx_nohandler);
603
604 static struct attribute *netstat_attrs[] __ro_after_init = {
605         &dev_attr_rx_packets.attr,
606         &dev_attr_tx_packets.attr,
607         &dev_attr_rx_bytes.attr,
608         &dev_attr_tx_bytes.attr,
609         &dev_attr_rx_errors.attr,
610         &dev_attr_tx_errors.attr,
611         &dev_attr_rx_dropped.attr,
612         &dev_attr_tx_dropped.attr,
613         &dev_attr_multicast.attr,
614         &dev_attr_collisions.attr,
615         &dev_attr_rx_length_errors.attr,
616         &dev_attr_rx_over_errors.attr,
617         &dev_attr_rx_crc_errors.attr,
618         &dev_attr_rx_frame_errors.attr,
619         &dev_attr_rx_fifo_errors.attr,
620         &dev_attr_rx_missed_errors.attr,
621         &dev_attr_tx_aborted_errors.attr,
622         &dev_attr_tx_carrier_errors.attr,
623         &dev_attr_tx_fifo_errors.attr,
624         &dev_attr_tx_heartbeat_errors.attr,
625         &dev_attr_tx_window_errors.attr,
626         &dev_attr_rx_compressed.attr,
627         &dev_attr_tx_compressed.attr,
628         &dev_attr_rx_nohandler.attr,
629         NULL
630 };
631
632 static const struct attribute_group netstat_group = {
633         .name  = "statistics",
634         .attrs  = netstat_attrs,
635 };
636
637 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
638 static struct attribute *wireless_attrs[] = {
639         NULL
640 };
641
642 static const struct attribute_group wireless_group = {
643         .name = "wireless",
644         .attrs = wireless_attrs,
645 };
646 #endif
647
648 #else /* CONFIG_SYSFS */
649 #define net_class_groups        NULL
650 #endif /* CONFIG_SYSFS */
651
652 #ifdef CONFIG_SYSFS
653 #define to_rx_queue_attr(_attr) \
654         container_of(_attr, struct rx_queue_attribute, attr)
655
656 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
657
658 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
659                                   char *buf)
660 {
661         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
662         struct netdev_rx_queue *queue = to_rx_queue(kobj);
663
664         if (!attribute->show)
665                 return -EIO;
666
667         return attribute->show(queue, buf);
668 }
669
670 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
671                                    const char *buf, size_t count)
672 {
673         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
674         struct netdev_rx_queue *queue = to_rx_queue(kobj);
675
676         if (!attribute->store)
677                 return -EIO;
678
679         return attribute->store(queue, buf, count);
680 }
681
682 static const struct sysfs_ops rx_queue_sysfs_ops = {
683         .show = rx_queue_attr_show,
684         .store = rx_queue_attr_store,
685 };
686
687 #ifdef CONFIG_RPS
688 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
689 {
690         struct rps_map *map;
691         cpumask_var_t mask;
692         int i, len;
693
694         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
695                 return -ENOMEM;
696
697         rcu_read_lock();
698         map = rcu_dereference(queue->rps_map);
699         if (map)
700                 for (i = 0; i < map->len; i++)
701                         cpumask_set_cpu(map->cpus[i], mask);
702
703         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
704         rcu_read_unlock();
705         free_cpumask_var(mask);
706
707         return len < PAGE_SIZE ? len : -EINVAL;
708 }
709
710 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
711                              const char *buf, size_t len)
712 {
713         struct rps_map *old_map, *map;
714         cpumask_var_t mask;
715         int err, cpu, i;
716         static DEFINE_MUTEX(rps_map_mutex);
717
718         if (!capable(CAP_NET_ADMIN))
719                 return -EPERM;
720
721         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
722                 return -ENOMEM;
723
724         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
725         if (err) {
726                 free_cpumask_var(mask);
727                 return err;
728         }
729
730         map = kzalloc(max_t(unsigned int,
731                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
732                       GFP_KERNEL);
733         if (!map) {
734                 free_cpumask_var(mask);
735                 return -ENOMEM;
736         }
737
738         i = 0;
739         for_each_cpu_and(cpu, mask, cpu_online_mask)
740                 map->cpus[i++] = cpu;
741
742         if (i) {
743                 map->len = i;
744         } else {
745                 kfree(map);
746                 map = NULL;
747         }
748
749         mutex_lock(&rps_map_mutex);
750         old_map = rcu_dereference_protected(queue->rps_map,
751                                             mutex_is_locked(&rps_map_mutex));
752         rcu_assign_pointer(queue->rps_map, map);
753
754         if (map)
755                 static_key_slow_inc(&rps_needed);
756         if (old_map)
757                 static_key_slow_dec(&rps_needed);
758
759         mutex_unlock(&rps_map_mutex);
760
761         if (old_map)
762                 kfree_rcu(old_map, rcu);
763
764         free_cpumask_var(mask);
765         return len;
766 }
767
768 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
769                                            char *buf)
770 {
771         struct rps_dev_flow_table *flow_table;
772         unsigned long val = 0;
773
774         rcu_read_lock();
775         flow_table = rcu_dereference(queue->rps_flow_table);
776         if (flow_table)
777                 val = (unsigned long)flow_table->mask + 1;
778         rcu_read_unlock();
779
780         return sprintf(buf, "%lu\n", val);
781 }
782
783 static void rps_dev_flow_table_release(struct rcu_head *rcu)
784 {
785         struct rps_dev_flow_table *table = container_of(rcu,
786             struct rps_dev_flow_table, rcu);
787         vfree(table);
788 }
789
790 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
791                                             const char *buf, size_t len)
792 {
793         unsigned long mask, count;
794         struct rps_dev_flow_table *table, *old_table;
795         static DEFINE_SPINLOCK(rps_dev_flow_lock);
796         int rc;
797
798         if (!capable(CAP_NET_ADMIN))
799                 return -EPERM;
800
801         rc = kstrtoul(buf, 0, &count);
802         if (rc < 0)
803                 return rc;
804
805         if (count) {
806                 mask = count - 1;
807                 /* mask = roundup_pow_of_two(count) - 1;
808                  * without overflows...
809                  */
810                 while ((mask | (mask >> 1)) != mask)
811                         mask |= (mask >> 1);
812                 /* On 64 bit arches, must check mask fits in table->mask (u32),
813                  * and on 32bit arches, must check
814                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
815                  */
816 #if BITS_PER_LONG > 32
817                 if (mask > (unsigned long)(u32)mask)
818                         return -EINVAL;
819 #else
820                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
821                                 / sizeof(struct rps_dev_flow)) {
822                         /* Enforce a limit to prevent overflow */
823                         return -EINVAL;
824                 }
825 #endif
826                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
827                 if (!table)
828                         return -ENOMEM;
829
830                 table->mask = mask;
831                 for (count = 0; count <= mask; count++)
832                         table->flows[count].cpu = RPS_NO_CPU;
833         } else {
834                 table = NULL;
835         }
836
837         spin_lock(&rps_dev_flow_lock);
838         old_table = rcu_dereference_protected(queue->rps_flow_table,
839                                               lockdep_is_held(&rps_dev_flow_lock));
840         rcu_assign_pointer(queue->rps_flow_table, table);
841         spin_unlock(&rps_dev_flow_lock);
842
843         if (old_table)
844                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
845
846         return len;
847 }
848
849 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
850         = __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
851
852 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
853         = __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
854                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
855 #endif /* CONFIG_RPS */
856
857 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
858 #ifdef CONFIG_RPS
859         &rps_cpus_attribute.attr,
860         &rps_dev_flow_table_cnt_attribute.attr,
861 #endif
862         NULL
863 };
864
865 static void rx_queue_release(struct kobject *kobj)
866 {
867         struct netdev_rx_queue *queue = to_rx_queue(kobj);
868 #ifdef CONFIG_RPS
869         struct rps_map *map;
870         struct rps_dev_flow_table *flow_table;
871
872         map = rcu_dereference_protected(queue->rps_map, 1);
873         if (map) {
874                 RCU_INIT_POINTER(queue->rps_map, NULL);
875                 kfree_rcu(map, rcu);
876         }
877
878         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
879         if (flow_table) {
880                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
881                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
882         }
883 #endif
884
885         memset(kobj, 0, sizeof(*kobj));
886         dev_put(queue->dev);
887 }
888
889 static const void *rx_queue_namespace(struct kobject *kobj)
890 {
891         struct netdev_rx_queue *queue = to_rx_queue(kobj);
892         struct device *dev = &queue->dev->dev;
893         const void *ns = NULL;
894
895         if (dev->class && dev->class->ns_type)
896                 ns = dev->class->namespace(dev);
897
898         return ns;
899 }
900
901 static struct kobj_type rx_queue_ktype __ro_after_init = {
902         .sysfs_ops = &rx_queue_sysfs_ops,
903         .release = rx_queue_release,
904         .default_attrs = rx_queue_default_attrs,
905         .namespace = rx_queue_namespace
906 };
907
908 static int rx_queue_add_kobject(struct net_device *dev, int index)
909 {
910         struct netdev_rx_queue *queue = dev->_rx + index;
911         struct kobject *kobj = &queue->kobj;
912         int error = 0;
913
914         /* Kobject_put later will trigger rx_queue_release call which
915          * decreases dev refcount: Take that reference here
916          */
917         dev_hold(queue->dev);
918
919         kobj->kset = dev->queues_kset;
920         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
921                                      "rx-%u", index);
922         if (error)
923                 goto err;
924
925         if (dev->sysfs_rx_queue_group) {
926                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
927                 if (error)
928                         goto err;
929         }
930
931         kobject_uevent(kobj, KOBJ_ADD);
932
933         return error;
934
935 err:
936         kobject_put(kobj);
937         return error;
938 }
939 #endif /* CONFIG_SYSFS */
940
941 int
942 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
943 {
944 #ifdef CONFIG_SYSFS
945         int i;
946         int error = 0;
947
948 #ifndef CONFIG_RPS
949         if (!dev->sysfs_rx_queue_group)
950                 return 0;
951 #endif
952         for (i = old_num; i < new_num; i++) {
953                 error = rx_queue_add_kobject(dev, i);
954                 if (error) {
955                         new_num = old_num;
956                         break;
957                 }
958         }
959
960         while (--i >= new_num) {
961                 struct kobject *kobj = &dev->_rx[i].kobj;
962
963                 if (!atomic_read(&dev_net(dev)->count))
964                         kobj->uevent_suppress = 1;
965                 if (dev->sysfs_rx_queue_group)
966                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
967                 kobject_put(kobj);
968         }
969
970         return error;
971 #else
972         return 0;
973 #endif
974 }
975
976 #ifdef CONFIG_SYSFS
977 /*
978  * netdev_queue sysfs structures and functions.
979  */
980 struct netdev_queue_attribute {
981         struct attribute attr;
982         ssize_t (*show)(struct netdev_queue *queue, char *buf);
983         ssize_t (*store)(struct netdev_queue *queue,
984                          const char *buf, size_t len);
985 };
986 #define to_netdev_queue_attr(_attr) \
987         container_of(_attr, struct netdev_queue_attribute, attr)
988
989 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
990
991 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
992                                       struct attribute *attr, char *buf)
993 {
994         const struct netdev_queue_attribute *attribute
995                 = to_netdev_queue_attr(attr);
996         struct netdev_queue *queue = to_netdev_queue(kobj);
997
998         if (!attribute->show)
999                 return -EIO;
1000
1001         return attribute->show(queue, buf);
1002 }
1003
1004 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1005                                        struct attribute *attr,
1006                                        const char *buf, size_t count)
1007 {
1008         const struct netdev_queue_attribute *attribute
1009                 = to_netdev_queue_attr(attr);
1010         struct netdev_queue *queue = to_netdev_queue(kobj);
1011
1012         if (!attribute->store)
1013                 return -EIO;
1014
1015         return attribute->store(queue, buf, count);
1016 }
1017
1018 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1019         .show = netdev_queue_attr_show,
1020         .store = netdev_queue_attr_store,
1021 };
1022
1023 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1024 {
1025         unsigned long trans_timeout;
1026
1027         spin_lock_irq(&queue->_xmit_lock);
1028         trans_timeout = queue->trans_timeout;
1029         spin_unlock_irq(&queue->_xmit_lock);
1030
1031         return sprintf(buf, fmt_ulong, trans_timeout);
1032 }
1033
1034 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1035 {
1036         struct net_device *dev = queue->dev;
1037         unsigned int i;
1038
1039         i = queue - dev->_tx;
1040         BUG_ON(i >= dev->num_tx_queues);
1041
1042         return i;
1043 }
1044
1045 static ssize_t traffic_class_show(struct netdev_queue *queue,
1046                                   char *buf)
1047 {
1048         struct net_device *dev = queue->dev;
1049         int index = get_netdev_queue_index(queue);
1050         int tc = netdev_txq_to_tc(dev, index);
1051
1052         if (tc < 0)
1053                 return -EINVAL;
1054
1055         return sprintf(buf, "%u\n", tc);
1056 }
1057
1058 #ifdef CONFIG_XPS
1059 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1060                                char *buf)
1061 {
1062         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1063 }
1064
1065 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1066                                 const char *buf, size_t len)
1067 {
1068         struct net_device *dev = queue->dev;
1069         int err, index = get_netdev_queue_index(queue);
1070         u32 rate = 0;
1071
1072         err = kstrtou32(buf, 10, &rate);
1073         if (err < 0)
1074                 return err;
1075
1076         if (!rtnl_trylock())
1077                 return restart_syscall();
1078
1079         err = -EOPNOTSUPP;
1080         if (dev->netdev_ops->ndo_set_tx_maxrate)
1081                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1082
1083         rtnl_unlock();
1084         if (!err) {
1085                 queue->tx_maxrate = rate;
1086                 return len;
1087         }
1088         return err;
1089 }
1090
1091 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1092         = __ATTR_RW(tx_maxrate);
1093 #endif
1094
1095 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1096         = __ATTR_RO(tx_timeout);
1097
1098 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1099         = __ATTR_RO(traffic_class);
1100
1101 #ifdef CONFIG_BQL
1102 /*
1103  * Byte queue limits sysfs structures and functions.
1104  */
1105 static ssize_t bql_show(char *buf, unsigned int value)
1106 {
1107         return sprintf(buf, "%u\n", value);
1108 }
1109
1110 static ssize_t bql_set(const char *buf, const size_t count,
1111                        unsigned int *pvalue)
1112 {
1113         unsigned int value;
1114         int err;
1115
1116         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1117                 value = DQL_MAX_LIMIT;
1118         } else {
1119                 err = kstrtouint(buf, 10, &value);
1120                 if (err < 0)
1121                         return err;
1122                 if (value > DQL_MAX_LIMIT)
1123                         return -EINVAL;
1124         }
1125
1126         *pvalue = value;
1127
1128         return count;
1129 }
1130
1131 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1132                                   char *buf)
1133 {
1134         struct dql *dql = &queue->dql;
1135
1136         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1137 }
1138
1139 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1140                                  const char *buf, size_t len)
1141 {
1142         struct dql *dql = &queue->dql;
1143         unsigned int value;
1144         int err;
1145
1146         err = kstrtouint(buf, 10, &value);
1147         if (err < 0)
1148                 return err;
1149
1150         dql->slack_hold_time = msecs_to_jiffies(value);
1151
1152         return len;
1153 }
1154
1155 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1156         = __ATTR(hold_time, S_IRUGO | S_IWUSR,
1157                  bql_show_hold_time, bql_set_hold_time);
1158
1159 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1160                                  char *buf)
1161 {
1162         struct dql *dql = &queue->dql;
1163
1164         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1165 }
1166
1167 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1168         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1169
1170 #define BQL_ATTR(NAME, FIELD)                                           \
1171 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1172                                  char *buf)                             \
1173 {                                                                       \
1174         return bql_show(buf, queue->dql.FIELD);                         \
1175 }                                                                       \
1176                                                                         \
1177 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1178                                 const char *buf, size_t len)            \
1179 {                                                                       \
1180         return bql_set(buf, len, &queue->dql.FIELD);                    \
1181 }                                                                       \
1182                                                                         \
1183 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1184         = __ATTR(NAME, S_IRUGO | S_IWUSR,                               \
1185                  bql_show_ ## NAME, bql_set_ ## NAME)
1186
1187 BQL_ATTR(limit, limit);
1188 BQL_ATTR(limit_max, max_limit);
1189 BQL_ATTR(limit_min, min_limit);
1190
1191 static struct attribute *dql_attrs[] __ro_after_init = {
1192         &bql_limit_attribute.attr,
1193         &bql_limit_max_attribute.attr,
1194         &bql_limit_min_attribute.attr,
1195         &bql_hold_time_attribute.attr,
1196         &bql_inflight_attribute.attr,
1197         NULL
1198 };
1199
1200 static const struct attribute_group dql_group = {
1201         .name  = "byte_queue_limits",
1202         .attrs  = dql_attrs,
1203 };
1204 #endif /* CONFIG_BQL */
1205
1206 #ifdef CONFIG_XPS
1207 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1208                              char *buf)
1209 {
1210         int cpu, len, ret, num_tc = 1, tc = 0;
1211         struct net_device *dev = queue->dev;
1212         struct xps_dev_maps *dev_maps;
1213         cpumask_var_t mask;
1214         unsigned long index;
1215
1216         index = get_netdev_queue_index(queue);
1217
1218         if (!rtnl_trylock())
1219                 return restart_syscall();
1220
1221         if (dev->num_tc) {
1222                 num_tc = dev->num_tc;
1223                 tc = netdev_txq_to_tc(dev, index);
1224                 if (tc < 0) {
1225                         ret = -EINVAL;
1226                         goto err_rtnl_unlock;
1227                 }
1228         }
1229
1230         if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1231                 ret = -ENOMEM;
1232                 goto err_rtnl_unlock;
1233         }
1234
1235         rcu_read_lock();
1236         dev_maps = rcu_dereference(dev->xps_maps);
1237         if (dev_maps) {
1238                 for_each_possible_cpu(cpu) {
1239                         int i, tci = cpu * num_tc + tc;
1240                         struct xps_map *map;
1241
1242                         map = rcu_dereference(dev_maps->cpu_map[tci]);
1243                         if (!map)
1244                                 continue;
1245
1246                         for (i = map->len; i--;) {
1247                                 if (map->queues[i] == index) {
1248                                         cpumask_set_cpu(cpu, mask);
1249                                         break;
1250                                 }
1251                         }
1252                 }
1253         }
1254         rcu_read_unlock();
1255
1256         rtnl_unlock();
1257
1258         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1259         free_cpumask_var(mask);
1260         return len < PAGE_SIZE ? len : -EINVAL;
1261
1262 err_rtnl_unlock:
1263         rtnl_unlock();
1264         return ret;
1265 }
1266
1267 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1268                               const char *buf, size_t len)
1269 {
1270         struct net_device *dev = queue->dev;
1271         unsigned long index;
1272         cpumask_var_t mask;
1273         int err;
1274
1275         if (!capable(CAP_NET_ADMIN))
1276                 return -EPERM;
1277
1278         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1279                 return -ENOMEM;
1280
1281         index = get_netdev_queue_index(queue);
1282
1283         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1284         if (err) {
1285                 free_cpumask_var(mask);
1286                 return err;
1287         }
1288
1289         if (!rtnl_trylock()) {
1290                 free_cpumask_var(mask);
1291                 return restart_syscall();
1292         }
1293
1294         err = netif_set_xps_queue(dev, mask, index);
1295         rtnl_unlock();
1296
1297         free_cpumask_var(mask);
1298
1299         return err ? : len;
1300 }
1301
1302 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1303         = __ATTR_RW(xps_cpus);
1304 #endif /* CONFIG_XPS */
1305
1306 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1307         &queue_trans_timeout.attr,
1308         &queue_traffic_class.attr,
1309 #ifdef CONFIG_XPS
1310         &xps_cpus_attribute.attr,
1311         &queue_tx_maxrate.attr,
1312 #endif
1313         NULL
1314 };
1315
1316 static void netdev_queue_release(struct kobject *kobj)
1317 {
1318         struct netdev_queue *queue = to_netdev_queue(kobj);
1319
1320         memset(kobj, 0, sizeof(*kobj));
1321         dev_put(queue->dev);
1322 }
1323
1324 static const void *netdev_queue_namespace(struct kobject *kobj)
1325 {
1326         struct netdev_queue *queue = to_netdev_queue(kobj);
1327         struct device *dev = &queue->dev->dev;
1328         const void *ns = NULL;
1329
1330         if (dev->class && dev->class->ns_type)
1331                 ns = dev->class->namespace(dev);
1332
1333         return ns;
1334 }
1335
1336 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1337         .sysfs_ops = &netdev_queue_sysfs_ops,
1338         .release = netdev_queue_release,
1339         .default_attrs = netdev_queue_default_attrs,
1340         .namespace = netdev_queue_namespace,
1341 };
1342
1343 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1344 {
1345         struct netdev_queue *queue = dev->_tx + index;
1346         struct kobject *kobj = &queue->kobj;
1347         int error = 0;
1348
1349         /* Kobject_put later will trigger netdev_queue_release call
1350          * which decreases dev refcount: Take that reference here
1351          */
1352         dev_hold(queue->dev);
1353
1354         kobj->kset = dev->queues_kset;
1355         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1356                                      "tx-%u", index);
1357         if (error)
1358                 goto err;
1359
1360 #ifdef CONFIG_BQL
1361         error = sysfs_create_group(kobj, &dql_group);
1362         if (error)
1363                 goto err;
1364 #endif
1365
1366         kobject_uevent(kobj, KOBJ_ADD);
1367         return 0;
1368
1369 err:
1370         kobject_put(kobj);
1371         return error;
1372 }
1373 #endif /* CONFIG_SYSFS */
1374
1375 int
1376 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1377 {
1378 #ifdef CONFIG_SYSFS
1379         int i;
1380         int error = 0;
1381
1382         for (i = old_num; i < new_num; i++) {
1383                 error = netdev_queue_add_kobject(dev, i);
1384                 if (error) {
1385                         new_num = old_num;
1386                         break;
1387                 }
1388         }
1389
1390         while (--i >= new_num) {
1391                 struct netdev_queue *queue = dev->_tx + i;
1392
1393                 if (!atomic_read(&dev_net(dev)->count))
1394                         queue->kobj.uevent_suppress = 1;
1395 #ifdef CONFIG_BQL
1396                 sysfs_remove_group(&queue->kobj, &dql_group);
1397 #endif
1398                 kobject_put(&queue->kobj);
1399         }
1400
1401         return error;
1402 #else
1403         return 0;
1404 #endif /* CONFIG_SYSFS */
1405 }
1406
1407 static int register_queue_kobjects(struct net_device *dev)
1408 {
1409         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1410
1411 #ifdef CONFIG_SYSFS
1412         dev->queues_kset = kset_create_and_add("queues",
1413                                                NULL, &dev->dev.kobj);
1414         if (!dev->queues_kset)
1415                 return -ENOMEM;
1416         real_rx = dev->real_num_rx_queues;
1417 #endif
1418         real_tx = dev->real_num_tx_queues;
1419
1420         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1421         if (error)
1422                 goto error;
1423         rxq = real_rx;
1424
1425         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1426         if (error)
1427                 goto error;
1428         txq = real_tx;
1429
1430         return 0;
1431
1432 error:
1433         netdev_queue_update_kobjects(dev, txq, 0);
1434         net_rx_queue_update_kobjects(dev, rxq, 0);
1435 #ifdef CONFIG_SYSFS
1436         kset_unregister(dev->queues_kset);
1437 #endif
1438         return error;
1439 }
1440
1441 static void remove_queue_kobjects(struct net_device *dev)
1442 {
1443         int real_rx = 0, real_tx = 0;
1444
1445 #ifdef CONFIG_SYSFS
1446         real_rx = dev->real_num_rx_queues;
1447 #endif
1448         real_tx = dev->real_num_tx_queues;
1449
1450         net_rx_queue_update_kobjects(dev, real_rx, 0);
1451         netdev_queue_update_kobjects(dev, real_tx, 0);
1452 #ifdef CONFIG_SYSFS
1453         kset_unregister(dev->queues_kset);
1454 #endif
1455 }
1456
1457 static bool net_current_may_mount(void)
1458 {
1459         struct net *net = current->nsproxy->net_ns;
1460
1461         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1462 }
1463
1464 static void *net_grab_current_ns(void)
1465 {
1466         struct net *ns = current->nsproxy->net_ns;
1467 #ifdef CONFIG_NET_NS
1468         if (ns)
1469                 refcount_inc(&ns->passive);
1470 #endif
1471         return ns;
1472 }
1473
1474 static const void *net_initial_ns(void)
1475 {
1476         return &init_net;
1477 }
1478
1479 static const void *net_netlink_ns(struct sock *sk)
1480 {
1481         return sock_net(sk);
1482 }
1483
1484 const struct kobj_ns_type_operations net_ns_type_operations = {
1485         .type = KOBJ_NS_TYPE_NET,
1486         .current_may_mount = net_current_may_mount,
1487         .grab_current_ns = net_grab_current_ns,
1488         .netlink_ns = net_netlink_ns,
1489         .initial_ns = net_initial_ns,
1490         .drop_ns = net_drop_ns,
1491 };
1492 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1493
1494 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1495 {
1496         struct net_device *dev = to_net_dev(d);
1497         int retval;
1498
1499         /* pass interface to uevent. */
1500         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1501         if (retval)
1502                 goto exit;
1503
1504         /* pass ifindex to uevent.
1505          * ifindex is useful as it won't change (interface name may change)
1506          * and is what RtNetlink uses natively.
1507          */
1508         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1509
1510 exit:
1511         return retval;
1512 }
1513
1514 /*
1515  *      netdev_release -- destroy and free a dead device.
1516  *      Called when last reference to device kobject is gone.
1517  */
1518 static void netdev_release(struct device *d)
1519 {
1520         struct net_device *dev = to_net_dev(d);
1521
1522         BUG_ON(dev->reg_state != NETREG_RELEASED);
1523
1524         kfree(dev->ifalias);
1525         netdev_freemem(dev);
1526 }
1527
1528 static const void *net_namespace(struct device *d)
1529 {
1530         struct net_device *dev = to_net_dev(d);
1531
1532         return dev_net(dev);
1533 }
1534
1535 static struct class net_class __ro_after_init = {
1536         .name = "net",
1537         .dev_release = netdev_release,
1538         .dev_groups = net_class_groups,
1539         .dev_uevent = netdev_uevent,
1540         .ns_type = &net_ns_type_operations,
1541         .namespace = net_namespace,
1542 };
1543
1544 #ifdef CONFIG_OF_NET
1545 static int of_dev_node_match(struct device *dev, const void *data)
1546 {
1547         int ret = 0;
1548
1549         if (dev->parent)
1550                 ret = dev->parent->of_node == data;
1551
1552         return ret == 0 ? dev->of_node == data : ret;
1553 }
1554
1555 /*
1556  * of_find_net_device_by_node - lookup the net device for the device node
1557  * @np: OF device node
1558  *
1559  * Looks up the net_device structure corresponding with the device node.
1560  * If successful, returns a pointer to the net_device with the embedded
1561  * struct device refcount incremented by one, or NULL on failure. The
1562  * refcount must be dropped when done with the net_device.
1563  */
1564 struct net_device *of_find_net_device_by_node(struct device_node *np)
1565 {
1566         struct device *dev;
1567
1568         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1569         if (!dev)
1570                 return NULL;
1571
1572         return to_net_dev(dev);
1573 }
1574 EXPORT_SYMBOL(of_find_net_device_by_node);
1575 #endif
1576
1577 /* Delete sysfs entries but hold kobject reference until after all
1578  * netdev references are gone.
1579  */
1580 void netdev_unregister_kobject(struct net_device *ndev)
1581 {
1582         struct device *dev = &ndev->dev;
1583
1584         if (!atomic_read(&dev_net(ndev)->count))
1585                 dev_set_uevent_suppress(dev, 1);
1586
1587         kobject_get(&dev->kobj);
1588
1589         remove_queue_kobjects(ndev);
1590
1591         pm_runtime_set_memalloc_noio(dev, false);
1592
1593         device_del(dev);
1594 }
1595
1596 /* Create sysfs entries for network device. */
1597 int netdev_register_kobject(struct net_device *ndev)
1598 {
1599         struct device *dev = &ndev->dev;
1600         const struct attribute_group **groups = ndev->sysfs_groups;
1601         int error = 0;
1602
1603         device_initialize(dev);
1604         dev->class = &net_class;
1605         dev->platform_data = ndev;
1606         dev->groups = groups;
1607
1608         dev_set_name(dev, "%s", ndev->name);
1609
1610 #ifdef CONFIG_SYSFS
1611         /* Allow for a device specific group */
1612         if (*groups)
1613                 groups++;
1614
1615         *groups++ = &netstat_group;
1616
1617 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1618         if (ndev->ieee80211_ptr)
1619                 *groups++ = &wireless_group;
1620 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1621         else if (ndev->wireless_handlers)
1622                 *groups++ = &wireless_group;
1623 #endif
1624 #endif
1625 #endif /* CONFIG_SYSFS */
1626
1627         error = device_add(dev);
1628         if (error)
1629                 return error;
1630
1631         error = register_queue_kobjects(ndev);
1632         if (error) {
1633                 device_del(dev);
1634                 return error;
1635         }
1636
1637         pm_runtime_set_memalloc_noio(dev, true);
1638
1639         return error;
1640 }
1641
1642 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1643                                 const void *ns)
1644 {
1645         return class_create_file_ns(&net_class, class_attr, ns);
1646 }
1647 EXPORT_SYMBOL(netdev_class_create_file_ns);
1648
1649 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1650                                  const void *ns)
1651 {
1652         class_remove_file_ns(&net_class, class_attr, ns);
1653 }
1654 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1655
1656 int __init netdev_kobject_init(void)
1657 {
1658         kobj_ns_type_register(&net_ns_type_operations);
1659         return class_register(&net_class);
1660 }