Merge tag 'pci-v4.7-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
[cascardo/linux.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/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 #include <linux/of_net.h>
28
29 #include "net-sysfs.h"
30
31 #ifdef CONFIG_SYSFS
32 static const char fmt_hex[] = "%#x\n";
33 static const char fmt_dec[] = "%d\n";
34 static const char fmt_ulong[] = "%lu\n";
35 static const char fmt_u64[] = "%llu\n";
36
37 static 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 */
43 static 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 *ndev = to_net_dev(dev);
48         ssize_t ret = -EINVAL;
49
50         read_lock(&dev_base_lock);
51         if (dev_isalive(ndev))
52                 ret = (*format)(ndev, 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)                            \
60 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
61 {                                                                       \
62         return sprintf(buf, format_string, dev->field);                 \
63 }                                                                       \
64 static ssize_t field##_show(struct device *dev,                         \
65                             struct device_attribute *attr, char *buf)   \
66 {                                                                       \
67         return netdev_show(dev, attr, buf, format_##field);             \
68 }                                                                       \
69
70 #define NETDEVICE_SHOW_RO(field, format_string)                         \
71 NETDEVICE_SHOW(field, format_string);                                   \
72 static DEVICE_ATTR_RO(field)
73
74 #define NETDEVICE_SHOW_RW(field, format_string)                         \
75 NETDEVICE_SHOW(field, format_string);                                   \
76 static DEVICE_ATTR_RW(field)
77
78 /* use same locking and permission rules as SIF* ioctl's */
79 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
80                             const char *buf, size_t len,
81                             int (*set)(struct net_device *, unsigned long))
82 {
83         struct net_device *netdev = to_net_dev(dev);
84         struct net *net = dev_net(netdev);
85         unsigned long new;
86         int ret = -EINVAL;
87
88         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
89                 return -EPERM;
90
91         ret = kstrtoul(buf, 0, &new);
92         if (ret)
93                 goto err;
94
95         if (!rtnl_trylock())
96                 return restart_syscall();
97
98         if (dev_isalive(netdev)) {
99                 if ((ret = (*set)(netdev, new)) == 0)
100                         ret = len;
101         }
102         rtnl_unlock();
103  err:
104         return ret;
105 }
106
107 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
108 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
109 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
110 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
111 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
112 NETDEVICE_SHOW_RO(type, fmt_dec);
113 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
114
115 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
116                            char *buf)
117 {
118         struct net_device *ndev = to_net_dev(dev);
119
120         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
121 }
122 static DEVICE_ATTR_RO(iflink);
123
124 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
125 {
126         return sprintf(buf, fmt_dec, dev->name_assign_type);
127 }
128
129 static ssize_t name_assign_type_show(struct device *dev,
130                                      struct device_attribute *attr,
131                                      char *buf)
132 {
133         struct net_device *ndev = to_net_dev(dev);
134         ssize_t ret = -EINVAL;
135
136         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
137                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
138
139         return ret;
140 }
141 static DEVICE_ATTR_RO(name_assign_type);
142
143 /* use same locking rules as GIFHWADDR ioctl's */
144 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
145                             char *buf)
146 {
147         struct net_device *ndev = to_net_dev(dev);
148         ssize_t ret = -EINVAL;
149
150         read_lock(&dev_base_lock);
151         if (dev_isalive(ndev))
152                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
153         read_unlock(&dev_base_lock);
154         return ret;
155 }
156 static DEVICE_ATTR_RO(address);
157
158 static ssize_t broadcast_show(struct device *dev,
159                               struct device_attribute *attr, char *buf)
160 {
161         struct net_device *ndev = to_net_dev(dev);
162         if (dev_isalive(ndev))
163                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
164         return -EINVAL;
165 }
166 static DEVICE_ATTR_RO(broadcast);
167
168 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
169 {
170         if (!netif_running(dev))
171                 return -EINVAL;
172         return dev_change_carrier(dev, (bool) new_carrier);
173 }
174
175 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
176                              const char *buf, size_t len)
177 {
178         return netdev_store(dev, attr, buf, len, change_carrier);
179 }
180
181 static ssize_t carrier_show(struct device *dev,
182                             struct device_attribute *attr, char *buf)
183 {
184         struct net_device *netdev = to_net_dev(dev);
185         if (netif_running(netdev)) {
186                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
187         }
188         return -EINVAL;
189 }
190 static DEVICE_ATTR_RW(carrier);
191
192 static ssize_t speed_show(struct device *dev,
193                           struct device_attribute *attr, char *buf)
194 {
195         struct net_device *netdev = to_net_dev(dev);
196         int ret = -EINVAL;
197
198         if (!rtnl_trylock())
199                 return restart_syscall();
200
201         if (netif_running(netdev)) {
202                 struct ethtool_link_ksettings cmd;
203
204                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
205                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
206         }
207         rtnl_unlock();
208         return ret;
209 }
210 static DEVICE_ATTR_RO(speed);
211
212 static ssize_t duplex_show(struct device *dev,
213                            struct device_attribute *attr, char *buf)
214 {
215         struct net_device *netdev = to_net_dev(dev);
216         int ret = -EINVAL;
217
218         if (!rtnl_trylock())
219                 return restart_syscall();
220
221         if (netif_running(netdev)) {
222                 struct ethtool_link_ksettings cmd;
223
224                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
225                         const char *duplex;
226
227                         switch (cmd.base.duplex) {
228                         case DUPLEX_HALF:
229                                 duplex = "half";
230                                 break;
231                         case DUPLEX_FULL:
232                                 duplex = "full";
233                                 break;
234                         default:
235                                 duplex = "unknown";
236                                 break;
237                         }
238                         ret = sprintf(buf, "%s\n", duplex);
239                 }
240         }
241         rtnl_unlock();
242         return ret;
243 }
244 static DEVICE_ATTR_RO(duplex);
245
246 static ssize_t dormant_show(struct device *dev,
247                             struct device_attribute *attr, char *buf)
248 {
249         struct net_device *netdev = to_net_dev(dev);
250
251         if (netif_running(netdev))
252                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
253
254         return -EINVAL;
255 }
256 static DEVICE_ATTR_RO(dormant);
257
258 static const char *const operstates[] = {
259         "unknown",
260         "notpresent", /* currently unused */
261         "down",
262         "lowerlayerdown",
263         "testing", /* currently unused */
264         "dormant",
265         "up"
266 };
267
268 static ssize_t operstate_show(struct device *dev,
269                               struct device_attribute *attr, char *buf)
270 {
271         const struct net_device *netdev = to_net_dev(dev);
272         unsigned char operstate;
273
274         read_lock(&dev_base_lock);
275         operstate = netdev->operstate;
276         if (!netif_running(netdev))
277                 operstate = IF_OPER_DOWN;
278         read_unlock(&dev_base_lock);
279
280         if (operstate >= ARRAY_SIZE(operstates))
281                 return -EINVAL; /* should not happen */
282
283         return sprintf(buf, "%s\n", operstates[operstate]);
284 }
285 static DEVICE_ATTR_RO(operstate);
286
287 static ssize_t carrier_changes_show(struct device *dev,
288                                     struct device_attribute *attr,
289                                     char *buf)
290 {
291         struct net_device *netdev = to_net_dev(dev);
292         return sprintf(buf, fmt_dec,
293                        atomic_read(&netdev->carrier_changes));
294 }
295 static DEVICE_ATTR_RO(carrier_changes);
296
297 /* read-write attributes */
298
299 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
300 {
301         return dev_set_mtu(dev, (int) new_mtu);
302 }
303
304 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
305                          const char *buf, size_t len)
306 {
307         return netdev_store(dev, attr, buf, len, change_mtu);
308 }
309 NETDEVICE_SHOW_RW(mtu, fmt_dec);
310
311 static int change_flags(struct net_device *dev, unsigned long new_flags)
312 {
313         return dev_change_flags(dev, (unsigned int) new_flags);
314 }
315
316 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
317                            const char *buf, size_t len)
318 {
319         return netdev_store(dev, attr, buf, len, change_flags);
320 }
321 NETDEVICE_SHOW_RW(flags, fmt_hex);
322
323 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
324 {
325         dev->tx_queue_len = new_len;
326         return 0;
327 }
328
329 static ssize_t tx_queue_len_store(struct device *dev,
330                                   struct device_attribute *attr,
331                                   const char *buf, size_t len)
332 {
333         if (!capable(CAP_NET_ADMIN))
334                 return -EPERM;
335
336         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
337 }
338 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
339
340 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
341 {
342         dev->gro_flush_timeout = val;
343         return 0;
344 }
345
346 static ssize_t gro_flush_timeout_store(struct device *dev,
347                                   struct device_attribute *attr,
348                                   const char *buf, size_t len)
349 {
350         if (!capable(CAP_NET_ADMIN))
351                 return -EPERM;
352
353         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
354 }
355 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
356
357 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
358                              const char *buf, size_t len)
359 {
360         struct net_device *netdev = to_net_dev(dev);
361         struct net *net = dev_net(netdev);
362         size_t count = len;
363         ssize_t ret;
364
365         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
366                 return -EPERM;
367
368         /* ignore trailing newline */
369         if (len >  0 && buf[len - 1] == '\n')
370                 --count;
371
372         if (!rtnl_trylock())
373                 return restart_syscall();
374         ret = dev_set_alias(netdev, buf, count);
375         rtnl_unlock();
376
377         return ret < 0 ? ret : len;
378 }
379
380 static ssize_t ifalias_show(struct device *dev,
381                             struct device_attribute *attr, char *buf)
382 {
383         const struct net_device *netdev = to_net_dev(dev);
384         ssize_t ret = 0;
385
386         if (!rtnl_trylock())
387                 return restart_syscall();
388         if (netdev->ifalias)
389                 ret = sprintf(buf, "%s\n", netdev->ifalias);
390         rtnl_unlock();
391         return ret;
392 }
393 static DEVICE_ATTR_RW(ifalias);
394
395 static int change_group(struct net_device *dev, unsigned long new_group)
396 {
397         dev_set_group(dev, (int) new_group);
398         return 0;
399 }
400
401 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
402                            const char *buf, size_t len)
403 {
404         return netdev_store(dev, attr, buf, len, change_group);
405 }
406 NETDEVICE_SHOW(group, fmt_dec);
407 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
408
409 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
410 {
411         return dev_change_proto_down(dev, (bool) proto_down);
412 }
413
414 static ssize_t proto_down_store(struct device *dev,
415                                 struct device_attribute *attr,
416                                 const char *buf, size_t len)
417 {
418         return netdev_store(dev, attr, buf, len, change_proto_down);
419 }
420 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
421
422 static ssize_t phys_port_id_show(struct device *dev,
423                                  struct device_attribute *attr, char *buf)
424 {
425         struct net_device *netdev = to_net_dev(dev);
426         ssize_t ret = -EINVAL;
427
428         if (!rtnl_trylock())
429                 return restart_syscall();
430
431         if (dev_isalive(netdev)) {
432                 struct netdev_phys_item_id ppid;
433
434                 ret = dev_get_phys_port_id(netdev, &ppid);
435                 if (!ret)
436                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
437         }
438         rtnl_unlock();
439
440         return ret;
441 }
442 static DEVICE_ATTR_RO(phys_port_id);
443
444 static ssize_t phys_port_name_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                 char name[IFNAMSIZ];
455
456                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
457                 if (!ret)
458                         ret = sprintf(buf, "%s\n", name);
459         }
460         rtnl_unlock();
461
462         return ret;
463 }
464 static DEVICE_ATTR_RO(phys_port_name);
465
466 static ssize_t phys_switch_id_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                 struct switchdev_attr attr = {
477                         .orig_dev = netdev,
478                         .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
479                         .flags = SWITCHDEV_F_NO_RECURSE,
480                 };
481
482                 ret = switchdev_port_attr_get(netdev, &attr);
483                 if (!ret)
484                         ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
485                                       attr.u.ppid.id);
486         }
487         rtnl_unlock();
488
489         return ret;
490 }
491 static DEVICE_ATTR_RO(phys_switch_id);
492
493 static struct attribute *net_class_attrs[] = {
494         &dev_attr_netdev_group.attr,
495         &dev_attr_type.attr,
496         &dev_attr_dev_id.attr,
497         &dev_attr_dev_port.attr,
498         &dev_attr_iflink.attr,
499         &dev_attr_ifindex.attr,
500         &dev_attr_name_assign_type.attr,
501         &dev_attr_addr_assign_type.attr,
502         &dev_attr_addr_len.attr,
503         &dev_attr_link_mode.attr,
504         &dev_attr_address.attr,
505         &dev_attr_broadcast.attr,
506         &dev_attr_speed.attr,
507         &dev_attr_duplex.attr,
508         &dev_attr_dormant.attr,
509         &dev_attr_operstate.attr,
510         &dev_attr_carrier_changes.attr,
511         &dev_attr_ifalias.attr,
512         &dev_attr_carrier.attr,
513         &dev_attr_mtu.attr,
514         &dev_attr_flags.attr,
515         &dev_attr_tx_queue_len.attr,
516         &dev_attr_gro_flush_timeout.attr,
517         &dev_attr_phys_port_id.attr,
518         &dev_attr_phys_port_name.attr,
519         &dev_attr_phys_switch_id.attr,
520         &dev_attr_proto_down.attr,
521         NULL,
522 };
523 ATTRIBUTE_GROUPS(net_class);
524
525 /* Show a given an attribute in the statistics group */
526 static ssize_t netstat_show(const struct device *d,
527                             struct device_attribute *attr, char *buf,
528                             unsigned long offset)
529 {
530         struct net_device *dev = to_net_dev(d);
531         ssize_t ret = -EINVAL;
532
533         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
534                         offset % sizeof(u64) != 0);
535
536         read_lock(&dev_base_lock);
537         if (dev_isalive(dev)) {
538                 struct rtnl_link_stats64 temp;
539                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
540
541                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
542         }
543         read_unlock(&dev_base_lock);
544         return ret;
545 }
546
547 /* generate a read-only statistics attribute */
548 #define NETSTAT_ENTRY(name)                                             \
549 static ssize_t name##_show(struct device *d,                            \
550                            struct device_attribute *attr, char *buf)    \
551 {                                                                       \
552         return netstat_show(d, attr, buf,                               \
553                             offsetof(struct rtnl_link_stats64, name));  \
554 }                                                                       \
555 static DEVICE_ATTR_RO(name)
556
557 NETSTAT_ENTRY(rx_packets);
558 NETSTAT_ENTRY(tx_packets);
559 NETSTAT_ENTRY(rx_bytes);
560 NETSTAT_ENTRY(tx_bytes);
561 NETSTAT_ENTRY(rx_errors);
562 NETSTAT_ENTRY(tx_errors);
563 NETSTAT_ENTRY(rx_dropped);
564 NETSTAT_ENTRY(tx_dropped);
565 NETSTAT_ENTRY(multicast);
566 NETSTAT_ENTRY(collisions);
567 NETSTAT_ENTRY(rx_length_errors);
568 NETSTAT_ENTRY(rx_over_errors);
569 NETSTAT_ENTRY(rx_crc_errors);
570 NETSTAT_ENTRY(rx_frame_errors);
571 NETSTAT_ENTRY(rx_fifo_errors);
572 NETSTAT_ENTRY(rx_missed_errors);
573 NETSTAT_ENTRY(tx_aborted_errors);
574 NETSTAT_ENTRY(tx_carrier_errors);
575 NETSTAT_ENTRY(tx_fifo_errors);
576 NETSTAT_ENTRY(tx_heartbeat_errors);
577 NETSTAT_ENTRY(tx_window_errors);
578 NETSTAT_ENTRY(rx_compressed);
579 NETSTAT_ENTRY(tx_compressed);
580 NETSTAT_ENTRY(rx_nohandler);
581
582 static struct attribute *netstat_attrs[] = {
583         &dev_attr_rx_packets.attr,
584         &dev_attr_tx_packets.attr,
585         &dev_attr_rx_bytes.attr,
586         &dev_attr_tx_bytes.attr,
587         &dev_attr_rx_errors.attr,
588         &dev_attr_tx_errors.attr,
589         &dev_attr_rx_dropped.attr,
590         &dev_attr_tx_dropped.attr,
591         &dev_attr_multicast.attr,
592         &dev_attr_collisions.attr,
593         &dev_attr_rx_length_errors.attr,
594         &dev_attr_rx_over_errors.attr,
595         &dev_attr_rx_crc_errors.attr,
596         &dev_attr_rx_frame_errors.attr,
597         &dev_attr_rx_fifo_errors.attr,
598         &dev_attr_rx_missed_errors.attr,
599         &dev_attr_tx_aborted_errors.attr,
600         &dev_attr_tx_carrier_errors.attr,
601         &dev_attr_tx_fifo_errors.attr,
602         &dev_attr_tx_heartbeat_errors.attr,
603         &dev_attr_tx_window_errors.attr,
604         &dev_attr_rx_compressed.attr,
605         &dev_attr_tx_compressed.attr,
606         &dev_attr_rx_nohandler.attr,
607         NULL
608 };
609
610
611 static struct attribute_group netstat_group = {
612         .name  = "statistics",
613         .attrs  = netstat_attrs,
614 };
615
616 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
617 static struct attribute *wireless_attrs[] = {
618         NULL
619 };
620
621 static struct attribute_group wireless_group = {
622         .name = "wireless",
623         .attrs = wireless_attrs,
624 };
625 #endif
626
627 #else /* CONFIG_SYSFS */
628 #define net_class_groups        NULL
629 #endif /* CONFIG_SYSFS */
630
631 #ifdef CONFIG_SYSFS
632 #define to_rx_queue_attr(_attr) container_of(_attr,             \
633     struct rx_queue_attribute, attr)
634
635 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
636
637 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
638                                   char *buf)
639 {
640         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
641         struct netdev_rx_queue *queue = to_rx_queue(kobj);
642
643         if (!attribute->show)
644                 return -EIO;
645
646         return attribute->show(queue, attribute, buf);
647 }
648
649 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
650                                    const char *buf, size_t count)
651 {
652         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
653         struct netdev_rx_queue *queue = to_rx_queue(kobj);
654
655         if (!attribute->store)
656                 return -EIO;
657
658         return attribute->store(queue, attribute, buf, count);
659 }
660
661 static const struct sysfs_ops rx_queue_sysfs_ops = {
662         .show = rx_queue_attr_show,
663         .store = rx_queue_attr_store,
664 };
665
666 #ifdef CONFIG_RPS
667 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
668                             struct rx_queue_attribute *attribute, char *buf)
669 {
670         struct rps_map *map;
671         cpumask_var_t mask;
672         int i, len;
673
674         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
675                 return -ENOMEM;
676
677         rcu_read_lock();
678         map = rcu_dereference(queue->rps_map);
679         if (map)
680                 for (i = 0; i < map->len; i++)
681                         cpumask_set_cpu(map->cpus[i], mask);
682
683         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
684         rcu_read_unlock();
685         free_cpumask_var(mask);
686
687         return len < PAGE_SIZE ? len : -EINVAL;
688 }
689
690 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
691                       struct rx_queue_attribute *attribute,
692                       const char *buf, size_t len)
693 {
694         struct rps_map *old_map, *map;
695         cpumask_var_t mask;
696         int err, cpu, i;
697         static DEFINE_MUTEX(rps_map_mutex);
698
699         if (!capable(CAP_NET_ADMIN))
700                 return -EPERM;
701
702         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
703                 return -ENOMEM;
704
705         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
706         if (err) {
707                 free_cpumask_var(mask);
708                 return err;
709         }
710
711         map = kzalloc(max_t(unsigned int,
712             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
713             GFP_KERNEL);
714         if (!map) {
715                 free_cpumask_var(mask);
716                 return -ENOMEM;
717         }
718
719         i = 0;
720         for_each_cpu_and(cpu, mask, cpu_online_mask)
721                 map->cpus[i++] = cpu;
722
723         if (i)
724                 map->len = i;
725         else {
726                 kfree(map);
727                 map = NULL;
728         }
729
730         mutex_lock(&rps_map_mutex);
731         old_map = rcu_dereference_protected(queue->rps_map,
732                                             mutex_is_locked(&rps_map_mutex));
733         rcu_assign_pointer(queue->rps_map, map);
734
735         if (map)
736                 static_key_slow_inc(&rps_needed);
737         if (old_map)
738                 static_key_slow_dec(&rps_needed);
739
740         mutex_unlock(&rps_map_mutex);
741
742         if (old_map)
743                 kfree_rcu(old_map, rcu);
744
745         free_cpumask_var(mask);
746         return len;
747 }
748
749 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
750                                            struct rx_queue_attribute *attr,
751                                            char *buf)
752 {
753         struct rps_dev_flow_table *flow_table;
754         unsigned long val = 0;
755
756         rcu_read_lock();
757         flow_table = rcu_dereference(queue->rps_flow_table);
758         if (flow_table)
759                 val = (unsigned long)flow_table->mask + 1;
760         rcu_read_unlock();
761
762         return sprintf(buf, "%lu\n", val);
763 }
764
765 static void rps_dev_flow_table_release(struct rcu_head *rcu)
766 {
767         struct rps_dev_flow_table *table = container_of(rcu,
768             struct rps_dev_flow_table, rcu);
769         vfree(table);
770 }
771
772 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
773                                      struct rx_queue_attribute *attr,
774                                      const char *buf, size_t len)
775 {
776         unsigned long mask, count;
777         struct rps_dev_flow_table *table, *old_table;
778         static DEFINE_SPINLOCK(rps_dev_flow_lock);
779         int rc;
780
781         if (!capable(CAP_NET_ADMIN))
782                 return -EPERM;
783
784         rc = kstrtoul(buf, 0, &count);
785         if (rc < 0)
786                 return rc;
787
788         if (count) {
789                 mask = count - 1;
790                 /* mask = roundup_pow_of_two(count) - 1;
791                  * without overflows...
792                  */
793                 while ((mask | (mask >> 1)) != mask)
794                         mask |= (mask >> 1);
795                 /* On 64 bit arches, must check mask fits in table->mask (u32),
796                  * and on 32bit arches, must check
797                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
798                  */
799 #if BITS_PER_LONG > 32
800                 if (mask > (unsigned long)(u32)mask)
801                         return -EINVAL;
802 #else
803                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
804                                 / sizeof(struct rps_dev_flow)) {
805                         /* Enforce a limit to prevent overflow */
806                         return -EINVAL;
807                 }
808 #endif
809                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
810                 if (!table)
811                         return -ENOMEM;
812
813                 table->mask = mask;
814                 for (count = 0; count <= mask; count++)
815                         table->flows[count].cpu = RPS_NO_CPU;
816         } else
817                 table = NULL;
818
819         spin_lock(&rps_dev_flow_lock);
820         old_table = rcu_dereference_protected(queue->rps_flow_table,
821                                               lockdep_is_held(&rps_dev_flow_lock));
822         rcu_assign_pointer(queue->rps_flow_table, table);
823         spin_unlock(&rps_dev_flow_lock);
824
825         if (old_table)
826                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
827
828         return len;
829 }
830
831 static struct rx_queue_attribute rps_cpus_attribute =
832         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
833
834
835 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
836         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
837             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
838 #endif /* CONFIG_RPS */
839
840 static struct attribute *rx_queue_default_attrs[] = {
841 #ifdef CONFIG_RPS
842         &rps_cpus_attribute.attr,
843         &rps_dev_flow_table_cnt_attribute.attr,
844 #endif
845         NULL
846 };
847
848 static void rx_queue_release(struct kobject *kobj)
849 {
850         struct netdev_rx_queue *queue = to_rx_queue(kobj);
851 #ifdef CONFIG_RPS
852         struct rps_map *map;
853         struct rps_dev_flow_table *flow_table;
854
855
856         map = rcu_dereference_protected(queue->rps_map, 1);
857         if (map) {
858                 RCU_INIT_POINTER(queue->rps_map, NULL);
859                 kfree_rcu(map, rcu);
860         }
861
862         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
863         if (flow_table) {
864                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
865                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
866         }
867 #endif
868
869         memset(kobj, 0, sizeof(*kobj));
870         dev_put(queue->dev);
871 }
872
873 static const void *rx_queue_namespace(struct kobject *kobj)
874 {
875         struct netdev_rx_queue *queue = to_rx_queue(kobj);
876         struct device *dev = &queue->dev->dev;
877         const void *ns = NULL;
878
879         if (dev->class && dev->class->ns_type)
880                 ns = dev->class->namespace(dev);
881
882         return ns;
883 }
884
885 static struct kobj_type rx_queue_ktype = {
886         .sysfs_ops = &rx_queue_sysfs_ops,
887         .release = rx_queue_release,
888         .default_attrs = rx_queue_default_attrs,
889         .namespace = rx_queue_namespace
890 };
891
892 static int rx_queue_add_kobject(struct net_device *dev, int index)
893 {
894         struct netdev_rx_queue *queue = dev->_rx + index;
895         struct kobject *kobj = &queue->kobj;
896         int error = 0;
897
898         kobj->kset = dev->queues_kset;
899         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
900             "rx-%u", index);
901         if (error)
902                 goto exit;
903
904         if (dev->sysfs_rx_queue_group) {
905                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
906                 if (error)
907                         goto exit;
908         }
909
910         kobject_uevent(kobj, KOBJ_ADD);
911         dev_hold(queue->dev);
912
913         return error;
914 exit:
915         kobject_put(kobj);
916         return error;
917 }
918 #endif /* CONFIG_SYSFS */
919
920 int
921 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
922 {
923 #ifdef CONFIG_SYSFS
924         int i;
925         int error = 0;
926
927 #ifndef CONFIG_RPS
928         if (!dev->sysfs_rx_queue_group)
929                 return 0;
930 #endif
931         for (i = old_num; i < new_num; i++) {
932                 error = rx_queue_add_kobject(dev, i);
933                 if (error) {
934                         new_num = old_num;
935                         break;
936                 }
937         }
938
939         while (--i >= new_num) {
940                 if (dev->sysfs_rx_queue_group)
941                         sysfs_remove_group(&dev->_rx[i].kobj,
942                                            dev->sysfs_rx_queue_group);
943                 kobject_put(&dev->_rx[i].kobj);
944         }
945
946         return error;
947 #else
948         return 0;
949 #endif
950 }
951
952 #ifdef CONFIG_SYSFS
953 /*
954  * netdev_queue sysfs structures and functions.
955  */
956 struct netdev_queue_attribute {
957         struct attribute attr;
958         ssize_t (*show)(struct netdev_queue *queue,
959             struct netdev_queue_attribute *attr, char *buf);
960         ssize_t (*store)(struct netdev_queue *queue,
961             struct netdev_queue_attribute *attr, const char *buf, size_t len);
962 };
963 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
964     struct netdev_queue_attribute, attr)
965
966 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
967
968 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
969                                       struct attribute *attr, char *buf)
970 {
971         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
972         struct netdev_queue *queue = to_netdev_queue(kobj);
973
974         if (!attribute->show)
975                 return -EIO;
976
977         return attribute->show(queue, attribute, buf);
978 }
979
980 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
981                                        struct attribute *attr,
982                                        const char *buf, size_t count)
983 {
984         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
985         struct netdev_queue *queue = to_netdev_queue(kobj);
986
987         if (!attribute->store)
988                 return -EIO;
989
990         return attribute->store(queue, attribute, buf, count);
991 }
992
993 static const struct sysfs_ops netdev_queue_sysfs_ops = {
994         .show = netdev_queue_attr_show,
995         .store = netdev_queue_attr_store,
996 };
997
998 static ssize_t show_trans_timeout(struct netdev_queue *queue,
999                                   struct netdev_queue_attribute *attribute,
1000                                   char *buf)
1001 {
1002         unsigned long trans_timeout;
1003
1004         spin_lock_irq(&queue->_xmit_lock);
1005         trans_timeout = queue->trans_timeout;
1006         spin_unlock_irq(&queue->_xmit_lock);
1007
1008         return sprintf(buf, "%lu", trans_timeout);
1009 }
1010
1011 #ifdef CONFIG_XPS
1012 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1013 {
1014         struct net_device *dev = queue->dev;
1015         unsigned int i;
1016
1017         i = queue - dev->_tx;
1018         BUG_ON(i >= dev->num_tx_queues);
1019
1020         return i;
1021 }
1022
1023 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1024                                struct netdev_queue_attribute *attribute,
1025                                char *buf)
1026 {
1027         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1028 }
1029
1030 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1031                               struct netdev_queue_attribute *attribute,
1032                               const char *buf, size_t len)
1033 {
1034         struct net_device *dev = queue->dev;
1035         int err, index = get_netdev_queue_index(queue);
1036         u32 rate = 0;
1037
1038         err = kstrtou32(buf, 10, &rate);
1039         if (err < 0)
1040                 return err;
1041
1042         if (!rtnl_trylock())
1043                 return restart_syscall();
1044
1045         err = -EOPNOTSUPP;
1046         if (dev->netdev_ops->ndo_set_tx_maxrate)
1047                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1048
1049         rtnl_unlock();
1050         if (!err) {
1051                 queue->tx_maxrate = rate;
1052                 return len;
1053         }
1054         return err;
1055 }
1056
1057 static struct netdev_queue_attribute queue_tx_maxrate =
1058         __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1059                show_tx_maxrate, set_tx_maxrate);
1060 #endif
1061
1062 static struct netdev_queue_attribute queue_trans_timeout =
1063         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1064
1065 #ifdef CONFIG_BQL
1066 /*
1067  * Byte queue limits sysfs structures and functions.
1068  */
1069 static ssize_t bql_show(char *buf, unsigned int value)
1070 {
1071         return sprintf(buf, "%u\n", value);
1072 }
1073
1074 static ssize_t bql_set(const char *buf, const size_t count,
1075                        unsigned int *pvalue)
1076 {
1077         unsigned int value;
1078         int err;
1079
1080         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1081                 value = DQL_MAX_LIMIT;
1082         else {
1083                 err = kstrtouint(buf, 10, &value);
1084                 if (err < 0)
1085                         return err;
1086                 if (value > DQL_MAX_LIMIT)
1087                         return -EINVAL;
1088         }
1089
1090         *pvalue = value;
1091
1092         return count;
1093 }
1094
1095 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1096                                   struct netdev_queue_attribute *attr,
1097                                   char *buf)
1098 {
1099         struct dql *dql = &queue->dql;
1100
1101         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1102 }
1103
1104 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1105                                  struct netdev_queue_attribute *attribute,
1106                                  const char *buf, size_t len)
1107 {
1108         struct dql *dql = &queue->dql;
1109         unsigned int value;
1110         int err;
1111
1112         err = kstrtouint(buf, 10, &value);
1113         if (err < 0)
1114                 return err;
1115
1116         dql->slack_hold_time = msecs_to_jiffies(value);
1117
1118         return len;
1119 }
1120
1121 static struct netdev_queue_attribute bql_hold_time_attribute =
1122         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1123             bql_set_hold_time);
1124
1125 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1126                                  struct netdev_queue_attribute *attr,
1127                                  char *buf)
1128 {
1129         struct dql *dql = &queue->dql;
1130
1131         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1132 }
1133
1134 static struct netdev_queue_attribute bql_inflight_attribute =
1135         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1136
1137 #define BQL_ATTR(NAME, FIELD)                                           \
1138 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1139                                  struct netdev_queue_attribute *attr,   \
1140                                  char *buf)                             \
1141 {                                                                       \
1142         return bql_show(buf, queue->dql.FIELD);                         \
1143 }                                                                       \
1144                                                                         \
1145 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1146                                 struct netdev_queue_attribute *attr,    \
1147                                 const char *buf, size_t len)            \
1148 {                                                                       \
1149         return bql_set(buf, len, &queue->dql.FIELD);                    \
1150 }                                                                       \
1151                                                                         \
1152 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1153         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1154             bql_set_ ## NAME);
1155
1156 BQL_ATTR(limit, limit)
1157 BQL_ATTR(limit_max, max_limit)
1158 BQL_ATTR(limit_min, min_limit)
1159
1160 static struct attribute *dql_attrs[] = {
1161         &bql_limit_attribute.attr,
1162         &bql_limit_max_attribute.attr,
1163         &bql_limit_min_attribute.attr,
1164         &bql_hold_time_attribute.attr,
1165         &bql_inflight_attribute.attr,
1166         NULL
1167 };
1168
1169 static struct attribute_group dql_group = {
1170         .name  = "byte_queue_limits",
1171         .attrs  = dql_attrs,
1172 };
1173 #endif /* CONFIG_BQL */
1174
1175 #ifdef CONFIG_XPS
1176 static ssize_t show_xps_map(struct netdev_queue *queue,
1177                             struct netdev_queue_attribute *attribute, char *buf)
1178 {
1179         struct net_device *dev = queue->dev;
1180         struct xps_dev_maps *dev_maps;
1181         cpumask_var_t mask;
1182         unsigned long index;
1183         int i, len;
1184
1185         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1186                 return -ENOMEM;
1187
1188         index = get_netdev_queue_index(queue);
1189
1190         rcu_read_lock();
1191         dev_maps = rcu_dereference(dev->xps_maps);
1192         if (dev_maps) {
1193                 for_each_possible_cpu(i) {
1194                         struct xps_map *map =
1195                             rcu_dereference(dev_maps->cpu_map[i]);
1196                         if (map) {
1197                                 int j;
1198                                 for (j = 0; j < map->len; j++) {
1199                                         if (map->queues[j] == index) {
1200                                                 cpumask_set_cpu(i, mask);
1201                                                 break;
1202                                         }
1203                                 }
1204                         }
1205                 }
1206         }
1207         rcu_read_unlock();
1208
1209         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1210         free_cpumask_var(mask);
1211         return len < PAGE_SIZE ? len : -EINVAL;
1212 }
1213
1214 static ssize_t store_xps_map(struct netdev_queue *queue,
1215                       struct netdev_queue_attribute *attribute,
1216                       const char *buf, size_t len)
1217 {
1218         struct net_device *dev = queue->dev;
1219         unsigned long index;
1220         cpumask_var_t mask;
1221         int err;
1222
1223         if (!capable(CAP_NET_ADMIN))
1224                 return -EPERM;
1225
1226         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1227                 return -ENOMEM;
1228
1229         index = get_netdev_queue_index(queue);
1230
1231         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1232         if (err) {
1233                 free_cpumask_var(mask);
1234                 return err;
1235         }
1236
1237         err = netif_set_xps_queue(dev, mask, index);
1238
1239         free_cpumask_var(mask);
1240
1241         return err ? : len;
1242 }
1243
1244 static struct netdev_queue_attribute xps_cpus_attribute =
1245     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1246 #endif /* CONFIG_XPS */
1247
1248 static struct attribute *netdev_queue_default_attrs[] = {
1249         &queue_trans_timeout.attr,
1250 #ifdef CONFIG_XPS
1251         &xps_cpus_attribute.attr,
1252         &queue_tx_maxrate.attr,
1253 #endif
1254         NULL
1255 };
1256
1257 static void netdev_queue_release(struct kobject *kobj)
1258 {
1259         struct netdev_queue *queue = to_netdev_queue(kobj);
1260
1261         memset(kobj, 0, sizeof(*kobj));
1262         dev_put(queue->dev);
1263 }
1264
1265 static const void *netdev_queue_namespace(struct kobject *kobj)
1266 {
1267         struct netdev_queue *queue = to_netdev_queue(kobj);
1268         struct device *dev = &queue->dev->dev;
1269         const void *ns = NULL;
1270
1271         if (dev->class && dev->class->ns_type)
1272                 ns = dev->class->namespace(dev);
1273
1274         return ns;
1275 }
1276
1277 static struct kobj_type netdev_queue_ktype = {
1278         .sysfs_ops = &netdev_queue_sysfs_ops,
1279         .release = netdev_queue_release,
1280         .default_attrs = netdev_queue_default_attrs,
1281         .namespace = netdev_queue_namespace,
1282 };
1283
1284 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1285 {
1286         struct netdev_queue *queue = dev->_tx + index;
1287         struct kobject *kobj = &queue->kobj;
1288         int error = 0;
1289
1290         kobj->kset = dev->queues_kset;
1291         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1292             "tx-%u", index);
1293         if (error)
1294                 goto exit;
1295
1296 #ifdef CONFIG_BQL
1297         error = sysfs_create_group(kobj, &dql_group);
1298         if (error)
1299                 goto exit;
1300 #endif
1301
1302         kobject_uevent(kobj, KOBJ_ADD);
1303         dev_hold(queue->dev);
1304
1305         return 0;
1306 exit:
1307         kobject_put(kobj);
1308         return error;
1309 }
1310 #endif /* CONFIG_SYSFS */
1311
1312 int
1313 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1314 {
1315 #ifdef CONFIG_SYSFS
1316         int i;
1317         int error = 0;
1318
1319         for (i = old_num; i < new_num; i++) {
1320                 error = netdev_queue_add_kobject(dev, i);
1321                 if (error) {
1322                         new_num = old_num;
1323                         break;
1324                 }
1325         }
1326
1327         while (--i >= new_num) {
1328                 struct netdev_queue *queue = dev->_tx + i;
1329
1330 #ifdef CONFIG_BQL
1331                 sysfs_remove_group(&queue->kobj, &dql_group);
1332 #endif
1333                 kobject_put(&queue->kobj);
1334         }
1335
1336         return error;
1337 #else
1338         return 0;
1339 #endif /* CONFIG_SYSFS */
1340 }
1341
1342 static int register_queue_kobjects(struct net_device *dev)
1343 {
1344         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1345
1346 #ifdef CONFIG_SYSFS
1347         dev->queues_kset = kset_create_and_add("queues",
1348             NULL, &dev->dev.kobj);
1349         if (!dev->queues_kset)
1350                 return -ENOMEM;
1351         real_rx = dev->real_num_rx_queues;
1352 #endif
1353         real_tx = dev->real_num_tx_queues;
1354
1355         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1356         if (error)
1357                 goto error;
1358         rxq = real_rx;
1359
1360         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1361         if (error)
1362                 goto error;
1363         txq = real_tx;
1364
1365         return 0;
1366
1367 error:
1368         netdev_queue_update_kobjects(dev, txq, 0);
1369         net_rx_queue_update_kobjects(dev, rxq, 0);
1370         return error;
1371 }
1372
1373 static void remove_queue_kobjects(struct net_device *dev)
1374 {
1375         int real_rx = 0, real_tx = 0;
1376
1377 #ifdef CONFIG_SYSFS
1378         real_rx = dev->real_num_rx_queues;
1379 #endif
1380         real_tx = dev->real_num_tx_queues;
1381
1382         net_rx_queue_update_kobjects(dev, real_rx, 0);
1383         netdev_queue_update_kobjects(dev, real_tx, 0);
1384 #ifdef CONFIG_SYSFS
1385         kset_unregister(dev->queues_kset);
1386 #endif
1387 }
1388
1389 static bool net_current_may_mount(void)
1390 {
1391         struct net *net = current->nsproxy->net_ns;
1392
1393         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1394 }
1395
1396 static void *net_grab_current_ns(void)
1397 {
1398         struct net *ns = current->nsproxy->net_ns;
1399 #ifdef CONFIG_NET_NS
1400         if (ns)
1401                 atomic_inc(&ns->passive);
1402 #endif
1403         return ns;
1404 }
1405
1406 static const void *net_initial_ns(void)
1407 {
1408         return &init_net;
1409 }
1410
1411 static const void *net_netlink_ns(struct sock *sk)
1412 {
1413         return sock_net(sk);
1414 }
1415
1416 struct kobj_ns_type_operations net_ns_type_operations = {
1417         .type = KOBJ_NS_TYPE_NET,
1418         .current_may_mount = net_current_may_mount,
1419         .grab_current_ns = net_grab_current_ns,
1420         .netlink_ns = net_netlink_ns,
1421         .initial_ns = net_initial_ns,
1422         .drop_ns = net_drop_ns,
1423 };
1424 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1425
1426 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1427 {
1428         struct net_device *dev = to_net_dev(d);
1429         int retval;
1430
1431         /* pass interface to uevent. */
1432         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1433         if (retval)
1434                 goto exit;
1435
1436         /* pass ifindex to uevent.
1437          * ifindex is useful as it won't change (interface name may change)
1438          * and is what RtNetlink uses natively. */
1439         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1440
1441 exit:
1442         return retval;
1443 }
1444
1445 /*
1446  *      netdev_release -- destroy and free a dead device.
1447  *      Called when last reference to device kobject is gone.
1448  */
1449 static void netdev_release(struct device *d)
1450 {
1451         struct net_device *dev = to_net_dev(d);
1452
1453         BUG_ON(dev->reg_state != NETREG_RELEASED);
1454
1455         kfree(dev->ifalias);
1456         netdev_freemem(dev);
1457 }
1458
1459 static const void *net_namespace(struct device *d)
1460 {
1461         struct net_device *dev = to_net_dev(d);
1462
1463         return dev_net(dev);
1464 }
1465
1466 static struct class net_class = {
1467         .name = "net",
1468         .dev_release = netdev_release,
1469         .dev_groups = net_class_groups,
1470         .dev_uevent = netdev_uevent,
1471         .ns_type = &net_ns_type_operations,
1472         .namespace = net_namespace,
1473 };
1474
1475 #ifdef CONFIG_OF_NET
1476 static int of_dev_node_match(struct device *dev, const void *data)
1477 {
1478         int ret = 0;
1479
1480         if (dev->parent)
1481                 ret = dev->parent->of_node == data;
1482
1483         return ret == 0 ? dev->of_node == data : ret;
1484 }
1485
1486 /*
1487  * of_find_net_device_by_node - lookup the net device for the device node
1488  * @np: OF device node
1489  *
1490  * Looks up the net_device structure corresponding with the device node.
1491  * If successful, returns a pointer to the net_device with the embedded
1492  * struct device refcount incremented by one, or NULL on failure. The
1493  * refcount must be dropped when done with the net_device.
1494  */
1495 struct net_device *of_find_net_device_by_node(struct device_node *np)
1496 {
1497         struct device *dev;
1498
1499         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1500         if (!dev)
1501                 return NULL;
1502
1503         return to_net_dev(dev);
1504 }
1505 EXPORT_SYMBOL(of_find_net_device_by_node);
1506 #endif
1507
1508 /* Delete sysfs entries but hold kobject reference until after all
1509  * netdev references are gone.
1510  */
1511 void netdev_unregister_kobject(struct net_device *ndev)
1512 {
1513         struct device *dev = &(ndev->dev);
1514
1515         kobject_get(&dev->kobj);
1516
1517         remove_queue_kobjects(ndev);
1518
1519         pm_runtime_set_memalloc_noio(dev, false);
1520
1521         device_del(dev);
1522 }
1523
1524 /* Create sysfs entries for network device. */
1525 int netdev_register_kobject(struct net_device *ndev)
1526 {
1527         struct device *dev = &(ndev->dev);
1528         const struct attribute_group **groups = ndev->sysfs_groups;
1529         int error = 0;
1530
1531         device_initialize(dev);
1532         dev->class = &net_class;
1533         dev->platform_data = ndev;
1534         dev->groups = groups;
1535
1536         dev_set_name(dev, "%s", ndev->name);
1537
1538 #ifdef CONFIG_SYSFS
1539         /* Allow for a device specific group */
1540         if (*groups)
1541                 groups++;
1542
1543         *groups++ = &netstat_group;
1544
1545 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1546         if (ndev->ieee80211_ptr)
1547                 *groups++ = &wireless_group;
1548 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1549         else if (ndev->wireless_handlers)
1550                 *groups++ = &wireless_group;
1551 #endif
1552 #endif
1553 #endif /* CONFIG_SYSFS */
1554
1555         error = device_add(dev);
1556         if (error)
1557                 return error;
1558
1559         error = register_queue_kobjects(ndev);
1560         if (error) {
1561                 device_del(dev);
1562                 return error;
1563         }
1564
1565         pm_runtime_set_memalloc_noio(dev, true);
1566
1567         return error;
1568 }
1569
1570 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1571                                 const void *ns)
1572 {
1573         return class_create_file_ns(&net_class, class_attr, ns);
1574 }
1575 EXPORT_SYMBOL(netdev_class_create_file_ns);
1576
1577 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1578                                  const void *ns)
1579 {
1580         class_remove_file_ns(&net_class, class_attr, ns);
1581 }
1582 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1583
1584 int __init netdev_kobject_init(void)
1585 {
1586         kobj_ns_type_register(&net_ns_type_operations);
1587         return class_register(&net_class);
1588 }