net-sysfs: try not to restart the syscall if it will fail eventually
[platform/kernel/linux-starfive.git] / net / core / net-sysfs.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net-sysfs.c - network device class and attributes
4  *
5  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
6  */
7
8 #include <linux/capability.h>
9 #include <linux/kernel.h>
10 #include <linux/netdevice.h>
11 #include <linux/if_arp.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/sched/isolation.h>
15 #include <linux/nsproxy.h>
16 #include <net/sock.h>
17 #include <net/net_namespace.h>
18 #include <linux/rtnetlink.h>
19 #include <linux/vmalloc.h>
20 #include <linux/export.h>
21 #include <linux/jiffies.h>
22 #include <linux/pm_runtime.h>
23 #include <linux/of.h>
24 #include <linux/of_net.h>
25 #include <linux/cpu.h>
26
27 #include "net-sysfs.h"
28
29 #ifdef CONFIG_SYSFS
30 static const char fmt_hex[] = "%#x\n";
31 static const char fmt_dec[] = "%d\n";
32 static const char fmt_ulong[] = "%lu\n";
33 static const char fmt_u64[] = "%llu\n";
34
35 static inline int dev_isalive(const struct net_device *dev)
36 {
37         return dev->reg_state <= NETREG_REGISTERED;
38 }
39
40 /* use same locking rules as GIF* ioctl's */
41 static ssize_t netdev_show(const struct device *dev,
42                            struct device_attribute *attr, char *buf,
43                            ssize_t (*format)(const struct net_device *, char *))
44 {
45         struct net_device *ndev = to_net_dev(dev);
46         ssize_t ret = -EINVAL;
47
48         read_lock(&dev_base_lock);
49         if (dev_isalive(ndev))
50                 ret = (*format)(ndev, buf);
51         read_unlock(&dev_base_lock);
52
53         return ret;
54 }
55
56 /* generate a show function for simple field */
57 #define NETDEVICE_SHOW(field, format_string)                            \
58 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
59 {                                                                       \
60         return sprintf(buf, format_string, dev->field);                 \
61 }                                                                       \
62 static ssize_t field##_show(struct device *dev,                         \
63                             struct device_attribute *attr, char *buf)   \
64 {                                                                       \
65         return netdev_show(dev, attr, buf, format_##field);             \
66 }                                                                       \
67
68 #define NETDEVICE_SHOW_RO(field, format_string)                         \
69 NETDEVICE_SHOW(field, format_string);                                   \
70 static DEVICE_ATTR_RO(field)
71
72 #define NETDEVICE_SHOW_RW(field, format_string)                         \
73 NETDEVICE_SHOW(field, format_string);                                   \
74 static DEVICE_ATTR_RW(field)
75
76 /* use same locking and permission rules as SIF* ioctl's */
77 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
78                             const char *buf, size_t len,
79                             int (*set)(struct net_device *, unsigned long))
80 {
81         struct net_device *netdev = to_net_dev(dev);
82         struct net *net = dev_net(netdev);
83         unsigned long new;
84         int ret;
85
86         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
87                 return -EPERM;
88
89         ret = kstrtoul(buf, 0, &new);
90         if (ret)
91                 goto err;
92
93         if (!rtnl_trylock())
94                 return restart_syscall();
95
96         if (dev_isalive(netdev)) {
97                 ret = (*set)(netdev, new);
98                 if (ret == 0)
99                         ret = len;
100         }
101         rtnl_unlock();
102  err:
103         return ret;
104 }
105
106 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
108 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
109 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
110 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
111 NETDEVICE_SHOW_RO(type, fmt_dec);
112 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
113
114 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
115                            char *buf)
116 {
117         struct net_device *ndev = to_net_dev(dev);
118
119         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
120 }
121 static DEVICE_ATTR_RO(iflink);
122
123 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
124 {
125         return sprintf(buf, fmt_dec, dev->name_assign_type);
126 }
127
128 static ssize_t name_assign_type_show(struct device *dev,
129                                      struct device_attribute *attr,
130                                      char *buf)
131 {
132         struct net_device *ndev = to_net_dev(dev);
133         ssize_t ret = -EINVAL;
134
135         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
136                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
137
138         return ret;
139 }
140 static DEVICE_ATTR_RO(name_assign_type);
141
142 /* use same locking rules as GIFHWADDR ioctl's */
143 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
144                             char *buf)
145 {
146         struct net_device *ndev = to_net_dev(dev);
147         ssize_t ret = -EINVAL;
148
149         read_lock(&dev_base_lock);
150         if (dev_isalive(ndev))
151                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
152         read_unlock(&dev_base_lock);
153         return ret;
154 }
155 static DEVICE_ATTR_RO(address);
156
157 static ssize_t broadcast_show(struct device *dev,
158                               struct device_attribute *attr, char *buf)
159 {
160         struct net_device *ndev = to_net_dev(dev);
161
162         if (dev_isalive(ndev))
163                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
164         return -EINVAL;
165 }
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         struct net_device *netdev = to_net_dev(dev);
179
180         /* The check is also done in change_carrier; this helps returning early
181          * without hitting the trylock/restart in netdev_store.
182          */
183         if (!netdev->netdev_ops->ndo_change_carrier)
184                 return -EOPNOTSUPP;
185
186         return netdev_store(dev, attr, buf, len, change_carrier);
187 }
188
189 static ssize_t carrier_show(struct device *dev,
190                             struct device_attribute *attr, char *buf)
191 {
192         struct net_device *netdev = to_net_dev(dev);
193
194         if (netif_running(netdev))
195                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
196
197         return -EINVAL;
198 }
199 static DEVICE_ATTR_RW(carrier);
200
201 static ssize_t speed_show(struct device *dev,
202                           struct device_attribute *attr, char *buf)
203 {
204         struct net_device *netdev = to_net_dev(dev);
205         int ret = -EINVAL;
206
207         /* The check is also done in __ethtool_get_link_ksettings; this helps
208          * returning early without hitting the trylock/restart below.
209          */
210         if (!netdev->ethtool_ops->get_link_ksettings)
211                 return ret;
212
213         if (!rtnl_trylock())
214                 return restart_syscall();
215
216         if (netif_running(netdev)) {
217                 struct ethtool_link_ksettings cmd;
218
219                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
220                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
221         }
222         rtnl_unlock();
223         return ret;
224 }
225 static DEVICE_ATTR_RO(speed);
226
227 static ssize_t duplex_show(struct device *dev,
228                            struct device_attribute *attr, char *buf)
229 {
230         struct net_device *netdev = to_net_dev(dev);
231         int ret = -EINVAL;
232
233         /* The check is also done in __ethtool_get_link_ksettings; this helps
234          * returning early without hitting the trylock/restart below.
235          */
236         if (!netdev->ethtool_ops->get_link_ksettings)
237                 return ret;
238
239         if (!rtnl_trylock())
240                 return restart_syscall();
241
242         if (netif_running(netdev)) {
243                 struct ethtool_link_ksettings cmd;
244
245                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
246                         const char *duplex;
247
248                         switch (cmd.base.duplex) {
249                         case DUPLEX_HALF:
250                                 duplex = "half";
251                                 break;
252                         case DUPLEX_FULL:
253                                 duplex = "full";
254                                 break;
255                         default:
256                                 duplex = "unknown";
257                                 break;
258                         }
259                         ret = sprintf(buf, "%s\n", duplex);
260                 }
261         }
262         rtnl_unlock();
263         return ret;
264 }
265 static DEVICE_ATTR_RO(duplex);
266
267 static ssize_t testing_show(struct device *dev,
268                             struct device_attribute *attr, char *buf)
269 {
270         struct net_device *netdev = to_net_dev(dev);
271
272         if (netif_running(netdev))
273                 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
274
275         return -EINVAL;
276 }
277 static DEVICE_ATTR_RO(testing);
278
279 static ssize_t dormant_show(struct device *dev,
280                             struct device_attribute *attr, char *buf)
281 {
282         struct net_device *netdev = to_net_dev(dev);
283
284         if (netif_running(netdev))
285                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
286
287         return -EINVAL;
288 }
289 static DEVICE_ATTR_RO(dormant);
290
291 static const char *const operstates[] = {
292         "unknown",
293         "notpresent", /* currently unused */
294         "down",
295         "lowerlayerdown",
296         "testing",
297         "dormant",
298         "up"
299 };
300
301 static ssize_t operstate_show(struct device *dev,
302                               struct device_attribute *attr, char *buf)
303 {
304         const struct net_device *netdev = to_net_dev(dev);
305         unsigned char operstate;
306
307         read_lock(&dev_base_lock);
308         operstate = netdev->operstate;
309         if (!netif_running(netdev))
310                 operstate = IF_OPER_DOWN;
311         read_unlock(&dev_base_lock);
312
313         if (operstate >= ARRAY_SIZE(operstates))
314                 return -EINVAL; /* should not happen */
315
316         return sprintf(buf, "%s\n", operstates[operstate]);
317 }
318 static DEVICE_ATTR_RO(operstate);
319
320 static ssize_t carrier_changes_show(struct device *dev,
321                                     struct device_attribute *attr,
322                                     char *buf)
323 {
324         struct net_device *netdev = to_net_dev(dev);
325
326         return sprintf(buf, fmt_dec,
327                        atomic_read(&netdev->carrier_up_count) +
328                        atomic_read(&netdev->carrier_down_count));
329 }
330 static DEVICE_ATTR_RO(carrier_changes);
331
332 static ssize_t carrier_up_count_show(struct device *dev,
333                                      struct device_attribute *attr,
334                                      char *buf)
335 {
336         struct net_device *netdev = to_net_dev(dev);
337
338         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
339 }
340 static DEVICE_ATTR_RO(carrier_up_count);
341
342 static ssize_t carrier_down_count_show(struct device *dev,
343                                        struct device_attribute *attr,
344                                        char *buf)
345 {
346         struct net_device *netdev = to_net_dev(dev);
347
348         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
349 }
350 static DEVICE_ATTR_RO(carrier_down_count);
351
352 /* read-write attributes */
353
354 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
355 {
356         return dev_set_mtu(dev, (int)new_mtu);
357 }
358
359 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
360                          const char *buf, size_t len)
361 {
362         return netdev_store(dev, attr, buf, len, change_mtu);
363 }
364 NETDEVICE_SHOW_RW(mtu, fmt_dec);
365
366 static int change_flags(struct net_device *dev, unsigned long new_flags)
367 {
368         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
369 }
370
371 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
372                            const char *buf, size_t len)
373 {
374         return netdev_store(dev, attr, buf, len, change_flags);
375 }
376 NETDEVICE_SHOW_RW(flags, fmt_hex);
377
378 static ssize_t tx_queue_len_store(struct device *dev,
379                                   struct device_attribute *attr,
380                                   const char *buf, size_t len)
381 {
382         if (!capable(CAP_NET_ADMIN))
383                 return -EPERM;
384
385         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
386 }
387 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
388
389 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
390 {
391         WRITE_ONCE(dev->gro_flush_timeout, val);
392         return 0;
393 }
394
395 static ssize_t gro_flush_timeout_store(struct device *dev,
396                                        struct device_attribute *attr,
397                                        const char *buf, size_t len)
398 {
399         if (!capable(CAP_NET_ADMIN))
400                 return -EPERM;
401
402         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
403 }
404 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
405
406 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
407 {
408         WRITE_ONCE(dev->napi_defer_hard_irqs, val);
409         return 0;
410 }
411
412 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
413                                           struct device_attribute *attr,
414                                           const char *buf, size_t len)
415 {
416         if (!capable(CAP_NET_ADMIN))
417                 return -EPERM;
418
419         return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
420 }
421 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
422
423 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
424                              const char *buf, size_t len)
425 {
426         struct net_device *netdev = to_net_dev(dev);
427         struct net *net = dev_net(netdev);
428         size_t count = len;
429         ssize_t ret = 0;
430
431         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
432                 return -EPERM;
433
434         /* ignore trailing newline */
435         if (len >  0 && buf[len - 1] == '\n')
436                 --count;
437
438         if (!rtnl_trylock())
439                 return restart_syscall();
440
441         if (dev_isalive(netdev)) {
442                 ret = dev_set_alias(netdev, buf, count);
443                 if (ret < 0)
444                         goto err;
445                 ret = len;
446                 netdev_state_change(netdev);
447         }
448 err:
449         rtnl_unlock();
450
451         return ret;
452 }
453
454 static ssize_t ifalias_show(struct device *dev,
455                             struct device_attribute *attr, char *buf)
456 {
457         const struct net_device *netdev = to_net_dev(dev);
458         char tmp[IFALIASZ];
459         ssize_t ret = 0;
460
461         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
462         if (ret > 0)
463                 ret = sprintf(buf, "%s\n", tmp);
464         return ret;
465 }
466 static DEVICE_ATTR_RW(ifalias);
467
468 static int change_group(struct net_device *dev, unsigned long new_group)
469 {
470         dev_set_group(dev, (int)new_group);
471         return 0;
472 }
473
474 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
475                            const char *buf, size_t len)
476 {
477         return netdev_store(dev, attr, buf, len, change_group);
478 }
479 NETDEVICE_SHOW(group, fmt_dec);
480 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
481
482 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
483 {
484         return dev_change_proto_down(dev, (bool)proto_down);
485 }
486
487 static ssize_t proto_down_store(struct device *dev,
488                                 struct device_attribute *attr,
489                                 const char *buf, size_t len)
490 {
491         struct net_device *netdev = to_net_dev(dev);
492
493         /* The check is also done in change_proto_down; this helps returning
494          * early without hitting the trylock/restart in netdev_store.
495          */
496         if (!netdev->netdev_ops->ndo_change_proto_down)
497                 return -EOPNOTSUPP;
498
499         return netdev_store(dev, attr, buf, len, change_proto_down);
500 }
501 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
502
503 static ssize_t phys_port_id_show(struct device *dev,
504                                  struct device_attribute *attr, char *buf)
505 {
506         struct net_device *netdev = to_net_dev(dev);
507         ssize_t ret = -EINVAL;
508
509         /* The check is also done in dev_get_phys_port_id; this helps returning
510          * early without hitting the trylock/restart below.
511          */
512         if (!netdev->netdev_ops->ndo_get_phys_port_id)
513                 return -EOPNOTSUPP;
514
515         if (!rtnl_trylock())
516                 return restart_syscall();
517
518         if (dev_isalive(netdev)) {
519                 struct netdev_phys_item_id ppid;
520
521                 ret = dev_get_phys_port_id(netdev, &ppid);
522                 if (!ret)
523                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
524         }
525         rtnl_unlock();
526
527         return ret;
528 }
529 static DEVICE_ATTR_RO(phys_port_id);
530
531 static ssize_t phys_port_name_show(struct device *dev,
532                                    struct device_attribute *attr, char *buf)
533 {
534         struct net_device *netdev = to_net_dev(dev);
535         ssize_t ret = -EINVAL;
536
537         /* The checks are also done in dev_get_phys_port_name; this helps
538          * returning early without hitting the trylock/restart below.
539          */
540         if (!netdev->netdev_ops->ndo_get_phys_port_name &&
541             !netdev->netdev_ops->ndo_get_devlink_port)
542                 return -EOPNOTSUPP;
543
544         if (!rtnl_trylock())
545                 return restart_syscall();
546
547         if (dev_isalive(netdev)) {
548                 char name[IFNAMSIZ];
549
550                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
551                 if (!ret)
552                         ret = sprintf(buf, "%s\n", name);
553         }
554         rtnl_unlock();
555
556         return ret;
557 }
558 static DEVICE_ATTR_RO(phys_port_name);
559
560 static ssize_t phys_switch_id_show(struct device *dev,
561                                    struct device_attribute *attr, char *buf)
562 {
563         struct net_device *netdev = to_net_dev(dev);
564         ssize_t ret = -EINVAL;
565
566         /* The checks are also done in dev_get_phys_port_name; this helps
567          * returning early without hitting the trylock/restart below. This works
568          * because recurse is false when calling dev_get_port_parent_id.
569          */
570         if (!netdev->netdev_ops->ndo_get_port_parent_id &&
571             !netdev->netdev_ops->ndo_get_devlink_port)
572                 return -EOPNOTSUPP;
573
574         if (!rtnl_trylock())
575                 return restart_syscall();
576
577         if (dev_isalive(netdev)) {
578                 struct netdev_phys_item_id ppid = { };
579
580                 ret = dev_get_port_parent_id(netdev, &ppid, false);
581                 if (!ret)
582                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
583         }
584         rtnl_unlock();
585
586         return ret;
587 }
588 static DEVICE_ATTR_RO(phys_switch_id);
589
590 static ssize_t threaded_show(struct device *dev,
591                              struct device_attribute *attr, char *buf)
592 {
593         struct net_device *netdev = to_net_dev(dev);
594         ssize_t ret = -EINVAL;
595
596         if (!rtnl_trylock())
597                 return restart_syscall();
598
599         if (dev_isalive(netdev))
600                 ret = sprintf(buf, fmt_dec, netdev->threaded);
601
602         rtnl_unlock();
603         return ret;
604 }
605
606 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
607 {
608         int ret;
609
610         if (list_empty(&dev->napi_list))
611                 return -EOPNOTSUPP;
612
613         if (val != 0 && val != 1)
614                 return -EOPNOTSUPP;
615
616         ret = dev_set_threaded(dev, val);
617
618         return ret;
619 }
620
621 static ssize_t threaded_store(struct device *dev,
622                               struct device_attribute *attr,
623                               const char *buf, size_t len)
624 {
625         return netdev_store(dev, attr, buf, len, modify_napi_threaded);
626 }
627 static DEVICE_ATTR_RW(threaded);
628
629 static struct attribute *net_class_attrs[] __ro_after_init = {
630         &dev_attr_netdev_group.attr,
631         &dev_attr_type.attr,
632         &dev_attr_dev_id.attr,
633         &dev_attr_dev_port.attr,
634         &dev_attr_iflink.attr,
635         &dev_attr_ifindex.attr,
636         &dev_attr_name_assign_type.attr,
637         &dev_attr_addr_assign_type.attr,
638         &dev_attr_addr_len.attr,
639         &dev_attr_link_mode.attr,
640         &dev_attr_address.attr,
641         &dev_attr_broadcast.attr,
642         &dev_attr_speed.attr,
643         &dev_attr_duplex.attr,
644         &dev_attr_dormant.attr,
645         &dev_attr_testing.attr,
646         &dev_attr_operstate.attr,
647         &dev_attr_carrier_changes.attr,
648         &dev_attr_ifalias.attr,
649         &dev_attr_carrier.attr,
650         &dev_attr_mtu.attr,
651         &dev_attr_flags.attr,
652         &dev_attr_tx_queue_len.attr,
653         &dev_attr_gro_flush_timeout.attr,
654         &dev_attr_napi_defer_hard_irqs.attr,
655         &dev_attr_phys_port_id.attr,
656         &dev_attr_phys_port_name.attr,
657         &dev_attr_phys_switch_id.attr,
658         &dev_attr_proto_down.attr,
659         &dev_attr_carrier_up_count.attr,
660         &dev_attr_carrier_down_count.attr,
661         &dev_attr_threaded.attr,
662         NULL,
663 };
664 ATTRIBUTE_GROUPS(net_class);
665
666 /* Show a given an attribute in the statistics group */
667 static ssize_t netstat_show(const struct device *d,
668                             struct device_attribute *attr, char *buf,
669                             unsigned long offset)
670 {
671         struct net_device *dev = to_net_dev(d);
672         ssize_t ret = -EINVAL;
673
674         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
675                 offset % sizeof(u64) != 0);
676
677         read_lock(&dev_base_lock);
678         if (dev_isalive(dev)) {
679                 struct rtnl_link_stats64 temp;
680                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
681
682                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
683         }
684         read_unlock(&dev_base_lock);
685         return ret;
686 }
687
688 /* generate a read-only statistics attribute */
689 #define NETSTAT_ENTRY(name)                                             \
690 static ssize_t name##_show(struct device *d,                            \
691                            struct device_attribute *attr, char *buf)    \
692 {                                                                       \
693         return netstat_show(d, attr, buf,                               \
694                             offsetof(struct rtnl_link_stats64, name));  \
695 }                                                                       \
696 static DEVICE_ATTR_RO(name)
697
698 NETSTAT_ENTRY(rx_packets);
699 NETSTAT_ENTRY(tx_packets);
700 NETSTAT_ENTRY(rx_bytes);
701 NETSTAT_ENTRY(tx_bytes);
702 NETSTAT_ENTRY(rx_errors);
703 NETSTAT_ENTRY(tx_errors);
704 NETSTAT_ENTRY(rx_dropped);
705 NETSTAT_ENTRY(tx_dropped);
706 NETSTAT_ENTRY(multicast);
707 NETSTAT_ENTRY(collisions);
708 NETSTAT_ENTRY(rx_length_errors);
709 NETSTAT_ENTRY(rx_over_errors);
710 NETSTAT_ENTRY(rx_crc_errors);
711 NETSTAT_ENTRY(rx_frame_errors);
712 NETSTAT_ENTRY(rx_fifo_errors);
713 NETSTAT_ENTRY(rx_missed_errors);
714 NETSTAT_ENTRY(tx_aborted_errors);
715 NETSTAT_ENTRY(tx_carrier_errors);
716 NETSTAT_ENTRY(tx_fifo_errors);
717 NETSTAT_ENTRY(tx_heartbeat_errors);
718 NETSTAT_ENTRY(tx_window_errors);
719 NETSTAT_ENTRY(rx_compressed);
720 NETSTAT_ENTRY(tx_compressed);
721 NETSTAT_ENTRY(rx_nohandler);
722
723 static struct attribute *netstat_attrs[] __ro_after_init = {
724         &dev_attr_rx_packets.attr,
725         &dev_attr_tx_packets.attr,
726         &dev_attr_rx_bytes.attr,
727         &dev_attr_tx_bytes.attr,
728         &dev_attr_rx_errors.attr,
729         &dev_attr_tx_errors.attr,
730         &dev_attr_rx_dropped.attr,
731         &dev_attr_tx_dropped.attr,
732         &dev_attr_multicast.attr,
733         &dev_attr_collisions.attr,
734         &dev_attr_rx_length_errors.attr,
735         &dev_attr_rx_over_errors.attr,
736         &dev_attr_rx_crc_errors.attr,
737         &dev_attr_rx_frame_errors.attr,
738         &dev_attr_rx_fifo_errors.attr,
739         &dev_attr_rx_missed_errors.attr,
740         &dev_attr_tx_aborted_errors.attr,
741         &dev_attr_tx_carrier_errors.attr,
742         &dev_attr_tx_fifo_errors.attr,
743         &dev_attr_tx_heartbeat_errors.attr,
744         &dev_attr_tx_window_errors.attr,
745         &dev_attr_rx_compressed.attr,
746         &dev_attr_tx_compressed.attr,
747         &dev_attr_rx_nohandler.attr,
748         NULL
749 };
750
751 static const struct attribute_group netstat_group = {
752         .name  = "statistics",
753         .attrs  = netstat_attrs,
754 };
755
756 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
757 static struct attribute *wireless_attrs[] = {
758         NULL
759 };
760
761 static const struct attribute_group wireless_group = {
762         .name = "wireless",
763         .attrs = wireless_attrs,
764 };
765 #endif
766
767 #else /* CONFIG_SYSFS */
768 #define net_class_groups        NULL
769 #endif /* CONFIG_SYSFS */
770
771 #ifdef CONFIG_SYSFS
772 #define to_rx_queue_attr(_attr) \
773         container_of(_attr, struct rx_queue_attribute, attr)
774
775 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
776
777 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
778                                   char *buf)
779 {
780         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
781         struct netdev_rx_queue *queue = to_rx_queue(kobj);
782
783         if (!attribute->show)
784                 return -EIO;
785
786         return attribute->show(queue, buf);
787 }
788
789 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
790                                    const char *buf, size_t count)
791 {
792         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
793         struct netdev_rx_queue *queue = to_rx_queue(kobj);
794
795         if (!attribute->store)
796                 return -EIO;
797
798         return attribute->store(queue, buf, count);
799 }
800
801 static const struct sysfs_ops rx_queue_sysfs_ops = {
802         .show = rx_queue_attr_show,
803         .store = rx_queue_attr_store,
804 };
805
806 #ifdef CONFIG_RPS
807 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
808 {
809         struct rps_map *map;
810         cpumask_var_t mask;
811         int i, len;
812
813         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
814                 return -ENOMEM;
815
816         rcu_read_lock();
817         map = rcu_dereference(queue->rps_map);
818         if (map)
819                 for (i = 0; i < map->len; i++)
820                         cpumask_set_cpu(map->cpus[i], mask);
821
822         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
823         rcu_read_unlock();
824         free_cpumask_var(mask);
825
826         return len < PAGE_SIZE ? len : -EINVAL;
827 }
828
829 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
830                              const char *buf, size_t len)
831 {
832         struct rps_map *old_map, *map;
833         cpumask_var_t mask;
834         int err, cpu, i, hk_flags;
835         static DEFINE_MUTEX(rps_map_mutex);
836
837         if (!capable(CAP_NET_ADMIN))
838                 return -EPERM;
839
840         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
841                 return -ENOMEM;
842
843         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
844         if (err) {
845                 free_cpumask_var(mask);
846                 return err;
847         }
848
849         if (!cpumask_empty(mask)) {
850                 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
851                 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
852                 if (cpumask_empty(mask)) {
853                         free_cpumask_var(mask);
854                         return -EINVAL;
855                 }
856         }
857
858         map = kzalloc(max_t(unsigned int,
859                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
860                       GFP_KERNEL);
861         if (!map) {
862                 free_cpumask_var(mask);
863                 return -ENOMEM;
864         }
865
866         i = 0;
867         for_each_cpu_and(cpu, mask, cpu_online_mask)
868                 map->cpus[i++] = cpu;
869
870         if (i) {
871                 map->len = i;
872         } else {
873                 kfree(map);
874                 map = NULL;
875         }
876
877         mutex_lock(&rps_map_mutex);
878         old_map = rcu_dereference_protected(queue->rps_map,
879                                             mutex_is_locked(&rps_map_mutex));
880         rcu_assign_pointer(queue->rps_map, map);
881
882         if (map)
883                 static_branch_inc(&rps_needed);
884         if (old_map)
885                 static_branch_dec(&rps_needed);
886
887         mutex_unlock(&rps_map_mutex);
888
889         if (old_map)
890                 kfree_rcu(old_map, rcu);
891
892         free_cpumask_var(mask);
893         return len;
894 }
895
896 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
897                                            char *buf)
898 {
899         struct rps_dev_flow_table *flow_table;
900         unsigned long val = 0;
901
902         rcu_read_lock();
903         flow_table = rcu_dereference(queue->rps_flow_table);
904         if (flow_table)
905                 val = (unsigned long)flow_table->mask + 1;
906         rcu_read_unlock();
907
908         return sprintf(buf, "%lu\n", val);
909 }
910
911 static void rps_dev_flow_table_release(struct rcu_head *rcu)
912 {
913         struct rps_dev_flow_table *table = container_of(rcu,
914             struct rps_dev_flow_table, rcu);
915         vfree(table);
916 }
917
918 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
919                                             const char *buf, size_t len)
920 {
921         unsigned long mask, count;
922         struct rps_dev_flow_table *table, *old_table;
923         static DEFINE_SPINLOCK(rps_dev_flow_lock);
924         int rc;
925
926         if (!capable(CAP_NET_ADMIN))
927                 return -EPERM;
928
929         rc = kstrtoul(buf, 0, &count);
930         if (rc < 0)
931                 return rc;
932
933         if (count) {
934                 mask = count - 1;
935                 /* mask = roundup_pow_of_two(count) - 1;
936                  * without overflows...
937                  */
938                 while ((mask | (mask >> 1)) != mask)
939                         mask |= (mask >> 1);
940                 /* On 64 bit arches, must check mask fits in table->mask (u32),
941                  * and on 32bit arches, must check
942                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
943                  */
944 #if BITS_PER_LONG > 32
945                 if (mask > (unsigned long)(u32)mask)
946                         return -EINVAL;
947 #else
948                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
949                                 / sizeof(struct rps_dev_flow)) {
950                         /* Enforce a limit to prevent overflow */
951                         return -EINVAL;
952                 }
953 #endif
954                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
955                 if (!table)
956                         return -ENOMEM;
957
958                 table->mask = mask;
959                 for (count = 0; count <= mask; count++)
960                         table->flows[count].cpu = RPS_NO_CPU;
961         } else {
962                 table = NULL;
963         }
964
965         spin_lock(&rps_dev_flow_lock);
966         old_table = rcu_dereference_protected(queue->rps_flow_table,
967                                               lockdep_is_held(&rps_dev_flow_lock));
968         rcu_assign_pointer(queue->rps_flow_table, table);
969         spin_unlock(&rps_dev_flow_lock);
970
971         if (old_table)
972                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
973
974         return len;
975 }
976
977 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
978         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
979
980 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
981         = __ATTR(rps_flow_cnt, 0644,
982                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
983 #endif /* CONFIG_RPS */
984
985 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
986 #ifdef CONFIG_RPS
987         &rps_cpus_attribute.attr,
988         &rps_dev_flow_table_cnt_attribute.attr,
989 #endif
990         NULL
991 };
992 ATTRIBUTE_GROUPS(rx_queue_default);
993
994 static void rx_queue_release(struct kobject *kobj)
995 {
996         struct netdev_rx_queue *queue = to_rx_queue(kobj);
997 #ifdef CONFIG_RPS
998         struct rps_map *map;
999         struct rps_dev_flow_table *flow_table;
1000
1001         map = rcu_dereference_protected(queue->rps_map, 1);
1002         if (map) {
1003                 RCU_INIT_POINTER(queue->rps_map, NULL);
1004                 kfree_rcu(map, rcu);
1005         }
1006
1007         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1008         if (flow_table) {
1009                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1010                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1011         }
1012 #endif
1013
1014         memset(kobj, 0, sizeof(*kobj));
1015         dev_put(queue->dev);
1016 }
1017
1018 static const void *rx_queue_namespace(struct kobject *kobj)
1019 {
1020         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1021         struct device *dev = &queue->dev->dev;
1022         const void *ns = NULL;
1023
1024         if (dev->class && dev->class->ns_type)
1025                 ns = dev->class->namespace(dev);
1026
1027         return ns;
1028 }
1029
1030 static void rx_queue_get_ownership(struct kobject *kobj,
1031                                    kuid_t *uid, kgid_t *gid)
1032 {
1033         const struct net *net = rx_queue_namespace(kobj);
1034
1035         net_ns_get_ownership(net, uid, gid);
1036 }
1037
1038 static struct kobj_type rx_queue_ktype __ro_after_init = {
1039         .sysfs_ops = &rx_queue_sysfs_ops,
1040         .release = rx_queue_release,
1041         .default_groups = rx_queue_default_groups,
1042         .namespace = rx_queue_namespace,
1043         .get_ownership = rx_queue_get_ownership,
1044 };
1045
1046 static int rx_queue_add_kobject(struct net_device *dev, int index)
1047 {
1048         struct netdev_rx_queue *queue = dev->_rx + index;
1049         struct kobject *kobj = &queue->kobj;
1050         int error = 0;
1051
1052         /* Kobject_put later will trigger rx_queue_release call which
1053          * decreases dev refcount: Take that reference here
1054          */
1055         dev_hold(queue->dev);
1056
1057         kobj->kset = dev->queues_kset;
1058         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1059                                      "rx-%u", index);
1060         if (error)
1061                 goto err;
1062
1063         if (dev->sysfs_rx_queue_group) {
1064                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1065                 if (error)
1066                         goto err;
1067         }
1068
1069         kobject_uevent(kobj, KOBJ_ADD);
1070
1071         return error;
1072
1073 err:
1074         kobject_put(kobj);
1075         return error;
1076 }
1077
1078 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1079                                  kgid_t kgid)
1080 {
1081         struct netdev_rx_queue *queue = dev->_rx + index;
1082         struct kobject *kobj = &queue->kobj;
1083         int error;
1084
1085         error = sysfs_change_owner(kobj, kuid, kgid);
1086         if (error)
1087                 return error;
1088
1089         if (dev->sysfs_rx_queue_group)
1090                 error = sysfs_group_change_owner(
1091                         kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1092
1093         return error;
1094 }
1095 #endif /* CONFIG_SYSFS */
1096
1097 int
1098 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1099 {
1100 #ifdef CONFIG_SYSFS
1101         int i;
1102         int error = 0;
1103
1104 #ifndef CONFIG_RPS
1105         if (!dev->sysfs_rx_queue_group)
1106                 return 0;
1107 #endif
1108         for (i = old_num; i < new_num; i++) {
1109                 error = rx_queue_add_kobject(dev, i);
1110                 if (error) {
1111                         new_num = old_num;
1112                         break;
1113                 }
1114         }
1115
1116         while (--i >= new_num) {
1117                 struct kobject *kobj = &dev->_rx[i].kobj;
1118
1119                 if (!refcount_read(&dev_net(dev)->ns.count))
1120                         kobj->uevent_suppress = 1;
1121                 if (dev->sysfs_rx_queue_group)
1122                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1123                 kobject_put(kobj);
1124         }
1125
1126         return error;
1127 #else
1128         return 0;
1129 #endif
1130 }
1131
1132 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1133                                      kuid_t kuid, kgid_t kgid)
1134 {
1135 #ifdef CONFIG_SYSFS
1136         int error = 0;
1137         int i;
1138
1139 #ifndef CONFIG_RPS
1140         if (!dev->sysfs_rx_queue_group)
1141                 return 0;
1142 #endif
1143         for (i = 0; i < num; i++) {
1144                 error = rx_queue_change_owner(dev, i, kuid, kgid);
1145                 if (error)
1146                         break;
1147         }
1148
1149         return error;
1150 #else
1151         return 0;
1152 #endif
1153 }
1154
1155 #ifdef CONFIG_SYSFS
1156 /*
1157  * netdev_queue sysfs structures and functions.
1158  */
1159 struct netdev_queue_attribute {
1160         struct attribute attr;
1161         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1162         ssize_t (*store)(struct netdev_queue *queue,
1163                          const char *buf, size_t len);
1164 };
1165 #define to_netdev_queue_attr(_attr) \
1166         container_of(_attr, struct netdev_queue_attribute, attr)
1167
1168 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1169
1170 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1171                                       struct attribute *attr, char *buf)
1172 {
1173         const struct netdev_queue_attribute *attribute
1174                 = to_netdev_queue_attr(attr);
1175         struct netdev_queue *queue = to_netdev_queue(kobj);
1176
1177         if (!attribute->show)
1178                 return -EIO;
1179
1180         return attribute->show(queue, buf);
1181 }
1182
1183 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1184                                        struct attribute *attr,
1185                                        const char *buf, size_t count)
1186 {
1187         const struct netdev_queue_attribute *attribute
1188                 = to_netdev_queue_attr(attr);
1189         struct netdev_queue *queue = to_netdev_queue(kobj);
1190
1191         if (!attribute->store)
1192                 return -EIO;
1193
1194         return attribute->store(queue, buf, count);
1195 }
1196
1197 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1198         .show = netdev_queue_attr_show,
1199         .store = netdev_queue_attr_store,
1200 };
1201
1202 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1203 {
1204         unsigned long trans_timeout;
1205
1206         spin_lock_irq(&queue->_xmit_lock);
1207         trans_timeout = queue->trans_timeout;
1208         spin_unlock_irq(&queue->_xmit_lock);
1209
1210         return sprintf(buf, fmt_ulong, trans_timeout);
1211 }
1212
1213 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1214 {
1215         struct net_device *dev = queue->dev;
1216         unsigned int i;
1217
1218         i = queue - dev->_tx;
1219         BUG_ON(i >= dev->num_tx_queues);
1220
1221         return i;
1222 }
1223
1224 static ssize_t traffic_class_show(struct netdev_queue *queue,
1225                                   char *buf)
1226 {
1227         struct net_device *dev = queue->dev;
1228         int num_tc, tc;
1229         int index;
1230
1231         if (!netif_is_multiqueue(dev))
1232                 return -ENOENT;
1233
1234         if (!rtnl_trylock())
1235                 return restart_syscall();
1236
1237         index = get_netdev_queue_index(queue);
1238
1239         /* If queue belongs to subordinate dev use its TC mapping */
1240         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1241
1242         num_tc = dev->num_tc;
1243         tc = netdev_txq_to_tc(dev, index);
1244
1245         rtnl_unlock();
1246
1247         if (tc < 0)
1248                 return -EINVAL;
1249
1250         /* We can report the traffic class one of two ways:
1251          * Subordinate device traffic classes are reported with the traffic
1252          * class first, and then the subordinate class so for example TC0 on
1253          * subordinate device 2 will be reported as "0-2". If the queue
1254          * belongs to the root device it will be reported with just the
1255          * traffic class, so just "0" for TC 0 for example.
1256          */
1257         return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1258                             sprintf(buf, "%d\n", tc);
1259 }
1260
1261 #ifdef CONFIG_XPS
1262 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1263                                char *buf)
1264 {
1265         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1266 }
1267
1268 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1269                                 const char *buf, size_t len)
1270 {
1271         struct net_device *dev = queue->dev;
1272         int err, index = get_netdev_queue_index(queue);
1273         u32 rate = 0;
1274
1275         if (!capable(CAP_NET_ADMIN))
1276                 return -EPERM;
1277
1278         /* The check is also done later; this helps returning early without
1279          * hitting the trylock/restart below.
1280          */
1281         if (!dev->netdev_ops->ndo_set_tx_maxrate)
1282                 return -EOPNOTSUPP;
1283
1284         err = kstrtou32(buf, 10, &rate);
1285         if (err < 0)
1286                 return err;
1287
1288         if (!rtnl_trylock())
1289                 return restart_syscall();
1290
1291         err = -EOPNOTSUPP;
1292         if (dev->netdev_ops->ndo_set_tx_maxrate)
1293                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1294
1295         rtnl_unlock();
1296         if (!err) {
1297                 queue->tx_maxrate = rate;
1298                 return len;
1299         }
1300         return err;
1301 }
1302
1303 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1304         = __ATTR_RW(tx_maxrate);
1305 #endif
1306
1307 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1308         = __ATTR_RO(tx_timeout);
1309
1310 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1311         = __ATTR_RO(traffic_class);
1312
1313 #ifdef CONFIG_BQL
1314 /*
1315  * Byte queue limits sysfs structures and functions.
1316  */
1317 static ssize_t bql_show(char *buf, unsigned int value)
1318 {
1319         return sprintf(buf, "%u\n", value);
1320 }
1321
1322 static ssize_t bql_set(const char *buf, const size_t count,
1323                        unsigned int *pvalue)
1324 {
1325         unsigned int value;
1326         int err;
1327
1328         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1329                 value = DQL_MAX_LIMIT;
1330         } else {
1331                 err = kstrtouint(buf, 10, &value);
1332                 if (err < 0)
1333                         return err;
1334                 if (value > DQL_MAX_LIMIT)
1335                         return -EINVAL;
1336         }
1337
1338         *pvalue = value;
1339
1340         return count;
1341 }
1342
1343 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1344                                   char *buf)
1345 {
1346         struct dql *dql = &queue->dql;
1347
1348         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1349 }
1350
1351 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1352                                  const char *buf, size_t len)
1353 {
1354         struct dql *dql = &queue->dql;
1355         unsigned int value;
1356         int err;
1357
1358         err = kstrtouint(buf, 10, &value);
1359         if (err < 0)
1360                 return err;
1361
1362         dql->slack_hold_time = msecs_to_jiffies(value);
1363
1364         return len;
1365 }
1366
1367 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1368         = __ATTR(hold_time, 0644,
1369                  bql_show_hold_time, bql_set_hold_time);
1370
1371 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1372                                  char *buf)
1373 {
1374         struct dql *dql = &queue->dql;
1375
1376         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1377 }
1378
1379 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1380         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1381
1382 #define BQL_ATTR(NAME, FIELD)                                           \
1383 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1384                                  char *buf)                             \
1385 {                                                                       \
1386         return bql_show(buf, queue->dql.FIELD);                         \
1387 }                                                                       \
1388                                                                         \
1389 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1390                                 const char *buf, size_t len)            \
1391 {                                                                       \
1392         return bql_set(buf, len, &queue->dql.FIELD);                    \
1393 }                                                                       \
1394                                                                         \
1395 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1396         = __ATTR(NAME, 0644,                            \
1397                  bql_show_ ## NAME, bql_set_ ## NAME)
1398
1399 BQL_ATTR(limit, limit);
1400 BQL_ATTR(limit_max, max_limit);
1401 BQL_ATTR(limit_min, min_limit);
1402
1403 static struct attribute *dql_attrs[] __ro_after_init = {
1404         &bql_limit_attribute.attr,
1405         &bql_limit_max_attribute.attr,
1406         &bql_limit_min_attribute.attr,
1407         &bql_hold_time_attribute.attr,
1408         &bql_inflight_attribute.attr,
1409         NULL
1410 };
1411
1412 static const struct attribute_group dql_group = {
1413         .name  = "byte_queue_limits",
1414         .attrs  = dql_attrs,
1415 };
1416 #endif /* CONFIG_BQL */
1417
1418 #ifdef CONFIG_XPS
1419 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1420                               int tc, char *buf, enum xps_map_type type)
1421 {
1422         struct xps_dev_maps *dev_maps;
1423         unsigned long *mask;
1424         unsigned int nr_ids;
1425         int j, len;
1426
1427         rcu_read_lock();
1428         dev_maps = rcu_dereference(dev->xps_maps[type]);
1429
1430         /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1431          * when dev_maps hasn't been allocated yet, to be backward compatible.
1432          */
1433         nr_ids = dev_maps ? dev_maps->nr_ids :
1434                  (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1435
1436         mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1437         if (!mask) {
1438                 rcu_read_unlock();
1439                 return -ENOMEM;
1440         }
1441
1442         if (!dev_maps || tc >= dev_maps->num_tc)
1443                 goto out_no_maps;
1444
1445         for (j = 0; j < nr_ids; j++) {
1446                 int i, tci = j * dev_maps->num_tc + tc;
1447                 struct xps_map *map;
1448
1449                 map = rcu_dereference(dev_maps->attr_map[tci]);
1450                 if (!map)
1451                         continue;
1452
1453                 for (i = map->len; i--;) {
1454                         if (map->queues[i] == index) {
1455                                 set_bit(j, mask);
1456                                 break;
1457                         }
1458                 }
1459         }
1460 out_no_maps:
1461         rcu_read_unlock();
1462
1463         len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1464         bitmap_free(mask);
1465
1466         return len < PAGE_SIZE ? len : -EINVAL;
1467 }
1468
1469 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1470 {
1471         struct net_device *dev = queue->dev;
1472         unsigned int index;
1473         int len, tc;
1474
1475         if (!netif_is_multiqueue(dev))
1476                 return -ENOENT;
1477
1478         index = get_netdev_queue_index(queue);
1479
1480         if (!rtnl_trylock())
1481                 return restart_syscall();
1482
1483         /* If queue belongs to subordinate dev use its map */
1484         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1485
1486         tc = netdev_txq_to_tc(dev, index);
1487         if (tc < 0) {
1488                 rtnl_unlock();
1489                 return -EINVAL;
1490         }
1491
1492         /* Make sure the subordinate device can't be freed */
1493         get_device(&dev->dev);
1494         rtnl_unlock();
1495
1496         len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1497
1498         put_device(&dev->dev);
1499         return len;
1500 }
1501
1502 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1503                               const char *buf, size_t len)
1504 {
1505         struct net_device *dev = queue->dev;
1506         unsigned int index;
1507         cpumask_var_t mask;
1508         int err;
1509
1510         if (!netif_is_multiqueue(dev))
1511                 return -ENOENT;
1512
1513         if (!capable(CAP_NET_ADMIN))
1514                 return -EPERM;
1515
1516         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1517                 return -ENOMEM;
1518
1519         index = get_netdev_queue_index(queue);
1520
1521         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1522         if (err) {
1523                 free_cpumask_var(mask);
1524                 return err;
1525         }
1526
1527         if (!rtnl_trylock()) {
1528                 free_cpumask_var(mask);
1529                 return restart_syscall();
1530         }
1531
1532         err = netif_set_xps_queue(dev, mask, index);
1533         rtnl_unlock();
1534
1535         free_cpumask_var(mask);
1536
1537         return err ? : len;
1538 }
1539
1540 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1541         = __ATTR_RW(xps_cpus);
1542
1543 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1544 {
1545         struct net_device *dev = queue->dev;
1546         unsigned int index;
1547         int tc;
1548
1549         index = get_netdev_queue_index(queue);
1550
1551         if (!rtnl_trylock())
1552                 return restart_syscall();
1553
1554         tc = netdev_txq_to_tc(dev, index);
1555         rtnl_unlock();
1556         if (tc < 0)
1557                 return -EINVAL;
1558
1559         return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1560 }
1561
1562 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1563                               size_t len)
1564 {
1565         struct net_device *dev = queue->dev;
1566         struct net *net = dev_net(dev);
1567         unsigned long *mask;
1568         unsigned int index;
1569         int err;
1570
1571         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1572                 return -EPERM;
1573
1574         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1575         if (!mask)
1576                 return -ENOMEM;
1577
1578         index = get_netdev_queue_index(queue);
1579
1580         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1581         if (err) {
1582                 bitmap_free(mask);
1583                 return err;
1584         }
1585
1586         if (!rtnl_trylock()) {
1587                 bitmap_free(mask);
1588                 return restart_syscall();
1589         }
1590
1591         cpus_read_lock();
1592         err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1593         cpus_read_unlock();
1594
1595         rtnl_unlock();
1596
1597         bitmap_free(mask);
1598         return err ? : len;
1599 }
1600
1601 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1602         = __ATTR_RW(xps_rxqs);
1603 #endif /* CONFIG_XPS */
1604
1605 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1606         &queue_trans_timeout.attr,
1607         &queue_traffic_class.attr,
1608 #ifdef CONFIG_XPS
1609         &xps_cpus_attribute.attr,
1610         &xps_rxqs_attribute.attr,
1611         &queue_tx_maxrate.attr,
1612 #endif
1613         NULL
1614 };
1615 ATTRIBUTE_GROUPS(netdev_queue_default);
1616
1617 static void netdev_queue_release(struct kobject *kobj)
1618 {
1619         struct netdev_queue *queue = to_netdev_queue(kobj);
1620
1621         memset(kobj, 0, sizeof(*kobj));
1622         dev_put(queue->dev);
1623 }
1624
1625 static const void *netdev_queue_namespace(struct kobject *kobj)
1626 {
1627         struct netdev_queue *queue = to_netdev_queue(kobj);
1628         struct device *dev = &queue->dev->dev;
1629         const void *ns = NULL;
1630
1631         if (dev->class && dev->class->ns_type)
1632                 ns = dev->class->namespace(dev);
1633
1634         return ns;
1635 }
1636
1637 static void netdev_queue_get_ownership(struct kobject *kobj,
1638                                        kuid_t *uid, kgid_t *gid)
1639 {
1640         const struct net *net = netdev_queue_namespace(kobj);
1641
1642         net_ns_get_ownership(net, uid, gid);
1643 }
1644
1645 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1646         .sysfs_ops = &netdev_queue_sysfs_ops,
1647         .release = netdev_queue_release,
1648         .default_groups = netdev_queue_default_groups,
1649         .namespace = netdev_queue_namespace,
1650         .get_ownership = netdev_queue_get_ownership,
1651 };
1652
1653 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1654 {
1655         struct netdev_queue *queue = dev->_tx + index;
1656         struct kobject *kobj = &queue->kobj;
1657         int error = 0;
1658
1659         /* Kobject_put later will trigger netdev_queue_release call
1660          * which decreases dev refcount: Take that reference here
1661          */
1662         dev_hold(queue->dev);
1663
1664         kobj->kset = dev->queues_kset;
1665         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1666                                      "tx-%u", index);
1667         if (error)
1668                 goto err;
1669
1670 #ifdef CONFIG_BQL
1671         error = sysfs_create_group(kobj, &dql_group);
1672         if (error)
1673                 goto err;
1674 #endif
1675
1676         kobject_uevent(kobj, KOBJ_ADD);
1677         return 0;
1678
1679 err:
1680         kobject_put(kobj);
1681         return error;
1682 }
1683
1684 static int tx_queue_change_owner(struct net_device *ndev, int index,
1685                                  kuid_t kuid, kgid_t kgid)
1686 {
1687         struct netdev_queue *queue = ndev->_tx + index;
1688         struct kobject *kobj = &queue->kobj;
1689         int error;
1690
1691         error = sysfs_change_owner(kobj, kuid, kgid);
1692         if (error)
1693                 return error;
1694
1695 #ifdef CONFIG_BQL
1696         error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1697 #endif
1698         return error;
1699 }
1700 #endif /* CONFIG_SYSFS */
1701
1702 int
1703 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1704 {
1705 #ifdef CONFIG_SYSFS
1706         int i;
1707         int error = 0;
1708
1709         for (i = old_num; i < new_num; i++) {
1710                 error = netdev_queue_add_kobject(dev, i);
1711                 if (error) {
1712                         new_num = old_num;
1713                         break;
1714                 }
1715         }
1716
1717         while (--i >= new_num) {
1718                 struct netdev_queue *queue = dev->_tx + i;
1719
1720                 if (!refcount_read(&dev_net(dev)->ns.count))
1721                         queue->kobj.uevent_suppress = 1;
1722 #ifdef CONFIG_BQL
1723                 sysfs_remove_group(&queue->kobj, &dql_group);
1724 #endif
1725                 kobject_put(&queue->kobj);
1726         }
1727
1728         return error;
1729 #else
1730         return 0;
1731 #endif /* CONFIG_SYSFS */
1732 }
1733
1734 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1735                                      kuid_t kuid, kgid_t kgid)
1736 {
1737 #ifdef CONFIG_SYSFS
1738         int error = 0;
1739         int i;
1740
1741         for (i = 0; i < num; i++) {
1742                 error = tx_queue_change_owner(dev, i, kuid, kgid);
1743                 if (error)
1744                         break;
1745         }
1746
1747         return error;
1748 #else
1749         return 0;
1750 #endif /* CONFIG_SYSFS */
1751 }
1752
1753 static int register_queue_kobjects(struct net_device *dev)
1754 {
1755         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1756
1757 #ifdef CONFIG_SYSFS
1758         dev->queues_kset = kset_create_and_add("queues",
1759                                                NULL, &dev->dev.kobj);
1760         if (!dev->queues_kset)
1761                 return -ENOMEM;
1762         real_rx = dev->real_num_rx_queues;
1763 #endif
1764         real_tx = dev->real_num_tx_queues;
1765
1766         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1767         if (error)
1768                 goto error;
1769         rxq = real_rx;
1770
1771         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1772         if (error)
1773                 goto error;
1774         txq = real_tx;
1775
1776         return 0;
1777
1778 error:
1779         netdev_queue_update_kobjects(dev, txq, 0);
1780         net_rx_queue_update_kobjects(dev, rxq, 0);
1781 #ifdef CONFIG_SYSFS
1782         kset_unregister(dev->queues_kset);
1783 #endif
1784         return error;
1785 }
1786
1787 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1788 {
1789         int error = 0, real_rx = 0, real_tx = 0;
1790
1791 #ifdef CONFIG_SYSFS
1792         if (ndev->queues_kset) {
1793                 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1794                 if (error)
1795                         return error;
1796         }
1797         real_rx = ndev->real_num_rx_queues;
1798 #endif
1799         real_tx = ndev->real_num_tx_queues;
1800
1801         error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1802         if (error)
1803                 return error;
1804
1805         error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1806         if (error)
1807                 return error;
1808
1809         return 0;
1810 }
1811
1812 static void remove_queue_kobjects(struct net_device *dev)
1813 {
1814         int real_rx = 0, real_tx = 0;
1815
1816 #ifdef CONFIG_SYSFS
1817         real_rx = dev->real_num_rx_queues;
1818 #endif
1819         real_tx = dev->real_num_tx_queues;
1820
1821         net_rx_queue_update_kobjects(dev, real_rx, 0);
1822         netdev_queue_update_kobjects(dev, real_tx, 0);
1823 #ifdef CONFIG_SYSFS
1824         kset_unregister(dev->queues_kset);
1825 #endif
1826 }
1827
1828 static bool net_current_may_mount(void)
1829 {
1830         struct net *net = current->nsproxy->net_ns;
1831
1832         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1833 }
1834
1835 static void *net_grab_current_ns(void)
1836 {
1837         struct net *ns = current->nsproxy->net_ns;
1838 #ifdef CONFIG_NET_NS
1839         if (ns)
1840                 refcount_inc(&ns->passive);
1841 #endif
1842         return ns;
1843 }
1844
1845 static const void *net_initial_ns(void)
1846 {
1847         return &init_net;
1848 }
1849
1850 static const void *net_netlink_ns(struct sock *sk)
1851 {
1852         return sock_net(sk);
1853 }
1854
1855 const struct kobj_ns_type_operations net_ns_type_operations = {
1856         .type = KOBJ_NS_TYPE_NET,
1857         .current_may_mount = net_current_may_mount,
1858         .grab_current_ns = net_grab_current_ns,
1859         .netlink_ns = net_netlink_ns,
1860         .initial_ns = net_initial_ns,
1861         .drop_ns = net_drop_ns,
1862 };
1863 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1864
1865 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1866 {
1867         struct net_device *dev = to_net_dev(d);
1868         int retval;
1869
1870         /* pass interface to uevent. */
1871         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1872         if (retval)
1873                 goto exit;
1874
1875         /* pass ifindex to uevent.
1876          * ifindex is useful as it won't change (interface name may change)
1877          * and is what RtNetlink uses natively.
1878          */
1879         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1880
1881 exit:
1882         return retval;
1883 }
1884
1885 /*
1886  *      netdev_release -- destroy and free a dead device.
1887  *      Called when last reference to device kobject is gone.
1888  */
1889 static void netdev_release(struct device *d)
1890 {
1891         struct net_device *dev = to_net_dev(d);
1892
1893         BUG_ON(dev->reg_state != NETREG_RELEASED);
1894
1895         /* no need to wait for rcu grace period:
1896          * device is dead and about to be freed.
1897          */
1898         kfree(rcu_access_pointer(dev->ifalias));
1899         netdev_freemem(dev);
1900 }
1901
1902 static const void *net_namespace(struct device *d)
1903 {
1904         struct net_device *dev = to_net_dev(d);
1905
1906         return dev_net(dev);
1907 }
1908
1909 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1910 {
1911         struct net_device *dev = to_net_dev(d);
1912         const struct net *net = dev_net(dev);
1913
1914         net_ns_get_ownership(net, uid, gid);
1915 }
1916
1917 static struct class net_class __ro_after_init = {
1918         .name = "net",
1919         .dev_release = netdev_release,
1920         .dev_groups = net_class_groups,
1921         .dev_uevent = netdev_uevent,
1922         .ns_type = &net_ns_type_operations,
1923         .namespace = net_namespace,
1924         .get_ownership = net_get_ownership,
1925 };
1926
1927 #ifdef CONFIG_OF
1928 static int of_dev_node_match(struct device *dev, const void *data)
1929 {
1930         for (; dev; dev = dev->parent) {
1931                 if (dev->of_node == data)
1932                         return 1;
1933         }
1934
1935         return 0;
1936 }
1937
1938 /*
1939  * of_find_net_device_by_node - lookup the net device for the device node
1940  * @np: OF device node
1941  *
1942  * Looks up the net_device structure corresponding with the device node.
1943  * If successful, returns a pointer to the net_device with the embedded
1944  * struct device refcount incremented by one, or NULL on failure. The
1945  * refcount must be dropped when done with the net_device.
1946  */
1947 struct net_device *of_find_net_device_by_node(struct device_node *np)
1948 {
1949         struct device *dev;
1950
1951         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1952         if (!dev)
1953                 return NULL;
1954
1955         return to_net_dev(dev);
1956 }
1957 EXPORT_SYMBOL(of_find_net_device_by_node);
1958 #endif
1959
1960 /* Delete sysfs entries but hold kobject reference until after all
1961  * netdev references are gone.
1962  */
1963 void netdev_unregister_kobject(struct net_device *ndev)
1964 {
1965         struct device *dev = &ndev->dev;
1966
1967         if (!refcount_read(&dev_net(ndev)->ns.count))
1968                 dev_set_uevent_suppress(dev, 1);
1969
1970         kobject_get(&dev->kobj);
1971
1972         remove_queue_kobjects(ndev);
1973
1974         pm_runtime_set_memalloc_noio(dev, false);
1975
1976         device_del(dev);
1977 }
1978
1979 /* Create sysfs entries for network device. */
1980 int netdev_register_kobject(struct net_device *ndev)
1981 {
1982         struct device *dev = &ndev->dev;
1983         const struct attribute_group **groups = ndev->sysfs_groups;
1984         int error = 0;
1985
1986         device_initialize(dev);
1987         dev->class = &net_class;
1988         dev->platform_data = ndev;
1989         dev->groups = groups;
1990
1991         dev_set_name(dev, "%s", ndev->name);
1992
1993 #ifdef CONFIG_SYSFS
1994         /* Allow for a device specific group */
1995         if (*groups)
1996                 groups++;
1997
1998         *groups++ = &netstat_group;
1999
2000 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
2001         if (ndev->ieee80211_ptr)
2002                 *groups++ = &wireless_group;
2003 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
2004         else if (ndev->wireless_handlers)
2005                 *groups++ = &wireless_group;
2006 #endif
2007 #endif
2008 #endif /* CONFIG_SYSFS */
2009
2010         error = device_add(dev);
2011         if (error)
2012                 return error;
2013
2014         error = register_queue_kobjects(ndev);
2015         if (error) {
2016                 device_del(dev);
2017                 return error;
2018         }
2019
2020         pm_runtime_set_memalloc_noio(dev, true);
2021
2022         return error;
2023 }
2024
2025 /* Change owner for sysfs entries when moving network devices across network
2026  * namespaces owned by different user namespaces.
2027  */
2028 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2029                         const struct net *net_new)
2030 {
2031         struct device *dev = &ndev->dev;
2032         kuid_t old_uid, new_uid;
2033         kgid_t old_gid, new_gid;
2034         int error;
2035
2036         net_ns_get_ownership(net_old, &old_uid, &old_gid);
2037         net_ns_get_ownership(net_new, &new_uid, &new_gid);
2038
2039         /* The network namespace was changed but the owning user namespace is
2040          * identical so there's no need to change the owner of sysfs entries.
2041          */
2042         if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2043                 return 0;
2044
2045         error = device_change_owner(dev, new_uid, new_gid);
2046         if (error)
2047                 return error;
2048
2049         error = queue_change_owner(ndev, new_uid, new_gid);
2050         if (error)
2051                 return error;
2052
2053         return 0;
2054 }
2055
2056 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2057                                 const void *ns)
2058 {
2059         return class_create_file_ns(&net_class, class_attr, ns);
2060 }
2061 EXPORT_SYMBOL(netdev_class_create_file_ns);
2062
2063 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2064                                  const void *ns)
2065 {
2066         class_remove_file_ns(&net_class, class_attr, ns);
2067 }
2068 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2069
2070 int __init netdev_kobject_init(void)
2071 {
2072         kobj_ns_type_register(&net_ns_type_operations);
2073         return class_register(&net_class);
2074 }