GNU Linux-libre 5.15.137-gnu
[releases.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 /* Caller holds RTNL or dev_base_lock */
36 static inline int dev_isalive(const struct net_device *dev)
37 {
38         return dev->reg_state <= NETREG_REGISTERED;
39 }
40
41 /* use same locking rules as GIF* ioctl's */
42 static ssize_t netdev_show(const struct device *dev,
43                            struct device_attribute *attr, char *buf,
44                            ssize_t (*format)(const struct net_device *, char *))
45 {
46         struct net_device *ndev = to_net_dev(dev);
47         ssize_t ret = -EINVAL;
48
49         read_lock(&dev_base_lock);
50         if (dev_isalive(ndev))
51                 ret = (*format)(ndev, buf);
52         read_unlock(&dev_base_lock);
53
54         return ret;
55 }
56
57 /* generate a show function for simple field */
58 #define NETDEVICE_SHOW(field, format_string)                            \
59 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
60 {                                                                       \
61         return sprintf(buf, format_string, dev->field);                 \
62 }                                                                       \
63 static ssize_t field##_show(struct device *dev,                         \
64                             struct device_attribute *attr, char *buf)   \
65 {                                                                       \
66         return netdev_show(dev, attr, buf, format_##field);             \
67 }                                                                       \
68
69 #define NETDEVICE_SHOW_RO(field, format_string)                         \
70 NETDEVICE_SHOW(field, format_string);                                   \
71 static DEVICE_ATTR_RO(field)
72
73 #define NETDEVICE_SHOW_RW(field, format_string)                         \
74 NETDEVICE_SHOW(field, format_string);                                   \
75 static DEVICE_ATTR_RW(field)
76
77 /* use same locking and permission rules as SIF* ioctl's */
78 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
79                             const char *buf, size_t len,
80                             int (*set)(struct net_device *, unsigned long))
81 {
82         struct net_device *netdev = to_net_dev(dev);
83         struct net *net = dev_net(netdev);
84         unsigned long new;
85         int ret;
86
87         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
88                 return -EPERM;
89
90         ret = kstrtoul(buf, 0, &new);
91         if (ret)
92                 goto err;
93
94         if (!rtnl_trylock())
95                 return restart_syscall();
96
97         if (dev_isalive(netdev)) {
98                 ret = (*set)(netdev, new);
99                 if (ret == 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
163         if (dev_isalive(ndev))
164                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
165         return -EINVAL;
166 }
167 static DEVICE_ATTR_RO(broadcast);
168
169 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
170 {
171         if (!netif_running(dev))
172                 return -EINVAL;
173         return dev_change_carrier(dev, (bool)new_carrier);
174 }
175
176 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
177                              const char *buf, size_t len)
178 {
179         struct net_device *netdev = to_net_dev(dev);
180
181         /* The check is also done in change_carrier; this helps returning early
182          * without hitting the trylock/restart in netdev_store.
183          */
184         if (!netdev->netdev_ops->ndo_change_carrier)
185                 return -EOPNOTSUPP;
186
187         return netdev_store(dev, attr, buf, len, change_carrier);
188 }
189
190 static ssize_t carrier_show(struct device *dev,
191                             struct device_attribute *attr, char *buf)
192 {
193         struct net_device *netdev = to_net_dev(dev);
194
195         if (netif_running(netdev))
196                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
197
198         return -EINVAL;
199 }
200 static DEVICE_ATTR_RW(carrier);
201
202 static ssize_t speed_show(struct device *dev,
203                           struct device_attribute *attr, char *buf)
204 {
205         struct net_device *netdev = to_net_dev(dev);
206         int ret = -EINVAL;
207
208         /* The check is also done in __ethtool_get_link_ksettings; this helps
209          * returning early without hitting the trylock/restart below.
210          */
211         if (!netdev->ethtool_ops->get_link_ksettings)
212                 return ret;
213
214         if (!rtnl_trylock())
215                 return restart_syscall();
216
217         if (netif_running(netdev) && netif_device_present(netdev)) {
218                 struct ethtool_link_ksettings cmd;
219
220                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
221                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
222         }
223         rtnl_unlock();
224         return ret;
225 }
226 static DEVICE_ATTR_RO(speed);
227
228 static ssize_t duplex_show(struct device *dev,
229                            struct device_attribute *attr, char *buf)
230 {
231         struct net_device *netdev = to_net_dev(dev);
232         int ret = -EINVAL;
233
234         /* The check is also done in __ethtool_get_link_ksettings; this helps
235          * returning early without hitting the trylock/restart below.
236          */
237         if (!netdev->ethtool_ops->get_link_ksettings)
238                 return ret;
239
240         if (!rtnl_trylock())
241                 return restart_syscall();
242
243         if (netif_running(netdev)) {
244                 struct ethtool_link_ksettings cmd;
245
246                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
247                         const char *duplex;
248
249                         switch (cmd.base.duplex) {
250                         case DUPLEX_HALF:
251                                 duplex = "half";
252                                 break;
253                         case DUPLEX_FULL:
254                                 duplex = "full";
255                                 break;
256                         default:
257                                 duplex = "unknown";
258                                 break;
259                         }
260                         ret = sprintf(buf, "%s\n", duplex);
261                 }
262         }
263         rtnl_unlock();
264         return ret;
265 }
266 static DEVICE_ATTR_RO(duplex);
267
268 static ssize_t testing_show(struct device *dev,
269                             struct device_attribute *attr, char *buf)
270 {
271         struct net_device *netdev = to_net_dev(dev);
272
273         if (netif_running(netdev))
274                 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
275
276         return -EINVAL;
277 }
278 static DEVICE_ATTR_RO(testing);
279
280 static ssize_t dormant_show(struct device *dev,
281                             struct device_attribute *attr, char *buf)
282 {
283         struct net_device *netdev = to_net_dev(dev);
284
285         if (netif_running(netdev))
286                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
287
288         return -EINVAL;
289 }
290 static DEVICE_ATTR_RO(dormant);
291
292 static const char *const operstates[] = {
293         "unknown",
294         "notpresent", /* currently unused */
295         "down",
296         "lowerlayerdown",
297         "testing",
298         "dormant",
299         "up"
300 };
301
302 static ssize_t operstate_show(struct device *dev,
303                               struct device_attribute *attr, char *buf)
304 {
305         const struct net_device *netdev = to_net_dev(dev);
306         unsigned char operstate;
307
308         read_lock(&dev_base_lock);
309         operstate = netdev->operstate;
310         if (!netif_running(netdev))
311                 operstate = IF_OPER_DOWN;
312         read_unlock(&dev_base_lock);
313
314         if (operstate >= ARRAY_SIZE(operstates))
315                 return -EINVAL; /* should not happen */
316
317         return sprintf(buf, "%s\n", operstates[operstate]);
318 }
319 static DEVICE_ATTR_RO(operstate);
320
321 static ssize_t carrier_changes_show(struct device *dev,
322                                     struct device_attribute *attr,
323                                     char *buf)
324 {
325         struct net_device *netdev = to_net_dev(dev);
326
327         return sprintf(buf, fmt_dec,
328                        atomic_read(&netdev->carrier_up_count) +
329                        atomic_read(&netdev->carrier_down_count));
330 }
331 static DEVICE_ATTR_RO(carrier_changes);
332
333 static ssize_t carrier_up_count_show(struct device *dev,
334                                      struct device_attribute *attr,
335                                      char *buf)
336 {
337         struct net_device *netdev = to_net_dev(dev);
338
339         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
340 }
341 static DEVICE_ATTR_RO(carrier_up_count);
342
343 static ssize_t carrier_down_count_show(struct device *dev,
344                                        struct device_attribute *attr,
345                                        char *buf)
346 {
347         struct net_device *netdev = to_net_dev(dev);
348
349         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
350 }
351 static DEVICE_ATTR_RO(carrier_down_count);
352
353 /* read-write attributes */
354
355 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
356 {
357         return dev_set_mtu(dev, (int)new_mtu);
358 }
359
360 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
361                          const char *buf, size_t len)
362 {
363         return netdev_store(dev, attr, buf, len, change_mtu);
364 }
365 NETDEVICE_SHOW_RW(mtu, fmt_dec);
366
367 static int change_flags(struct net_device *dev, unsigned long new_flags)
368 {
369         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
370 }
371
372 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
373                            const char *buf, size_t len)
374 {
375         return netdev_store(dev, attr, buf, len, change_flags);
376 }
377 NETDEVICE_SHOW_RW(flags, fmt_hex);
378
379 static ssize_t tx_queue_len_store(struct device *dev,
380                                   struct device_attribute *attr,
381                                   const char *buf, size_t len)
382 {
383         if (!capable(CAP_NET_ADMIN))
384                 return -EPERM;
385
386         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
387 }
388 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
389
390 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
391 {
392         WRITE_ONCE(dev->gro_flush_timeout, val);
393         return 0;
394 }
395
396 static ssize_t gro_flush_timeout_store(struct device *dev,
397                                        struct device_attribute *attr,
398                                        const char *buf, size_t len)
399 {
400         if (!capable(CAP_NET_ADMIN))
401                 return -EPERM;
402
403         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
404 }
405 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
406
407 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
408 {
409         WRITE_ONCE(dev->napi_defer_hard_irqs, val);
410         return 0;
411 }
412
413 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
414                                           struct device_attribute *attr,
415                                           const char *buf, size_t len)
416 {
417         if (!capable(CAP_NET_ADMIN))
418                 return -EPERM;
419
420         return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
421 }
422 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
423
424 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
425                              const char *buf, size_t len)
426 {
427         struct net_device *netdev = to_net_dev(dev);
428         struct net *net = dev_net(netdev);
429         size_t count = len;
430         ssize_t ret = 0;
431
432         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
433                 return -EPERM;
434
435         /* ignore trailing newline */
436         if (len >  0 && buf[len - 1] == '\n')
437                 --count;
438
439         if (!rtnl_trylock())
440                 return restart_syscall();
441
442         if (dev_isalive(netdev)) {
443                 ret = dev_set_alias(netdev, buf, count);
444                 if (ret < 0)
445                         goto err;
446                 ret = len;
447                 netdev_state_change(netdev);
448         }
449 err:
450         rtnl_unlock();
451
452         return ret;
453 }
454
455 static ssize_t ifalias_show(struct device *dev,
456                             struct device_attribute *attr, char *buf)
457 {
458         const struct net_device *netdev = to_net_dev(dev);
459         char tmp[IFALIASZ];
460         ssize_t ret = 0;
461
462         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
463         if (ret > 0)
464                 ret = sprintf(buf, "%s\n", tmp);
465         return ret;
466 }
467 static DEVICE_ATTR_RW(ifalias);
468
469 static int change_group(struct net_device *dev, unsigned long new_group)
470 {
471         dev_set_group(dev, (int)new_group);
472         return 0;
473 }
474
475 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
476                            const char *buf, size_t len)
477 {
478         return netdev_store(dev, attr, buf, len, change_group);
479 }
480 NETDEVICE_SHOW(group, fmt_dec);
481 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
482
483 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
484 {
485         return dev_change_proto_down(dev, (bool)proto_down);
486 }
487
488 static ssize_t proto_down_store(struct device *dev,
489                                 struct device_attribute *attr,
490                                 const char *buf, size_t len)
491 {
492         struct net_device *netdev = to_net_dev(dev);
493
494         /* The check is also done in change_proto_down; this helps returning
495          * early without hitting the trylock/restart in netdev_store.
496          */
497         if (!netdev->netdev_ops->ndo_change_proto_down)
498                 return -EOPNOTSUPP;
499
500         return netdev_store(dev, attr, buf, len, change_proto_down);
501 }
502 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
503
504 static ssize_t phys_port_id_show(struct device *dev,
505                                  struct device_attribute *attr, char *buf)
506 {
507         struct net_device *netdev = to_net_dev(dev);
508         ssize_t ret = -EINVAL;
509
510         /* The check is also done in dev_get_phys_port_id; this helps returning
511          * early without hitting the trylock/restart below.
512          */
513         if (!netdev->netdev_ops->ndo_get_phys_port_id)
514                 return -EOPNOTSUPP;
515
516         if (!rtnl_trylock())
517                 return restart_syscall();
518
519         if (dev_isalive(netdev)) {
520                 struct netdev_phys_item_id ppid;
521
522                 ret = dev_get_phys_port_id(netdev, &ppid);
523                 if (!ret)
524                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
525         }
526         rtnl_unlock();
527
528         return ret;
529 }
530 static DEVICE_ATTR_RO(phys_port_id);
531
532 static ssize_t phys_port_name_show(struct device *dev,
533                                    struct device_attribute *attr, char *buf)
534 {
535         struct net_device *netdev = to_net_dev(dev);
536         ssize_t ret = -EINVAL;
537
538         /* The checks are also done in dev_get_phys_port_name; this helps
539          * returning early without hitting the trylock/restart below.
540          */
541         if (!netdev->netdev_ops->ndo_get_phys_port_name &&
542             !netdev->netdev_ops->ndo_get_devlink_port)
543                 return -EOPNOTSUPP;
544
545         if (!rtnl_trylock())
546                 return restart_syscall();
547
548         if (dev_isalive(netdev)) {
549                 char name[IFNAMSIZ];
550
551                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
552                 if (!ret)
553                         ret = sprintf(buf, "%s\n", name);
554         }
555         rtnl_unlock();
556
557         return ret;
558 }
559 static DEVICE_ATTR_RO(phys_port_name);
560
561 static ssize_t phys_switch_id_show(struct device *dev,
562                                    struct device_attribute *attr, char *buf)
563 {
564         struct net_device *netdev = to_net_dev(dev);
565         ssize_t ret = -EINVAL;
566
567         /* The checks are also done in dev_get_phys_port_name; this helps
568          * returning early without hitting the trylock/restart below. This works
569          * because recurse is false when calling dev_get_port_parent_id.
570          */
571         if (!netdev->netdev_ops->ndo_get_port_parent_id &&
572             !netdev->netdev_ops->ndo_get_devlink_port)
573                 return -EOPNOTSUPP;
574
575         if (!rtnl_trylock())
576                 return restart_syscall();
577
578         if (dev_isalive(netdev)) {
579                 struct netdev_phys_item_id ppid = { };
580
581                 ret = dev_get_port_parent_id(netdev, &ppid, false);
582                 if (!ret)
583                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
584         }
585         rtnl_unlock();
586
587         return ret;
588 }
589 static DEVICE_ATTR_RO(phys_switch_id);
590
591 static ssize_t threaded_show(struct device *dev,
592                              struct device_attribute *attr, char *buf)
593 {
594         struct net_device *netdev = to_net_dev(dev);
595         ssize_t ret = -EINVAL;
596
597         if (!rtnl_trylock())
598                 return restart_syscall();
599
600         if (dev_isalive(netdev))
601                 ret = sprintf(buf, fmt_dec, netdev->threaded);
602
603         rtnl_unlock();
604         return ret;
605 }
606
607 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
608 {
609         int ret;
610
611         if (list_empty(&dev->napi_list))
612                 return -EOPNOTSUPP;
613
614         if (val != 0 && val != 1)
615                 return -EOPNOTSUPP;
616
617         ret = dev_set_threaded(dev, val);
618
619         return ret;
620 }
621
622 static ssize_t threaded_store(struct device *dev,
623                               struct device_attribute *attr,
624                               const char *buf, size_t len)
625 {
626         return netdev_store(dev, attr, buf, len, modify_napi_threaded);
627 }
628 static DEVICE_ATTR_RW(threaded);
629
630 static struct attribute *net_class_attrs[] __ro_after_init = {
631         &dev_attr_netdev_group.attr,
632         &dev_attr_type.attr,
633         &dev_attr_dev_id.attr,
634         &dev_attr_dev_port.attr,
635         &dev_attr_iflink.attr,
636         &dev_attr_ifindex.attr,
637         &dev_attr_name_assign_type.attr,
638         &dev_attr_addr_assign_type.attr,
639         &dev_attr_addr_len.attr,
640         &dev_attr_link_mode.attr,
641         &dev_attr_address.attr,
642         &dev_attr_broadcast.attr,
643         &dev_attr_speed.attr,
644         &dev_attr_duplex.attr,
645         &dev_attr_dormant.attr,
646         &dev_attr_testing.attr,
647         &dev_attr_operstate.attr,
648         &dev_attr_carrier_changes.attr,
649         &dev_attr_ifalias.attr,
650         &dev_attr_carrier.attr,
651         &dev_attr_mtu.attr,
652         &dev_attr_flags.attr,
653         &dev_attr_tx_queue_len.attr,
654         &dev_attr_gro_flush_timeout.attr,
655         &dev_attr_napi_defer_hard_irqs.attr,
656         &dev_attr_phys_port_id.attr,
657         &dev_attr_phys_port_name.attr,
658         &dev_attr_phys_switch_id.attr,
659         &dev_attr_proto_down.attr,
660         &dev_attr_carrier_up_count.attr,
661         &dev_attr_carrier_down_count.attr,
662         &dev_attr_threaded.attr,
663         NULL,
664 };
665 ATTRIBUTE_GROUPS(net_class);
666
667 /* Show a given an attribute in the statistics group */
668 static ssize_t netstat_show(const struct device *d,
669                             struct device_attribute *attr, char *buf,
670                             unsigned long offset)
671 {
672         struct net_device *dev = to_net_dev(d);
673         ssize_t ret = -EINVAL;
674
675         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
676                 offset % sizeof(u64) != 0);
677
678         read_lock(&dev_base_lock);
679         if (dev_isalive(dev)) {
680                 struct rtnl_link_stats64 temp;
681                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
682
683                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
684         }
685         read_unlock(&dev_base_lock);
686         return ret;
687 }
688
689 /* generate a read-only statistics attribute */
690 #define NETSTAT_ENTRY(name)                                             \
691 static ssize_t name##_show(struct device *d,                            \
692                            struct device_attribute *attr, char *buf)    \
693 {                                                                       \
694         return netstat_show(d, attr, buf,                               \
695                             offsetof(struct rtnl_link_stats64, name));  \
696 }                                                                       \
697 static DEVICE_ATTR_RO(name)
698
699 NETSTAT_ENTRY(rx_packets);
700 NETSTAT_ENTRY(tx_packets);
701 NETSTAT_ENTRY(rx_bytes);
702 NETSTAT_ENTRY(tx_bytes);
703 NETSTAT_ENTRY(rx_errors);
704 NETSTAT_ENTRY(tx_errors);
705 NETSTAT_ENTRY(rx_dropped);
706 NETSTAT_ENTRY(tx_dropped);
707 NETSTAT_ENTRY(multicast);
708 NETSTAT_ENTRY(collisions);
709 NETSTAT_ENTRY(rx_length_errors);
710 NETSTAT_ENTRY(rx_over_errors);
711 NETSTAT_ENTRY(rx_crc_errors);
712 NETSTAT_ENTRY(rx_frame_errors);
713 NETSTAT_ENTRY(rx_fifo_errors);
714 NETSTAT_ENTRY(rx_missed_errors);
715 NETSTAT_ENTRY(tx_aborted_errors);
716 NETSTAT_ENTRY(tx_carrier_errors);
717 NETSTAT_ENTRY(tx_fifo_errors);
718 NETSTAT_ENTRY(tx_heartbeat_errors);
719 NETSTAT_ENTRY(tx_window_errors);
720 NETSTAT_ENTRY(rx_compressed);
721 NETSTAT_ENTRY(tx_compressed);
722 NETSTAT_ENTRY(rx_nohandler);
723
724 static struct attribute *netstat_attrs[] __ro_after_init = {
725         &dev_attr_rx_packets.attr,
726         &dev_attr_tx_packets.attr,
727         &dev_attr_rx_bytes.attr,
728         &dev_attr_tx_bytes.attr,
729         &dev_attr_rx_errors.attr,
730         &dev_attr_tx_errors.attr,
731         &dev_attr_rx_dropped.attr,
732         &dev_attr_tx_dropped.attr,
733         &dev_attr_multicast.attr,
734         &dev_attr_collisions.attr,
735         &dev_attr_rx_length_errors.attr,
736         &dev_attr_rx_over_errors.attr,
737         &dev_attr_rx_crc_errors.attr,
738         &dev_attr_rx_frame_errors.attr,
739         &dev_attr_rx_fifo_errors.attr,
740         &dev_attr_rx_missed_errors.attr,
741         &dev_attr_tx_aborted_errors.attr,
742         &dev_attr_tx_carrier_errors.attr,
743         &dev_attr_tx_fifo_errors.attr,
744         &dev_attr_tx_heartbeat_errors.attr,
745         &dev_attr_tx_window_errors.attr,
746         &dev_attr_rx_compressed.attr,
747         &dev_attr_tx_compressed.attr,
748         &dev_attr_rx_nohandler.attr,
749         NULL
750 };
751
752 static const struct attribute_group netstat_group = {
753         .name  = "statistics",
754         .attrs  = netstat_attrs,
755 };
756
757 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
758 static struct attribute *wireless_attrs[] = {
759         NULL
760 };
761
762 static const struct attribute_group wireless_group = {
763         .name = "wireless",
764         .attrs = wireless_attrs,
765 };
766 #endif
767
768 #else /* CONFIG_SYSFS */
769 #define net_class_groups        NULL
770 #endif /* CONFIG_SYSFS */
771
772 #ifdef CONFIG_SYSFS
773 #define to_rx_queue_attr(_attr) \
774         container_of(_attr, struct rx_queue_attribute, attr)
775
776 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
777
778 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
779                                   char *buf)
780 {
781         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
782         struct netdev_rx_queue *queue = to_rx_queue(kobj);
783
784         if (!attribute->show)
785                 return -EIO;
786
787         return attribute->show(queue, buf);
788 }
789
790 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
791                                    const char *buf, size_t count)
792 {
793         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
794         struct netdev_rx_queue *queue = to_rx_queue(kobj);
795
796         if (!attribute->store)
797                 return -EIO;
798
799         return attribute->store(queue, buf, count);
800 }
801
802 static const struct sysfs_ops rx_queue_sysfs_ops = {
803         .show = rx_queue_attr_show,
804         .store = rx_queue_attr_store,
805 };
806
807 #ifdef CONFIG_RPS
808 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
809 {
810         struct rps_map *map;
811         cpumask_var_t mask;
812         int i, len;
813
814         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
815                 return -ENOMEM;
816
817         rcu_read_lock();
818         map = rcu_dereference(queue->rps_map);
819         if (map)
820                 for (i = 0; i < map->len; i++)
821                         cpumask_set_cpu(map->cpus[i], mask);
822
823         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
824         rcu_read_unlock();
825         free_cpumask_var(mask);
826
827         return len < PAGE_SIZE ? len : -EINVAL;
828 }
829
830 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
831                              const char *buf, size_t len)
832 {
833         struct rps_map *old_map, *map;
834         cpumask_var_t mask;
835         int err, cpu, i, hk_flags;
836         static DEFINE_MUTEX(rps_map_mutex);
837
838         if (!capable(CAP_NET_ADMIN))
839                 return -EPERM;
840
841         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
842                 return -ENOMEM;
843
844         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
845         if (err) {
846                 free_cpumask_var(mask);
847                 return err;
848         }
849
850         if (!cpumask_empty(mask)) {
851                 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
852                 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
853                 if (cpumask_empty(mask)) {
854                         free_cpumask_var(mask);
855                         return -EINVAL;
856                 }
857         }
858
859         map = kzalloc(max_t(unsigned int,
860                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
861                       GFP_KERNEL);
862         if (!map) {
863                 free_cpumask_var(mask);
864                 return -ENOMEM;
865         }
866
867         i = 0;
868         for_each_cpu_and(cpu, mask, cpu_online_mask)
869                 map->cpus[i++] = cpu;
870
871         if (i) {
872                 map->len = i;
873         } else {
874                 kfree(map);
875                 map = NULL;
876         }
877
878         mutex_lock(&rps_map_mutex);
879         old_map = rcu_dereference_protected(queue->rps_map,
880                                             mutex_is_locked(&rps_map_mutex));
881         rcu_assign_pointer(queue->rps_map, map);
882
883         if (map)
884                 static_branch_inc(&rps_needed);
885         if (old_map)
886                 static_branch_dec(&rps_needed);
887
888         mutex_unlock(&rps_map_mutex);
889
890         if (old_map)
891                 kfree_rcu(old_map, rcu);
892
893         free_cpumask_var(mask);
894         return len;
895 }
896
897 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
898                                            char *buf)
899 {
900         struct rps_dev_flow_table *flow_table;
901         unsigned long val = 0;
902
903         rcu_read_lock();
904         flow_table = rcu_dereference(queue->rps_flow_table);
905         if (flow_table)
906                 val = (unsigned long)flow_table->mask + 1;
907         rcu_read_unlock();
908
909         return sprintf(buf, "%lu\n", val);
910 }
911
912 static void rps_dev_flow_table_release(struct rcu_head *rcu)
913 {
914         struct rps_dev_flow_table *table = container_of(rcu,
915             struct rps_dev_flow_table, rcu);
916         vfree(table);
917 }
918
919 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
920                                             const char *buf, size_t len)
921 {
922         unsigned long mask, count;
923         struct rps_dev_flow_table *table, *old_table;
924         static DEFINE_SPINLOCK(rps_dev_flow_lock);
925         int rc;
926
927         if (!capable(CAP_NET_ADMIN))
928                 return -EPERM;
929
930         rc = kstrtoul(buf, 0, &count);
931         if (rc < 0)
932                 return rc;
933
934         if (count) {
935                 mask = count - 1;
936                 /* mask = roundup_pow_of_two(count) - 1;
937                  * without overflows...
938                  */
939                 while ((mask | (mask >> 1)) != mask)
940                         mask |= (mask >> 1);
941                 /* On 64 bit arches, must check mask fits in table->mask (u32),
942                  * and on 32bit arches, must check
943                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
944                  */
945 #if BITS_PER_LONG > 32
946                 if (mask > (unsigned long)(u32)mask)
947                         return -EINVAL;
948 #else
949                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
950                                 / sizeof(struct rps_dev_flow)) {
951                         /* Enforce a limit to prevent overflow */
952                         return -EINVAL;
953                 }
954 #endif
955                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
956                 if (!table)
957                         return -ENOMEM;
958
959                 table->mask = mask;
960                 for (count = 0; count <= mask; count++)
961                         table->flows[count].cpu = RPS_NO_CPU;
962         } else {
963                 table = NULL;
964         }
965
966         spin_lock(&rps_dev_flow_lock);
967         old_table = rcu_dereference_protected(queue->rps_flow_table,
968                                               lockdep_is_held(&rps_dev_flow_lock));
969         rcu_assign_pointer(queue->rps_flow_table, table);
970         spin_unlock(&rps_dev_flow_lock);
971
972         if (old_table)
973                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
974
975         return len;
976 }
977
978 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
979         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
980
981 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
982         = __ATTR(rps_flow_cnt, 0644,
983                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
984 #endif /* CONFIG_RPS */
985
986 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
987 #ifdef CONFIG_RPS
988         &rps_cpus_attribute.attr,
989         &rps_dev_flow_table_cnt_attribute.attr,
990 #endif
991         NULL
992 };
993 ATTRIBUTE_GROUPS(rx_queue_default);
994
995 static void rx_queue_release(struct kobject *kobj)
996 {
997         struct netdev_rx_queue *queue = to_rx_queue(kobj);
998 #ifdef CONFIG_RPS
999         struct rps_map *map;
1000         struct rps_dev_flow_table *flow_table;
1001
1002         map = rcu_dereference_protected(queue->rps_map, 1);
1003         if (map) {
1004                 RCU_INIT_POINTER(queue->rps_map, NULL);
1005                 kfree_rcu(map, rcu);
1006         }
1007
1008         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1009         if (flow_table) {
1010                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1011                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1012         }
1013 #endif
1014
1015         memset(kobj, 0, sizeof(*kobj));
1016         dev_put(queue->dev);
1017 }
1018
1019 static const void *rx_queue_namespace(struct kobject *kobj)
1020 {
1021         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1022         struct device *dev = &queue->dev->dev;
1023         const void *ns = NULL;
1024
1025         if (dev->class && dev->class->ns_type)
1026                 ns = dev->class->namespace(dev);
1027
1028         return ns;
1029 }
1030
1031 static void rx_queue_get_ownership(struct kobject *kobj,
1032                                    kuid_t *uid, kgid_t *gid)
1033 {
1034         const struct net *net = rx_queue_namespace(kobj);
1035
1036         net_ns_get_ownership(net, uid, gid);
1037 }
1038
1039 static struct kobj_type rx_queue_ktype __ro_after_init = {
1040         .sysfs_ops = &rx_queue_sysfs_ops,
1041         .release = rx_queue_release,
1042         .default_groups = rx_queue_default_groups,
1043         .namespace = rx_queue_namespace,
1044         .get_ownership = rx_queue_get_ownership,
1045 };
1046
1047 static int rx_queue_add_kobject(struct net_device *dev, int index)
1048 {
1049         struct netdev_rx_queue *queue = dev->_rx + index;
1050         struct kobject *kobj = &queue->kobj;
1051         int error = 0;
1052
1053         /* Kobject_put later will trigger rx_queue_release call which
1054          * decreases dev refcount: Take that reference here
1055          */
1056         dev_hold(queue->dev);
1057
1058         kobj->kset = dev->queues_kset;
1059         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1060                                      "rx-%u", index);
1061         if (error)
1062                 goto err;
1063
1064         if (dev->sysfs_rx_queue_group) {
1065                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1066                 if (error)
1067                         goto err;
1068         }
1069
1070         kobject_uevent(kobj, KOBJ_ADD);
1071
1072         return error;
1073
1074 err:
1075         kobject_put(kobj);
1076         return error;
1077 }
1078
1079 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1080                                  kgid_t kgid)
1081 {
1082         struct netdev_rx_queue *queue = dev->_rx + index;
1083         struct kobject *kobj = &queue->kobj;
1084         int error;
1085
1086         error = sysfs_change_owner(kobj, kuid, kgid);
1087         if (error)
1088                 return error;
1089
1090         if (dev->sysfs_rx_queue_group)
1091                 error = sysfs_group_change_owner(
1092                         kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1093
1094         return error;
1095 }
1096 #endif /* CONFIG_SYSFS */
1097
1098 int
1099 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1100 {
1101 #ifdef CONFIG_SYSFS
1102         int i;
1103         int error = 0;
1104
1105 #ifndef CONFIG_RPS
1106         if (!dev->sysfs_rx_queue_group)
1107                 return 0;
1108 #endif
1109         for (i = old_num; i < new_num; i++) {
1110                 error = rx_queue_add_kobject(dev, i);
1111                 if (error) {
1112                         new_num = old_num;
1113                         break;
1114                 }
1115         }
1116
1117         while (--i >= new_num) {
1118                 struct kobject *kobj = &dev->_rx[i].kobj;
1119
1120                 if (!refcount_read(&dev_net(dev)->ns.count))
1121                         kobj->uevent_suppress = 1;
1122                 if (dev->sysfs_rx_queue_group)
1123                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1124                 kobject_put(kobj);
1125         }
1126
1127         return error;
1128 #else
1129         return 0;
1130 #endif
1131 }
1132
1133 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1134                                      kuid_t kuid, kgid_t kgid)
1135 {
1136 #ifdef CONFIG_SYSFS
1137         int error = 0;
1138         int i;
1139
1140 #ifndef CONFIG_RPS
1141         if (!dev->sysfs_rx_queue_group)
1142                 return 0;
1143 #endif
1144         for (i = 0; i < num; i++) {
1145                 error = rx_queue_change_owner(dev, i, kuid, kgid);
1146                 if (error)
1147                         break;
1148         }
1149
1150         return error;
1151 #else
1152         return 0;
1153 #endif
1154 }
1155
1156 #ifdef CONFIG_SYSFS
1157 /*
1158  * netdev_queue sysfs structures and functions.
1159  */
1160 struct netdev_queue_attribute {
1161         struct attribute attr;
1162         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1163         ssize_t (*store)(struct netdev_queue *queue,
1164                          const char *buf, size_t len);
1165 };
1166 #define to_netdev_queue_attr(_attr) \
1167         container_of(_attr, struct netdev_queue_attribute, attr)
1168
1169 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1170
1171 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1172                                       struct attribute *attr, char *buf)
1173 {
1174         const struct netdev_queue_attribute *attribute
1175                 = to_netdev_queue_attr(attr);
1176         struct netdev_queue *queue = to_netdev_queue(kobj);
1177
1178         if (!attribute->show)
1179                 return -EIO;
1180
1181         return attribute->show(queue, buf);
1182 }
1183
1184 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1185                                        struct attribute *attr,
1186                                        const char *buf, size_t count)
1187 {
1188         const struct netdev_queue_attribute *attribute
1189                 = to_netdev_queue_attr(attr);
1190         struct netdev_queue *queue = to_netdev_queue(kobj);
1191
1192         if (!attribute->store)
1193                 return -EIO;
1194
1195         return attribute->store(queue, buf, count);
1196 }
1197
1198 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1199         .show = netdev_queue_attr_show,
1200         .store = netdev_queue_attr_store,
1201 };
1202
1203 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1204 {
1205         unsigned long trans_timeout;
1206
1207         spin_lock_irq(&queue->_xmit_lock);
1208         trans_timeout = queue->trans_timeout;
1209         spin_unlock_irq(&queue->_xmit_lock);
1210
1211         return sprintf(buf, fmt_ulong, trans_timeout);
1212 }
1213
1214 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1215 {
1216         struct net_device *dev = queue->dev;
1217         unsigned int i;
1218
1219         i = queue - dev->_tx;
1220         BUG_ON(i >= dev->num_tx_queues);
1221
1222         return i;
1223 }
1224
1225 static ssize_t traffic_class_show(struct netdev_queue *queue,
1226                                   char *buf)
1227 {
1228         struct net_device *dev = queue->dev;
1229         int num_tc, tc;
1230         int index;
1231
1232         if (!netif_is_multiqueue(dev))
1233                 return -ENOENT;
1234
1235         if (!rtnl_trylock())
1236                 return restart_syscall();
1237
1238         index = get_netdev_queue_index(queue);
1239
1240         /* If queue belongs to subordinate dev use its TC mapping */
1241         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1242
1243         num_tc = dev->num_tc;
1244         tc = netdev_txq_to_tc(dev, index);
1245
1246         rtnl_unlock();
1247
1248         if (tc < 0)
1249                 return -EINVAL;
1250
1251         /* We can report the traffic class one of two ways:
1252          * Subordinate device traffic classes are reported with the traffic
1253          * class first, and then the subordinate class so for example TC0 on
1254          * subordinate device 2 will be reported as "0-2". If the queue
1255          * belongs to the root device it will be reported with just the
1256          * traffic class, so just "0" for TC 0 for example.
1257          */
1258         return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1259                             sprintf(buf, "%d\n", tc);
1260 }
1261
1262 #ifdef CONFIG_XPS
1263 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1264                                char *buf)
1265 {
1266         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1267 }
1268
1269 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1270                                 const char *buf, size_t len)
1271 {
1272         struct net_device *dev = queue->dev;
1273         int err, index = get_netdev_queue_index(queue);
1274         u32 rate = 0;
1275
1276         if (!capable(CAP_NET_ADMIN))
1277                 return -EPERM;
1278
1279         /* The check is also done later; this helps returning early without
1280          * hitting the trylock/restart below.
1281          */
1282         if (!dev->netdev_ops->ndo_set_tx_maxrate)
1283                 return -EOPNOTSUPP;
1284
1285         err = kstrtou32(buf, 10, &rate);
1286         if (err < 0)
1287                 return err;
1288
1289         if (!rtnl_trylock())
1290                 return restart_syscall();
1291
1292         err = -EOPNOTSUPP;
1293         if (dev->netdev_ops->ndo_set_tx_maxrate)
1294                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1295
1296         rtnl_unlock();
1297         if (!err) {
1298                 queue->tx_maxrate = rate;
1299                 return len;
1300         }
1301         return err;
1302 }
1303
1304 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1305         = __ATTR_RW(tx_maxrate);
1306 #endif
1307
1308 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1309         = __ATTR_RO(tx_timeout);
1310
1311 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1312         = __ATTR_RO(traffic_class);
1313
1314 #ifdef CONFIG_BQL
1315 /*
1316  * Byte queue limits sysfs structures and functions.
1317  */
1318 static ssize_t bql_show(char *buf, unsigned int value)
1319 {
1320         return sprintf(buf, "%u\n", value);
1321 }
1322
1323 static ssize_t bql_set(const char *buf, const size_t count,
1324                        unsigned int *pvalue)
1325 {
1326         unsigned int value;
1327         int err;
1328
1329         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1330                 value = DQL_MAX_LIMIT;
1331         } else {
1332                 err = kstrtouint(buf, 10, &value);
1333                 if (err < 0)
1334                         return err;
1335                 if (value > DQL_MAX_LIMIT)
1336                         return -EINVAL;
1337         }
1338
1339         *pvalue = value;
1340
1341         return count;
1342 }
1343
1344 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1345                                   char *buf)
1346 {
1347         struct dql *dql = &queue->dql;
1348
1349         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1350 }
1351
1352 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1353                                  const char *buf, size_t len)
1354 {
1355         struct dql *dql = &queue->dql;
1356         unsigned int value;
1357         int err;
1358
1359         err = kstrtouint(buf, 10, &value);
1360         if (err < 0)
1361                 return err;
1362
1363         dql->slack_hold_time = msecs_to_jiffies(value);
1364
1365         return len;
1366 }
1367
1368 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1369         = __ATTR(hold_time, 0644,
1370                  bql_show_hold_time, bql_set_hold_time);
1371
1372 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1373                                  char *buf)
1374 {
1375         struct dql *dql = &queue->dql;
1376
1377         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1378 }
1379
1380 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1381         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1382
1383 #define BQL_ATTR(NAME, FIELD)                                           \
1384 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1385                                  char *buf)                             \
1386 {                                                                       \
1387         return bql_show(buf, queue->dql.FIELD);                         \
1388 }                                                                       \
1389                                                                         \
1390 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1391                                 const char *buf, size_t len)            \
1392 {                                                                       \
1393         return bql_set(buf, len, &queue->dql.FIELD);                    \
1394 }                                                                       \
1395                                                                         \
1396 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1397         = __ATTR(NAME, 0644,                            \
1398                  bql_show_ ## NAME, bql_set_ ## NAME)
1399
1400 BQL_ATTR(limit, limit);
1401 BQL_ATTR(limit_max, max_limit);
1402 BQL_ATTR(limit_min, min_limit);
1403
1404 static struct attribute *dql_attrs[] __ro_after_init = {
1405         &bql_limit_attribute.attr,
1406         &bql_limit_max_attribute.attr,
1407         &bql_limit_min_attribute.attr,
1408         &bql_hold_time_attribute.attr,
1409         &bql_inflight_attribute.attr,
1410         NULL
1411 };
1412
1413 static const struct attribute_group dql_group = {
1414         .name  = "byte_queue_limits",
1415         .attrs  = dql_attrs,
1416 };
1417 #endif /* CONFIG_BQL */
1418
1419 #ifdef CONFIG_XPS
1420 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1421                               int tc, char *buf, enum xps_map_type type)
1422 {
1423         struct xps_dev_maps *dev_maps;
1424         unsigned long *mask;
1425         unsigned int nr_ids;
1426         int j, len;
1427
1428         rcu_read_lock();
1429         dev_maps = rcu_dereference(dev->xps_maps[type]);
1430
1431         /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1432          * when dev_maps hasn't been allocated yet, to be backward compatible.
1433          */
1434         nr_ids = dev_maps ? dev_maps->nr_ids :
1435                  (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1436
1437         mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1438         if (!mask) {
1439                 rcu_read_unlock();
1440                 return -ENOMEM;
1441         }
1442
1443         if (!dev_maps || tc >= dev_maps->num_tc)
1444                 goto out_no_maps;
1445
1446         for (j = 0; j < nr_ids; j++) {
1447                 int i, tci = j * dev_maps->num_tc + tc;
1448                 struct xps_map *map;
1449
1450                 map = rcu_dereference(dev_maps->attr_map[tci]);
1451                 if (!map)
1452                         continue;
1453
1454                 for (i = map->len; i--;) {
1455                         if (map->queues[i] == index) {
1456                                 set_bit(j, mask);
1457                                 break;
1458                         }
1459                 }
1460         }
1461 out_no_maps:
1462         rcu_read_unlock();
1463
1464         len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1465         bitmap_free(mask);
1466
1467         return len < PAGE_SIZE ? len : -EINVAL;
1468 }
1469
1470 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1471 {
1472         struct net_device *dev = queue->dev;
1473         unsigned int index;
1474         int len, tc;
1475
1476         if (!netif_is_multiqueue(dev))
1477                 return -ENOENT;
1478
1479         index = get_netdev_queue_index(queue);
1480
1481         if (!rtnl_trylock())
1482                 return restart_syscall();
1483
1484         /* If queue belongs to subordinate dev use its map */
1485         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1486
1487         tc = netdev_txq_to_tc(dev, index);
1488         if (tc < 0) {
1489                 rtnl_unlock();
1490                 return -EINVAL;
1491         }
1492
1493         /* Make sure the subordinate device can't be freed */
1494         get_device(&dev->dev);
1495         rtnl_unlock();
1496
1497         len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1498
1499         put_device(&dev->dev);
1500         return len;
1501 }
1502
1503 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1504                               const char *buf, size_t len)
1505 {
1506         struct net_device *dev = queue->dev;
1507         unsigned int index;
1508         cpumask_var_t mask;
1509         int err;
1510
1511         if (!netif_is_multiqueue(dev))
1512                 return -ENOENT;
1513
1514         if (!capable(CAP_NET_ADMIN))
1515                 return -EPERM;
1516
1517         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1518                 return -ENOMEM;
1519
1520         index = get_netdev_queue_index(queue);
1521
1522         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1523         if (err) {
1524                 free_cpumask_var(mask);
1525                 return err;
1526         }
1527
1528         if (!rtnl_trylock()) {
1529                 free_cpumask_var(mask);
1530                 return restart_syscall();
1531         }
1532
1533         err = netif_set_xps_queue(dev, mask, index);
1534         rtnl_unlock();
1535
1536         free_cpumask_var(mask);
1537
1538         return err ? : len;
1539 }
1540
1541 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1542         = __ATTR_RW(xps_cpus);
1543
1544 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1545 {
1546         struct net_device *dev = queue->dev;
1547         unsigned int index;
1548         int tc;
1549
1550         index = get_netdev_queue_index(queue);
1551
1552         if (!rtnl_trylock())
1553                 return restart_syscall();
1554
1555         tc = netdev_txq_to_tc(dev, index);
1556         rtnl_unlock();
1557         if (tc < 0)
1558                 return -EINVAL;
1559
1560         return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1561 }
1562
1563 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1564                               size_t len)
1565 {
1566         struct net_device *dev = queue->dev;
1567         struct net *net = dev_net(dev);
1568         unsigned long *mask;
1569         unsigned int index;
1570         int err;
1571
1572         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1573                 return -EPERM;
1574
1575         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1576         if (!mask)
1577                 return -ENOMEM;
1578
1579         index = get_netdev_queue_index(queue);
1580
1581         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1582         if (err) {
1583                 bitmap_free(mask);
1584                 return err;
1585         }
1586
1587         if (!rtnl_trylock()) {
1588                 bitmap_free(mask);
1589                 return restart_syscall();
1590         }
1591
1592         cpus_read_lock();
1593         err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1594         cpus_read_unlock();
1595
1596         rtnl_unlock();
1597
1598         bitmap_free(mask);
1599         return err ? : len;
1600 }
1601
1602 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1603         = __ATTR_RW(xps_rxqs);
1604 #endif /* CONFIG_XPS */
1605
1606 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1607         &queue_trans_timeout.attr,
1608         &queue_traffic_class.attr,
1609 #ifdef CONFIG_XPS
1610         &xps_cpus_attribute.attr,
1611         &xps_rxqs_attribute.attr,
1612         &queue_tx_maxrate.attr,
1613 #endif
1614         NULL
1615 };
1616 ATTRIBUTE_GROUPS(netdev_queue_default);
1617
1618 static void netdev_queue_release(struct kobject *kobj)
1619 {
1620         struct netdev_queue *queue = to_netdev_queue(kobj);
1621
1622         memset(kobj, 0, sizeof(*kobj));
1623         dev_put(queue->dev);
1624 }
1625
1626 static const void *netdev_queue_namespace(struct kobject *kobj)
1627 {
1628         struct netdev_queue *queue = to_netdev_queue(kobj);
1629         struct device *dev = &queue->dev->dev;
1630         const void *ns = NULL;
1631
1632         if (dev->class && dev->class->ns_type)
1633                 ns = dev->class->namespace(dev);
1634
1635         return ns;
1636 }
1637
1638 static void netdev_queue_get_ownership(struct kobject *kobj,
1639                                        kuid_t *uid, kgid_t *gid)
1640 {
1641         const struct net *net = netdev_queue_namespace(kobj);
1642
1643         net_ns_get_ownership(net, uid, gid);
1644 }
1645
1646 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1647         .sysfs_ops = &netdev_queue_sysfs_ops,
1648         .release = netdev_queue_release,
1649         .default_groups = netdev_queue_default_groups,
1650         .namespace = netdev_queue_namespace,
1651         .get_ownership = netdev_queue_get_ownership,
1652 };
1653
1654 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1655 {
1656         struct netdev_queue *queue = dev->_tx + index;
1657         struct kobject *kobj = &queue->kobj;
1658         int error = 0;
1659
1660         /* Kobject_put later will trigger netdev_queue_release call
1661          * which decreases dev refcount: Take that reference here
1662          */
1663         dev_hold(queue->dev);
1664
1665         kobj->kset = dev->queues_kset;
1666         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1667                                      "tx-%u", index);
1668         if (error)
1669                 goto err;
1670
1671 #ifdef CONFIG_BQL
1672         error = sysfs_create_group(kobj, &dql_group);
1673         if (error)
1674                 goto err;
1675 #endif
1676
1677         kobject_uevent(kobj, KOBJ_ADD);
1678         return 0;
1679
1680 err:
1681         kobject_put(kobj);
1682         return error;
1683 }
1684
1685 static int tx_queue_change_owner(struct net_device *ndev, int index,
1686                                  kuid_t kuid, kgid_t kgid)
1687 {
1688         struct netdev_queue *queue = ndev->_tx + index;
1689         struct kobject *kobj = &queue->kobj;
1690         int error;
1691
1692         error = sysfs_change_owner(kobj, kuid, kgid);
1693         if (error)
1694                 return error;
1695
1696 #ifdef CONFIG_BQL
1697         error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1698 #endif
1699         return error;
1700 }
1701 #endif /* CONFIG_SYSFS */
1702
1703 int
1704 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1705 {
1706 #ifdef CONFIG_SYSFS
1707         int i;
1708         int error = 0;
1709
1710         for (i = old_num; i < new_num; i++) {
1711                 error = netdev_queue_add_kobject(dev, i);
1712                 if (error) {
1713                         new_num = old_num;
1714                         break;
1715                 }
1716         }
1717
1718         while (--i >= new_num) {
1719                 struct netdev_queue *queue = dev->_tx + i;
1720
1721                 if (!refcount_read(&dev_net(dev)->ns.count))
1722                         queue->kobj.uevent_suppress = 1;
1723 #ifdef CONFIG_BQL
1724                 sysfs_remove_group(&queue->kobj, &dql_group);
1725 #endif
1726                 kobject_put(&queue->kobj);
1727         }
1728
1729         return error;
1730 #else
1731         return 0;
1732 #endif /* CONFIG_SYSFS */
1733 }
1734
1735 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1736                                      kuid_t kuid, kgid_t kgid)
1737 {
1738 #ifdef CONFIG_SYSFS
1739         int error = 0;
1740         int i;
1741
1742         for (i = 0; i < num; i++) {
1743                 error = tx_queue_change_owner(dev, i, kuid, kgid);
1744                 if (error)
1745                         break;
1746         }
1747
1748         return error;
1749 #else
1750         return 0;
1751 #endif /* CONFIG_SYSFS */
1752 }
1753
1754 static int register_queue_kobjects(struct net_device *dev)
1755 {
1756         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1757
1758 #ifdef CONFIG_SYSFS
1759         dev->queues_kset = kset_create_and_add("queues",
1760                                                NULL, &dev->dev.kobj);
1761         if (!dev->queues_kset)
1762                 return -ENOMEM;
1763         real_rx = dev->real_num_rx_queues;
1764 #endif
1765         real_tx = dev->real_num_tx_queues;
1766
1767         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1768         if (error)
1769                 goto error;
1770         rxq = real_rx;
1771
1772         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1773         if (error)
1774                 goto error;
1775         txq = real_tx;
1776
1777         return 0;
1778
1779 error:
1780         netdev_queue_update_kobjects(dev, txq, 0);
1781         net_rx_queue_update_kobjects(dev, rxq, 0);
1782 #ifdef CONFIG_SYSFS
1783         kset_unregister(dev->queues_kset);
1784 #endif
1785         return error;
1786 }
1787
1788 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1789 {
1790         int error = 0, real_rx = 0, real_tx = 0;
1791
1792 #ifdef CONFIG_SYSFS
1793         if (ndev->queues_kset) {
1794                 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1795                 if (error)
1796                         return error;
1797         }
1798         real_rx = ndev->real_num_rx_queues;
1799 #endif
1800         real_tx = ndev->real_num_tx_queues;
1801
1802         error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1803         if (error)
1804                 return error;
1805
1806         error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1807         if (error)
1808                 return error;
1809
1810         return 0;
1811 }
1812
1813 static void remove_queue_kobjects(struct net_device *dev)
1814 {
1815         int real_rx = 0, real_tx = 0;
1816
1817 #ifdef CONFIG_SYSFS
1818         real_rx = dev->real_num_rx_queues;
1819 #endif
1820         real_tx = dev->real_num_tx_queues;
1821
1822         net_rx_queue_update_kobjects(dev, real_rx, 0);
1823         netdev_queue_update_kobjects(dev, real_tx, 0);
1824
1825         dev->real_num_rx_queues = 0;
1826         dev->real_num_tx_queues = 0;
1827 #ifdef CONFIG_SYSFS
1828         kset_unregister(dev->queues_kset);
1829 #endif
1830 }
1831
1832 static bool net_current_may_mount(void)
1833 {
1834         struct net *net = current->nsproxy->net_ns;
1835
1836         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1837 }
1838
1839 static void *net_grab_current_ns(void)
1840 {
1841         struct net *ns = current->nsproxy->net_ns;
1842 #ifdef CONFIG_NET_NS
1843         if (ns)
1844                 refcount_inc(&ns->passive);
1845 #endif
1846         return ns;
1847 }
1848
1849 static const void *net_initial_ns(void)
1850 {
1851         return &init_net;
1852 }
1853
1854 static const void *net_netlink_ns(struct sock *sk)
1855 {
1856         return sock_net(sk);
1857 }
1858
1859 const struct kobj_ns_type_operations net_ns_type_operations = {
1860         .type = KOBJ_NS_TYPE_NET,
1861         .current_may_mount = net_current_may_mount,
1862         .grab_current_ns = net_grab_current_ns,
1863         .netlink_ns = net_netlink_ns,
1864         .initial_ns = net_initial_ns,
1865         .drop_ns = net_drop_ns,
1866 };
1867 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1868
1869 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1870 {
1871         struct net_device *dev = to_net_dev(d);
1872         int retval;
1873
1874         /* pass interface to uevent. */
1875         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1876         if (retval)
1877                 goto exit;
1878
1879         /* pass ifindex to uevent.
1880          * ifindex is useful as it won't change (interface name may change)
1881          * and is what RtNetlink uses natively.
1882          */
1883         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1884
1885 exit:
1886         return retval;
1887 }
1888
1889 /*
1890  *      netdev_release -- destroy and free a dead device.
1891  *      Called when last reference to device kobject is gone.
1892  */
1893 static void netdev_release(struct device *d)
1894 {
1895         struct net_device *dev = to_net_dev(d);
1896
1897         BUG_ON(dev->reg_state != NETREG_RELEASED);
1898
1899         /* no need to wait for rcu grace period:
1900          * device is dead and about to be freed.
1901          */
1902         kfree(rcu_access_pointer(dev->ifalias));
1903         netdev_freemem(dev);
1904 }
1905
1906 static const void *net_namespace(struct device *d)
1907 {
1908         struct net_device *dev = to_net_dev(d);
1909
1910         return dev_net(dev);
1911 }
1912
1913 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1914 {
1915         struct net_device *dev = to_net_dev(d);
1916         const struct net *net = dev_net(dev);
1917
1918         net_ns_get_ownership(net, uid, gid);
1919 }
1920
1921 static struct class net_class __ro_after_init = {
1922         .name = "net",
1923         .dev_release = netdev_release,
1924         .dev_groups = net_class_groups,
1925         .dev_uevent = netdev_uevent,
1926         .ns_type = &net_ns_type_operations,
1927         .namespace = net_namespace,
1928         .get_ownership = net_get_ownership,
1929 };
1930
1931 #ifdef CONFIG_OF
1932 static int of_dev_node_match(struct device *dev, const void *data)
1933 {
1934         for (; dev; dev = dev->parent) {
1935                 if (dev->of_node == data)
1936                         return 1;
1937         }
1938
1939         return 0;
1940 }
1941
1942 /*
1943  * of_find_net_device_by_node - lookup the net device for the device node
1944  * @np: OF device node
1945  *
1946  * Looks up the net_device structure corresponding with the device node.
1947  * If successful, returns a pointer to the net_device with the embedded
1948  * struct device refcount incremented by one, or NULL on failure. The
1949  * refcount must be dropped when done with the net_device.
1950  */
1951 struct net_device *of_find_net_device_by_node(struct device_node *np)
1952 {
1953         struct device *dev;
1954
1955         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1956         if (!dev)
1957                 return NULL;
1958
1959         return to_net_dev(dev);
1960 }
1961 EXPORT_SYMBOL(of_find_net_device_by_node);
1962 #endif
1963
1964 /* Delete sysfs entries but hold kobject reference until after all
1965  * netdev references are gone.
1966  */
1967 void netdev_unregister_kobject(struct net_device *ndev)
1968 {
1969         struct device *dev = &ndev->dev;
1970
1971         if (!refcount_read(&dev_net(ndev)->ns.count))
1972                 dev_set_uevent_suppress(dev, 1);
1973
1974         kobject_get(&dev->kobj);
1975
1976         remove_queue_kobjects(ndev);
1977
1978         pm_runtime_set_memalloc_noio(dev, false);
1979
1980         device_del(dev);
1981 }
1982
1983 /* Create sysfs entries for network device. */
1984 int netdev_register_kobject(struct net_device *ndev)
1985 {
1986         struct device *dev = &ndev->dev;
1987         const struct attribute_group **groups = ndev->sysfs_groups;
1988         int error = 0;
1989
1990         device_initialize(dev);
1991         dev->class = &net_class;
1992         dev->platform_data = ndev;
1993         dev->groups = groups;
1994
1995         dev_set_name(dev, "%s", ndev->name);
1996
1997 #ifdef CONFIG_SYSFS
1998         /* Allow for a device specific group */
1999         if (*groups)
2000                 groups++;
2001
2002         *groups++ = &netstat_group;
2003
2004 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
2005         if (ndev->ieee80211_ptr)
2006                 *groups++ = &wireless_group;
2007 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
2008         else if (ndev->wireless_handlers)
2009                 *groups++ = &wireless_group;
2010 #endif
2011 #endif
2012 #endif /* CONFIG_SYSFS */
2013
2014         error = device_add(dev);
2015         if (error)
2016                 return error;
2017
2018         error = register_queue_kobjects(ndev);
2019         if (error) {
2020                 device_del(dev);
2021                 return error;
2022         }
2023
2024         pm_runtime_set_memalloc_noio(dev, true);
2025
2026         return error;
2027 }
2028
2029 /* Change owner for sysfs entries when moving network devices across network
2030  * namespaces owned by different user namespaces.
2031  */
2032 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2033                         const struct net *net_new)
2034 {
2035         kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2036         kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2037         struct device *dev = &ndev->dev;
2038         int error;
2039
2040         net_ns_get_ownership(net_old, &old_uid, &old_gid);
2041         net_ns_get_ownership(net_new, &new_uid, &new_gid);
2042
2043         /* The network namespace was changed but the owning user namespace is
2044          * identical so there's no need to change the owner of sysfs entries.
2045          */
2046         if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2047                 return 0;
2048
2049         error = device_change_owner(dev, new_uid, new_gid);
2050         if (error)
2051                 return error;
2052
2053         error = queue_change_owner(ndev, new_uid, new_gid);
2054         if (error)
2055                 return error;
2056
2057         return 0;
2058 }
2059
2060 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2061                                 const void *ns)
2062 {
2063         return class_create_file_ns(&net_class, class_attr, ns);
2064 }
2065 EXPORT_SYMBOL(netdev_class_create_file_ns);
2066
2067 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2068                                  const void *ns)
2069 {
2070         class_remove_file_ns(&net_class, class_attr, ns);
2071 }
2072 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2073
2074 int __init netdev_kobject_init(void)
2075 {
2076         kobj_ns_type_register(&net_ns_type_operations);
2077         return class_register(&net_class);
2078 }