net-sysfs: add support for device-specific rx queue sysfs attributes
[firefly-linux-kernel-4.4.55.git] / net / core / net-sysfs.c
1 /*
2  * net-sysfs.c - network device class and attributes
3  *
4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5  *
6  *      This program is free software; you can redistribute it and/or
7  *      modify it under the terms of the GNU General Public License
8  *      as published by the Free Software Foundation; either version
9  *      2 of the License, or (at your option) any later version.
10  */
11
12 #include <linux/capability.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/if_arp.h>
16 #include <linux/slab.h>
17 #include <linux/nsproxy.h>
18 #include <net/sock.h>
19 #include <net/net_namespace.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/vmalloc.h>
22 #include <linux/export.h>
23 #include <linux/jiffies.h>
24 #include <linux/pm_runtime.h>
25
26 #include "net-sysfs.h"
27
28 #ifdef CONFIG_SYSFS
29 static const char fmt_hex[] = "%#x\n";
30 static const char fmt_long_hex[] = "%#lx\n";
31 static const char fmt_dec[] = "%d\n";
32 static const char fmt_udec[] = "%u\n";
33 static const char fmt_ulong[] = "%lu\n";
34 static const char fmt_u64[] = "%llu\n";
35
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 *net = to_net_dev(dev);
47         ssize_t ret = -EINVAL;
48
49         read_lock(&dev_base_lock);
50         if (dev_isalive(net))
51                 ret = (*format)(net, 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 *net, char *buf)  \
60 {                                                                       \
61         return sprintf(buf, format_string, net->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 = -EINVAL;
86
87         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
88                 return -EPERM;
89
90         ret = kstrtoul(buf, 0, &new);
91         if (ret)
92                 goto err;
93
94         if (!rtnl_trylock())
95                 return restart_syscall();
96
97         if (dev_isalive(netdev)) {
98                 if ((ret = (*set)(netdev, new)) == 0)
99                         ret = len;
100         }
101         rtnl_unlock();
102  err:
103         return ret;
104 }
105
106 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
108 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
109 NETDEVICE_SHOW_RO(iflink, 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 /* use same locking rules as GIFHWADDR ioctl's */
115 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
116                             char *buf)
117 {
118         struct net_device *net = to_net_dev(dev);
119         ssize_t ret = -EINVAL;
120
121         read_lock(&dev_base_lock);
122         if (dev_isalive(net))
123                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
124         read_unlock(&dev_base_lock);
125         return ret;
126 }
127 static DEVICE_ATTR_RO(address);
128
129 static ssize_t broadcast_show(struct device *dev,
130                               struct device_attribute *attr, char *buf)
131 {
132         struct net_device *net = to_net_dev(dev);
133         if (dev_isalive(net))
134                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
135         return -EINVAL;
136 }
137 static DEVICE_ATTR_RO(broadcast);
138
139 static int change_carrier(struct net_device *net, unsigned long new_carrier)
140 {
141         if (!netif_running(net))
142                 return -EINVAL;
143         return dev_change_carrier(net, (bool) new_carrier);
144 }
145
146 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
147                              const char *buf, size_t len)
148 {
149         return netdev_store(dev, attr, buf, len, change_carrier);
150 }
151
152 static ssize_t carrier_show(struct device *dev,
153                             struct device_attribute *attr, char *buf)
154 {
155         struct net_device *netdev = to_net_dev(dev);
156         if (netif_running(netdev)) {
157                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
158         }
159         return -EINVAL;
160 }
161 static DEVICE_ATTR_RW(carrier);
162
163 static ssize_t speed_show(struct device *dev,
164                           struct device_attribute *attr, char *buf)
165 {
166         struct net_device *netdev = to_net_dev(dev);
167         int ret = -EINVAL;
168
169         if (!rtnl_trylock())
170                 return restart_syscall();
171
172         if (netif_running(netdev)) {
173                 struct ethtool_cmd cmd;
174                 if (!__ethtool_get_settings(netdev, &cmd))
175                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
176         }
177         rtnl_unlock();
178         return ret;
179 }
180 static DEVICE_ATTR_RO(speed);
181
182 static ssize_t duplex_show(struct device *dev,
183                            struct device_attribute *attr, char *buf)
184 {
185         struct net_device *netdev = to_net_dev(dev);
186         int ret = -EINVAL;
187
188         if (!rtnl_trylock())
189                 return restart_syscall();
190
191         if (netif_running(netdev)) {
192                 struct ethtool_cmd cmd;
193                 if (!__ethtool_get_settings(netdev, &cmd)) {
194                         const char *duplex;
195                         switch (cmd.duplex) {
196                         case DUPLEX_HALF:
197                                 duplex = "half";
198                                 break;
199                         case DUPLEX_FULL:
200                                 duplex = "full";
201                                 break;
202                         default:
203                                 duplex = "unknown";
204                                 break;
205                         }
206                         ret = sprintf(buf, "%s\n", duplex);
207                 }
208         }
209         rtnl_unlock();
210         return ret;
211 }
212 static DEVICE_ATTR_RO(duplex);
213
214 static ssize_t dormant_show(struct device *dev,
215                             struct device_attribute *attr, char *buf)
216 {
217         struct net_device *netdev = to_net_dev(dev);
218
219         if (netif_running(netdev))
220                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
221
222         return -EINVAL;
223 }
224 static DEVICE_ATTR_RO(dormant);
225
226 static const char *const operstates[] = {
227         "unknown",
228         "notpresent", /* currently unused */
229         "down",
230         "lowerlayerdown",
231         "testing", /* currently unused */
232         "dormant",
233         "up"
234 };
235
236 static ssize_t operstate_show(struct device *dev,
237                               struct device_attribute *attr, char *buf)
238 {
239         const struct net_device *netdev = to_net_dev(dev);
240         unsigned char operstate;
241
242         read_lock(&dev_base_lock);
243         operstate = netdev->operstate;
244         if (!netif_running(netdev))
245                 operstate = IF_OPER_DOWN;
246         read_unlock(&dev_base_lock);
247
248         if (operstate >= ARRAY_SIZE(operstates))
249                 return -EINVAL; /* should not happen */
250
251         return sprintf(buf, "%s\n", operstates[operstate]);
252 }
253 static DEVICE_ATTR_RO(operstate);
254
255 /* read-write attributes */
256
257 static int change_mtu(struct net_device *net, unsigned long new_mtu)
258 {
259         return dev_set_mtu(net, (int) new_mtu);
260 }
261
262 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
263                          const char *buf, size_t len)
264 {
265         return netdev_store(dev, attr, buf, len, change_mtu);
266 }
267 NETDEVICE_SHOW_RW(mtu, fmt_dec);
268
269 static int change_flags(struct net_device *net, unsigned long new_flags)
270 {
271         return dev_change_flags(net, (unsigned int) new_flags);
272 }
273
274 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
275                            const char *buf, size_t len)
276 {
277         return netdev_store(dev, attr, buf, len, change_flags);
278 }
279 NETDEVICE_SHOW_RW(flags, fmt_hex);
280
281 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
282 {
283         net->tx_queue_len = new_len;
284         return 0;
285 }
286
287 static ssize_t tx_queue_len_store(struct device *dev,
288                                   struct device_attribute *attr,
289                                   const char *buf, size_t len)
290 {
291         if (!capable(CAP_NET_ADMIN))
292                 return -EPERM;
293
294         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
295 }
296 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
297
298 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
299                              const char *buf, size_t len)
300 {
301         struct net_device *netdev = to_net_dev(dev);
302         struct net *net = dev_net(netdev);
303         size_t count = len;
304         ssize_t ret;
305
306         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
307                 return -EPERM;
308
309         /* ignore trailing newline */
310         if (len >  0 && buf[len - 1] == '\n')
311                 --count;
312
313         if (!rtnl_trylock())
314                 return restart_syscall();
315         ret = dev_set_alias(netdev, buf, count);
316         rtnl_unlock();
317
318         return ret < 0 ? ret : len;
319 }
320
321 static ssize_t ifalias_show(struct device *dev,
322                             struct device_attribute *attr, char *buf)
323 {
324         const struct net_device *netdev = to_net_dev(dev);
325         ssize_t ret = 0;
326
327         if (!rtnl_trylock())
328                 return restart_syscall();
329         if (netdev->ifalias)
330                 ret = sprintf(buf, "%s\n", netdev->ifalias);
331         rtnl_unlock();
332         return ret;
333 }
334 static DEVICE_ATTR_RW(ifalias);
335
336 static int change_group(struct net_device *net, unsigned long new_group)
337 {
338         dev_set_group(net, (int) new_group);
339         return 0;
340 }
341
342 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
343                            const char *buf, size_t len)
344 {
345         return netdev_store(dev, attr, buf, len, change_group);
346 }
347 NETDEVICE_SHOW(group, fmt_dec);
348 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
349
350 static ssize_t phys_port_id_show(struct device *dev,
351                                  struct device_attribute *attr, char *buf)
352 {
353         struct net_device *netdev = to_net_dev(dev);
354         ssize_t ret = -EINVAL;
355
356         if (!rtnl_trylock())
357                 return restart_syscall();
358
359         if (dev_isalive(netdev)) {
360                 struct netdev_phys_port_id ppid;
361
362                 ret = dev_get_phys_port_id(netdev, &ppid);
363                 if (!ret)
364                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
365         }
366         rtnl_unlock();
367
368         return ret;
369 }
370 static DEVICE_ATTR_RO(phys_port_id);
371
372 static struct attribute *net_class_attrs[] = {
373         &dev_attr_netdev_group.attr,
374         &dev_attr_type.attr,
375         &dev_attr_dev_id.attr,
376         &dev_attr_iflink.attr,
377         &dev_attr_ifindex.attr,
378         &dev_attr_addr_assign_type.attr,
379         &dev_attr_addr_len.attr,
380         &dev_attr_link_mode.attr,
381         &dev_attr_address.attr,
382         &dev_attr_broadcast.attr,
383         &dev_attr_speed.attr,
384         &dev_attr_duplex.attr,
385         &dev_attr_dormant.attr,
386         &dev_attr_operstate.attr,
387         &dev_attr_ifalias.attr,
388         &dev_attr_carrier.attr,
389         &dev_attr_mtu.attr,
390         &dev_attr_flags.attr,
391         &dev_attr_tx_queue_len.attr,
392         &dev_attr_phys_port_id.attr,
393         NULL,
394 };
395 ATTRIBUTE_GROUPS(net_class);
396
397 /* Show a given an attribute in the statistics group */
398 static ssize_t netstat_show(const struct device *d,
399                             struct device_attribute *attr, char *buf,
400                             unsigned long offset)
401 {
402         struct net_device *dev = to_net_dev(d);
403         ssize_t ret = -EINVAL;
404
405         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
406                         offset % sizeof(u64) != 0);
407
408         read_lock(&dev_base_lock);
409         if (dev_isalive(dev)) {
410                 struct rtnl_link_stats64 temp;
411                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
412
413                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
414         }
415         read_unlock(&dev_base_lock);
416         return ret;
417 }
418
419 /* generate a read-only statistics attribute */
420 #define NETSTAT_ENTRY(name)                                             \
421 static ssize_t name##_show(struct device *d,                            \
422                            struct device_attribute *attr, char *buf)    \
423 {                                                                       \
424         return netstat_show(d, attr, buf,                               \
425                             offsetof(struct rtnl_link_stats64, name));  \
426 }                                                                       \
427 static DEVICE_ATTR_RO(name)
428
429 NETSTAT_ENTRY(rx_packets);
430 NETSTAT_ENTRY(tx_packets);
431 NETSTAT_ENTRY(rx_bytes);
432 NETSTAT_ENTRY(tx_bytes);
433 NETSTAT_ENTRY(rx_errors);
434 NETSTAT_ENTRY(tx_errors);
435 NETSTAT_ENTRY(rx_dropped);
436 NETSTAT_ENTRY(tx_dropped);
437 NETSTAT_ENTRY(multicast);
438 NETSTAT_ENTRY(collisions);
439 NETSTAT_ENTRY(rx_length_errors);
440 NETSTAT_ENTRY(rx_over_errors);
441 NETSTAT_ENTRY(rx_crc_errors);
442 NETSTAT_ENTRY(rx_frame_errors);
443 NETSTAT_ENTRY(rx_fifo_errors);
444 NETSTAT_ENTRY(rx_missed_errors);
445 NETSTAT_ENTRY(tx_aborted_errors);
446 NETSTAT_ENTRY(tx_carrier_errors);
447 NETSTAT_ENTRY(tx_fifo_errors);
448 NETSTAT_ENTRY(tx_heartbeat_errors);
449 NETSTAT_ENTRY(tx_window_errors);
450 NETSTAT_ENTRY(rx_compressed);
451 NETSTAT_ENTRY(tx_compressed);
452
453 static struct attribute *netstat_attrs[] = {
454         &dev_attr_rx_packets.attr,
455         &dev_attr_tx_packets.attr,
456         &dev_attr_rx_bytes.attr,
457         &dev_attr_tx_bytes.attr,
458         &dev_attr_rx_errors.attr,
459         &dev_attr_tx_errors.attr,
460         &dev_attr_rx_dropped.attr,
461         &dev_attr_tx_dropped.attr,
462         &dev_attr_multicast.attr,
463         &dev_attr_collisions.attr,
464         &dev_attr_rx_length_errors.attr,
465         &dev_attr_rx_over_errors.attr,
466         &dev_attr_rx_crc_errors.attr,
467         &dev_attr_rx_frame_errors.attr,
468         &dev_attr_rx_fifo_errors.attr,
469         &dev_attr_rx_missed_errors.attr,
470         &dev_attr_tx_aborted_errors.attr,
471         &dev_attr_tx_carrier_errors.attr,
472         &dev_attr_tx_fifo_errors.attr,
473         &dev_attr_tx_heartbeat_errors.attr,
474         &dev_attr_tx_window_errors.attr,
475         &dev_attr_rx_compressed.attr,
476         &dev_attr_tx_compressed.attr,
477         NULL
478 };
479
480
481 static struct attribute_group netstat_group = {
482         .name  = "statistics",
483         .attrs  = netstat_attrs,
484 };
485
486 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
487 static struct attribute *wireless_attrs[] = {
488         NULL
489 };
490
491 static struct attribute_group wireless_group = {
492         .name = "wireless",
493         .attrs = wireless_attrs,
494 };
495 #endif
496
497 #else /* CONFIG_SYSFS */
498 #define net_class_groups        NULL
499 #endif /* CONFIG_SYSFS */
500
501 #ifdef CONFIG_SYSFS
502 #define to_rx_queue_attr(_attr) container_of(_attr,             \
503     struct rx_queue_attribute, attr)
504
505 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
506
507 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
508                                   char *buf)
509 {
510         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
511         struct netdev_rx_queue *queue = to_rx_queue(kobj);
512
513         if (!attribute->show)
514                 return -EIO;
515
516         return attribute->show(queue, attribute, buf);
517 }
518
519 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
520                                    const char *buf, size_t count)
521 {
522         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
523         struct netdev_rx_queue *queue = to_rx_queue(kobj);
524
525         if (!attribute->store)
526                 return -EIO;
527
528         return attribute->store(queue, attribute, buf, count);
529 }
530
531 static const struct sysfs_ops rx_queue_sysfs_ops = {
532         .show = rx_queue_attr_show,
533         .store = rx_queue_attr_store,
534 };
535
536 #ifdef CONFIG_RPS
537 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
538                             struct rx_queue_attribute *attribute, char *buf)
539 {
540         struct rps_map *map;
541         cpumask_var_t mask;
542         size_t len = 0;
543         int i;
544
545         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
546                 return -ENOMEM;
547
548         rcu_read_lock();
549         map = rcu_dereference(queue->rps_map);
550         if (map)
551                 for (i = 0; i < map->len; i++)
552                         cpumask_set_cpu(map->cpus[i], mask);
553
554         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
555         if (PAGE_SIZE - len < 3) {
556                 rcu_read_unlock();
557                 free_cpumask_var(mask);
558                 return -EINVAL;
559         }
560         rcu_read_unlock();
561
562         free_cpumask_var(mask);
563         len += sprintf(buf + len, "\n");
564         return len;
565 }
566
567 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
568                       struct rx_queue_attribute *attribute,
569                       const char *buf, size_t len)
570 {
571         struct rps_map *old_map, *map;
572         cpumask_var_t mask;
573         int err, cpu, i;
574         static DEFINE_SPINLOCK(rps_map_lock);
575
576         if (!capable(CAP_NET_ADMIN))
577                 return -EPERM;
578
579         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
580                 return -ENOMEM;
581
582         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
583         if (err) {
584                 free_cpumask_var(mask);
585                 return err;
586         }
587
588         map = kzalloc(max_t(unsigned int,
589             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
590             GFP_KERNEL);
591         if (!map) {
592                 free_cpumask_var(mask);
593                 return -ENOMEM;
594         }
595
596         i = 0;
597         for_each_cpu_and(cpu, mask, cpu_online_mask)
598                 map->cpus[i++] = cpu;
599
600         if (i)
601                 map->len = i;
602         else {
603                 kfree(map);
604                 map = NULL;
605         }
606
607         spin_lock(&rps_map_lock);
608         old_map = rcu_dereference_protected(queue->rps_map,
609                                             lockdep_is_held(&rps_map_lock));
610         rcu_assign_pointer(queue->rps_map, map);
611         spin_unlock(&rps_map_lock);
612
613         if (map)
614                 static_key_slow_inc(&rps_needed);
615         if (old_map) {
616                 kfree_rcu(old_map, rcu);
617                 static_key_slow_dec(&rps_needed);
618         }
619         free_cpumask_var(mask);
620         return len;
621 }
622
623 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
624                                            struct rx_queue_attribute *attr,
625                                            char *buf)
626 {
627         struct rps_dev_flow_table *flow_table;
628         unsigned long val = 0;
629
630         rcu_read_lock();
631         flow_table = rcu_dereference(queue->rps_flow_table);
632         if (flow_table)
633                 val = (unsigned long)flow_table->mask + 1;
634         rcu_read_unlock();
635
636         return sprintf(buf, "%lu\n", val);
637 }
638
639 static void rps_dev_flow_table_release(struct rcu_head *rcu)
640 {
641         struct rps_dev_flow_table *table = container_of(rcu,
642             struct rps_dev_flow_table, rcu);
643         vfree(table);
644 }
645
646 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
647                                      struct rx_queue_attribute *attr,
648                                      const char *buf, size_t len)
649 {
650         unsigned long mask, count;
651         struct rps_dev_flow_table *table, *old_table;
652         static DEFINE_SPINLOCK(rps_dev_flow_lock);
653         int rc;
654
655         if (!capable(CAP_NET_ADMIN))
656                 return -EPERM;
657
658         rc = kstrtoul(buf, 0, &count);
659         if (rc < 0)
660                 return rc;
661
662         if (count) {
663                 mask = count - 1;
664                 /* mask = roundup_pow_of_two(count) - 1;
665                  * without overflows...
666                  */
667                 while ((mask | (mask >> 1)) != mask)
668                         mask |= (mask >> 1);
669                 /* On 64 bit arches, must check mask fits in table->mask (u32),
670                  * and on 32bit arches, must check
671                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
672                  */
673 #if BITS_PER_LONG > 32
674                 if (mask > (unsigned long)(u32)mask)
675                         return -EINVAL;
676 #else
677                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
678                                 / sizeof(struct rps_dev_flow)) {
679                         /* Enforce a limit to prevent overflow */
680                         return -EINVAL;
681                 }
682 #endif
683                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
684                 if (!table)
685                         return -ENOMEM;
686
687                 table->mask = mask;
688                 for (count = 0; count <= mask; count++)
689                         table->flows[count].cpu = RPS_NO_CPU;
690         } else
691                 table = NULL;
692
693         spin_lock(&rps_dev_flow_lock);
694         old_table = rcu_dereference_protected(queue->rps_flow_table,
695                                               lockdep_is_held(&rps_dev_flow_lock));
696         rcu_assign_pointer(queue->rps_flow_table, table);
697         spin_unlock(&rps_dev_flow_lock);
698
699         if (old_table)
700                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
701
702         return len;
703 }
704
705 static struct rx_queue_attribute rps_cpus_attribute =
706         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
707
708
709 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
710         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
711             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
712 #endif /* CONFIG_RPS */
713
714 static struct attribute *rx_queue_default_attrs[] = {
715 #ifdef CONFIG_RPS
716         &rps_cpus_attribute.attr,
717         &rps_dev_flow_table_cnt_attribute.attr,
718 #endif
719         NULL
720 };
721
722 static void rx_queue_release(struct kobject *kobj)
723 {
724         struct netdev_rx_queue *queue = to_rx_queue(kobj);
725 #ifdef CONFIG_RPS
726         struct rps_map *map;
727         struct rps_dev_flow_table *flow_table;
728
729
730         map = rcu_dereference_protected(queue->rps_map, 1);
731         if (map) {
732                 RCU_INIT_POINTER(queue->rps_map, NULL);
733                 kfree_rcu(map, rcu);
734         }
735
736         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
737         if (flow_table) {
738                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
739                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
740         }
741 #endif
742
743         memset(kobj, 0, sizeof(*kobj));
744         dev_put(queue->dev);
745 }
746
747 static struct kobj_type rx_queue_ktype = {
748         .sysfs_ops = &rx_queue_sysfs_ops,
749         .release = rx_queue_release,
750         .default_attrs = rx_queue_default_attrs,
751 };
752
753 static int rx_queue_add_kobject(struct net_device *net, int index)
754 {
755         struct netdev_rx_queue *queue = net->_rx + index;
756         struct kobject *kobj = &queue->kobj;
757         int error = 0;
758
759         kobj->kset = net->queues_kset;
760         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
761             "rx-%u", index);
762         if (error)
763                 goto exit;
764
765         if (net->sysfs_rx_queue_group) {
766                 error = sysfs_create_group(kobj, net->sysfs_rx_queue_group);
767                 if (error)
768                         goto exit;
769         }
770
771         kobject_uevent(kobj, KOBJ_ADD);
772         dev_hold(queue->dev);
773
774         return error;
775 exit:
776         kobject_put(kobj);
777         return error;
778 }
779 #endif /* CONFIG_SYFS */
780
781 int
782 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
783 {
784 #ifdef CONFIG_SYSFS
785         int i;
786         int error = 0;
787
788 #ifndef CONFIG_RPS
789         if (!net->sysfs_rx_queue_group)
790                 return 0;
791 #endif
792         for (i = old_num; i < new_num; i++) {
793                 error = rx_queue_add_kobject(net, i);
794                 if (error) {
795                         new_num = old_num;
796                         break;
797                 }
798         }
799
800         while (--i >= new_num) {
801                 if (net->sysfs_rx_queue_group)
802                         sysfs_remove_group(&net->_rx[i].kobj,
803                                            net->sysfs_rx_queue_group);
804                 kobject_put(&net->_rx[i].kobj);
805         }
806
807         return error;
808 #else
809         return 0;
810 #endif
811 }
812
813 #ifdef CONFIG_SYSFS
814 /*
815  * netdev_queue sysfs structures and functions.
816  */
817 struct netdev_queue_attribute {
818         struct attribute attr;
819         ssize_t (*show)(struct netdev_queue *queue,
820             struct netdev_queue_attribute *attr, char *buf);
821         ssize_t (*store)(struct netdev_queue *queue,
822             struct netdev_queue_attribute *attr, const char *buf, size_t len);
823 };
824 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
825     struct netdev_queue_attribute, attr)
826
827 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
828
829 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
830                                       struct attribute *attr, char *buf)
831 {
832         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
833         struct netdev_queue *queue = to_netdev_queue(kobj);
834
835         if (!attribute->show)
836                 return -EIO;
837
838         return attribute->show(queue, attribute, buf);
839 }
840
841 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
842                                        struct attribute *attr,
843                                        const char *buf, size_t count)
844 {
845         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
846         struct netdev_queue *queue = to_netdev_queue(kobj);
847
848         if (!attribute->store)
849                 return -EIO;
850
851         return attribute->store(queue, attribute, buf, count);
852 }
853
854 static const struct sysfs_ops netdev_queue_sysfs_ops = {
855         .show = netdev_queue_attr_show,
856         .store = netdev_queue_attr_store,
857 };
858
859 static ssize_t show_trans_timeout(struct netdev_queue *queue,
860                                   struct netdev_queue_attribute *attribute,
861                                   char *buf)
862 {
863         unsigned long trans_timeout;
864
865         spin_lock_irq(&queue->_xmit_lock);
866         trans_timeout = queue->trans_timeout;
867         spin_unlock_irq(&queue->_xmit_lock);
868
869         return sprintf(buf, "%lu", trans_timeout);
870 }
871
872 static struct netdev_queue_attribute queue_trans_timeout =
873         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
874
875 #ifdef CONFIG_BQL
876 /*
877  * Byte queue limits sysfs structures and functions.
878  */
879 static ssize_t bql_show(char *buf, unsigned int value)
880 {
881         return sprintf(buf, "%u\n", value);
882 }
883
884 static ssize_t bql_set(const char *buf, const size_t count,
885                        unsigned int *pvalue)
886 {
887         unsigned int value;
888         int err;
889
890         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
891                 value = DQL_MAX_LIMIT;
892         else {
893                 err = kstrtouint(buf, 10, &value);
894                 if (err < 0)
895                         return err;
896                 if (value > DQL_MAX_LIMIT)
897                         return -EINVAL;
898         }
899
900         *pvalue = value;
901
902         return count;
903 }
904
905 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
906                                   struct netdev_queue_attribute *attr,
907                                   char *buf)
908 {
909         struct dql *dql = &queue->dql;
910
911         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
912 }
913
914 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
915                                  struct netdev_queue_attribute *attribute,
916                                  const char *buf, size_t len)
917 {
918         struct dql *dql = &queue->dql;
919         unsigned int value;
920         int err;
921
922         err = kstrtouint(buf, 10, &value);
923         if (err < 0)
924                 return err;
925
926         dql->slack_hold_time = msecs_to_jiffies(value);
927
928         return len;
929 }
930
931 static struct netdev_queue_attribute bql_hold_time_attribute =
932         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
933             bql_set_hold_time);
934
935 static ssize_t bql_show_inflight(struct netdev_queue *queue,
936                                  struct netdev_queue_attribute *attr,
937                                  char *buf)
938 {
939         struct dql *dql = &queue->dql;
940
941         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
942 }
943
944 static struct netdev_queue_attribute bql_inflight_attribute =
945         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
946
947 #define BQL_ATTR(NAME, FIELD)                                           \
948 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
949                                  struct netdev_queue_attribute *attr,   \
950                                  char *buf)                             \
951 {                                                                       \
952         return bql_show(buf, queue->dql.FIELD);                         \
953 }                                                                       \
954                                                                         \
955 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
956                                 struct netdev_queue_attribute *attr,    \
957                                 const char *buf, size_t len)            \
958 {                                                                       \
959         return bql_set(buf, len, &queue->dql.FIELD);                    \
960 }                                                                       \
961                                                                         \
962 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
963         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
964             bql_set_ ## NAME);
965
966 BQL_ATTR(limit, limit)
967 BQL_ATTR(limit_max, max_limit)
968 BQL_ATTR(limit_min, min_limit)
969
970 static struct attribute *dql_attrs[] = {
971         &bql_limit_attribute.attr,
972         &bql_limit_max_attribute.attr,
973         &bql_limit_min_attribute.attr,
974         &bql_hold_time_attribute.attr,
975         &bql_inflight_attribute.attr,
976         NULL
977 };
978
979 static struct attribute_group dql_group = {
980         .name  = "byte_queue_limits",
981         .attrs  = dql_attrs,
982 };
983 #endif /* CONFIG_BQL */
984
985 #ifdef CONFIG_XPS
986 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
987 {
988         struct net_device *dev = queue->dev;
989         int i;
990
991         for (i = 0; i < dev->num_tx_queues; i++)
992                 if (queue == &dev->_tx[i])
993                         break;
994
995         BUG_ON(i >= dev->num_tx_queues);
996
997         return i;
998 }
999
1000
1001 static ssize_t show_xps_map(struct netdev_queue *queue,
1002                             struct netdev_queue_attribute *attribute, char *buf)
1003 {
1004         struct net_device *dev = queue->dev;
1005         struct xps_dev_maps *dev_maps;
1006         cpumask_var_t mask;
1007         unsigned long index;
1008         size_t len = 0;
1009         int i;
1010
1011         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1012                 return -ENOMEM;
1013
1014         index = get_netdev_queue_index(queue);
1015
1016         rcu_read_lock();
1017         dev_maps = rcu_dereference(dev->xps_maps);
1018         if (dev_maps) {
1019                 for_each_possible_cpu(i) {
1020                         struct xps_map *map =
1021                             rcu_dereference(dev_maps->cpu_map[i]);
1022                         if (map) {
1023                                 int j;
1024                                 for (j = 0; j < map->len; j++) {
1025                                         if (map->queues[j] == index) {
1026                                                 cpumask_set_cpu(i, mask);
1027                                                 break;
1028                                         }
1029                                 }
1030                         }
1031                 }
1032         }
1033         rcu_read_unlock();
1034
1035         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1036         if (PAGE_SIZE - len < 3) {
1037                 free_cpumask_var(mask);
1038                 return -EINVAL;
1039         }
1040
1041         free_cpumask_var(mask);
1042         len += sprintf(buf + len, "\n");
1043         return len;
1044 }
1045
1046 static ssize_t store_xps_map(struct netdev_queue *queue,
1047                       struct netdev_queue_attribute *attribute,
1048                       const char *buf, size_t len)
1049 {
1050         struct net_device *dev = queue->dev;
1051         unsigned long index;
1052         cpumask_var_t mask;
1053         int err;
1054
1055         if (!capable(CAP_NET_ADMIN))
1056                 return -EPERM;
1057
1058         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1059                 return -ENOMEM;
1060
1061         index = get_netdev_queue_index(queue);
1062
1063         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1064         if (err) {
1065                 free_cpumask_var(mask);
1066                 return err;
1067         }
1068
1069         err = netif_set_xps_queue(dev, mask, index);
1070
1071         free_cpumask_var(mask);
1072
1073         return err ? : len;
1074 }
1075
1076 static struct netdev_queue_attribute xps_cpus_attribute =
1077     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1078 #endif /* CONFIG_XPS */
1079
1080 static struct attribute *netdev_queue_default_attrs[] = {
1081         &queue_trans_timeout.attr,
1082 #ifdef CONFIG_XPS
1083         &xps_cpus_attribute.attr,
1084 #endif
1085         NULL
1086 };
1087
1088 static void netdev_queue_release(struct kobject *kobj)
1089 {
1090         struct netdev_queue *queue = to_netdev_queue(kobj);
1091
1092         memset(kobj, 0, sizeof(*kobj));
1093         dev_put(queue->dev);
1094 }
1095
1096 static struct kobj_type netdev_queue_ktype = {
1097         .sysfs_ops = &netdev_queue_sysfs_ops,
1098         .release = netdev_queue_release,
1099         .default_attrs = netdev_queue_default_attrs,
1100 };
1101
1102 static int netdev_queue_add_kobject(struct net_device *net, int index)
1103 {
1104         struct netdev_queue *queue = net->_tx + index;
1105         struct kobject *kobj = &queue->kobj;
1106         int error = 0;
1107
1108         kobj->kset = net->queues_kset;
1109         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1110             "tx-%u", index);
1111         if (error)
1112                 goto exit;
1113
1114 #ifdef CONFIG_BQL
1115         error = sysfs_create_group(kobj, &dql_group);
1116         if (error)
1117                 goto exit;
1118 #endif
1119
1120         kobject_uevent(kobj, KOBJ_ADD);
1121         dev_hold(queue->dev);
1122
1123         return 0;
1124 exit:
1125         kobject_put(kobj);
1126         return error;
1127 }
1128 #endif /* CONFIG_SYSFS */
1129
1130 int
1131 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1132 {
1133 #ifdef CONFIG_SYSFS
1134         int i;
1135         int error = 0;
1136
1137         for (i = old_num; i < new_num; i++) {
1138                 error = netdev_queue_add_kobject(net, i);
1139                 if (error) {
1140                         new_num = old_num;
1141                         break;
1142                 }
1143         }
1144
1145         while (--i >= new_num) {
1146                 struct netdev_queue *queue = net->_tx + i;
1147
1148 #ifdef CONFIG_BQL
1149                 sysfs_remove_group(&queue->kobj, &dql_group);
1150 #endif
1151                 kobject_put(&queue->kobj);
1152         }
1153
1154         return error;
1155 #else
1156         return 0;
1157 #endif /* CONFIG_SYSFS */
1158 }
1159
1160 static int register_queue_kobjects(struct net_device *net)
1161 {
1162         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1163
1164 #ifdef CONFIG_SYSFS
1165         net->queues_kset = kset_create_and_add("queues",
1166             NULL, &net->dev.kobj);
1167         if (!net->queues_kset)
1168                 return -ENOMEM;
1169         real_rx = net->real_num_rx_queues;
1170 #endif
1171         real_tx = net->real_num_tx_queues;
1172
1173         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1174         if (error)
1175                 goto error;
1176         rxq = real_rx;
1177
1178         error = netdev_queue_update_kobjects(net, 0, real_tx);
1179         if (error)
1180                 goto error;
1181         txq = real_tx;
1182
1183         return 0;
1184
1185 error:
1186         netdev_queue_update_kobjects(net, txq, 0);
1187         net_rx_queue_update_kobjects(net, rxq, 0);
1188         return error;
1189 }
1190
1191 static void remove_queue_kobjects(struct net_device *net)
1192 {
1193         int real_rx = 0, real_tx = 0;
1194
1195 #ifdef CONFIG_SYSFS
1196         real_rx = net->real_num_rx_queues;
1197 #endif
1198         real_tx = net->real_num_tx_queues;
1199
1200         net_rx_queue_update_kobjects(net, real_rx, 0);
1201         netdev_queue_update_kobjects(net, real_tx, 0);
1202 #ifdef CONFIG_SYSFS
1203         kset_unregister(net->queues_kset);
1204 #endif
1205 }
1206
1207 static bool net_current_may_mount(void)
1208 {
1209         struct net *net = current->nsproxy->net_ns;
1210
1211         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1212 }
1213
1214 static void *net_grab_current_ns(void)
1215 {
1216         struct net *ns = current->nsproxy->net_ns;
1217 #ifdef CONFIG_NET_NS
1218         if (ns)
1219                 atomic_inc(&ns->passive);
1220 #endif
1221         return ns;
1222 }
1223
1224 static const void *net_initial_ns(void)
1225 {
1226         return &init_net;
1227 }
1228
1229 static const void *net_netlink_ns(struct sock *sk)
1230 {
1231         return sock_net(sk);
1232 }
1233
1234 struct kobj_ns_type_operations net_ns_type_operations = {
1235         .type = KOBJ_NS_TYPE_NET,
1236         .current_may_mount = net_current_may_mount,
1237         .grab_current_ns = net_grab_current_ns,
1238         .netlink_ns = net_netlink_ns,
1239         .initial_ns = net_initial_ns,
1240         .drop_ns = net_drop_ns,
1241 };
1242 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1243
1244 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1245 {
1246         struct net_device *dev = to_net_dev(d);
1247         int retval;
1248
1249         /* pass interface to uevent. */
1250         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1251         if (retval)
1252                 goto exit;
1253
1254         /* pass ifindex to uevent.
1255          * ifindex is useful as it won't change (interface name may change)
1256          * and is what RtNetlink uses natively. */
1257         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1258
1259 exit:
1260         return retval;
1261 }
1262
1263 /*
1264  *      netdev_release -- destroy and free a dead device.
1265  *      Called when last reference to device kobject is gone.
1266  */
1267 static void netdev_release(struct device *d)
1268 {
1269         struct net_device *dev = to_net_dev(d);
1270
1271         BUG_ON(dev->reg_state != NETREG_RELEASED);
1272
1273         kfree(dev->ifalias);
1274         netdev_freemem(dev);
1275 }
1276
1277 static const void *net_namespace(struct device *d)
1278 {
1279         struct net_device *dev;
1280         dev = container_of(d, struct net_device, dev);
1281         return dev_net(dev);
1282 }
1283
1284 static struct class net_class = {
1285         .name = "net",
1286         .dev_release = netdev_release,
1287         .dev_groups = net_class_groups,
1288         .dev_uevent = netdev_uevent,
1289         .ns_type = &net_ns_type_operations,
1290         .namespace = net_namespace,
1291 };
1292
1293 /* Delete sysfs entries but hold kobject reference until after all
1294  * netdev references are gone.
1295  */
1296 void netdev_unregister_kobject(struct net_device * net)
1297 {
1298         struct device *dev = &(net->dev);
1299
1300         kobject_get(&dev->kobj);
1301
1302         remove_queue_kobjects(net);
1303
1304         pm_runtime_set_memalloc_noio(dev, false);
1305
1306         device_del(dev);
1307 }
1308
1309 /* Create sysfs entries for network device. */
1310 int netdev_register_kobject(struct net_device *net)
1311 {
1312         struct device *dev = &(net->dev);
1313         const struct attribute_group **groups = net->sysfs_groups;
1314         int error = 0;
1315
1316         device_initialize(dev);
1317         dev->class = &net_class;
1318         dev->platform_data = net;
1319         dev->groups = groups;
1320
1321         dev_set_name(dev, "%s", net->name);
1322
1323 #ifdef CONFIG_SYSFS
1324         /* Allow for a device specific group */
1325         if (*groups)
1326                 groups++;
1327
1328         *groups++ = &netstat_group;
1329
1330 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1331         if (net->ieee80211_ptr)
1332                 *groups++ = &wireless_group;
1333 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1334         else if (net->wireless_handlers)
1335                 *groups++ = &wireless_group;
1336 #endif
1337 #endif
1338 #endif /* CONFIG_SYSFS */
1339
1340         error = device_add(dev);
1341         if (error)
1342                 return error;
1343
1344         error = register_queue_kobjects(net);
1345         if (error) {
1346                 device_del(dev);
1347                 return error;
1348         }
1349
1350         pm_runtime_set_memalloc_noio(dev, true);
1351
1352         return error;
1353 }
1354
1355 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1356                                 const void *ns)
1357 {
1358         return class_create_file_ns(&net_class, class_attr, ns);
1359 }
1360 EXPORT_SYMBOL(netdev_class_create_file_ns);
1361
1362 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1363                                  const void *ns)
1364 {
1365         class_remove_file_ns(&net_class, class_attr, ns);
1366 }
1367 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1368
1369 int __init netdev_kobject_init(void)
1370 {
1371         kobj_ns_type_register(&net_ns_type_operations);
1372         return class_register(&net_class);
1373 }