Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial
[pandora-kernel.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/wireless.h>
22 #include <linux/vmalloc.h>
23 #include <net/wext.h>
24
25 #include "net-sysfs.h"
26
27 #ifdef CONFIG_SYSFS
28 static const char fmt_hex[] = "%#x\n";
29 static const char fmt_long_hex[] = "%#lx\n";
30 static const char fmt_dec[] = "%d\n";
31 static const char fmt_ulong[] = "%lu\n";
32 static const char fmt_u64[] = "%llu\n";
33
34 static inline int dev_isalive(const struct net_device *dev)
35 {
36         return dev->reg_state <= NETREG_REGISTERED;
37 }
38
39 /* use same locking rules as GIF* ioctl's */
40 static ssize_t netdev_show(const struct device *dev,
41                            struct device_attribute *attr, char *buf,
42                            ssize_t (*format)(const struct net_device *, char *))
43 {
44         struct net_device *net = to_net_dev(dev);
45         ssize_t ret = -EINVAL;
46
47         read_lock(&dev_base_lock);
48         if (dev_isalive(net))
49                 ret = (*format)(net, buf);
50         read_unlock(&dev_base_lock);
51
52         return ret;
53 }
54
55 /* generate a show function for simple field */
56 #define NETDEVICE_SHOW(field, format_string)                            \
57 static ssize_t format_##field(const struct net_device *net, char *buf)  \
58 {                                                                       \
59         return sprintf(buf, format_string, net->field);                 \
60 }                                                                       \
61 static ssize_t show_##field(struct device *dev,                         \
62                             struct device_attribute *attr, char *buf)   \
63 {                                                                       \
64         return netdev_show(dev, attr, buf, format_##field);             \
65 }
66
67
68 /* use same locking and permission rules as SIF* ioctl's */
69 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
70                             const char *buf, size_t len,
71                             int (*set)(struct net_device *, unsigned long))
72 {
73         struct net_device *net = to_net_dev(dev);
74         char *endp;
75         unsigned long new;
76         int ret = -EINVAL;
77
78         if (!capable(CAP_NET_ADMIN))
79                 return -EPERM;
80
81         new = simple_strtoul(buf, &endp, 0);
82         if (endp == buf)
83                 goto err;
84
85         if (!rtnl_trylock())
86                 return restart_syscall();
87
88         if (dev_isalive(net)) {
89                 if ((ret = (*set)(net, new)) == 0)
90                         ret = len;
91         }
92         rtnl_unlock();
93  err:
94         return ret;
95 }
96
97 NETDEVICE_SHOW(dev_id, fmt_hex);
98 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
99 NETDEVICE_SHOW(addr_len, fmt_dec);
100 NETDEVICE_SHOW(iflink, fmt_dec);
101 NETDEVICE_SHOW(ifindex, fmt_dec);
102 NETDEVICE_SHOW(features, fmt_long_hex);
103 NETDEVICE_SHOW(type, fmt_dec);
104 NETDEVICE_SHOW(link_mode, fmt_dec);
105
106 /* use same locking rules as GIFHWADDR ioctl's */
107 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
108                             char *buf)
109 {
110         struct net_device *net = to_net_dev(dev);
111         ssize_t ret = -EINVAL;
112
113         read_lock(&dev_base_lock);
114         if (dev_isalive(net))
115                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
116         read_unlock(&dev_base_lock);
117         return ret;
118 }
119
120 static ssize_t show_broadcast(struct device *dev,
121                             struct device_attribute *attr, char *buf)
122 {
123         struct net_device *net = to_net_dev(dev);
124         if (dev_isalive(net))
125                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
126         return -EINVAL;
127 }
128
129 static ssize_t show_carrier(struct device *dev,
130                             struct device_attribute *attr, char *buf)
131 {
132         struct net_device *netdev = to_net_dev(dev);
133         if (netif_running(netdev)) {
134                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
135         }
136         return -EINVAL;
137 }
138
139 static ssize_t show_speed(struct device *dev,
140                           struct device_attribute *attr, char *buf)
141 {
142         struct net_device *netdev = to_net_dev(dev);
143         int ret = -EINVAL;
144
145         if (!rtnl_trylock())
146                 return restart_syscall();
147
148         if (netif_running(netdev) &&
149             netdev->ethtool_ops &&
150             netdev->ethtool_ops->get_settings) {
151                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
152
153                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
154                         ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
155         }
156         rtnl_unlock();
157         return ret;
158 }
159
160 static ssize_t show_duplex(struct device *dev,
161                            struct device_attribute *attr, char *buf)
162 {
163         struct net_device *netdev = to_net_dev(dev);
164         int ret = -EINVAL;
165
166         if (!rtnl_trylock())
167                 return restart_syscall();
168
169         if (netif_running(netdev) &&
170             netdev->ethtool_ops &&
171             netdev->ethtool_ops->get_settings) {
172                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
173
174                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
175                         ret = sprintf(buf, "%s\n", cmd.duplex ? "full" : "half");
176         }
177         rtnl_unlock();
178         return ret;
179 }
180
181 static ssize_t show_dormant(struct device *dev,
182                             struct device_attribute *attr, char *buf)
183 {
184         struct net_device *netdev = to_net_dev(dev);
185
186         if (netif_running(netdev))
187                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
188
189         return -EINVAL;
190 }
191
192 static const char *const operstates[] = {
193         "unknown",
194         "notpresent", /* currently unused */
195         "down",
196         "lowerlayerdown",
197         "testing", /* currently unused */
198         "dormant",
199         "up"
200 };
201
202 static ssize_t show_operstate(struct device *dev,
203                               struct device_attribute *attr, char *buf)
204 {
205         const struct net_device *netdev = to_net_dev(dev);
206         unsigned char operstate;
207
208         read_lock(&dev_base_lock);
209         operstate = netdev->operstate;
210         if (!netif_running(netdev))
211                 operstate = IF_OPER_DOWN;
212         read_unlock(&dev_base_lock);
213
214         if (operstate >= ARRAY_SIZE(operstates))
215                 return -EINVAL; /* should not happen */
216
217         return sprintf(buf, "%s\n", operstates[operstate]);
218 }
219
220 /* read-write attributes */
221 NETDEVICE_SHOW(mtu, fmt_dec);
222
223 static int change_mtu(struct net_device *net, unsigned long new_mtu)
224 {
225         return dev_set_mtu(net, (int) new_mtu);
226 }
227
228 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
229                          const char *buf, size_t len)
230 {
231         return netdev_store(dev, attr, buf, len, change_mtu);
232 }
233
234 NETDEVICE_SHOW(flags, fmt_hex);
235
236 static int change_flags(struct net_device *net, unsigned long new_flags)
237 {
238         return dev_change_flags(net, (unsigned) new_flags);
239 }
240
241 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
242                            const char *buf, size_t len)
243 {
244         return netdev_store(dev, attr, buf, len, change_flags);
245 }
246
247 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
248
249 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
250 {
251         net->tx_queue_len = new_len;
252         return 0;
253 }
254
255 static ssize_t store_tx_queue_len(struct device *dev,
256                                   struct device_attribute *attr,
257                                   const char *buf, size_t len)
258 {
259         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
260 }
261
262 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
263                              const char *buf, size_t len)
264 {
265         struct net_device *netdev = to_net_dev(dev);
266         size_t count = len;
267         ssize_t ret;
268
269         if (!capable(CAP_NET_ADMIN))
270                 return -EPERM;
271
272         /* ignore trailing newline */
273         if (len >  0 && buf[len - 1] == '\n')
274                 --count;
275
276         if (!rtnl_trylock())
277                 return restart_syscall();
278         ret = dev_set_alias(netdev, buf, count);
279         rtnl_unlock();
280
281         return ret < 0 ? ret : len;
282 }
283
284 static ssize_t show_ifalias(struct device *dev,
285                             struct device_attribute *attr, char *buf)
286 {
287         const struct net_device *netdev = to_net_dev(dev);
288         ssize_t ret = 0;
289
290         if (!rtnl_trylock())
291                 return restart_syscall();
292         if (netdev->ifalias)
293                 ret = sprintf(buf, "%s\n", netdev->ifalias);
294         rtnl_unlock();
295         return ret;
296 }
297
298 static struct device_attribute net_class_attributes[] = {
299         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
300         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
301         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
302         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
303         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
304         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
305         __ATTR(features, S_IRUGO, show_features, NULL),
306         __ATTR(type, S_IRUGO, show_type, NULL),
307         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
308         __ATTR(address, S_IRUGO, show_address, NULL),
309         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
310         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
311         __ATTR(speed, S_IRUGO, show_speed, NULL),
312         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
313         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
314         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
315         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
316         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
317         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
318                store_tx_queue_len),
319         {}
320 };
321
322 /* Show a given an attribute in the statistics group */
323 static ssize_t netstat_show(const struct device *d,
324                             struct device_attribute *attr, char *buf,
325                             unsigned long offset)
326 {
327         struct net_device *dev = to_net_dev(d);
328         ssize_t ret = -EINVAL;
329
330         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
331                         offset % sizeof(u64) != 0);
332
333         read_lock(&dev_base_lock);
334         if (dev_isalive(dev)) {
335                 struct rtnl_link_stats64 temp;
336                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
337
338                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
339         }
340         read_unlock(&dev_base_lock);
341         return ret;
342 }
343
344 /* generate a read-only statistics attribute */
345 #define NETSTAT_ENTRY(name)                                             \
346 static ssize_t show_##name(struct device *d,                            \
347                            struct device_attribute *attr, char *buf)    \
348 {                                                                       \
349         return netstat_show(d, attr, buf,                               \
350                             offsetof(struct rtnl_link_stats64, name));  \
351 }                                                                       \
352 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
353
354 NETSTAT_ENTRY(rx_packets);
355 NETSTAT_ENTRY(tx_packets);
356 NETSTAT_ENTRY(rx_bytes);
357 NETSTAT_ENTRY(tx_bytes);
358 NETSTAT_ENTRY(rx_errors);
359 NETSTAT_ENTRY(tx_errors);
360 NETSTAT_ENTRY(rx_dropped);
361 NETSTAT_ENTRY(tx_dropped);
362 NETSTAT_ENTRY(multicast);
363 NETSTAT_ENTRY(collisions);
364 NETSTAT_ENTRY(rx_length_errors);
365 NETSTAT_ENTRY(rx_over_errors);
366 NETSTAT_ENTRY(rx_crc_errors);
367 NETSTAT_ENTRY(rx_frame_errors);
368 NETSTAT_ENTRY(rx_fifo_errors);
369 NETSTAT_ENTRY(rx_missed_errors);
370 NETSTAT_ENTRY(tx_aborted_errors);
371 NETSTAT_ENTRY(tx_carrier_errors);
372 NETSTAT_ENTRY(tx_fifo_errors);
373 NETSTAT_ENTRY(tx_heartbeat_errors);
374 NETSTAT_ENTRY(tx_window_errors);
375 NETSTAT_ENTRY(rx_compressed);
376 NETSTAT_ENTRY(tx_compressed);
377
378 static struct attribute *netstat_attrs[] = {
379         &dev_attr_rx_packets.attr,
380         &dev_attr_tx_packets.attr,
381         &dev_attr_rx_bytes.attr,
382         &dev_attr_tx_bytes.attr,
383         &dev_attr_rx_errors.attr,
384         &dev_attr_tx_errors.attr,
385         &dev_attr_rx_dropped.attr,
386         &dev_attr_tx_dropped.attr,
387         &dev_attr_multicast.attr,
388         &dev_attr_collisions.attr,
389         &dev_attr_rx_length_errors.attr,
390         &dev_attr_rx_over_errors.attr,
391         &dev_attr_rx_crc_errors.attr,
392         &dev_attr_rx_frame_errors.attr,
393         &dev_attr_rx_fifo_errors.attr,
394         &dev_attr_rx_missed_errors.attr,
395         &dev_attr_tx_aborted_errors.attr,
396         &dev_attr_tx_carrier_errors.attr,
397         &dev_attr_tx_fifo_errors.attr,
398         &dev_attr_tx_heartbeat_errors.attr,
399         &dev_attr_tx_window_errors.attr,
400         &dev_attr_rx_compressed.attr,
401         &dev_attr_tx_compressed.attr,
402         NULL
403 };
404
405
406 static struct attribute_group netstat_group = {
407         .name  = "statistics",
408         .attrs  = netstat_attrs,
409 };
410
411 #ifdef CONFIG_WIRELESS_EXT_SYSFS
412 /* helper function that does all the locking etc for wireless stats */
413 static ssize_t wireless_show(struct device *d, char *buf,
414                              ssize_t (*format)(const struct iw_statistics *,
415                                                char *))
416 {
417         struct net_device *dev = to_net_dev(d);
418         const struct iw_statistics *iw;
419         ssize_t ret = -EINVAL;
420
421         if (!rtnl_trylock())
422                 return restart_syscall();
423         if (dev_isalive(dev)) {
424                 iw = get_wireless_stats(dev);
425                 if (iw)
426                         ret = (*format)(iw, buf);
427         }
428         rtnl_unlock();
429
430         return ret;
431 }
432
433 /* show function template for wireless fields */
434 #define WIRELESS_SHOW(name, field, format_string)                       \
435 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
436 {                                                                       \
437         return sprintf(buf, format_string, iw->field);                  \
438 }                                                                       \
439 static ssize_t show_iw_##name(struct device *d,                         \
440                               struct device_attribute *attr, char *buf) \
441 {                                                                       \
442         return wireless_show(d, buf, format_iw_##name);                 \
443 }                                                                       \
444 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
445
446 WIRELESS_SHOW(status, status, fmt_hex);
447 WIRELESS_SHOW(link, qual.qual, fmt_dec);
448 WIRELESS_SHOW(level, qual.level, fmt_dec);
449 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
450 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
451 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
452 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
453 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
454 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
455 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
456
457 static struct attribute *wireless_attrs[] = {
458         &dev_attr_status.attr,
459         &dev_attr_link.attr,
460         &dev_attr_level.attr,
461         &dev_attr_noise.attr,
462         &dev_attr_nwid.attr,
463         &dev_attr_crypt.attr,
464         &dev_attr_fragment.attr,
465         &dev_attr_retries.attr,
466         &dev_attr_misc.attr,
467         &dev_attr_beacon.attr,
468         NULL
469 };
470
471 static struct attribute_group wireless_group = {
472         .name = "wireless",
473         .attrs = wireless_attrs,
474 };
475 #endif
476 #endif /* CONFIG_SYSFS */
477
478 #ifdef CONFIG_RPS
479 /*
480  * RX queue sysfs structures and functions.
481  */
482 struct rx_queue_attribute {
483         struct attribute attr;
484         ssize_t (*show)(struct netdev_rx_queue *queue,
485             struct rx_queue_attribute *attr, char *buf);
486         ssize_t (*store)(struct netdev_rx_queue *queue,
487             struct rx_queue_attribute *attr, const char *buf, size_t len);
488 };
489 #define to_rx_queue_attr(_attr) container_of(_attr,             \
490     struct rx_queue_attribute, attr)
491
492 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
493
494 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
495                                   char *buf)
496 {
497         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
498         struct netdev_rx_queue *queue = to_rx_queue(kobj);
499
500         if (!attribute->show)
501                 return -EIO;
502
503         return attribute->show(queue, attribute, buf);
504 }
505
506 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
507                                    const char *buf, size_t count)
508 {
509         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
510         struct netdev_rx_queue *queue = to_rx_queue(kobj);
511
512         if (!attribute->store)
513                 return -EIO;
514
515         return attribute->store(queue, attribute, buf, count);
516 }
517
518 static const struct sysfs_ops rx_queue_sysfs_ops = {
519         .show = rx_queue_attr_show,
520         .store = rx_queue_attr_store,
521 };
522
523 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
524                             struct rx_queue_attribute *attribute, char *buf)
525 {
526         struct rps_map *map;
527         cpumask_var_t mask;
528         size_t len = 0;
529         int i;
530
531         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
532                 return -ENOMEM;
533
534         rcu_read_lock();
535         map = rcu_dereference(queue->rps_map);
536         if (map)
537                 for (i = 0; i < map->len; i++)
538                         cpumask_set_cpu(map->cpus[i], mask);
539
540         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
541         if (PAGE_SIZE - len < 3) {
542                 rcu_read_unlock();
543                 free_cpumask_var(mask);
544                 return -EINVAL;
545         }
546         rcu_read_unlock();
547
548         free_cpumask_var(mask);
549         len += sprintf(buf + len, "\n");
550         return len;
551 }
552
553 static void rps_map_release(struct rcu_head *rcu)
554 {
555         struct rps_map *map = container_of(rcu, struct rps_map, rcu);
556
557         kfree(map);
558 }
559
560 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
561                       struct rx_queue_attribute *attribute,
562                       const char *buf, size_t len)
563 {
564         struct rps_map *old_map, *map;
565         cpumask_var_t mask;
566         int err, cpu, i;
567         static DEFINE_SPINLOCK(rps_map_lock);
568
569         if (!capable(CAP_NET_ADMIN))
570                 return -EPERM;
571
572         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
573                 return -ENOMEM;
574
575         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
576         if (err) {
577                 free_cpumask_var(mask);
578                 return err;
579         }
580
581         map = kzalloc(max_t(unsigned,
582             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
583             GFP_KERNEL);
584         if (!map) {
585                 free_cpumask_var(mask);
586                 return -ENOMEM;
587         }
588
589         i = 0;
590         for_each_cpu_and(cpu, mask, cpu_online_mask)
591                 map->cpus[i++] = cpu;
592
593         if (i)
594                 map->len = i;
595         else {
596                 kfree(map);
597                 map = NULL;
598         }
599
600         spin_lock(&rps_map_lock);
601         old_map = rcu_dereference_protected(queue->rps_map,
602                                             lockdep_is_held(&rps_map_lock));
603         rcu_assign_pointer(queue->rps_map, map);
604         spin_unlock(&rps_map_lock);
605
606         if (old_map)
607                 call_rcu(&old_map->rcu, rps_map_release);
608
609         free_cpumask_var(mask);
610         return len;
611 }
612
613 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
614                                            struct rx_queue_attribute *attr,
615                                            char *buf)
616 {
617         struct rps_dev_flow_table *flow_table;
618         unsigned int val = 0;
619
620         rcu_read_lock();
621         flow_table = rcu_dereference(queue->rps_flow_table);
622         if (flow_table)
623                 val = flow_table->mask + 1;
624         rcu_read_unlock();
625
626         return sprintf(buf, "%u\n", val);
627 }
628
629 static void rps_dev_flow_table_release_work(struct work_struct *work)
630 {
631         struct rps_dev_flow_table *table = container_of(work,
632             struct rps_dev_flow_table, free_work);
633
634         vfree(table);
635 }
636
637 static void rps_dev_flow_table_release(struct rcu_head *rcu)
638 {
639         struct rps_dev_flow_table *table = container_of(rcu,
640             struct rps_dev_flow_table, rcu);
641
642         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
643         schedule_work(&table->free_work);
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 int count;
651         char *endp;
652         struct rps_dev_flow_table *table, *old_table;
653         static DEFINE_SPINLOCK(rps_dev_flow_lock);
654
655         if (!capable(CAP_NET_ADMIN))
656                 return -EPERM;
657
658         count = simple_strtoul(buf, &endp, 0);
659         if (endp == buf)
660                 return -EINVAL;
661
662         if (count) {
663                 int i;
664
665                 if (count > 1<<30) {
666                         /* Enforce a limit to prevent overflow */
667                         return -EINVAL;
668                 }
669                 count = roundup_pow_of_two(count);
670                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
671                 if (!table)
672                         return -ENOMEM;
673
674                 table->mask = count - 1;
675                 for (i = 0; i < count; i++)
676                         table->flows[i].cpu = RPS_NO_CPU;
677         } else
678                 table = NULL;
679
680         spin_lock(&rps_dev_flow_lock);
681         old_table = rcu_dereference_protected(queue->rps_flow_table,
682                                               lockdep_is_held(&rps_dev_flow_lock));
683         rcu_assign_pointer(queue->rps_flow_table, table);
684         spin_unlock(&rps_dev_flow_lock);
685
686         if (old_table)
687                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
688
689         return len;
690 }
691
692 static struct rx_queue_attribute rps_cpus_attribute =
693         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
694
695
696 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
697         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
698             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
699
700 static struct attribute *rx_queue_default_attrs[] = {
701         &rps_cpus_attribute.attr,
702         &rps_dev_flow_table_cnt_attribute.attr,
703         NULL
704 };
705
706 static void rx_queue_release(struct kobject *kobj)
707 {
708         struct netdev_rx_queue *queue = to_rx_queue(kobj);
709         struct rps_map *map;
710         struct rps_dev_flow_table *flow_table;
711
712
713         map = rcu_dereference_raw(queue->rps_map);
714         if (map) {
715                 RCU_INIT_POINTER(queue->rps_map, NULL);
716                 call_rcu(&map->rcu, rps_map_release);
717         }
718
719         flow_table = rcu_dereference_raw(queue->rps_flow_table);
720         if (flow_table) {
721                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
722                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
723         }
724
725         memset(kobj, 0, sizeof(*kobj));
726         dev_put(queue->dev);
727 }
728
729 static struct kobj_type rx_queue_ktype = {
730         .sysfs_ops = &rx_queue_sysfs_ops,
731         .release = rx_queue_release,
732         .default_attrs = rx_queue_default_attrs,
733 };
734
735 static int rx_queue_add_kobject(struct net_device *net, int index)
736 {
737         struct netdev_rx_queue *queue = net->_rx + index;
738         struct kobject *kobj = &queue->kobj;
739         int error = 0;
740
741         kobj->kset = net->queues_kset;
742         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
743             "rx-%u", index);
744         if (error) {
745                 kobject_put(kobj);
746                 return error;
747         }
748
749         kobject_uevent(kobj, KOBJ_ADD);
750         dev_hold(queue->dev);
751
752         return error;
753 }
754 #endif /* CONFIG_RPS */
755
756 int
757 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
758 {
759 #ifdef CONFIG_RPS
760         int i;
761         int error = 0;
762
763         for (i = old_num; i < new_num; i++) {
764                 error = rx_queue_add_kobject(net, i);
765                 if (error) {
766                         new_num = old_num;
767                         break;
768                 }
769         }
770
771         while (--i >= new_num)
772                 kobject_put(&net->_rx[i].kobj);
773
774         return error;
775 #else
776         return 0;
777 #endif
778 }
779
780 #ifdef CONFIG_XPS
781 /*
782  * netdev_queue sysfs structures and functions.
783  */
784 struct netdev_queue_attribute {
785         struct attribute attr;
786         ssize_t (*show)(struct netdev_queue *queue,
787             struct netdev_queue_attribute *attr, char *buf);
788         ssize_t (*store)(struct netdev_queue *queue,
789             struct netdev_queue_attribute *attr, const char *buf, size_t len);
790 };
791 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
792     struct netdev_queue_attribute, attr)
793
794 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
795
796 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
797                                       struct attribute *attr, char *buf)
798 {
799         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
800         struct netdev_queue *queue = to_netdev_queue(kobj);
801
802         if (!attribute->show)
803                 return -EIO;
804
805         return attribute->show(queue, attribute, buf);
806 }
807
808 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
809                                        struct attribute *attr,
810                                        const char *buf, size_t count)
811 {
812         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
813         struct netdev_queue *queue = to_netdev_queue(kobj);
814
815         if (!attribute->store)
816                 return -EIO;
817
818         return attribute->store(queue, attribute, buf, count);
819 }
820
821 static const struct sysfs_ops netdev_queue_sysfs_ops = {
822         .show = netdev_queue_attr_show,
823         .store = netdev_queue_attr_store,
824 };
825
826 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
827 {
828         struct net_device *dev = queue->dev;
829         int i;
830
831         for (i = 0; i < dev->num_tx_queues; i++)
832                 if (queue == &dev->_tx[i])
833                         break;
834
835         BUG_ON(i >= dev->num_tx_queues);
836
837         return i;
838 }
839
840
841 static ssize_t show_xps_map(struct netdev_queue *queue,
842                             struct netdev_queue_attribute *attribute, char *buf)
843 {
844         struct net_device *dev = queue->dev;
845         struct xps_dev_maps *dev_maps;
846         cpumask_var_t mask;
847         unsigned long index;
848         size_t len = 0;
849         int i;
850
851         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
852                 return -ENOMEM;
853
854         index = get_netdev_queue_index(queue);
855
856         rcu_read_lock();
857         dev_maps = rcu_dereference(dev->xps_maps);
858         if (dev_maps) {
859                 for_each_possible_cpu(i) {
860                         struct xps_map *map =
861                             rcu_dereference(dev_maps->cpu_map[i]);
862                         if (map) {
863                                 int j;
864                                 for (j = 0; j < map->len; j++) {
865                                         if (map->queues[j] == index) {
866                                                 cpumask_set_cpu(i, mask);
867                                                 break;
868                                         }
869                                 }
870                         }
871                 }
872         }
873         rcu_read_unlock();
874
875         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
876         if (PAGE_SIZE - len < 3) {
877                 free_cpumask_var(mask);
878                 return -EINVAL;
879         }
880
881         free_cpumask_var(mask);
882         len += sprintf(buf + len, "\n");
883         return len;
884 }
885
886 static void xps_map_release(struct rcu_head *rcu)
887 {
888         struct xps_map *map = container_of(rcu, struct xps_map, rcu);
889
890         kfree(map);
891 }
892
893 static void xps_dev_maps_release(struct rcu_head *rcu)
894 {
895         struct xps_dev_maps *dev_maps =
896             container_of(rcu, struct xps_dev_maps, rcu);
897
898         kfree(dev_maps);
899 }
900
901 static DEFINE_MUTEX(xps_map_mutex);
902 #define xmap_dereference(P)             \
903         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
904
905 static ssize_t store_xps_map(struct netdev_queue *queue,
906                       struct netdev_queue_attribute *attribute,
907                       const char *buf, size_t len)
908 {
909         struct net_device *dev = queue->dev;
910         cpumask_var_t mask;
911         int err, i, cpu, pos, map_len, alloc_len, need_set;
912         unsigned long index;
913         struct xps_map *map, *new_map;
914         struct xps_dev_maps *dev_maps, *new_dev_maps;
915         int nonempty = 0;
916         int numa_node = -2;
917
918         if (!capable(CAP_NET_ADMIN))
919                 return -EPERM;
920
921         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
922                 return -ENOMEM;
923
924         index = get_netdev_queue_index(queue);
925
926         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
927         if (err) {
928                 free_cpumask_var(mask);
929                 return err;
930         }
931
932         new_dev_maps = kzalloc(max_t(unsigned,
933             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
934         if (!new_dev_maps) {
935                 free_cpumask_var(mask);
936                 return -ENOMEM;
937         }
938
939         mutex_lock(&xps_map_mutex);
940
941         dev_maps = xmap_dereference(dev->xps_maps);
942
943         for_each_possible_cpu(cpu) {
944                 map = dev_maps ?
945                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
946                 new_map = map;
947                 if (map) {
948                         for (pos = 0; pos < map->len; pos++)
949                                 if (map->queues[pos] == index)
950                                         break;
951                         map_len = map->len;
952                         alloc_len = map->alloc_len;
953                 } else
954                         pos = map_len = alloc_len = 0;
955
956                 need_set = cpu_isset(cpu, *mask) && cpu_online(cpu);
957 #ifdef CONFIG_NUMA
958                 if (need_set) {
959                         if (numa_node == -2)
960                                 numa_node = cpu_to_node(cpu);
961                         else if (numa_node != cpu_to_node(cpu))
962                                 numa_node = -1;
963                 }
964 #endif
965                 if (need_set && pos >= map_len) {
966                         /* Need to add queue to this CPU's map */
967                         if (map_len >= alloc_len) {
968                                 alloc_len = alloc_len ?
969                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
970                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
971                                                        GFP_KERNEL,
972                                                        cpu_to_node(cpu));
973                                 if (!new_map)
974                                         goto error;
975                                 new_map->alloc_len = alloc_len;
976                                 for (i = 0; i < map_len; i++)
977                                         new_map->queues[i] = map->queues[i];
978                                 new_map->len = map_len;
979                         }
980                         new_map->queues[new_map->len++] = index;
981                 } else if (!need_set && pos < map_len) {
982                         /* Need to remove queue from this CPU's map */
983                         if (map_len > 1)
984                                 new_map->queues[pos] =
985                                     new_map->queues[--new_map->len];
986                         else
987                                 new_map = NULL;
988                 }
989                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
990         }
991
992         /* Cleanup old maps */
993         for_each_possible_cpu(cpu) {
994                 map = dev_maps ?
995                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
996                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
997                         call_rcu(&map->rcu, xps_map_release);
998                 if (new_dev_maps->cpu_map[cpu])
999                         nonempty = 1;
1000         }
1001
1002         if (nonempty)
1003                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1004         else {
1005                 kfree(new_dev_maps);
1006                 rcu_assign_pointer(dev->xps_maps, NULL);
1007         }
1008
1009         if (dev_maps)
1010                 call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1011
1012         netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1013                                             NUMA_NO_NODE);
1014
1015         mutex_unlock(&xps_map_mutex);
1016
1017         free_cpumask_var(mask);
1018         return len;
1019
1020 error:
1021         mutex_unlock(&xps_map_mutex);
1022
1023         if (new_dev_maps)
1024                 for_each_possible_cpu(i)
1025                         kfree(rcu_dereference_protected(
1026                                 new_dev_maps->cpu_map[i],
1027                                 1));
1028         kfree(new_dev_maps);
1029         free_cpumask_var(mask);
1030         return -ENOMEM;
1031 }
1032
1033 static struct netdev_queue_attribute xps_cpus_attribute =
1034     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1035
1036 static struct attribute *netdev_queue_default_attrs[] = {
1037         &xps_cpus_attribute.attr,
1038         NULL
1039 };
1040
1041 static void netdev_queue_release(struct kobject *kobj)
1042 {
1043         struct netdev_queue *queue = to_netdev_queue(kobj);
1044         struct net_device *dev = queue->dev;
1045         struct xps_dev_maps *dev_maps;
1046         struct xps_map *map;
1047         unsigned long index;
1048         int i, pos, nonempty = 0;
1049
1050         index = get_netdev_queue_index(queue);
1051
1052         mutex_lock(&xps_map_mutex);
1053         dev_maps = xmap_dereference(dev->xps_maps);
1054
1055         if (dev_maps) {
1056                 for_each_possible_cpu(i) {
1057                         map = xmap_dereference(dev_maps->cpu_map[i]);
1058                         if (!map)
1059                                 continue;
1060
1061                         for (pos = 0; pos < map->len; pos++)
1062                                 if (map->queues[pos] == index)
1063                                         break;
1064
1065                         if (pos < map->len) {
1066                                 if (map->len > 1)
1067                                         map->queues[pos] =
1068                                             map->queues[--map->len];
1069                                 else {
1070                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1071                                             NULL);
1072                                         call_rcu(&map->rcu, xps_map_release);
1073                                         map = NULL;
1074                                 }
1075                         }
1076                         if (map)
1077                                 nonempty = 1;
1078                 }
1079
1080                 if (!nonempty) {
1081                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1082                         call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1083                 }
1084         }
1085
1086         mutex_unlock(&xps_map_mutex);
1087
1088         memset(kobj, 0, sizeof(*kobj));
1089         dev_put(queue->dev);
1090 }
1091
1092 static struct kobj_type netdev_queue_ktype = {
1093         .sysfs_ops = &netdev_queue_sysfs_ops,
1094         .release = netdev_queue_release,
1095         .default_attrs = netdev_queue_default_attrs,
1096 };
1097
1098 static int netdev_queue_add_kobject(struct net_device *net, int index)
1099 {
1100         struct netdev_queue *queue = net->_tx + index;
1101         struct kobject *kobj = &queue->kobj;
1102         int error = 0;
1103
1104         kobj->kset = net->queues_kset;
1105         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1106             "tx-%u", index);
1107         if (error) {
1108                 kobject_put(kobj);
1109                 return error;
1110         }
1111
1112         kobject_uevent(kobj, KOBJ_ADD);
1113         dev_hold(queue->dev);
1114
1115         return error;
1116 }
1117 #endif /* CONFIG_XPS */
1118
1119 int
1120 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1121 {
1122 #ifdef CONFIG_XPS
1123         int i;
1124         int error = 0;
1125
1126         for (i = old_num; i < new_num; i++) {
1127                 error = netdev_queue_add_kobject(net, i);
1128                 if (error) {
1129                         new_num = old_num;
1130                         break;
1131                 }
1132         }
1133
1134         while (--i >= new_num)
1135                 kobject_put(&net->_tx[i].kobj);
1136
1137         return error;
1138 #else
1139         return 0;
1140 #endif
1141 }
1142
1143 static int register_queue_kobjects(struct net_device *net)
1144 {
1145         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1146
1147 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1148         net->queues_kset = kset_create_and_add("queues",
1149             NULL, &net->dev.kobj);
1150         if (!net->queues_kset)
1151                 return -ENOMEM;
1152 #endif
1153
1154 #ifdef CONFIG_RPS
1155         real_rx = net->real_num_rx_queues;
1156 #endif
1157         real_tx = net->real_num_tx_queues;
1158
1159         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1160         if (error)
1161                 goto error;
1162         rxq = real_rx;
1163
1164         error = netdev_queue_update_kobjects(net, 0, real_tx);
1165         if (error)
1166                 goto error;
1167         txq = real_tx;
1168
1169         return 0;
1170
1171 error:
1172         netdev_queue_update_kobjects(net, txq, 0);
1173         net_rx_queue_update_kobjects(net, rxq, 0);
1174         return error;
1175 }
1176
1177 static void remove_queue_kobjects(struct net_device *net)
1178 {
1179         int real_rx = 0, real_tx = 0;
1180
1181 #ifdef CONFIG_RPS
1182         real_rx = net->real_num_rx_queues;
1183 #endif
1184         real_tx = net->real_num_tx_queues;
1185
1186         net_rx_queue_update_kobjects(net, real_rx, 0);
1187         netdev_queue_update_kobjects(net, real_tx, 0);
1188 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1189         kset_unregister(net->queues_kset);
1190 #endif
1191 }
1192
1193 static const void *net_current_ns(void)
1194 {
1195         return current->nsproxy->net_ns;
1196 }
1197
1198 static const void *net_initial_ns(void)
1199 {
1200         return &init_net;
1201 }
1202
1203 static const void *net_netlink_ns(struct sock *sk)
1204 {
1205         return sock_net(sk);
1206 }
1207
1208 struct kobj_ns_type_operations net_ns_type_operations = {
1209         .type = KOBJ_NS_TYPE_NET,
1210         .current_ns = net_current_ns,
1211         .netlink_ns = net_netlink_ns,
1212         .initial_ns = net_initial_ns,
1213 };
1214 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1215
1216 static void net_kobj_ns_exit(struct net *net)
1217 {
1218         kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
1219 }
1220
1221 static struct pernet_operations kobj_net_ops = {
1222         .exit = net_kobj_ns_exit,
1223 };
1224
1225
1226 #ifdef CONFIG_HOTPLUG
1227 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1228 {
1229         struct net_device *dev = to_net_dev(d);
1230         int retval;
1231
1232         /* pass interface to uevent. */
1233         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1234         if (retval)
1235                 goto exit;
1236
1237         /* pass ifindex to uevent.
1238          * ifindex is useful as it won't change (interface name may change)
1239          * and is what RtNetlink uses natively. */
1240         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1241
1242 exit:
1243         return retval;
1244 }
1245 #endif
1246
1247 /*
1248  *      netdev_release -- destroy and free a dead device.
1249  *      Called when last reference to device kobject is gone.
1250  */
1251 static void netdev_release(struct device *d)
1252 {
1253         struct net_device *dev = to_net_dev(d);
1254
1255         BUG_ON(dev->reg_state != NETREG_RELEASED);
1256
1257         kfree(dev->ifalias);
1258         kfree((char *)dev - dev->padded);
1259 }
1260
1261 static const void *net_namespace(struct device *d)
1262 {
1263         struct net_device *dev;
1264         dev = container_of(d, struct net_device, dev);
1265         return dev_net(dev);
1266 }
1267
1268 static struct class net_class = {
1269         .name = "net",
1270         .dev_release = netdev_release,
1271 #ifdef CONFIG_SYSFS
1272         .dev_attrs = net_class_attributes,
1273 #endif /* CONFIG_SYSFS */
1274 #ifdef CONFIG_HOTPLUG
1275         .dev_uevent = netdev_uevent,
1276 #endif
1277         .ns_type = &net_ns_type_operations,
1278         .namespace = net_namespace,
1279 };
1280
1281 /* Delete sysfs entries but hold kobject reference until after all
1282  * netdev references are gone.
1283  */
1284 void netdev_unregister_kobject(struct net_device * net)
1285 {
1286         struct device *dev = &(net->dev);
1287
1288         kobject_get(&dev->kobj);
1289
1290         remove_queue_kobjects(net);
1291
1292         device_del(dev);
1293 }
1294
1295 /* Create sysfs entries for network device. */
1296 int netdev_register_kobject(struct net_device *net)
1297 {
1298         struct device *dev = &(net->dev);
1299         const struct attribute_group **groups = net->sysfs_groups;
1300         int error = 0;
1301
1302         device_initialize(dev);
1303         dev->class = &net_class;
1304         dev->platform_data = net;
1305         dev->groups = groups;
1306
1307         dev_set_name(dev, "%s", net->name);
1308
1309 #ifdef CONFIG_SYSFS
1310         /* Allow for a device specific group */
1311         if (*groups)
1312                 groups++;
1313
1314         *groups++ = &netstat_group;
1315 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1316         if (net->ieee80211_ptr)
1317                 *groups++ = &wireless_group;
1318 #ifdef CONFIG_WIRELESS_EXT
1319         else if (net->wireless_handlers)
1320                 *groups++ = &wireless_group;
1321 #endif
1322 #endif
1323 #endif /* CONFIG_SYSFS */
1324
1325         error = device_add(dev);
1326         if (error)
1327                 return error;
1328
1329         error = register_queue_kobjects(net);
1330         if (error) {
1331                 device_del(dev);
1332                 return error;
1333         }
1334
1335         return error;
1336 }
1337
1338 int netdev_class_create_file(struct class_attribute *class_attr)
1339 {
1340         return class_create_file(&net_class, class_attr);
1341 }
1342 EXPORT_SYMBOL(netdev_class_create_file);
1343
1344 void netdev_class_remove_file(struct class_attribute *class_attr)
1345 {
1346         class_remove_file(&net_class, class_attr);
1347 }
1348 EXPORT_SYMBOL(netdev_class_remove_file);
1349
1350 int netdev_kobject_init(void)
1351 {
1352         kobj_ns_type_register(&net_ns_type_operations);
1353         register_pernet_subsys(&kobj_net_ops);
1354         return class_register(&net_class);
1355 }