Merge branches 'acpica', 'aml-custom', 'bugzilla-16548', 'bugzilla-20242', 'd3-cold...
[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_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 NETDEVICE_SHOW(group, fmt_dec);
299
300 static int change_group(struct net_device *net, unsigned long new_group)
301 {
302         dev_set_group(net, (int) new_group);
303         return 0;
304 }
305
306 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
307                          const char *buf, size_t len)
308 {
309         return netdev_store(dev, attr, buf, len, change_group);
310 }
311
312 static struct device_attribute net_class_attributes[] = {
313         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
314         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
315         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
316         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
317         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
318         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
319         __ATTR(features, S_IRUGO, show_features, NULL),
320         __ATTR(type, S_IRUGO, show_type, NULL),
321         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
322         __ATTR(address, S_IRUGO, show_address, NULL),
323         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
324         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
325         __ATTR(speed, S_IRUGO, show_speed, NULL),
326         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
327         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
328         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
329         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
330         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
331         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
332                store_tx_queue_len),
333         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
334         {}
335 };
336
337 /* Show a given an attribute in the statistics group */
338 static ssize_t netstat_show(const struct device *d,
339                             struct device_attribute *attr, char *buf,
340                             unsigned long offset)
341 {
342         struct net_device *dev = to_net_dev(d);
343         ssize_t ret = -EINVAL;
344
345         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
346                         offset % sizeof(u64) != 0);
347
348         read_lock(&dev_base_lock);
349         if (dev_isalive(dev)) {
350                 struct rtnl_link_stats64 temp;
351                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
352
353                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
354         }
355         read_unlock(&dev_base_lock);
356         return ret;
357 }
358
359 /* generate a read-only statistics attribute */
360 #define NETSTAT_ENTRY(name)                                             \
361 static ssize_t show_##name(struct device *d,                            \
362                            struct device_attribute *attr, char *buf)    \
363 {                                                                       \
364         return netstat_show(d, attr, buf,                               \
365                             offsetof(struct rtnl_link_stats64, name));  \
366 }                                                                       \
367 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
368
369 NETSTAT_ENTRY(rx_packets);
370 NETSTAT_ENTRY(tx_packets);
371 NETSTAT_ENTRY(rx_bytes);
372 NETSTAT_ENTRY(tx_bytes);
373 NETSTAT_ENTRY(rx_errors);
374 NETSTAT_ENTRY(tx_errors);
375 NETSTAT_ENTRY(rx_dropped);
376 NETSTAT_ENTRY(tx_dropped);
377 NETSTAT_ENTRY(multicast);
378 NETSTAT_ENTRY(collisions);
379 NETSTAT_ENTRY(rx_length_errors);
380 NETSTAT_ENTRY(rx_over_errors);
381 NETSTAT_ENTRY(rx_crc_errors);
382 NETSTAT_ENTRY(rx_frame_errors);
383 NETSTAT_ENTRY(rx_fifo_errors);
384 NETSTAT_ENTRY(rx_missed_errors);
385 NETSTAT_ENTRY(tx_aborted_errors);
386 NETSTAT_ENTRY(tx_carrier_errors);
387 NETSTAT_ENTRY(tx_fifo_errors);
388 NETSTAT_ENTRY(tx_heartbeat_errors);
389 NETSTAT_ENTRY(tx_window_errors);
390 NETSTAT_ENTRY(rx_compressed);
391 NETSTAT_ENTRY(tx_compressed);
392
393 static struct attribute *netstat_attrs[] = {
394         &dev_attr_rx_packets.attr,
395         &dev_attr_tx_packets.attr,
396         &dev_attr_rx_bytes.attr,
397         &dev_attr_tx_bytes.attr,
398         &dev_attr_rx_errors.attr,
399         &dev_attr_tx_errors.attr,
400         &dev_attr_rx_dropped.attr,
401         &dev_attr_tx_dropped.attr,
402         &dev_attr_multicast.attr,
403         &dev_attr_collisions.attr,
404         &dev_attr_rx_length_errors.attr,
405         &dev_attr_rx_over_errors.attr,
406         &dev_attr_rx_crc_errors.attr,
407         &dev_attr_rx_frame_errors.attr,
408         &dev_attr_rx_fifo_errors.attr,
409         &dev_attr_rx_missed_errors.attr,
410         &dev_attr_tx_aborted_errors.attr,
411         &dev_attr_tx_carrier_errors.attr,
412         &dev_attr_tx_fifo_errors.attr,
413         &dev_attr_tx_heartbeat_errors.attr,
414         &dev_attr_tx_window_errors.attr,
415         &dev_attr_rx_compressed.attr,
416         &dev_attr_tx_compressed.attr,
417         NULL
418 };
419
420
421 static struct attribute_group netstat_group = {
422         .name  = "statistics",
423         .attrs  = netstat_attrs,
424 };
425
426 #ifdef CONFIG_WIRELESS_EXT_SYSFS
427 /* helper function that does all the locking etc for wireless stats */
428 static ssize_t wireless_show(struct device *d, char *buf,
429                              ssize_t (*format)(const struct iw_statistics *,
430                                                char *))
431 {
432         struct net_device *dev = to_net_dev(d);
433         const struct iw_statistics *iw;
434         ssize_t ret = -EINVAL;
435
436         if (!rtnl_trylock())
437                 return restart_syscall();
438         if (dev_isalive(dev)) {
439                 iw = get_wireless_stats(dev);
440                 if (iw)
441                         ret = (*format)(iw, buf);
442         }
443         rtnl_unlock();
444
445         return ret;
446 }
447
448 /* show function template for wireless fields */
449 #define WIRELESS_SHOW(name, field, format_string)                       \
450 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
451 {                                                                       \
452         return sprintf(buf, format_string, iw->field);                  \
453 }                                                                       \
454 static ssize_t show_iw_##name(struct device *d,                         \
455                               struct device_attribute *attr, char *buf) \
456 {                                                                       \
457         return wireless_show(d, buf, format_iw_##name);                 \
458 }                                                                       \
459 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
460
461 WIRELESS_SHOW(status, status, fmt_hex);
462 WIRELESS_SHOW(link, qual.qual, fmt_dec);
463 WIRELESS_SHOW(level, qual.level, fmt_dec);
464 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
465 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
466 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
467 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
468 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
469 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
470 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
471
472 static struct attribute *wireless_attrs[] = {
473         &dev_attr_status.attr,
474         &dev_attr_link.attr,
475         &dev_attr_level.attr,
476         &dev_attr_noise.attr,
477         &dev_attr_nwid.attr,
478         &dev_attr_crypt.attr,
479         &dev_attr_fragment.attr,
480         &dev_attr_retries.attr,
481         &dev_attr_misc.attr,
482         &dev_attr_beacon.attr,
483         NULL
484 };
485
486 static struct attribute_group wireless_group = {
487         .name = "wireless",
488         .attrs = wireless_attrs,
489 };
490 #endif
491 #endif /* CONFIG_SYSFS */
492
493 #ifdef CONFIG_RPS
494 /*
495  * RX queue sysfs structures and functions.
496  */
497 struct rx_queue_attribute {
498         struct attribute attr;
499         ssize_t (*show)(struct netdev_rx_queue *queue,
500             struct rx_queue_attribute *attr, char *buf);
501         ssize_t (*store)(struct netdev_rx_queue *queue,
502             struct rx_queue_attribute *attr, const char *buf, size_t len);
503 };
504 #define to_rx_queue_attr(_attr) container_of(_attr,             \
505     struct rx_queue_attribute, attr)
506
507 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
508
509 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
510                                   char *buf)
511 {
512         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
513         struct netdev_rx_queue *queue = to_rx_queue(kobj);
514
515         if (!attribute->show)
516                 return -EIO;
517
518         return attribute->show(queue, attribute, buf);
519 }
520
521 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
522                                    const char *buf, size_t count)
523 {
524         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
525         struct netdev_rx_queue *queue = to_rx_queue(kobj);
526
527         if (!attribute->store)
528                 return -EIO;
529
530         return attribute->store(queue, attribute, buf, count);
531 }
532
533 static const struct sysfs_ops rx_queue_sysfs_ops = {
534         .show = rx_queue_attr_show,
535         .store = rx_queue_attr_store,
536 };
537
538 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
539                             struct rx_queue_attribute *attribute, char *buf)
540 {
541         struct rps_map *map;
542         cpumask_var_t mask;
543         size_t len = 0;
544         int i;
545
546         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
547                 return -ENOMEM;
548
549         rcu_read_lock();
550         map = rcu_dereference(queue->rps_map);
551         if (map)
552                 for (i = 0; i < map->len; i++)
553                         cpumask_set_cpu(map->cpus[i], mask);
554
555         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
556         if (PAGE_SIZE - len < 3) {
557                 rcu_read_unlock();
558                 free_cpumask_var(mask);
559                 return -EINVAL;
560         }
561         rcu_read_unlock();
562
563         free_cpumask_var(mask);
564         len += sprintf(buf + len, "\n");
565         return len;
566 }
567
568 static void rps_map_release(struct rcu_head *rcu)
569 {
570         struct rps_map *map = container_of(rcu, struct rps_map, rcu);
571
572         kfree(map);
573 }
574
575 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
576                       struct rx_queue_attribute *attribute,
577                       const char *buf, size_t len)
578 {
579         struct rps_map *old_map, *map;
580         cpumask_var_t mask;
581         int err, cpu, i;
582         static DEFINE_SPINLOCK(rps_map_lock);
583
584         if (!capable(CAP_NET_ADMIN))
585                 return -EPERM;
586
587         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
588                 return -ENOMEM;
589
590         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
591         if (err) {
592                 free_cpumask_var(mask);
593                 return err;
594         }
595
596         map = kzalloc(max_t(unsigned,
597             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
598             GFP_KERNEL);
599         if (!map) {
600                 free_cpumask_var(mask);
601                 return -ENOMEM;
602         }
603
604         i = 0;
605         for_each_cpu_and(cpu, mask, cpu_online_mask)
606                 map->cpus[i++] = cpu;
607
608         if (i)
609                 map->len = i;
610         else {
611                 kfree(map);
612                 map = NULL;
613         }
614
615         spin_lock(&rps_map_lock);
616         old_map = rcu_dereference_protected(queue->rps_map,
617                                             lockdep_is_held(&rps_map_lock));
618         rcu_assign_pointer(queue->rps_map, map);
619         spin_unlock(&rps_map_lock);
620
621         if (old_map)
622                 call_rcu(&old_map->rcu, rps_map_release);
623
624         free_cpumask_var(mask);
625         return len;
626 }
627
628 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
629                                            struct rx_queue_attribute *attr,
630                                            char *buf)
631 {
632         struct rps_dev_flow_table *flow_table;
633         unsigned int val = 0;
634
635         rcu_read_lock();
636         flow_table = rcu_dereference(queue->rps_flow_table);
637         if (flow_table)
638                 val = flow_table->mask + 1;
639         rcu_read_unlock();
640
641         return sprintf(buf, "%u\n", val);
642 }
643
644 static void rps_dev_flow_table_release_work(struct work_struct *work)
645 {
646         struct rps_dev_flow_table *table = container_of(work,
647             struct rps_dev_flow_table, free_work);
648
649         vfree(table);
650 }
651
652 static void rps_dev_flow_table_release(struct rcu_head *rcu)
653 {
654         struct rps_dev_flow_table *table = container_of(rcu,
655             struct rps_dev_flow_table, rcu);
656
657         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
658         schedule_work(&table->free_work);
659 }
660
661 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
662                                      struct rx_queue_attribute *attr,
663                                      const char *buf, size_t len)
664 {
665         unsigned int count;
666         char *endp;
667         struct rps_dev_flow_table *table, *old_table;
668         static DEFINE_SPINLOCK(rps_dev_flow_lock);
669
670         if (!capable(CAP_NET_ADMIN))
671                 return -EPERM;
672
673         count = simple_strtoul(buf, &endp, 0);
674         if (endp == buf)
675                 return -EINVAL;
676
677         if (count) {
678                 int i;
679
680                 if (count > 1<<30) {
681                         /* Enforce a limit to prevent overflow */
682                         return -EINVAL;
683                 }
684                 count = roundup_pow_of_two(count);
685                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
686                 if (!table)
687                         return -ENOMEM;
688
689                 table->mask = count - 1;
690                 for (i = 0; i < count; i++)
691                         table->flows[i].cpu = RPS_NO_CPU;
692         } else
693                 table = NULL;
694
695         spin_lock(&rps_dev_flow_lock);
696         old_table = rcu_dereference_protected(queue->rps_flow_table,
697                                               lockdep_is_held(&rps_dev_flow_lock));
698         rcu_assign_pointer(queue->rps_flow_table, table);
699         spin_unlock(&rps_dev_flow_lock);
700
701         if (old_table)
702                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
703
704         return len;
705 }
706
707 static struct rx_queue_attribute rps_cpus_attribute =
708         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
709
710
711 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
712         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
713             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
714
715 static struct attribute *rx_queue_default_attrs[] = {
716         &rps_cpus_attribute.attr,
717         &rps_dev_flow_table_cnt_attribute.attr,
718         NULL
719 };
720
721 static void rx_queue_release(struct kobject *kobj)
722 {
723         struct netdev_rx_queue *queue = to_rx_queue(kobj);
724         struct rps_map *map;
725         struct rps_dev_flow_table *flow_table;
726
727
728         map = rcu_dereference_raw(queue->rps_map);
729         if (map) {
730                 RCU_INIT_POINTER(queue->rps_map, NULL);
731                 call_rcu(&map->rcu, rps_map_release);
732         }
733
734         flow_table = rcu_dereference_raw(queue->rps_flow_table);
735         if (flow_table) {
736                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
737                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
738         }
739
740         memset(kobj, 0, sizeof(*kobj));
741         dev_put(queue->dev);
742 }
743
744 static struct kobj_type rx_queue_ktype = {
745         .sysfs_ops = &rx_queue_sysfs_ops,
746         .release = rx_queue_release,
747         .default_attrs = rx_queue_default_attrs,
748 };
749
750 static int rx_queue_add_kobject(struct net_device *net, int index)
751 {
752         struct netdev_rx_queue *queue = net->_rx + index;
753         struct kobject *kobj = &queue->kobj;
754         int error = 0;
755
756         kobj->kset = net->queues_kset;
757         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
758             "rx-%u", index);
759         if (error) {
760                 kobject_put(kobj);
761                 return error;
762         }
763
764         kobject_uevent(kobj, KOBJ_ADD);
765         dev_hold(queue->dev);
766
767         return error;
768 }
769 #endif /* CONFIG_RPS */
770
771 int
772 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
773 {
774 #ifdef CONFIG_RPS
775         int i;
776         int error = 0;
777
778         for (i = old_num; i < new_num; i++) {
779                 error = rx_queue_add_kobject(net, i);
780                 if (error) {
781                         new_num = old_num;
782                         break;
783                 }
784         }
785
786         while (--i >= new_num)
787                 kobject_put(&net->_rx[i].kobj);
788
789         return error;
790 #else
791         return 0;
792 #endif
793 }
794
795 #ifdef CONFIG_XPS
796 /*
797  * netdev_queue sysfs structures and functions.
798  */
799 struct netdev_queue_attribute {
800         struct attribute attr;
801         ssize_t (*show)(struct netdev_queue *queue,
802             struct netdev_queue_attribute *attr, char *buf);
803         ssize_t (*store)(struct netdev_queue *queue,
804             struct netdev_queue_attribute *attr, const char *buf, size_t len);
805 };
806 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
807     struct netdev_queue_attribute, attr)
808
809 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
810
811 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
812                                       struct attribute *attr, char *buf)
813 {
814         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
815         struct netdev_queue *queue = to_netdev_queue(kobj);
816
817         if (!attribute->show)
818                 return -EIO;
819
820         return attribute->show(queue, attribute, buf);
821 }
822
823 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
824                                        struct attribute *attr,
825                                        const char *buf, size_t count)
826 {
827         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
828         struct netdev_queue *queue = to_netdev_queue(kobj);
829
830         if (!attribute->store)
831                 return -EIO;
832
833         return attribute->store(queue, attribute, buf, count);
834 }
835
836 static const struct sysfs_ops netdev_queue_sysfs_ops = {
837         .show = netdev_queue_attr_show,
838         .store = netdev_queue_attr_store,
839 };
840
841 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
842 {
843         struct net_device *dev = queue->dev;
844         int i;
845
846         for (i = 0; i < dev->num_tx_queues; i++)
847                 if (queue == &dev->_tx[i])
848                         break;
849
850         BUG_ON(i >= dev->num_tx_queues);
851
852         return i;
853 }
854
855
856 static ssize_t show_xps_map(struct netdev_queue *queue,
857                             struct netdev_queue_attribute *attribute, char *buf)
858 {
859         struct net_device *dev = queue->dev;
860         struct xps_dev_maps *dev_maps;
861         cpumask_var_t mask;
862         unsigned long index;
863         size_t len = 0;
864         int i;
865
866         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
867                 return -ENOMEM;
868
869         index = get_netdev_queue_index(queue);
870
871         rcu_read_lock();
872         dev_maps = rcu_dereference(dev->xps_maps);
873         if (dev_maps) {
874                 for_each_possible_cpu(i) {
875                         struct xps_map *map =
876                             rcu_dereference(dev_maps->cpu_map[i]);
877                         if (map) {
878                                 int j;
879                                 for (j = 0; j < map->len; j++) {
880                                         if (map->queues[j] == index) {
881                                                 cpumask_set_cpu(i, mask);
882                                                 break;
883                                         }
884                                 }
885                         }
886                 }
887         }
888         rcu_read_unlock();
889
890         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
891         if (PAGE_SIZE - len < 3) {
892                 free_cpumask_var(mask);
893                 return -EINVAL;
894         }
895
896         free_cpumask_var(mask);
897         len += sprintf(buf + len, "\n");
898         return len;
899 }
900
901 static void xps_map_release(struct rcu_head *rcu)
902 {
903         struct xps_map *map = container_of(rcu, struct xps_map, rcu);
904
905         kfree(map);
906 }
907
908 static void xps_dev_maps_release(struct rcu_head *rcu)
909 {
910         struct xps_dev_maps *dev_maps =
911             container_of(rcu, struct xps_dev_maps, rcu);
912
913         kfree(dev_maps);
914 }
915
916 static DEFINE_MUTEX(xps_map_mutex);
917 #define xmap_dereference(P)             \
918         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
919
920 static ssize_t store_xps_map(struct netdev_queue *queue,
921                       struct netdev_queue_attribute *attribute,
922                       const char *buf, size_t len)
923 {
924         struct net_device *dev = queue->dev;
925         cpumask_var_t mask;
926         int err, i, cpu, pos, map_len, alloc_len, need_set;
927         unsigned long index;
928         struct xps_map *map, *new_map;
929         struct xps_dev_maps *dev_maps, *new_dev_maps;
930         int nonempty = 0;
931         int numa_node = -2;
932
933         if (!capable(CAP_NET_ADMIN))
934                 return -EPERM;
935
936         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
937                 return -ENOMEM;
938
939         index = get_netdev_queue_index(queue);
940
941         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
942         if (err) {
943                 free_cpumask_var(mask);
944                 return err;
945         }
946
947         new_dev_maps = kzalloc(max_t(unsigned,
948             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
949         if (!new_dev_maps) {
950                 free_cpumask_var(mask);
951                 return -ENOMEM;
952         }
953
954         mutex_lock(&xps_map_mutex);
955
956         dev_maps = xmap_dereference(dev->xps_maps);
957
958         for_each_possible_cpu(cpu) {
959                 map = dev_maps ?
960                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
961                 new_map = map;
962                 if (map) {
963                         for (pos = 0; pos < map->len; pos++)
964                                 if (map->queues[pos] == index)
965                                         break;
966                         map_len = map->len;
967                         alloc_len = map->alloc_len;
968                 } else
969                         pos = map_len = alloc_len = 0;
970
971                 need_set = cpu_isset(cpu, *mask) && cpu_online(cpu);
972 #ifdef CONFIG_NUMA
973                 if (need_set) {
974                         if (numa_node == -2)
975                                 numa_node = cpu_to_node(cpu);
976                         else if (numa_node != cpu_to_node(cpu))
977                                 numa_node = -1;
978                 }
979 #endif
980                 if (need_set && pos >= map_len) {
981                         /* Need to add queue to this CPU's map */
982                         if (map_len >= alloc_len) {
983                                 alloc_len = alloc_len ?
984                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
985                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
986                                                        GFP_KERNEL,
987                                                        cpu_to_node(cpu));
988                                 if (!new_map)
989                                         goto error;
990                                 new_map->alloc_len = alloc_len;
991                                 for (i = 0; i < map_len; i++)
992                                         new_map->queues[i] = map->queues[i];
993                                 new_map->len = map_len;
994                         }
995                         new_map->queues[new_map->len++] = index;
996                 } else if (!need_set && pos < map_len) {
997                         /* Need to remove queue from this CPU's map */
998                         if (map_len > 1)
999                                 new_map->queues[pos] =
1000                                     new_map->queues[--new_map->len];
1001                         else
1002                                 new_map = NULL;
1003                 }
1004                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1005         }
1006
1007         /* Cleanup old maps */
1008         for_each_possible_cpu(cpu) {
1009                 map = dev_maps ?
1010                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1011                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1012                         call_rcu(&map->rcu, xps_map_release);
1013                 if (new_dev_maps->cpu_map[cpu])
1014                         nonempty = 1;
1015         }
1016
1017         if (nonempty)
1018                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1019         else {
1020                 kfree(new_dev_maps);
1021                 rcu_assign_pointer(dev->xps_maps, NULL);
1022         }
1023
1024         if (dev_maps)
1025                 call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1026
1027         netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1028                                             NUMA_NO_NODE);
1029
1030         mutex_unlock(&xps_map_mutex);
1031
1032         free_cpumask_var(mask);
1033         return len;
1034
1035 error:
1036         mutex_unlock(&xps_map_mutex);
1037
1038         if (new_dev_maps)
1039                 for_each_possible_cpu(i)
1040                         kfree(rcu_dereference_protected(
1041                                 new_dev_maps->cpu_map[i],
1042                                 1));
1043         kfree(new_dev_maps);
1044         free_cpumask_var(mask);
1045         return -ENOMEM;
1046 }
1047
1048 static struct netdev_queue_attribute xps_cpus_attribute =
1049     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1050
1051 static struct attribute *netdev_queue_default_attrs[] = {
1052         &xps_cpus_attribute.attr,
1053         NULL
1054 };
1055
1056 static void netdev_queue_release(struct kobject *kobj)
1057 {
1058         struct netdev_queue *queue = to_netdev_queue(kobj);
1059         struct net_device *dev = queue->dev;
1060         struct xps_dev_maps *dev_maps;
1061         struct xps_map *map;
1062         unsigned long index;
1063         int i, pos, nonempty = 0;
1064
1065         index = get_netdev_queue_index(queue);
1066
1067         mutex_lock(&xps_map_mutex);
1068         dev_maps = xmap_dereference(dev->xps_maps);
1069
1070         if (dev_maps) {
1071                 for_each_possible_cpu(i) {
1072                         map = xmap_dereference(dev_maps->cpu_map[i]);
1073                         if (!map)
1074                                 continue;
1075
1076                         for (pos = 0; pos < map->len; pos++)
1077                                 if (map->queues[pos] == index)
1078                                         break;
1079
1080                         if (pos < map->len) {
1081                                 if (map->len > 1)
1082                                         map->queues[pos] =
1083                                             map->queues[--map->len];
1084                                 else {
1085                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1086                                             NULL);
1087                                         call_rcu(&map->rcu, xps_map_release);
1088                                         map = NULL;
1089                                 }
1090                         }
1091                         if (map)
1092                                 nonempty = 1;
1093                 }
1094
1095                 if (!nonempty) {
1096                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1097                         call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1098                 }
1099         }
1100
1101         mutex_unlock(&xps_map_mutex);
1102
1103         memset(kobj, 0, sizeof(*kobj));
1104         dev_put(queue->dev);
1105 }
1106
1107 static struct kobj_type netdev_queue_ktype = {
1108         .sysfs_ops = &netdev_queue_sysfs_ops,
1109         .release = netdev_queue_release,
1110         .default_attrs = netdev_queue_default_attrs,
1111 };
1112
1113 static int netdev_queue_add_kobject(struct net_device *net, int index)
1114 {
1115         struct netdev_queue *queue = net->_tx + index;
1116         struct kobject *kobj = &queue->kobj;
1117         int error = 0;
1118
1119         kobj->kset = net->queues_kset;
1120         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1121             "tx-%u", index);
1122         if (error) {
1123                 kobject_put(kobj);
1124                 return error;
1125         }
1126
1127         kobject_uevent(kobj, KOBJ_ADD);
1128         dev_hold(queue->dev);
1129
1130         return error;
1131 }
1132 #endif /* CONFIG_XPS */
1133
1134 int
1135 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1136 {
1137 #ifdef CONFIG_XPS
1138         int i;
1139         int error = 0;
1140
1141         for (i = old_num; i < new_num; i++) {
1142                 error = netdev_queue_add_kobject(net, i);
1143                 if (error) {
1144                         new_num = old_num;
1145                         break;
1146                 }
1147         }
1148
1149         while (--i >= new_num)
1150                 kobject_put(&net->_tx[i].kobj);
1151
1152         return error;
1153 #else
1154         return 0;
1155 #endif
1156 }
1157
1158 static int register_queue_kobjects(struct net_device *net)
1159 {
1160         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1161
1162 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1163         net->queues_kset = kset_create_and_add("queues",
1164             NULL, &net->dev.kobj);
1165         if (!net->queues_kset)
1166                 return -ENOMEM;
1167 #endif
1168
1169 #ifdef CONFIG_RPS
1170         real_rx = net->real_num_rx_queues;
1171 #endif
1172         real_tx = net->real_num_tx_queues;
1173
1174         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1175         if (error)
1176                 goto error;
1177         rxq = real_rx;
1178
1179         error = netdev_queue_update_kobjects(net, 0, real_tx);
1180         if (error)
1181                 goto error;
1182         txq = real_tx;
1183
1184         return 0;
1185
1186 error:
1187         netdev_queue_update_kobjects(net, txq, 0);
1188         net_rx_queue_update_kobjects(net, rxq, 0);
1189         return error;
1190 }
1191
1192 static void remove_queue_kobjects(struct net_device *net)
1193 {
1194         int real_rx = 0, real_tx = 0;
1195
1196 #ifdef CONFIG_RPS
1197         real_rx = net->real_num_rx_queues;
1198 #endif
1199         real_tx = net->real_num_tx_queues;
1200
1201         net_rx_queue_update_kobjects(net, real_rx, 0);
1202         netdev_queue_update_kobjects(net, real_tx, 0);
1203 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1204         kset_unregister(net->queues_kset);
1205 #endif
1206 }
1207
1208 static const void *net_current_ns(void)
1209 {
1210         return current->nsproxy->net_ns;
1211 }
1212
1213 static const void *net_initial_ns(void)
1214 {
1215         return &init_net;
1216 }
1217
1218 static const void *net_netlink_ns(struct sock *sk)
1219 {
1220         return sock_net(sk);
1221 }
1222
1223 struct kobj_ns_type_operations net_ns_type_operations = {
1224         .type = KOBJ_NS_TYPE_NET,
1225         .current_ns = net_current_ns,
1226         .netlink_ns = net_netlink_ns,
1227         .initial_ns = net_initial_ns,
1228 };
1229 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1230
1231 static void net_kobj_ns_exit(struct net *net)
1232 {
1233         kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
1234 }
1235
1236 static struct pernet_operations kobj_net_ops = {
1237         .exit = net_kobj_ns_exit,
1238 };
1239
1240
1241 #ifdef CONFIG_HOTPLUG
1242 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1243 {
1244         struct net_device *dev = to_net_dev(d);
1245         int retval;
1246
1247         /* pass interface to uevent. */
1248         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1249         if (retval)
1250                 goto exit;
1251
1252         /* pass ifindex to uevent.
1253          * ifindex is useful as it won't change (interface name may change)
1254          * and is what RtNetlink uses natively. */
1255         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1256
1257 exit:
1258         return retval;
1259 }
1260 #endif
1261
1262 /*
1263  *      netdev_release -- destroy and free a dead device.
1264  *      Called when last reference to device kobject is gone.
1265  */
1266 static void netdev_release(struct device *d)
1267 {
1268         struct net_device *dev = to_net_dev(d);
1269
1270         BUG_ON(dev->reg_state != NETREG_RELEASED);
1271
1272         kfree(dev->ifalias);
1273         kfree((char *)dev - dev->padded);
1274 }
1275
1276 static const void *net_namespace(struct device *d)
1277 {
1278         struct net_device *dev;
1279         dev = container_of(d, struct net_device, dev);
1280         return dev_net(dev);
1281 }
1282
1283 static struct class net_class = {
1284         .name = "net",
1285         .dev_release = netdev_release,
1286 #ifdef CONFIG_SYSFS
1287         .dev_attrs = net_class_attributes,
1288 #endif /* CONFIG_SYSFS */
1289 #ifdef CONFIG_HOTPLUG
1290         .dev_uevent = netdev_uevent,
1291 #endif
1292         .ns_type = &net_ns_type_operations,
1293         .namespace = net_namespace,
1294 };
1295
1296 /* Delete sysfs entries but hold kobject reference until after all
1297  * netdev references are gone.
1298  */
1299 void netdev_unregister_kobject(struct net_device * net)
1300 {
1301         struct device *dev = &(net->dev);
1302
1303         kobject_get(&dev->kobj);
1304
1305         remove_queue_kobjects(net);
1306
1307         device_del(dev);
1308 }
1309
1310 /* Create sysfs entries for network device. */
1311 int netdev_register_kobject(struct net_device *net)
1312 {
1313         struct device *dev = &(net->dev);
1314         const struct attribute_group **groups = net->sysfs_groups;
1315         int error = 0;
1316
1317         device_initialize(dev);
1318         dev->class = &net_class;
1319         dev->platform_data = net;
1320         dev->groups = groups;
1321
1322         dev_set_name(dev, "%s", net->name);
1323
1324 #ifdef CONFIG_SYSFS
1325         /* Allow for a device specific group */
1326         if (*groups)
1327                 groups++;
1328
1329         *groups++ = &netstat_group;
1330 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1331         if (net->ieee80211_ptr)
1332                 *groups++ = &wireless_group;
1333 #ifdef 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         return error;
1351 }
1352
1353 int netdev_class_create_file(struct class_attribute *class_attr)
1354 {
1355         return class_create_file(&net_class, class_attr);
1356 }
1357 EXPORT_SYMBOL(netdev_class_create_file);
1358
1359 void netdev_class_remove_file(struct class_attribute *class_attr)
1360 {
1361         class_remove_file(&net_class, class_attr);
1362 }
1363 EXPORT_SYMBOL(netdev_class_remove_file);
1364
1365 int netdev_kobject_init(void)
1366 {
1367         kobj_ns_type_register(&net_ns_type_operations);
1368         register_pernet_subsys(&kobj_net_ops);
1369         return class_register(&net_class);
1370 }