net,rcu: convert call_rcu(xps_map_release) to kfree_rcu()
[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 ssize_t store_rps_map(struct netdev_rx_queue *queue,
569                       struct rx_queue_attribute *attribute,
570                       const char *buf, size_t len)
571 {
572         struct rps_map *old_map, *map;
573         cpumask_var_t mask;
574         int err, cpu, i;
575         static DEFINE_SPINLOCK(rps_map_lock);
576
577         if (!capable(CAP_NET_ADMIN))
578                 return -EPERM;
579
580         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
581                 return -ENOMEM;
582
583         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
584         if (err) {
585                 free_cpumask_var(mask);
586                 return err;
587         }
588
589         map = kzalloc(max_t(unsigned,
590             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
591             GFP_KERNEL);
592         if (!map) {
593                 free_cpumask_var(mask);
594                 return -ENOMEM;
595         }
596
597         i = 0;
598         for_each_cpu_and(cpu, mask, cpu_online_mask)
599                 map->cpus[i++] = cpu;
600
601         if (i)
602                 map->len = i;
603         else {
604                 kfree(map);
605                 map = NULL;
606         }
607
608         spin_lock(&rps_map_lock);
609         old_map = rcu_dereference_protected(queue->rps_map,
610                                             lockdep_is_held(&rps_map_lock));
611         rcu_assign_pointer(queue->rps_map, map);
612         spin_unlock(&rps_map_lock);
613
614         if (old_map)
615                 kfree_rcu(old_map, rcu);
616
617         free_cpumask_var(mask);
618         return len;
619 }
620
621 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
622                                            struct rx_queue_attribute *attr,
623                                            char *buf)
624 {
625         struct rps_dev_flow_table *flow_table;
626         unsigned int val = 0;
627
628         rcu_read_lock();
629         flow_table = rcu_dereference(queue->rps_flow_table);
630         if (flow_table)
631                 val = flow_table->mask + 1;
632         rcu_read_unlock();
633
634         return sprintf(buf, "%u\n", val);
635 }
636
637 static void rps_dev_flow_table_release_work(struct work_struct *work)
638 {
639         struct rps_dev_flow_table *table = container_of(work,
640             struct rps_dev_flow_table, free_work);
641
642         vfree(table);
643 }
644
645 static void rps_dev_flow_table_release(struct rcu_head *rcu)
646 {
647         struct rps_dev_flow_table *table = container_of(rcu,
648             struct rps_dev_flow_table, rcu);
649
650         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
651         schedule_work(&table->free_work);
652 }
653
654 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
655                                      struct rx_queue_attribute *attr,
656                                      const char *buf, size_t len)
657 {
658         unsigned int count;
659         char *endp;
660         struct rps_dev_flow_table *table, *old_table;
661         static DEFINE_SPINLOCK(rps_dev_flow_lock);
662
663         if (!capable(CAP_NET_ADMIN))
664                 return -EPERM;
665
666         count = simple_strtoul(buf, &endp, 0);
667         if (endp == buf)
668                 return -EINVAL;
669
670         if (count) {
671                 int i;
672
673                 if (count > 1<<30) {
674                         /* Enforce a limit to prevent overflow */
675                         return -EINVAL;
676                 }
677                 count = roundup_pow_of_two(count);
678                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
679                 if (!table)
680                         return -ENOMEM;
681
682                 table->mask = count - 1;
683                 for (i = 0; i < count; i++)
684                         table->flows[i].cpu = RPS_NO_CPU;
685         } else
686                 table = NULL;
687
688         spin_lock(&rps_dev_flow_lock);
689         old_table = rcu_dereference_protected(queue->rps_flow_table,
690                                               lockdep_is_held(&rps_dev_flow_lock));
691         rcu_assign_pointer(queue->rps_flow_table, table);
692         spin_unlock(&rps_dev_flow_lock);
693
694         if (old_table)
695                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
696
697         return len;
698 }
699
700 static struct rx_queue_attribute rps_cpus_attribute =
701         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
702
703
704 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
705         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
706             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
707
708 static struct attribute *rx_queue_default_attrs[] = {
709         &rps_cpus_attribute.attr,
710         &rps_dev_flow_table_cnt_attribute.attr,
711         NULL
712 };
713
714 static void rx_queue_release(struct kobject *kobj)
715 {
716         struct netdev_rx_queue *queue = to_rx_queue(kobj);
717         struct rps_map *map;
718         struct rps_dev_flow_table *flow_table;
719
720
721         map = rcu_dereference_raw(queue->rps_map);
722         if (map) {
723                 RCU_INIT_POINTER(queue->rps_map, NULL);
724                 kfree_rcu(map, rcu);
725         }
726
727         flow_table = rcu_dereference_raw(queue->rps_flow_table);
728         if (flow_table) {
729                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
730                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
731         }
732
733         memset(kobj, 0, sizeof(*kobj));
734         dev_put(queue->dev);
735 }
736
737 static struct kobj_type rx_queue_ktype = {
738         .sysfs_ops = &rx_queue_sysfs_ops,
739         .release = rx_queue_release,
740         .default_attrs = rx_queue_default_attrs,
741 };
742
743 static int rx_queue_add_kobject(struct net_device *net, int index)
744 {
745         struct netdev_rx_queue *queue = net->_rx + index;
746         struct kobject *kobj = &queue->kobj;
747         int error = 0;
748
749         kobj->kset = net->queues_kset;
750         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
751             "rx-%u", index);
752         if (error) {
753                 kobject_put(kobj);
754                 return error;
755         }
756
757         kobject_uevent(kobj, KOBJ_ADD);
758         dev_hold(queue->dev);
759
760         return error;
761 }
762 #endif /* CONFIG_RPS */
763
764 int
765 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
766 {
767 #ifdef CONFIG_RPS
768         int i;
769         int error = 0;
770
771         for (i = old_num; i < new_num; i++) {
772                 error = rx_queue_add_kobject(net, i);
773                 if (error) {
774                         new_num = old_num;
775                         break;
776                 }
777         }
778
779         while (--i >= new_num)
780                 kobject_put(&net->_rx[i].kobj);
781
782         return error;
783 #else
784         return 0;
785 #endif
786 }
787
788 #ifdef CONFIG_XPS
789 /*
790  * netdev_queue sysfs structures and functions.
791  */
792 struct netdev_queue_attribute {
793         struct attribute attr;
794         ssize_t (*show)(struct netdev_queue *queue,
795             struct netdev_queue_attribute *attr, char *buf);
796         ssize_t (*store)(struct netdev_queue *queue,
797             struct netdev_queue_attribute *attr, const char *buf, size_t len);
798 };
799 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
800     struct netdev_queue_attribute, attr)
801
802 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
803
804 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
805                                       struct attribute *attr, char *buf)
806 {
807         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
808         struct netdev_queue *queue = to_netdev_queue(kobj);
809
810         if (!attribute->show)
811                 return -EIO;
812
813         return attribute->show(queue, attribute, buf);
814 }
815
816 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
817                                        struct attribute *attr,
818                                        const char *buf, size_t count)
819 {
820         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
821         struct netdev_queue *queue = to_netdev_queue(kobj);
822
823         if (!attribute->store)
824                 return -EIO;
825
826         return attribute->store(queue, attribute, buf, count);
827 }
828
829 static const struct sysfs_ops netdev_queue_sysfs_ops = {
830         .show = netdev_queue_attr_show,
831         .store = netdev_queue_attr_store,
832 };
833
834 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
835 {
836         struct net_device *dev = queue->dev;
837         int i;
838
839         for (i = 0; i < dev->num_tx_queues; i++)
840                 if (queue == &dev->_tx[i])
841                         break;
842
843         BUG_ON(i >= dev->num_tx_queues);
844
845         return i;
846 }
847
848
849 static ssize_t show_xps_map(struct netdev_queue *queue,
850                             struct netdev_queue_attribute *attribute, char *buf)
851 {
852         struct net_device *dev = queue->dev;
853         struct xps_dev_maps *dev_maps;
854         cpumask_var_t mask;
855         unsigned long index;
856         size_t len = 0;
857         int i;
858
859         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
860                 return -ENOMEM;
861
862         index = get_netdev_queue_index(queue);
863
864         rcu_read_lock();
865         dev_maps = rcu_dereference(dev->xps_maps);
866         if (dev_maps) {
867                 for_each_possible_cpu(i) {
868                         struct xps_map *map =
869                             rcu_dereference(dev_maps->cpu_map[i]);
870                         if (map) {
871                                 int j;
872                                 for (j = 0; j < map->len; j++) {
873                                         if (map->queues[j] == index) {
874                                                 cpumask_set_cpu(i, mask);
875                                                 break;
876                                         }
877                                 }
878                         }
879                 }
880         }
881         rcu_read_unlock();
882
883         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
884         if (PAGE_SIZE - len < 3) {
885                 free_cpumask_var(mask);
886                 return -EINVAL;
887         }
888
889         free_cpumask_var(mask);
890         len += sprintf(buf + len, "\n");
891         return len;
892 }
893
894 static void xps_dev_maps_release(struct rcu_head *rcu)
895 {
896         struct xps_dev_maps *dev_maps =
897             container_of(rcu, struct xps_dev_maps, rcu);
898
899         kfree(dev_maps);
900 }
901
902 static DEFINE_MUTEX(xps_map_mutex);
903 #define xmap_dereference(P)             \
904         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
905
906 static ssize_t store_xps_map(struct netdev_queue *queue,
907                       struct netdev_queue_attribute *attribute,
908                       const char *buf, size_t len)
909 {
910         struct net_device *dev = queue->dev;
911         cpumask_var_t mask;
912         int err, i, cpu, pos, map_len, alloc_len, need_set;
913         unsigned long index;
914         struct xps_map *map, *new_map;
915         struct xps_dev_maps *dev_maps, *new_dev_maps;
916         int nonempty = 0;
917         int numa_node = -2;
918
919         if (!capable(CAP_NET_ADMIN))
920                 return -EPERM;
921
922         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
923                 return -ENOMEM;
924
925         index = get_netdev_queue_index(queue);
926
927         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
928         if (err) {
929                 free_cpumask_var(mask);
930                 return err;
931         }
932
933         new_dev_maps = kzalloc(max_t(unsigned,
934             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
935         if (!new_dev_maps) {
936                 free_cpumask_var(mask);
937                 return -ENOMEM;
938         }
939
940         mutex_lock(&xps_map_mutex);
941
942         dev_maps = xmap_dereference(dev->xps_maps);
943
944         for_each_possible_cpu(cpu) {
945                 map = dev_maps ?
946                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
947                 new_map = map;
948                 if (map) {
949                         for (pos = 0; pos < map->len; pos++)
950                                 if (map->queues[pos] == index)
951                                         break;
952                         map_len = map->len;
953                         alloc_len = map->alloc_len;
954                 } else
955                         pos = map_len = alloc_len = 0;
956
957                 need_set = cpu_isset(cpu, *mask) && cpu_online(cpu);
958 #ifdef CONFIG_NUMA
959                 if (need_set) {
960                         if (numa_node == -2)
961                                 numa_node = cpu_to_node(cpu);
962                         else if (numa_node != cpu_to_node(cpu))
963                                 numa_node = -1;
964                 }
965 #endif
966                 if (need_set && pos >= map_len) {
967                         /* Need to add queue to this CPU's map */
968                         if (map_len >= alloc_len) {
969                                 alloc_len = alloc_len ?
970                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
971                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
972                                                        GFP_KERNEL,
973                                                        cpu_to_node(cpu));
974                                 if (!new_map)
975                                         goto error;
976                                 new_map->alloc_len = alloc_len;
977                                 for (i = 0; i < map_len; i++)
978                                         new_map->queues[i] = map->queues[i];
979                                 new_map->len = map_len;
980                         }
981                         new_map->queues[new_map->len++] = index;
982                 } else if (!need_set && pos < map_len) {
983                         /* Need to remove queue from this CPU's map */
984                         if (map_len > 1)
985                                 new_map->queues[pos] =
986                                     new_map->queues[--new_map->len];
987                         else
988                                 new_map = NULL;
989                 }
990                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
991         }
992
993         /* Cleanup old maps */
994         for_each_possible_cpu(cpu) {
995                 map = dev_maps ?
996                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
997                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
998                         kfree_rcu(map, rcu);
999                 if (new_dev_maps->cpu_map[cpu])
1000                         nonempty = 1;
1001         }
1002
1003         if (nonempty)
1004                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1005         else {
1006                 kfree(new_dev_maps);
1007                 rcu_assign_pointer(dev->xps_maps, NULL);
1008         }
1009
1010         if (dev_maps)
1011                 call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1012
1013         netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1014                                             NUMA_NO_NODE);
1015
1016         mutex_unlock(&xps_map_mutex);
1017
1018         free_cpumask_var(mask);
1019         return len;
1020
1021 error:
1022         mutex_unlock(&xps_map_mutex);
1023
1024         if (new_dev_maps)
1025                 for_each_possible_cpu(i)
1026                         kfree(rcu_dereference_protected(
1027                                 new_dev_maps->cpu_map[i],
1028                                 1));
1029         kfree(new_dev_maps);
1030         free_cpumask_var(mask);
1031         return -ENOMEM;
1032 }
1033
1034 static struct netdev_queue_attribute xps_cpus_attribute =
1035     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1036
1037 static struct attribute *netdev_queue_default_attrs[] = {
1038         &xps_cpus_attribute.attr,
1039         NULL
1040 };
1041
1042 static void netdev_queue_release(struct kobject *kobj)
1043 {
1044         struct netdev_queue *queue = to_netdev_queue(kobj);
1045         struct net_device *dev = queue->dev;
1046         struct xps_dev_maps *dev_maps;
1047         struct xps_map *map;
1048         unsigned long index;
1049         int i, pos, nonempty = 0;
1050
1051         index = get_netdev_queue_index(queue);
1052
1053         mutex_lock(&xps_map_mutex);
1054         dev_maps = xmap_dereference(dev->xps_maps);
1055
1056         if (dev_maps) {
1057                 for_each_possible_cpu(i) {
1058                         map = xmap_dereference(dev_maps->cpu_map[i]);
1059                         if (!map)
1060                                 continue;
1061
1062                         for (pos = 0; pos < map->len; pos++)
1063                                 if (map->queues[pos] == index)
1064                                         break;
1065
1066                         if (pos < map->len) {
1067                                 if (map->len > 1)
1068                                         map->queues[pos] =
1069                                             map->queues[--map->len];
1070                                 else {
1071                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1072                                             NULL);
1073                                         kfree_rcu(map, rcu);
1074                                         map = NULL;
1075                                 }
1076                         }
1077                         if (map)
1078                                 nonempty = 1;
1079                 }
1080
1081                 if (!nonempty) {
1082                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1083                         call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1084                 }
1085         }
1086
1087         mutex_unlock(&xps_map_mutex);
1088
1089         memset(kobj, 0, sizeof(*kobj));
1090         dev_put(queue->dev);
1091 }
1092
1093 static struct kobj_type netdev_queue_ktype = {
1094         .sysfs_ops = &netdev_queue_sysfs_ops,
1095         .release = netdev_queue_release,
1096         .default_attrs = netdev_queue_default_attrs,
1097 };
1098
1099 static int netdev_queue_add_kobject(struct net_device *net, int index)
1100 {
1101         struct netdev_queue *queue = net->_tx + index;
1102         struct kobject *kobj = &queue->kobj;
1103         int error = 0;
1104
1105         kobj->kset = net->queues_kset;
1106         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1107             "tx-%u", index);
1108         if (error) {
1109                 kobject_put(kobj);
1110                 return error;
1111         }
1112
1113         kobject_uevent(kobj, KOBJ_ADD);
1114         dev_hold(queue->dev);
1115
1116         return error;
1117 }
1118 #endif /* CONFIG_XPS */
1119
1120 int
1121 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1122 {
1123 #ifdef CONFIG_XPS
1124         int i;
1125         int error = 0;
1126
1127         for (i = old_num; i < new_num; i++) {
1128                 error = netdev_queue_add_kobject(net, i);
1129                 if (error) {
1130                         new_num = old_num;
1131                         break;
1132                 }
1133         }
1134
1135         while (--i >= new_num)
1136                 kobject_put(&net->_tx[i].kobj);
1137
1138         return error;
1139 #else
1140         return 0;
1141 #endif
1142 }
1143
1144 static int register_queue_kobjects(struct net_device *net)
1145 {
1146         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1147
1148 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1149         net->queues_kset = kset_create_and_add("queues",
1150             NULL, &net->dev.kobj);
1151         if (!net->queues_kset)
1152                 return -ENOMEM;
1153 #endif
1154
1155 #ifdef CONFIG_RPS
1156         real_rx = net->real_num_rx_queues;
1157 #endif
1158         real_tx = net->real_num_tx_queues;
1159
1160         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1161         if (error)
1162                 goto error;
1163         rxq = real_rx;
1164
1165         error = netdev_queue_update_kobjects(net, 0, real_tx);
1166         if (error)
1167                 goto error;
1168         txq = real_tx;
1169
1170         return 0;
1171
1172 error:
1173         netdev_queue_update_kobjects(net, txq, 0);
1174         net_rx_queue_update_kobjects(net, rxq, 0);
1175         return error;
1176 }
1177
1178 static void remove_queue_kobjects(struct net_device *net)
1179 {
1180         int real_rx = 0, real_tx = 0;
1181
1182 #ifdef CONFIG_RPS
1183         real_rx = net->real_num_rx_queues;
1184 #endif
1185         real_tx = net->real_num_tx_queues;
1186
1187         net_rx_queue_update_kobjects(net, real_rx, 0);
1188         netdev_queue_update_kobjects(net, real_tx, 0);
1189 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1190         kset_unregister(net->queues_kset);
1191 #endif
1192 }
1193
1194 static const void *net_current_ns(void)
1195 {
1196         return current->nsproxy->net_ns;
1197 }
1198
1199 static const void *net_initial_ns(void)
1200 {
1201         return &init_net;
1202 }
1203
1204 static const void *net_netlink_ns(struct sock *sk)
1205 {
1206         return sock_net(sk);
1207 }
1208
1209 struct kobj_ns_type_operations net_ns_type_operations = {
1210         .type = KOBJ_NS_TYPE_NET,
1211         .current_ns = net_current_ns,
1212         .netlink_ns = net_netlink_ns,
1213         .initial_ns = net_initial_ns,
1214 };
1215 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1216
1217 static void net_kobj_ns_exit(struct net *net)
1218 {
1219         kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
1220 }
1221
1222 static struct pernet_operations kobj_net_ops = {
1223         .exit = net_kobj_ns_exit,
1224 };
1225
1226
1227 #ifdef CONFIG_HOTPLUG
1228 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1229 {
1230         struct net_device *dev = to_net_dev(d);
1231         int retval;
1232
1233         /* pass interface to uevent. */
1234         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1235         if (retval)
1236                 goto exit;
1237
1238         /* pass ifindex to uevent.
1239          * ifindex is useful as it won't change (interface name may change)
1240          * and is what RtNetlink uses natively. */
1241         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1242
1243 exit:
1244         return retval;
1245 }
1246 #endif
1247
1248 /*
1249  *      netdev_release -- destroy and free a dead device.
1250  *      Called when last reference to device kobject is gone.
1251  */
1252 static void netdev_release(struct device *d)
1253 {
1254         struct net_device *dev = to_net_dev(d);
1255
1256         BUG_ON(dev->reg_state != NETREG_RELEASED);
1257
1258         kfree(dev->ifalias);
1259         kfree((char *)dev - dev->padded);
1260 }
1261
1262 static const void *net_namespace(struct device *d)
1263 {
1264         struct net_device *dev;
1265         dev = container_of(d, struct net_device, dev);
1266         return dev_net(dev);
1267 }
1268
1269 static struct class net_class = {
1270         .name = "net",
1271         .dev_release = netdev_release,
1272 #ifdef CONFIG_SYSFS
1273         .dev_attrs = net_class_attributes,
1274 #endif /* CONFIG_SYSFS */
1275 #ifdef CONFIG_HOTPLUG
1276         .dev_uevent = netdev_uevent,
1277 #endif
1278         .ns_type = &net_ns_type_operations,
1279         .namespace = net_namespace,
1280 };
1281
1282 /* Delete sysfs entries but hold kobject reference until after all
1283  * netdev references are gone.
1284  */
1285 void netdev_unregister_kobject(struct net_device * net)
1286 {
1287         struct device *dev = &(net->dev);
1288
1289         kobject_get(&dev->kobj);
1290
1291         remove_queue_kobjects(net);
1292
1293         device_del(dev);
1294 }
1295
1296 /* Create sysfs entries for network device. */
1297 int netdev_register_kobject(struct net_device *net)
1298 {
1299         struct device *dev = &(net->dev);
1300         const struct attribute_group **groups = net->sysfs_groups;
1301         int error = 0;
1302
1303         device_initialize(dev);
1304         dev->class = &net_class;
1305         dev->platform_data = net;
1306         dev->groups = groups;
1307
1308         dev_set_name(dev, "%s", net->name);
1309
1310 #ifdef CONFIG_SYSFS
1311         /* Allow for a device specific group */
1312         if (*groups)
1313                 groups++;
1314
1315         *groups++ = &netstat_group;
1316 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1317         if (net->ieee80211_ptr)
1318                 *groups++ = &wireless_group;
1319 #ifdef CONFIG_WIRELESS_EXT
1320         else if (net->wireless_handlers)
1321                 *groups++ = &wireless_group;
1322 #endif
1323 #endif
1324 #endif /* CONFIG_SYSFS */
1325
1326         error = device_add(dev);
1327         if (error)
1328                 return error;
1329
1330         error = register_queue_kobjects(net);
1331         if (error) {
1332                 device_del(dev);
1333                 return error;
1334         }
1335
1336         return error;
1337 }
1338
1339 int netdev_class_create_file(struct class_attribute *class_attr)
1340 {
1341         return class_create_file(&net_class, class_attr);
1342 }
1343 EXPORT_SYMBOL(netdev_class_create_file);
1344
1345 void netdev_class_remove_file(struct class_attribute *class_attr)
1346 {
1347         class_remove_file(&net_class, class_attr);
1348 }
1349 EXPORT_SYMBOL(netdev_class_remove_file);
1350
1351 int netdev_kobject_init(void)
1352 {
1353         kobj_ns_type_register(&net_ns_type_operations);
1354         register_pernet_subsys(&kobj_net_ops);
1355         return class_register(&net_class);
1356 }