net: Fix rxq ref counting
[pandora-kernel.git] / net / core / net-sysfs.c
1 /*
2  * net-sysfs.c - network device class and attributes
3  *
4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5  *
6  *      This program is free software; you can redistribute it and/or
7  *      modify it under the terms of the GNU General Public License
8  *      as published by the Free Software Foundation; either version
9  *      2 of the License, or (at your option) any later version.
10  */
11
12 #include <linux/capability.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/if_arp.h>
16 #include <linux/slab.h>
17 #include <linux/nsproxy.h>
18 #include <net/sock.h>
19 #include <net/net_namespace.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/wireless.h>
22 #include <linux/vmalloc.h>
23 #include <net/wext.h>
24
25 #include "net-sysfs.h"
26
27 #ifdef CONFIG_SYSFS
28 static const char fmt_hex[] = "%#x\n";
29 static const char fmt_long_hex[] = "%#lx\n";
30 static const char fmt_dec[] = "%d\n";
31 static const char fmt_ulong[] = "%lu\n";
32 static const char fmt_u64[] = "%llu\n";
33
34 static inline int dev_isalive(const struct net_device *dev)
35 {
36         return dev->reg_state <= NETREG_REGISTERED;
37 }
38
39 /* use same locking rules as GIF* ioctl's */
40 static ssize_t netdev_show(const struct device *dev,
41                            struct device_attribute *attr, char *buf,
42                            ssize_t (*format)(const struct net_device *, char *))
43 {
44         struct net_device *net = to_net_dev(dev);
45         ssize_t ret = -EINVAL;
46
47         read_lock(&dev_base_lock);
48         if (dev_isalive(net))
49                 ret = (*format)(net, buf);
50         read_unlock(&dev_base_lock);
51
52         return ret;
53 }
54
55 /* generate a show function for simple field */
56 #define NETDEVICE_SHOW(field, format_string)                            \
57 static ssize_t format_##field(const struct net_device *net, char *buf)  \
58 {                                                                       \
59         return sprintf(buf, format_string, net->field);                 \
60 }                                                                       \
61 static ssize_t show_##field(struct device *dev,                         \
62                             struct device_attribute *attr, char *buf)   \
63 {                                                                       \
64         return netdev_show(dev, attr, buf, format_##field);             \
65 }
66
67
68 /* use same locking and permission rules as SIF* ioctl's */
69 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
70                             const char *buf, size_t len,
71                             int (*set)(struct net_device *, unsigned long))
72 {
73         struct net_device *net = to_net_dev(dev);
74         char *endp;
75         unsigned long new;
76         int ret = -EINVAL;
77
78         if (!capable(CAP_NET_ADMIN))
79                 return -EPERM;
80
81         new = simple_strtoul(buf, &endp, 0);
82         if (endp == buf)
83                 goto err;
84
85         if (!rtnl_trylock())
86                 return restart_syscall();
87
88         if (dev_isalive(net)) {
89                 if ((ret = (*set)(net, new)) == 0)
90                         ret = len;
91         }
92         rtnl_unlock();
93  err:
94         return ret;
95 }
96
97 NETDEVICE_SHOW(dev_id, fmt_hex);
98 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
99 NETDEVICE_SHOW(addr_len, fmt_dec);
100 NETDEVICE_SHOW(iflink, fmt_dec);
101 NETDEVICE_SHOW(ifindex, fmt_dec);
102 NETDEVICE_SHOW(features, fmt_long_hex);
103 NETDEVICE_SHOW(type, fmt_dec);
104 NETDEVICE_SHOW(link_mode, fmt_dec);
105
106 /* use same locking rules as GIFHWADDR ioctl's */
107 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
108                             char *buf)
109 {
110         struct net_device *net = to_net_dev(dev);
111         ssize_t ret = -EINVAL;
112
113         read_lock(&dev_base_lock);
114         if (dev_isalive(net))
115                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
116         read_unlock(&dev_base_lock);
117         return ret;
118 }
119
120 static ssize_t show_broadcast(struct device *dev,
121                             struct device_attribute *attr, char *buf)
122 {
123         struct net_device *net = to_net_dev(dev);
124         if (dev_isalive(net))
125                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
126         return -EINVAL;
127 }
128
129 static ssize_t show_carrier(struct device *dev,
130                             struct device_attribute *attr, char *buf)
131 {
132         struct net_device *netdev = to_net_dev(dev);
133         if (netif_running(netdev)) {
134                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
135         }
136         return -EINVAL;
137 }
138
139 static ssize_t show_speed(struct device *dev,
140                           struct device_attribute *attr, char *buf)
141 {
142         struct net_device *netdev = to_net_dev(dev);
143         int ret = -EINVAL;
144
145         if (!rtnl_trylock())
146                 return restart_syscall();
147
148         if (netif_running(netdev) &&
149             netdev->ethtool_ops &&
150             netdev->ethtool_ops->get_settings) {
151                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
152
153                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
154                         ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
155         }
156         rtnl_unlock();
157         return ret;
158 }
159
160 static ssize_t show_duplex(struct device *dev,
161                            struct device_attribute *attr, char *buf)
162 {
163         struct net_device *netdev = to_net_dev(dev);
164         int ret = -EINVAL;
165
166         if (!rtnl_trylock())
167                 return restart_syscall();
168
169         if (netif_running(netdev) &&
170             netdev->ethtool_ops &&
171             netdev->ethtool_ops->get_settings) {
172                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
173
174                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
175                         ret = sprintf(buf, "%s\n", cmd.duplex ? "full" : "half");
176         }
177         rtnl_unlock();
178         return ret;
179 }
180
181 static ssize_t show_dormant(struct device *dev,
182                             struct device_attribute *attr, char *buf)
183 {
184         struct net_device *netdev = to_net_dev(dev);
185
186         if (netif_running(netdev))
187                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
188
189         return -EINVAL;
190 }
191
192 static const char *const operstates[] = {
193         "unknown",
194         "notpresent", /* currently unused */
195         "down",
196         "lowerlayerdown",
197         "testing", /* currently unused */
198         "dormant",
199         "up"
200 };
201
202 static ssize_t show_operstate(struct device *dev,
203                               struct device_attribute *attr, char *buf)
204 {
205         const struct net_device *netdev = to_net_dev(dev);
206         unsigned char operstate;
207
208         read_lock(&dev_base_lock);
209         operstate = netdev->operstate;
210         if (!netif_running(netdev))
211                 operstate = IF_OPER_DOWN;
212         read_unlock(&dev_base_lock);
213
214         if (operstate >= ARRAY_SIZE(operstates))
215                 return -EINVAL; /* should not happen */
216
217         return sprintf(buf, "%s\n", operstates[operstate]);
218 }
219
220 /* read-write attributes */
221 NETDEVICE_SHOW(mtu, fmt_dec);
222
223 static int change_mtu(struct net_device *net, unsigned long new_mtu)
224 {
225         return dev_set_mtu(net, (int) new_mtu);
226 }
227
228 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
229                          const char *buf, size_t len)
230 {
231         return netdev_store(dev, attr, buf, len, change_mtu);
232 }
233
234 NETDEVICE_SHOW(flags, fmt_hex);
235
236 static int change_flags(struct net_device *net, unsigned long new_flags)
237 {
238         return dev_change_flags(net, (unsigned) new_flags);
239 }
240
241 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
242                            const char *buf, size_t len)
243 {
244         return netdev_store(dev, attr, buf, len, change_flags);
245 }
246
247 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
248
249 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
250 {
251         net->tx_queue_len = new_len;
252         return 0;
253 }
254
255 static ssize_t store_tx_queue_len(struct device *dev,
256                                   struct device_attribute *attr,
257                                   const char *buf, size_t len)
258 {
259         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
260 }
261
262 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
263                              const char *buf, size_t len)
264 {
265         struct net_device *netdev = to_net_dev(dev);
266         size_t count = len;
267         ssize_t ret;
268
269         if (!capable(CAP_NET_ADMIN))
270                 return -EPERM;
271
272         /* ignore trailing newline */
273         if (len >  0 && buf[len - 1] == '\n')
274                 --count;
275
276         if (!rtnl_trylock())
277                 return restart_syscall();
278         ret = dev_set_alias(netdev, buf, count);
279         rtnl_unlock();
280
281         return ret < 0 ? ret : len;
282 }
283
284 static ssize_t show_ifalias(struct device *dev,
285                             struct device_attribute *attr, char *buf)
286 {
287         const struct net_device *netdev = to_net_dev(dev);
288         ssize_t ret = 0;
289
290         if (!rtnl_trylock())
291                 return restart_syscall();
292         if (netdev->ifalias)
293                 ret = sprintf(buf, "%s\n", netdev->ifalias);
294         rtnl_unlock();
295         return ret;
296 }
297
298 static struct device_attribute net_class_attributes[] = {
299         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
300         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
301         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
302         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
303         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
304         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
305         __ATTR(features, S_IRUGO, show_features, NULL),
306         __ATTR(type, S_IRUGO, show_type, NULL),
307         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
308         __ATTR(address, S_IRUGO, show_address, NULL),
309         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
310         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
311         __ATTR(speed, S_IRUGO, show_speed, NULL),
312         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
313         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
314         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
315         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
316         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
317         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
318                store_tx_queue_len),
319         {}
320 };
321
322 /* Show a given an attribute in the statistics group */
323 static ssize_t netstat_show(const struct device *d,
324                             struct device_attribute *attr, char *buf,
325                             unsigned long offset)
326 {
327         struct net_device *dev = to_net_dev(d);
328         ssize_t ret = -EINVAL;
329
330         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
331                         offset % sizeof(u64) != 0);
332
333         read_lock(&dev_base_lock);
334         if (dev_isalive(dev)) {
335                 struct rtnl_link_stats64 temp;
336                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
337
338                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
339         }
340         read_unlock(&dev_base_lock);
341         return ret;
342 }
343
344 /* generate a read-only statistics attribute */
345 #define NETSTAT_ENTRY(name)                                             \
346 static ssize_t show_##name(struct device *d,                            \
347                            struct device_attribute *attr, char *buf)    \
348 {                                                                       \
349         return netstat_show(d, attr, buf,                               \
350                             offsetof(struct rtnl_link_stats64, name));  \
351 }                                                                       \
352 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
353
354 NETSTAT_ENTRY(rx_packets);
355 NETSTAT_ENTRY(tx_packets);
356 NETSTAT_ENTRY(rx_bytes);
357 NETSTAT_ENTRY(tx_bytes);
358 NETSTAT_ENTRY(rx_errors);
359 NETSTAT_ENTRY(tx_errors);
360 NETSTAT_ENTRY(rx_dropped);
361 NETSTAT_ENTRY(tx_dropped);
362 NETSTAT_ENTRY(multicast);
363 NETSTAT_ENTRY(collisions);
364 NETSTAT_ENTRY(rx_length_errors);
365 NETSTAT_ENTRY(rx_over_errors);
366 NETSTAT_ENTRY(rx_crc_errors);
367 NETSTAT_ENTRY(rx_frame_errors);
368 NETSTAT_ENTRY(rx_fifo_errors);
369 NETSTAT_ENTRY(rx_missed_errors);
370 NETSTAT_ENTRY(tx_aborted_errors);
371 NETSTAT_ENTRY(tx_carrier_errors);
372 NETSTAT_ENTRY(tx_fifo_errors);
373 NETSTAT_ENTRY(tx_heartbeat_errors);
374 NETSTAT_ENTRY(tx_window_errors);
375 NETSTAT_ENTRY(rx_compressed);
376 NETSTAT_ENTRY(tx_compressed);
377
378 static struct attribute *netstat_attrs[] = {
379         &dev_attr_rx_packets.attr,
380         &dev_attr_tx_packets.attr,
381         &dev_attr_rx_bytes.attr,
382         &dev_attr_tx_bytes.attr,
383         &dev_attr_rx_errors.attr,
384         &dev_attr_tx_errors.attr,
385         &dev_attr_rx_dropped.attr,
386         &dev_attr_tx_dropped.attr,
387         &dev_attr_multicast.attr,
388         &dev_attr_collisions.attr,
389         &dev_attr_rx_length_errors.attr,
390         &dev_attr_rx_over_errors.attr,
391         &dev_attr_rx_crc_errors.attr,
392         &dev_attr_rx_frame_errors.attr,
393         &dev_attr_rx_fifo_errors.attr,
394         &dev_attr_rx_missed_errors.attr,
395         &dev_attr_tx_aborted_errors.attr,
396         &dev_attr_tx_carrier_errors.attr,
397         &dev_attr_tx_fifo_errors.attr,
398         &dev_attr_tx_heartbeat_errors.attr,
399         &dev_attr_tx_window_errors.attr,
400         &dev_attr_rx_compressed.attr,
401         &dev_attr_tx_compressed.attr,
402         NULL
403 };
404
405
406 static struct attribute_group netstat_group = {
407         .name  = "statistics",
408         .attrs  = netstat_attrs,
409 };
410
411 #ifdef CONFIG_WIRELESS_EXT_SYSFS
412 /* helper function that does all the locking etc for wireless stats */
413 static ssize_t wireless_show(struct device *d, char *buf,
414                              ssize_t (*format)(const struct iw_statistics *,
415                                                char *))
416 {
417         struct net_device *dev = to_net_dev(d);
418         const struct iw_statistics *iw;
419         ssize_t ret = -EINVAL;
420
421         if (!rtnl_trylock())
422                 return restart_syscall();
423         if (dev_isalive(dev)) {
424                 iw = get_wireless_stats(dev);
425                 if (iw)
426                         ret = (*format)(iw, buf);
427         }
428         rtnl_unlock();
429
430         return ret;
431 }
432
433 /* show function template for wireless fields */
434 #define WIRELESS_SHOW(name, field, format_string)                       \
435 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
436 {                                                                       \
437         return sprintf(buf, format_string, iw->field);                  \
438 }                                                                       \
439 static ssize_t show_iw_##name(struct device *d,                         \
440                               struct device_attribute *attr, char *buf) \
441 {                                                                       \
442         return wireless_show(d, buf, format_iw_##name);                 \
443 }                                                                       \
444 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
445
446 WIRELESS_SHOW(status, status, fmt_hex);
447 WIRELESS_SHOW(link, qual.qual, fmt_dec);
448 WIRELESS_SHOW(level, qual.level, fmt_dec);
449 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
450 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
451 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
452 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
453 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
454 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
455 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
456
457 static struct attribute *wireless_attrs[] = {
458         &dev_attr_status.attr,
459         &dev_attr_link.attr,
460         &dev_attr_level.attr,
461         &dev_attr_noise.attr,
462         &dev_attr_nwid.attr,
463         &dev_attr_crypt.attr,
464         &dev_attr_fragment.attr,
465         &dev_attr_retries.attr,
466         &dev_attr_misc.attr,
467         &dev_attr_beacon.attr,
468         NULL
469 };
470
471 static struct attribute_group wireless_group = {
472         .name = "wireless",
473         .attrs = wireless_attrs,
474 };
475 #endif
476 #endif /* CONFIG_SYSFS */
477
478 #ifdef CONFIG_RPS
479 /*
480  * RX queue sysfs structures and functions.
481  */
482 struct rx_queue_attribute {
483         struct attribute attr;
484         ssize_t (*show)(struct netdev_rx_queue *queue,
485             struct rx_queue_attribute *attr, char *buf);
486         ssize_t (*store)(struct netdev_rx_queue *queue,
487             struct rx_queue_attribute *attr, const char *buf, size_t len);
488 };
489 #define to_rx_queue_attr(_attr) container_of(_attr,             \
490     struct rx_queue_attribute, attr)
491
492 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
493
494 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
495                                   char *buf)
496 {
497         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
498         struct netdev_rx_queue *queue = to_rx_queue(kobj);
499
500         if (!attribute->show)
501                 return -EIO;
502
503         return attribute->show(queue, attribute, buf);
504 }
505
506 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
507                                    const char *buf, size_t count)
508 {
509         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
510         struct netdev_rx_queue *queue = to_rx_queue(kobj);
511
512         if (!attribute->store)
513                 return -EIO;
514
515         return attribute->store(queue, attribute, buf, count);
516 }
517
518 static const struct sysfs_ops rx_queue_sysfs_ops = {
519         .show = rx_queue_attr_show,
520         .store = rx_queue_attr_store,
521 };
522
523 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
524                             struct rx_queue_attribute *attribute, char *buf)
525 {
526         struct rps_map *map;
527         cpumask_var_t mask;
528         size_t len = 0;
529         int i;
530
531         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
532                 return -ENOMEM;
533
534         rcu_read_lock();
535         map = rcu_dereference(queue->rps_map);
536         if (map)
537                 for (i = 0; i < map->len; i++)
538                         cpumask_set_cpu(map->cpus[i], mask);
539
540         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
541         if (PAGE_SIZE - len < 3) {
542                 rcu_read_unlock();
543                 free_cpumask_var(mask);
544                 return -EINVAL;
545         }
546         rcu_read_unlock();
547
548         free_cpumask_var(mask);
549         len += sprintf(buf + len, "\n");
550         return len;
551 }
552
553 static void rps_map_release(struct rcu_head *rcu)
554 {
555         struct rps_map *map = container_of(rcu, struct rps_map, rcu);
556
557         kfree(map);
558 }
559
560 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
561                       struct rx_queue_attribute *attribute,
562                       const char *buf, size_t len)
563 {
564         struct rps_map *old_map, *map;
565         cpumask_var_t mask;
566         int err, cpu, i;
567         static DEFINE_SPINLOCK(rps_map_lock);
568
569         if (!capable(CAP_NET_ADMIN))
570                 return -EPERM;
571
572         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
573                 return -ENOMEM;
574
575         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
576         if (err) {
577                 free_cpumask_var(mask);
578                 return err;
579         }
580
581         map = kzalloc(max_t(unsigned,
582             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
583             GFP_KERNEL);
584         if (!map) {
585                 free_cpumask_var(mask);
586                 return -ENOMEM;
587         }
588
589         i = 0;
590         for_each_cpu_and(cpu, mask, cpu_online_mask)
591                 map->cpus[i++] = cpu;
592
593         if (i)
594                 map->len = i;
595         else {
596                 kfree(map);
597                 map = NULL;
598         }
599
600         spin_lock(&rps_map_lock);
601         old_map = queue->rps_map;
602         rcu_assign_pointer(queue->rps_map, map);
603         spin_unlock(&rps_map_lock);
604
605         if (old_map)
606                 call_rcu(&old_map->rcu, rps_map_release);
607
608         free_cpumask_var(mask);
609         return len;
610 }
611
612 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
613                                            struct rx_queue_attribute *attr,
614                                            char *buf)
615 {
616         struct rps_dev_flow_table *flow_table;
617         unsigned int val = 0;
618
619         rcu_read_lock();
620         flow_table = rcu_dereference(queue->rps_flow_table);
621         if (flow_table)
622                 val = flow_table->mask + 1;
623         rcu_read_unlock();
624
625         return sprintf(buf, "%u\n", val);
626 }
627
628 static void rps_dev_flow_table_release_work(struct work_struct *work)
629 {
630         struct rps_dev_flow_table *table = container_of(work,
631             struct rps_dev_flow_table, free_work);
632
633         vfree(table);
634 }
635
636 static void rps_dev_flow_table_release(struct rcu_head *rcu)
637 {
638         struct rps_dev_flow_table *table = container_of(rcu,
639             struct rps_dev_flow_table, rcu);
640
641         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
642         schedule_work(&table->free_work);
643 }
644
645 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
646                                      struct rx_queue_attribute *attr,
647                                      const char *buf, size_t len)
648 {
649         unsigned int count;
650         char *endp;
651         struct rps_dev_flow_table *table, *old_table;
652         static DEFINE_SPINLOCK(rps_dev_flow_lock);
653
654         if (!capable(CAP_NET_ADMIN))
655                 return -EPERM;
656
657         count = simple_strtoul(buf, &endp, 0);
658         if (endp == buf)
659                 return -EINVAL;
660
661         if (count) {
662                 int i;
663
664                 if (count > 1<<30) {
665                         /* Enforce a limit to prevent overflow */
666                         return -EINVAL;
667                 }
668                 count = roundup_pow_of_two(count);
669                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
670                 if (!table)
671                         return -ENOMEM;
672
673                 table->mask = count - 1;
674                 for (i = 0; i < count; i++)
675                         table->flows[i].cpu = RPS_NO_CPU;
676         } else
677                 table = NULL;
678
679         spin_lock(&rps_dev_flow_lock);
680         old_table = queue->rps_flow_table;
681         rcu_assign_pointer(queue->rps_flow_table, table);
682         spin_unlock(&rps_dev_flow_lock);
683
684         if (old_table)
685                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
686
687         return len;
688 }
689
690 static struct rx_queue_attribute rps_cpus_attribute =
691         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
692
693
694 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
695         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
696             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
697
698 static struct attribute *rx_queue_default_attrs[] = {
699         &rps_cpus_attribute.attr,
700         &rps_dev_flow_table_cnt_attribute.attr,
701         NULL
702 };
703
704 static void rx_queue_release(struct kobject *kobj)
705 {
706         struct netdev_rx_queue *queue = to_rx_queue(kobj);
707         struct netdev_rx_queue *first = queue->first;
708
709         if (queue->rps_map)
710                 call_rcu(&queue->rps_map->rcu, rps_map_release);
711
712         if (queue->rps_flow_table)
713                 call_rcu(&queue->rps_flow_table->rcu,
714                     rps_dev_flow_table_release);
715
716         if (atomic_dec_and_test(&first->count))
717                 kfree(first);
718 }
719
720 static struct kobj_type rx_queue_ktype = {
721         .sysfs_ops = &rx_queue_sysfs_ops,
722         .release = rx_queue_release,
723         .default_attrs = rx_queue_default_attrs,
724 };
725
726 static int rx_queue_add_kobject(struct net_device *net, int index)
727 {
728         struct netdev_rx_queue *queue = net->_rx + index;
729         struct netdev_rx_queue *first = queue->first;
730         struct kobject *kobj = &queue->kobj;
731         int error = 0;
732
733         kobj->kset = net->queues_kset;
734         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
735             "rx-%u", index);
736         if (error) {
737                 kobject_put(kobj);
738                 return error;
739         }
740
741         kobject_uevent(kobj, KOBJ_ADD);
742         atomic_inc(&first->count);
743
744         return error;
745 }
746
747 int
748 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
749 {
750         int i;
751         int error = 0;
752
753         for (i = old_num; i < new_num; i++) {
754                 error = rx_queue_add_kobject(net, i);
755                 if (error) {
756                         new_num = old_num;
757                         break;
758                 }
759         }
760
761         while (--i >= new_num)
762                 kobject_put(&net->_rx[i].kobj);
763
764         return error;
765 }
766
767 static int rx_queue_register_kobjects(struct net_device *net)
768 {
769         net->queues_kset = kset_create_and_add("queues",
770             NULL, &net->dev.kobj);
771         if (!net->queues_kset)
772                 return -ENOMEM;
773         return net_rx_queue_update_kobjects(net, 0, net->real_num_rx_queues);
774 }
775
776 static void rx_queue_remove_kobjects(struct net_device *net)
777 {
778         net_rx_queue_update_kobjects(net, net->real_num_rx_queues, 0);
779         kset_unregister(net->queues_kset);
780 }
781 #endif /* CONFIG_RPS */
782
783 static const void *net_current_ns(void)
784 {
785         return current->nsproxy->net_ns;
786 }
787
788 static const void *net_initial_ns(void)
789 {
790         return &init_net;
791 }
792
793 static const void *net_netlink_ns(struct sock *sk)
794 {
795         return sock_net(sk);
796 }
797
798 struct kobj_ns_type_operations net_ns_type_operations = {
799         .type = KOBJ_NS_TYPE_NET,
800         .current_ns = net_current_ns,
801         .netlink_ns = net_netlink_ns,
802         .initial_ns = net_initial_ns,
803 };
804 EXPORT_SYMBOL_GPL(net_ns_type_operations);
805
806 static void net_kobj_ns_exit(struct net *net)
807 {
808         kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
809 }
810
811 static struct pernet_operations kobj_net_ops = {
812         .exit = net_kobj_ns_exit,
813 };
814
815
816 #ifdef CONFIG_HOTPLUG
817 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
818 {
819         struct net_device *dev = to_net_dev(d);
820         int retval;
821
822         /* pass interface to uevent. */
823         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
824         if (retval)
825                 goto exit;
826
827         /* pass ifindex to uevent.
828          * ifindex is useful as it won't change (interface name may change)
829          * and is what RtNetlink uses natively. */
830         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
831
832 exit:
833         return retval;
834 }
835 #endif
836
837 /*
838  *      netdev_release -- destroy and free a dead device.
839  *      Called when last reference to device kobject is gone.
840  */
841 static void netdev_release(struct device *d)
842 {
843         struct net_device *dev = to_net_dev(d);
844
845         BUG_ON(dev->reg_state != NETREG_RELEASED);
846
847         kfree(dev->ifalias);
848         kfree((char *)dev - dev->padded);
849 }
850
851 static const void *net_namespace(struct device *d)
852 {
853         struct net_device *dev;
854         dev = container_of(d, struct net_device, dev);
855         return dev_net(dev);
856 }
857
858 static struct class net_class = {
859         .name = "net",
860         .dev_release = netdev_release,
861 #ifdef CONFIG_SYSFS
862         .dev_attrs = net_class_attributes,
863 #endif /* CONFIG_SYSFS */
864 #ifdef CONFIG_HOTPLUG
865         .dev_uevent = netdev_uevent,
866 #endif
867         .ns_type = &net_ns_type_operations,
868         .namespace = net_namespace,
869 };
870
871 /* Delete sysfs entries but hold kobject reference until after all
872  * netdev references are gone.
873  */
874 void netdev_unregister_kobject(struct net_device * net)
875 {
876         struct device *dev = &(net->dev);
877
878         kobject_get(&dev->kobj);
879
880 #ifdef CONFIG_RPS
881         rx_queue_remove_kobjects(net);
882 #endif
883
884         device_del(dev);
885 }
886
887 /* Create sysfs entries for network device. */
888 int netdev_register_kobject(struct net_device *net)
889 {
890         struct device *dev = &(net->dev);
891         const struct attribute_group **groups = net->sysfs_groups;
892         int error = 0;
893
894         device_initialize(dev);
895         dev->class = &net_class;
896         dev->platform_data = net;
897         dev->groups = groups;
898
899         dev_set_name(dev, "%s", net->name);
900
901 #ifdef CONFIG_SYSFS
902         /* Allow for a device specific group */
903         if (*groups)
904                 groups++;
905
906         *groups++ = &netstat_group;
907 #ifdef CONFIG_WIRELESS_EXT_SYSFS
908         if (net->ieee80211_ptr)
909                 *groups++ = &wireless_group;
910 #ifdef CONFIG_WIRELESS_EXT
911         else if (net->wireless_handlers)
912                 *groups++ = &wireless_group;
913 #endif
914 #endif
915 #endif /* CONFIG_SYSFS */
916
917         error = device_add(dev);
918         if (error)
919                 return error;
920
921 #ifdef CONFIG_RPS
922         error = rx_queue_register_kobjects(net);
923         if (error) {
924                 device_del(dev);
925                 return error;
926         }
927 #endif
928
929         return error;
930 }
931
932 int netdev_class_create_file(struct class_attribute *class_attr)
933 {
934         return class_create_file(&net_class, class_attr);
935 }
936 EXPORT_SYMBOL(netdev_class_create_file);
937
938 void netdev_class_remove_file(struct class_attribute *class_attr)
939 {
940         class_remove_file(&net_class, class_attr);
941 }
942 EXPORT_SYMBOL(netdev_class_remove_file);
943
944 int netdev_kobject_init(void)
945 {
946         kobj_ns_type_register(&net_ns_type_operations);
947         register_pernet_subsys(&kobj_net_ops);
948         return class_register(&net_class);
949 }