Delay struct net freeing while there's a sysfs instance refering to it
[linux-2.6.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_udec[] = "%u\n";
32 static const char fmt_ulong[] = "%lu\n";
33 static const char fmt_u64[] = "%llu\n";
34
35 static inline int dev_isalive(const struct net_device *dev)
36 {
37         return dev->reg_state <= NETREG_REGISTERED;
38 }
39
40 /* use same locking rules as GIF* ioctl's */
41 static ssize_t netdev_show(const struct device *dev,
42                            struct device_attribute *attr, char *buf,
43                            ssize_t (*format)(const struct net_device *, char *))
44 {
45         struct net_device *net = to_net_dev(dev);
46         ssize_t ret = -EINVAL;
47
48         read_lock(&dev_base_lock);
49         if (dev_isalive(net))
50                 ret = (*format)(net, buf);
51         read_unlock(&dev_base_lock);
52
53         return ret;
54 }
55
56 /* generate a show function for simple field */
57 #define NETDEVICE_SHOW(field, format_string)                            \
58 static ssize_t format_##field(const struct net_device *net, char *buf)  \
59 {                                                                       \
60         return sprintf(buf, format_string, net->field);                 \
61 }                                                                       \
62 static ssize_t show_##field(struct device *dev,                         \
63                             struct device_attribute *attr, char *buf)   \
64 {                                                                       \
65         return netdev_show(dev, attr, buf, format_##field);             \
66 }
67
68
69 /* use same locking and permission rules as SIF* ioctl's */
70 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
71                             const char *buf, size_t len,
72                             int (*set)(struct net_device *, unsigned long))
73 {
74         struct net_device *net = to_net_dev(dev);
75         char *endp;
76         unsigned long new;
77         int ret = -EINVAL;
78
79         if (!capable(CAP_NET_ADMIN))
80                 return -EPERM;
81
82         new = simple_strtoul(buf, &endp, 0);
83         if (endp == buf)
84                 goto err;
85
86         if (!rtnl_trylock())
87                 return restart_syscall();
88
89         if (dev_isalive(net)) {
90                 if ((ret = (*set)(net, new)) == 0)
91                         ret = len;
92         }
93         rtnl_unlock();
94  err:
95         return ret;
96 }
97
98 NETDEVICE_SHOW(dev_id, fmt_hex);
99 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
100 NETDEVICE_SHOW(addr_len, fmt_dec);
101 NETDEVICE_SHOW(iflink, fmt_dec);
102 NETDEVICE_SHOW(ifindex, fmt_dec);
103 NETDEVICE_SHOW(features, fmt_hex);
104 NETDEVICE_SHOW(type, fmt_dec);
105 NETDEVICE_SHOW(link_mode, fmt_dec);
106
107 /* use same locking rules as GIFHWADDR ioctl's */
108 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
109                             char *buf)
110 {
111         struct net_device *net = to_net_dev(dev);
112         ssize_t ret = -EINVAL;
113
114         read_lock(&dev_base_lock);
115         if (dev_isalive(net))
116                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
117         read_unlock(&dev_base_lock);
118         return ret;
119 }
120
121 static ssize_t show_broadcast(struct device *dev,
122                             struct device_attribute *attr, char *buf)
123 {
124         struct net_device *net = to_net_dev(dev);
125         if (dev_isalive(net))
126                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
127         return -EINVAL;
128 }
129
130 static ssize_t show_carrier(struct device *dev,
131                             struct device_attribute *attr, char *buf)
132 {
133         struct net_device *netdev = to_net_dev(dev);
134         if (netif_running(netdev)) {
135                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
136         }
137         return -EINVAL;
138 }
139
140 static ssize_t show_speed(struct device *dev,
141                           struct device_attribute *attr, char *buf)
142 {
143         struct net_device *netdev = to_net_dev(dev);
144         int ret = -EINVAL;
145
146         if (!rtnl_trylock())
147                 return restart_syscall();
148
149         if (netif_running(netdev)) {
150                 struct ethtool_cmd cmd;
151                 if (!dev_ethtool_get_settings(netdev, &cmd))
152                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
153         }
154         rtnl_unlock();
155         return ret;
156 }
157
158 static ssize_t show_duplex(struct device *dev,
159                            struct device_attribute *attr, char *buf)
160 {
161         struct net_device *netdev = to_net_dev(dev);
162         int ret = -EINVAL;
163
164         if (!rtnl_trylock())
165                 return restart_syscall();
166
167         if (netif_running(netdev)) {
168                 struct ethtool_cmd cmd;
169                 if (!dev_ethtool_get_settings(netdev, &cmd))
170                         ret = sprintf(buf, "%s\n",
171                                       cmd.duplex ? "full" : "half");
172         }
173         rtnl_unlock();
174         return ret;
175 }
176
177 static ssize_t show_dormant(struct device *dev,
178                             struct device_attribute *attr, char *buf)
179 {
180         struct net_device *netdev = to_net_dev(dev);
181
182         if (netif_running(netdev))
183                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
184
185         return -EINVAL;
186 }
187
188 static const char *const operstates[] = {
189         "unknown",
190         "notpresent", /* currently unused */
191         "down",
192         "lowerlayerdown",
193         "testing", /* currently unused */
194         "dormant",
195         "up"
196 };
197
198 static ssize_t show_operstate(struct device *dev,
199                               struct device_attribute *attr, char *buf)
200 {
201         const struct net_device *netdev = to_net_dev(dev);
202         unsigned char operstate;
203
204         read_lock(&dev_base_lock);
205         operstate = netdev->operstate;
206         if (!netif_running(netdev))
207                 operstate = IF_OPER_DOWN;
208         read_unlock(&dev_base_lock);
209
210         if (operstate >= ARRAY_SIZE(operstates))
211                 return -EINVAL; /* should not happen */
212
213         return sprintf(buf, "%s\n", operstates[operstate]);
214 }
215
216 /* read-write attributes */
217 NETDEVICE_SHOW(mtu, fmt_dec);
218
219 static int change_mtu(struct net_device *net, unsigned long new_mtu)
220 {
221         return dev_set_mtu(net, (int) new_mtu);
222 }
223
224 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
225                          const char *buf, size_t len)
226 {
227         return netdev_store(dev, attr, buf, len, change_mtu);
228 }
229
230 NETDEVICE_SHOW(flags, fmt_hex);
231
232 static int change_flags(struct net_device *net, unsigned long new_flags)
233 {
234         return dev_change_flags(net, (unsigned) new_flags);
235 }
236
237 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
238                            const char *buf, size_t len)
239 {
240         return netdev_store(dev, attr, buf, len, change_flags);
241 }
242
243 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
244
245 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
246 {
247         net->tx_queue_len = new_len;
248         return 0;
249 }
250
251 static ssize_t store_tx_queue_len(struct device *dev,
252                                   struct device_attribute *attr,
253                                   const char *buf, size_t len)
254 {
255         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
256 }
257
258 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
259                              const char *buf, size_t len)
260 {
261         struct net_device *netdev = to_net_dev(dev);
262         size_t count = len;
263         ssize_t ret;
264
265         if (!capable(CAP_NET_ADMIN))
266                 return -EPERM;
267
268         /* ignore trailing newline */
269         if (len >  0 && buf[len - 1] == '\n')
270                 --count;
271
272         if (!rtnl_trylock())
273                 return restart_syscall();
274         ret = dev_set_alias(netdev, buf, count);
275         rtnl_unlock();
276
277         return ret < 0 ? ret : len;
278 }
279
280 static ssize_t show_ifalias(struct device *dev,
281                             struct device_attribute *attr, char *buf)
282 {
283         const struct net_device *netdev = to_net_dev(dev);
284         ssize_t ret = 0;
285
286         if (!rtnl_trylock())
287                 return restart_syscall();
288         if (netdev->ifalias)
289                 ret = sprintf(buf, "%s\n", netdev->ifalias);
290         rtnl_unlock();
291         return ret;
292 }
293
294 NETDEVICE_SHOW(group, fmt_dec);
295
296 static int change_group(struct net_device *net, unsigned long new_group)
297 {
298         dev_set_group(net, (int) new_group);
299         return 0;
300 }
301
302 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
303                          const char *buf, size_t len)
304 {
305         return netdev_store(dev, attr, buf, len, change_group);
306 }
307
308 static struct device_attribute net_class_attributes[] = {
309         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
310         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
311         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
312         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
313         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
314         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
315         __ATTR(features, S_IRUGO, show_features, NULL),
316         __ATTR(type, S_IRUGO, show_type, NULL),
317         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
318         __ATTR(address, S_IRUGO, show_address, NULL),
319         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
320         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
321         __ATTR(speed, S_IRUGO, show_speed, NULL),
322         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
323         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
324         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
325         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
326         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
327         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
328                store_tx_queue_len),
329         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
330         {}
331 };
332
333 /* Show a given an attribute in the statistics group */
334 static ssize_t netstat_show(const struct device *d,
335                             struct device_attribute *attr, char *buf,
336                             unsigned long offset)
337 {
338         struct net_device *dev = to_net_dev(d);
339         ssize_t ret = -EINVAL;
340
341         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
342                         offset % sizeof(u64) != 0);
343
344         read_lock(&dev_base_lock);
345         if (dev_isalive(dev)) {
346                 struct rtnl_link_stats64 temp;
347                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
348
349                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
350         }
351         read_unlock(&dev_base_lock);
352         return ret;
353 }
354
355 /* generate a read-only statistics attribute */
356 #define NETSTAT_ENTRY(name)                                             \
357 static ssize_t show_##name(struct device *d,                            \
358                            struct device_attribute *attr, char *buf)    \
359 {                                                                       \
360         return netstat_show(d, attr, buf,                               \
361                             offsetof(struct rtnl_link_stats64, name));  \
362 }                                                                       \
363 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
364
365 NETSTAT_ENTRY(rx_packets);
366 NETSTAT_ENTRY(tx_packets);
367 NETSTAT_ENTRY(rx_bytes);
368 NETSTAT_ENTRY(tx_bytes);
369 NETSTAT_ENTRY(rx_errors);
370 NETSTAT_ENTRY(tx_errors);
371 NETSTAT_ENTRY(rx_dropped);
372 NETSTAT_ENTRY(tx_dropped);
373 NETSTAT_ENTRY(multicast);
374 NETSTAT_ENTRY(collisions);
375 NETSTAT_ENTRY(rx_length_errors);
376 NETSTAT_ENTRY(rx_over_errors);
377 NETSTAT_ENTRY(rx_crc_errors);
378 NETSTAT_ENTRY(rx_frame_errors);
379 NETSTAT_ENTRY(rx_fifo_errors);
380 NETSTAT_ENTRY(rx_missed_errors);
381 NETSTAT_ENTRY(tx_aborted_errors);
382 NETSTAT_ENTRY(tx_carrier_errors);
383 NETSTAT_ENTRY(tx_fifo_errors);
384 NETSTAT_ENTRY(tx_heartbeat_errors);
385 NETSTAT_ENTRY(tx_window_errors);
386 NETSTAT_ENTRY(rx_compressed);
387 NETSTAT_ENTRY(tx_compressed);
388
389 static struct attribute *netstat_attrs[] = {
390         &dev_attr_rx_packets.attr,
391         &dev_attr_tx_packets.attr,
392         &dev_attr_rx_bytes.attr,
393         &dev_attr_tx_bytes.attr,
394         &dev_attr_rx_errors.attr,
395         &dev_attr_tx_errors.attr,
396         &dev_attr_rx_dropped.attr,
397         &dev_attr_tx_dropped.attr,
398         &dev_attr_multicast.attr,
399         &dev_attr_collisions.attr,
400         &dev_attr_rx_length_errors.attr,
401         &dev_attr_rx_over_errors.attr,
402         &dev_attr_rx_crc_errors.attr,
403         &dev_attr_rx_frame_errors.attr,
404         &dev_attr_rx_fifo_errors.attr,
405         &dev_attr_rx_missed_errors.attr,
406         &dev_attr_tx_aborted_errors.attr,
407         &dev_attr_tx_carrier_errors.attr,
408         &dev_attr_tx_fifo_errors.attr,
409         &dev_attr_tx_heartbeat_errors.attr,
410         &dev_attr_tx_window_errors.attr,
411         &dev_attr_rx_compressed.attr,
412         &dev_attr_tx_compressed.attr,
413         NULL
414 };
415
416
417 static struct attribute_group netstat_group = {
418         .name  = "statistics",
419         .attrs  = netstat_attrs,
420 };
421
422 #ifdef CONFIG_WIRELESS_EXT_SYSFS
423 /* helper function that does all the locking etc for wireless stats */
424 static ssize_t wireless_show(struct device *d, char *buf,
425                              ssize_t (*format)(const struct iw_statistics *,
426                                                char *))
427 {
428         struct net_device *dev = to_net_dev(d);
429         const struct iw_statistics *iw;
430         ssize_t ret = -EINVAL;
431
432         if (!rtnl_trylock())
433                 return restart_syscall();
434         if (dev_isalive(dev)) {
435                 iw = get_wireless_stats(dev);
436                 if (iw)
437                         ret = (*format)(iw, buf);
438         }
439         rtnl_unlock();
440
441         return ret;
442 }
443
444 /* show function template for wireless fields */
445 #define WIRELESS_SHOW(name, field, format_string)                       \
446 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
447 {                                                                       \
448         return sprintf(buf, format_string, iw->field);                  \
449 }                                                                       \
450 static ssize_t show_iw_##name(struct device *d,                         \
451                               struct device_attribute *attr, char *buf) \
452 {                                                                       \
453         return wireless_show(d, buf, format_iw_##name);                 \
454 }                                                                       \
455 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
456
457 WIRELESS_SHOW(status, status, fmt_hex);
458 WIRELESS_SHOW(link, qual.qual, fmt_dec);
459 WIRELESS_SHOW(level, qual.level, fmt_dec);
460 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
461 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
462 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
463 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
464 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
465 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
466 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
467
468 static struct attribute *wireless_attrs[] = {
469         &dev_attr_status.attr,
470         &dev_attr_link.attr,
471         &dev_attr_level.attr,
472         &dev_attr_noise.attr,
473         &dev_attr_nwid.attr,
474         &dev_attr_crypt.attr,
475         &dev_attr_fragment.attr,
476         &dev_attr_retries.attr,
477         &dev_attr_misc.attr,
478         &dev_attr_beacon.attr,
479         NULL
480 };
481
482 static struct attribute_group wireless_group = {
483         .name = "wireless",
484         .attrs = wireless_attrs,
485 };
486 #endif
487 #endif /* CONFIG_SYSFS */
488
489 #ifdef CONFIG_RPS
490 /*
491  * RX queue sysfs structures and functions.
492  */
493 struct rx_queue_attribute {
494         struct attribute attr;
495         ssize_t (*show)(struct netdev_rx_queue *queue,
496             struct rx_queue_attribute *attr, char *buf);
497         ssize_t (*store)(struct netdev_rx_queue *queue,
498             struct rx_queue_attribute *attr, const char *buf, size_t len);
499 };
500 #define to_rx_queue_attr(_attr) container_of(_attr,             \
501     struct rx_queue_attribute, attr)
502
503 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
504
505 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
506                                   char *buf)
507 {
508         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
509         struct netdev_rx_queue *queue = to_rx_queue(kobj);
510
511         if (!attribute->show)
512                 return -EIO;
513
514         return attribute->show(queue, attribute, buf);
515 }
516
517 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
518                                    const char *buf, size_t count)
519 {
520         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
521         struct netdev_rx_queue *queue = to_rx_queue(kobj);
522
523         if (!attribute->store)
524                 return -EIO;
525
526         return attribute->store(queue, attribute, buf, count);
527 }
528
529 static const struct sysfs_ops rx_queue_sysfs_ops = {
530         .show = rx_queue_attr_show,
531         .store = rx_queue_attr_store,
532 };
533
534 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
535                             struct rx_queue_attribute *attribute, char *buf)
536 {
537         struct rps_map *map;
538         cpumask_var_t mask;
539         size_t len = 0;
540         int i;
541
542         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
543                 return -ENOMEM;
544
545         rcu_read_lock();
546         map = rcu_dereference(queue->rps_map);
547         if (map)
548                 for (i = 0; i < map->len; i++)
549                         cpumask_set_cpu(map->cpus[i], mask);
550
551         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
552         if (PAGE_SIZE - len < 3) {
553                 rcu_read_unlock();
554                 free_cpumask_var(mask);
555                 return -EINVAL;
556         }
557         rcu_read_unlock();
558
559         free_cpumask_var(mask);
560         len += sprintf(buf + len, "\n");
561         return len;
562 }
563
564 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
565                       struct rx_queue_attribute *attribute,
566                       const char *buf, size_t len)
567 {
568         struct rps_map *old_map, *map;
569         cpumask_var_t mask;
570         int err, cpu, i;
571         static DEFINE_SPINLOCK(rps_map_lock);
572
573         if (!capable(CAP_NET_ADMIN))
574                 return -EPERM;
575
576         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
577                 return -ENOMEM;
578
579         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
580         if (err) {
581                 free_cpumask_var(mask);
582                 return err;
583         }
584
585         map = kzalloc(max_t(unsigned,
586             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
587             GFP_KERNEL);
588         if (!map) {
589                 free_cpumask_var(mask);
590                 return -ENOMEM;
591         }
592
593         i = 0;
594         for_each_cpu_and(cpu, mask, cpu_online_mask)
595                 map->cpus[i++] = cpu;
596
597         if (i)
598                 map->len = i;
599         else {
600                 kfree(map);
601                 map = NULL;
602         }
603
604         spin_lock(&rps_map_lock);
605         old_map = rcu_dereference_protected(queue->rps_map,
606                                             lockdep_is_held(&rps_map_lock));
607         rcu_assign_pointer(queue->rps_map, map);
608         spin_unlock(&rps_map_lock);
609
610         if (old_map)
611                 kfree_rcu(old_map, rcu);
612
613         free_cpumask_var(mask);
614         return len;
615 }
616
617 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
618                                            struct rx_queue_attribute *attr,
619                                            char *buf)
620 {
621         struct rps_dev_flow_table *flow_table;
622         unsigned int val = 0;
623
624         rcu_read_lock();
625         flow_table = rcu_dereference(queue->rps_flow_table);
626         if (flow_table)
627                 val = flow_table->mask + 1;
628         rcu_read_unlock();
629
630         return sprintf(buf, "%u\n", val);
631 }
632
633 static void rps_dev_flow_table_release_work(struct work_struct *work)
634 {
635         struct rps_dev_flow_table *table = container_of(work,
636             struct rps_dev_flow_table, free_work);
637
638         vfree(table);
639 }
640
641 static void rps_dev_flow_table_release(struct rcu_head *rcu)
642 {
643         struct rps_dev_flow_table *table = container_of(rcu,
644             struct rps_dev_flow_table, rcu);
645
646         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
647         schedule_work(&table->free_work);
648 }
649
650 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
651                                      struct rx_queue_attribute *attr,
652                                      const char *buf, size_t len)
653 {
654         unsigned int count;
655         char *endp;
656         struct rps_dev_flow_table *table, *old_table;
657         static DEFINE_SPINLOCK(rps_dev_flow_lock);
658
659         if (!capable(CAP_NET_ADMIN))
660                 return -EPERM;
661
662         count = simple_strtoul(buf, &endp, 0);
663         if (endp == buf)
664                 return -EINVAL;
665
666         if (count) {
667                 int i;
668
669                 if (count > 1<<30) {
670                         /* Enforce a limit to prevent overflow */
671                         return -EINVAL;
672                 }
673                 count = roundup_pow_of_two(count);
674                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
675                 if (!table)
676                         return -ENOMEM;
677
678                 table->mask = count - 1;
679                 for (i = 0; i < count; i++)
680                         table->flows[i].cpu = RPS_NO_CPU;
681         } else
682                 table = NULL;
683
684         spin_lock(&rps_dev_flow_lock);
685         old_table = rcu_dereference_protected(queue->rps_flow_table,
686                                               lockdep_is_held(&rps_dev_flow_lock));
687         rcu_assign_pointer(queue->rps_flow_table, table);
688         spin_unlock(&rps_dev_flow_lock);
689
690         if (old_table)
691                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
692
693         return len;
694 }
695
696 static struct rx_queue_attribute rps_cpus_attribute =
697         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
698
699
700 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
701         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
702             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
703
704 static struct attribute *rx_queue_default_attrs[] = {
705         &rps_cpus_attribute.attr,
706         &rps_dev_flow_table_cnt_attribute.attr,
707         NULL
708 };
709
710 static void rx_queue_release(struct kobject *kobj)
711 {
712         struct netdev_rx_queue *queue = to_rx_queue(kobj);
713         struct rps_map *map;
714         struct rps_dev_flow_table *flow_table;
715
716
717         map = rcu_dereference_raw(queue->rps_map);
718         if (map) {
719                 RCU_INIT_POINTER(queue->rps_map, NULL);
720                 kfree_rcu(map, rcu);
721         }
722
723         flow_table = rcu_dereference_raw(queue->rps_flow_table);
724         if (flow_table) {
725                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
726                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
727         }
728
729         memset(kobj, 0, sizeof(*kobj));
730         dev_put(queue->dev);
731 }
732
733 static struct kobj_type rx_queue_ktype = {
734         .sysfs_ops = &rx_queue_sysfs_ops,
735         .release = rx_queue_release,
736         .default_attrs = rx_queue_default_attrs,
737 };
738
739 static int rx_queue_add_kobject(struct net_device *net, int index)
740 {
741         struct netdev_rx_queue *queue = net->_rx + index;
742         struct kobject *kobj = &queue->kobj;
743         int error = 0;
744
745         kobj->kset = net->queues_kset;
746         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
747             "rx-%u", index);
748         if (error) {
749                 kobject_put(kobj);
750                 return error;
751         }
752
753         kobject_uevent(kobj, KOBJ_ADD);
754         dev_hold(queue->dev);
755
756         return error;
757 }
758 #endif /* CONFIG_RPS */
759
760 int
761 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
762 {
763 #ifdef CONFIG_RPS
764         int i;
765         int error = 0;
766
767         for (i = old_num; i < new_num; i++) {
768                 error = rx_queue_add_kobject(net, i);
769                 if (error) {
770                         new_num = old_num;
771                         break;
772                 }
773         }
774
775         while (--i >= new_num)
776                 kobject_put(&net->_rx[i].kobj);
777
778         return error;
779 #else
780         return 0;
781 #endif
782 }
783
784 #ifdef CONFIG_XPS
785 /*
786  * netdev_queue sysfs structures and functions.
787  */
788 struct netdev_queue_attribute {
789         struct attribute attr;
790         ssize_t (*show)(struct netdev_queue *queue,
791             struct netdev_queue_attribute *attr, char *buf);
792         ssize_t (*store)(struct netdev_queue *queue,
793             struct netdev_queue_attribute *attr, const char *buf, size_t len);
794 };
795 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
796     struct netdev_queue_attribute, attr)
797
798 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
799
800 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
801                                       struct attribute *attr, char *buf)
802 {
803         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
804         struct netdev_queue *queue = to_netdev_queue(kobj);
805
806         if (!attribute->show)
807                 return -EIO;
808
809         return attribute->show(queue, attribute, buf);
810 }
811
812 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
813                                        struct attribute *attr,
814                                        const char *buf, size_t count)
815 {
816         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
817         struct netdev_queue *queue = to_netdev_queue(kobj);
818
819         if (!attribute->store)
820                 return -EIO;
821
822         return attribute->store(queue, attribute, buf, count);
823 }
824
825 static const struct sysfs_ops netdev_queue_sysfs_ops = {
826         .show = netdev_queue_attr_show,
827         .store = netdev_queue_attr_store,
828 };
829
830 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
831 {
832         struct net_device *dev = queue->dev;
833         int i;
834
835         for (i = 0; i < dev->num_tx_queues; i++)
836                 if (queue == &dev->_tx[i])
837                         break;
838
839         BUG_ON(i >= dev->num_tx_queues);
840
841         return i;
842 }
843
844
845 static ssize_t show_xps_map(struct netdev_queue *queue,
846                             struct netdev_queue_attribute *attribute, char *buf)
847 {
848         struct net_device *dev = queue->dev;
849         struct xps_dev_maps *dev_maps;
850         cpumask_var_t mask;
851         unsigned long index;
852         size_t len = 0;
853         int i;
854
855         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
856                 return -ENOMEM;
857
858         index = get_netdev_queue_index(queue);
859
860         rcu_read_lock();
861         dev_maps = rcu_dereference(dev->xps_maps);
862         if (dev_maps) {
863                 for_each_possible_cpu(i) {
864                         struct xps_map *map =
865                             rcu_dereference(dev_maps->cpu_map[i]);
866                         if (map) {
867                                 int j;
868                                 for (j = 0; j < map->len; j++) {
869                                         if (map->queues[j] == index) {
870                                                 cpumask_set_cpu(i, mask);
871                                                 break;
872                                         }
873                                 }
874                         }
875                 }
876         }
877         rcu_read_unlock();
878
879         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
880         if (PAGE_SIZE - len < 3) {
881                 free_cpumask_var(mask);
882                 return -EINVAL;
883         }
884
885         free_cpumask_var(mask);
886         len += sprintf(buf + len, "\n");
887         return len;
888 }
889
890 static DEFINE_MUTEX(xps_map_mutex);
891 #define xmap_dereference(P)             \
892         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
893
894 static ssize_t store_xps_map(struct netdev_queue *queue,
895                       struct netdev_queue_attribute *attribute,
896                       const char *buf, size_t len)
897 {
898         struct net_device *dev = queue->dev;
899         cpumask_var_t mask;
900         int err, i, cpu, pos, map_len, alloc_len, need_set;
901         unsigned long index;
902         struct xps_map *map, *new_map;
903         struct xps_dev_maps *dev_maps, *new_dev_maps;
904         int nonempty = 0;
905         int numa_node = -2;
906
907         if (!capable(CAP_NET_ADMIN))
908                 return -EPERM;
909
910         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
911                 return -ENOMEM;
912
913         index = get_netdev_queue_index(queue);
914
915         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
916         if (err) {
917                 free_cpumask_var(mask);
918                 return err;
919         }
920
921         new_dev_maps = kzalloc(max_t(unsigned,
922             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
923         if (!new_dev_maps) {
924                 free_cpumask_var(mask);
925                 return -ENOMEM;
926         }
927
928         mutex_lock(&xps_map_mutex);
929
930         dev_maps = xmap_dereference(dev->xps_maps);
931
932         for_each_possible_cpu(cpu) {
933                 map = dev_maps ?
934                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
935                 new_map = map;
936                 if (map) {
937                         for (pos = 0; pos < map->len; pos++)
938                                 if (map->queues[pos] == index)
939                                         break;
940                         map_len = map->len;
941                         alloc_len = map->alloc_len;
942                 } else
943                         pos = map_len = alloc_len = 0;
944
945                 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
946 #ifdef CONFIG_NUMA
947                 if (need_set) {
948                         if (numa_node == -2)
949                                 numa_node = cpu_to_node(cpu);
950                         else if (numa_node != cpu_to_node(cpu))
951                                 numa_node = -1;
952                 }
953 #endif
954                 if (need_set && pos >= map_len) {
955                         /* Need to add queue to this CPU's map */
956                         if (map_len >= alloc_len) {
957                                 alloc_len = alloc_len ?
958                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
959                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
960                                                        GFP_KERNEL,
961                                                        cpu_to_node(cpu));
962                                 if (!new_map)
963                                         goto error;
964                                 new_map->alloc_len = alloc_len;
965                                 for (i = 0; i < map_len; i++)
966                                         new_map->queues[i] = map->queues[i];
967                                 new_map->len = map_len;
968                         }
969                         new_map->queues[new_map->len++] = index;
970                 } else if (!need_set && pos < map_len) {
971                         /* Need to remove queue from this CPU's map */
972                         if (map_len > 1)
973                                 new_map->queues[pos] =
974                                     new_map->queues[--new_map->len];
975                         else
976                                 new_map = NULL;
977                 }
978                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
979         }
980
981         /* Cleanup old maps */
982         for_each_possible_cpu(cpu) {
983                 map = dev_maps ?
984                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
985                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
986                         kfree_rcu(map, rcu);
987                 if (new_dev_maps->cpu_map[cpu])
988                         nonempty = 1;
989         }
990
991         if (nonempty)
992                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
993         else {
994                 kfree(new_dev_maps);
995                 rcu_assign_pointer(dev->xps_maps, NULL);
996         }
997
998         if (dev_maps)
999                 kfree_rcu(dev_maps, rcu);
1000
1001         netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1002                                             NUMA_NO_NODE);
1003
1004         mutex_unlock(&xps_map_mutex);
1005
1006         free_cpumask_var(mask);
1007         return len;
1008
1009 error:
1010         mutex_unlock(&xps_map_mutex);
1011
1012         if (new_dev_maps)
1013                 for_each_possible_cpu(i)
1014                         kfree(rcu_dereference_protected(
1015                                 new_dev_maps->cpu_map[i],
1016                                 1));
1017         kfree(new_dev_maps);
1018         free_cpumask_var(mask);
1019         return -ENOMEM;
1020 }
1021
1022 static struct netdev_queue_attribute xps_cpus_attribute =
1023     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1024
1025 static struct attribute *netdev_queue_default_attrs[] = {
1026         &xps_cpus_attribute.attr,
1027         NULL
1028 };
1029
1030 static void netdev_queue_release(struct kobject *kobj)
1031 {
1032         struct netdev_queue *queue = to_netdev_queue(kobj);
1033         struct net_device *dev = queue->dev;
1034         struct xps_dev_maps *dev_maps;
1035         struct xps_map *map;
1036         unsigned long index;
1037         int i, pos, nonempty = 0;
1038
1039         index = get_netdev_queue_index(queue);
1040
1041         mutex_lock(&xps_map_mutex);
1042         dev_maps = xmap_dereference(dev->xps_maps);
1043
1044         if (dev_maps) {
1045                 for_each_possible_cpu(i) {
1046                         map = xmap_dereference(dev_maps->cpu_map[i]);
1047                         if (!map)
1048                                 continue;
1049
1050                         for (pos = 0; pos < map->len; pos++)
1051                                 if (map->queues[pos] == index)
1052                                         break;
1053
1054                         if (pos < map->len) {
1055                                 if (map->len > 1)
1056                                         map->queues[pos] =
1057                                             map->queues[--map->len];
1058                                 else {
1059                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1060                                             NULL);
1061                                         kfree_rcu(map, rcu);
1062                                         map = NULL;
1063                                 }
1064                         }
1065                         if (map)
1066                                 nonempty = 1;
1067                 }
1068
1069                 if (!nonempty) {
1070                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1071                         kfree_rcu(dev_maps, rcu);
1072                 }
1073         }
1074
1075         mutex_unlock(&xps_map_mutex);
1076
1077         memset(kobj, 0, sizeof(*kobj));
1078         dev_put(queue->dev);
1079 }
1080
1081 static struct kobj_type netdev_queue_ktype = {
1082         .sysfs_ops = &netdev_queue_sysfs_ops,
1083         .release = netdev_queue_release,
1084         .default_attrs = netdev_queue_default_attrs,
1085 };
1086
1087 static int netdev_queue_add_kobject(struct net_device *net, int index)
1088 {
1089         struct netdev_queue *queue = net->_tx + index;
1090         struct kobject *kobj = &queue->kobj;
1091         int error = 0;
1092
1093         kobj->kset = net->queues_kset;
1094         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1095             "tx-%u", index);
1096         if (error) {
1097                 kobject_put(kobj);
1098                 return error;
1099         }
1100
1101         kobject_uevent(kobj, KOBJ_ADD);
1102         dev_hold(queue->dev);
1103
1104         return error;
1105 }
1106 #endif /* CONFIG_XPS */
1107
1108 int
1109 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1110 {
1111 #ifdef CONFIG_XPS
1112         int i;
1113         int error = 0;
1114
1115         for (i = old_num; i < new_num; i++) {
1116                 error = netdev_queue_add_kobject(net, i);
1117                 if (error) {
1118                         new_num = old_num;
1119                         break;
1120                 }
1121         }
1122
1123         while (--i >= new_num)
1124                 kobject_put(&net->_tx[i].kobj);
1125
1126         return error;
1127 #else
1128         return 0;
1129 #endif
1130 }
1131
1132 static int register_queue_kobjects(struct net_device *net)
1133 {
1134         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1135
1136 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1137         net->queues_kset = kset_create_and_add("queues",
1138             NULL, &net->dev.kobj);
1139         if (!net->queues_kset)
1140                 return -ENOMEM;
1141 #endif
1142
1143 #ifdef CONFIG_RPS
1144         real_rx = net->real_num_rx_queues;
1145 #endif
1146         real_tx = net->real_num_tx_queues;
1147
1148         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1149         if (error)
1150                 goto error;
1151         rxq = real_rx;
1152
1153         error = netdev_queue_update_kobjects(net, 0, real_tx);
1154         if (error)
1155                 goto error;
1156         txq = real_tx;
1157
1158         return 0;
1159
1160 error:
1161         netdev_queue_update_kobjects(net, txq, 0);
1162         net_rx_queue_update_kobjects(net, rxq, 0);
1163         return error;
1164 }
1165
1166 static void remove_queue_kobjects(struct net_device *net)
1167 {
1168         int real_rx = 0, real_tx = 0;
1169
1170 #ifdef CONFIG_RPS
1171         real_rx = net->real_num_rx_queues;
1172 #endif
1173         real_tx = net->real_num_tx_queues;
1174
1175         net_rx_queue_update_kobjects(net, real_rx, 0);
1176         netdev_queue_update_kobjects(net, real_tx, 0);
1177 #if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1178         kset_unregister(net->queues_kset);
1179 #endif
1180 }
1181
1182 static void *net_grab_current_ns(void)
1183 {
1184         struct net *ns = current->nsproxy->net_ns;
1185 #ifdef CONFIG_NET_NS
1186         if (ns)
1187                 atomic_inc(&ns->passive);
1188 #endif
1189         return ns;
1190 }
1191
1192 static const void *net_initial_ns(void)
1193 {
1194         return &init_net;
1195 }
1196
1197 static const void *net_netlink_ns(struct sock *sk)
1198 {
1199         return sock_net(sk);
1200 }
1201
1202 struct kobj_ns_type_operations net_ns_type_operations = {
1203         .type = KOBJ_NS_TYPE_NET,
1204         .grab_current_ns = net_grab_current_ns,
1205         .netlink_ns = net_netlink_ns,
1206         .initial_ns = net_initial_ns,
1207         .drop_ns = net_drop_ns,
1208 };
1209 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1210
1211 #ifdef CONFIG_HOTPLUG
1212 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1213 {
1214         struct net_device *dev = to_net_dev(d);
1215         int retval;
1216
1217         /* pass interface to uevent. */
1218         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1219         if (retval)
1220                 goto exit;
1221
1222         /* pass ifindex to uevent.
1223          * ifindex is useful as it won't change (interface name may change)
1224          * and is what RtNetlink uses natively. */
1225         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1226
1227 exit:
1228         return retval;
1229 }
1230 #endif
1231
1232 /*
1233  *      netdev_release -- destroy and free a dead device.
1234  *      Called when last reference to device kobject is gone.
1235  */
1236 static void netdev_release(struct device *d)
1237 {
1238         struct net_device *dev = to_net_dev(d);
1239
1240         BUG_ON(dev->reg_state != NETREG_RELEASED);
1241
1242         kfree(dev->ifalias);
1243         kfree((char *)dev - dev->padded);
1244 }
1245
1246 static const void *net_namespace(struct device *d)
1247 {
1248         struct net_device *dev;
1249         dev = container_of(d, struct net_device, dev);
1250         return dev_net(dev);
1251 }
1252
1253 static struct class net_class = {
1254         .name = "net",
1255         .dev_release = netdev_release,
1256 #ifdef CONFIG_SYSFS
1257         .dev_attrs = net_class_attributes,
1258 #endif /* CONFIG_SYSFS */
1259 #ifdef CONFIG_HOTPLUG
1260         .dev_uevent = netdev_uevent,
1261 #endif
1262         .ns_type = &net_ns_type_operations,
1263         .namespace = net_namespace,
1264 };
1265
1266 /* Delete sysfs entries but hold kobject reference until after all
1267  * netdev references are gone.
1268  */
1269 void netdev_unregister_kobject(struct net_device * net)
1270 {
1271         struct device *dev = &(net->dev);
1272
1273         kobject_get(&dev->kobj);
1274
1275         remove_queue_kobjects(net);
1276
1277         device_del(dev);
1278 }
1279
1280 /* Create sysfs entries for network device. */
1281 int netdev_register_kobject(struct net_device *net)
1282 {
1283         struct device *dev = &(net->dev);
1284         const struct attribute_group **groups = net->sysfs_groups;
1285         int error = 0;
1286
1287         device_initialize(dev);
1288         dev->class = &net_class;
1289         dev->platform_data = net;
1290         dev->groups = groups;
1291
1292         dev_set_name(dev, "%s", net->name);
1293
1294 #ifdef CONFIG_SYSFS
1295         /* Allow for a device specific group */
1296         if (*groups)
1297                 groups++;
1298
1299         *groups++ = &netstat_group;
1300 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1301         if (net->ieee80211_ptr)
1302                 *groups++ = &wireless_group;
1303 #ifdef CONFIG_WIRELESS_EXT
1304         else if (net->wireless_handlers)
1305                 *groups++ = &wireless_group;
1306 #endif
1307 #endif
1308 #endif /* CONFIG_SYSFS */
1309
1310         error = device_add(dev);
1311         if (error)
1312                 return error;
1313
1314         error = register_queue_kobjects(net);
1315         if (error) {
1316                 device_del(dev);
1317                 return error;
1318         }
1319
1320         return error;
1321 }
1322
1323 int netdev_class_create_file(struct class_attribute *class_attr)
1324 {
1325         return class_create_file(&net_class, class_attr);
1326 }
1327 EXPORT_SYMBOL(netdev_class_create_file);
1328
1329 void netdev_class_remove_file(struct class_attribute *class_attr)
1330 {
1331         class_remove_file(&net_class, class_attr);
1332 }
1333 EXPORT_SYMBOL(netdev_class_remove_file);
1334
1335 int netdev_kobject_init(void)
1336 {
1337         kobj_ns_type_register(&net_ns_type_operations);
1338         return class_register(&net_class);
1339 }