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