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 <linux/export.h>
  24 #include <linux/jiffies.h>
  25 #include <net/wext.h>
  26
  27 #include "net-sysfs.h"
  28
  29 #ifdef CONFIG_SYSFS
  30 static const char fmt_hex[] = "%#x\n";
  31 static const char fmt_long_hex[] = "%#lx\n";
  32 static const char fmt_dec[] = "%d\n";
  33 static const char fmt_udec[] = "%u\n";
  34 static const char fmt_ulong[] = "%lu\n";
  35 static const char fmt_u64[] = "%llu\n";
  36
  37 static inline int dev_isalive(const struct net_device *dev)
  38 {
  39         return dev->reg_state <= NETREG_REGISTERED;
  40 }
  41
  42 /* use same locking rules as GIF* ioctl's */
  43 static ssize_t netdev_show(const struct device *dev,
  44                            struct device_attribute *attr, char *buf,
  45                            ssize_t (*format)(const struct net_device *, char *))
  46 {
  47         struct net_device *net = to_net_dev(dev);
  48         ssize_t ret = -EINVAL;
  49
  50         read_lock(&dev_base_lock);
  51         if (dev_isalive(net))
  52                 ret = (*format)(net, buf);
  53         read_unlock(&dev_base_lock);
  54
  55         return ret;
  56 }
  57
  58 /* generate a show function for simple field */
  59 #define NETDEVICE_SHOW(field, format_string)                            \
  60 static ssize_t format_##field(const struct net_device *net, char *buf)  \
  61 {                                                                       \
  62         return sprintf(buf, format_string, net->field);                 \
  63 }                                                                       \
  64 static ssize_t show_##field(struct device *dev,                         \
  65                             struct device_attribute *attr, char *buf)   \
  66 {                                                                       \
  67         return netdev_show(dev, attr, buf, format_##field);             \
  68 }
  69
  70
  71 /* use same locking and permission rules as SIF* ioctl's */
  72 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  73                             const char *buf, size_t len,
  74                             int (*set)(struct net_device *, unsigned long))
  75 {
  76         struct net_device *net = to_net_dev(dev);
  77         unsigned long new;
  78         int ret = -EINVAL;
  79
  80         if (!capable(CAP_NET_ADMIN))
  81                 return -EPERM;
  82
  83         ret = kstrtoul(buf, 0, &new);
  84         if (ret)
  85                 goto err;
  86
  87         if (!rtnl_trylock())
  88                 return restart_syscall();
  89
  90         if (dev_isalive(net)) {
  91                 if ((ret = (*set)(net, new)) == 0)
  92                         ret = len;
  93         }
  94         rtnl_unlock();
  95  err:
  96         return ret;
  97 }
  98
  99 NETDEVICE_SHOW(dev_id, fmt_hex);
 100 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 101 NETDEVICE_SHOW(addr_len, fmt_dec);
 102 NETDEVICE_SHOW(iflink, fmt_dec);
 103 NETDEVICE_SHOW(ifindex, fmt_dec);
 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 (!__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 (!__ethtool_get_settings(netdev, &cmd)) {
 170                         const char *duplex;
 171                         switch (cmd.duplex) {
 172                         case DUPLEX_HALF:
 173                                 duplex = "half";
 174                                 break;
 175                         case DUPLEX_FULL:
 176                                 duplex = "full";
 177                                 break;
 178                         default:
 179                                 duplex = "unknown";
 180                                 break;
 181                         }
 182                         ret = sprintf(buf, "%s\n", duplex);
 183                 }
 184         }
 185         rtnl_unlock();
 186         return ret;
 187 }
 188
 189 static ssize_t show_dormant(struct device *dev,
 190                             struct device_attribute *attr, char *buf)
 191 {
 192         struct net_device *netdev = to_net_dev(dev);
 193
 194         if (netif_running(netdev))
 195                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 196
 197         return -EINVAL;
 198 }
 199
 200 static const char *const operstates[] = {
 201         "unknown",
 202         "notpresent", /* currently unused */
 203         "down",
 204         "lowerlayerdown",
 205         "testing", /* currently unused */
 206         "dormant",
 207         "up"
 208 };
 209
 210 static ssize_t show_operstate(struct device *dev,
 211                               struct device_attribute *attr, char *buf)
 212 {
 213         const struct net_device *netdev = to_net_dev(dev);
 214         unsigned char operstate;
 215
 216         read_lock(&dev_base_lock);
 217         operstate = netdev->operstate;
 218         if (!netif_running(netdev))
 219                 operstate = IF_OPER_DOWN;
 220         read_unlock(&dev_base_lock);
 221
 222         if (operstate >= ARRAY_SIZE(operstates))
 223                 return -EINVAL; /* should not happen */
 224
 225         return sprintf(buf, "%s\n", operstates[operstate]);
 226 }
 227
 228 /* read-write attributes */
 229 NETDEVICE_SHOW(mtu, fmt_dec);
 230
 231 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 232 {
 233         return dev_set_mtu(net, (int) new_mtu);
 234 }
 235
 236 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 237                          const char *buf, size_t len)
 238 {
 239         return netdev_store(dev, attr, buf, len, change_mtu);
 240 }
 241
 242 NETDEVICE_SHOW(flags, fmt_hex);
 243
 244 static int change_flags(struct net_device *net, unsigned long new_flags)
 245 {
 246         return dev_change_flags(net, (unsigned int) new_flags);
 247 }
 248
 249 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 250                            const char *buf, size_t len)
 251 {
 252         return netdev_store(dev, attr, buf, len, change_flags);
 253 }
 254
 255 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 256
 257 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 258 {
 259         net->tx_queue_len = new_len;
 260         return 0;
 261 }
 262
 263 static ssize_t store_tx_queue_len(struct device *dev,
 264                                   struct device_attribute *attr,
 265                                   const char *buf, size_t len)
 266 {
 267         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 268 }
 269
 270 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 271                              const char *buf, size_t len)
 272 {
 273         struct net_device *netdev = to_net_dev(dev);
 274         size_t count = len;
 275         ssize_t ret;
 276
 277         if (!capable(CAP_NET_ADMIN))
 278                 return -EPERM;
 279
 280         /* ignore trailing newline */
 281         if (len >  0 && buf[len - 1] == '\n')
 282                 --count;
 283
 284         if (!rtnl_trylock())
 285                 return restart_syscall();
 286         ret = dev_set_alias(netdev, buf, count);
 287         rtnl_unlock();
 288
 289         return ret < 0 ? ret : len;
 290 }
 291
 292 static ssize_t show_ifalias(struct device *dev,
 293                             struct device_attribute *attr, char *buf)
 294 {
 295         const struct net_device *netdev = to_net_dev(dev);
 296         ssize_t ret = 0;
 297
 298         if (!rtnl_trylock())
 299                 return restart_syscall();
 300         if (netdev->ifalias)
 301                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 302         rtnl_unlock();
 303         return ret;
 304 }
 305
 306 NETDEVICE_SHOW(group, fmt_dec);
 307
 308 static int change_group(struct net_device *net, unsigned long new_group)
 309 {
 310         dev_set_group(net, (int) new_group);
 311         return 0;
 312 }
 313
 314 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 315                          const char *buf, size_t len)
 316 {
 317         return netdev_store(dev, attr, buf, len, change_group);
 318 }
 319
 320 static struct device_attribute net_class_attributes[] = {
 321         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 322         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 323         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 324         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 325         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 326         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 327         __ATTR(type, S_IRUGO, show_type, NULL),
 328         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 329         __ATTR(address, S_IRUGO, show_address, NULL),
 330         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 331         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
 332         __ATTR(speed, S_IRUGO, show_speed, NULL),
 333         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 334         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 335         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 336         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 337         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 338         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 339                store_tx_queue_len),
 340         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 341         {}
 342 };
 343
 344 /* Show a given an attribute in the statistics group */
 345 static ssize_t netstat_show(const struct device *d,
 346                             struct device_attribute *attr, char *buf,
 347                             unsigned long offset)
 348 {
 349         struct net_device *dev = to_net_dev(d);
 350         ssize_t ret = -EINVAL;
 351
 352         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 353                         offset % sizeof(u64) != 0);
 354
 355         read_lock(&dev_base_lock);
 356         if (dev_isalive(dev)) {
 357                 struct rtnl_link_stats64 temp;
 358                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 359
 360                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 361         }
 362         read_unlock(&dev_base_lock);
 363         return ret;
 364 }
 365
 366 /* generate a read-only statistics attribute */
 367 #define NETSTAT_ENTRY(name)                                             \
 368 static ssize_t show_##name(struct device *d,                            \
 369                            struct device_attribute *attr, char *buf)    \
 370 {                                                                       \
 371         return netstat_show(d, attr, buf,                               \
 372                             offsetof(struct rtnl_link_stats64, name));  \
 373 }                                                                       \
 374 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 375
 376 NETSTAT_ENTRY(rx_packets);
 377 NETSTAT_ENTRY(tx_packets);
 378 NETSTAT_ENTRY(rx_bytes);
 379 NETSTAT_ENTRY(tx_bytes);
 380 NETSTAT_ENTRY(rx_errors);
 381 NETSTAT_ENTRY(tx_errors);
 382 NETSTAT_ENTRY(rx_dropped);
 383 NETSTAT_ENTRY(tx_dropped);
 384 NETSTAT_ENTRY(multicast);
 385 NETSTAT_ENTRY(collisions);
 386 NETSTAT_ENTRY(rx_length_errors);
 387 NETSTAT_ENTRY(rx_over_errors);
 388 NETSTAT_ENTRY(rx_crc_errors);
 389 NETSTAT_ENTRY(rx_frame_errors);
 390 NETSTAT_ENTRY(rx_fifo_errors);
 391 NETSTAT_ENTRY(rx_missed_errors);
 392 NETSTAT_ENTRY(tx_aborted_errors);
 393 NETSTAT_ENTRY(tx_carrier_errors);
 394 NETSTAT_ENTRY(tx_fifo_errors);
 395 NETSTAT_ENTRY(tx_heartbeat_errors);
 396 NETSTAT_ENTRY(tx_window_errors);
 397 NETSTAT_ENTRY(rx_compressed);
 398 NETSTAT_ENTRY(tx_compressed);
 399
 400 static struct attribute *netstat_attrs[] = {
 401         &dev_attr_rx_packets.attr,
 402         &dev_attr_tx_packets.attr,
 403         &dev_attr_rx_bytes.attr,
 404         &dev_attr_tx_bytes.attr,
 405         &dev_attr_rx_errors.attr,
 406         &dev_attr_tx_errors.attr,
 407         &dev_attr_rx_dropped.attr,
 408         &dev_attr_tx_dropped.attr,
 409         &dev_attr_multicast.attr,
 410         &dev_attr_collisions.attr,
 411         &dev_attr_rx_length_errors.attr,
 412         &dev_attr_rx_over_errors.attr,
 413         &dev_attr_rx_crc_errors.attr,
 414         &dev_attr_rx_frame_errors.attr,
 415         &dev_attr_rx_fifo_errors.attr,
 416         &dev_attr_rx_missed_errors.attr,
 417         &dev_attr_tx_aborted_errors.attr,
 418         &dev_attr_tx_carrier_errors.attr,
 419         &dev_attr_tx_fifo_errors.attr,
 420         &dev_attr_tx_heartbeat_errors.attr,
 421         &dev_attr_tx_window_errors.attr,
 422         &dev_attr_rx_compressed.attr,
 423         &dev_attr_tx_compressed.attr,
 424         NULL
 425 };
 426
 427
 428 static struct attribute_group netstat_group = {
 429         .name  = "statistics",
 430         .attrs  = netstat_attrs,
 431 };
 432 #endif /* CONFIG_SYSFS */
 433
 434 #ifdef CONFIG_RPS
 435 /*
 436  * RX queue sysfs structures and functions.
 437  */
 438 struct rx_queue_attribute {
 439         struct attribute attr;
 440         ssize_t (*show)(struct netdev_rx_queue *queue,
 441             struct rx_queue_attribute *attr, char *buf);
 442         ssize_t (*store)(struct netdev_rx_queue *queue,
 443             struct rx_queue_attribute *attr, const char *buf, size_t len);
 444 };
 445 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 446     struct rx_queue_attribute, attr)
 447
 448 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 449
 450 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 451                                   char *buf)
 452 {
 453         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 454         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 455
 456         if (!attribute->show)
 457                 return -EIO;
 458
 459         return attribute->show(queue, attribute, buf);
 460 }
 461
 462 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 463                                    const char *buf, size_t count)
 464 {
 465         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 466         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 467
 468         if (!attribute->store)
 469                 return -EIO;
 470
 471         return attribute->store(queue, attribute, buf, count);
 472 }
 473
 474 static const struct sysfs_ops rx_queue_sysfs_ops = {
 475         .show = rx_queue_attr_show,
 476         .store = rx_queue_attr_store,
 477 };
 478
 479 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 480                             struct rx_queue_attribute *attribute, char *buf)
 481 {
 482         struct rps_map *map;
 483         cpumask_var_t mask;
 484         size_t len = 0;
 485         int i;
 486
 487         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 488                 return -ENOMEM;
 489
 490         rcu_read_lock();
 491         map = rcu_dereference(queue->rps_map);
 492         if (map)
 493                 for (i = 0; i < map->len; i++)
 494                         cpumask_set_cpu(map->cpus[i], mask);
 495
 496         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 497         if (PAGE_SIZE - len < 3) {
 498                 rcu_read_unlock();
 499                 free_cpumask_var(mask);
 500                 return -EINVAL;
 501         }
 502         rcu_read_unlock();
 503
 504         free_cpumask_var(mask);
 505         len += sprintf(buf + len, "\n");
 506         return len;
 507 }
 508
 509 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 510                       struct rx_queue_attribute *attribute,
 511                       const char *buf, size_t len)
 512 {
 513         struct rps_map *old_map, *map;
 514         cpumask_var_t mask;
 515         int err, cpu, i;
 516         static DEFINE_SPINLOCK(rps_map_lock);
 517
 518         if (!capable(CAP_NET_ADMIN))
 519                 return -EPERM;
 520
 521         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 522                 return -ENOMEM;
 523
 524         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 525         if (err) {
 526                 free_cpumask_var(mask);
 527                 return err;
 528         }
 529
 530         map = kzalloc(max_t(unsigned int,
 531             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 532             GFP_KERNEL);
 533         if (!map) {
 534                 free_cpumask_var(mask);
 535                 return -ENOMEM;
 536         }
 537
 538         i = 0;
 539         for_each_cpu_and(cpu, mask, cpu_online_mask)
 540                 map->cpus[i++] = cpu;
 541
 542         if (i)
 543                 map->len = i;
 544         else {
 545                 kfree(map);
 546                 map = NULL;
 547         }
 548
 549         spin_lock(&rps_map_lock);
 550         old_map = rcu_dereference_protected(queue->rps_map,
 551                                             lockdep_is_held(&rps_map_lock));
 552         rcu_assign_pointer(queue->rps_map, map);
 553         spin_unlock(&rps_map_lock);
 554
 555         if (map)
 556                 static_key_slow_inc(&rps_needed);
 557         if (old_map) {
 558                 kfree_rcu(old_map, rcu);
 559                 static_key_slow_dec(&rps_needed);
 560         }
 561         free_cpumask_var(mask);
 562         return len;
 563 }
 564
 565 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 566                                            struct rx_queue_attribute *attr,
 567                                            char *buf)
 568 {
 569         struct rps_dev_flow_table *flow_table;
 570         unsigned long val = 0;
 571
 572         rcu_read_lock();
 573         flow_table = rcu_dereference(queue->rps_flow_table);
 574         if (flow_table)
 575                 val = (unsigned long)flow_table->mask + 1;
 576         rcu_read_unlock();
 577
 578         return sprintf(buf, "%lu\n", val);
 579 }
 580
 581 static void rps_dev_flow_table_release_work(struct work_struct *work)
 582 {
 583         struct rps_dev_flow_table *table = container_of(work,
 584             struct rps_dev_flow_table, free_work);
 585
 586         vfree(table);
 587 }
 588
 589 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 590 {
 591         struct rps_dev_flow_table *table = container_of(rcu,
 592             struct rps_dev_flow_table, rcu);
 593
 594         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 595         schedule_work(&table->free_work);
 596 }
 597
 598 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 599                                      struct rx_queue_attribute *attr,
 600                                      const char *buf, size_t len)
 601 {
 602         unsigned long mask, count;
 603         struct rps_dev_flow_table *table, *old_table;
 604         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 605         int rc;
 606
 607         if (!capable(CAP_NET_ADMIN))
 608                 return -EPERM;
 609
 610         rc = kstrtoul(buf, 0, &count);
 611         if (rc < 0)
 612                 return rc;
 613
 614         if (count) {
 615                 mask = count - 1;
 616                 /* mask = roundup_pow_of_two(count) - 1;
 617                  * without overflows...
 618                  */
 619                 while ((mask | (mask >> 1)) != mask)
 620                         mask |= (mask >> 1);
 621                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 622                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 623                  * doesnt overflow.
 624                  */
 625 #if BITS_PER_LONG > 32
 626                 if (mask > (unsigned long)(u32)mask)
 627                         return -EINVAL;
 628 #else
 629                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 630                                 / sizeof(struct rps_dev_flow)) {
 631                         /* Enforce a limit to prevent overflow */
 632                         return -EINVAL;
 633                 }
 634 #endif
 635                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 636                 if (!table)
 637                         return -ENOMEM;
 638
 639                 table->mask = mask;
 640                 for (count = 0; count <= mask; count++)
 641                         table->flows[count].cpu = RPS_NO_CPU;
 642         } else
 643                 table = NULL;
 644
 645         spin_lock(&rps_dev_flow_lock);
 646         old_table = rcu_dereference_protected(queue->rps_flow_table,
 647                                               lockdep_is_held(&rps_dev_flow_lock));
 648         rcu_assign_pointer(queue->rps_flow_table, table);
 649         spin_unlock(&rps_dev_flow_lock);
 650
 651         if (old_table)
 652                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 653
 654         return len;
 655 }
 656
 657 static struct rx_queue_attribute rps_cpus_attribute =
 658         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 659
 660
 661 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 662         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 663             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 664
 665 static struct attribute *rx_queue_default_attrs[] = {
 666         &rps_cpus_attribute.attr,
 667         &rps_dev_flow_table_cnt_attribute.attr,
 668         NULL
 669 };
 670
 671 static void rx_queue_release(struct kobject *kobj)
 672 {
 673         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 674         struct rps_map *map;
 675         struct rps_dev_flow_table *flow_table;
 676
 677
 678         map = rcu_dereference_protected(queue->rps_map, 1);
 679         if (map) {
 680                 RCU_INIT_POINTER(queue->rps_map, NULL);
 681                 kfree_rcu(map, rcu);
 682         }
 683
 684         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 685         if (flow_table) {
 686                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 687                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 688         }
 689
 690         memset(kobj, 0, sizeof(*kobj));
 691         dev_put(queue->dev);
 692 }
 693
 694 static struct kobj_type rx_queue_ktype = {
 695         .sysfs_ops = &rx_queue_sysfs_ops,
 696         .release = rx_queue_release,
 697         .default_attrs = rx_queue_default_attrs,
 698 };
 699
 700 static int rx_queue_add_kobject(struct net_device *net, int index)
 701 {
 702         struct netdev_rx_queue *queue = net->_rx + index;
 703         struct kobject *kobj = &queue->kobj;
 704         int error = 0;
 705
 706         kobj->kset = net->queues_kset;
 707         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 708             "rx-%u", index);
 709         if (error) {
 710                 kobject_put(kobj);
 711                 return error;
 712         }
 713
 714         kobject_uevent(kobj, KOBJ_ADD);
 715         dev_hold(queue->dev);
 716
 717         return error;
 718 }
 719 #endif /* CONFIG_RPS */
 720
 721 int
 722 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 723 {
 724 #ifdef CONFIG_RPS
 725         int i;
 726         int error = 0;
 727
 728         for (i = old_num; i < new_num; i++) {
 729                 error = rx_queue_add_kobject(net, i);
 730                 if (error) {
 731                         new_num = old_num;
 732                         break;
 733                 }
 734         }
 735
 736         while (--i >= new_num)
 737                 kobject_put(&net->_rx[i].kobj);
 738
 739         return error;
 740 #else
 741         return 0;
 742 #endif
 743 }
 744
 745 #ifdef CONFIG_SYSFS
 746 /*
 747  * netdev_queue sysfs structures and functions.
 748  */
 749 struct netdev_queue_attribute {
 750         struct attribute attr;
 751         ssize_t (*show)(struct netdev_queue *queue,
 752             struct netdev_queue_attribute *attr, char *buf);
 753         ssize_t (*store)(struct netdev_queue *queue,
 754             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 755 };
 756 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 757     struct netdev_queue_attribute, attr)
 758
 759 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 760
 761 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 762                                       struct attribute *attr, char *buf)
 763 {
 764         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 765         struct netdev_queue *queue = to_netdev_queue(kobj);
 766
 767         if (!attribute->show)
 768                 return -EIO;
 769
 770         return attribute->show(queue, attribute, buf);
 771 }
 772
 773 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 774                                        struct attribute *attr,
 775                                        const char *buf, size_t count)
 776 {
 777         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 778         struct netdev_queue *queue = to_netdev_queue(kobj);
 779
 780         if (!attribute->store)
 781                 return -EIO;
 782
 783         return attribute->store(queue, attribute, buf, count);
 784 }
 785
 786 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 787         .show = netdev_queue_attr_show,
 788         .store = netdev_queue_attr_store,
 789 };
 790
 791 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 792                                   struct netdev_queue_attribute *attribute,
 793                                   char *buf)
 794 {
 795         unsigned long trans_timeout;
 796
 797         spin_lock_irq(&queue->_xmit_lock);
 798         trans_timeout = queue->trans_timeout;
 799         spin_unlock_irq(&queue->_xmit_lock);
 800
 801         return sprintf(buf, "%lu", trans_timeout);
 802 }
 803
 804 static struct netdev_queue_attribute queue_trans_timeout =
 805         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 806
 807 #ifdef CONFIG_BQL
 808 /*
 809  * Byte queue limits sysfs structures and functions.
 810  */
 811 static ssize_t bql_show(char *buf, unsigned int value)
 812 {
 813         return sprintf(buf, "%u\n", value);
 814 }
 815
 816 static ssize_t bql_set(const char *buf, const size_t count,
 817                        unsigned int *pvalue)
 818 {
 819         unsigned int value;
 820         int err;
 821
 822         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 823                 value = DQL_MAX_LIMIT;
 824         else {
 825                 err = kstrtouint(buf, 10, &value);
 826                 if (err < 0)
 827                         return err;
 828                 if (value > DQL_MAX_LIMIT)
 829                         return -EINVAL;
 830         }
 831
 832         *pvalue = value;
 833
 834         return count;
 835 }
 836
 837 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 838                                   struct netdev_queue_attribute *attr,
 839                                   char *buf)
 840 {
 841         struct dql *dql = &queue->dql;
 842
 843         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 844 }
 845
 846 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 847                                  struct netdev_queue_attribute *attribute,
 848                                  const char *buf, size_t len)
 849 {
 850         struct dql *dql = &queue->dql;
 851         unsigned int value;
 852         int err;
 853
 854         err = kstrtouint(buf, 10, &value);
 855         if (err < 0)
 856                 return err;
 857
 858         dql->slack_hold_time = msecs_to_jiffies(value);
 859
 860         return len;
 861 }
 862
 863 static struct netdev_queue_attribute bql_hold_time_attribute =
 864         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 865             bql_set_hold_time);
 866
 867 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 868                                  struct netdev_queue_attribute *attr,
 869                                  char *buf)
 870 {
 871         struct dql *dql = &queue->dql;
 872
 873         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 874 }
 875
 876 static struct netdev_queue_attribute bql_inflight_attribute =
 877         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 878
 879 #define BQL_ATTR(NAME, FIELD)                                           \
 880 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 881                                  struct netdev_queue_attribute *attr,   \
 882                                  char *buf)                             \
 883 {                                                                       \
 884         return bql_show(buf, queue->dql.FIELD);                         \
 885 }                                                                       \
 886                                                                         \
 887 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 888                                 struct netdev_queue_attribute *attr,    \
 889                                 const char *buf, size_t len)            \
 890 {                                                                       \
 891         return bql_set(buf, len, &queue->dql.FIELD);                    \
 892 }                                                                       \
 893                                                                         \
 894 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 895         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 896             bql_set_ ## NAME);
 897
 898 BQL_ATTR(limit, limit)
 899 BQL_ATTR(limit_max, max_limit)
 900 BQL_ATTR(limit_min, min_limit)
 901
 902 static struct attribute *dql_attrs[] = {
 903         &bql_limit_attribute.attr,
 904         &bql_limit_max_attribute.attr,
 905         &bql_limit_min_attribute.attr,
 906         &bql_hold_time_attribute.attr,
 907         &bql_inflight_attribute.attr,
 908         NULL
 909 };
 910
 911 static struct attribute_group dql_group = {
 912         .name  = "byte_queue_limits",
 913         .attrs  = dql_attrs,
 914 };
 915 #endif /* CONFIG_BQL */
 916
 917 #ifdef CONFIG_XPS
 918 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 919 {
 920         struct net_device *dev = queue->dev;
 921         int i;
 922
 923         for (i = 0; i < dev->num_tx_queues; i++)
 924                 if (queue == &dev->_tx[i])
 925                         break;
 926
 927         BUG_ON(i >= dev->num_tx_queues);
 928
 929         return i;
 930 }
 931
 932
 933 static ssize_t show_xps_map(struct netdev_queue *queue,
 934                             struct netdev_queue_attribute *attribute, char *buf)
 935 {
 936         struct net_device *dev = queue->dev;
 937         struct xps_dev_maps *dev_maps;
 938         cpumask_var_t mask;
 939         unsigned long index;
 940         size_t len = 0;
 941         int i;
 942
 943         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 944                 return -ENOMEM;
 945
 946         index = get_netdev_queue_index(queue);
 947
 948         rcu_read_lock();
 949         dev_maps = rcu_dereference(dev->xps_maps);
 950         if (dev_maps) {
 951                 for_each_possible_cpu(i) {
 952                         struct xps_map *map =
 953                             rcu_dereference(dev_maps->cpu_map[i]);
 954                         if (map) {
 955                                 int j;
 956                                 for (j = 0; j < map->len; j++) {
 957                                         if (map->queues[j] == index) {
 958                                                 cpumask_set_cpu(i, mask);
 959                                                 break;
 960                                         }
 961                                 }
 962                         }
 963                 }
 964         }
 965         rcu_read_unlock();
 966
 967         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 968         if (PAGE_SIZE - len < 3) {
 969                 free_cpumask_var(mask);
 970                 return -EINVAL;
 971         }
 972
 973         free_cpumask_var(mask);
 974         len += sprintf(buf + len, "\n");
 975         return len;
 976 }
 977
 978 static DEFINE_MUTEX(xps_map_mutex);
 979 #define xmap_dereference(P)             \
 980         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
 981
 982 static void xps_queue_release(struct netdev_queue *queue)
 983 {
 984         struct net_device *dev = queue->dev;
 985         struct xps_dev_maps *dev_maps;
 986         struct xps_map *map;
 987         unsigned long index;
 988         int i, pos, nonempty = 0;
 989
 990         index = get_netdev_queue_index(queue);
 991
 992         mutex_lock(&xps_map_mutex);
 993         dev_maps = xmap_dereference(dev->xps_maps);
 994
 995         if (dev_maps) {
 996                 for_each_possible_cpu(i) {
 997                         map = xmap_dereference(dev_maps->cpu_map[i]);
 998                         if (!map)
 999                                 continue;
1000
1001                         for (pos = 0; pos < map->len; pos++)
1002                                 if (map->queues[pos] == index)
1003                                         break;
1004
1005                         if (pos < map->len) {
1006                                 if (map->len > 1)
1007                                         map->queues[pos] =
1008                                             map->queues[--map->len];
1009                                 else {
1010                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1011                                             NULL);
1012                                         kfree_rcu(map, rcu);
1013                                         map = NULL;
1014                                 }
1015                         }
1016                         if (map)
1017                                 nonempty = 1;
1018                 }
1019
1020                 if (!nonempty) {
1021                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1022                         kfree_rcu(dev_maps, rcu);
1023                 }
1024         }
1025         mutex_unlock(&xps_map_mutex);
1026 }
1027
1028 static ssize_t store_xps_map(struct netdev_queue *queue,
1029                       struct netdev_queue_attribute *attribute,
1030                       const char *buf, size_t len)
1031 {
1032         struct net_device *dev = queue->dev;
1033         cpumask_var_t mask;
1034         int err, i, cpu, pos, map_len, alloc_len, need_set;
1035         unsigned long index;
1036         struct xps_map *map, *new_map;
1037         struct xps_dev_maps *dev_maps, *new_dev_maps;
1038         int nonempty = 0;
1039         int numa_node_id = -2;
1040
1041         if (!capable(CAP_NET_ADMIN))
1042                 return -EPERM;
1043
1044         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1045                 return -ENOMEM;
1046
1047         index = get_netdev_queue_index(queue);
1048
1049         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1050         if (err) {
1051                 free_cpumask_var(mask);
1052                 return err;
1053         }
1054
1055         new_dev_maps = kzalloc(max_t(unsigned int,
1056             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1057         if (!new_dev_maps) {
1058                 free_cpumask_var(mask);
1059                 return -ENOMEM;
1060         }
1061
1062         mutex_lock(&xps_map_mutex);
1063
1064         dev_maps = xmap_dereference(dev->xps_maps);
1065
1066         for_each_possible_cpu(cpu) {
1067                 map = dev_maps ?
1068                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1069                 new_map = map;
1070                 if (map) {
1071                         for (pos = 0; pos < map->len; pos++)
1072                                 if (map->queues[pos] == index)
1073                                         break;
1074                         map_len = map->len;
1075                         alloc_len = map->alloc_len;
1076                 } else
1077                         pos = map_len = alloc_len = 0;
1078
1079                 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1080 #ifdef CONFIG_NUMA
1081                 if (need_set) {
1082                         if (numa_node_id == -2)
1083                                 numa_node_id = cpu_to_node(cpu);
1084                         else if (numa_node_id != cpu_to_node(cpu))
1085                                 numa_node_id = -1;
1086                 }
1087 #endif
1088                 if (need_set && pos >= map_len) {
1089                         /* Need to add queue to this CPU's map */
1090                         if (map_len >= alloc_len) {
1091                                 alloc_len = alloc_len ?
1092                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
1093                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1094                                                        GFP_KERNEL,
1095                                                        cpu_to_node(cpu));
1096                                 if (!new_map)
1097                                         goto error;
1098                                 new_map->alloc_len = alloc_len;
1099                                 for (i = 0; i < map_len; i++)
1100                                         new_map->queues[i] = map->queues[i];
1101                                 new_map->len = map_len;
1102                         }
1103                         new_map->queues[new_map->len++] = index;
1104                 } else if (!need_set && pos < map_len) {
1105                         /* Need to remove queue from this CPU's map */
1106                         if (map_len > 1)
1107                                 new_map->queues[pos] =
1108                                     new_map->queues[--new_map->len];
1109                         else
1110                                 new_map = NULL;
1111                 }
1112                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1113         }
1114
1115         /* Cleanup old maps */
1116         for_each_possible_cpu(cpu) {
1117                 map = dev_maps ?
1118                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1119                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1120                         kfree_rcu(map, rcu);
1121                 if (new_dev_maps->cpu_map[cpu])
1122                         nonempty = 1;
1123         }
1124
1125         if (nonempty) {
1126                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1127         } else {
1128                 kfree(new_dev_maps);
1129                 RCU_INIT_POINTER(dev->xps_maps, NULL);
1130         }
1131
1132         if (dev_maps)
1133                 kfree_rcu(dev_maps, rcu);
1134
1135         netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1136                                             NUMA_NO_NODE);
1137
1138         mutex_unlock(&xps_map_mutex);
1139
1140         free_cpumask_var(mask);
1141         return len;
1142
1143 error:
1144         mutex_unlock(&xps_map_mutex);
1145
1146         if (new_dev_maps)
1147                 for_each_possible_cpu(i)
1148                         kfree(rcu_dereference_protected(
1149                                 new_dev_maps->cpu_map[i],
1150                                 1));
1151         kfree(new_dev_maps);
1152         free_cpumask_var(mask);
1153         return -ENOMEM;
1154 }
1155
1156 static struct netdev_queue_attribute xps_cpus_attribute =
1157     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1158 #endif /* CONFIG_XPS */
1159
1160 static struct attribute *netdev_queue_default_attrs[] = {
1161         &queue_trans_timeout.attr,
1162 #ifdef CONFIG_XPS
1163         &xps_cpus_attribute.attr,
1164 #endif
1165         NULL
1166 };
1167
1168 static void netdev_queue_release(struct kobject *kobj)
1169 {
1170         struct netdev_queue *queue = to_netdev_queue(kobj);
1171
1172 #ifdef CONFIG_XPS
1173         xps_queue_release(queue);
1174 #endif
1175
1176         memset(kobj, 0, sizeof(*kobj));
1177         dev_put(queue->dev);
1178 }
1179
1180 static struct kobj_type netdev_queue_ktype = {
1181         .sysfs_ops = &netdev_queue_sysfs_ops,
1182         .release = netdev_queue_release,
1183         .default_attrs = netdev_queue_default_attrs,
1184 };
1185
1186 static int netdev_queue_add_kobject(struct net_device *net, int index)
1187 {
1188         struct netdev_queue *queue = net->_tx + index;
1189         struct kobject *kobj = &queue->kobj;
1190         int error = 0;
1191
1192         kobj->kset = net->queues_kset;
1193         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1194             "tx-%u", index);
1195         if (error)
1196                 goto exit;
1197
1198 #ifdef CONFIG_BQL
1199         error = sysfs_create_group(kobj, &dql_group);
1200         if (error)
1201                 goto exit;
1202 #endif
1203
1204         kobject_uevent(kobj, KOBJ_ADD);
1205         dev_hold(queue->dev);
1206
1207         return 0;
1208 exit:
1209         kobject_put(kobj);
1210         return error;
1211 }
1212 #endif /* CONFIG_SYSFS */
1213
1214 int
1215 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1216 {
1217 #ifdef CONFIG_SYSFS
1218         int i;
1219         int error = 0;
1220
1221         for (i = old_num; i < new_num; i++) {
1222                 error = netdev_queue_add_kobject(net, i);
1223                 if (error) {
1224                         new_num = old_num;
1225                         break;
1226                 }
1227         }
1228
1229         while (--i >= new_num) {
1230                 struct netdev_queue *queue = net->_tx + i;
1231
1232 #ifdef CONFIG_BQL
1233                 sysfs_remove_group(&queue->kobj, &dql_group);
1234 #endif
1235                 kobject_put(&queue->kobj);
1236         }
1237
1238         return error;
1239 #else
1240         return 0;
1241 #endif /* CONFIG_SYSFS */
1242 }
1243
1244 static int register_queue_kobjects(struct net_device *net)
1245 {
1246         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1247
1248 #ifdef CONFIG_SYSFS
1249         net->queues_kset = kset_create_and_add("queues",
1250             NULL, &net->dev.kobj);
1251         if (!net->queues_kset)
1252                 return -ENOMEM;
1253 #endif
1254
1255 #ifdef CONFIG_RPS
1256         real_rx = net->real_num_rx_queues;
1257 #endif
1258         real_tx = net->real_num_tx_queues;
1259
1260         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1261         if (error)
1262                 goto error;
1263         rxq = real_rx;
1264
1265         error = netdev_queue_update_kobjects(net, 0, real_tx);
1266         if (error)
1267                 goto error;
1268         txq = real_tx;
1269
1270         return 0;
1271
1272 error:
1273         netdev_queue_update_kobjects(net, txq, 0);
1274         net_rx_queue_update_kobjects(net, rxq, 0);
1275         return error;
1276 }
1277
1278 static void remove_queue_kobjects(struct net_device *net)
1279 {
1280         int real_rx = 0, real_tx = 0;
1281
1282 #ifdef CONFIG_RPS
1283         real_rx = net->real_num_rx_queues;
1284 #endif
1285         real_tx = net->real_num_tx_queues;
1286
1287         net_rx_queue_update_kobjects(net, real_rx, 0);
1288         netdev_queue_update_kobjects(net, real_tx, 0);
1289 #ifdef CONFIG_SYSFS
1290         kset_unregister(net->queues_kset);
1291 #endif
1292 }
1293
1294 static void *net_grab_current_ns(void)
1295 {
1296         struct net *ns = current->nsproxy->net_ns;
1297 #ifdef CONFIG_NET_NS
1298         if (ns)
1299                 atomic_inc(&ns->passive);
1300 #endif
1301         return ns;
1302 }
1303
1304 static const void *net_initial_ns(void)
1305 {
1306         return &init_net;
1307 }
1308
1309 static const void *net_netlink_ns(struct sock *sk)
1310 {
1311         return sock_net(sk);
1312 }
1313
1314 struct kobj_ns_type_operations net_ns_type_operations = {
1315         .type = KOBJ_NS_TYPE_NET,
1316         .grab_current_ns = net_grab_current_ns,
1317         .netlink_ns = net_netlink_ns,
1318         .initial_ns = net_initial_ns,
1319         .drop_ns = net_drop_ns,
1320 };
1321 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1322
1323 #ifdef CONFIG_HOTPLUG
1324 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1325 {
1326         struct net_device *dev = to_net_dev(d);
1327         int retval;
1328
1329         /* pass interface to uevent. */
1330         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1331         if (retval)
1332                 goto exit;
1333
1334         /* pass ifindex to uevent.
1335          * ifindex is useful as it won't change (interface name may change)
1336          * and is what RtNetlink uses natively. */
1337         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1338
1339 exit:
1340         return retval;
1341 }
1342 #endif
1343
1344 /*
1345  *      netdev_release -- destroy and free a dead device.
1346  *      Called when last reference to device kobject is gone.
1347  */
1348 static void netdev_release(struct device *d)
1349 {
1350         struct net_device *dev = to_net_dev(d);
1351
1352         BUG_ON(dev->reg_state != NETREG_RELEASED);
1353
1354         kfree(dev->ifalias);
1355         kfree((char *)dev - dev->padded);
1356 }
1357
1358 static const void *net_namespace(struct device *d)
1359 {
1360         struct net_device *dev;
1361         dev = container_of(d, struct net_device, dev);
1362         return dev_net(dev);
1363 }
1364
1365 static struct class net_class = {
1366         .name = "net",
1367         .dev_release = netdev_release,
1368 #ifdef CONFIG_SYSFS
1369         .dev_attrs = net_class_attributes,
1370 #endif /* CONFIG_SYSFS */
1371 #ifdef CONFIG_HOTPLUG
1372         .dev_uevent = netdev_uevent,
1373 #endif
1374         .ns_type = &net_ns_type_operations,
1375         .namespace = net_namespace,
1376 };
1377
1378 /* Delete sysfs entries but hold kobject reference until after all
1379  * netdev references are gone.
1380  */
1381 void netdev_unregister_kobject(struct net_device * net)
1382 {
1383         struct device *dev = &(net->dev);
1384
1385         kobject_get(&dev->kobj);
1386
1387         remove_queue_kobjects(net);
1388
1389         device_del(dev);
1390 }
1391
1392 /* Create sysfs entries for network device. */
1393 int netdev_register_kobject(struct net_device *net)
1394 {
1395         struct device *dev = &(net->dev);
1396         const struct attribute_group **groups = net->sysfs_groups;
1397         int error = 0;
1398
1399         device_initialize(dev);
1400         dev->class = &net_class;
1401         dev->platform_data = net;
1402         dev->groups = groups;
1403
1404         dev_set_name(dev, "%s", net->name);
1405
1406 #ifdef CONFIG_SYSFS
1407         /* Allow for a device specific group */
1408         if (*groups)
1409                 groups++;
1410
1411         *groups++ = &netstat_group;
1412 #endif /* CONFIG_SYSFS */
1413
1414         error = device_add(dev);
1415         if (error)
1416                 return error;
1417
1418         error = register_queue_kobjects(net);
1419         if (error) {
1420                 device_del(dev);
1421                 return error;
1422         }
1423
1424         return error;
1425 }
1426
1427 int netdev_class_create_file(struct class_attribute *class_attr)
1428 {
1429         return class_create_file(&net_class, class_attr);
1430 }
1431 EXPORT_SYMBOL(netdev_class_create_file);
1432
1433 void netdev_class_remove_file(struct class_attribute *class_attr)
1434 {
1435         class_remove_file(&net_class, class_attr);
1436 }
1437 EXPORT_SYMBOL(netdev_class_remove_file);
1438
1439 int netdev_kobject_init(void)
1440 {
1441         kobj_ns_type_register(&net_ns_type_operations);
1442         return class_register(&net_class);
1443 }