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
 433 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 434 static struct attribute *wireless_attrs[] = {
 435         NULL
 436 };
 437
 438 static struct attribute_group wireless_group = {
 439         .name = "wireless",
 440         .attrs = wireless_attrs,
 441 };
 442 #endif
 443 #endif /* CONFIG_SYSFS */
 444
 445 #ifdef CONFIG_RPS
 446 /*
 447  * RX queue sysfs structures and functions.
 448  */
 449 struct rx_queue_attribute {
 450         struct attribute attr;
 451         ssize_t (*show)(struct netdev_rx_queue *queue,
 452             struct rx_queue_attribute *attr, char *buf);
 453         ssize_t (*store)(struct netdev_rx_queue *queue,
 454             struct rx_queue_attribute *attr, const char *buf, size_t len);
 455 };
 456 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 457     struct rx_queue_attribute, attr)
 458
 459 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 460
 461 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 462                                   char *buf)
 463 {
 464         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 465         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 466
 467         if (!attribute->show)
 468                 return -EIO;
 469
 470         return attribute->show(queue, attribute, buf);
 471 }
 472
 473 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 474                                    const char *buf, size_t count)
 475 {
 476         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 477         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 478
 479         if (!attribute->store)
 480                 return -EIO;
 481
 482         return attribute->store(queue, attribute, buf, count);
 483 }
 484
 485 static const struct sysfs_ops rx_queue_sysfs_ops = {
 486         .show = rx_queue_attr_show,
 487         .store = rx_queue_attr_store,
 488 };
 489
 490 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 491                             struct rx_queue_attribute *attribute, char *buf)
 492 {
 493         struct rps_map *map;
 494         cpumask_var_t mask;
 495         size_t len = 0;
 496         int i;
 497
 498         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 499                 return -ENOMEM;
 500
 501         rcu_read_lock();
 502         map = rcu_dereference(queue->rps_map);
 503         if (map)
 504                 for (i = 0; i < map->len; i++)
 505                         cpumask_set_cpu(map->cpus[i], mask);
 506
 507         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 508         if (PAGE_SIZE - len < 3) {
 509                 rcu_read_unlock();
 510                 free_cpumask_var(mask);
 511                 return -EINVAL;
 512         }
 513         rcu_read_unlock();
 514
 515         free_cpumask_var(mask);
 516         len += sprintf(buf + len, "\n");
 517         return len;
 518 }
 519
 520 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 521                       struct rx_queue_attribute *attribute,
 522                       const char *buf, size_t len)
 523 {
 524         struct rps_map *old_map, *map;
 525         cpumask_var_t mask;
 526         int err, cpu, i;
 527         static DEFINE_SPINLOCK(rps_map_lock);
 528
 529         if (!capable(CAP_NET_ADMIN))
 530                 return -EPERM;
 531
 532         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 533                 return -ENOMEM;
 534
 535         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 536         if (err) {
 537                 free_cpumask_var(mask);
 538                 return err;
 539         }
 540
 541         map = kzalloc(max_t(unsigned int,
 542             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 543             GFP_KERNEL);
 544         if (!map) {
 545                 free_cpumask_var(mask);
 546                 return -ENOMEM;
 547         }
 548
 549         i = 0;
 550         for_each_cpu_and(cpu, mask, cpu_online_mask)
 551                 map->cpus[i++] = cpu;
 552
 553         if (i)
 554                 map->len = i;
 555         else {
 556                 kfree(map);
 557                 map = NULL;
 558         }
 559
 560         spin_lock(&rps_map_lock);
 561         old_map = rcu_dereference_protected(queue->rps_map,
 562                                             lockdep_is_held(&rps_map_lock));
 563         rcu_assign_pointer(queue->rps_map, map);
 564         spin_unlock(&rps_map_lock);
 565
 566         if (map)
 567                 static_key_slow_inc(&rps_needed);
 568         if (old_map) {
 569                 kfree_rcu(old_map, rcu);
 570                 static_key_slow_dec(&rps_needed);
 571         }
 572         free_cpumask_var(mask);
 573         return len;
 574 }
 575
 576 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 577                                            struct rx_queue_attribute *attr,
 578                                            char *buf)
 579 {
 580         struct rps_dev_flow_table *flow_table;
 581         unsigned long val = 0;
 582
 583         rcu_read_lock();
 584         flow_table = rcu_dereference(queue->rps_flow_table);
 585         if (flow_table)
 586                 val = (unsigned long)flow_table->mask + 1;
 587         rcu_read_unlock();
 588
 589         return sprintf(buf, "%lu\n", val);
 590 }
 591
 592 static void rps_dev_flow_table_release_work(struct work_struct *work)
 593 {
 594         struct rps_dev_flow_table *table = container_of(work,
 595             struct rps_dev_flow_table, free_work);
 596
 597         vfree(table);
 598 }
 599
 600 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 601 {
 602         struct rps_dev_flow_table *table = container_of(rcu,
 603             struct rps_dev_flow_table, rcu);
 604
 605         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 606         schedule_work(&table->free_work);
 607 }
 608
 609 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 610                                      struct rx_queue_attribute *attr,
 611                                      const char *buf, size_t len)
 612 {
 613         unsigned long mask, count;
 614         struct rps_dev_flow_table *table, *old_table;
 615         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 616         int rc;
 617
 618         if (!capable(CAP_NET_ADMIN))
 619                 return -EPERM;
 620
 621         rc = kstrtoul(buf, 0, &count);
 622         if (rc < 0)
 623                 return rc;
 624
 625         if (count) {
 626                 mask = count - 1;
 627                 /* mask = roundup_pow_of_two(count) - 1;
 628                  * without overflows...
 629                  */
 630                 while ((mask | (mask >> 1)) != mask)
 631                         mask |= (mask >> 1);
 632                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 633                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 634                  * doesnt overflow.
 635                  */
 636 #if BITS_PER_LONG > 32
 637                 if (mask > (unsigned long)(u32)mask)
 638                         return -EINVAL;
 639 #else
 640                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 641                                 / sizeof(struct rps_dev_flow)) {
 642                         /* Enforce a limit to prevent overflow */
 643                         return -EINVAL;
 644                 }
 645 #endif
 646                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 647                 if (!table)
 648                         return -ENOMEM;
 649
 650                 table->mask = mask;
 651                 for (count = 0; count <= mask; count++)
 652                         table->flows[count].cpu = RPS_NO_CPU;
 653         } else
 654                 table = NULL;
 655
 656         spin_lock(&rps_dev_flow_lock);
 657         old_table = rcu_dereference_protected(queue->rps_flow_table,
 658                                               lockdep_is_held(&rps_dev_flow_lock));
 659         rcu_assign_pointer(queue->rps_flow_table, table);
 660         spin_unlock(&rps_dev_flow_lock);
 661
 662         if (old_table)
 663                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 664
 665         return len;
 666 }
 667
 668 static struct rx_queue_attribute rps_cpus_attribute =
 669         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 670
 671
 672 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 673         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 674             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 675
 676 static struct attribute *rx_queue_default_attrs[] = {
 677         &rps_cpus_attribute.attr,
 678         &rps_dev_flow_table_cnt_attribute.attr,
 679         NULL
 680 };
 681
 682 static void rx_queue_release(struct kobject *kobj)
 683 {
 684         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 685         struct rps_map *map;
 686         struct rps_dev_flow_table *flow_table;
 687
 688
 689         map = rcu_dereference_protected(queue->rps_map, 1);
 690         if (map) {
 691                 RCU_INIT_POINTER(queue->rps_map, NULL);
 692                 kfree_rcu(map, rcu);
 693         }
 694
 695         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 696         if (flow_table) {
 697                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 698                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 699         }
 700
 701         memset(kobj, 0, sizeof(*kobj));
 702         dev_put(queue->dev);
 703 }
 704
 705 static struct kobj_type rx_queue_ktype = {
 706         .sysfs_ops = &rx_queue_sysfs_ops,
 707         .release = rx_queue_release,
 708         .default_attrs = rx_queue_default_attrs,
 709 };
 710
 711 static int rx_queue_add_kobject(struct net_device *net, int index)
 712 {
 713         struct netdev_rx_queue *queue = net->_rx + index;
 714         struct kobject *kobj = &queue->kobj;
 715         int error = 0;
 716
 717         kobj->kset = net->queues_kset;
 718         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 719             "rx-%u", index);
 720         if (error) {
 721                 kobject_put(kobj);
 722                 return error;
 723         }
 724
 725         kobject_uevent(kobj, KOBJ_ADD);
 726         dev_hold(queue->dev);
 727
 728         return error;
 729 }
 730 #endif /* CONFIG_RPS */
 731
 732 int
 733 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 734 {
 735 #ifdef CONFIG_RPS
 736         int i;
 737         int error = 0;
 738
 739         for (i = old_num; i < new_num; i++) {
 740                 error = rx_queue_add_kobject(net, i);
 741                 if (error) {
 742                         new_num = old_num;
 743                         break;
 744                 }
 745         }
 746
 747         while (--i >= new_num)
 748                 kobject_put(&net->_rx[i].kobj);
 749
 750         return error;
 751 #else
 752         return 0;
 753 #endif
 754 }
 755
 756 #ifdef CONFIG_SYSFS
 757 /*
 758  * netdev_queue sysfs structures and functions.
 759  */
 760 struct netdev_queue_attribute {
 761         struct attribute attr;
 762         ssize_t (*show)(struct netdev_queue *queue,
 763             struct netdev_queue_attribute *attr, char *buf);
 764         ssize_t (*store)(struct netdev_queue *queue,
 765             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 766 };
 767 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 768     struct netdev_queue_attribute, attr)
 769
 770 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 771
 772 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 773                                       struct attribute *attr, char *buf)
 774 {
 775         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 776         struct netdev_queue *queue = to_netdev_queue(kobj);
 777
 778         if (!attribute->show)
 779                 return -EIO;
 780
 781         return attribute->show(queue, attribute, buf);
 782 }
 783
 784 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 785                                        struct attribute *attr,
 786                                        const char *buf, size_t count)
 787 {
 788         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 789         struct netdev_queue *queue = to_netdev_queue(kobj);
 790
 791         if (!attribute->store)
 792                 return -EIO;
 793
 794         return attribute->store(queue, attribute, buf, count);
 795 }
 796
 797 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 798         .show = netdev_queue_attr_show,
 799         .store = netdev_queue_attr_store,
 800 };
 801
 802 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 803                                   struct netdev_queue_attribute *attribute,
 804                                   char *buf)
 805 {
 806         unsigned long trans_timeout;
 807
 808         spin_lock_irq(&queue->_xmit_lock);
 809         trans_timeout = queue->trans_timeout;
 810         spin_unlock_irq(&queue->_xmit_lock);
 811
 812         return sprintf(buf, "%lu", trans_timeout);
 813 }
 814
 815 static struct netdev_queue_attribute queue_trans_timeout =
 816         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 817
 818 #ifdef CONFIG_BQL
 819 /*
 820  * Byte queue limits sysfs structures and functions.
 821  */
 822 static ssize_t bql_show(char *buf, unsigned int value)
 823 {
 824         return sprintf(buf, "%u\n", value);
 825 }
 826
 827 static ssize_t bql_set(const char *buf, const size_t count,
 828                        unsigned int *pvalue)
 829 {
 830         unsigned int value;
 831         int err;
 832
 833         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 834                 value = DQL_MAX_LIMIT;
 835         else {
 836                 err = kstrtouint(buf, 10, &value);
 837                 if (err < 0)
 838                         return err;
 839                 if (value > DQL_MAX_LIMIT)
 840                         return -EINVAL;
 841         }
 842
 843         *pvalue = value;
 844
 845         return count;
 846 }
 847
 848 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 849                                   struct netdev_queue_attribute *attr,
 850                                   char *buf)
 851 {
 852         struct dql *dql = &queue->dql;
 853
 854         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 855 }
 856
 857 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 858                                  struct netdev_queue_attribute *attribute,
 859                                  const char *buf, size_t len)
 860 {
 861         struct dql *dql = &queue->dql;
 862         unsigned int value;
 863         int err;
 864
 865         err = kstrtouint(buf, 10, &value);
 866         if (err < 0)
 867                 return err;
 868
 869         dql->slack_hold_time = msecs_to_jiffies(value);
 870
 871         return len;
 872 }
 873
 874 static struct netdev_queue_attribute bql_hold_time_attribute =
 875         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 876             bql_set_hold_time);
 877
 878 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 879                                  struct netdev_queue_attribute *attr,
 880                                  char *buf)
 881 {
 882         struct dql *dql = &queue->dql;
 883
 884         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 885 }
 886
 887 static struct netdev_queue_attribute bql_inflight_attribute =
 888         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 889
 890 #define BQL_ATTR(NAME, FIELD)                                           \
 891 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 892                                  struct netdev_queue_attribute *attr,   \
 893                                  char *buf)                             \
 894 {                                                                       \
 895         return bql_show(buf, queue->dql.FIELD);                         \
 896 }                                                                       \
 897                                                                         \
 898 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 899                                 struct netdev_queue_attribute *attr,    \
 900                                 const char *buf, size_t len)            \
 901 {                                                                       \
 902         return bql_set(buf, len, &queue->dql.FIELD);                    \
 903 }                                                                       \
 904                                                                         \
 905 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 906         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 907             bql_set_ ## NAME);
 908
 909 BQL_ATTR(limit, limit)
 910 BQL_ATTR(limit_max, max_limit)
 911 BQL_ATTR(limit_min, min_limit)
 912
 913 static struct attribute *dql_attrs[] = {
 914         &bql_limit_attribute.attr,
 915         &bql_limit_max_attribute.attr,
 916         &bql_limit_min_attribute.attr,
 917         &bql_hold_time_attribute.attr,
 918         &bql_inflight_attribute.attr,
 919         NULL
 920 };
 921
 922 static struct attribute_group dql_group = {
 923         .name  = "byte_queue_limits",
 924         .attrs  = dql_attrs,
 925 };
 926 #endif /* CONFIG_BQL */
 927
 928 #ifdef CONFIG_XPS
 929 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 930 {
 931         struct net_device *dev = queue->dev;
 932         int i;
 933
 934         for (i = 0; i < dev->num_tx_queues; i++)
 935                 if (queue == &dev->_tx[i])
 936                         break;
 937
 938         BUG_ON(i >= dev->num_tx_queues);
 939
 940         return i;
 941 }
 942
 943
 944 static ssize_t show_xps_map(struct netdev_queue *queue,
 945                             struct netdev_queue_attribute *attribute, char *buf)
 946 {
 947         struct net_device *dev = queue->dev;
 948         struct xps_dev_maps *dev_maps;
 949         cpumask_var_t mask;
 950         unsigned long index;
 951         size_t len = 0;
 952         int i;
 953
 954         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 955                 return -ENOMEM;
 956
 957         index = get_netdev_queue_index(queue);
 958
 959         rcu_read_lock();
 960         dev_maps = rcu_dereference(dev->xps_maps);
 961         if (dev_maps) {
 962                 for_each_possible_cpu(i) {
 963                         struct xps_map *map =
 964                             rcu_dereference(dev_maps->cpu_map[i]);
 965                         if (map) {
 966                                 int j;
 967                                 for (j = 0; j < map->len; j++) {
 968                                         if (map->queues[j] == index) {
 969                                                 cpumask_set_cpu(i, mask);
 970                                                 break;
 971                                         }
 972                                 }
 973                         }
 974                 }
 975         }
 976         rcu_read_unlock();
 977
 978         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 979         if (PAGE_SIZE - len < 3) {
 980                 free_cpumask_var(mask);
 981                 return -EINVAL;
 982         }
 983
 984         free_cpumask_var(mask);
 985         len += sprintf(buf + len, "\n");
 986         return len;
 987 }
 988
 989 static DEFINE_MUTEX(xps_map_mutex);
 990 #define xmap_dereference(P)             \
 991         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
 992
 993 static void xps_queue_release(struct netdev_queue *queue)
 994 {
 995         struct net_device *dev = queue->dev;
 996         struct xps_dev_maps *dev_maps;
 997         struct xps_map *map;
 998         unsigned long index;
 999         int i, pos, nonempty = 0;
1000
1001         index = get_netdev_queue_index(queue);
1002
1003         mutex_lock(&xps_map_mutex);
1004         dev_maps = xmap_dereference(dev->xps_maps);
1005
1006         if (dev_maps) {
1007                 for_each_possible_cpu(i) {
1008                         map = xmap_dereference(dev_maps->cpu_map[i]);
1009                         if (!map)
1010                                 continue;
1011
1012                         for (pos = 0; pos < map->len; pos++)
1013                                 if (map->queues[pos] == index)
1014                                         break;
1015
1016                         if (pos < map->len) {
1017                                 if (map->len > 1)
1018                                         map->queues[pos] =
1019                                             map->queues[--map->len];
1020                                 else {
1021                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1022                                             NULL);
1023                                         kfree_rcu(map, rcu);
1024                                         map = NULL;
1025                                 }
1026                         }
1027                         if (map)
1028                                 nonempty = 1;
1029                 }
1030
1031                 if (!nonempty) {
1032                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1033                         kfree_rcu(dev_maps, rcu);
1034                 }
1035         }
1036         mutex_unlock(&xps_map_mutex);
1037 }
1038
1039 static ssize_t store_xps_map(struct netdev_queue *queue,
1040                       struct netdev_queue_attribute *attribute,
1041                       const char *buf, size_t len)
1042 {
1043         struct net_device *dev = queue->dev;
1044         cpumask_var_t mask;
1045         int err, i, cpu, pos, map_len, alloc_len, need_set;
1046         unsigned long index;
1047         struct xps_map *map, *new_map;
1048         struct xps_dev_maps *dev_maps, *new_dev_maps;
1049         int nonempty = 0;
1050         int numa_node_id = -2;
1051
1052         if (!capable(CAP_NET_ADMIN))
1053                 return -EPERM;
1054
1055         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1056                 return -ENOMEM;
1057
1058         index = get_netdev_queue_index(queue);
1059
1060         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1061         if (err) {
1062                 free_cpumask_var(mask);
1063                 return err;
1064         }
1065
1066         new_dev_maps = kzalloc(max_t(unsigned int,
1067             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1068         if (!new_dev_maps) {
1069                 free_cpumask_var(mask);
1070                 return -ENOMEM;
1071         }
1072
1073         mutex_lock(&xps_map_mutex);
1074
1075         dev_maps = xmap_dereference(dev->xps_maps);
1076
1077         for_each_possible_cpu(cpu) {
1078                 map = dev_maps ?
1079                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1080                 new_map = map;
1081                 if (map) {
1082                         for (pos = 0; pos < map->len; pos++)
1083                                 if (map->queues[pos] == index)
1084                                         break;
1085                         map_len = map->len;
1086                         alloc_len = map->alloc_len;
1087                 } else
1088                         pos = map_len = alloc_len = 0;
1089
1090                 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1091 #ifdef CONFIG_NUMA
1092                 if (need_set) {
1093                         if (numa_node_id == -2)
1094                                 numa_node_id = cpu_to_node(cpu);
1095                         else if (numa_node_id != cpu_to_node(cpu))
1096                                 numa_node_id = -1;
1097                 }
1098 #endif
1099                 if (need_set && pos >= map_len) {
1100                         /* Need to add queue to this CPU's map */
1101                         if (map_len >= alloc_len) {
1102                                 alloc_len = alloc_len ?
1103                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
1104                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1105                                                        GFP_KERNEL,
1106                                                        cpu_to_node(cpu));
1107                                 if (!new_map)
1108                                         goto error;
1109                                 new_map->alloc_len = alloc_len;
1110                                 for (i = 0; i < map_len; i++)
1111                                         new_map->queues[i] = map->queues[i];
1112                                 new_map->len = map_len;
1113                         }
1114                         new_map->queues[new_map->len++] = index;
1115                 } else if (!need_set && pos < map_len) {
1116                         /* Need to remove queue from this CPU's map */
1117                         if (map_len > 1)
1118                                 new_map->queues[pos] =
1119                                     new_map->queues[--new_map->len];
1120                         else
1121                                 new_map = NULL;
1122                 }
1123                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1124         }
1125
1126         /* Cleanup old maps */
1127         for_each_possible_cpu(cpu) {
1128                 map = dev_maps ?
1129                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1130                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1131                         kfree_rcu(map, rcu);
1132                 if (new_dev_maps->cpu_map[cpu])
1133                         nonempty = 1;
1134         }
1135
1136         if (nonempty) {
1137                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1138         } else {
1139                 kfree(new_dev_maps);
1140                 RCU_INIT_POINTER(dev->xps_maps, NULL);
1141         }
1142
1143         if (dev_maps)
1144                 kfree_rcu(dev_maps, rcu);
1145
1146         netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1147                                             NUMA_NO_NODE);
1148
1149         mutex_unlock(&xps_map_mutex);
1150
1151         free_cpumask_var(mask);
1152         return len;
1153
1154 error:
1155         mutex_unlock(&xps_map_mutex);
1156
1157         if (new_dev_maps)
1158                 for_each_possible_cpu(i)
1159                         kfree(rcu_dereference_protected(
1160                                 new_dev_maps->cpu_map[i],
1161                                 1));
1162         kfree(new_dev_maps);
1163         free_cpumask_var(mask);
1164         return -ENOMEM;
1165 }
1166
1167 static struct netdev_queue_attribute xps_cpus_attribute =
1168     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1169 #endif /* CONFIG_XPS */
1170
1171 static struct attribute *netdev_queue_default_attrs[] = {
1172         &queue_trans_timeout.attr,
1173 #ifdef CONFIG_XPS
1174         &xps_cpus_attribute.attr,
1175 #endif
1176         NULL
1177 };
1178
1179 static void netdev_queue_release(struct kobject *kobj)
1180 {
1181         struct netdev_queue *queue = to_netdev_queue(kobj);
1182
1183 #ifdef CONFIG_XPS
1184         xps_queue_release(queue);
1185 #endif
1186
1187         memset(kobj, 0, sizeof(*kobj));
1188         dev_put(queue->dev);
1189 }
1190
1191 static struct kobj_type netdev_queue_ktype = {
1192         .sysfs_ops = &netdev_queue_sysfs_ops,
1193         .release = netdev_queue_release,
1194         .default_attrs = netdev_queue_default_attrs,
1195 };
1196
1197 static int netdev_queue_add_kobject(struct net_device *net, int index)
1198 {
1199         struct netdev_queue *queue = net->_tx + index;
1200         struct kobject *kobj = &queue->kobj;
1201         int error = 0;
1202
1203         kobj->kset = net->queues_kset;
1204         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1205             "tx-%u", index);
1206         if (error)
1207                 goto exit;
1208
1209 #ifdef CONFIG_BQL
1210         error = sysfs_create_group(kobj, &dql_group);
1211         if (error)
1212                 goto exit;
1213 #endif
1214
1215         kobject_uevent(kobj, KOBJ_ADD);
1216         dev_hold(queue->dev);
1217
1218         return 0;
1219 exit:
1220         kobject_put(kobj);
1221         return error;
1222 }
1223 #endif /* CONFIG_SYSFS */
1224
1225 int
1226 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1227 {
1228 #ifdef CONFIG_SYSFS
1229         int i;
1230         int error = 0;
1231
1232         for (i = old_num; i < new_num; i++) {
1233                 error = netdev_queue_add_kobject(net, i);
1234                 if (error) {
1235                         new_num = old_num;
1236                         break;
1237                 }
1238         }
1239
1240         while (--i >= new_num) {
1241                 struct netdev_queue *queue = net->_tx + i;
1242
1243 #ifdef CONFIG_BQL
1244                 sysfs_remove_group(&queue->kobj, &dql_group);
1245 #endif
1246                 kobject_put(&queue->kobj);
1247         }
1248
1249         return error;
1250 #else
1251         return 0;
1252 #endif /* CONFIG_SYSFS */
1253 }
1254
1255 static int register_queue_kobjects(struct net_device *net)
1256 {
1257         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1258
1259 #ifdef CONFIG_SYSFS
1260         net->queues_kset = kset_create_and_add("queues",
1261             NULL, &net->dev.kobj);
1262         if (!net->queues_kset)
1263                 return -ENOMEM;
1264 #endif
1265
1266 #ifdef CONFIG_RPS
1267         real_rx = net->real_num_rx_queues;
1268 #endif
1269         real_tx = net->real_num_tx_queues;
1270
1271         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1272         if (error)
1273                 goto error;
1274         rxq = real_rx;
1275
1276         error = netdev_queue_update_kobjects(net, 0, real_tx);
1277         if (error)
1278                 goto error;
1279         txq = real_tx;
1280
1281         return 0;
1282
1283 error:
1284         netdev_queue_update_kobjects(net, txq, 0);
1285         net_rx_queue_update_kobjects(net, rxq, 0);
1286         return error;
1287 }
1288
1289 static void remove_queue_kobjects(struct net_device *net)
1290 {
1291         int real_rx = 0, real_tx = 0;
1292
1293 #ifdef CONFIG_RPS
1294         real_rx = net->real_num_rx_queues;
1295 #endif
1296         real_tx = net->real_num_tx_queues;
1297
1298         net_rx_queue_update_kobjects(net, real_rx, 0);
1299         netdev_queue_update_kobjects(net, real_tx, 0);
1300 #ifdef CONFIG_SYSFS
1301         kset_unregister(net->queues_kset);
1302 #endif
1303 }
1304
1305 static void *net_grab_current_ns(void)
1306 {
1307         struct net *ns = current->nsproxy->net_ns;
1308 #ifdef CONFIG_NET_NS
1309         if (ns)
1310                 atomic_inc(&ns->passive);
1311 #endif
1312         return ns;
1313 }
1314
1315 static const void *net_initial_ns(void)
1316 {
1317         return &init_net;
1318 }
1319
1320 static const void *net_netlink_ns(struct sock *sk)
1321 {
1322         return sock_net(sk);
1323 }
1324
1325 struct kobj_ns_type_operations net_ns_type_operations = {
1326         .type = KOBJ_NS_TYPE_NET,
1327         .grab_current_ns = net_grab_current_ns,
1328         .netlink_ns = net_netlink_ns,
1329         .initial_ns = net_initial_ns,
1330         .drop_ns = net_drop_ns,
1331 };
1332 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1333
1334 #ifdef CONFIG_HOTPLUG
1335 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1336 {
1337         struct net_device *dev = to_net_dev(d);
1338         int retval;
1339
1340         /* pass interface to uevent. */
1341         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1342         if (retval)
1343                 goto exit;
1344
1345         /* pass ifindex to uevent.
1346          * ifindex is useful as it won't change (interface name may change)
1347          * and is what RtNetlink uses natively. */
1348         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1349
1350 exit:
1351         return retval;
1352 }
1353 #endif
1354
1355 /*
1356  *      netdev_release -- destroy and free a dead device.
1357  *      Called when last reference to device kobject is gone.
1358  */
1359 static void netdev_release(struct device *d)
1360 {
1361         struct net_device *dev = to_net_dev(d);
1362
1363         BUG_ON(dev->reg_state != NETREG_RELEASED);
1364
1365         kfree(dev->ifalias);
1366         kfree((char *)dev - dev->padded);
1367 }
1368
1369 static const void *net_namespace(struct device *d)
1370 {
1371         struct net_device *dev;
1372         dev = container_of(d, struct net_device, dev);
1373         return dev_net(dev);
1374 }
1375
1376 static struct class net_class = {
1377         .name = "net",
1378         .dev_release = netdev_release,
1379 #ifdef CONFIG_SYSFS
1380         .dev_attrs = net_class_attributes,
1381 #endif /* CONFIG_SYSFS */
1382 #ifdef CONFIG_HOTPLUG
1383         .dev_uevent = netdev_uevent,
1384 #endif
1385         .ns_type = &net_ns_type_operations,
1386         .namespace = net_namespace,
1387 };
1388
1389 /* Delete sysfs entries but hold kobject reference until after all
1390  * netdev references are gone.
1391  */
1392 void netdev_unregister_kobject(struct net_device * net)
1393 {
1394         struct device *dev = &(net->dev);
1395
1396         kobject_get(&dev->kobj);
1397
1398         remove_queue_kobjects(net);
1399
1400         device_del(dev);
1401 }
1402
1403 /* Create sysfs entries for network device. */
1404 int netdev_register_kobject(struct net_device *net)
1405 {
1406         struct device *dev = &(net->dev);
1407         const struct attribute_group **groups = net->sysfs_groups;
1408         int error = 0;
1409
1410         device_initialize(dev);
1411         dev->class = &net_class;
1412         dev->platform_data = net;
1413         dev->groups = groups;
1414
1415         dev_set_name(dev, "%s", net->name);
1416
1417 #ifdef CONFIG_SYSFS
1418         /* Allow for a device specific group */
1419         if (*groups)
1420                 groups++;
1421
1422         *groups++ = &netstat_group;
1423
1424 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1425         if (net->ieee80211_ptr)
1426                 *groups++ = &wireless_group;
1427 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1428         else if (net->wireless_handlers)
1429                 *groups++ = &wireless_group;
1430 #endif
1431 #endif
1432 #endif /* CONFIG_SYSFS */
1433
1434         error = device_add(dev);
1435         if (error)
1436                 return error;
1437
1438         error = register_queue_kobjects(net);
1439         if (error) {
1440                 device_del(dev);
1441                 return error;
1442         }
1443
1444         return error;
1445 }
1446
1447 int netdev_class_create_file(struct class_attribute *class_attr)
1448 {
1449         return class_create_file(&net_class, class_attr);
1450 }
1451 EXPORT_SYMBOL(netdev_class_create_file);
1452
1453 void netdev_class_remove_file(struct class_attribute *class_attr)
1454 {
1455         class_remove_file(&net_class, class_attr);
1456 }
1457 EXPORT_SYMBOL(netdev_class_remove_file);
1458
1459 int netdev_kobject_init(void)
1460 {
1461         kobj_ns_type_register(&net_ns_type_operations);
1462         return class_register(&net_class);
1463 }