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