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         char *endp;
  78         unsigned long new;
  79         int ret = -EINVAL;
  80
  81         if (!capable(CAP_NET_ADMIN))
  82                 return -EPERM;
  83
  84         new = simple_strtoul(buf, &endp, 0);
  85         if (endp == buf)
  86                 goto err;
  87
  88         if (!rtnl_trylock())
  89                 return restart_syscall();
  90
  91         if (dev_isalive(net)) {
  92                 if ((ret = (*set)(net, new)) == 0)
  93                         ret = len;
  94         }
  95         rtnl_unlock();
  96  err:
  97         return ret;
  98 }
  99
 100 NETDEVICE_SHOW(dev_id, fmt_hex);
 101 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 102 NETDEVICE_SHOW(addr_len, fmt_dec);
 103 NETDEVICE_SHOW(iflink, fmt_dec);
 104 NETDEVICE_SHOW(ifindex, fmt_dec);
 105 NETDEVICE_SHOW(type, fmt_dec);
 106 NETDEVICE_SHOW(link_mode, fmt_dec);
 107
 108 /* use same locking rules as GIFHWADDR ioctl's */
 109 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 110                             char *buf)
 111 {
 112         struct net_device *net = to_net_dev(dev);
 113         ssize_t ret = -EINVAL;
 114
 115         read_lock(&dev_base_lock);
 116         if (dev_isalive(net))
 117                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 118         read_unlock(&dev_base_lock);
 119         return ret;
 120 }
 121
 122 static ssize_t show_broadcast(struct device *dev,
 123                             struct device_attribute *attr, char *buf)
 124 {
 125         struct net_device *net = to_net_dev(dev);
 126         if (dev_isalive(net))
 127                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 128         return -EINVAL;
 129 }
 130
 131 static ssize_t show_carrier(struct device *dev,
 132                             struct device_attribute *attr, char *buf)
 133 {
 134         struct net_device *netdev = to_net_dev(dev);
 135         if (netif_running(netdev)) {
 136                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 137         }
 138         return -EINVAL;
 139 }
 140
 141 static ssize_t show_speed(struct device *dev,
 142                           struct device_attribute *attr, char *buf)
 143 {
 144         struct net_device *netdev = to_net_dev(dev);
 145         int ret = -EINVAL;
 146
 147         if (!rtnl_trylock())
 148                 return restart_syscall();
 149
 150         if (netif_running(netdev)) {
 151                 struct ethtool_cmd cmd;
 152                 if (!__ethtool_get_settings(netdev, &cmd))
 153                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 154         }
 155         rtnl_unlock();
 156         return ret;
 157 }
 158
 159 static ssize_t show_duplex(struct device *dev,
 160                            struct device_attribute *attr, char *buf)
 161 {
 162         struct net_device *netdev = to_net_dev(dev);
 163         int ret = -EINVAL;
 164
 165         if (!rtnl_trylock())
 166                 return restart_syscall();
 167
 168         if (netif_running(netdev)) {
 169                 struct ethtool_cmd cmd;
 170                 if (!__ethtool_get_settings(netdev, &cmd))
 171                         ret = sprintf(buf, "%s\n",
 172                                       cmd.duplex ? "full" : "half");
 173         }
 174         rtnl_unlock();
 175         return ret;
 176 }
 177
 178 static ssize_t show_dormant(struct device *dev,
 179                             struct device_attribute *attr, char *buf)
 180 {
 181         struct net_device *netdev = to_net_dev(dev);
 182
 183         if (netif_running(netdev))
 184                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 185
 186         return -EINVAL;
 187 }
 188
 189 static const char *const operstates[] = {
 190         "unknown",
 191         "notpresent", /* currently unused */
 192         "down",
 193         "lowerlayerdown",
 194         "testing", /* currently unused */
 195         "dormant",
 196         "up"
 197 };
 198
 199 static ssize_t show_operstate(struct device *dev,
 200                               struct device_attribute *attr, char *buf)
 201 {
 202         const struct net_device *netdev = to_net_dev(dev);
 203         unsigned char operstate;
 204
 205         read_lock(&dev_base_lock);
 206         operstate = netdev->operstate;
 207         if (!netif_running(netdev))
 208                 operstate = IF_OPER_DOWN;
 209         read_unlock(&dev_base_lock);
 210
 211         if (operstate >= ARRAY_SIZE(operstates))
 212                 return -EINVAL; /* should not happen */
 213
 214         return sprintf(buf, "%s\n", operstates[operstate]);
 215 }
 216
 217 /* read-write attributes */
 218 NETDEVICE_SHOW(mtu, fmt_dec);
 219
 220 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 221 {
 222         return dev_set_mtu(net, (int) new_mtu);
 223 }
 224
 225 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 226                          const char *buf, size_t len)
 227 {
 228         return netdev_store(dev, attr, buf, len, change_mtu);
 229 }
 230
 231 NETDEVICE_SHOW(flags, fmt_hex);
 232
 233 static int change_flags(struct net_device *net, unsigned long new_flags)
 234 {
 235         return dev_change_flags(net, (unsigned) new_flags);
 236 }
 237
 238 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 239                            const char *buf, size_t len)
 240 {
 241         return netdev_store(dev, attr, buf, len, change_flags);
 242 }
 243
 244 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 245
 246 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 247 {
 248         net->tx_queue_len = new_len;
 249         return 0;
 250 }
 251
 252 static ssize_t store_tx_queue_len(struct device *dev,
 253                                   struct device_attribute *attr,
 254                                   const char *buf, size_t len)
 255 {
 256         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 257 }
 258
 259 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 260                              const char *buf, size_t len)
 261 {
 262         struct net_device *netdev = to_net_dev(dev);
 263         size_t count = len;
 264         ssize_t ret;
 265
 266         if (!capable(CAP_NET_ADMIN))
 267                 return -EPERM;
 268
 269         /* ignore trailing newline */
 270         if (len >  0 && buf[len - 1] == '\n')
 271                 --count;
 272
 273         if (!rtnl_trylock())
 274                 return restart_syscall();
 275         ret = dev_set_alias(netdev, buf, count);
 276         rtnl_unlock();
 277
 278         return ret < 0 ? ret : len;
 279 }
 280
 281 static ssize_t show_ifalias(struct device *dev,
 282                             struct device_attribute *attr, char *buf)
 283 {
 284         const struct net_device *netdev = to_net_dev(dev);
 285         ssize_t ret = 0;
 286
 287         if (!rtnl_trylock())
 288                 return restart_syscall();
 289         if (netdev->ifalias)
 290                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 291         rtnl_unlock();
 292         return ret;
 293 }
 294
 295 NETDEVICE_SHOW(group, fmt_dec);
 296
 297 static int change_group(struct net_device *net, unsigned long new_group)
 298 {
 299         dev_set_group(net, (int) new_group);
 300         return 0;
 301 }
 302
 303 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 304                          const char *buf, size_t len)
 305 {
 306         return netdev_store(dev, attr, buf, len, change_group);
 307 }
 308
 309 static struct device_attribute net_class_attributes[] = {
 310         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 311         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 312         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 313         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 314         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 315         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 316         __ATTR(type, S_IRUGO, show_type, NULL),
 317         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 318         __ATTR(address, S_IRUGO, show_address, NULL),
 319         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 320         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
 321         __ATTR(speed, S_IRUGO, show_speed, NULL),
 322         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 323         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 324         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 325         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 326         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 327         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 328                store_tx_queue_len),
 329         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 330         {}
 331 };
 332
 333 /* Show a given an attribute in the statistics group */
 334 static ssize_t netstat_show(const struct device *d,
 335                             struct device_attribute *attr, char *buf,
 336                             unsigned long offset)
 337 {
 338         struct net_device *dev = to_net_dev(d);
 339         ssize_t ret = -EINVAL;
 340
 341         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 342                         offset % sizeof(u64) != 0);
 343
 344         read_lock(&dev_base_lock);
 345         if (dev_isalive(dev)) {
 346                 struct rtnl_link_stats64 temp;
 347                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 348
 349                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 350         }
 351         read_unlock(&dev_base_lock);
 352         return ret;
 353 }
 354
 355 /* generate a read-only statistics attribute */
 356 #define NETSTAT_ENTRY(name)                                             \
 357 static ssize_t show_##name(struct device *d,                            \
 358                            struct device_attribute *attr, char *buf)    \
 359 {                                                                       \
 360         return netstat_show(d, attr, buf,                               \
 361                             offsetof(struct rtnl_link_stats64, name));  \
 362 }                                                                       \
 363 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 364
 365 NETSTAT_ENTRY(rx_packets);
 366 NETSTAT_ENTRY(tx_packets);
 367 NETSTAT_ENTRY(rx_bytes);
 368 NETSTAT_ENTRY(tx_bytes);
 369 NETSTAT_ENTRY(rx_errors);
 370 NETSTAT_ENTRY(tx_errors);
 371 NETSTAT_ENTRY(rx_dropped);
 372 NETSTAT_ENTRY(tx_dropped);
 373 NETSTAT_ENTRY(multicast);
 374 NETSTAT_ENTRY(collisions);
 375 NETSTAT_ENTRY(rx_length_errors);
 376 NETSTAT_ENTRY(rx_over_errors);
 377 NETSTAT_ENTRY(rx_crc_errors);
 378 NETSTAT_ENTRY(rx_frame_errors);
 379 NETSTAT_ENTRY(rx_fifo_errors);
 380 NETSTAT_ENTRY(rx_missed_errors);
 381 NETSTAT_ENTRY(tx_aborted_errors);
 382 NETSTAT_ENTRY(tx_carrier_errors);
 383 NETSTAT_ENTRY(tx_fifo_errors);
 384 NETSTAT_ENTRY(tx_heartbeat_errors);
 385 NETSTAT_ENTRY(tx_window_errors);
 386 NETSTAT_ENTRY(rx_compressed);
 387 NETSTAT_ENTRY(tx_compressed);
 388
 389 static struct attribute *netstat_attrs[] = {
 390         &dev_attr_rx_packets.attr,
 391         &dev_attr_tx_packets.attr,
 392         &dev_attr_rx_bytes.attr,
 393         &dev_attr_tx_bytes.attr,
 394         &dev_attr_rx_errors.attr,
 395         &dev_attr_tx_errors.attr,
 396         &dev_attr_rx_dropped.attr,
 397         &dev_attr_tx_dropped.attr,
 398         &dev_attr_multicast.attr,
 399         &dev_attr_collisions.attr,
 400         &dev_attr_rx_length_errors.attr,
 401         &dev_attr_rx_over_errors.attr,
 402         &dev_attr_rx_crc_errors.attr,
 403         &dev_attr_rx_frame_errors.attr,
 404         &dev_attr_rx_fifo_errors.attr,
 405         &dev_attr_rx_missed_errors.attr,
 406         &dev_attr_tx_aborted_errors.attr,
 407         &dev_attr_tx_carrier_errors.attr,
 408         &dev_attr_tx_fifo_errors.attr,
 409         &dev_attr_tx_heartbeat_errors.attr,
 410         &dev_attr_tx_window_errors.attr,
 411         &dev_attr_rx_compressed.attr,
 412         &dev_attr_tx_compressed.attr,
 413         NULL
 414 };
 415
 416
 417 static struct attribute_group netstat_group = {
 418         .name  = "statistics",
 419         .attrs  = netstat_attrs,
 420 };
 421
 422 #ifdef CONFIG_WIRELESS_EXT_SYSFS
 423 /* helper function that does all the locking etc for wireless stats */
 424 static ssize_t wireless_show(struct device *d, char *buf,
 425                              ssize_t (*format)(const struct iw_statistics *,
 426                                                char *))
 427 {
 428         struct net_device *dev = to_net_dev(d);
 429         const struct iw_statistics *iw;
 430         ssize_t ret = -EINVAL;
 431
 432         if (!rtnl_trylock())
 433                 return restart_syscall();
 434         if (dev_isalive(dev)) {
 435                 iw = get_wireless_stats(dev);
 436                 if (iw)
 437                         ret = (*format)(iw, buf);
 438         }
 439         rtnl_unlock();
 440
 441         return ret;
 442 }
 443
 444 /* show function template for wireless fields */
 445 #define WIRELESS_SHOW(name, field, format_string)                       \
 446 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
 447 {                                                                       \
 448         return sprintf(buf, format_string, iw->field);                  \
 449 }                                                                       \
 450 static ssize_t show_iw_##name(struct device *d,                         \
 451                               struct device_attribute *attr, char *buf) \
 452 {                                                                       \
 453         return wireless_show(d, buf, format_iw_##name);                 \
 454 }                                                                       \
 455 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
 456
 457 WIRELESS_SHOW(status, status, fmt_hex);
 458 WIRELESS_SHOW(link, qual.qual, fmt_dec);
 459 WIRELESS_SHOW(level, qual.level, fmt_dec);
 460 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
 461 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
 462 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
 463 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
 464 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
 465 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
 466 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
 467
 468 static struct attribute *wireless_attrs[] = {
 469         &dev_attr_status.attr,
 470         &dev_attr_link.attr,
 471         &dev_attr_level.attr,
 472         &dev_attr_noise.attr,
 473         &dev_attr_nwid.attr,
 474         &dev_attr_crypt.attr,
 475         &dev_attr_fragment.attr,
 476         &dev_attr_retries.attr,
 477         &dev_attr_misc.attr,
 478         &dev_attr_beacon.attr,
 479         NULL
 480 };
 481
 482 static struct attribute_group wireless_group = {
 483         .name = "wireless",
 484         .attrs = wireless_attrs,
 485 };
 486 #endif
 487 #endif /* CONFIG_SYSFS */
 488
 489 #ifdef CONFIG_RPS
 490 /*
 491  * RX queue sysfs structures and functions.
 492  */
 493 struct rx_queue_attribute {
 494         struct attribute attr;
 495         ssize_t (*show)(struct netdev_rx_queue *queue,
 496             struct rx_queue_attribute *attr, char *buf);
 497         ssize_t (*store)(struct netdev_rx_queue *queue,
 498             struct rx_queue_attribute *attr, const char *buf, size_t len);
 499 };
 500 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 501     struct rx_queue_attribute, attr)
 502
 503 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 504
 505 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 506                                   char *buf)
 507 {
 508         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 509         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 510
 511         if (!attribute->show)
 512                 return -EIO;
 513
 514         return attribute->show(queue, attribute, buf);
 515 }
 516
 517 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 518                                    const char *buf, size_t count)
 519 {
 520         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 521         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 522
 523         if (!attribute->store)
 524                 return -EIO;
 525
 526         return attribute->store(queue, attribute, buf, count);
 527 }
 528
 529 static const struct sysfs_ops rx_queue_sysfs_ops = {
 530         .show = rx_queue_attr_show,
 531         .store = rx_queue_attr_store,
 532 };
 533
 534 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 535                             struct rx_queue_attribute *attribute, char *buf)
 536 {
 537         struct rps_map *map;
 538         cpumask_var_t mask;
 539         size_t len = 0;
 540         int i;
 541
 542         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 543                 return -ENOMEM;
 544
 545         rcu_read_lock();
 546         map = rcu_dereference(queue->rps_map);
 547         if (map)
 548                 for (i = 0; i < map->len; i++)
 549                         cpumask_set_cpu(map->cpus[i], mask);
 550
 551         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 552         if (PAGE_SIZE - len < 3) {
 553                 rcu_read_unlock();
 554                 free_cpumask_var(mask);
 555                 return -EINVAL;
 556         }
 557         rcu_read_unlock();
 558
 559         free_cpumask_var(mask);
 560         len += sprintf(buf + len, "\n");
 561         return len;
 562 }
 563
 564 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 565                       struct rx_queue_attribute *attribute,
 566                       const char *buf, size_t len)
 567 {
 568         struct rps_map *old_map, *map;
 569         cpumask_var_t mask;
 570         int err, cpu, i;
 571         static DEFINE_SPINLOCK(rps_map_lock);
 572
 573         if (!capable(CAP_NET_ADMIN))
 574                 return -EPERM;
 575
 576         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 577                 return -ENOMEM;
 578
 579         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 580         if (err) {
 581                 free_cpumask_var(mask);
 582                 return err;
 583         }
 584
 585         map = kzalloc(max_t(unsigned,
 586             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 587             GFP_KERNEL);
 588         if (!map) {
 589                 free_cpumask_var(mask);
 590                 return -ENOMEM;
 591         }
 592
 593         i = 0;
 594         for_each_cpu_and(cpu, mask, cpu_online_mask)
 595                 map->cpus[i++] = cpu;
 596
 597         if (i)
 598                 map->len = i;
 599         else {
 600                 kfree(map);
 601                 map = NULL;
 602         }
 603
 604         spin_lock(&rps_map_lock);
 605         old_map = rcu_dereference_protected(queue->rps_map,
 606                                             lockdep_is_held(&rps_map_lock));
 607         rcu_assign_pointer(queue->rps_map, map);
 608         spin_unlock(&rps_map_lock);
 609
 610         if (map)
 611                 static_key_slow_inc(&rps_needed);
 612         if (old_map) {
 613                 kfree_rcu(old_map, rcu);
 614                 static_key_slow_dec(&rps_needed);
 615         }
 616         free_cpumask_var(mask);
 617         return len;
 618 }
 619
 620 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 621                                            struct rx_queue_attribute *attr,
 622                                            char *buf)
 623 {
 624         struct rps_dev_flow_table *flow_table;
 625         unsigned long val = 0;
 626
 627         rcu_read_lock();
 628         flow_table = rcu_dereference(queue->rps_flow_table);
 629         if (flow_table)
 630                 val = (unsigned long)flow_table->mask + 1;
 631         rcu_read_unlock();
 632
 633         return sprintf(buf, "%lu\n", val);
 634 }
 635
 636 static void rps_dev_flow_table_release_work(struct work_struct *work)
 637 {
 638         struct rps_dev_flow_table *table = container_of(work,
 639             struct rps_dev_flow_table, free_work);
 640
 641         vfree(table);
 642 }
 643
 644 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 645 {
 646         struct rps_dev_flow_table *table = container_of(rcu,
 647             struct rps_dev_flow_table, rcu);
 648
 649         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 650         schedule_work(&table->free_work);
 651 }
 652
 653 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 654                                      struct rx_queue_attribute *attr,
 655                                      const char *buf, size_t len)
 656 {
 657         unsigned long mask, count;
 658         struct rps_dev_flow_table *table, *old_table;
 659         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 660         int rc;
 661
 662         if (!capable(CAP_NET_ADMIN))
 663                 return -EPERM;
 664
 665         rc = kstrtoul(buf, 0, &count);
 666         if (rc < 0)
 667                 return rc;
 668
 669         if (count) {
 670                 mask = count - 1;
 671                 /* mask = roundup_pow_of_two(count) - 1;
 672                  * without overflows...
 673                  */
 674                 while ((mask | (mask >> 1)) != mask)
 675                         mask |= (mask >> 1);
 676                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 677                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 678                  * doesnt overflow.
 679                  */
 680 #if BITS_PER_LONG > 32
 681                 if (mask > (unsigned long)(u32)mask)
 682                         return -EINVAL;
 683 #else
 684                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 685                                 / sizeof(struct rps_dev_flow)) {
 686                         /* Enforce a limit to prevent overflow */
 687                         return -EINVAL;
 688                 }
 689 #endif
 690                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 691                 if (!table)
 692                         return -ENOMEM;
 693
 694                 table->mask = mask;
 695                 for (count = 0; count <= mask; count++)
 696                         table->flows[count].cpu = RPS_NO_CPU;
 697         } else
 698                 table = NULL;
 699
 700         spin_lock(&rps_dev_flow_lock);
 701         old_table = rcu_dereference_protected(queue->rps_flow_table,
 702                                               lockdep_is_held(&rps_dev_flow_lock));
 703         rcu_assign_pointer(queue->rps_flow_table, table);
 704         spin_unlock(&rps_dev_flow_lock);
 705
 706         if (old_table)
 707                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 708
 709         return len;
 710 }
 711
 712 static struct rx_queue_attribute rps_cpus_attribute =
 713         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 714
 715
 716 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 717         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 718             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 719
 720 static struct attribute *rx_queue_default_attrs[] = {
 721         &rps_cpus_attribute.attr,
 722         &rps_dev_flow_table_cnt_attribute.attr,
 723         NULL
 724 };
 725
 726 static void rx_queue_release(struct kobject *kobj)
 727 {
 728         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 729         struct rps_map *map;
 730         struct rps_dev_flow_table *flow_table;
 731
 732
 733         map = rcu_dereference_protected(queue->rps_map, 1);
 734         if (map) {
 735                 RCU_INIT_POINTER(queue->rps_map, NULL);
 736                 kfree_rcu(map, rcu);
 737         }
 738
 739         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 740         if (flow_table) {
 741                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 742                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 743         }
 744
 745         memset(kobj, 0, sizeof(*kobj));
 746         dev_put(queue->dev);
 747 }
 748
 749 static struct kobj_type rx_queue_ktype = {
 750         .sysfs_ops = &rx_queue_sysfs_ops,
 751         .release = rx_queue_release,
 752         .default_attrs = rx_queue_default_attrs,
 753 };
 754
 755 static int rx_queue_add_kobject(struct net_device *net, int index)
 756 {
 757         struct netdev_rx_queue *queue = net->_rx + index;
 758         struct kobject *kobj = &queue->kobj;
 759         int error = 0;
 760
 761         kobj->kset = net->queues_kset;
 762         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 763             "rx-%u", index);
 764         if (error) {
 765                 kobject_put(kobj);
 766                 return error;
 767         }
 768
 769         kobject_uevent(kobj, KOBJ_ADD);
 770         dev_hold(queue->dev);
 771
 772         return error;
 773 }
 774 #endif /* CONFIG_RPS */
 775
 776 int
 777 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 778 {
 779 #ifdef CONFIG_RPS
 780         int i;
 781         int error = 0;
 782
 783         for (i = old_num; i < new_num; i++) {
 784                 error = rx_queue_add_kobject(net, i);
 785                 if (error) {
 786                         new_num = old_num;
 787                         break;
 788                 }
 789         }
 790
 791         while (--i >= new_num)
 792                 kobject_put(&net->_rx[i].kobj);
 793
 794         return error;
 795 #else
 796         return 0;
 797 #endif
 798 }
 799
 800 #ifdef CONFIG_SYSFS
 801 /*
 802  * netdev_queue sysfs structures and functions.
 803  */
 804 struct netdev_queue_attribute {
 805         struct attribute attr;
 806         ssize_t (*show)(struct netdev_queue *queue,
 807             struct netdev_queue_attribute *attr, char *buf);
 808         ssize_t (*store)(struct netdev_queue *queue,
 809             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 810 };
 811 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 812     struct netdev_queue_attribute, attr)
 813
 814 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 815
 816 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 817                                       struct attribute *attr, char *buf)
 818 {
 819         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 820         struct netdev_queue *queue = to_netdev_queue(kobj);
 821
 822         if (!attribute->show)
 823                 return -EIO;
 824
 825         return attribute->show(queue, attribute, buf);
 826 }
 827
 828 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 829                                        struct attribute *attr,
 830                                        const char *buf, size_t count)
 831 {
 832         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 833         struct netdev_queue *queue = to_netdev_queue(kobj);
 834
 835         if (!attribute->store)
 836                 return -EIO;
 837
 838         return attribute->store(queue, attribute, buf, count);
 839 }
 840
 841 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 842         .show = netdev_queue_attr_show,
 843         .store = netdev_queue_attr_store,
 844 };
 845
 846 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 847                                   struct netdev_queue_attribute *attribute,
 848                                   char *buf)
 849 {
 850         unsigned long trans_timeout;
 851
 852         spin_lock_irq(&queue->_xmit_lock);
 853         trans_timeout = queue->trans_timeout;
 854         spin_unlock_irq(&queue->_xmit_lock);
 855
 856         return sprintf(buf, "%lu", trans_timeout);
 857 }
 858
 859 static struct netdev_queue_attribute queue_trans_timeout =
 860         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 861
 862 #ifdef CONFIG_BQL
 863 /*
 864  * Byte queue limits sysfs structures and functions.
 865  */
 866 static ssize_t bql_show(char *buf, unsigned int value)
 867 {
 868         return sprintf(buf, "%u\n", value);
 869 }
 870
 871 static ssize_t bql_set(const char *buf, const size_t count,
 872                        unsigned int *pvalue)
 873 {
 874         unsigned int value;
 875         int err;
 876
 877         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 878                 value = DQL_MAX_LIMIT;
 879         else {
 880                 err = kstrtouint(buf, 10, &value);
 881                 if (err < 0)
 882                         return err;
 883                 if (value > DQL_MAX_LIMIT)
 884                         return -EINVAL;
 885         }
 886
 887         *pvalue = value;
 888
 889         return count;
 890 }
 891
 892 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 893                                   struct netdev_queue_attribute *attr,
 894                                   char *buf)
 895 {
 896         struct dql *dql = &queue->dql;
 897
 898         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 899 }
 900
 901 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 902                                  struct netdev_queue_attribute *attribute,
 903                                  const char *buf, size_t len)
 904 {
 905         struct dql *dql = &queue->dql;
 906         unsigned value;
 907         int err;
 908
 909         err = kstrtouint(buf, 10, &value);
 910         if (err < 0)
 911                 return err;
 912
 913         dql->slack_hold_time = msecs_to_jiffies(value);
 914
 915         return len;
 916 }
 917
 918 static struct netdev_queue_attribute bql_hold_time_attribute =
 919         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 920             bql_set_hold_time);
 921
 922 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 923                                  struct netdev_queue_attribute *attr,
 924                                  char *buf)
 925 {
 926         struct dql *dql = &queue->dql;
 927
 928         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 929 }
 930
 931 static struct netdev_queue_attribute bql_inflight_attribute =
 932         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 933
 934 #define BQL_ATTR(NAME, FIELD)                                           \
 935 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 936                                  struct netdev_queue_attribute *attr,   \
 937                                  char *buf)                             \
 938 {                                                                       \
 939         return bql_show(buf, queue->dql.FIELD);                         \
 940 }                                                                       \
 941                                                                         \
 942 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 943                                 struct netdev_queue_attribute *attr,    \
 944                                 const char *buf, size_t len)            \
 945 {                                                                       \
 946         return bql_set(buf, len, &queue->dql.FIELD);                    \
 947 }                                                                       \
 948                                                                         \
 949 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 950         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 951             bql_set_ ## NAME);
 952
 953 BQL_ATTR(limit, limit)
 954 BQL_ATTR(limit_max, max_limit)
 955 BQL_ATTR(limit_min, min_limit)
 956
 957 static struct attribute *dql_attrs[] = {
 958         &bql_limit_attribute.attr,
 959         &bql_limit_max_attribute.attr,
 960         &bql_limit_min_attribute.attr,
 961         &bql_hold_time_attribute.attr,
 962         &bql_inflight_attribute.attr,
 963         NULL
 964 };
 965
 966 static struct attribute_group dql_group = {
 967         .name  = "byte_queue_limits",
 968         .attrs  = dql_attrs,
 969 };
 970 #endif /* CONFIG_BQL */
 971
 972 #ifdef CONFIG_XPS
 973 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 974 {
 975         struct net_device *dev = queue->dev;
 976         int i;
 977
 978         for (i = 0; i < dev->num_tx_queues; i++)
 979                 if (queue == &dev->_tx[i])
 980                         break;
 981
 982         BUG_ON(i >= dev->num_tx_queues);
 983
 984         return i;
 985 }
 986
 987
 988 static ssize_t show_xps_map(struct netdev_queue *queue,
 989                             struct netdev_queue_attribute *attribute, char *buf)
 990 {
 991         struct net_device *dev = queue->dev;
 992         struct xps_dev_maps *dev_maps;
 993         cpumask_var_t mask;
 994         unsigned long index;
 995         size_t len = 0;
 996         int i;
 997
 998         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 999                 return -ENOMEM;
1000
1001         index = get_netdev_queue_index(queue);
1002
1003         rcu_read_lock();
1004         dev_maps = rcu_dereference(dev->xps_maps);
1005         if (dev_maps) {
1006                 for_each_possible_cpu(i) {
1007                         struct xps_map *map =
1008                             rcu_dereference(dev_maps->cpu_map[i]);
1009                         if (map) {
1010                                 int j;
1011                                 for (j = 0; j < map->len; j++) {
1012                                         if (map->queues[j] == index) {
1013                                                 cpumask_set_cpu(i, mask);
1014                                                 break;
1015                                         }
1016                                 }
1017                         }
1018                 }
1019         }
1020         rcu_read_unlock();
1021
1022         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1023         if (PAGE_SIZE - len < 3) {
1024                 free_cpumask_var(mask);
1025                 return -EINVAL;
1026         }
1027
1028         free_cpumask_var(mask);
1029         len += sprintf(buf + len, "\n");
1030         return len;
1031 }
1032
1033 static DEFINE_MUTEX(xps_map_mutex);
1034 #define xmap_dereference(P)             \
1035         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1036
1037 static void xps_queue_release(struct netdev_queue *queue)
1038 {
1039         struct net_device *dev = queue->dev;
1040         struct xps_dev_maps *dev_maps;
1041         struct xps_map *map;
1042         unsigned long index;
1043         int i, pos, nonempty = 0;
1044
1045         index = get_netdev_queue_index(queue);
1046
1047         mutex_lock(&xps_map_mutex);
1048         dev_maps = xmap_dereference(dev->xps_maps);
1049
1050         if (dev_maps) {
1051                 for_each_possible_cpu(i) {
1052                         map = xmap_dereference(dev_maps->cpu_map[i]);
1053                         if (!map)
1054                                 continue;
1055
1056                         for (pos = 0; pos < map->len; pos++)
1057                                 if (map->queues[pos] == index)
1058                                         break;
1059
1060                         if (pos < map->len) {
1061                                 if (map->len > 1)
1062                                         map->queues[pos] =
1063                                             map->queues[--map->len];
1064                                 else {
1065                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1066                                             NULL);
1067                                         kfree_rcu(map, rcu);
1068                                         map = NULL;
1069                                 }
1070                         }
1071                         if (map)
1072                                 nonempty = 1;
1073                 }
1074
1075                 if (!nonempty) {
1076                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1077                         kfree_rcu(dev_maps, rcu);
1078                 }
1079         }
1080         mutex_unlock(&xps_map_mutex);
1081 }
1082
1083 static ssize_t store_xps_map(struct netdev_queue *queue,
1084                       struct netdev_queue_attribute *attribute,
1085                       const char *buf, size_t len)
1086 {
1087         struct net_device *dev = queue->dev;
1088         cpumask_var_t mask;
1089         int err, i, cpu, pos, map_len, alloc_len, need_set;
1090         unsigned long index;
1091         struct xps_map *map, *new_map;
1092         struct xps_dev_maps *dev_maps, *new_dev_maps;
1093         int nonempty = 0;
1094         int numa_node_id = -2;
1095
1096         if (!capable(CAP_NET_ADMIN))
1097                 return -EPERM;
1098
1099         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1100                 return -ENOMEM;
1101
1102         index = get_netdev_queue_index(queue);
1103
1104         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1105         if (err) {
1106                 free_cpumask_var(mask);
1107                 return err;
1108         }
1109
1110         new_dev_maps = kzalloc(max_t(unsigned,
1111             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1112         if (!new_dev_maps) {
1113                 free_cpumask_var(mask);
1114                 return -ENOMEM;
1115         }
1116
1117         mutex_lock(&xps_map_mutex);
1118
1119         dev_maps = xmap_dereference(dev->xps_maps);
1120
1121         for_each_possible_cpu(cpu) {
1122                 map = dev_maps ?
1123                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1124                 new_map = map;
1125                 if (map) {
1126                         for (pos = 0; pos < map->len; pos++)
1127                                 if (map->queues[pos] == index)
1128                                         break;
1129                         map_len = map->len;
1130                         alloc_len = map->alloc_len;
1131                 } else
1132                         pos = map_len = alloc_len = 0;
1133
1134                 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1135 #ifdef CONFIG_NUMA
1136                 if (need_set) {
1137                         if (numa_node_id == -2)
1138                                 numa_node_id = cpu_to_node(cpu);
1139                         else if (numa_node_id != cpu_to_node(cpu))
1140                                 numa_node_id = -1;
1141                 }
1142 #endif
1143                 if (need_set && pos >= map_len) {
1144                         /* Need to add queue to this CPU's map */
1145                         if (map_len >= alloc_len) {
1146                                 alloc_len = alloc_len ?
1147                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
1148                                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1149                                                        GFP_KERNEL,
1150                                                        cpu_to_node(cpu));
1151                                 if (!new_map)
1152                                         goto error;
1153                                 new_map->alloc_len = alloc_len;
1154                                 for (i = 0; i < map_len; i++)
1155                                         new_map->queues[i] = map->queues[i];
1156                                 new_map->len = map_len;
1157                         }
1158                         new_map->queues[new_map->len++] = index;
1159                 } else if (!need_set && pos < map_len) {
1160                         /* Need to remove queue from this CPU's map */
1161                         if (map_len > 1)
1162                                 new_map->queues[pos] =
1163                                     new_map->queues[--new_map->len];
1164                         else
1165                                 new_map = NULL;
1166                 }
1167                 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1168         }
1169
1170         /* Cleanup old maps */
1171         for_each_possible_cpu(cpu) {
1172                 map = dev_maps ?
1173                         xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1174                 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1175                         kfree_rcu(map, rcu);
1176                 if (new_dev_maps->cpu_map[cpu])
1177                         nonempty = 1;
1178         }
1179
1180         if (nonempty) {
1181                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1182         } else {
1183                 kfree(new_dev_maps);
1184                 RCU_INIT_POINTER(dev->xps_maps, NULL);
1185         }
1186
1187         if (dev_maps)
1188                 kfree_rcu(dev_maps, rcu);
1189
1190         netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1191                                             NUMA_NO_NODE);
1192
1193         mutex_unlock(&xps_map_mutex);
1194
1195         free_cpumask_var(mask);
1196         return len;
1197
1198 error:
1199         mutex_unlock(&xps_map_mutex);
1200
1201         if (new_dev_maps)
1202                 for_each_possible_cpu(i)
1203                         kfree(rcu_dereference_protected(
1204                                 new_dev_maps->cpu_map[i],
1205                                 1));
1206         kfree(new_dev_maps);
1207         free_cpumask_var(mask);
1208         return -ENOMEM;
1209 }
1210
1211 static struct netdev_queue_attribute xps_cpus_attribute =
1212     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1213 #endif /* CONFIG_XPS */
1214
1215 static struct attribute *netdev_queue_default_attrs[] = {
1216         &queue_trans_timeout.attr,
1217 #ifdef CONFIG_XPS
1218         &xps_cpus_attribute.attr,
1219 #endif
1220         NULL
1221 };
1222
1223 static void netdev_queue_release(struct kobject *kobj)
1224 {
1225         struct netdev_queue *queue = to_netdev_queue(kobj);
1226
1227 #ifdef CONFIG_XPS
1228         xps_queue_release(queue);
1229 #endif
1230
1231         memset(kobj, 0, sizeof(*kobj));
1232         dev_put(queue->dev);
1233 }
1234
1235 static struct kobj_type netdev_queue_ktype = {
1236         .sysfs_ops = &netdev_queue_sysfs_ops,
1237         .release = netdev_queue_release,
1238         .default_attrs = netdev_queue_default_attrs,
1239 };
1240
1241 static int netdev_queue_add_kobject(struct net_device *net, int index)
1242 {
1243         struct netdev_queue *queue = net->_tx + index;
1244         struct kobject *kobj = &queue->kobj;
1245         int error = 0;
1246
1247         kobj->kset = net->queues_kset;
1248         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1249             "tx-%u", index);
1250         if (error)
1251                 goto exit;
1252
1253 #ifdef CONFIG_BQL
1254         error = sysfs_create_group(kobj, &dql_group);
1255         if (error)
1256                 goto exit;
1257 #endif
1258
1259         kobject_uevent(kobj, KOBJ_ADD);
1260         dev_hold(queue->dev);
1261
1262         return 0;
1263 exit:
1264         kobject_put(kobj);
1265         return error;
1266 }
1267 #endif /* CONFIG_SYSFS */
1268
1269 int
1270 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1271 {
1272 #ifdef CONFIG_SYSFS
1273         int i;
1274         int error = 0;
1275
1276         for (i = old_num; i < new_num; i++) {
1277                 error = netdev_queue_add_kobject(net, i);
1278                 if (error) {
1279                         new_num = old_num;
1280                         break;
1281                 }
1282         }
1283
1284         while (--i >= new_num) {
1285                 struct netdev_queue *queue = net->_tx + i;
1286
1287 #ifdef CONFIG_BQL
1288                 sysfs_remove_group(&queue->kobj, &dql_group);
1289 #endif
1290                 kobject_put(&queue->kobj);
1291         }
1292
1293         return error;
1294 #else
1295         return 0;
1296 #endif /* CONFIG_SYSFS */
1297 }
1298
1299 static int register_queue_kobjects(struct net_device *net)
1300 {
1301         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1302
1303 #ifdef CONFIG_SYSFS
1304         net->queues_kset = kset_create_and_add("queues",
1305             NULL, &net->dev.kobj);
1306         if (!net->queues_kset)
1307                 return -ENOMEM;
1308 #endif
1309
1310 #ifdef CONFIG_RPS
1311         real_rx = net->real_num_rx_queues;
1312 #endif
1313         real_tx = net->real_num_tx_queues;
1314
1315         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1316         if (error)
1317                 goto error;
1318         rxq = real_rx;
1319
1320         error = netdev_queue_update_kobjects(net, 0, real_tx);
1321         if (error)
1322                 goto error;
1323         txq = real_tx;
1324
1325         return 0;
1326
1327 error:
1328         netdev_queue_update_kobjects(net, txq, 0);
1329         net_rx_queue_update_kobjects(net, rxq, 0);
1330         return error;
1331 }
1332
1333 static void remove_queue_kobjects(struct net_device *net)
1334 {
1335         int real_rx = 0, real_tx = 0;
1336
1337 #ifdef CONFIG_RPS
1338         real_rx = net->real_num_rx_queues;
1339 #endif
1340         real_tx = net->real_num_tx_queues;
1341
1342         net_rx_queue_update_kobjects(net, real_rx, 0);
1343         netdev_queue_update_kobjects(net, real_tx, 0);
1344 #ifdef CONFIG_SYSFS
1345         kset_unregister(net->queues_kset);
1346 #endif
1347 }
1348
1349 static void *net_grab_current_ns(void)
1350 {
1351         struct net *ns = current->nsproxy->net_ns;
1352 #ifdef CONFIG_NET_NS
1353         if (ns)
1354                 atomic_inc(&ns->passive);
1355 #endif
1356         return ns;
1357 }
1358
1359 static const void *net_initial_ns(void)
1360 {
1361         return &init_net;
1362 }
1363
1364 static const void *net_netlink_ns(struct sock *sk)
1365 {
1366         return sock_net(sk);
1367 }
1368
1369 struct kobj_ns_type_operations net_ns_type_operations = {
1370         .type = KOBJ_NS_TYPE_NET,
1371         .grab_current_ns = net_grab_current_ns,
1372         .netlink_ns = net_netlink_ns,
1373         .initial_ns = net_initial_ns,
1374         .drop_ns = net_drop_ns,
1375 };
1376 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1377
1378 #ifdef CONFIG_HOTPLUG
1379 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1380 {
1381         struct net_device *dev = to_net_dev(d);
1382         int retval;
1383
1384         /* pass interface to uevent. */
1385         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1386         if (retval)
1387                 goto exit;
1388
1389         /* pass ifindex to uevent.
1390          * ifindex is useful as it won't change (interface name may change)
1391          * and is what RtNetlink uses natively. */
1392         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1393
1394 exit:
1395         return retval;
1396 }
1397 #endif
1398
1399 /*
1400  *      netdev_release -- destroy and free a dead device.
1401  *      Called when last reference to device kobject is gone.
1402  */
1403 static void netdev_release(struct device *d)
1404 {
1405         struct net_device *dev = to_net_dev(d);
1406
1407         BUG_ON(dev->reg_state != NETREG_RELEASED);
1408
1409         kfree(dev->ifalias);
1410         kfree((char *)dev - dev->padded);
1411 }
1412
1413 static const void *net_namespace(struct device *d)
1414 {
1415         struct net_device *dev;
1416         dev = container_of(d, struct net_device, dev);
1417         return dev_net(dev);
1418 }
1419
1420 static struct class net_class = {
1421         .name = "net",
1422         .dev_release = netdev_release,
1423 #ifdef CONFIG_SYSFS
1424         .dev_attrs = net_class_attributes,
1425 #endif /* CONFIG_SYSFS */
1426 #ifdef CONFIG_HOTPLUG
1427         .dev_uevent = netdev_uevent,
1428 #endif
1429         .ns_type = &net_ns_type_operations,
1430         .namespace = net_namespace,
1431 };
1432
1433 /* Delete sysfs entries but hold kobject reference until after all
1434  * netdev references are gone.
1435  */
1436 void netdev_unregister_kobject(struct net_device * net)
1437 {
1438         struct device *dev = &(net->dev);
1439
1440         kobject_get(&dev->kobj);
1441
1442         remove_queue_kobjects(net);
1443
1444         device_del(dev);
1445 }
1446
1447 /* Create sysfs entries for network device. */
1448 int netdev_register_kobject(struct net_device *net)
1449 {
1450         struct device *dev = &(net->dev);
1451         const struct attribute_group **groups = net->sysfs_groups;
1452         int error = 0;
1453
1454         device_initialize(dev);
1455         dev->class = &net_class;
1456         dev->platform_data = net;
1457         dev->groups = groups;
1458
1459         dev_set_name(dev, "%s", net->name);
1460
1461 #ifdef CONFIG_SYSFS
1462         /* Allow for a device specific group */
1463         if (*groups)
1464                 groups++;
1465
1466         *groups++ = &netstat_group;
1467 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1468         if (net->ieee80211_ptr)
1469                 *groups++ = &wireless_group;
1470 #ifdef CONFIG_WIRELESS_EXT
1471         else if (net->wireless_handlers)
1472                 *groups++ = &wireless_group;
1473 #endif
1474 #endif
1475 #endif /* CONFIG_SYSFS */
1476
1477         error = device_add(dev);
1478         if (error)
1479                 return error;
1480
1481         error = register_queue_kobjects(net);
1482         if (error) {
1483                 device_del(dev);
1484                 return error;
1485         }
1486
1487         return error;
1488 }
1489
1490 int netdev_class_create_file(struct class_attribute *class_attr)
1491 {
1492         return class_create_file(&net_class, class_attr);
1493 }
1494 EXPORT_SYMBOL(netdev_class_create_file);
1495
1496 void netdev_class_remove_file(struct class_attribute *class_attr)
1497 {
1498         class_remove_file(&net_class, class_attr);
1499 }
1500 EXPORT_SYMBOL(netdev_class_remove_file);
1501
1502 int netdev_kobject_init(void)
1503 {
1504         kobj_ns_type_register(&net_ns_type_operations);
1505         return class_register(&net_class);
1506 }