net/core/net-sysfs.c

   1 /*
   2  * net-sysfs.c - network device class and attributes
   3  *
   4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/kernel.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/slab.h>
  17 #include <linux/nsproxy.h>
  18 #include <net/sock.h>
  19 #include <net/net_namespace.h>
  20 #include <linux/rtnetlink.h>
  21 #include <linux/wireless.h>
  22 #include <linux/vmalloc.h>
  23 #include <net/wext.h>
  24
  25 #include "net-sysfs.h"
  26
  27 #ifdef CONFIG_SYSFS
  28 static const char fmt_hex[] = "%#x\n";
  29 static const char fmt_long_hex[] = "%#lx\n";
  30 static const char fmt_dec[] = "%d\n";
  31 static const char fmt_ulong[] = "%lu\n";
  32 static const char fmt_u64[] = "%llu\n";
  33
  34 static inline int dev_isalive(const struct net_device *dev)
  35 {
  36         return dev->reg_state <= NETREG_REGISTERED;
  37 }
  38
  39 /* use same locking rules as GIF* ioctl's */
  40 static ssize_t netdev_show(const struct device *dev,
  41                            struct device_attribute *attr, char *buf,
  42                            ssize_t (*format)(const struct net_device *, char *))
  43 {
  44         struct net_device *net = to_net_dev(dev);
  45         ssize_t ret = -EINVAL;
  46
  47         read_lock(&dev_base_lock);
  48         if (dev_isalive(net))
  49                 ret = (*format)(net, buf);
  50         read_unlock(&dev_base_lock);
  51
  52         return ret;
  53 }
  54
  55 /* generate a show function for simple field */
  56 #define NETDEVICE_SHOW(field, format_string)                            \
  57 static ssize_t format_##field(const struct net_device *net, char *buf)  \
  58 {                                                                       \
  59         return sprintf(buf, format_string, net->field);                 \
  60 }                                                                       \
  61 static ssize_t show_##field(struct device *dev,                         \
  62                             struct device_attribute *attr, char *buf)   \
  63 {                                                                       \
  64         return netdev_show(dev, attr, buf, format_##field);             \
  65 }
  66
  67
  68 /* use same locking and permission rules as SIF* ioctl's */
  69 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  70                             const char *buf, size_t len,
  71                             int (*set)(struct net_device *, unsigned long))
  72 {
  73         struct net_device *net = to_net_dev(dev);
  74         char *endp;
  75         unsigned long new;
  76         int ret = -EINVAL;
  77
  78         if (!capable(CAP_NET_ADMIN))
  79                 return -EPERM;
  80
  81         new = simple_strtoul(buf, &endp, 0);
  82         if (endp == buf)
  83                 goto err;
  84
  85         if (!rtnl_trylock())
  86                 return restart_syscall();
  87
  88         if (dev_isalive(net)) {
  89                 if ((ret = (*set)(net, new)) == 0)
  90                         ret = len;
  91         }
  92         rtnl_unlock();
  93  err:
  94         return ret;
  95 }
  96
  97 NETDEVICE_SHOW(dev_id, fmt_hex);
  98 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
  99 NETDEVICE_SHOW(addr_len, fmt_dec);
 100 NETDEVICE_SHOW(iflink, fmt_dec);
 101 NETDEVICE_SHOW(ifindex, fmt_dec);
 102 NETDEVICE_SHOW(features, fmt_long_hex);
 103 NETDEVICE_SHOW(type, fmt_dec);
 104 NETDEVICE_SHOW(link_mode, fmt_dec);
 105
 106 /* use same locking rules as GIFHWADDR ioctl's */
 107 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 108                             char *buf)
 109 {
 110         struct net_device *net = to_net_dev(dev);
 111         ssize_t ret = -EINVAL;
 112
 113         read_lock(&dev_base_lock);
 114         if (dev_isalive(net))
 115                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 116         read_unlock(&dev_base_lock);
 117         return ret;
 118 }
 119
 120 static ssize_t show_broadcast(struct device *dev,
 121                             struct device_attribute *attr, char *buf)
 122 {
 123         struct net_device *net = to_net_dev(dev);
 124         if (dev_isalive(net))
 125                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 126         return -EINVAL;
 127 }
 128
 129 static ssize_t show_carrier(struct device *dev,
 130                             struct device_attribute *attr, char *buf)
 131 {
 132         struct net_device *netdev = to_net_dev(dev);
 133         if (netif_running(netdev)) {
 134                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 135         }
 136         return -EINVAL;
 137 }
 138
 139 static ssize_t show_speed(struct device *dev,
 140                           struct device_attribute *attr, char *buf)
 141 {
 142         struct net_device *netdev = to_net_dev(dev);
 143         int ret = -EINVAL;
 144
 145         if (!rtnl_trylock())
 146                 return restart_syscall();
 147
 148         if (netif_running(netdev) &&
 149             netdev->ethtool_ops &&
 150             netdev->ethtool_ops->get_settings) {
 151                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
 152
 153                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
 154                         ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
 155         }
 156         rtnl_unlock();
 157         return ret;
 158 }
 159
 160 static ssize_t show_duplex(struct device *dev,
 161                            struct device_attribute *attr, char *buf)
 162 {
 163         struct net_device *netdev = to_net_dev(dev);
 164         int ret = -EINVAL;
 165
 166         if (!rtnl_trylock())
 167                 return restart_syscall();
 168
 169         if (netif_running(netdev) &&
 170             netdev->ethtool_ops &&
 171             netdev->ethtool_ops->get_settings) {
 172                 struct ethtool_cmd cmd = { ETHTOOL_GSET };
 173
 174                 if (!netdev->ethtool_ops->get_settings(netdev, &cmd))
 175                         ret = sprintf(buf, "%s\n", cmd.duplex ? "full" : "half");
 176         }
 177         rtnl_unlock();
 178         return ret;
 179 }
 180
 181 static ssize_t show_dormant(struct device *dev,
 182                             struct device_attribute *attr, char *buf)
 183 {
 184         struct net_device *netdev = to_net_dev(dev);
 185
 186         if (netif_running(netdev))
 187                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 188
 189         return -EINVAL;
 190 }
 191
 192 static const char *const operstates[] = {
 193         "unknown",
 194         "notpresent", /* currently unused */
 195         "down",
 196         "lowerlayerdown",
 197         "testing", /* currently unused */
 198         "dormant",
 199         "up"
 200 };
 201
 202 static ssize_t show_operstate(struct device *dev,
 203                               struct device_attribute *attr, char *buf)
 204 {
 205         const struct net_device *netdev = to_net_dev(dev);
 206         unsigned char operstate;
 207
 208         read_lock(&dev_base_lock);
 209         operstate = netdev->operstate;
 210         if (!netif_running(netdev))
 211                 operstate = IF_OPER_DOWN;
 212         read_unlock(&dev_base_lock);
 213
 214         if (operstate >= ARRAY_SIZE(operstates))
 215                 return -EINVAL; /* should not happen */
 216
 217         return sprintf(buf, "%s\n", operstates[operstate]);
 218 }
 219
 220 /* read-write attributes */
 221 NETDEVICE_SHOW(mtu, fmt_dec);
 222
 223 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 224 {
 225         return dev_set_mtu(net, (int) new_mtu);
 226 }
 227
 228 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 229                          const char *buf, size_t len)
 230 {
 231         return netdev_store(dev, attr, buf, len, change_mtu);
 232 }
 233
 234 NETDEVICE_SHOW(flags, fmt_hex);
 235
 236 static int change_flags(struct net_device *net, unsigned long new_flags)
 237 {
 238         return dev_change_flags(net, (unsigned) new_flags);
 239 }
 240
 241 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 242                            const char *buf, size_t len)
 243 {
 244         return netdev_store(dev, attr, buf, len, change_flags);
 245 }
 246
 247 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 248
 249 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 250 {
 251         net->tx_queue_len = new_len;
 252         return 0;
 253 }
 254
 255 static ssize_t store_tx_queue_len(struct device *dev,
 256                                   struct device_attribute *attr,
 257                                   const char *buf, size_t len)
 258 {
 259         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 260 }
 261
 262 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 263                              const char *buf, size_t len)
 264 {
 265         struct net_device *netdev = to_net_dev(dev);
 266         size_t count = len;
 267         ssize_t ret;
 268
 269         if (!capable(CAP_NET_ADMIN))
 270                 return -EPERM;
 271
 272         /* ignore trailing newline */
 273         if (len >  0 && buf[len - 1] == '\n')
 274                 --count;
 275
 276         if (!rtnl_trylock())
 277                 return restart_syscall();
 278         ret = dev_set_alias(netdev, buf, count);
 279         rtnl_unlock();
 280
 281         return ret < 0 ? ret : len;
 282 }
 283
 284 static ssize_t show_ifalias(struct device *dev,
 285                             struct device_attribute *attr, char *buf)
 286 {
 287         const struct net_device *netdev = to_net_dev(dev);
 288         ssize_t ret = 0;
 289
 290         if (!rtnl_trylock())
 291                 return restart_syscall();
 292         if (netdev->ifalias)
 293                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 294         rtnl_unlock();
 295         return ret;
 296 }
 297
 298 static struct device_attribute net_class_attributes[] = {
 299         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 300         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 301         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 302         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 303         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 304         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 305         __ATTR(features, S_IRUGO, show_features, NULL),
 306         __ATTR(type, S_IRUGO, show_type, NULL),
 307         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 308         __ATTR(address, S_IRUGO, show_address, NULL),
 309         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 310         __ATTR(carrier, S_IRUGO, show_carrier, NULL),
 311         __ATTR(speed, S_IRUGO, show_speed, NULL),
 312         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 313         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 314         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 315         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 316         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 317         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 318                store_tx_queue_len),
 319         {}
 320 };
 321
 322 /* Show a given an attribute in the statistics group */
 323 static ssize_t netstat_show(const struct device *d,
 324                             struct device_attribute *attr, char *buf,
 325                             unsigned long offset)
 326 {
 327         struct net_device *dev = to_net_dev(d);
 328         ssize_t ret = -EINVAL;
 329
 330         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 331                         offset % sizeof(u64) != 0);
 332
 333         read_lock(&dev_base_lock);
 334         if (dev_isalive(dev)) {
 335                 struct rtnl_link_stats64 temp;
 336                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 337
 338                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 339         }
 340         read_unlock(&dev_base_lock);
 341         return ret;
 342 }
 343
 344 /* generate a read-only statistics attribute */
 345 #define NETSTAT_ENTRY(name)                                             \
 346 static ssize_t show_##name(struct device *d,                            \
 347                            struct device_attribute *attr, char *buf)    \
 348 {                                                                       \
 349         return netstat_show(d, attr, buf,                               \
 350                             offsetof(struct rtnl_link_stats64, name));  \
 351 }                                                                       \
 352 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 353
 354 NETSTAT_ENTRY(rx_packets);
 355 NETSTAT_ENTRY(tx_packets);
 356 NETSTAT_ENTRY(rx_bytes);
 357 NETSTAT_ENTRY(tx_bytes);
 358 NETSTAT_ENTRY(rx_errors);
 359 NETSTAT_ENTRY(tx_errors);
 360 NETSTAT_ENTRY(rx_dropped);
 361 NETSTAT_ENTRY(tx_dropped);
 362 NETSTAT_ENTRY(multicast);
 363 NETSTAT_ENTRY(collisions);
 364 NETSTAT_ENTRY(rx_length_errors);
 365 NETSTAT_ENTRY(rx_over_errors);
 366 NETSTAT_ENTRY(rx_crc_errors);
 367 NETSTAT_ENTRY(rx_frame_errors);
 368 NETSTAT_ENTRY(rx_fifo_errors);
 369 NETSTAT_ENTRY(rx_missed_errors);
 370 NETSTAT_ENTRY(tx_aborted_errors);
 371 NETSTAT_ENTRY(tx_carrier_errors);
 372 NETSTAT_ENTRY(tx_fifo_errors);
 373 NETSTAT_ENTRY(tx_heartbeat_errors);
 374 NETSTAT_ENTRY(tx_window_errors);
 375 NETSTAT_ENTRY(rx_compressed);
 376 NETSTAT_ENTRY(tx_compressed);
 377
 378 static struct attribute *netstat_attrs[] = {
 379         &dev_attr_rx_packets.attr,
 380         &dev_attr_tx_packets.attr,
 381         &dev_attr_rx_bytes.attr,
 382         &dev_attr_tx_bytes.attr,
 383         &dev_attr_rx_errors.attr,
 384         &dev_attr_tx_errors.attr,
 385         &dev_attr_rx_dropped.attr,
 386         &dev_attr_tx_dropped.attr,
 387         &dev_attr_multicast.attr,
 388         &dev_attr_collisions.attr,
 389         &dev_attr_rx_length_errors.attr,
 390         &dev_attr_rx_over_errors.attr,
 391         &dev_attr_rx_crc_errors.attr,
 392         &dev_attr_rx_frame_errors.attr,
 393         &dev_attr_rx_fifo_errors.attr,
 394         &dev_attr_rx_missed_errors.attr,
 395         &dev_attr_tx_aborted_errors.attr,
 396         &dev_attr_tx_carrier_errors.attr,
 397         &dev_attr_tx_fifo_errors.attr,
 398         &dev_attr_tx_heartbeat_errors.attr,
 399         &dev_attr_tx_window_errors.attr,
 400         &dev_attr_rx_compressed.attr,
 401         &dev_attr_tx_compressed.attr,
 402         NULL
 403 };
 404
 405
 406 static struct attribute_group netstat_group = {
 407         .name  = "statistics",
 408         .attrs  = netstat_attrs,
 409 };
 410
 411 #ifdef CONFIG_WIRELESS_EXT_SYSFS
 412 /* helper function that does all the locking etc for wireless stats */
 413 static ssize_t wireless_show(struct device *d, char *buf,
 414                              ssize_t (*format)(const struct iw_statistics *,
 415                                                char *))
 416 {
 417         struct net_device *dev = to_net_dev(d);
 418         const struct iw_statistics *iw;
 419         ssize_t ret = -EINVAL;
 420
 421         if (!rtnl_trylock())
 422                 return restart_syscall();
 423         if (dev_isalive(dev)) {
 424                 iw = get_wireless_stats(dev);
 425                 if (iw)
 426                         ret = (*format)(iw, buf);
 427         }
 428         rtnl_unlock();
 429
 430         return ret;
 431 }
 432
 433 /* show function template for wireless fields */
 434 #define WIRELESS_SHOW(name, field, format_string)                       \
 435 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
 436 {                                                                       \
 437         return sprintf(buf, format_string, iw->field);                  \
 438 }                                                                       \
 439 static ssize_t show_iw_##name(struct device *d,                         \
 440                               struct device_attribute *attr, char *buf) \
 441 {                                                                       \
 442         return wireless_show(d, buf, format_iw_##name);                 \
 443 }                                                                       \
 444 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
 445
 446 WIRELESS_SHOW(status, status, fmt_hex);
 447 WIRELESS_SHOW(link, qual.qual, fmt_dec);
 448 WIRELESS_SHOW(level, qual.level, fmt_dec);
 449 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
 450 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
 451 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
 452 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
 453 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
 454 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
 455 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
 456
 457 static struct attribute *wireless_attrs[] = {
 458         &dev_attr_status.attr,
 459         &dev_attr_link.attr,
 460         &dev_attr_level.attr,
 461         &dev_attr_noise.attr,
 462         &dev_attr_nwid.attr,
 463         &dev_attr_crypt.attr,
 464         &dev_attr_fragment.attr,
 465         &dev_attr_retries.attr,
 466         &dev_attr_misc.attr,
 467         &dev_attr_beacon.attr,
 468         NULL
 469 };
 470
 471 static struct attribute_group wireless_group = {
 472         .name = "wireless",
 473         .attrs = wireless_attrs,
 474 };
 475 #endif
 476 #endif /* CONFIG_SYSFS */
 477
 478 #ifdef CONFIG_RPS
 479 /*
 480  * RX queue sysfs structures and functions.
 481  */
 482 struct rx_queue_attribute {
 483         struct attribute attr;
 484         ssize_t (*show)(struct netdev_rx_queue *queue,
 485             struct rx_queue_attribute *attr, char *buf);
 486         ssize_t (*store)(struct netdev_rx_queue *queue,
 487             struct rx_queue_attribute *attr, const char *buf, size_t len);
 488 };
 489 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 490     struct rx_queue_attribute, attr)
 491
 492 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 493
 494 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 495                                   char *buf)
 496 {
 497         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 498         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 499
 500         if (!attribute->show)
 501                 return -EIO;
 502
 503         return attribute->show(queue, attribute, buf);
 504 }
 505
 506 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 507                                    const char *buf, size_t count)
 508 {
 509         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 510         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 511
 512         if (!attribute->store)
 513                 return -EIO;
 514
 515         return attribute->store(queue, attribute, buf, count);
 516 }
 517
 518 static const struct sysfs_ops rx_queue_sysfs_ops = {
 519         .show = rx_queue_attr_show,
 520         .store = rx_queue_attr_store,
 521 };
 522
 523 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 524                             struct rx_queue_attribute *attribute, char *buf)
 525 {
 526         struct rps_map *map;
 527         cpumask_var_t mask;
 528         size_t len = 0;
 529         int i;
 530
 531         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 532                 return -ENOMEM;
 533
 534         rcu_read_lock();
 535         map = rcu_dereference(queue->rps_map);
 536         if (map)
 537                 for (i = 0; i < map->len; i++)
 538                         cpumask_set_cpu(map->cpus[i], mask);
 539
 540         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 541         if (PAGE_SIZE - len < 3) {
 542                 rcu_read_unlock();
 543                 free_cpumask_var(mask);
 544                 return -EINVAL;
 545         }
 546         rcu_read_unlock();
 547
 548         free_cpumask_var(mask);
 549         len += sprintf(buf + len, "\n");
 550         return len;
 551 }
 552
 553 static void rps_map_release(struct rcu_head *rcu)
 554 {
 555         struct rps_map *map = container_of(rcu, struct rps_map, rcu);
 556
 557         kfree(map);
 558 }
 559
 560 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 561                       struct rx_queue_attribute *attribute,
 562                       const char *buf, size_t len)
 563 {
 564         struct rps_map *old_map, *map;
 565         cpumask_var_t mask;
 566         int err, cpu, i;
 567         static DEFINE_SPINLOCK(rps_map_lock);
 568
 569         if (!capable(CAP_NET_ADMIN))
 570                 return -EPERM;
 571
 572         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 573                 return -ENOMEM;
 574
 575         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 576         if (err) {
 577                 free_cpumask_var(mask);
 578                 return err;
 579         }
 580
 581         map = kzalloc(max_t(unsigned,
 582             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 583             GFP_KERNEL);
 584         if (!map) {
 585                 free_cpumask_var(mask);
 586                 return -ENOMEM;
 587         }
 588
 589         i = 0;
 590         for_each_cpu_and(cpu, mask, cpu_online_mask)
 591                 map->cpus[i++] = cpu;
 592
 593         if (i)
 594                 map->len = i;
 595         else {
 596                 kfree(map);
 597                 map = NULL;
 598         }
 599
 600         spin_lock(&rps_map_lock);
 601         old_map = rcu_dereference_protected(queue->rps_map,
 602                                             lockdep_is_held(&rps_map_lock));
 603         rcu_assign_pointer(queue->rps_map, map);
 604         spin_unlock(&rps_map_lock);
 605
 606         if (old_map)
 607                 call_rcu(&old_map->rcu, rps_map_release);
 608
 609         free_cpumask_var(mask);
 610         return len;
 611 }
 612
 613 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 614                                            struct rx_queue_attribute *attr,
 615                                            char *buf)
 616 {
 617         struct rps_dev_flow_table *flow_table;
 618         unsigned int val = 0;
 619
 620         rcu_read_lock();
 621         flow_table = rcu_dereference(queue->rps_flow_table);
 622         if (flow_table)
 623                 val = flow_table->mask + 1;
 624         rcu_read_unlock();
 625
 626         return sprintf(buf, "%u\n", val);
 627 }
 628
 629 static void rps_dev_flow_table_release_work(struct work_struct *work)
 630 {
 631         struct rps_dev_flow_table *table = container_of(work,
 632             struct rps_dev_flow_table, free_work);
 633
 634         vfree(table);
 635 }
 636
 637 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 638 {
 639         struct rps_dev_flow_table *table = container_of(rcu,
 640             struct rps_dev_flow_table, rcu);
 641
 642         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 643         schedule_work(&table->free_work);
 644 }
 645
 646 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 647                                      struct rx_queue_attribute *attr,
 648                                      const char *buf, size_t len)
 649 {
 650         unsigned int count;
 651         char *endp;
 652         struct rps_dev_flow_table *table, *old_table;
 653         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 654
 655         if (!capable(CAP_NET_ADMIN))
 656                 return -EPERM;
 657
 658         count = simple_strtoul(buf, &endp, 0);
 659         if (endp == buf)
 660                 return -EINVAL;
 661
 662         if (count) {
 663                 int i;
 664
 665                 if (count > 1<<30) {
 666                         /* Enforce a limit to prevent overflow */
 667                         return -EINVAL;
 668                 }
 669                 count = roundup_pow_of_two(count);
 670                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
 671                 if (!table)
 672                         return -ENOMEM;
 673
 674                 table->mask = count - 1;
 675                 for (i = 0; i < count; i++)
 676                         table->flows[i].cpu = RPS_NO_CPU;
 677         } else
 678                 table = NULL;
 679
 680         spin_lock(&rps_dev_flow_lock);
 681         old_table = rcu_dereference_protected(queue->rps_flow_table,
 682                                               lockdep_is_held(&rps_dev_flow_lock));
 683         rcu_assign_pointer(queue->rps_flow_table, table);
 684         spin_unlock(&rps_dev_flow_lock);
 685
 686         if (old_table)
 687                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 688
 689         return len;
 690 }
 691
 692 static struct rx_queue_attribute rps_cpus_attribute =
 693         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 694
 695
 696 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 697         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 698             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 699
 700 static struct attribute *rx_queue_default_attrs[] = {
 701         &rps_cpus_attribute.attr,
 702         &rps_dev_flow_table_cnt_attribute.attr,
 703         NULL
 704 };
 705
 706 static void rx_queue_release(struct kobject *kobj)
 707 {
 708         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 709         struct rps_map *map;
 710         struct rps_dev_flow_table *flow_table;
 711
 712
 713         map = rcu_dereference_raw(queue->rps_map);
 714         if (map) {
 715                 RCU_INIT_POINTER(queue->rps_map, NULL);
 716                 call_rcu(&map->rcu, rps_map_release);
 717         }
 718
 719         flow_table = rcu_dereference_raw(queue->rps_flow_table);
 720         if (flow_table) {
 721                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 722                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 723         }
 724
 725         memset(kobj, 0, sizeof(*kobj));
 726         dev_put(queue->dev);
 727 }
 728
 729 static struct kobj_type rx_queue_ktype = {
 730         .sysfs_ops = &rx_queue_sysfs_ops,
 731         .release = rx_queue_release,
 732         .default_attrs = rx_queue_default_attrs,
 733 };
 734
 735 static int rx_queue_add_kobject(struct net_device *net, int index)
 736 {
 737         struct netdev_rx_queue *queue = net->_rx + index;
 738         struct kobject *kobj = &queue->kobj;
 739         int error = 0;
 740
 741         kobj->kset = net->queues_kset;
 742         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 743             "rx-%u", index);
 744         if (error) {
 745                 kobject_put(kobj);
 746                 return error;
 747         }
 748
 749         kobject_uevent(kobj, KOBJ_ADD);
 750         dev_hold(queue->dev);
 751
 752         return error;
 753 }
 754
 755 int
 756 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 757 {
 758         int i;
 759         int error = 0;
 760
 761         for (i = old_num; i < new_num; i++) {
 762                 error = rx_queue_add_kobject(net, i);
 763                 if (error) {
 764                         new_num = old_num;
 765                         break;
 766                 }
 767         }
 768
 769         while (--i >= new_num)
 770                 kobject_put(&net->_rx[i].kobj);
 771
 772         return error;
 773 }
 774
 775 /*
 776  * netdev_queue sysfs structures and functions.
 777  */
 778 struct netdev_queue_attribute {
 779         struct attribute attr;
 780         ssize_t (*show)(struct netdev_queue *queue,
 781             struct netdev_queue_attribute *attr, char *buf);
 782         ssize_t (*store)(struct netdev_queue *queue,
 783             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 784 };
 785 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 786     struct netdev_queue_attribute, attr)
 787
 788 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 789
 790 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 791                                       struct attribute *attr, char *buf)
 792 {
 793         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 794         struct netdev_queue *queue = to_netdev_queue(kobj);
 795
 796         if (!attribute->show)
 797                 return -EIO;
 798
 799         return attribute->show(queue, attribute, buf);
 800 }
 801
 802 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 803                                        struct attribute *attr,
 804                                        const char *buf, size_t count)
 805 {
 806         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 807         struct netdev_queue *queue = to_netdev_queue(kobj);
 808
 809         if (!attribute->store)
 810                 return -EIO;
 811
 812         return attribute->store(queue, attribute, buf, count);
 813 }
 814
 815 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 816         .show = netdev_queue_attr_show,
 817         .store = netdev_queue_attr_store,
 818 };
 819
 820 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 821 {
 822         struct net_device *dev = queue->dev;
 823         int i;
 824
 825         for (i = 0; i < dev->num_tx_queues; i++)
 826                 if (queue == &dev->_tx[i])
 827                         break;
 828
 829         BUG_ON(i >= dev->num_tx_queues);
 830
 831         return i;
 832 }
 833
 834
 835 static ssize_t show_xps_map(struct netdev_queue *queue,
 836                             struct netdev_queue_attribute *attribute, char *buf)
 837 {
 838         struct net_device *dev = queue->dev;
 839         struct xps_dev_maps *dev_maps;
 840         cpumask_var_t mask;
 841         unsigned long index;
 842         size_t len = 0;
 843         int i;
 844
 845         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 846                 return -ENOMEM;
 847
 848         index = get_netdev_queue_index(queue);
 849
 850         rcu_read_lock();
 851         dev_maps = rcu_dereference(dev->xps_maps);
 852         if (dev_maps) {
 853                 for_each_possible_cpu(i) {
 854                         struct xps_map *map =
 855                             rcu_dereference(dev_maps->cpu_map[i]);
 856                         if (map) {
 857                                 int j;
 858                                 for (j = 0; j < map->len; j++) {
 859                                         if (map->queues[j] == index) {
 860                                                 cpumask_set_cpu(i, mask);
 861                                                 break;
 862                                         }
 863                                 }
 864                         }
 865                 }
 866         }
 867         rcu_read_unlock();
 868
 869         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 870         if (PAGE_SIZE - len < 3) {
 871                 free_cpumask_var(mask);
 872                 return -EINVAL;
 873         }
 874
 875         free_cpumask_var(mask);
 876         len += sprintf(buf + len, "\n");
 877         return len;
 878 }
 879
 880 static void xps_map_release(struct rcu_head *rcu)
 881 {
 882         struct xps_map *map = container_of(rcu, struct xps_map, rcu);
 883
 884         kfree(map);
 885 }
 886
 887 static void xps_dev_maps_release(struct rcu_head *rcu)
 888 {
 889         struct xps_dev_maps *dev_maps =
 890             container_of(rcu, struct xps_dev_maps, rcu);
 891
 892         kfree(dev_maps);
 893 }
 894
 895 static DEFINE_MUTEX(xps_map_mutex);
 896
 897 static ssize_t store_xps_map(struct netdev_queue *queue,
 898                       struct netdev_queue_attribute *attribute,
 899                       const char *buf, size_t len)
 900 {
 901         struct net_device *dev = queue->dev;
 902         cpumask_var_t mask;
 903         int err, i, cpu, pos, map_len, alloc_len, need_set;
 904         unsigned long index;
 905         struct xps_map *map, *new_map;
 906         struct xps_dev_maps *dev_maps, *new_dev_maps;
 907         int nonempty = 0;
 908
 909         if (!capable(CAP_NET_ADMIN))
 910                 return -EPERM;
 911
 912         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 913                 return -ENOMEM;
 914
 915         index = get_netdev_queue_index(queue);
 916
 917         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 918         if (err) {
 919                 free_cpumask_var(mask);
 920                 return err;
 921         }
 922
 923         new_dev_maps = kzalloc(max_t(unsigned,
 924             XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
 925         if (!new_dev_maps) {
 926                 free_cpumask_var(mask);
 927                 return -ENOMEM;
 928         }
 929
 930         mutex_lock(&xps_map_mutex);
 931
 932         dev_maps = dev->xps_maps;
 933
 934         for_each_possible_cpu(cpu) {
 935                 new_map = map = dev_maps ? dev_maps->cpu_map[cpu] : NULL;
 936
 937                 if (map) {
 938                         for (pos = 0; pos < map->len; pos++)
 939                                 if (map->queues[pos] == index)
 940                                         break;
 941                         map_len = map->len;
 942                         alloc_len = map->alloc_len;
 943                 } else
 944                         pos = map_len = alloc_len = 0;
 945
 946                 need_set = cpu_isset(cpu, *mask) && cpu_online(cpu);
 947
 948                 if (need_set && pos >= map_len) {
 949                         /* Need to add queue to this CPU's map */
 950                         if (map_len >= alloc_len) {
 951                                 alloc_len = alloc_len ?
 952                                     2 * alloc_len : XPS_MIN_MAP_ALLOC;
 953                                 new_map = kzalloc(XPS_MAP_SIZE(alloc_len),
 954                                     GFP_KERNEL);
 955                                 if (!new_map)
 956                                         goto error;
 957                                 new_map->alloc_len = alloc_len;
 958                                 for (i = 0; i < map_len; i++)
 959                                         new_map->queues[i] = map->queues[i];
 960                                 new_map->len = map_len;
 961                         }
 962                         new_map->queues[new_map->len++] = index;
 963                 } else if (!need_set && pos < map_len) {
 964                         /* Need to remove queue from this CPU's map */
 965                         if (map_len > 1)
 966                                 new_map->queues[pos] =
 967                                     new_map->queues[--new_map->len];
 968                         else
 969                                 new_map = NULL;
 970                 }
 971                 new_dev_maps->cpu_map[cpu] = new_map;
 972         }
 973
 974         /* Cleanup old maps */
 975         for_each_possible_cpu(cpu) {
 976                 map = dev_maps ? dev_maps->cpu_map[cpu] : NULL;
 977                 if (map && new_dev_maps->cpu_map[cpu] != map)
 978                         call_rcu(&map->rcu, xps_map_release);
 979                 if (new_dev_maps->cpu_map[cpu])
 980                         nonempty = 1;
 981         }
 982
 983         if (nonempty)
 984                 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
 985         else {
 986                 kfree(new_dev_maps);
 987                 rcu_assign_pointer(dev->xps_maps, NULL);
 988         }
 989
 990         if (dev_maps)
 991                 call_rcu(&dev_maps->rcu, xps_dev_maps_release);
 992
 993         mutex_unlock(&xps_map_mutex);
 994
 995         free_cpumask_var(mask);
 996         return len;
 997
 998 error:
 999         mutex_unlock(&xps_map_mutex);
1000
1001         if (new_dev_maps)
1002                 for_each_possible_cpu(i)
1003                         kfree(new_dev_maps->cpu_map[i]);
1004         kfree(new_dev_maps);
1005         free_cpumask_var(mask);
1006         return -ENOMEM;
1007 }
1008
1009 static struct netdev_queue_attribute xps_cpus_attribute =
1010     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1011
1012 static struct attribute *netdev_queue_default_attrs[] = {
1013         &xps_cpus_attribute.attr,
1014         NULL
1015 };
1016
1017 static void netdev_queue_release(struct kobject *kobj)
1018 {
1019         struct netdev_queue *queue = to_netdev_queue(kobj);
1020         struct net_device *dev = queue->dev;
1021         struct xps_dev_maps *dev_maps;
1022         struct xps_map *map;
1023         unsigned long index;
1024         int i, pos, nonempty = 0;
1025
1026         index = get_netdev_queue_index(queue);
1027
1028         mutex_lock(&xps_map_mutex);
1029         dev_maps = dev->xps_maps;
1030
1031         if (dev_maps) {
1032                 for_each_possible_cpu(i) {
1033                         map  = dev_maps->cpu_map[i];
1034                         if (!map)
1035                                 continue;
1036
1037                         for (pos = 0; pos < map->len; pos++)
1038                                 if (map->queues[pos] == index)
1039                                         break;
1040
1041                         if (pos < map->len) {
1042                                 if (map->len > 1)
1043                                         map->queues[pos] =
1044                                             map->queues[--map->len];
1045                                 else {
1046                                         RCU_INIT_POINTER(dev_maps->cpu_map[i],
1047                                             NULL);
1048                                         call_rcu(&map->rcu, xps_map_release);
1049                                         map = NULL;
1050                                 }
1051                         }
1052                         if (map)
1053                                 nonempty = 1;
1054                 }
1055
1056                 if (!nonempty) {
1057                         RCU_INIT_POINTER(dev->xps_maps, NULL);
1058                         call_rcu(&dev_maps->rcu, xps_dev_maps_release);
1059                 }
1060         }
1061
1062         mutex_unlock(&xps_map_mutex);
1063
1064         memset(kobj, 0, sizeof(*kobj));
1065         dev_put(queue->dev);
1066 }
1067
1068 static struct kobj_type netdev_queue_ktype = {
1069         .sysfs_ops = &netdev_queue_sysfs_ops,
1070         .release = netdev_queue_release,
1071         .default_attrs = netdev_queue_default_attrs,
1072 };
1073
1074 static int netdev_queue_add_kobject(struct net_device *net, int index)
1075 {
1076         struct netdev_queue *queue = net->_tx + index;
1077         struct kobject *kobj = &queue->kobj;
1078         int error = 0;
1079
1080         kobj->kset = net->queues_kset;
1081         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1082             "tx-%u", index);
1083         if (error) {
1084                 kobject_put(kobj);
1085                 return error;
1086         }
1087
1088         kobject_uevent(kobj, KOBJ_ADD);
1089         dev_hold(queue->dev);
1090
1091         return error;
1092 }
1093
1094 int
1095 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1096 {
1097         int i;
1098         int error = 0;
1099
1100         for (i = old_num; i < new_num; i++) {
1101                 error = netdev_queue_add_kobject(net, i);
1102                 if (error) {
1103                         new_num = old_num;
1104                         break;
1105                 }
1106         }
1107
1108         while (--i >= new_num)
1109                 kobject_put(&net->_tx[i].kobj);
1110
1111         return error;
1112 }
1113
1114 static int register_queue_kobjects(struct net_device *net)
1115 {
1116         int error = 0, txq = 0, rxq = 0;
1117
1118         net->queues_kset = kset_create_and_add("queues",
1119             NULL, &net->dev.kobj);
1120         if (!net->queues_kset)
1121                 return -ENOMEM;
1122
1123         error = net_rx_queue_update_kobjects(net, 0, net->real_num_rx_queues);
1124         if (error)
1125                 goto error;
1126         rxq = net->real_num_rx_queues;
1127
1128         error = netdev_queue_update_kobjects(net, 0,
1129                                              net->real_num_tx_queues);
1130         if (error)
1131                 goto error;
1132         txq = net->real_num_tx_queues;
1133
1134         return 0;
1135
1136 error:
1137         netdev_queue_update_kobjects(net, txq, 0);
1138         net_rx_queue_update_kobjects(net, rxq, 0);
1139         return error;
1140 }
1141
1142 static void remove_queue_kobjects(struct net_device *net)
1143 {
1144         net_rx_queue_update_kobjects(net, net->real_num_rx_queues, 0);
1145         netdev_queue_update_kobjects(net, net->real_num_tx_queues, 0);
1146         kset_unregister(net->queues_kset);
1147 }
1148 #endif /* CONFIG_RPS */
1149
1150 static const void *net_current_ns(void)
1151 {
1152         return current->nsproxy->net_ns;
1153 }
1154
1155 static const void *net_initial_ns(void)
1156 {
1157         return &init_net;
1158 }
1159
1160 static const void *net_netlink_ns(struct sock *sk)
1161 {
1162         return sock_net(sk);
1163 }
1164
1165 struct kobj_ns_type_operations net_ns_type_operations = {
1166         .type = KOBJ_NS_TYPE_NET,
1167         .current_ns = net_current_ns,
1168         .netlink_ns = net_netlink_ns,
1169         .initial_ns = net_initial_ns,
1170 };
1171 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1172
1173 static void net_kobj_ns_exit(struct net *net)
1174 {
1175         kobj_ns_exit(KOBJ_NS_TYPE_NET, net);
1176 }
1177
1178 static struct pernet_operations kobj_net_ops = {
1179         .exit = net_kobj_ns_exit,
1180 };
1181
1182
1183 #ifdef CONFIG_HOTPLUG
1184 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1185 {
1186         struct net_device *dev = to_net_dev(d);
1187         int retval;
1188
1189         /* pass interface to uevent. */
1190         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1191         if (retval)
1192                 goto exit;
1193
1194         /* pass ifindex to uevent.
1195          * ifindex is useful as it won't change (interface name may change)
1196          * and is what RtNetlink uses natively. */
1197         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1198
1199 exit:
1200         return retval;
1201 }
1202 #endif
1203
1204 /*
1205  *      netdev_release -- destroy and free a dead device.
1206  *      Called when last reference to device kobject is gone.
1207  */
1208 static void netdev_release(struct device *d)
1209 {
1210         struct net_device *dev = to_net_dev(d);
1211
1212         BUG_ON(dev->reg_state != NETREG_RELEASED);
1213
1214         kfree(dev->ifalias);
1215         kfree((char *)dev - dev->padded);
1216 }
1217
1218 static const void *net_namespace(struct device *d)
1219 {
1220         struct net_device *dev;
1221         dev = container_of(d, struct net_device, dev);
1222         return dev_net(dev);
1223 }
1224
1225 static struct class net_class = {
1226         .name = "net",
1227         .dev_release = netdev_release,
1228 #ifdef CONFIG_SYSFS
1229         .dev_attrs = net_class_attributes,
1230 #endif /* CONFIG_SYSFS */
1231 #ifdef CONFIG_HOTPLUG
1232         .dev_uevent = netdev_uevent,
1233 #endif
1234         .ns_type = &net_ns_type_operations,
1235         .namespace = net_namespace,
1236 };
1237
1238 /* Delete sysfs entries but hold kobject reference until after all
1239  * netdev references are gone.
1240  */
1241 void netdev_unregister_kobject(struct net_device * net)
1242 {
1243         struct device *dev = &(net->dev);
1244
1245         kobject_get(&dev->kobj);
1246
1247 #ifdef CONFIG_RPS
1248         remove_queue_kobjects(net);
1249 #endif
1250
1251         device_del(dev);
1252 }
1253
1254 /* Create sysfs entries for network device. */
1255 int netdev_register_kobject(struct net_device *net)
1256 {
1257         struct device *dev = &(net->dev);
1258         const struct attribute_group **groups = net->sysfs_groups;
1259         int error = 0;
1260
1261         device_initialize(dev);
1262         dev->class = &net_class;
1263         dev->platform_data = net;
1264         dev->groups = groups;
1265
1266         dev_set_name(dev, "%s", net->name);
1267
1268 #ifdef CONFIG_SYSFS
1269         /* Allow for a device specific group */
1270         if (*groups)
1271                 groups++;
1272
1273         *groups++ = &netstat_group;
1274 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1275         if (net->ieee80211_ptr)
1276                 *groups++ = &wireless_group;
1277 #ifdef CONFIG_WIRELESS_EXT
1278         else if (net->wireless_handlers)
1279                 *groups++ = &wireless_group;
1280 #endif
1281 #endif
1282 #endif /* CONFIG_SYSFS */
1283
1284         error = device_add(dev);
1285         if (error)
1286                 return error;
1287
1288 #ifdef CONFIG_RPS
1289         error = register_queue_kobjects(net);
1290         if (error) {
1291                 device_del(dev);
1292                 return error;
1293         }
1294 #endif
1295
1296         return error;
1297 }
1298
1299 int netdev_class_create_file(struct class_attribute *class_attr)
1300 {
1301         return class_create_file(&net_class, class_attr);
1302 }
1303 EXPORT_SYMBOL(netdev_class_create_file);
1304
1305 void netdev_class_remove_file(struct class_attribute *class_attr)
1306 {
1307         class_remove_file(&net_class, class_attr);
1308 }
1309 EXPORT_SYMBOL(netdev_class_remove_file);
1310
1311 int netdev_kobject_init(void)
1312 {
1313         kobj_ns_type_register(&net_ns_type_operations);
1314         register_pernet_subsys(&kobj_net_ops);
1315         return class_register(&net_class);
1316 }