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 <net/switchdev.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/slab.h>
  18 #include <linux/nsproxy.h>
  19 #include <net/sock.h>
  20 #include <net/net_namespace.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/export.h>
  24 #include <linux/jiffies.h>
  25 #include <linux/pm_runtime.h>
  26 #include <linux/of.h>
  27
  28 #include "net-sysfs.h"
  29
  30 #ifdef CONFIG_SYSFS
  31 static const char fmt_hex[] = "%#x\n";
  32 static const char fmt_long_hex[] = "%#lx\n";
  33 static const char fmt_dec[] = "%d\n";
  34 static const char fmt_udec[] = "%u\n";
  35 static const char fmt_ulong[] = "%lu\n";
  36 static const char fmt_u64[] = "%llu\n";
  37
  38 static inline int dev_isalive(const struct net_device *dev)
  39 {
  40         return dev->reg_state <= NETREG_REGISTERED;
  41 }
  42
  43 /* use same locking rules as GIF* ioctl's */
  44 static ssize_t netdev_show(const struct device *dev,
  45                            struct device_attribute *attr, char *buf,
  46                            ssize_t (*format)(const struct net_device *, char *))
  47 {
  48         struct net_device *ndev = to_net_dev(dev);
  49         ssize_t ret = -EINVAL;
  50
  51         read_lock(&dev_base_lock);
  52         if (dev_isalive(ndev))
  53                 ret = (*format)(ndev, buf);
  54         read_unlock(&dev_base_lock);
  55
  56         return ret;
  57 }
  58
  59 /* generate a show function for simple field */
  60 #define NETDEVICE_SHOW(field, format_string)                            \
  61 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  62 {                                                                       \
  63         return sprintf(buf, format_string, dev->field);                 \
  64 }                                                                       \
  65 static ssize_t field##_show(struct device *dev,                         \
  66                             struct device_attribute *attr, char *buf)   \
  67 {                                                                       \
  68         return netdev_show(dev, attr, buf, format_##field);             \
  69 }                                                                       \
  70
  71 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  72 NETDEVICE_SHOW(field, format_string);                                   \
  73 static DEVICE_ATTR_RO(field)
  74
  75 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  76 NETDEVICE_SHOW(field, format_string);                                   \
  77 static DEVICE_ATTR_RW(field)
  78
  79 /* use same locking and permission rules as SIF* ioctl's */
  80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  81                             const char *buf, size_t len,
  82                             int (*set)(struct net_device *, unsigned long))
  83 {
  84         struct net_device *netdev = to_net_dev(dev);
  85         struct net *net = dev_net(netdev);
  86         unsigned long new;
  87         int ret = -EINVAL;
  88
  89         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  90                 return -EPERM;
  91
  92         ret = kstrtoul(buf, 0, &new);
  93         if (ret)
  94                 goto err;
  95
  96         if (!rtnl_trylock())
  97                 return restart_syscall();
  98
  99         if (dev_isalive(netdev)) {
 100                 if ((ret = (*set)(netdev, new)) == 0)
 101                         ret = len;
 102         }
 103         rtnl_unlock();
 104  err:
 105         return ret;
 106 }
 107
 108 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 109 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 110 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 111 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 112 NETDEVICE_SHOW_RO(iflink, fmt_dec);
 113 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 114 NETDEVICE_SHOW_RO(type, fmt_dec);
 115 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 116
 117 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 118 {
 119         return sprintf(buf, fmt_dec, dev->name_assign_type);
 120 }
 121
 122 static ssize_t name_assign_type_show(struct device *dev,
 123                                      struct device_attribute *attr,
 124                                      char *buf)
 125 {
 126         struct net_device *ndev = to_net_dev(dev);
 127         ssize_t ret = -EINVAL;
 128
 129         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 130                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 131
 132         return ret;
 133 }
 134 static DEVICE_ATTR_RO(name_assign_type);
 135
 136 /* use same locking rules as GIFHWADDR ioctl's */
 137 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 138                             char *buf)
 139 {
 140         struct net_device *ndev = to_net_dev(dev);
 141         ssize_t ret = -EINVAL;
 142
 143         read_lock(&dev_base_lock);
 144         if (dev_isalive(ndev))
 145                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 146         read_unlock(&dev_base_lock);
 147         return ret;
 148 }
 149 static DEVICE_ATTR_RO(address);
 150
 151 static ssize_t broadcast_show(struct device *dev,
 152                               struct device_attribute *attr, char *buf)
 153 {
 154         struct net_device *ndev = to_net_dev(dev);
 155         if (dev_isalive(ndev))
 156                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 157         return -EINVAL;
 158 }
 159 static DEVICE_ATTR_RO(broadcast);
 160
 161 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 162 {
 163         if (!netif_running(dev))
 164                 return -EINVAL;
 165         return dev_change_carrier(dev, (bool) new_carrier);
 166 }
 167
 168 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 169                              const char *buf, size_t len)
 170 {
 171         return netdev_store(dev, attr, buf, len, change_carrier);
 172 }
 173
 174 static ssize_t carrier_show(struct device *dev,
 175                             struct device_attribute *attr, char *buf)
 176 {
 177         struct net_device *netdev = to_net_dev(dev);
 178         if (netif_running(netdev)) {
 179                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 180         }
 181         return -EINVAL;
 182 }
 183 static DEVICE_ATTR_RW(carrier);
 184
 185 static ssize_t speed_show(struct device *dev,
 186                           struct device_attribute *attr, char *buf)
 187 {
 188         struct net_device *netdev = to_net_dev(dev);
 189         int ret = -EINVAL;
 190
 191         if (!rtnl_trylock())
 192                 return restart_syscall();
 193
 194         if (netif_running(netdev)) {
 195                 struct ethtool_cmd cmd;
 196                 if (!__ethtool_get_settings(netdev, &cmd))
 197                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 198         }
 199         rtnl_unlock();
 200         return ret;
 201 }
 202 static DEVICE_ATTR_RO(speed);
 203
 204 static ssize_t duplex_show(struct device *dev,
 205                            struct device_attribute *attr, char *buf)
 206 {
 207         struct net_device *netdev = to_net_dev(dev);
 208         int ret = -EINVAL;
 209
 210         if (!rtnl_trylock())
 211                 return restart_syscall();
 212
 213         if (netif_running(netdev)) {
 214                 struct ethtool_cmd cmd;
 215                 if (!__ethtool_get_settings(netdev, &cmd)) {
 216                         const char *duplex;
 217                         switch (cmd.duplex) {
 218                         case DUPLEX_HALF:
 219                                 duplex = "half";
 220                                 break;
 221                         case DUPLEX_FULL:
 222                                 duplex = "full";
 223                                 break;
 224                         default:
 225                                 duplex = "unknown";
 226                                 break;
 227                         }
 228                         ret = sprintf(buf, "%s\n", duplex);
 229                 }
 230         }
 231         rtnl_unlock();
 232         return ret;
 233 }
 234 static DEVICE_ATTR_RO(duplex);
 235
 236 static ssize_t dormant_show(struct device *dev,
 237                             struct device_attribute *attr, char *buf)
 238 {
 239         struct net_device *netdev = to_net_dev(dev);
 240
 241         if (netif_running(netdev))
 242                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 243
 244         return -EINVAL;
 245 }
 246 static DEVICE_ATTR_RO(dormant);
 247
 248 static const char *const operstates[] = {
 249         "unknown",
 250         "notpresent", /* currently unused */
 251         "down",
 252         "lowerlayerdown",
 253         "testing", /* currently unused */
 254         "dormant",
 255         "up"
 256 };
 257
 258 static ssize_t operstate_show(struct device *dev,
 259                               struct device_attribute *attr, char *buf)
 260 {
 261         const struct net_device *netdev = to_net_dev(dev);
 262         unsigned char operstate;
 263
 264         read_lock(&dev_base_lock);
 265         operstate = netdev->operstate;
 266         if (!netif_running(netdev))
 267                 operstate = IF_OPER_DOWN;
 268         read_unlock(&dev_base_lock);
 269
 270         if (operstate >= ARRAY_SIZE(operstates))
 271                 return -EINVAL; /* should not happen */
 272
 273         return sprintf(buf, "%s\n", operstates[operstate]);
 274 }
 275 static DEVICE_ATTR_RO(operstate);
 276
 277 static ssize_t carrier_changes_show(struct device *dev,
 278                                     struct device_attribute *attr,
 279                                     char *buf)
 280 {
 281         struct net_device *netdev = to_net_dev(dev);
 282         return sprintf(buf, fmt_dec,
 283                        atomic_read(&netdev->carrier_changes));
 284 }
 285 static DEVICE_ATTR_RO(carrier_changes);
 286
 287 /* read-write attributes */
 288
 289 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 290 {
 291         return dev_set_mtu(dev, (int) new_mtu);
 292 }
 293
 294 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 295                          const char *buf, size_t len)
 296 {
 297         return netdev_store(dev, attr, buf, len, change_mtu);
 298 }
 299 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 300
 301 static int change_flags(struct net_device *dev, unsigned long new_flags)
 302 {
 303         return dev_change_flags(dev, (unsigned int) new_flags);
 304 }
 305
 306 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 307                            const char *buf, size_t len)
 308 {
 309         return netdev_store(dev, attr, buf, len, change_flags);
 310 }
 311 NETDEVICE_SHOW_RW(flags, fmt_hex);
 312
 313 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 314 {
 315         dev->tx_queue_len = new_len;
 316         return 0;
 317 }
 318
 319 static ssize_t tx_queue_len_store(struct device *dev,
 320                                   struct device_attribute *attr,
 321                                   const char *buf, size_t len)
 322 {
 323         if (!capable(CAP_NET_ADMIN))
 324                 return -EPERM;
 325
 326         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 327 }
 328 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
 329
 330 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 331 {
 332         dev->gro_flush_timeout = val;
 333         return 0;
 334 }
 335
 336 static ssize_t gro_flush_timeout_store(struct device *dev,
 337                                   struct device_attribute *attr,
 338                                   const char *buf, size_t len)
 339 {
 340         if (!capable(CAP_NET_ADMIN))
 341                 return -EPERM;
 342
 343         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 344 }
 345 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 346
 347 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 348                              const char *buf, size_t len)
 349 {
 350         struct net_device *netdev = to_net_dev(dev);
 351         struct net *net = dev_net(netdev);
 352         size_t count = len;
 353         ssize_t ret;
 354
 355         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 356                 return -EPERM;
 357
 358         /* ignore trailing newline */
 359         if (len >  0 && buf[len - 1] == '\n')
 360                 --count;
 361
 362         if (!rtnl_trylock())
 363                 return restart_syscall();
 364         ret = dev_set_alias(netdev, buf, count);
 365         rtnl_unlock();
 366
 367         return ret < 0 ? ret : len;
 368 }
 369
 370 static ssize_t ifalias_show(struct device *dev,
 371                             struct device_attribute *attr, char *buf)
 372 {
 373         const struct net_device *netdev = to_net_dev(dev);
 374         ssize_t ret = 0;
 375
 376         if (!rtnl_trylock())
 377                 return restart_syscall();
 378         if (netdev->ifalias)
 379                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 380         rtnl_unlock();
 381         return ret;
 382 }
 383 static DEVICE_ATTR_RW(ifalias);
 384
 385 static int change_group(struct net_device *dev, unsigned long new_group)
 386 {
 387         dev_set_group(dev, (int) new_group);
 388         return 0;
 389 }
 390
 391 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 392                            const char *buf, size_t len)
 393 {
 394         return netdev_store(dev, attr, buf, len, change_group);
 395 }
 396 NETDEVICE_SHOW(group, fmt_dec);
 397 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 398
 399 static ssize_t phys_port_id_show(struct device *dev,
 400                                  struct device_attribute *attr, char *buf)
 401 {
 402         struct net_device *netdev = to_net_dev(dev);
 403         ssize_t ret = -EINVAL;
 404
 405         if (!rtnl_trylock())
 406                 return restart_syscall();
 407
 408         if (dev_isalive(netdev)) {
 409                 struct netdev_phys_item_id ppid;
 410
 411                 ret = dev_get_phys_port_id(netdev, &ppid);
 412                 if (!ret)
 413                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 414         }
 415         rtnl_unlock();
 416
 417         return ret;
 418 }
 419 static DEVICE_ATTR_RO(phys_port_id);
 420
 421 static ssize_t phys_port_name_show(struct device *dev,
 422                                    struct device_attribute *attr, char *buf)
 423 {
 424         struct net_device *netdev = to_net_dev(dev);
 425         ssize_t ret = -EINVAL;
 426
 427         if (!rtnl_trylock())
 428                 return restart_syscall();
 429
 430         if (dev_isalive(netdev)) {
 431                 char name[IFNAMSIZ];
 432
 433                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 434                 if (!ret)
 435                         ret = sprintf(buf, "%s\n", name);
 436         }
 437         rtnl_unlock();
 438
 439         return ret;
 440 }
 441 static DEVICE_ATTR_RO(phys_port_name);
 442
 443 static ssize_t phys_switch_id_show(struct device *dev,
 444                                    struct device_attribute *attr, char *buf)
 445 {
 446         struct net_device *netdev = to_net_dev(dev);
 447         ssize_t ret = -EINVAL;
 448
 449         if (!rtnl_trylock())
 450                 return restart_syscall();
 451
 452         if (dev_isalive(netdev)) {
 453                 struct netdev_phys_item_id ppid;
 454
 455                 ret = netdev_switch_parent_id_get(netdev, &ppid);
 456                 if (!ret)
 457                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 458         }
 459         rtnl_unlock();
 460
 461         return ret;
 462 }
 463 static DEVICE_ATTR_RO(phys_switch_id);
 464
 465 static struct attribute *net_class_attrs[] = {
 466         &dev_attr_netdev_group.attr,
 467         &dev_attr_type.attr,
 468         &dev_attr_dev_id.attr,
 469         &dev_attr_dev_port.attr,
 470         &dev_attr_iflink.attr,
 471         &dev_attr_ifindex.attr,
 472         &dev_attr_name_assign_type.attr,
 473         &dev_attr_addr_assign_type.attr,
 474         &dev_attr_addr_len.attr,
 475         &dev_attr_link_mode.attr,
 476         &dev_attr_address.attr,
 477         &dev_attr_broadcast.attr,
 478         &dev_attr_speed.attr,
 479         &dev_attr_duplex.attr,
 480         &dev_attr_dormant.attr,
 481         &dev_attr_operstate.attr,
 482         &dev_attr_carrier_changes.attr,
 483         &dev_attr_ifalias.attr,
 484         &dev_attr_carrier.attr,
 485         &dev_attr_mtu.attr,
 486         &dev_attr_flags.attr,
 487         &dev_attr_tx_queue_len.attr,
 488         &dev_attr_gro_flush_timeout.attr,
 489         &dev_attr_phys_port_id.attr,
 490         &dev_attr_phys_port_name.attr,
 491         &dev_attr_phys_switch_id.attr,
 492         NULL,
 493 };
 494 ATTRIBUTE_GROUPS(net_class);
 495
 496 /* Show a given an attribute in the statistics group */
 497 static ssize_t netstat_show(const struct device *d,
 498                             struct device_attribute *attr, char *buf,
 499                             unsigned long offset)
 500 {
 501         struct net_device *dev = to_net_dev(d);
 502         ssize_t ret = -EINVAL;
 503
 504         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 505                         offset % sizeof(u64) != 0);
 506
 507         read_lock(&dev_base_lock);
 508         if (dev_isalive(dev)) {
 509                 struct rtnl_link_stats64 temp;
 510                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 511
 512                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 513         }
 514         read_unlock(&dev_base_lock);
 515         return ret;
 516 }
 517
 518 /* generate a read-only statistics attribute */
 519 #define NETSTAT_ENTRY(name)                                             \
 520 static ssize_t name##_show(struct device *d,                            \
 521                            struct device_attribute *attr, char *buf)    \
 522 {                                                                       \
 523         return netstat_show(d, attr, buf,                               \
 524                             offsetof(struct rtnl_link_stats64, name));  \
 525 }                                                                       \
 526 static DEVICE_ATTR_RO(name)
 527
 528 NETSTAT_ENTRY(rx_packets);
 529 NETSTAT_ENTRY(tx_packets);
 530 NETSTAT_ENTRY(rx_bytes);
 531 NETSTAT_ENTRY(tx_bytes);
 532 NETSTAT_ENTRY(rx_errors);
 533 NETSTAT_ENTRY(tx_errors);
 534 NETSTAT_ENTRY(rx_dropped);
 535 NETSTAT_ENTRY(tx_dropped);
 536 NETSTAT_ENTRY(multicast);
 537 NETSTAT_ENTRY(collisions);
 538 NETSTAT_ENTRY(rx_length_errors);
 539 NETSTAT_ENTRY(rx_over_errors);
 540 NETSTAT_ENTRY(rx_crc_errors);
 541 NETSTAT_ENTRY(rx_frame_errors);
 542 NETSTAT_ENTRY(rx_fifo_errors);
 543 NETSTAT_ENTRY(rx_missed_errors);
 544 NETSTAT_ENTRY(tx_aborted_errors);
 545 NETSTAT_ENTRY(tx_carrier_errors);
 546 NETSTAT_ENTRY(tx_fifo_errors);
 547 NETSTAT_ENTRY(tx_heartbeat_errors);
 548 NETSTAT_ENTRY(tx_window_errors);
 549 NETSTAT_ENTRY(rx_compressed);
 550 NETSTAT_ENTRY(tx_compressed);
 551
 552 static struct attribute *netstat_attrs[] = {
 553         &dev_attr_rx_packets.attr,
 554         &dev_attr_tx_packets.attr,
 555         &dev_attr_rx_bytes.attr,
 556         &dev_attr_tx_bytes.attr,
 557         &dev_attr_rx_errors.attr,
 558         &dev_attr_tx_errors.attr,
 559         &dev_attr_rx_dropped.attr,
 560         &dev_attr_tx_dropped.attr,
 561         &dev_attr_multicast.attr,
 562         &dev_attr_collisions.attr,
 563         &dev_attr_rx_length_errors.attr,
 564         &dev_attr_rx_over_errors.attr,
 565         &dev_attr_rx_crc_errors.attr,
 566         &dev_attr_rx_frame_errors.attr,
 567         &dev_attr_rx_fifo_errors.attr,
 568         &dev_attr_rx_missed_errors.attr,
 569         &dev_attr_tx_aborted_errors.attr,
 570         &dev_attr_tx_carrier_errors.attr,
 571         &dev_attr_tx_fifo_errors.attr,
 572         &dev_attr_tx_heartbeat_errors.attr,
 573         &dev_attr_tx_window_errors.attr,
 574         &dev_attr_rx_compressed.attr,
 575         &dev_attr_tx_compressed.attr,
 576         NULL
 577 };
 578
 579
 580 static struct attribute_group netstat_group = {
 581         .name  = "statistics",
 582         .attrs  = netstat_attrs,
 583 };
 584
 585 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 586 static struct attribute *wireless_attrs[] = {
 587         NULL
 588 };
 589
 590 static struct attribute_group wireless_group = {
 591         .name = "wireless",
 592         .attrs = wireless_attrs,
 593 };
 594 #endif
 595
 596 #else /* CONFIG_SYSFS */
 597 #define net_class_groups        NULL
 598 #endif /* CONFIG_SYSFS */
 599
 600 #ifdef CONFIG_SYSFS
 601 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 602     struct rx_queue_attribute, attr)
 603
 604 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 605
 606 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 607                                   char *buf)
 608 {
 609         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 610         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 611
 612         if (!attribute->show)
 613                 return -EIO;
 614
 615         return attribute->show(queue, attribute, buf);
 616 }
 617
 618 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 619                                    const char *buf, size_t count)
 620 {
 621         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 622         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 623
 624         if (!attribute->store)
 625                 return -EIO;
 626
 627         return attribute->store(queue, attribute, buf, count);
 628 }
 629
 630 static const struct sysfs_ops rx_queue_sysfs_ops = {
 631         .show = rx_queue_attr_show,
 632         .store = rx_queue_attr_store,
 633 };
 634
 635 #ifdef CONFIG_RPS
 636 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 637                             struct rx_queue_attribute *attribute, char *buf)
 638 {
 639         struct rps_map *map;
 640         cpumask_var_t mask;
 641         int i, len;
 642
 643         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 644                 return -ENOMEM;
 645
 646         rcu_read_lock();
 647         map = rcu_dereference(queue->rps_map);
 648         if (map)
 649                 for (i = 0; i < map->len; i++)
 650                         cpumask_set_cpu(map->cpus[i], mask);
 651
 652         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 653         rcu_read_unlock();
 654         free_cpumask_var(mask);
 655
 656         return len < PAGE_SIZE ? len : -EINVAL;
 657 }
 658
 659 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 660                       struct rx_queue_attribute *attribute,
 661                       const char *buf, size_t len)
 662 {
 663         struct rps_map *old_map, *map;
 664         cpumask_var_t mask;
 665         int err, cpu, i;
 666         static DEFINE_SPINLOCK(rps_map_lock);
 667
 668         if (!capable(CAP_NET_ADMIN))
 669                 return -EPERM;
 670
 671         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 672                 return -ENOMEM;
 673
 674         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 675         if (err) {
 676                 free_cpumask_var(mask);
 677                 return err;
 678         }
 679
 680         map = kzalloc(max_t(unsigned int,
 681             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 682             GFP_KERNEL);
 683         if (!map) {
 684                 free_cpumask_var(mask);
 685                 return -ENOMEM;
 686         }
 687
 688         i = 0;
 689         for_each_cpu_and(cpu, mask, cpu_online_mask)
 690                 map->cpus[i++] = cpu;
 691
 692         if (i)
 693                 map->len = i;
 694         else {
 695                 kfree(map);
 696                 map = NULL;
 697         }
 698
 699         spin_lock(&rps_map_lock);
 700         old_map = rcu_dereference_protected(queue->rps_map,
 701                                             lockdep_is_held(&rps_map_lock));
 702         rcu_assign_pointer(queue->rps_map, map);
 703         spin_unlock(&rps_map_lock);
 704
 705         if (map)
 706                 static_key_slow_inc(&rps_needed);
 707         if (old_map) {
 708                 kfree_rcu(old_map, rcu);
 709                 static_key_slow_dec(&rps_needed);
 710         }
 711         free_cpumask_var(mask);
 712         return len;
 713 }
 714
 715 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 716                                            struct rx_queue_attribute *attr,
 717                                            char *buf)
 718 {
 719         struct rps_dev_flow_table *flow_table;
 720         unsigned long val = 0;
 721
 722         rcu_read_lock();
 723         flow_table = rcu_dereference(queue->rps_flow_table);
 724         if (flow_table)
 725                 val = (unsigned long)flow_table->mask + 1;
 726         rcu_read_unlock();
 727
 728         return sprintf(buf, "%lu\n", val);
 729 }
 730
 731 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 732 {
 733         struct rps_dev_flow_table *table = container_of(rcu,
 734             struct rps_dev_flow_table, rcu);
 735         vfree(table);
 736 }
 737
 738 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 739                                      struct rx_queue_attribute *attr,
 740                                      const char *buf, size_t len)
 741 {
 742         unsigned long mask, count;
 743         struct rps_dev_flow_table *table, *old_table;
 744         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 745         int rc;
 746
 747         if (!capable(CAP_NET_ADMIN))
 748                 return -EPERM;
 749
 750         rc = kstrtoul(buf, 0, &count);
 751         if (rc < 0)
 752                 return rc;
 753
 754         if (count) {
 755                 mask = count - 1;
 756                 /* mask = roundup_pow_of_two(count) - 1;
 757                  * without overflows...
 758                  */
 759                 while ((mask | (mask >> 1)) != mask)
 760                         mask |= (mask >> 1);
 761                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 762                  * and on 32bit arches, must check
 763                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 764                  */
 765 #if BITS_PER_LONG > 32
 766                 if (mask > (unsigned long)(u32)mask)
 767                         return -EINVAL;
 768 #else
 769                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 770                                 / sizeof(struct rps_dev_flow)) {
 771                         /* Enforce a limit to prevent overflow */
 772                         return -EINVAL;
 773                 }
 774 #endif
 775                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 776                 if (!table)
 777                         return -ENOMEM;
 778
 779                 table->mask = mask;
 780                 for (count = 0; count <= mask; count++)
 781                         table->flows[count].cpu = RPS_NO_CPU;
 782         } else
 783                 table = NULL;
 784
 785         spin_lock(&rps_dev_flow_lock);
 786         old_table = rcu_dereference_protected(queue->rps_flow_table,
 787                                               lockdep_is_held(&rps_dev_flow_lock));
 788         rcu_assign_pointer(queue->rps_flow_table, table);
 789         spin_unlock(&rps_dev_flow_lock);
 790
 791         if (old_table)
 792                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 793
 794         return len;
 795 }
 796
 797 static struct rx_queue_attribute rps_cpus_attribute =
 798         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 799
 800
 801 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 802         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 803             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 804 #endif /* CONFIG_RPS */
 805
 806 static struct attribute *rx_queue_default_attrs[] = {
 807 #ifdef CONFIG_RPS
 808         &rps_cpus_attribute.attr,
 809         &rps_dev_flow_table_cnt_attribute.attr,
 810 #endif
 811         NULL
 812 };
 813
 814 static void rx_queue_release(struct kobject *kobj)
 815 {
 816         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 817 #ifdef CONFIG_RPS
 818         struct rps_map *map;
 819         struct rps_dev_flow_table *flow_table;
 820
 821
 822         map = rcu_dereference_protected(queue->rps_map, 1);
 823         if (map) {
 824                 RCU_INIT_POINTER(queue->rps_map, NULL);
 825                 kfree_rcu(map, rcu);
 826         }
 827
 828         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 829         if (flow_table) {
 830                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 831                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 832         }
 833 #endif
 834
 835         memset(kobj, 0, sizeof(*kobj));
 836         dev_put(queue->dev);
 837 }
 838
 839 static const void *rx_queue_namespace(struct kobject *kobj)
 840 {
 841         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 842         struct device *dev = &queue->dev->dev;
 843         const void *ns = NULL;
 844
 845         if (dev->class && dev->class->ns_type)
 846                 ns = dev->class->namespace(dev);
 847
 848         return ns;
 849 }
 850
 851 static struct kobj_type rx_queue_ktype = {
 852         .sysfs_ops = &rx_queue_sysfs_ops,
 853         .release = rx_queue_release,
 854         .default_attrs = rx_queue_default_attrs,
 855         .namespace = rx_queue_namespace
 856 };
 857
 858 static int rx_queue_add_kobject(struct net_device *dev, int index)
 859 {
 860         struct netdev_rx_queue *queue = dev->_rx + index;
 861         struct kobject *kobj = &queue->kobj;
 862         int error = 0;
 863
 864         kobj->kset = dev->queues_kset;
 865         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 866             "rx-%u", index);
 867         if (error)
 868                 goto exit;
 869
 870         if (dev->sysfs_rx_queue_group) {
 871                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 872                 if (error)
 873                         goto exit;
 874         }
 875
 876         kobject_uevent(kobj, KOBJ_ADD);
 877         dev_hold(queue->dev);
 878
 879         return error;
 880 exit:
 881         kobject_put(kobj);
 882         return error;
 883 }
 884 #endif /* CONFIG_SYSFS */
 885
 886 int
 887 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 888 {
 889 #ifdef CONFIG_SYSFS
 890         int i;
 891         int error = 0;
 892
 893 #ifndef CONFIG_RPS
 894         if (!dev->sysfs_rx_queue_group)
 895                 return 0;
 896 #endif
 897         for (i = old_num; i < new_num; i++) {
 898                 error = rx_queue_add_kobject(dev, i);
 899                 if (error) {
 900                         new_num = old_num;
 901                         break;
 902                 }
 903         }
 904
 905         while (--i >= new_num) {
 906                 if (dev->sysfs_rx_queue_group)
 907                         sysfs_remove_group(&dev->_rx[i].kobj,
 908                                            dev->sysfs_rx_queue_group);
 909                 kobject_put(&dev->_rx[i].kobj);
 910         }
 911
 912         return error;
 913 #else
 914         return 0;
 915 #endif
 916 }
 917
 918 #ifdef CONFIG_SYSFS
 919 /*
 920  * netdev_queue sysfs structures and functions.
 921  */
 922 struct netdev_queue_attribute {
 923         struct attribute attr;
 924         ssize_t (*show)(struct netdev_queue *queue,
 925             struct netdev_queue_attribute *attr, char *buf);
 926         ssize_t (*store)(struct netdev_queue *queue,
 927             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 928 };
 929 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 930     struct netdev_queue_attribute, attr)
 931
 932 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 933
 934 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 935                                       struct attribute *attr, char *buf)
 936 {
 937         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 938         struct netdev_queue *queue = to_netdev_queue(kobj);
 939
 940         if (!attribute->show)
 941                 return -EIO;
 942
 943         return attribute->show(queue, attribute, buf);
 944 }
 945
 946 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 947                                        struct attribute *attr,
 948                                        const char *buf, size_t count)
 949 {
 950         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 951         struct netdev_queue *queue = to_netdev_queue(kobj);
 952
 953         if (!attribute->store)
 954                 return -EIO;
 955
 956         return attribute->store(queue, attribute, buf, count);
 957 }
 958
 959 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 960         .show = netdev_queue_attr_show,
 961         .store = netdev_queue_attr_store,
 962 };
 963
 964 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 965                                   struct netdev_queue_attribute *attribute,
 966                                   char *buf)
 967 {
 968         unsigned long trans_timeout;
 969
 970         spin_lock_irq(&queue->_xmit_lock);
 971         trans_timeout = queue->trans_timeout;
 972         spin_unlock_irq(&queue->_xmit_lock);
 973
 974         return sprintf(buf, "%lu", trans_timeout);
 975 }
 976
 977 #ifdef CONFIG_XPS
 978 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 979 {
 980         struct net_device *dev = queue->dev;
 981         int i;
 982
 983         for (i = 0; i < dev->num_tx_queues; i++)
 984                 if (queue == &dev->_tx[i])
 985                         break;
 986
 987         BUG_ON(i >= dev->num_tx_queues);
 988
 989         return i;
 990 }
 991
 992 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
 993                                struct netdev_queue_attribute *attribute,
 994                                char *buf)
 995 {
 996         return sprintf(buf, "%lu\n", queue->tx_maxrate);
 997 }
 998
 999 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1000                               struct netdev_queue_attribute *attribute,
1001                               const char *buf, size_t len)
1002 {
1003         struct net_device *dev = queue->dev;
1004         int err, index = get_netdev_queue_index(queue);
1005         u32 rate = 0;
1006
1007         err = kstrtou32(buf, 10, &rate);
1008         if (err < 0)
1009                 return err;
1010
1011         if (!rtnl_trylock())
1012                 return restart_syscall();
1013
1014         err = -EOPNOTSUPP;
1015         if (dev->netdev_ops->ndo_set_tx_maxrate)
1016                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1017
1018         rtnl_unlock();
1019         if (!err) {
1020                 queue->tx_maxrate = rate;
1021                 return len;
1022         }
1023         return err;
1024 }
1025
1026 static struct netdev_queue_attribute queue_tx_maxrate =
1027         __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1028                show_tx_maxrate, set_tx_maxrate);
1029 #endif
1030
1031 static struct netdev_queue_attribute queue_trans_timeout =
1032         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1033
1034 #ifdef CONFIG_BQL
1035 /*
1036  * Byte queue limits sysfs structures and functions.
1037  */
1038 static ssize_t bql_show(char *buf, unsigned int value)
1039 {
1040         return sprintf(buf, "%u\n", value);
1041 }
1042
1043 static ssize_t bql_set(const char *buf, const size_t count,
1044                        unsigned int *pvalue)
1045 {
1046         unsigned int value;
1047         int err;
1048
1049         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1050                 value = DQL_MAX_LIMIT;
1051         else {
1052                 err = kstrtouint(buf, 10, &value);
1053                 if (err < 0)
1054                         return err;
1055                 if (value > DQL_MAX_LIMIT)
1056                         return -EINVAL;
1057         }
1058
1059         *pvalue = value;
1060
1061         return count;
1062 }
1063
1064 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1065                                   struct netdev_queue_attribute *attr,
1066                                   char *buf)
1067 {
1068         struct dql *dql = &queue->dql;
1069
1070         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1071 }
1072
1073 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1074                                  struct netdev_queue_attribute *attribute,
1075                                  const char *buf, size_t len)
1076 {
1077         struct dql *dql = &queue->dql;
1078         unsigned int value;
1079         int err;
1080
1081         err = kstrtouint(buf, 10, &value);
1082         if (err < 0)
1083                 return err;
1084
1085         dql->slack_hold_time = msecs_to_jiffies(value);
1086
1087         return len;
1088 }
1089
1090 static struct netdev_queue_attribute bql_hold_time_attribute =
1091         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1092             bql_set_hold_time);
1093
1094 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1095                                  struct netdev_queue_attribute *attr,
1096                                  char *buf)
1097 {
1098         struct dql *dql = &queue->dql;
1099
1100         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1101 }
1102
1103 static struct netdev_queue_attribute bql_inflight_attribute =
1104         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1105
1106 #define BQL_ATTR(NAME, FIELD)                                           \
1107 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1108                                  struct netdev_queue_attribute *attr,   \
1109                                  char *buf)                             \
1110 {                                                                       \
1111         return bql_show(buf, queue->dql.FIELD);                         \
1112 }                                                                       \
1113                                                                         \
1114 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1115                                 struct netdev_queue_attribute *attr,    \
1116                                 const char *buf, size_t len)            \
1117 {                                                                       \
1118         return bql_set(buf, len, &queue->dql.FIELD);                    \
1119 }                                                                       \
1120                                                                         \
1121 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1122         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1123             bql_set_ ## NAME);
1124
1125 BQL_ATTR(limit, limit)
1126 BQL_ATTR(limit_max, max_limit)
1127 BQL_ATTR(limit_min, min_limit)
1128
1129 static struct attribute *dql_attrs[] = {
1130         &bql_limit_attribute.attr,
1131         &bql_limit_max_attribute.attr,
1132         &bql_limit_min_attribute.attr,
1133         &bql_hold_time_attribute.attr,
1134         &bql_inflight_attribute.attr,
1135         NULL
1136 };
1137
1138 static struct attribute_group dql_group = {
1139         .name  = "byte_queue_limits",
1140         .attrs  = dql_attrs,
1141 };
1142 #endif /* CONFIG_BQL */
1143
1144 #ifdef CONFIG_XPS
1145 static ssize_t show_xps_map(struct netdev_queue *queue,
1146                             struct netdev_queue_attribute *attribute, char *buf)
1147 {
1148         struct net_device *dev = queue->dev;
1149         struct xps_dev_maps *dev_maps;
1150         cpumask_var_t mask;
1151         unsigned long index;
1152         int i, len;
1153
1154         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1155                 return -ENOMEM;
1156
1157         index = get_netdev_queue_index(queue);
1158
1159         rcu_read_lock();
1160         dev_maps = rcu_dereference(dev->xps_maps);
1161         if (dev_maps) {
1162                 for_each_possible_cpu(i) {
1163                         struct xps_map *map =
1164                             rcu_dereference(dev_maps->cpu_map[i]);
1165                         if (map) {
1166                                 int j;
1167                                 for (j = 0; j < map->len; j++) {
1168                                         if (map->queues[j] == index) {
1169                                                 cpumask_set_cpu(i, mask);
1170                                                 break;
1171                                         }
1172                                 }
1173                         }
1174                 }
1175         }
1176         rcu_read_unlock();
1177
1178         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1179         free_cpumask_var(mask);
1180         return len < PAGE_SIZE ? len : -EINVAL;
1181 }
1182
1183 static ssize_t store_xps_map(struct netdev_queue *queue,
1184                       struct netdev_queue_attribute *attribute,
1185                       const char *buf, size_t len)
1186 {
1187         struct net_device *dev = queue->dev;
1188         unsigned long index;
1189         cpumask_var_t mask;
1190         int err;
1191
1192         if (!capable(CAP_NET_ADMIN))
1193                 return -EPERM;
1194
1195         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1196                 return -ENOMEM;
1197
1198         index = get_netdev_queue_index(queue);
1199
1200         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1201         if (err) {
1202                 free_cpumask_var(mask);
1203                 return err;
1204         }
1205
1206         err = netif_set_xps_queue(dev, mask, index);
1207
1208         free_cpumask_var(mask);
1209
1210         return err ? : len;
1211 }
1212
1213 static struct netdev_queue_attribute xps_cpus_attribute =
1214     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1215 #endif /* CONFIG_XPS */
1216
1217 static struct attribute *netdev_queue_default_attrs[] = {
1218         &queue_trans_timeout.attr,
1219 #ifdef CONFIG_XPS
1220         &xps_cpus_attribute.attr,
1221         &queue_tx_maxrate.attr,
1222 #endif
1223         NULL
1224 };
1225
1226 static void netdev_queue_release(struct kobject *kobj)
1227 {
1228         struct netdev_queue *queue = to_netdev_queue(kobj);
1229
1230         memset(kobj, 0, sizeof(*kobj));
1231         dev_put(queue->dev);
1232 }
1233
1234 static const void *netdev_queue_namespace(struct kobject *kobj)
1235 {
1236         struct netdev_queue *queue = to_netdev_queue(kobj);
1237         struct device *dev = &queue->dev->dev;
1238         const void *ns = NULL;
1239
1240         if (dev->class && dev->class->ns_type)
1241                 ns = dev->class->namespace(dev);
1242
1243         return ns;
1244 }
1245
1246 static struct kobj_type netdev_queue_ktype = {
1247         .sysfs_ops = &netdev_queue_sysfs_ops,
1248         .release = netdev_queue_release,
1249         .default_attrs = netdev_queue_default_attrs,
1250         .namespace = netdev_queue_namespace,
1251 };
1252
1253 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1254 {
1255         struct netdev_queue *queue = dev->_tx + index;
1256         struct kobject *kobj = &queue->kobj;
1257         int error = 0;
1258
1259         kobj->kset = dev->queues_kset;
1260         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1261             "tx-%u", index);
1262         if (error)
1263                 goto exit;
1264
1265 #ifdef CONFIG_BQL
1266         error = sysfs_create_group(kobj, &dql_group);
1267         if (error)
1268                 goto exit;
1269 #endif
1270
1271         kobject_uevent(kobj, KOBJ_ADD);
1272         dev_hold(queue->dev);
1273
1274         return 0;
1275 exit:
1276         kobject_put(kobj);
1277         return error;
1278 }
1279 #endif /* CONFIG_SYSFS */
1280
1281 int
1282 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1283 {
1284 #ifdef CONFIG_SYSFS
1285         int i;
1286         int error = 0;
1287
1288         for (i = old_num; i < new_num; i++) {
1289                 error = netdev_queue_add_kobject(dev, i);
1290                 if (error) {
1291                         new_num = old_num;
1292                         break;
1293                 }
1294         }
1295
1296         while (--i >= new_num) {
1297                 struct netdev_queue *queue = dev->_tx + i;
1298
1299 #ifdef CONFIG_BQL
1300                 sysfs_remove_group(&queue->kobj, &dql_group);
1301 #endif
1302                 kobject_put(&queue->kobj);
1303         }
1304
1305         return error;
1306 #else
1307         return 0;
1308 #endif /* CONFIG_SYSFS */
1309 }
1310
1311 static int register_queue_kobjects(struct net_device *dev)
1312 {
1313         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1314
1315 #ifdef CONFIG_SYSFS
1316         dev->queues_kset = kset_create_and_add("queues",
1317             NULL, &dev->dev.kobj);
1318         if (!dev->queues_kset)
1319                 return -ENOMEM;
1320         real_rx = dev->real_num_rx_queues;
1321 #endif
1322         real_tx = dev->real_num_tx_queues;
1323
1324         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1325         if (error)
1326                 goto error;
1327         rxq = real_rx;
1328
1329         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1330         if (error)
1331                 goto error;
1332         txq = real_tx;
1333
1334         return 0;
1335
1336 error:
1337         netdev_queue_update_kobjects(dev, txq, 0);
1338         net_rx_queue_update_kobjects(dev, rxq, 0);
1339         return error;
1340 }
1341
1342 static void remove_queue_kobjects(struct net_device *dev)
1343 {
1344         int real_rx = 0, real_tx = 0;
1345
1346 #ifdef CONFIG_SYSFS
1347         real_rx = dev->real_num_rx_queues;
1348 #endif
1349         real_tx = dev->real_num_tx_queues;
1350
1351         net_rx_queue_update_kobjects(dev, real_rx, 0);
1352         netdev_queue_update_kobjects(dev, real_tx, 0);
1353 #ifdef CONFIG_SYSFS
1354         kset_unregister(dev->queues_kset);
1355 #endif
1356 }
1357
1358 static bool net_current_may_mount(void)
1359 {
1360         struct net *net = current->nsproxy->net_ns;
1361
1362         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1363 }
1364
1365 static void *net_grab_current_ns(void)
1366 {
1367         struct net *ns = current->nsproxy->net_ns;
1368 #ifdef CONFIG_NET_NS
1369         if (ns)
1370                 atomic_inc(&ns->passive);
1371 #endif
1372         return ns;
1373 }
1374
1375 static const void *net_initial_ns(void)
1376 {
1377         return &init_net;
1378 }
1379
1380 static const void *net_netlink_ns(struct sock *sk)
1381 {
1382         return sock_net(sk);
1383 }
1384
1385 struct kobj_ns_type_operations net_ns_type_operations = {
1386         .type = KOBJ_NS_TYPE_NET,
1387         .current_may_mount = net_current_may_mount,
1388         .grab_current_ns = net_grab_current_ns,
1389         .netlink_ns = net_netlink_ns,
1390         .initial_ns = net_initial_ns,
1391         .drop_ns = net_drop_ns,
1392 };
1393 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1394
1395 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1396 {
1397         struct net_device *dev = to_net_dev(d);
1398         int retval;
1399
1400         /* pass interface to uevent. */
1401         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1402         if (retval)
1403                 goto exit;
1404
1405         /* pass ifindex to uevent.
1406          * ifindex is useful as it won't change (interface name may change)
1407          * and is what RtNetlink uses natively. */
1408         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1409
1410 exit:
1411         return retval;
1412 }
1413
1414 /*
1415  *      netdev_release -- destroy and free a dead device.
1416  *      Called when last reference to device kobject is gone.
1417  */
1418 static void netdev_release(struct device *d)
1419 {
1420         struct net_device *dev = to_net_dev(d);
1421
1422         BUG_ON(dev->reg_state != NETREG_RELEASED);
1423
1424         kfree(dev->ifalias);
1425         netdev_freemem(dev);
1426 }
1427
1428 static const void *net_namespace(struct device *d)
1429 {
1430         struct net_device *dev;
1431         dev = container_of(d, struct net_device, dev);
1432         return dev_net(dev);
1433 }
1434
1435 static struct class net_class = {
1436         .name = "net",
1437         .dev_release = netdev_release,
1438         .dev_groups = net_class_groups,
1439         .dev_uevent = netdev_uevent,
1440         .ns_type = &net_ns_type_operations,
1441         .namespace = net_namespace,
1442 };
1443
1444 #ifdef CONFIG_OF_NET
1445 static int of_dev_node_match(struct device *dev, const void *data)
1446 {
1447         int ret = 0;
1448
1449         if (dev->parent)
1450                 ret = dev->parent->of_node == data;
1451
1452         return ret == 0 ? dev->of_node == data : ret;
1453 }
1454
1455 struct net_device *of_find_net_device_by_node(struct device_node *np)
1456 {
1457         struct device *dev;
1458
1459         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1460         if (!dev)
1461                 return NULL;
1462
1463         return to_net_dev(dev);
1464 }
1465 EXPORT_SYMBOL(of_find_net_device_by_node);
1466 #endif
1467
1468 /* Delete sysfs entries but hold kobject reference until after all
1469  * netdev references are gone.
1470  */
1471 void netdev_unregister_kobject(struct net_device *ndev)
1472 {
1473         struct device *dev = &(ndev->dev);
1474
1475         kobject_get(&dev->kobj);
1476
1477         remove_queue_kobjects(ndev);
1478
1479         pm_runtime_set_memalloc_noio(dev, false);
1480
1481         device_del(dev);
1482 }
1483
1484 /* Create sysfs entries for network device. */
1485 int netdev_register_kobject(struct net_device *ndev)
1486 {
1487         struct device *dev = &(ndev->dev);
1488         const struct attribute_group **groups = ndev->sysfs_groups;
1489         int error = 0;
1490
1491         device_initialize(dev);
1492         dev->class = &net_class;
1493         dev->platform_data = ndev;
1494         dev->groups = groups;
1495
1496         dev_set_name(dev, "%s", ndev->name);
1497
1498 #ifdef CONFIG_SYSFS
1499         /* Allow for a device specific group */
1500         if (*groups)
1501                 groups++;
1502
1503         *groups++ = &netstat_group;
1504
1505 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1506         if (ndev->ieee80211_ptr)
1507                 *groups++ = &wireless_group;
1508 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1509         else if (ndev->wireless_handlers)
1510                 *groups++ = &wireless_group;
1511 #endif
1512 #endif
1513 #endif /* CONFIG_SYSFS */
1514
1515         error = device_add(dev);
1516         if (error)
1517                 return error;
1518
1519         error = register_queue_kobjects(ndev);
1520         if (error) {
1521                 device_del(dev);
1522                 return error;
1523         }
1524
1525         pm_runtime_set_memalloc_noio(dev, true);
1526
1527         return error;
1528 }
1529
1530 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1531                                 const void *ns)
1532 {
1533         return class_create_file_ns(&net_class, class_attr, ns);
1534 }
1535 EXPORT_SYMBOL(netdev_class_create_file_ns);
1536
1537 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1538                                  const void *ns)
1539 {
1540         class_remove_file_ns(&net_class, class_attr, ns);
1541 }
1542 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1543
1544 int __init netdev_kobject_init(void)
1545 {
1546         kobj_ns_type_register(&net_ns_type_operations);
1547         return class_register(&net_class);
1548 }