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_ulong[] = "%lu\n";
  35 static const char fmt_u64[] = "%llu\n";
  36
  37 static inline int dev_isalive(const struct net_device *dev)
  38 {
  39         return dev->reg_state <= NETREG_REGISTERED;
  40 }
  41
  42 /* use same locking rules as GIF* ioctl's */
  43 static ssize_t netdev_show(const struct device *dev,
  44                            struct device_attribute *attr, char *buf,
  45                            ssize_t (*format)(const struct net_device *, char *))
  46 {
  47         struct net_device *ndev = to_net_dev(dev);
  48         ssize_t ret = -EINVAL;
  49
  50         read_lock(&dev_base_lock);
  51         if (dev_isalive(ndev))
  52                 ret = (*format)(ndev, buf);
  53         read_unlock(&dev_base_lock);
  54
  55         return ret;
  56 }
  57
  58 /* generate a show function for simple field */
  59 #define NETDEVICE_SHOW(field, format_string)                            \
  60 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  61 {                                                                       \
  62         return sprintf(buf, format_string, dev->field);                 \
  63 }                                                                       \
  64 static ssize_t field##_show(struct device *dev,                         \
  65                             struct device_attribute *attr, char *buf)   \
  66 {                                                                       \
  67         return netdev_show(dev, attr, buf, format_##field);             \
  68 }                                                                       \
  69
  70 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  71 NETDEVICE_SHOW(field, format_string);                                   \
  72 static DEVICE_ATTR_RO(field)
  73
  74 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  75 NETDEVICE_SHOW(field, format_string);                                   \
  76 static DEVICE_ATTR_RW(field)
  77
  78 /* use same locking and permission rules as SIF* ioctl's */
  79 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  80                             const char *buf, size_t len,
  81                             int (*set)(struct net_device *, unsigned long))
  82 {
  83         struct net_device *netdev = to_net_dev(dev);
  84         struct net *net = dev_net(netdev);
  85         unsigned long new;
  86         int ret = -EINVAL;
  87
  88         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  89                 return -EPERM;
  90
  91         ret = kstrtoul(buf, 0, &new);
  92         if (ret)
  93                 goto err;
  94
  95         if (!rtnl_trylock())
  96                 return restart_syscall();
  97
  98         if (dev_isalive(netdev)) {
  99                 if ((ret = (*set)(netdev, new)) == 0)
 100                         ret = len;
 101         }
 102         rtnl_unlock();
 103  err:
 104         return ret;
 105 }
 106
 107 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 108 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 109 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 110 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 111 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 112 NETDEVICE_SHOW_RO(type, fmt_dec);
 113 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 114
 115 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 116                            char *buf)
 117 {
 118         struct net_device *ndev = to_net_dev(dev);
 119
 120         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
 121 }
 122 static DEVICE_ATTR_RO(iflink);
 123
 124 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 125 {
 126         return sprintf(buf, fmt_dec, dev->name_assign_type);
 127 }
 128
 129 static ssize_t name_assign_type_show(struct device *dev,
 130                                      struct device_attribute *attr,
 131                                      char *buf)
 132 {
 133         struct net_device *ndev = to_net_dev(dev);
 134         ssize_t ret = -EINVAL;
 135
 136         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 137                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 138
 139         return ret;
 140 }
 141 static DEVICE_ATTR_RO(name_assign_type);
 142
 143 /* use same locking rules as GIFHWADDR ioctl's */
 144 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 145                             char *buf)
 146 {
 147         struct net_device *ndev = to_net_dev(dev);
 148         ssize_t ret = -EINVAL;
 149
 150         read_lock(&dev_base_lock);
 151         if (dev_isalive(ndev))
 152                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 153         read_unlock(&dev_base_lock);
 154         return ret;
 155 }
 156 static DEVICE_ATTR_RO(address);
 157
 158 static ssize_t broadcast_show(struct device *dev,
 159                               struct device_attribute *attr, char *buf)
 160 {
 161         struct net_device *ndev = to_net_dev(dev);
 162         if (dev_isalive(ndev))
 163                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 164         return -EINVAL;
 165 }
 166 static DEVICE_ATTR_RO(broadcast);
 167
 168 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 169 {
 170         if (!netif_running(dev))
 171                 return -EINVAL;
 172         return dev_change_carrier(dev, (bool) new_carrier);
 173 }
 174
 175 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 176                              const char *buf, size_t len)
 177 {
 178         return netdev_store(dev, attr, buf, len, change_carrier);
 179 }
 180
 181 static ssize_t carrier_show(struct device *dev,
 182                             struct device_attribute *attr, char *buf)
 183 {
 184         struct net_device *netdev = to_net_dev(dev);
 185         if (netif_running(netdev)) {
 186                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 187         }
 188         return -EINVAL;
 189 }
 190 static DEVICE_ATTR_RW(carrier);
 191
 192 static ssize_t speed_show(struct device *dev,
 193                           struct device_attribute *attr, char *buf)
 194 {
 195         struct net_device *netdev = to_net_dev(dev);
 196         int ret = -EINVAL;
 197
 198         if (!rtnl_trylock())
 199                 return restart_syscall();
 200
 201         if (netif_running(netdev)) {
 202                 struct ethtool_cmd cmd;
 203                 if (!__ethtool_get_settings(netdev, &cmd))
 204                         ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
 205         }
 206         rtnl_unlock();
 207         return ret;
 208 }
 209 static DEVICE_ATTR_RO(speed);
 210
 211 static ssize_t duplex_show(struct device *dev,
 212                            struct device_attribute *attr, char *buf)
 213 {
 214         struct net_device *netdev = to_net_dev(dev);
 215         int ret = -EINVAL;
 216
 217         if (!rtnl_trylock())
 218                 return restart_syscall();
 219
 220         if (netif_running(netdev)) {
 221                 struct ethtool_cmd cmd;
 222                 if (!__ethtool_get_settings(netdev, &cmd)) {
 223                         const char *duplex;
 224                         switch (cmd.duplex) {
 225                         case DUPLEX_HALF:
 226                                 duplex = "half";
 227                                 break;
 228                         case DUPLEX_FULL:
 229                                 duplex = "full";
 230                                 break;
 231                         default:
 232                                 duplex = "unknown";
 233                                 break;
 234                         }
 235                         ret = sprintf(buf, "%s\n", duplex);
 236                 }
 237         }
 238         rtnl_unlock();
 239         return ret;
 240 }
 241 static DEVICE_ATTR_RO(duplex);
 242
 243 static ssize_t dormant_show(struct device *dev,
 244                             struct device_attribute *attr, char *buf)
 245 {
 246         struct net_device *netdev = to_net_dev(dev);
 247
 248         if (netif_running(netdev))
 249                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 250
 251         return -EINVAL;
 252 }
 253 static DEVICE_ATTR_RO(dormant);
 254
 255 static const char *const operstates[] = {
 256         "unknown",
 257         "notpresent", /* currently unused */
 258         "down",
 259         "lowerlayerdown",
 260         "testing", /* currently unused */
 261         "dormant",
 262         "up"
 263 };
 264
 265 static ssize_t operstate_show(struct device *dev,
 266                               struct device_attribute *attr, char *buf)
 267 {
 268         const struct net_device *netdev = to_net_dev(dev);
 269         unsigned char operstate;
 270
 271         read_lock(&dev_base_lock);
 272         operstate = netdev->operstate;
 273         if (!netif_running(netdev))
 274                 operstate = IF_OPER_DOWN;
 275         read_unlock(&dev_base_lock);
 276
 277         if (operstate >= ARRAY_SIZE(operstates))
 278                 return -EINVAL; /* should not happen */
 279
 280         return sprintf(buf, "%s\n", operstates[operstate]);
 281 }
 282 static DEVICE_ATTR_RO(operstate);
 283
 284 static ssize_t carrier_changes_show(struct device *dev,
 285                                     struct device_attribute *attr,
 286                                     char *buf)
 287 {
 288         struct net_device *netdev = to_net_dev(dev);
 289         return sprintf(buf, fmt_dec,
 290                        atomic_read(&netdev->carrier_changes));
 291 }
 292 static DEVICE_ATTR_RO(carrier_changes);
 293
 294 /* read-write attributes */
 295
 296 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 297 {
 298         return dev_set_mtu(dev, (int) new_mtu);
 299 }
 300
 301 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 302                          const char *buf, size_t len)
 303 {
 304         return netdev_store(dev, attr, buf, len, change_mtu);
 305 }
 306 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 307
 308 static int change_flags(struct net_device *dev, unsigned long new_flags)
 309 {
 310         return dev_change_flags(dev, (unsigned int) new_flags);
 311 }
 312
 313 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 314                            const char *buf, size_t len)
 315 {
 316         return netdev_store(dev, attr, buf, len, change_flags);
 317 }
 318 NETDEVICE_SHOW_RW(flags, fmt_hex);
 319
 320 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 321 {
 322         dev->tx_queue_len = new_len;
 323         return 0;
 324 }
 325
 326 static ssize_t tx_queue_len_store(struct device *dev,
 327                                   struct device_attribute *attr,
 328                                   const char *buf, size_t len)
 329 {
 330         if (!capable(CAP_NET_ADMIN))
 331                 return -EPERM;
 332
 333         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 334 }
 335 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
 336
 337 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 338 {
 339         dev->gro_flush_timeout = val;
 340         return 0;
 341 }
 342
 343 static ssize_t gro_flush_timeout_store(struct device *dev,
 344                                   struct device_attribute *attr,
 345                                   const char *buf, size_t len)
 346 {
 347         if (!capable(CAP_NET_ADMIN))
 348                 return -EPERM;
 349
 350         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 351 }
 352 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 353
 354 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 355                              const char *buf, size_t len)
 356 {
 357         struct net_device *netdev = to_net_dev(dev);
 358         struct net *net = dev_net(netdev);
 359         size_t count = len;
 360         ssize_t ret;
 361
 362         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 363                 return -EPERM;
 364
 365         /* ignore trailing newline */
 366         if (len >  0 && buf[len - 1] == '\n')
 367                 --count;
 368
 369         if (!rtnl_trylock())
 370                 return restart_syscall();
 371         ret = dev_set_alias(netdev, buf, count);
 372         rtnl_unlock();
 373
 374         return ret < 0 ? ret : len;
 375 }
 376
 377 static ssize_t ifalias_show(struct device *dev,
 378                             struct device_attribute *attr, char *buf)
 379 {
 380         const struct net_device *netdev = to_net_dev(dev);
 381         ssize_t ret = 0;
 382
 383         if (!rtnl_trylock())
 384                 return restart_syscall();
 385         if (netdev->ifalias)
 386                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 387         rtnl_unlock();
 388         return ret;
 389 }
 390 static DEVICE_ATTR_RW(ifalias);
 391
 392 static int change_group(struct net_device *dev, unsigned long new_group)
 393 {
 394         dev_set_group(dev, (int) new_group);
 395         return 0;
 396 }
 397
 398 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 399                            const char *buf, size_t len)
 400 {
 401         return netdev_store(dev, attr, buf, len, change_group);
 402 }
 403 NETDEVICE_SHOW(group, fmt_dec);
 404 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 405
 406 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 407 {
 408         return dev_change_proto_down(dev, (bool) proto_down);
 409 }
 410
 411 static ssize_t proto_down_store(struct device *dev,
 412                                 struct device_attribute *attr,
 413                                 const char *buf, size_t len)
 414 {
 415         return netdev_store(dev, attr, buf, len, change_proto_down);
 416 }
 417 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 418
 419 static ssize_t phys_port_id_show(struct device *dev,
 420                                  struct device_attribute *attr, char *buf)
 421 {
 422         struct net_device *netdev = to_net_dev(dev);
 423         ssize_t ret = -EINVAL;
 424
 425         if (!rtnl_trylock())
 426                 return restart_syscall();
 427
 428         if (dev_isalive(netdev)) {
 429                 struct netdev_phys_item_id ppid;
 430
 431                 ret = dev_get_phys_port_id(netdev, &ppid);
 432                 if (!ret)
 433                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 434         }
 435         rtnl_unlock();
 436
 437         return ret;
 438 }
 439 static DEVICE_ATTR_RO(phys_port_id);
 440
 441 static ssize_t phys_port_name_show(struct device *dev,
 442                                    struct device_attribute *attr, char *buf)
 443 {
 444         struct net_device *netdev = to_net_dev(dev);
 445         ssize_t ret = -EINVAL;
 446
 447         if (!rtnl_trylock())
 448                 return restart_syscall();
 449
 450         if (dev_isalive(netdev)) {
 451                 char name[IFNAMSIZ];
 452
 453                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 454                 if (!ret)
 455                         ret = sprintf(buf, "%s\n", name);
 456         }
 457         rtnl_unlock();
 458
 459         return ret;
 460 }
 461 static DEVICE_ATTR_RO(phys_port_name);
 462
 463 static ssize_t phys_switch_id_show(struct device *dev,
 464                                    struct device_attribute *attr, char *buf)
 465 {
 466         struct net_device *netdev = to_net_dev(dev);
 467         ssize_t ret = -EINVAL;
 468
 469         if (!rtnl_trylock())
 470                 return restart_syscall();
 471
 472         if (dev_isalive(netdev)) {
 473                 struct switchdev_attr attr = {
 474                         .orig_dev = netdev,
 475                         .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 476                         .flags = SWITCHDEV_F_NO_RECURSE,
 477                 };
 478
 479                 ret = switchdev_port_attr_get(netdev, &attr);
 480                 if (!ret)
 481                         ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
 482                                       attr.u.ppid.id);
 483         }
 484         rtnl_unlock();
 485
 486         return ret;
 487 }
 488 static DEVICE_ATTR_RO(phys_switch_id);
 489
 490 static struct attribute *net_class_attrs[] = {
 491         &dev_attr_netdev_group.attr,
 492         &dev_attr_type.attr,
 493         &dev_attr_dev_id.attr,
 494         &dev_attr_dev_port.attr,
 495         &dev_attr_iflink.attr,
 496         &dev_attr_ifindex.attr,
 497         &dev_attr_name_assign_type.attr,
 498         &dev_attr_addr_assign_type.attr,
 499         &dev_attr_addr_len.attr,
 500         &dev_attr_link_mode.attr,
 501         &dev_attr_address.attr,
 502         &dev_attr_broadcast.attr,
 503         &dev_attr_speed.attr,
 504         &dev_attr_duplex.attr,
 505         &dev_attr_dormant.attr,
 506         &dev_attr_operstate.attr,
 507         &dev_attr_carrier_changes.attr,
 508         &dev_attr_ifalias.attr,
 509         &dev_attr_carrier.attr,
 510         &dev_attr_mtu.attr,
 511         &dev_attr_flags.attr,
 512         &dev_attr_tx_queue_len.attr,
 513         &dev_attr_gro_flush_timeout.attr,
 514         &dev_attr_phys_port_id.attr,
 515         &dev_attr_phys_port_name.attr,
 516         &dev_attr_phys_switch_id.attr,
 517         &dev_attr_proto_down.attr,
 518         NULL,
 519 };
 520 ATTRIBUTE_GROUPS(net_class);
 521
 522 /* Show a given an attribute in the statistics group */
 523 static ssize_t netstat_show(const struct device *d,
 524                             struct device_attribute *attr, char *buf,
 525                             unsigned long offset)
 526 {
 527         struct net_device *dev = to_net_dev(d);
 528         ssize_t ret = -EINVAL;
 529
 530         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 531                         offset % sizeof(u64) != 0);
 532
 533         read_lock(&dev_base_lock);
 534         if (dev_isalive(dev)) {
 535                 struct rtnl_link_stats64 temp;
 536                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 537
 538                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 539         }
 540         read_unlock(&dev_base_lock);
 541         return ret;
 542 }
 543
 544 /* generate a read-only statistics attribute */
 545 #define NETSTAT_ENTRY(name)                                             \
 546 static ssize_t name##_show(struct device *d,                            \
 547                            struct device_attribute *attr, char *buf)    \
 548 {                                                                       \
 549         return netstat_show(d, attr, buf,                               \
 550                             offsetof(struct rtnl_link_stats64, name));  \
 551 }                                                                       \
 552 static DEVICE_ATTR_RO(name)
 553
 554 NETSTAT_ENTRY(rx_packets);
 555 NETSTAT_ENTRY(tx_packets);
 556 NETSTAT_ENTRY(rx_bytes);
 557 NETSTAT_ENTRY(tx_bytes);
 558 NETSTAT_ENTRY(rx_errors);
 559 NETSTAT_ENTRY(tx_errors);
 560 NETSTAT_ENTRY(rx_dropped);
 561 NETSTAT_ENTRY(tx_dropped);
 562 NETSTAT_ENTRY(multicast);
 563 NETSTAT_ENTRY(collisions);
 564 NETSTAT_ENTRY(rx_length_errors);
 565 NETSTAT_ENTRY(rx_over_errors);
 566 NETSTAT_ENTRY(rx_crc_errors);
 567 NETSTAT_ENTRY(rx_frame_errors);
 568 NETSTAT_ENTRY(rx_fifo_errors);
 569 NETSTAT_ENTRY(rx_missed_errors);
 570 NETSTAT_ENTRY(tx_aborted_errors);
 571 NETSTAT_ENTRY(tx_carrier_errors);
 572 NETSTAT_ENTRY(tx_fifo_errors);
 573 NETSTAT_ENTRY(tx_heartbeat_errors);
 574 NETSTAT_ENTRY(tx_window_errors);
 575 NETSTAT_ENTRY(rx_compressed);
 576 NETSTAT_ENTRY(tx_compressed);
 577
 578 static struct attribute *netstat_attrs[] = {
 579         &dev_attr_rx_packets.attr,
 580         &dev_attr_tx_packets.attr,
 581         &dev_attr_rx_bytes.attr,
 582         &dev_attr_tx_bytes.attr,
 583         &dev_attr_rx_errors.attr,
 584         &dev_attr_tx_errors.attr,
 585         &dev_attr_rx_dropped.attr,
 586         &dev_attr_tx_dropped.attr,
 587         &dev_attr_multicast.attr,
 588         &dev_attr_collisions.attr,
 589         &dev_attr_rx_length_errors.attr,
 590         &dev_attr_rx_over_errors.attr,
 591         &dev_attr_rx_crc_errors.attr,
 592         &dev_attr_rx_frame_errors.attr,
 593         &dev_attr_rx_fifo_errors.attr,
 594         &dev_attr_rx_missed_errors.attr,
 595         &dev_attr_tx_aborted_errors.attr,
 596         &dev_attr_tx_carrier_errors.attr,
 597         &dev_attr_tx_fifo_errors.attr,
 598         &dev_attr_tx_heartbeat_errors.attr,
 599         &dev_attr_tx_window_errors.attr,
 600         &dev_attr_rx_compressed.attr,
 601         &dev_attr_tx_compressed.attr,
 602         NULL
 603 };
 604
 605
 606 static struct attribute_group netstat_group = {
 607         .name  = "statistics",
 608         .attrs  = netstat_attrs,
 609 };
 610
 611 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 612 static struct attribute *wireless_attrs[] = {
 613         NULL
 614 };
 615
 616 static struct attribute_group wireless_group = {
 617         .name = "wireless",
 618         .attrs = wireless_attrs,
 619 };
 620 #endif
 621
 622 #else /* CONFIG_SYSFS */
 623 #define net_class_groups        NULL
 624 #endif /* CONFIG_SYSFS */
 625
 626 #ifdef CONFIG_SYSFS
 627 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 628     struct rx_queue_attribute, attr)
 629
 630 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 631
 632 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 633                                   char *buf)
 634 {
 635         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 636         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 637
 638         if (!attribute->show)
 639                 return -EIO;
 640
 641         return attribute->show(queue, attribute, buf);
 642 }
 643
 644 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 645                                    const char *buf, size_t count)
 646 {
 647         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 648         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 649
 650         if (!attribute->store)
 651                 return -EIO;
 652
 653         return attribute->store(queue, attribute, buf, count);
 654 }
 655
 656 static const struct sysfs_ops rx_queue_sysfs_ops = {
 657         .show = rx_queue_attr_show,
 658         .store = rx_queue_attr_store,
 659 };
 660
 661 #ifdef CONFIG_RPS
 662 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 663                             struct rx_queue_attribute *attribute, char *buf)
 664 {
 665         struct rps_map *map;
 666         cpumask_var_t mask;
 667         int i, len;
 668
 669         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 670                 return -ENOMEM;
 671
 672         rcu_read_lock();
 673         map = rcu_dereference(queue->rps_map);
 674         if (map)
 675                 for (i = 0; i < map->len; i++)
 676                         cpumask_set_cpu(map->cpus[i], mask);
 677
 678         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 679         rcu_read_unlock();
 680         free_cpumask_var(mask);
 681
 682         return len < PAGE_SIZE ? len : -EINVAL;
 683 }
 684
 685 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 686                       struct rx_queue_attribute *attribute,
 687                       const char *buf, size_t len)
 688 {
 689         struct rps_map *old_map, *map;
 690         cpumask_var_t mask;
 691         int err, cpu, i;
 692         static DEFINE_MUTEX(rps_map_mutex);
 693
 694         if (!capable(CAP_NET_ADMIN))
 695                 return -EPERM;
 696
 697         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 698                 return -ENOMEM;
 699
 700         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 701         if (err) {
 702                 free_cpumask_var(mask);
 703                 return err;
 704         }
 705
 706         map = kzalloc(max_t(unsigned int,
 707             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 708             GFP_KERNEL);
 709         if (!map) {
 710                 free_cpumask_var(mask);
 711                 return -ENOMEM;
 712         }
 713
 714         i = 0;
 715         for_each_cpu_and(cpu, mask, cpu_online_mask)
 716                 map->cpus[i++] = cpu;
 717
 718         if (i)
 719                 map->len = i;
 720         else {
 721                 kfree(map);
 722                 map = NULL;
 723         }
 724
 725         mutex_lock(&rps_map_mutex);
 726         old_map = rcu_dereference_protected(queue->rps_map,
 727                                             mutex_is_locked(&rps_map_mutex));
 728         rcu_assign_pointer(queue->rps_map, map);
 729
 730         if (map)
 731                 static_key_slow_inc(&rps_needed);
 732         if (old_map)
 733                 static_key_slow_dec(&rps_needed);
 734
 735         mutex_unlock(&rps_map_mutex);
 736
 737         if (old_map)
 738                 kfree_rcu(old_map, rcu);
 739
 740         free_cpumask_var(mask);
 741         return len;
 742 }
 743
 744 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 745                                            struct rx_queue_attribute *attr,
 746                                            char *buf)
 747 {
 748         struct rps_dev_flow_table *flow_table;
 749         unsigned long val = 0;
 750
 751         rcu_read_lock();
 752         flow_table = rcu_dereference(queue->rps_flow_table);
 753         if (flow_table)
 754                 val = (unsigned long)flow_table->mask + 1;
 755         rcu_read_unlock();
 756
 757         return sprintf(buf, "%lu\n", val);
 758 }
 759
 760 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 761 {
 762         struct rps_dev_flow_table *table = container_of(rcu,
 763             struct rps_dev_flow_table, rcu);
 764         vfree(table);
 765 }
 766
 767 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 768                                      struct rx_queue_attribute *attr,
 769                                      const char *buf, size_t len)
 770 {
 771         unsigned long mask, count;
 772         struct rps_dev_flow_table *table, *old_table;
 773         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 774         int rc;
 775
 776         if (!capable(CAP_NET_ADMIN))
 777                 return -EPERM;
 778
 779         rc = kstrtoul(buf, 0, &count);
 780         if (rc < 0)
 781                 return rc;
 782
 783         if (count) {
 784                 mask = count - 1;
 785                 /* mask = roundup_pow_of_two(count) - 1;
 786                  * without overflows...
 787                  */
 788                 while ((mask | (mask >> 1)) != mask)
 789                         mask |= (mask >> 1);
 790                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 791                  * and on 32bit arches, must check
 792                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 793                  */
 794 #if BITS_PER_LONG > 32
 795                 if (mask > (unsigned long)(u32)mask)
 796                         return -EINVAL;
 797 #else
 798                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 799                                 / sizeof(struct rps_dev_flow)) {
 800                         /* Enforce a limit to prevent overflow */
 801                         return -EINVAL;
 802                 }
 803 #endif
 804                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 805                 if (!table)
 806                         return -ENOMEM;
 807
 808                 table->mask = mask;
 809                 for (count = 0; count <= mask; count++)
 810                         table->flows[count].cpu = RPS_NO_CPU;
 811         } else
 812                 table = NULL;
 813
 814         spin_lock(&rps_dev_flow_lock);
 815         old_table = rcu_dereference_protected(queue->rps_flow_table,
 816                                               lockdep_is_held(&rps_dev_flow_lock));
 817         rcu_assign_pointer(queue->rps_flow_table, table);
 818         spin_unlock(&rps_dev_flow_lock);
 819
 820         if (old_table)
 821                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 822
 823         return len;
 824 }
 825
 826 static struct rx_queue_attribute rps_cpus_attribute =
 827         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 828
 829
 830 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 831         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 832             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 833 #endif /* CONFIG_RPS */
 834
 835 static struct attribute *rx_queue_default_attrs[] = {
 836 #ifdef CONFIG_RPS
 837         &rps_cpus_attribute.attr,
 838         &rps_dev_flow_table_cnt_attribute.attr,
 839 #endif
 840         NULL
 841 };
 842
 843 static void rx_queue_release(struct kobject *kobj)
 844 {
 845         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 846 #ifdef CONFIG_RPS
 847         struct rps_map *map;
 848         struct rps_dev_flow_table *flow_table;
 849
 850
 851         map = rcu_dereference_protected(queue->rps_map, 1);
 852         if (map) {
 853                 RCU_INIT_POINTER(queue->rps_map, NULL);
 854                 kfree_rcu(map, rcu);
 855         }
 856
 857         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 858         if (flow_table) {
 859                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 860                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 861         }
 862 #endif
 863
 864         memset(kobj, 0, sizeof(*kobj));
 865         dev_put(queue->dev);
 866 }
 867
 868 static const void *rx_queue_namespace(struct kobject *kobj)
 869 {
 870         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 871         struct device *dev = &queue->dev->dev;
 872         const void *ns = NULL;
 873
 874         if (dev->class && dev->class->ns_type)
 875                 ns = dev->class->namespace(dev);
 876
 877         return ns;
 878 }
 879
 880 static struct kobj_type rx_queue_ktype = {
 881         .sysfs_ops = &rx_queue_sysfs_ops,
 882         .release = rx_queue_release,
 883         .default_attrs = rx_queue_default_attrs,
 884         .namespace = rx_queue_namespace
 885 };
 886
 887 static int rx_queue_add_kobject(struct net_device *dev, int index)
 888 {
 889         struct netdev_rx_queue *queue = dev->_rx + index;
 890         struct kobject *kobj = &queue->kobj;
 891         int error = 0;
 892
 893         kobj->kset = dev->queues_kset;
 894         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 895             "rx-%u", index);
 896         if (error)
 897                 goto exit;
 898
 899         if (dev->sysfs_rx_queue_group) {
 900                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 901                 if (error)
 902                         goto exit;
 903         }
 904
 905         kobject_uevent(kobj, KOBJ_ADD);
 906         dev_hold(queue->dev);
 907
 908         return error;
 909 exit:
 910         kobject_put(kobj);
 911         return error;
 912 }
 913 #endif /* CONFIG_SYSFS */
 914
 915 int
 916 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 917 {
 918 #ifdef CONFIG_SYSFS
 919         int i;
 920         int error = 0;
 921
 922 #ifndef CONFIG_RPS
 923         if (!dev->sysfs_rx_queue_group)
 924                 return 0;
 925 #endif
 926         for (i = old_num; i < new_num; i++) {
 927                 error = rx_queue_add_kobject(dev, i);
 928                 if (error) {
 929                         new_num = old_num;
 930                         break;
 931                 }
 932         }
 933
 934         while (--i >= new_num) {
 935                 if (dev->sysfs_rx_queue_group)
 936                         sysfs_remove_group(&dev->_rx[i].kobj,
 937                                            dev->sysfs_rx_queue_group);
 938                 kobject_put(&dev->_rx[i].kobj);
 939         }
 940
 941         return error;
 942 #else
 943         return 0;
 944 #endif
 945 }
 946
 947 #ifdef CONFIG_SYSFS
 948 /*
 949  * netdev_queue sysfs structures and functions.
 950  */
 951 struct netdev_queue_attribute {
 952         struct attribute attr;
 953         ssize_t (*show)(struct netdev_queue *queue,
 954             struct netdev_queue_attribute *attr, char *buf);
 955         ssize_t (*store)(struct netdev_queue *queue,
 956             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 957 };
 958 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 959     struct netdev_queue_attribute, attr)
 960
 961 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 962
 963 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 964                                       struct attribute *attr, char *buf)
 965 {
 966         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 967         struct netdev_queue *queue = to_netdev_queue(kobj);
 968
 969         if (!attribute->show)
 970                 return -EIO;
 971
 972         return attribute->show(queue, attribute, buf);
 973 }
 974
 975 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 976                                        struct attribute *attr,
 977                                        const char *buf, size_t count)
 978 {
 979         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 980         struct netdev_queue *queue = to_netdev_queue(kobj);
 981
 982         if (!attribute->store)
 983                 return -EIO;
 984
 985         return attribute->store(queue, attribute, buf, count);
 986 }
 987
 988 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 989         .show = netdev_queue_attr_show,
 990         .store = netdev_queue_attr_store,
 991 };
 992
 993 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 994                                   struct netdev_queue_attribute *attribute,
 995                                   char *buf)
 996 {
 997         unsigned long trans_timeout;
 998
 999         spin_lock_irq(&queue->_xmit_lock);
1000         trans_timeout = queue->trans_timeout;
1001         spin_unlock_irq(&queue->_xmit_lock);
1002
1003         return sprintf(buf, "%lu", trans_timeout);
1004 }
1005
1006 #ifdef CONFIG_XPS
1007 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1008 {
1009         struct net_device *dev = queue->dev;
1010         unsigned int i;
1011
1012         i = queue - dev->_tx;
1013         BUG_ON(i >= dev->num_tx_queues);
1014
1015         return i;
1016 }
1017
1018 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1019                                struct netdev_queue_attribute *attribute,
1020                                char *buf)
1021 {
1022         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1023 }
1024
1025 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1026                               struct netdev_queue_attribute *attribute,
1027                               const char *buf, size_t len)
1028 {
1029         struct net_device *dev = queue->dev;
1030         int err, index = get_netdev_queue_index(queue);
1031         u32 rate = 0;
1032
1033         err = kstrtou32(buf, 10, &rate);
1034         if (err < 0)
1035                 return err;
1036
1037         if (!rtnl_trylock())
1038                 return restart_syscall();
1039
1040         err = -EOPNOTSUPP;
1041         if (dev->netdev_ops->ndo_set_tx_maxrate)
1042                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1043
1044         rtnl_unlock();
1045         if (!err) {
1046                 queue->tx_maxrate = rate;
1047                 return len;
1048         }
1049         return err;
1050 }
1051
1052 static struct netdev_queue_attribute queue_tx_maxrate =
1053         __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1054                show_tx_maxrate, set_tx_maxrate);
1055 #endif
1056
1057 static struct netdev_queue_attribute queue_trans_timeout =
1058         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1059
1060 #ifdef CONFIG_BQL
1061 /*
1062  * Byte queue limits sysfs structures and functions.
1063  */
1064 static ssize_t bql_show(char *buf, unsigned int value)
1065 {
1066         return sprintf(buf, "%u\n", value);
1067 }
1068
1069 static ssize_t bql_set(const char *buf, const size_t count,
1070                        unsigned int *pvalue)
1071 {
1072         unsigned int value;
1073         int err;
1074
1075         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1076                 value = DQL_MAX_LIMIT;
1077         else {
1078                 err = kstrtouint(buf, 10, &value);
1079                 if (err < 0)
1080                         return err;
1081                 if (value > DQL_MAX_LIMIT)
1082                         return -EINVAL;
1083         }
1084
1085         *pvalue = value;
1086
1087         return count;
1088 }
1089
1090 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1091                                   struct netdev_queue_attribute *attr,
1092                                   char *buf)
1093 {
1094         struct dql *dql = &queue->dql;
1095
1096         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1097 }
1098
1099 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1100                                  struct netdev_queue_attribute *attribute,
1101                                  const char *buf, size_t len)
1102 {
1103         struct dql *dql = &queue->dql;
1104         unsigned int value;
1105         int err;
1106
1107         err = kstrtouint(buf, 10, &value);
1108         if (err < 0)
1109                 return err;
1110
1111         dql->slack_hold_time = msecs_to_jiffies(value);
1112
1113         return len;
1114 }
1115
1116 static struct netdev_queue_attribute bql_hold_time_attribute =
1117         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1118             bql_set_hold_time);
1119
1120 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1121                                  struct netdev_queue_attribute *attr,
1122                                  char *buf)
1123 {
1124         struct dql *dql = &queue->dql;
1125
1126         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1127 }
1128
1129 static struct netdev_queue_attribute bql_inflight_attribute =
1130         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1131
1132 #define BQL_ATTR(NAME, FIELD)                                           \
1133 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1134                                  struct netdev_queue_attribute *attr,   \
1135                                  char *buf)                             \
1136 {                                                                       \
1137         return bql_show(buf, queue->dql.FIELD);                         \
1138 }                                                                       \
1139                                                                         \
1140 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1141                                 struct netdev_queue_attribute *attr,    \
1142                                 const char *buf, size_t len)            \
1143 {                                                                       \
1144         return bql_set(buf, len, &queue->dql.FIELD);                    \
1145 }                                                                       \
1146                                                                         \
1147 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1148         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1149             bql_set_ ## NAME);
1150
1151 BQL_ATTR(limit, limit)
1152 BQL_ATTR(limit_max, max_limit)
1153 BQL_ATTR(limit_min, min_limit)
1154
1155 static struct attribute *dql_attrs[] = {
1156         &bql_limit_attribute.attr,
1157         &bql_limit_max_attribute.attr,
1158         &bql_limit_min_attribute.attr,
1159         &bql_hold_time_attribute.attr,
1160         &bql_inflight_attribute.attr,
1161         NULL
1162 };
1163
1164 static struct attribute_group dql_group = {
1165         .name  = "byte_queue_limits",
1166         .attrs  = dql_attrs,
1167 };
1168 #endif /* CONFIG_BQL */
1169
1170 #ifdef CONFIG_XPS
1171 static ssize_t show_xps_map(struct netdev_queue *queue,
1172                             struct netdev_queue_attribute *attribute, char *buf)
1173 {
1174         struct net_device *dev = queue->dev;
1175         struct xps_dev_maps *dev_maps;
1176         cpumask_var_t mask;
1177         unsigned long index;
1178         int i, len;
1179
1180         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1181                 return -ENOMEM;
1182
1183         index = get_netdev_queue_index(queue);
1184
1185         rcu_read_lock();
1186         dev_maps = rcu_dereference(dev->xps_maps);
1187         if (dev_maps) {
1188                 for_each_possible_cpu(i) {
1189                         struct xps_map *map =
1190                             rcu_dereference(dev_maps->cpu_map[i]);
1191                         if (map) {
1192                                 int j;
1193                                 for (j = 0; j < map->len; j++) {
1194                                         if (map->queues[j] == index) {
1195                                                 cpumask_set_cpu(i, mask);
1196                                                 break;
1197                                         }
1198                                 }
1199                         }
1200                 }
1201         }
1202         rcu_read_unlock();
1203
1204         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1205         free_cpumask_var(mask);
1206         return len < PAGE_SIZE ? len : -EINVAL;
1207 }
1208
1209 static ssize_t store_xps_map(struct netdev_queue *queue,
1210                       struct netdev_queue_attribute *attribute,
1211                       const char *buf, size_t len)
1212 {
1213         struct net_device *dev = queue->dev;
1214         unsigned long index;
1215         cpumask_var_t mask;
1216         int err;
1217
1218         if (!capable(CAP_NET_ADMIN))
1219                 return -EPERM;
1220
1221         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1222                 return -ENOMEM;
1223
1224         index = get_netdev_queue_index(queue);
1225
1226         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1227         if (err) {
1228                 free_cpumask_var(mask);
1229                 return err;
1230         }
1231
1232         err = netif_set_xps_queue(dev, mask, index);
1233
1234         free_cpumask_var(mask);
1235
1236         return err ? : len;
1237 }
1238
1239 static struct netdev_queue_attribute xps_cpus_attribute =
1240     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1241 #endif /* CONFIG_XPS */
1242
1243 static struct attribute *netdev_queue_default_attrs[] = {
1244         &queue_trans_timeout.attr,
1245 #ifdef CONFIG_XPS
1246         &xps_cpus_attribute.attr,
1247         &queue_tx_maxrate.attr,
1248 #endif
1249         NULL
1250 };
1251
1252 static void netdev_queue_release(struct kobject *kobj)
1253 {
1254         struct netdev_queue *queue = to_netdev_queue(kobj);
1255
1256         memset(kobj, 0, sizeof(*kobj));
1257         dev_put(queue->dev);
1258 }
1259
1260 static const void *netdev_queue_namespace(struct kobject *kobj)
1261 {
1262         struct netdev_queue *queue = to_netdev_queue(kobj);
1263         struct device *dev = &queue->dev->dev;
1264         const void *ns = NULL;
1265
1266         if (dev->class && dev->class->ns_type)
1267                 ns = dev->class->namespace(dev);
1268
1269         return ns;
1270 }
1271
1272 static struct kobj_type netdev_queue_ktype = {
1273         .sysfs_ops = &netdev_queue_sysfs_ops,
1274         .release = netdev_queue_release,
1275         .default_attrs = netdev_queue_default_attrs,
1276         .namespace = netdev_queue_namespace,
1277 };
1278
1279 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1280 {
1281         struct netdev_queue *queue = dev->_tx + index;
1282         struct kobject *kobj = &queue->kobj;
1283         int error = 0;
1284
1285         kobj->kset = dev->queues_kset;
1286         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1287             "tx-%u", index);
1288         if (error)
1289                 goto exit;
1290
1291 #ifdef CONFIG_BQL
1292         error = sysfs_create_group(kobj, &dql_group);
1293         if (error)
1294                 goto exit;
1295 #endif
1296
1297         kobject_uevent(kobj, KOBJ_ADD);
1298         dev_hold(queue->dev);
1299
1300         return 0;
1301 exit:
1302         kobject_put(kobj);
1303         return error;
1304 }
1305 #endif /* CONFIG_SYSFS */
1306
1307 int
1308 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1309 {
1310 #ifdef CONFIG_SYSFS
1311         int i;
1312         int error = 0;
1313
1314         for (i = old_num; i < new_num; i++) {
1315                 error = netdev_queue_add_kobject(dev, i);
1316                 if (error) {
1317                         new_num = old_num;
1318                         break;
1319                 }
1320         }
1321
1322         while (--i >= new_num) {
1323                 struct netdev_queue *queue = dev->_tx + i;
1324
1325 #ifdef CONFIG_BQL
1326                 sysfs_remove_group(&queue->kobj, &dql_group);
1327 #endif
1328                 kobject_put(&queue->kobj);
1329         }
1330
1331         return error;
1332 #else
1333         return 0;
1334 #endif /* CONFIG_SYSFS */
1335 }
1336
1337 static int register_queue_kobjects(struct net_device *dev)
1338 {
1339         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1340
1341 #ifdef CONFIG_SYSFS
1342         dev->queues_kset = kset_create_and_add("queues",
1343             NULL, &dev->dev.kobj);
1344         if (!dev->queues_kset)
1345                 return -ENOMEM;
1346         real_rx = dev->real_num_rx_queues;
1347 #endif
1348         real_tx = dev->real_num_tx_queues;
1349
1350         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1351         if (error)
1352                 goto error;
1353         rxq = real_rx;
1354
1355         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1356         if (error)
1357                 goto error;
1358         txq = real_tx;
1359
1360         return 0;
1361
1362 error:
1363         netdev_queue_update_kobjects(dev, txq, 0);
1364         net_rx_queue_update_kobjects(dev, rxq, 0);
1365         return error;
1366 }
1367
1368 static void remove_queue_kobjects(struct net_device *dev)
1369 {
1370         int real_rx = 0, real_tx = 0;
1371
1372 #ifdef CONFIG_SYSFS
1373         real_rx = dev->real_num_rx_queues;
1374 #endif
1375         real_tx = dev->real_num_tx_queues;
1376
1377         net_rx_queue_update_kobjects(dev, real_rx, 0);
1378         netdev_queue_update_kobjects(dev, real_tx, 0);
1379 #ifdef CONFIG_SYSFS
1380         kset_unregister(dev->queues_kset);
1381 #endif
1382 }
1383
1384 static bool net_current_may_mount(void)
1385 {
1386         struct net *net = current->nsproxy->net_ns;
1387
1388         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1389 }
1390
1391 static void *net_grab_current_ns(void)
1392 {
1393         struct net *ns = current->nsproxy->net_ns;
1394 #ifdef CONFIG_NET_NS
1395         if (ns)
1396                 atomic_inc(&ns->passive);
1397 #endif
1398         return ns;
1399 }
1400
1401 static const void *net_initial_ns(void)
1402 {
1403         return &init_net;
1404 }
1405
1406 static const void *net_netlink_ns(struct sock *sk)
1407 {
1408         return sock_net(sk);
1409 }
1410
1411 struct kobj_ns_type_operations net_ns_type_operations = {
1412         .type = KOBJ_NS_TYPE_NET,
1413         .current_may_mount = net_current_may_mount,
1414         .grab_current_ns = net_grab_current_ns,
1415         .netlink_ns = net_netlink_ns,
1416         .initial_ns = net_initial_ns,
1417         .drop_ns = net_drop_ns,
1418 };
1419 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1420
1421 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1422 {
1423         struct net_device *dev = to_net_dev(d);
1424         int retval;
1425
1426         /* pass interface to uevent. */
1427         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1428         if (retval)
1429                 goto exit;
1430
1431         /* pass ifindex to uevent.
1432          * ifindex is useful as it won't change (interface name may change)
1433          * and is what RtNetlink uses natively. */
1434         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1435
1436 exit:
1437         return retval;
1438 }
1439
1440 /*
1441  *      netdev_release -- destroy and free a dead device.
1442  *      Called when last reference to device kobject is gone.
1443  */
1444 static void netdev_release(struct device *d)
1445 {
1446         struct net_device *dev = to_net_dev(d);
1447
1448         BUG_ON(dev->reg_state != NETREG_RELEASED);
1449
1450         kfree(dev->ifalias);
1451         netdev_freemem(dev);
1452 }
1453
1454 static const void *net_namespace(struct device *d)
1455 {
1456         struct net_device *dev = to_net_dev(d);
1457
1458         return dev_net(dev);
1459 }
1460
1461 static struct class net_class = {
1462         .name = "net",
1463         .dev_release = netdev_release,
1464         .dev_groups = net_class_groups,
1465         .dev_uevent = netdev_uevent,
1466         .ns_type = &net_ns_type_operations,
1467         .namespace = net_namespace,
1468 };
1469
1470 #ifdef CONFIG_OF_NET
1471 static int of_dev_node_match(struct device *dev, const void *data)
1472 {
1473         int ret = 0;
1474
1475         if (dev->parent)
1476                 ret = dev->parent->of_node == data;
1477
1478         return ret == 0 ? dev->of_node == data : ret;
1479 }
1480
1481 /*
1482  * of_find_net_device_by_node - lookup the net device for the device node
1483  * @np: OF device node
1484  *
1485  * Looks up the net_device structure corresponding with the device node.
1486  * If successful, returns a pointer to the net_device with the embedded
1487  * struct device refcount incremented by one, or NULL on failure. The
1488  * refcount must be dropped when done with the net_device.
1489  */
1490 struct net_device *of_find_net_device_by_node(struct device_node *np)
1491 {
1492         struct device *dev;
1493
1494         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1495         if (!dev)
1496                 return NULL;
1497
1498         return to_net_dev(dev);
1499 }
1500 EXPORT_SYMBOL(of_find_net_device_by_node);
1501 #endif
1502
1503 /* Delete sysfs entries but hold kobject reference until after all
1504  * netdev references are gone.
1505  */
1506 void netdev_unregister_kobject(struct net_device *ndev)
1507 {
1508         struct device *dev = &(ndev->dev);
1509
1510         kobject_get(&dev->kobj);
1511
1512         remove_queue_kobjects(ndev);
1513
1514         pm_runtime_set_memalloc_noio(dev, false);
1515
1516         device_del(dev);
1517 }
1518
1519 /* Create sysfs entries for network device. */
1520 int netdev_register_kobject(struct net_device *ndev)
1521 {
1522         struct device *dev = &(ndev->dev);
1523         const struct attribute_group **groups = ndev->sysfs_groups;
1524         int error = 0;
1525
1526         device_initialize(dev);
1527         dev->class = &net_class;
1528         dev->platform_data = ndev;
1529         dev->groups = groups;
1530
1531         dev_set_name(dev, "%s", ndev->name);
1532
1533 #ifdef CONFIG_SYSFS
1534         /* Allow for a device specific group */
1535         if (*groups)
1536                 groups++;
1537
1538         *groups++ = &netstat_group;
1539
1540 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1541         if (ndev->ieee80211_ptr)
1542                 *groups++ = &wireless_group;
1543 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1544         else if (ndev->wireless_handlers)
1545                 *groups++ = &wireless_group;
1546 #endif
1547 #endif
1548 #endif /* CONFIG_SYSFS */
1549
1550         error = device_add(dev);
1551         if (error)
1552                 return error;
1553
1554         error = register_queue_kobjects(ndev);
1555         if (error) {
1556                 device_del(dev);
1557                 return error;
1558         }
1559
1560         pm_runtime_set_memalloc_noio(dev, true);
1561
1562         return error;
1563 }
1564
1565 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1566                                 const void *ns)
1567 {
1568         return class_create_file_ns(&net_class, class_attr, ns);
1569 }
1570 EXPORT_SYMBOL(netdev_class_create_file_ns);
1571
1572 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1573                                  const void *ns)
1574 {
1575         class_remove_file_ns(&net_class, class_attr, ns);
1576 }
1577 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1578
1579 int __init netdev_kobject_init(void)
1580 {
1581         kobj_ns_type_register(&net_ns_type_operations);
1582         return class_register(&net_class);
1583 }