net/core/net-sysfs.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net-sysfs.c - network device class and attributes
   4  *
   5  * Copyright (c) 2003 Stephen Hemminger <[email protected]>
   6  */
   7
   8 #include <linux/capability.h>
   9 #include <linux/kernel.h>
  10 #include <linux/netdevice.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/sched/isolation.h>
  15 #include <linux/nsproxy.h>
  16 #include <net/sock.h>
  17 #include <net/net_namespace.h>
  18 #include <linux/rtnetlink.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/export.h>
  21 #include <linux/jiffies.h>
  22 #include <linux/pm_runtime.h>
  23 #include <linux/of.h>
  24 #include <linux/of_net.h>
  25 #include <linux/cpu.h>
  26
  27 #include "dev.h"
  28 #include "net-sysfs.h"
  29
  30 #ifdef CONFIG_SYSFS
  31 static const char fmt_hex[] = "%#x\n";
  32 static const char fmt_dec[] = "%d\n";
  33 static const char fmt_ulong[] = "%lu\n";
  34 static const char fmt_u64[] = "%llu\n";
  35
  36 /* Caller holds RTNL or dev_base_lock */
  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 sysfs_emit(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;
  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                 ret = (*set)(netdev, new);
 100                 if (ret == 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(ifindex, fmt_dec);
 113 NETDEVICE_SHOW_RO(type, fmt_dec);
 114 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 115
 116 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 117                            char *buf)
 118 {
 119         struct net_device *ndev = to_net_dev(dev);
 120
 121         return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
 122 }
 123 static DEVICE_ATTR_RO(iflink);
 124
 125 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 126 {
 127         return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
 128 }
 129
 130 static ssize_t name_assign_type_show(struct device *dev,
 131                                      struct device_attribute *attr,
 132                                      char *buf)
 133 {
 134         struct net_device *ndev = to_net_dev(dev);
 135         ssize_t ret = -EINVAL;
 136
 137         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 138                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 139
 140         return ret;
 141 }
 142 static DEVICE_ATTR_RO(name_assign_type);
 143
 144 /* use same locking rules as GIFHWADDR ioctl's */
 145 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 146                             char *buf)
 147 {
 148         struct net_device *ndev = to_net_dev(dev);
 149         ssize_t ret = -EINVAL;
 150
 151         read_lock(&dev_base_lock);
 152         if (dev_isalive(ndev))
 153                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 154         read_unlock(&dev_base_lock);
 155         return ret;
 156 }
 157 static DEVICE_ATTR_RO(address);
 158
 159 static ssize_t broadcast_show(struct device *dev,
 160                               struct device_attribute *attr, char *buf)
 161 {
 162         struct net_device *ndev = to_net_dev(dev);
 163
 164         if (dev_isalive(ndev))
 165                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 166         return -EINVAL;
 167 }
 168 static DEVICE_ATTR_RO(broadcast);
 169
 170 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 171 {
 172         if (!netif_running(dev))
 173                 return -EINVAL;
 174         return dev_change_carrier(dev, (bool)new_carrier);
 175 }
 176
 177 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 178                              const char *buf, size_t len)
 179 {
 180         struct net_device *netdev = to_net_dev(dev);
 181
 182         /* The check is also done in change_carrier; this helps returning early
 183          * without hitting the trylock/restart in netdev_store.
 184          */
 185         if (!netdev->netdev_ops->ndo_change_carrier)
 186                 return -EOPNOTSUPP;
 187
 188         return netdev_store(dev, attr, buf, len, change_carrier);
 189 }
 190
 191 static ssize_t carrier_show(struct device *dev,
 192                             struct device_attribute *attr, char *buf)
 193 {
 194         struct net_device *netdev = to_net_dev(dev);
 195
 196         if (netif_running(netdev))
 197                 return sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
 198
 199         return -EINVAL;
 200 }
 201 static DEVICE_ATTR_RW(carrier);
 202
 203 static ssize_t speed_show(struct device *dev,
 204                           struct device_attribute *attr, char *buf)
 205 {
 206         struct net_device *netdev = to_net_dev(dev);
 207         int ret = -EINVAL;
 208
 209         /* The check is also done in __ethtool_get_link_ksettings; this helps
 210          * returning early without hitting the trylock/restart below.
 211          */
 212         if (!netdev->ethtool_ops->get_link_ksettings)
 213                 return ret;
 214
 215         if (!rtnl_trylock())
 216                 return restart_syscall();
 217
 218         if (netif_running(netdev) && netif_device_present(netdev)) {
 219                 struct ethtool_link_ksettings cmd;
 220
 221                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
 222                         ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
 223         }
 224         rtnl_unlock();
 225         return ret;
 226 }
 227 static DEVICE_ATTR_RO(speed);
 228
 229 static ssize_t duplex_show(struct device *dev,
 230                            struct device_attribute *attr, char *buf)
 231 {
 232         struct net_device *netdev = to_net_dev(dev);
 233         int ret = -EINVAL;
 234
 235         /* The check is also done in __ethtool_get_link_ksettings; this helps
 236          * returning early without hitting the trylock/restart below.
 237          */
 238         if (!netdev->ethtool_ops->get_link_ksettings)
 239                 return ret;
 240
 241         if (!rtnl_trylock())
 242                 return restart_syscall();
 243
 244         if (netif_running(netdev)) {
 245                 struct ethtool_link_ksettings cmd;
 246
 247                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
 248                         const char *duplex;
 249
 250                         switch (cmd.base.duplex) {
 251                         case DUPLEX_HALF:
 252                                 duplex = "half";
 253                                 break;
 254                         case DUPLEX_FULL:
 255                                 duplex = "full";
 256                                 break;
 257                         default:
 258                                 duplex = "unknown";
 259                                 break;
 260                         }
 261                         ret = sysfs_emit(buf, "%s\n", duplex);
 262                 }
 263         }
 264         rtnl_unlock();
 265         return ret;
 266 }
 267 static DEVICE_ATTR_RO(duplex);
 268
 269 static ssize_t testing_show(struct device *dev,
 270                             struct device_attribute *attr, char *buf)
 271 {
 272         struct net_device *netdev = to_net_dev(dev);
 273
 274         if (netif_running(netdev))
 275                 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
 276
 277         return -EINVAL;
 278 }
 279 static DEVICE_ATTR_RO(testing);
 280
 281 static ssize_t dormant_show(struct device *dev,
 282                             struct device_attribute *attr, char *buf)
 283 {
 284         struct net_device *netdev = to_net_dev(dev);
 285
 286         if (netif_running(netdev))
 287                 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
 288
 289         return -EINVAL;
 290 }
 291 static DEVICE_ATTR_RO(dormant);
 292
 293 static const char *const operstates[] = {
 294         "unknown",
 295         "notpresent", /* currently unused */
 296         "down",
 297         "lowerlayerdown",
 298         "testing",
 299         "dormant",
 300         "up"
 301 };
 302
 303 static ssize_t operstate_show(struct device *dev,
 304                               struct device_attribute *attr, char *buf)
 305 {
 306         const struct net_device *netdev = to_net_dev(dev);
 307         unsigned char operstate;
 308
 309         read_lock(&dev_base_lock);
 310         operstate = netdev->operstate;
 311         if (!netif_running(netdev))
 312                 operstate = IF_OPER_DOWN;
 313         read_unlock(&dev_base_lock);
 314
 315         if (operstate >= ARRAY_SIZE(operstates))
 316                 return -EINVAL; /* should not happen */
 317
 318         return sysfs_emit(buf, "%s\n", operstates[operstate]);
 319 }
 320 static DEVICE_ATTR_RO(operstate);
 321
 322 static ssize_t carrier_changes_show(struct device *dev,
 323                                     struct device_attribute *attr,
 324                                     char *buf)
 325 {
 326         struct net_device *netdev = to_net_dev(dev);
 327
 328         return sysfs_emit(buf, fmt_dec,
 329                           atomic_read(&netdev->carrier_up_count) +
 330                           atomic_read(&netdev->carrier_down_count));
 331 }
 332 static DEVICE_ATTR_RO(carrier_changes);
 333
 334 static ssize_t carrier_up_count_show(struct device *dev,
 335                                      struct device_attribute *attr,
 336                                      char *buf)
 337 {
 338         struct net_device *netdev = to_net_dev(dev);
 339
 340         return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
 341 }
 342 static DEVICE_ATTR_RO(carrier_up_count);
 343
 344 static ssize_t carrier_down_count_show(struct device *dev,
 345                                        struct device_attribute *attr,
 346                                        char *buf)
 347 {
 348         struct net_device *netdev = to_net_dev(dev);
 349
 350         return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
 351 }
 352 static DEVICE_ATTR_RO(carrier_down_count);
 353
 354 /* read-write attributes */
 355
 356 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 357 {
 358         return dev_set_mtu(dev, (int)new_mtu);
 359 }
 360
 361 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 362                          const char *buf, size_t len)
 363 {
 364         return netdev_store(dev, attr, buf, len, change_mtu);
 365 }
 366 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 367
 368 static int change_flags(struct net_device *dev, unsigned long new_flags)
 369 {
 370         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
 371 }
 372
 373 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 374                            const char *buf, size_t len)
 375 {
 376         return netdev_store(dev, attr, buf, len, change_flags);
 377 }
 378 NETDEVICE_SHOW_RW(flags, fmt_hex);
 379
 380 static ssize_t tx_queue_len_store(struct device *dev,
 381                                   struct device_attribute *attr,
 382                                   const char *buf, size_t len)
 383 {
 384         if (!capable(CAP_NET_ADMIN))
 385                 return -EPERM;
 386
 387         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
 388 }
 389 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 390
 391 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 392 {
 393         WRITE_ONCE(dev->gro_flush_timeout, val);
 394         return 0;
 395 }
 396
 397 static ssize_t gro_flush_timeout_store(struct device *dev,
 398                                        struct device_attribute *attr,
 399                                        const char *buf, size_t len)
 400 {
 401         if (!capable(CAP_NET_ADMIN))
 402                 return -EPERM;
 403
 404         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 405 }
 406 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 407
 408 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
 409 {
 410         WRITE_ONCE(dev->napi_defer_hard_irqs, val);
 411         return 0;
 412 }
 413
 414 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
 415                                           struct device_attribute *attr,
 416                                           const char *buf, size_t len)
 417 {
 418         if (!capable(CAP_NET_ADMIN))
 419                 return -EPERM;
 420
 421         return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
 422 }
 423 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
 424
 425 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 426                              const char *buf, size_t len)
 427 {
 428         struct net_device *netdev = to_net_dev(dev);
 429         struct net *net = dev_net(netdev);
 430         size_t count = len;
 431         ssize_t ret = 0;
 432
 433         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 434                 return -EPERM;
 435
 436         /* ignore trailing newline */
 437         if (len >  0 && buf[len - 1] == '\n')
 438                 --count;
 439
 440         if (!rtnl_trylock())
 441                 return restart_syscall();
 442
 443         if (dev_isalive(netdev)) {
 444                 ret = dev_set_alias(netdev, buf, count);
 445                 if (ret < 0)
 446                         goto err;
 447                 ret = len;
 448                 netdev_state_change(netdev);
 449         }
 450 err:
 451         rtnl_unlock();
 452
 453         return ret;
 454 }
 455
 456 static ssize_t ifalias_show(struct device *dev,
 457                             struct device_attribute *attr, char *buf)
 458 {
 459         const struct net_device *netdev = to_net_dev(dev);
 460         char tmp[IFALIASZ];
 461         ssize_t ret = 0;
 462
 463         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
 464         if (ret > 0)
 465                 ret = sysfs_emit(buf, "%s\n", tmp);
 466         return ret;
 467 }
 468 static DEVICE_ATTR_RW(ifalias);
 469
 470 static int change_group(struct net_device *dev, unsigned long new_group)
 471 {
 472         dev_set_group(dev, (int)new_group);
 473         return 0;
 474 }
 475
 476 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 477                            const char *buf, size_t len)
 478 {
 479         return netdev_store(dev, attr, buf, len, change_group);
 480 }
 481 NETDEVICE_SHOW(group, fmt_dec);
 482 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
 483
 484 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 485 {
 486         return dev_change_proto_down(dev, (bool)proto_down);
 487 }
 488
 489 static ssize_t proto_down_store(struct device *dev,
 490                                 struct device_attribute *attr,
 491                                 const char *buf, size_t len)
 492 {
 493         return netdev_store(dev, attr, buf, len, change_proto_down);
 494 }
 495 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 496
 497 static ssize_t phys_port_id_show(struct device *dev,
 498                                  struct device_attribute *attr, char *buf)
 499 {
 500         struct net_device *netdev = to_net_dev(dev);
 501         ssize_t ret = -EINVAL;
 502
 503         /* The check is also done in dev_get_phys_port_id; this helps returning
 504          * early without hitting the trylock/restart below.
 505          */
 506         if (!netdev->netdev_ops->ndo_get_phys_port_id)
 507                 return -EOPNOTSUPP;
 508
 509         if (!rtnl_trylock())
 510                 return restart_syscall();
 511
 512         if (dev_isalive(netdev)) {
 513                 struct netdev_phys_item_id ppid;
 514
 515                 ret = dev_get_phys_port_id(netdev, &ppid);
 516                 if (!ret)
 517                         ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 518         }
 519         rtnl_unlock();
 520
 521         return ret;
 522 }
 523 static DEVICE_ATTR_RO(phys_port_id);
 524
 525 static ssize_t phys_port_name_show(struct device *dev,
 526                                    struct device_attribute *attr, char *buf)
 527 {
 528         struct net_device *netdev = to_net_dev(dev);
 529         ssize_t ret = -EINVAL;
 530
 531         /* The checks are also done in dev_get_phys_port_name; this helps
 532          * returning early without hitting the trylock/restart below.
 533          */
 534         if (!netdev->netdev_ops->ndo_get_phys_port_name &&
 535             !netdev->devlink_port)
 536                 return -EOPNOTSUPP;
 537
 538         if (!rtnl_trylock())
 539                 return restart_syscall();
 540
 541         if (dev_isalive(netdev)) {
 542                 char name[IFNAMSIZ];
 543
 544                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 545                 if (!ret)
 546                         ret = sysfs_emit(buf, "%s\n", name);
 547         }
 548         rtnl_unlock();
 549
 550         return ret;
 551 }
 552 static DEVICE_ATTR_RO(phys_port_name);
 553
 554 static ssize_t phys_switch_id_show(struct device *dev,
 555                                    struct device_attribute *attr, char *buf)
 556 {
 557         struct net_device *netdev = to_net_dev(dev);
 558         ssize_t ret = -EINVAL;
 559
 560         /* The checks are also done in dev_get_phys_port_name; this helps
 561          * returning early without hitting the trylock/restart below. This works
 562          * because recurse is false when calling dev_get_port_parent_id.
 563          */
 564         if (!netdev->netdev_ops->ndo_get_port_parent_id &&
 565             !netdev->devlink_port)
 566                 return -EOPNOTSUPP;
 567
 568         if (!rtnl_trylock())
 569                 return restart_syscall();
 570
 571         if (dev_isalive(netdev)) {
 572                 struct netdev_phys_item_id ppid = { };
 573
 574                 ret = dev_get_port_parent_id(netdev, &ppid, false);
 575                 if (!ret)
 576                         ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 577         }
 578         rtnl_unlock();
 579
 580         return ret;
 581 }
 582 static DEVICE_ATTR_RO(phys_switch_id);
 583
 584 static ssize_t threaded_show(struct device *dev,
 585                              struct device_attribute *attr, char *buf)
 586 {
 587         struct net_device *netdev = to_net_dev(dev);
 588         ssize_t ret = -EINVAL;
 589
 590         if (!rtnl_trylock())
 591                 return restart_syscall();
 592
 593         if (dev_isalive(netdev))
 594                 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
 595
 596         rtnl_unlock();
 597         return ret;
 598 }
 599
 600 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
 601 {
 602         int ret;
 603
 604         if (list_empty(&dev->napi_list))
 605                 return -EOPNOTSUPP;
 606
 607         if (val != 0 && val != 1)
 608                 return -EOPNOTSUPP;
 609
 610         ret = dev_set_threaded(dev, val);
 611
 612         return ret;
 613 }
 614
 615 static ssize_t threaded_store(struct device *dev,
 616                               struct device_attribute *attr,
 617                               const char *buf, size_t len)
 618 {
 619         return netdev_store(dev, attr, buf, len, modify_napi_threaded);
 620 }
 621 static DEVICE_ATTR_RW(threaded);
 622
 623 static struct attribute *net_class_attrs[] __ro_after_init = {
 624         &dev_attr_netdev_group.attr,
 625         &dev_attr_type.attr,
 626         &dev_attr_dev_id.attr,
 627         &dev_attr_dev_port.attr,
 628         &dev_attr_iflink.attr,
 629         &dev_attr_ifindex.attr,
 630         &dev_attr_name_assign_type.attr,
 631         &dev_attr_addr_assign_type.attr,
 632         &dev_attr_addr_len.attr,
 633         &dev_attr_link_mode.attr,
 634         &dev_attr_address.attr,
 635         &dev_attr_broadcast.attr,
 636         &dev_attr_speed.attr,
 637         &dev_attr_duplex.attr,
 638         &dev_attr_dormant.attr,
 639         &dev_attr_testing.attr,
 640         &dev_attr_operstate.attr,
 641         &dev_attr_carrier_changes.attr,
 642         &dev_attr_ifalias.attr,
 643         &dev_attr_carrier.attr,
 644         &dev_attr_mtu.attr,
 645         &dev_attr_flags.attr,
 646         &dev_attr_tx_queue_len.attr,
 647         &dev_attr_gro_flush_timeout.attr,
 648         &dev_attr_napi_defer_hard_irqs.attr,
 649         &dev_attr_phys_port_id.attr,
 650         &dev_attr_phys_port_name.attr,
 651         &dev_attr_phys_switch_id.attr,
 652         &dev_attr_proto_down.attr,
 653         &dev_attr_carrier_up_count.attr,
 654         &dev_attr_carrier_down_count.attr,
 655         &dev_attr_threaded.attr,
 656         NULL,
 657 };
 658 ATTRIBUTE_GROUPS(net_class);
 659
 660 /* Show a given an attribute in the statistics group */
 661 static ssize_t netstat_show(const struct device *d,
 662                             struct device_attribute *attr, char *buf,
 663                             unsigned long offset)
 664 {
 665         struct net_device *dev = to_net_dev(d);
 666         ssize_t ret = -EINVAL;
 667
 668         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 669                 offset % sizeof(u64) != 0);
 670
 671         read_lock(&dev_base_lock);
 672         if (dev_isalive(dev)) {
 673                 struct rtnl_link_stats64 temp;
 674                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 675
 676                 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
 677         }
 678         read_unlock(&dev_base_lock);
 679         return ret;
 680 }
 681
 682 /* generate a read-only statistics attribute */
 683 #define NETSTAT_ENTRY(name)                                             \
 684 static ssize_t name##_show(struct device *d,                            \
 685                            struct device_attribute *attr, char *buf)    \
 686 {                                                                       \
 687         return netstat_show(d, attr, buf,                               \
 688                             offsetof(struct rtnl_link_stats64, name));  \
 689 }                                                                       \
 690 static DEVICE_ATTR_RO(name)
 691
 692 NETSTAT_ENTRY(rx_packets);
 693 NETSTAT_ENTRY(tx_packets);
 694 NETSTAT_ENTRY(rx_bytes);
 695 NETSTAT_ENTRY(tx_bytes);
 696 NETSTAT_ENTRY(rx_errors);
 697 NETSTAT_ENTRY(tx_errors);
 698 NETSTAT_ENTRY(rx_dropped);
 699 NETSTAT_ENTRY(tx_dropped);
 700 NETSTAT_ENTRY(multicast);
 701 NETSTAT_ENTRY(collisions);
 702 NETSTAT_ENTRY(rx_length_errors);
 703 NETSTAT_ENTRY(rx_over_errors);
 704 NETSTAT_ENTRY(rx_crc_errors);
 705 NETSTAT_ENTRY(rx_frame_errors);
 706 NETSTAT_ENTRY(rx_fifo_errors);
 707 NETSTAT_ENTRY(rx_missed_errors);
 708 NETSTAT_ENTRY(tx_aborted_errors);
 709 NETSTAT_ENTRY(tx_carrier_errors);
 710 NETSTAT_ENTRY(tx_fifo_errors);
 711 NETSTAT_ENTRY(tx_heartbeat_errors);
 712 NETSTAT_ENTRY(tx_window_errors);
 713 NETSTAT_ENTRY(rx_compressed);
 714 NETSTAT_ENTRY(tx_compressed);
 715 NETSTAT_ENTRY(rx_nohandler);
 716
 717 static struct attribute *netstat_attrs[] __ro_after_init = {
 718         &dev_attr_rx_packets.attr,
 719         &dev_attr_tx_packets.attr,
 720         &dev_attr_rx_bytes.attr,
 721         &dev_attr_tx_bytes.attr,
 722         &dev_attr_rx_errors.attr,
 723         &dev_attr_tx_errors.attr,
 724         &dev_attr_rx_dropped.attr,
 725         &dev_attr_tx_dropped.attr,
 726         &dev_attr_multicast.attr,
 727         &dev_attr_collisions.attr,
 728         &dev_attr_rx_length_errors.attr,
 729         &dev_attr_rx_over_errors.attr,
 730         &dev_attr_rx_crc_errors.attr,
 731         &dev_attr_rx_frame_errors.attr,
 732         &dev_attr_rx_fifo_errors.attr,
 733         &dev_attr_rx_missed_errors.attr,
 734         &dev_attr_tx_aborted_errors.attr,
 735         &dev_attr_tx_carrier_errors.attr,
 736         &dev_attr_tx_fifo_errors.attr,
 737         &dev_attr_tx_heartbeat_errors.attr,
 738         &dev_attr_tx_window_errors.attr,
 739         &dev_attr_rx_compressed.attr,
 740         &dev_attr_tx_compressed.attr,
 741         &dev_attr_rx_nohandler.attr,
 742         NULL
 743 };
 744
 745 static const struct attribute_group netstat_group = {
 746         .name  = "statistics",
 747         .attrs  = netstat_attrs,
 748 };
 749
 750 static struct attribute *wireless_attrs[] = {
 751         NULL
 752 };
 753
 754 static const struct attribute_group wireless_group = {
 755         .name = "wireless",
 756         .attrs = wireless_attrs,
 757 };
 758
 759 static bool wireless_group_needed(struct net_device *ndev)
 760 {
 761 #if IS_ENABLED(CONFIG_CFG80211)
 762         if (ndev->ieee80211_ptr)
 763                 return true;
 764 #endif
 765 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
 766         if (ndev->wireless_handlers)
 767                 return true;
 768 #endif
 769         return false;
 770 }
 771
 772 #else /* CONFIG_SYSFS */
 773 #define net_class_groups        NULL
 774 #endif /* CONFIG_SYSFS */
 775
 776 #ifdef CONFIG_SYSFS
 777 #define to_rx_queue_attr(_attr) \
 778         container_of(_attr, struct rx_queue_attribute, attr)
 779
 780 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 781
 782 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 783                                   char *buf)
 784 {
 785         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 786         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 787
 788         if (!attribute->show)
 789                 return -EIO;
 790
 791         return attribute->show(queue, buf);
 792 }
 793
 794 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 795                                    const char *buf, size_t count)
 796 {
 797         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 798         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 799
 800         if (!attribute->store)
 801                 return -EIO;
 802
 803         return attribute->store(queue, buf, count);
 804 }
 805
 806 static const struct sysfs_ops rx_queue_sysfs_ops = {
 807         .show = rx_queue_attr_show,
 808         .store = rx_queue_attr_store,
 809 };
 810
 811 #ifdef CONFIG_RPS
 812 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
 813 {
 814         struct rps_map *map;
 815         cpumask_var_t mask;
 816         int i, len;
 817
 818         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 819                 return -ENOMEM;
 820
 821         rcu_read_lock();
 822         map = rcu_dereference(queue->rps_map);
 823         if (map)
 824                 for (i = 0; i < map->len; i++)
 825                         cpumask_set_cpu(map->cpus[i], mask);
 826
 827         len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
 828         rcu_read_unlock();
 829         free_cpumask_var(mask);
 830
 831         return len < PAGE_SIZE ? len : -EINVAL;
 832 }
 833
 834 static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
 835                                         cpumask_var_t mask)
 836 {
 837         static DEFINE_MUTEX(rps_map_mutex);
 838         struct rps_map *old_map, *map;
 839         int cpu, i;
 840
 841         map = kzalloc(max_t(unsigned int,
 842                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 843                       GFP_KERNEL);
 844         if (!map)
 845                 return -ENOMEM;
 846
 847         i = 0;
 848         for_each_cpu_and(cpu, mask, cpu_online_mask)
 849                 map->cpus[i++] = cpu;
 850
 851         if (i) {
 852                 map->len = i;
 853         } else {
 854                 kfree(map);
 855                 map = NULL;
 856         }
 857
 858         mutex_lock(&rps_map_mutex);
 859         old_map = rcu_dereference_protected(queue->rps_map,
 860                                             mutex_is_locked(&rps_map_mutex));
 861         rcu_assign_pointer(queue->rps_map, map);
 862
 863         if (map)
 864                 static_branch_inc(&rps_needed);
 865         if (old_map)
 866                 static_branch_dec(&rps_needed);
 867
 868         mutex_unlock(&rps_map_mutex);
 869
 870         if (old_map)
 871                 kfree_rcu(old_map, rcu);
 872         return 0;
 873 }
 874
 875 int rps_cpumask_housekeeping(struct cpumask *mask)
 876 {
 877         if (!cpumask_empty(mask)) {
 878                 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
 879                 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
 880                 if (cpumask_empty(mask))
 881                         return -EINVAL;
 882         }
 883         return 0;
 884 }
 885
 886 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 887                              const char *buf, size_t len)
 888 {
 889         cpumask_var_t mask;
 890         int err;
 891
 892         if (!capable(CAP_NET_ADMIN))
 893                 return -EPERM;
 894
 895         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 896                 return -ENOMEM;
 897
 898         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 899         if (err)
 900                 goto out;
 901
 902         err = rps_cpumask_housekeeping(mask);
 903         if (err)
 904                 goto out;
 905
 906         err = netdev_rx_queue_set_rps_mask(queue, mask);
 907
 908 out:
 909         free_cpumask_var(mask);
 910         return err ? : len;
 911 }
 912
 913 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 914                                            char *buf)
 915 {
 916         struct rps_dev_flow_table *flow_table;
 917         unsigned long val = 0;
 918
 919         rcu_read_lock();
 920         flow_table = rcu_dereference(queue->rps_flow_table);
 921         if (flow_table)
 922                 val = (unsigned long)flow_table->mask + 1;
 923         rcu_read_unlock();
 924
 925         return sysfs_emit(buf, "%lu\n", val);
 926 }
 927
 928 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 929 {
 930         struct rps_dev_flow_table *table = container_of(rcu,
 931             struct rps_dev_flow_table, rcu);
 932         vfree(table);
 933 }
 934
 935 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 936                                             const char *buf, size_t len)
 937 {
 938         unsigned long mask, count;
 939         struct rps_dev_flow_table *table, *old_table;
 940         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 941         int rc;
 942
 943         if (!capable(CAP_NET_ADMIN))
 944                 return -EPERM;
 945
 946         rc = kstrtoul(buf, 0, &count);
 947         if (rc < 0)
 948                 return rc;
 949
 950         if (count) {
 951                 mask = count - 1;
 952                 /* mask = roundup_pow_of_two(count) - 1;
 953                  * without overflows...
 954                  */
 955                 while ((mask | (mask >> 1)) != mask)
 956                         mask |= (mask >> 1);
 957                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 958                  * and on 32bit arches, must check
 959                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 960                  */
 961 #if BITS_PER_LONG > 32
 962                 if (mask > (unsigned long)(u32)mask)
 963                         return -EINVAL;
 964 #else
 965                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 966                                 / sizeof(struct rps_dev_flow)) {
 967                         /* Enforce a limit to prevent overflow */
 968                         return -EINVAL;
 969                 }
 970 #endif
 971                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 972                 if (!table)
 973                         return -ENOMEM;
 974
 975                 table->mask = mask;
 976                 for (count = 0; count <= mask; count++)
 977                         table->flows[count].cpu = RPS_NO_CPU;
 978         } else {
 979                 table = NULL;
 980         }
 981
 982         spin_lock(&rps_dev_flow_lock);
 983         old_table = rcu_dereference_protected(queue->rps_flow_table,
 984                                               lockdep_is_held(&rps_dev_flow_lock));
 985         rcu_assign_pointer(queue->rps_flow_table, table);
 986         spin_unlock(&rps_dev_flow_lock);
 987
 988         if (old_table)
 989                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 990
 991         return len;
 992 }
 993
 994 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
 995         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
 996
 997 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
 998         = __ATTR(rps_flow_cnt, 0644,
 999                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1000 #endif /* CONFIG_RPS */
1001
1002 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1003 #ifdef CONFIG_RPS
1004         &rps_cpus_attribute.attr,
1005         &rps_dev_flow_table_cnt_attribute.attr,
1006 #endif
1007         NULL
1008 };
1009 ATTRIBUTE_GROUPS(rx_queue_default);
1010
1011 static void rx_queue_release(struct kobject *kobj)
1012 {
1013         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1014 #ifdef CONFIG_RPS
1015         struct rps_map *map;
1016         struct rps_dev_flow_table *flow_table;
1017
1018         map = rcu_dereference_protected(queue->rps_map, 1);
1019         if (map) {
1020                 RCU_INIT_POINTER(queue->rps_map, NULL);
1021                 kfree_rcu(map, rcu);
1022         }
1023
1024         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1025         if (flow_table) {
1026                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1027                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1028         }
1029 #endif
1030
1031         memset(kobj, 0, sizeof(*kobj));
1032         netdev_put(queue->dev, &queue->dev_tracker);
1033 }
1034
1035 static const void *rx_queue_namespace(const struct kobject *kobj)
1036 {
1037         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1038         struct device *dev = &queue->dev->dev;
1039         const void *ns = NULL;
1040
1041         if (dev->class && dev->class->ns_type)
1042                 ns = dev->class->namespace(dev);
1043
1044         return ns;
1045 }
1046
1047 static void rx_queue_get_ownership(const struct kobject *kobj,
1048                                    kuid_t *uid, kgid_t *gid)
1049 {
1050         const struct net *net = rx_queue_namespace(kobj);
1051
1052         net_ns_get_ownership(net, uid, gid);
1053 }
1054
1055 static const struct kobj_type rx_queue_ktype = {
1056         .sysfs_ops = &rx_queue_sysfs_ops,
1057         .release = rx_queue_release,
1058         .default_groups = rx_queue_default_groups,
1059         .namespace = rx_queue_namespace,
1060         .get_ownership = rx_queue_get_ownership,
1061 };
1062
1063 static int rx_queue_default_mask(struct net_device *dev,
1064                                  struct netdev_rx_queue *queue)
1065 {
1066 #if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1067         struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1068
1069         if (rps_default_mask && !cpumask_empty(rps_default_mask))
1070                 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1071 #endif
1072         return 0;
1073 }
1074
1075 static int rx_queue_add_kobject(struct net_device *dev, int index)
1076 {
1077         struct netdev_rx_queue *queue = dev->_rx + index;
1078         struct kobject *kobj = &queue->kobj;
1079         int error = 0;
1080
1081         /* Kobject_put later will trigger rx_queue_release call which
1082          * decreases dev refcount: Take that reference here
1083          */
1084         netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1085
1086         kobj->kset = dev->queues_kset;
1087         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1088                                      "rx-%u", index);
1089         if (error)
1090                 goto err;
1091
1092         if (dev->sysfs_rx_queue_group) {
1093                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1094                 if (error)
1095                         goto err;
1096         }
1097
1098         error = rx_queue_default_mask(dev, queue);
1099         if (error)
1100                 goto err;
1101
1102         kobject_uevent(kobj, KOBJ_ADD);
1103
1104         return error;
1105
1106 err:
1107         kobject_put(kobj);
1108         return error;
1109 }
1110
1111 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1112                                  kgid_t kgid)
1113 {
1114         struct netdev_rx_queue *queue = dev->_rx + index;
1115         struct kobject *kobj = &queue->kobj;
1116         int error;
1117
1118         error = sysfs_change_owner(kobj, kuid, kgid);
1119         if (error)
1120                 return error;
1121
1122         if (dev->sysfs_rx_queue_group)
1123                 error = sysfs_group_change_owner(
1124                         kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1125
1126         return error;
1127 }
1128 #endif /* CONFIG_SYSFS */
1129
1130 int
1131 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1132 {
1133 #ifdef CONFIG_SYSFS
1134         int i;
1135         int error = 0;
1136
1137 #ifndef CONFIG_RPS
1138         if (!dev->sysfs_rx_queue_group)
1139                 return 0;
1140 #endif
1141         for (i = old_num; i < new_num; i++) {
1142                 error = rx_queue_add_kobject(dev, i);
1143                 if (error) {
1144                         new_num = old_num;
1145                         break;
1146                 }
1147         }
1148
1149         while (--i >= new_num) {
1150                 struct kobject *kobj = &dev->_rx[i].kobj;
1151
1152                 if (!refcount_read(&dev_net(dev)->ns.count))
1153                         kobj->uevent_suppress = 1;
1154                 if (dev->sysfs_rx_queue_group)
1155                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1156                 kobject_put(kobj);
1157         }
1158
1159         return error;
1160 #else
1161         return 0;
1162 #endif
1163 }
1164
1165 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1166                                      kuid_t kuid, kgid_t kgid)
1167 {
1168 #ifdef CONFIG_SYSFS
1169         int error = 0;
1170         int i;
1171
1172 #ifndef CONFIG_RPS
1173         if (!dev->sysfs_rx_queue_group)
1174                 return 0;
1175 #endif
1176         for (i = 0; i < num; i++) {
1177                 error = rx_queue_change_owner(dev, i, kuid, kgid);
1178                 if (error)
1179                         break;
1180         }
1181
1182         return error;
1183 #else
1184         return 0;
1185 #endif
1186 }
1187
1188 #ifdef CONFIG_SYSFS
1189 /*
1190  * netdev_queue sysfs structures and functions.
1191  */
1192 struct netdev_queue_attribute {
1193         struct attribute attr;
1194         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1195         ssize_t (*store)(struct netdev_queue *queue,
1196                          const char *buf, size_t len);
1197 };
1198 #define to_netdev_queue_attr(_attr) \
1199         container_of(_attr, struct netdev_queue_attribute, attr)
1200
1201 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1202
1203 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1204                                       struct attribute *attr, char *buf)
1205 {
1206         const struct netdev_queue_attribute *attribute
1207                 = to_netdev_queue_attr(attr);
1208         struct netdev_queue *queue = to_netdev_queue(kobj);
1209
1210         if (!attribute->show)
1211                 return -EIO;
1212
1213         return attribute->show(queue, buf);
1214 }
1215
1216 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1217                                        struct attribute *attr,
1218                                        const char *buf, size_t count)
1219 {
1220         const struct netdev_queue_attribute *attribute
1221                 = to_netdev_queue_attr(attr);
1222         struct netdev_queue *queue = to_netdev_queue(kobj);
1223
1224         if (!attribute->store)
1225                 return -EIO;
1226
1227         return attribute->store(queue, buf, count);
1228 }
1229
1230 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1231         .show = netdev_queue_attr_show,
1232         .store = netdev_queue_attr_store,
1233 };
1234
1235 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1236 {
1237         unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1238
1239         return sysfs_emit(buf, fmt_ulong, trans_timeout);
1240 }
1241
1242 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1243 {
1244         struct net_device *dev = queue->dev;
1245         unsigned int i;
1246
1247         i = queue - dev->_tx;
1248         BUG_ON(i >= dev->num_tx_queues);
1249
1250         return i;
1251 }
1252
1253 static ssize_t traffic_class_show(struct netdev_queue *queue,
1254                                   char *buf)
1255 {
1256         struct net_device *dev = queue->dev;
1257         int num_tc, tc;
1258         int index;
1259
1260         if (!netif_is_multiqueue(dev))
1261                 return -ENOENT;
1262
1263         if (!rtnl_trylock())
1264                 return restart_syscall();
1265
1266         index = get_netdev_queue_index(queue);
1267
1268         /* If queue belongs to subordinate dev use its TC mapping */
1269         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1270
1271         num_tc = dev->num_tc;
1272         tc = netdev_txq_to_tc(dev, index);
1273
1274         rtnl_unlock();
1275
1276         if (tc < 0)
1277                 return -EINVAL;
1278
1279         /* We can report the traffic class one of two ways:
1280          * Subordinate device traffic classes are reported with the traffic
1281          * class first, and then the subordinate class so for example TC0 on
1282          * subordinate device 2 will be reported as "0-2". If the queue
1283          * belongs to the root device it will be reported with just the
1284          * traffic class, so just "0" for TC 0 for example.
1285          */
1286         return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1287                             sysfs_emit(buf, "%d\n", tc);
1288 }
1289
1290 #ifdef CONFIG_XPS
1291 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1292                                char *buf)
1293 {
1294         return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1295 }
1296
1297 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1298                                 const char *buf, size_t len)
1299 {
1300         struct net_device *dev = queue->dev;
1301         int err, index = get_netdev_queue_index(queue);
1302         u32 rate = 0;
1303
1304         if (!capable(CAP_NET_ADMIN))
1305                 return -EPERM;
1306
1307         /* The check is also done later; this helps returning early without
1308          * hitting the trylock/restart below.
1309          */
1310         if (!dev->netdev_ops->ndo_set_tx_maxrate)
1311                 return -EOPNOTSUPP;
1312
1313         err = kstrtou32(buf, 10, &rate);
1314         if (err < 0)
1315                 return err;
1316
1317         if (!rtnl_trylock())
1318                 return restart_syscall();
1319
1320         err = -EOPNOTSUPP;
1321         if (dev->netdev_ops->ndo_set_tx_maxrate)
1322                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1323
1324         rtnl_unlock();
1325         if (!err) {
1326                 queue->tx_maxrate = rate;
1327                 return len;
1328         }
1329         return err;
1330 }
1331
1332 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1333         = __ATTR_RW(tx_maxrate);
1334 #endif
1335
1336 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1337         = __ATTR_RO(tx_timeout);
1338
1339 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1340         = __ATTR_RO(traffic_class);
1341
1342 #ifdef CONFIG_BQL
1343 /*
1344  * Byte queue limits sysfs structures and functions.
1345  */
1346 static ssize_t bql_show(char *buf, unsigned int value)
1347 {
1348         return sysfs_emit(buf, "%u\n", value);
1349 }
1350
1351 static ssize_t bql_set(const char *buf, const size_t count,
1352                        unsigned int *pvalue)
1353 {
1354         unsigned int value;
1355         int err;
1356
1357         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1358                 value = DQL_MAX_LIMIT;
1359         } else {
1360                 err = kstrtouint(buf, 10, &value);
1361                 if (err < 0)
1362                         return err;
1363                 if (value > DQL_MAX_LIMIT)
1364                         return -EINVAL;
1365         }
1366
1367         *pvalue = value;
1368
1369         return count;
1370 }
1371
1372 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1373                                   char *buf)
1374 {
1375         struct dql *dql = &queue->dql;
1376
1377         return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1378 }
1379
1380 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1381                                  const char *buf, size_t len)
1382 {
1383         struct dql *dql = &queue->dql;
1384         unsigned int value;
1385         int err;
1386
1387         err = kstrtouint(buf, 10, &value);
1388         if (err < 0)
1389                 return err;
1390
1391         dql->slack_hold_time = msecs_to_jiffies(value);
1392
1393         return len;
1394 }
1395
1396 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1397         = __ATTR(hold_time, 0644,
1398                  bql_show_hold_time, bql_set_hold_time);
1399
1400 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1401                                  char *buf)
1402 {
1403         struct dql *dql = &queue->dql;
1404
1405         return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1406 }
1407
1408 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1409         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1410
1411 #define BQL_ATTR(NAME, FIELD)                                           \
1412 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1413                                  char *buf)                             \
1414 {                                                                       \
1415         return bql_show(buf, queue->dql.FIELD);                         \
1416 }                                                                       \
1417                                                                         \
1418 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1419                                 const char *buf, size_t len)            \
1420 {                                                                       \
1421         return bql_set(buf, len, &queue->dql.FIELD);                    \
1422 }                                                                       \
1423                                                                         \
1424 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1425         = __ATTR(NAME, 0644,                            \
1426                  bql_show_ ## NAME, bql_set_ ## NAME)
1427
1428 BQL_ATTR(limit, limit);
1429 BQL_ATTR(limit_max, max_limit);
1430 BQL_ATTR(limit_min, min_limit);
1431
1432 static struct attribute *dql_attrs[] __ro_after_init = {
1433         &bql_limit_attribute.attr,
1434         &bql_limit_max_attribute.attr,
1435         &bql_limit_min_attribute.attr,
1436         &bql_hold_time_attribute.attr,
1437         &bql_inflight_attribute.attr,
1438         NULL
1439 };
1440
1441 static const struct attribute_group dql_group = {
1442         .name  = "byte_queue_limits",
1443         .attrs  = dql_attrs,
1444 };
1445 #endif /* CONFIG_BQL */
1446
1447 #ifdef CONFIG_XPS
1448 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1449                               int tc, char *buf, enum xps_map_type type)
1450 {
1451         struct xps_dev_maps *dev_maps;
1452         unsigned long *mask;
1453         unsigned int nr_ids;
1454         int j, len;
1455
1456         rcu_read_lock();
1457         dev_maps = rcu_dereference(dev->xps_maps[type]);
1458
1459         /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1460          * when dev_maps hasn't been allocated yet, to be backward compatible.
1461          */
1462         nr_ids = dev_maps ? dev_maps->nr_ids :
1463                  (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1464
1465         mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1466         if (!mask) {
1467                 rcu_read_unlock();
1468                 return -ENOMEM;
1469         }
1470
1471         if (!dev_maps || tc >= dev_maps->num_tc)
1472                 goto out_no_maps;
1473
1474         for (j = 0; j < nr_ids; j++) {
1475                 int i, tci = j * dev_maps->num_tc + tc;
1476                 struct xps_map *map;
1477
1478                 map = rcu_dereference(dev_maps->attr_map[tci]);
1479                 if (!map)
1480                         continue;
1481
1482                 for (i = map->len; i--;) {
1483                         if (map->queues[i] == index) {
1484                                 __set_bit(j, mask);
1485                                 break;
1486                         }
1487                 }
1488         }
1489 out_no_maps:
1490         rcu_read_unlock();
1491
1492         len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1493         bitmap_free(mask);
1494
1495         return len < PAGE_SIZE ? len : -EINVAL;
1496 }
1497
1498 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1499 {
1500         struct net_device *dev = queue->dev;
1501         unsigned int index;
1502         int len, tc;
1503
1504         if (!netif_is_multiqueue(dev))
1505                 return -ENOENT;
1506
1507         index = get_netdev_queue_index(queue);
1508
1509         if (!rtnl_trylock())
1510                 return restart_syscall();
1511
1512         /* If queue belongs to subordinate dev use its map */
1513         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1514
1515         tc = netdev_txq_to_tc(dev, index);
1516         if (tc < 0) {
1517                 rtnl_unlock();
1518                 return -EINVAL;
1519         }
1520
1521         /* Make sure the subordinate device can't be freed */
1522         get_device(&dev->dev);
1523         rtnl_unlock();
1524
1525         len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1526
1527         put_device(&dev->dev);
1528         return len;
1529 }
1530
1531 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1532                               const char *buf, size_t len)
1533 {
1534         struct net_device *dev = queue->dev;
1535         unsigned int index;
1536         cpumask_var_t mask;
1537         int err;
1538
1539         if (!netif_is_multiqueue(dev))
1540                 return -ENOENT;
1541
1542         if (!capable(CAP_NET_ADMIN))
1543                 return -EPERM;
1544
1545         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1546                 return -ENOMEM;
1547
1548         index = get_netdev_queue_index(queue);
1549
1550         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1551         if (err) {
1552                 free_cpumask_var(mask);
1553                 return err;
1554         }
1555
1556         if (!rtnl_trylock()) {
1557                 free_cpumask_var(mask);
1558                 return restart_syscall();
1559         }
1560
1561         err = netif_set_xps_queue(dev, mask, index);
1562         rtnl_unlock();
1563
1564         free_cpumask_var(mask);
1565
1566         return err ? : len;
1567 }
1568
1569 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1570         = __ATTR_RW(xps_cpus);
1571
1572 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1573 {
1574         struct net_device *dev = queue->dev;
1575         unsigned int index;
1576         int tc;
1577
1578         index = get_netdev_queue_index(queue);
1579
1580         if (!rtnl_trylock())
1581                 return restart_syscall();
1582
1583         tc = netdev_txq_to_tc(dev, index);
1584         rtnl_unlock();
1585         if (tc < 0)
1586                 return -EINVAL;
1587
1588         return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1589 }
1590
1591 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1592                               size_t len)
1593 {
1594         struct net_device *dev = queue->dev;
1595         struct net *net = dev_net(dev);
1596         unsigned long *mask;
1597         unsigned int index;
1598         int err;
1599
1600         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1601                 return -EPERM;
1602
1603         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1604         if (!mask)
1605                 return -ENOMEM;
1606
1607         index = get_netdev_queue_index(queue);
1608
1609         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1610         if (err) {
1611                 bitmap_free(mask);
1612                 return err;
1613         }
1614
1615         if (!rtnl_trylock()) {
1616                 bitmap_free(mask);
1617                 return restart_syscall();
1618         }
1619
1620         cpus_read_lock();
1621         err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1622         cpus_read_unlock();
1623
1624         rtnl_unlock();
1625
1626         bitmap_free(mask);
1627         return err ? : len;
1628 }
1629
1630 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1631         = __ATTR_RW(xps_rxqs);
1632 #endif /* CONFIG_XPS */
1633
1634 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1635         &queue_trans_timeout.attr,
1636         &queue_traffic_class.attr,
1637 #ifdef CONFIG_XPS
1638         &xps_cpus_attribute.attr,
1639         &xps_rxqs_attribute.attr,
1640         &queue_tx_maxrate.attr,
1641 #endif
1642         NULL
1643 };
1644 ATTRIBUTE_GROUPS(netdev_queue_default);
1645
1646 static void netdev_queue_release(struct kobject *kobj)
1647 {
1648         struct netdev_queue *queue = to_netdev_queue(kobj);
1649
1650         memset(kobj, 0, sizeof(*kobj));
1651         netdev_put(queue->dev, &queue->dev_tracker);
1652 }
1653
1654 static const void *netdev_queue_namespace(const struct kobject *kobj)
1655 {
1656         struct netdev_queue *queue = to_netdev_queue(kobj);
1657         struct device *dev = &queue->dev->dev;
1658         const void *ns = NULL;
1659
1660         if (dev->class && dev->class->ns_type)
1661                 ns = dev->class->namespace(dev);
1662
1663         return ns;
1664 }
1665
1666 static void netdev_queue_get_ownership(const struct kobject *kobj,
1667                                        kuid_t *uid, kgid_t *gid)
1668 {
1669         const struct net *net = netdev_queue_namespace(kobj);
1670
1671         net_ns_get_ownership(net, uid, gid);
1672 }
1673
1674 static const struct kobj_type netdev_queue_ktype = {
1675         .sysfs_ops = &netdev_queue_sysfs_ops,
1676         .release = netdev_queue_release,
1677         .default_groups = netdev_queue_default_groups,
1678         .namespace = netdev_queue_namespace,
1679         .get_ownership = netdev_queue_get_ownership,
1680 };
1681
1682 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1683 {
1684         struct netdev_queue *queue = dev->_tx + index;
1685         struct kobject *kobj = &queue->kobj;
1686         int error = 0;
1687
1688         /* Kobject_put later will trigger netdev_queue_release call
1689          * which decreases dev refcount: Take that reference here
1690          */
1691         netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1692
1693         kobj->kset = dev->queues_kset;
1694         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1695                                      "tx-%u", index);
1696         if (error)
1697                 goto err;
1698
1699 #ifdef CONFIG_BQL
1700         error = sysfs_create_group(kobj, &dql_group);
1701         if (error)
1702                 goto err;
1703 #endif
1704
1705         kobject_uevent(kobj, KOBJ_ADD);
1706         return 0;
1707
1708 err:
1709         kobject_put(kobj);
1710         return error;
1711 }
1712
1713 static int tx_queue_change_owner(struct net_device *ndev, int index,
1714                                  kuid_t kuid, kgid_t kgid)
1715 {
1716         struct netdev_queue *queue = ndev->_tx + index;
1717         struct kobject *kobj = &queue->kobj;
1718         int error;
1719
1720         error = sysfs_change_owner(kobj, kuid, kgid);
1721         if (error)
1722                 return error;
1723
1724 #ifdef CONFIG_BQL
1725         error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1726 #endif
1727         return error;
1728 }
1729 #endif /* CONFIG_SYSFS */
1730
1731 int
1732 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1733 {
1734 #ifdef CONFIG_SYSFS
1735         int i;
1736         int error = 0;
1737
1738         /* Tx queue kobjects are allowed to be updated when a device is being
1739          * unregistered, but solely to remove queues from qdiscs. Any path
1740          * adding queues should be fixed.
1741          */
1742         WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1743              "New queues can't be registered after device unregistration.");
1744
1745         for (i = old_num; i < new_num; i++) {
1746                 error = netdev_queue_add_kobject(dev, i);
1747                 if (error) {
1748                         new_num = old_num;
1749                         break;
1750                 }
1751         }
1752
1753         while (--i >= new_num) {
1754                 struct netdev_queue *queue = dev->_tx + i;
1755
1756                 if (!refcount_read(&dev_net(dev)->ns.count))
1757                         queue->kobj.uevent_suppress = 1;
1758 #ifdef CONFIG_BQL
1759                 sysfs_remove_group(&queue->kobj, &dql_group);
1760 #endif
1761                 kobject_put(&queue->kobj);
1762         }
1763
1764         return error;
1765 #else
1766         return 0;
1767 #endif /* CONFIG_SYSFS */
1768 }
1769
1770 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1771                                      kuid_t kuid, kgid_t kgid)
1772 {
1773 #ifdef CONFIG_SYSFS
1774         int error = 0;
1775         int i;
1776
1777         for (i = 0; i < num; i++) {
1778                 error = tx_queue_change_owner(dev, i, kuid, kgid);
1779                 if (error)
1780                         break;
1781         }
1782
1783         return error;
1784 #else
1785         return 0;
1786 #endif /* CONFIG_SYSFS */
1787 }
1788
1789 static int register_queue_kobjects(struct net_device *dev)
1790 {
1791         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1792
1793 #ifdef CONFIG_SYSFS
1794         dev->queues_kset = kset_create_and_add("queues",
1795                                                NULL, &dev->dev.kobj);
1796         if (!dev->queues_kset)
1797                 return -ENOMEM;
1798         real_rx = dev->real_num_rx_queues;
1799 #endif
1800         real_tx = dev->real_num_tx_queues;
1801
1802         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1803         if (error)
1804                 goto error;
1805         rxq = real_rx;
1806
1807         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1808         if (error)
1809                 goto error;
1810         txq = real_tx;
1811
1812         return 0;
1813
1814 error:
1815         netdev_queue_update_kobjects(dev, txq, 0);
1816         net_rx_queue_update_kobjects(dev, rxq, 0);
1817 #ifdef CONFIG_SYSFS
1818         kset_unregister(dev->queues_kset);
1819 #endif
1820         return error;
1821 }
1822
1823 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1824 {
1825         int error = 0, real_rx = 0, real_tx = 0;
1826
1827 #ifdef CONFIG_SYSFS
1828         if (ndev->queues_kset) {
1829                 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1830                 if (error)
1831                         return error;
1832         }
1833         real_rx = ndev->real_num_rx_queues;
1834 #endif
1835         real_tx = ndev->real_num_tx_queues;
1836
1837         error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1838         if (error)
1839                 return error;
1840
1841         error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1842         if (error)
1843                 return error;
1844
1845         return 0;
1846 }
1847
1848 static void remove_queue_kobjects(struct net_device *dev)
1849 {
1850         int real_rx = 0, real_tx = 0;
1851
1852 #ifdef CONFIG_SYSFS
1853         real_rx = dev->real_num_rx_queues;
1854 #endif
1855         real_tx = dev->real_num_tx_queues;
1856
1857         net_rx_queue_update_kobjects(dev, real_rx, 0);
1858         netdev_queue_update_kobjects(dev, real_tx, 0);
1859
1860         dev->real_num_rx_queues = 0;
1861         dev->real_num_tx_queues = 0;
1862 #ifdef CONFIG_SYSFS
1863         kset_unregister(dev->queues_kset);
1864 #endif
1865 }
1866
1867 static bool net_current_may_mount(void)
1868 {
1869         struct net *net = current->nsproxy->net_ns;
1870
1871         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1872 }
1873
1874 static void *net_grab_current_ns(void)
1875 {
1876         struct net *ns = current->nsproxy->net_ns;
1877 #ifdef CONFIG_NET_NS
1878         if (ns)
1879                 refcount_inc(&ns->passive);
1880 #endif
1881         return ns;
1882 }
1883
1884 static const void *net_initial_ns(void)
1885 {
1886         return &init_net;
1887 }
1888
1889 static const void *net_netlink_ns(struct sock *sk)
1890 {
1891         return sock_net(sk);
1892 }
1893
1894 const struct kobj_ns_type_operations net_ns_type_operations = {
1895         .type = KOBJ_NS_TYPE_NET,
1896         .current_may_mount = net_current_may_mount,
1897         .grab_current_ns = net_grab_current_ns,
1898         .netlink_ns = net_netlink_ns,
1899         .initial_ns = net_initial_ns,
1900         .drop_ns = net_drop_ns,
1901 };
1902 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1903
1904 static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1905 {
1906         const struct net_device *dev = to_net_dev(d);
1907         int retval;
1908
1909         /* pass interface to uevent. */
1910         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1911         if (retval)
1912                 goto exit;
1913
1914         /* pass ifindex to uevent.
1915          * ifindex is useful as it won't change (interface name may change)
1916          * and is what RtNetlink uses natively.
1917          */
1918         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1919
1920 exit:
1921         return retval;
1922 }
1923
1924 /*
1925  *      netdev_release -- destroy and free a dead device.
1926  *      Called when last reference to device kobject is gone.
1927  */
1928 static void netdev_release(struct device *d)
1929 {
1930         struct net_device *dev = to_net_dev(d);
1931
1932         BUG_ON(dev->reg_state != NETREG_RELEASED);
1933
1934         /* no need to wait for rcu grace period:
1935          * device is dead and about to be freed.
1936          */
1937         kfree(rcu_access_pointer(dev->ifalias));
1938         netdev_freemem(dev);
1939 }
1940
1941 static const void *net_namespace(const struct device *d)
1942 {
1943         const struct net_device *dev = to_net_dev(d);
1944
1945         return dev_net(dev);
1946 }
1947
1948 static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1949 {
1950         const struct net_device *dev = to_net_dev(d);
1951         const struct net *net = dev_net(dev);
1952
1953         net_ns_get_ownership(net, uid, gid);
1954 }
1955
1956 static struct class net_class __ro_after_init = {
1957         .name = "net",
1958         .dev_release = netdev_release,
1959         .dev_groups = net_class_groups,
1960         .dev_uevent = netdev_uevent,
1961         .ns_type = &net_ns_type_operations,
1962         .namespace = net_namespace,
1963         .get_ownership = net_get_ownership,
1964 };
1965
1966 #ifdef CONFIG_OF
1967 static int of_dev_node_match(struct device *dev, const void *data)
1968 {
1969         for (; dev; dev = dev->parent) {
1970                 if (dev->of_node == data)
1971                         return 1;
1972         }
1973
1974         return 0;
1975 }
1976
1977 /*
1978  * of_find_net_device_by_node - lookup the net device for the device node
1979  * @np: OF device node
1980  *
1981  * Looks up the net_device structure corresponding with the device node.
1982  * If successful, returns a pointer to the net_device with the embedded
1983  * struct device refcount incremented by one, or NULL on failure. The
1984  * refcount must be dropped when done with the net_device.
1985  */
1986 struct net_device *of_find_net_device_by_node(struct device_node *np)
1987 {
1988         struct device *dev;
1989
1990         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1991         if (!dev)
1992                 return NULL;
1993
1994         return to_net_dev(dev);
1995 }
1996 EXPORT_SYMBOL(of_find_net_device_by_node);
1997 #endif
1998
1999 /* Delete sysfs entries but hold kobject reference until after all
2000  * netdev references are gone.
2001  */
2002 void netdev_unregister_kobject(struct net_device *ndev)
2003 {
2004         struct device *dev = &ndev->dev;
2005
2006         if (!refcount_read(&dev_net(ndev)->ns.count))
2007                 dev_set_uevent_suppress(dev, 1);
2008
2009         kobject_get(&dev->kobj);
2010
2011         remove_queue_kobjects(ndev);
2012
2013         pm_runtime_set_memalloc_noio(dev, false);
2014
2015         device_del(dev);
2016 }
2017
2018 /* Create sysfs entries for network device. */
2019 int netdev_register_kobject(struct net_device *ndev)
2020 {
2021         struct device *dev = &ndev->dev;
2022         const struct attribute_group **groups = ndev->sysfs_groups;
2023         int error = 0;
2024
2025         device_initialize(dev);
2026         dev->class = &net_class;
2027         dev->platform_data = ndev;
2028         dev->groups = groups;
2029
2030         dev_set_name(dev, "%s", ndev->name);
2031
2032 #ifdef CONFIG_SYSFS
2033         /* Allow for a device specific group */
2034         if (*groups)
2035                 groups++;
2036
2037         *groups++ = &netstat_group;
2038
2039         if (wireless_group_needed(ndev))
2040                 *groups++ = &wireless_group;
2041 #endif /* CONFIG_SYSFS */
2042
2043         error = device_add(dev);
2044         if (error)
2045                 return error;
2046
2047         error = register_queue_kobjects(ndev);
2048         if (error) {
2049                 device_del(dev);
2050                 return error;
2051         }
2052
2053         pm_runtime_set_memalloc_noio(dev, true);
2054
2055         return error;
2056 }
2057
2058 /* Change owner for sysfs entries when moving network devices across network
2059  * namespaces owned by different user namespaces.
2060  */
2061 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2062                         const struct net *net_new)
2063 {
2064         kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2065         kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2066         struct device *dev = &ndev->dev;
2067         int error;
2068
2069         net_ns_get_ownership(net_old, &old_uid, &old_gid);
2070         net_ns_get_ownership(net_new, &new_uid, &new_gid);
2071
2072         /* The network namespace was changed but the owning user namespace is
2073          * identical so there's no need to change the owner of sysfs entries.
2074          */
2075         if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2076                 return 0;
2077
2078         error = device_change_owner(dev, new_uid, new_gid);
2079         if (error)
2080                 return error;
2081
2082         error = queue_change_owner(ndev, new_uid, new_gid);
2083         if (error)
2084                 return error;
2085
2086         return 0;
2087 }
2088
2089 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2090                                 const void *ns)
2091 {
2092         return class_create_file_ns(&net_class, class_attr, ns);
2093 }
2094 EXPORT_SYMBOL(netdev_class_create_file_ns);
2095
2096 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2097                                  const void *ns)
2098 {
2099         class_remove_file_ns(&net_class, class_attr, ns);
2100 }
2101 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2102
2103 int __init netdev_kobject_init(void)
2104 {
2105         kobj_ns_type_register(&net_ns_type_operations);
2106         return class_register(&net_class);
2107 }