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 #include <net/netdev_rx_queue.h>
  27 #include <net/rps.h>
  28
  29 #include "dev.h"
  30 #include "net-sysfs.h"
  31
  32 #ifdef CONFIG_SYSFS
  33 static const char fmt_hex[] = "%#x\n";
  34 static const char fmt_dec[] = "%d\n";
  35 static const char fmt_uint[] = "%u\n";
  36 static const char fmt_ulong[] = "%lu\n";
  37 static const char fmt_u64[] = "%llu\n";
  38
  39 /* Caller holds RTNL or RCU */
  40 static inline int dev_isalive(const struct net_device *dev)
  41 {
  42         return READ_ONCE(dev->reg_state) <= NETREG_REGISTERED;
  43 }
  44
  45 /* use same locking rules as GIF* ioctl's */
  46 static ssize_t netdev_show(const struct device *dev,
  47                            struct device_attribute *attr, char *buf,
  48                            ssize_t (*format)(const struct net_device *, char *))
  49 {
  50         struct net_device *ndev = to_net_dev(dev);
  51         ssize_t ret = -EINVAL;
  52
  53         rcu_read_lock();
  54         if (dev_isalive(ndev))
  55                 ret = (*format)(ndev, buf);
  56         rcu_read_unlock();
  57
  58         return ret;
  59 }
  60
  61 /* generate a show function for simple field */
  62 #define NETDEVICE_SHOW(field, format_string)                            \
  63 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  64 {                                                                       \
  65         return sysfs_emit(buf, format_string, READ_ONCE(dev->field));           \
  66 }                                                                       \
  67 static ssize_t field##_show(struct device *dev,                         \
  68                             struct device_attribute *attr, char *buf)   \
  69 {                                                                       \
  70         return netdev_show(dev, attr, buf, format_##field);             \
  71 }                                                                       \
  72
  73 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  74 NETDEVICE_SHOW(field, format_string);                                   \
  75 static DEVICE_ATTR_RO(field)
  76
  77 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  78 NETDEVICE_SHOW(field, format_string);                                   \
  79 static DEVICE_ATTR_RW(field)
  80
  81 /* use same locking and permission rules as SIF* ioctl's */
  82 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  83                             const char *buf, size_t len,
  84                             int (*set)(struct net_device *, unsigned long))
  85 {
  86         struct net_device *netdev = to_net_dev(dev);
  87         struct net *net = dev_net(netdev);
  88         unsigned long new;
  89         int ret;
  90
  91         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  92                 return -EPERM;
  93
  94         ret = kstrtoul(buf, 0, &new);
  95         if (ret)
  96                 goto err;
  97
  98         if (!rtnl_trylock())
  99                 return restart_syscall();
 100
 101         if (dev_isalive(netdev)) {
 102                 ret = (*set)(netdev, new);
 103                 if (ret == 0)
 104                         ret = len;
 105         }
 106         rtnl_unlock();
 107  err:
 108         return ret;
 109 }
 110
 111 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 112 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 113 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 114 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 115 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 116 NETDEVICE_SHOW_RO(type, fmt_dec);
 117 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 118
 119 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 120                            char *buf)
 121 {
 122         struct net_device *ndev = to_net_dev(dev);
 123
 124         return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
 125 }
 126 static DEVICE_ATTR_RO(iflink);
 127
 128 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 129 {
 130         return sysfs_emit(buf, fmt_dec, READ_ONCE(dev->name_assign_type));
 131 }
 132
 133 static ssize_t name_assign_type_show(struct device *dev,
 134                                      struct device_attribute *attr,
 135                                      char *buf)
 136 {
 137         struct net_device *ndev = to_net_dev(dev);
 138         ssize_t ret = -EINVAL;
 139
 140         if (READ_ONCE(ndev->name_assign_type) != NET_NAME_UNKNOWN)
 141                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 142
 143         return ret;
 144 }
 145 static DEVICE_ATTR_RO(name_assign_type);
 146
 147 /* use same locking rules as GIFHWADDR ioctl's (dev_get_mac_address()) */
 148 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 149                             char *buf)
 150 {
 151         struct net_device *ndev = to_net_dev(dev);
 152         ssize_t ret = -EINVAL;
 153
 154         down_read(&dev_addr_sem);
 155
 156         rcu_read_lock();
 157         if (dev_isalive(ndev))
 158                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 159         rcu_read_unlock();
 160
 161         up_read(&dev_addr_sem);
 162         return ret;
 163 }
 164 static DEVICE_ATTR_RO(address);
 165
 166 static ssize_t broadcast_show(struct device *dev,
 167                               struct device_attribute *attr, char *buf)
 168 {
 169         struct net_device *ndev = to_net_dev(dev);
 170         int ret = -EINVAL;
 171
 172         rcu_read_lock();
 173         if (dev_isalive(ndev))
 174                 ret = sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 175         rcu_read_unlock();
 176         return ret;
 177 }
 178 static DEVICE_ATTR_RO(broadcast);
 179
 180 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 181 {
 182         if (!netif_running(dev))
 183                 return -EINVAL;
 184         return dev_change_carrier(dev, (bool)new_carrier);
 185 }
 186
 187 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 188                              const char *buf, size_t len)
 189 {
 190         struct net_device *netdev = to_net_dev(dev);
 191
 192         /* The check is also done in change_carrier; this helps returning early
 193          * without hitting the trylock/restart in netdev_store.
 194          */
 195         if (!netdev->netdev_ops->ndo_change_carrier)
 196                 return -EOPNOTSUPP;
 197
 198         return netdev_store(dev, attr, buf, len, change_carrier);
 199 }
 200
 201 static ssize_t carrier_show(struct device *dev,
 202                             struct device_attribute *attr, char *buf)
 203 {
 204         struct net_device *netdev = to_net_dev(dev);
 205         int ret = -EINVAL;
 206
 207         if (!rtnl_trylock())
 208                 return restart_syscall();
 209
 210         if (netif_running(netdev)) {
 211                 /* Synchronize carrier state with link watch,
 212                  * see also rtnl_getlink().
 213                  */
 214                 linkwatch_sync_dev(netdev);
 215
 216                 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
 217         }
 218         rtnl_unlock();
 219
 220         return ret;
 221 }
 222 static DEVICE_ATTR_RW(carrier);
 223
 224 static ssize_t speed_show(struct device *dev,
 225                           struct device_attribute *attr, char *buf)
 226 {
 227         struct net_device *netdev = to_net_dev(dev);
 228         int ret = -EINVAL;
 229
 230         /* The check is also done in __ethtool_get_link_ksettings; this helps
 231          * returning early without hitting the trylock/restart below.
 232          */
 233         if (!netdev->ethtool_ops->get_link_ksettings)
 234                 return ret;
 235
 236         if (!rtnl_trylock())
 237                 return restart_syscall();
 238
 239         if (netif_running(netdev)) {
 240                 struct ethtool_link_ksettings cmd;
 241
 242                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
 243                         ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
 244         }
 245         rtnl_unlock();
 246         return ret;
 247 }
 248 static DEVICE_ATTR_RO(speed);
 249
 250 static ssize_t duplex_show(struct device *dev,
 251                            struct device_attribute *attr, char *buf)
 252 {
 253         struct net_device *netdev = to_net_dev(dev);
 254         int ret = -EINVAL;
 255
 256         /* The check is also done in __ethtool_get_link_ksettings; this helps
 257          * returning early without hitting the trylock/restart below.
 258          */
 259         if (!netdev->ethtool_ops->get_link_ksettings)
 260                 return ret;
 261
 262         if (!rtnl_trylock())
 263                 return restart_syscall();
 264
 265         if (netif_running(netdev)) {
 266                 struct ethtool_link_ksettings cmd;
 267
 268                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
 269                         const char *duplex;
 270
 271                         switch (cmd.base.duplex) {
 272                         case DUPLEX_HALF:
 273                                 duplex = "half";
 274                                 break;
 275                         case DUPLEX_FULL:
 276                                 duplex = "full";
 277                                 break;
 278                         default:
 279                                 duplex = "unknown";
 280                                 break;
 281                         }
 282                         ret = sysfs_emit(buf, "%s\n", duplex);
 283                 }
 284         }
 285         rtnl_unlock();
 286         return ret;
 287 }
 288 static DEVICE_ATTR_RO(duplex);
 289
 290 static ssize_t testing_show(struct device *dev,
 291                             struct device_attribute *attr, char *buf)
 292 {
 293         struct net_device *netdev = to_net_dev(dev);
 294
 295         if (netif_running(netdev))
 296                 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
 297
 298         return -EINVAL;
 299 }
 300 static DEVICE_ATTR_RO(testing);
 301
 302 static ssize_t dormant_show(struct device *dev,
 303                             struct device_attribute *attr, char *buf)
 304 {
 305         struct net_device *netdev = to_net_dev(dev);
 306
 307         if (netif_running(netdev))
 308                 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
 309
 310         return -EINVAL;
 311 }
 312 static DEVICE_ATTR_RO(dormant);
 313
 314 static const char *const operstates[] = {
 315         "unknown",
 316         "notpresent", /* currently unused */
 317         "down",
 318         "lowerlayerdown",
 319         "testing",
 320         "dormant",
 321         "up"
 322 };
 323
 324 static ssize_t operstate_show(struct device *dev,
 325                               struct device_attribute *attr, char *buf)
 326 {
 327         const struct net_device *netdev = to_net_dev(dev);
 328         unsigned char operstate;
 329
 330         operstate = READ_ONCE(netdev->operstate);
 331         if (!netif_running(netdev))
 332                 operstate = IF_OPER_DOWN;
 333
 334         if (operstate >= ARRAY_SIZE(operstates))
 335                 return -EINVAL; /* should not happen */
 336
 337         return sysfs_emit(buf, "%s\n", operstates[operstate]);
 338 }
 339 static DEVICE_ATTR_RO(operstate);
 340
 341 static ssize_t carrier_changes_show(struct device *dev,
 342                                     struct device_attribute *attr,
 343                                     char *buf)
 344 {
 345         struct net_device *netdev = to_net_dev(dev);
 346
 347         return sysfs_emit(buf, fmt_dec,
 348                           atomic_read(&netdev->carrier_up_count) +
 349                           atomic_read(&netdev->carrier_down_count));
 350 }
 351 static DEVICE_ATTR_RO(carrier_changes);
 352
 353 static ssize_t carrier_up_count_show(struct device *dev,
 354                                      struct device_attribute *attr,
 355                                      char *buf)
 356 {
 357         struct net_device *netdev = to_net_dev(dev);
 358
 359         return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
 360 }
 361 static DEVICE_ATTR_RO(carrier_up_count);
 362
 363 static ssize_t carrier_down_count_show(struct device *dev,
 364                                        struct device_attribute *attr,
 365                                        char *buf)
 366 {
 367         struct net_device *netdev = to_net_dev(dev);
 368
 369         return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
 370 }
 371 static DEVICE_ATTR_RO(carrier_down_count);
 372
 373 /* read-write attributes */
 374
 375 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 376 {
 377         return dev_set_mtu(dev, (int)new_mtu);
 378 }
 379
 380 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 381                          const char *buf, size_t len)
 382 {
 383         return netdev_store(dev, attr, buf, len, change_mtu);
 384 }
 385 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 386
 387 static int change_flags(struct net_device *dev, unsigned long new_flags)
 388 {
 389         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
 390 }
 391
 392 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 393                            const char *buf, size_t len)
 394 {
 395         return netdev_store(dev, attr, buf, len, change_flags);
 396 }
 397 NETDEVICE_SHOW_RW(flags, fmt_hex);
 398
 399 static ssize_t tx_queue_len_store(struct device *dev,
 400                                   struct device_attribute *attr,
 401                                   const char *buf, size_t len)
 402 {
 403         if (!capable(CAP_NET_ADMIN))
 404                 return -EPERM;
 405
 406         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
 407 }
 408 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 409
 410 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 411 {
 412         netdev_set_gro_flush_timeout(dev, val);
 413         return 0;
 414 }
 415
 416 static ssize_t gro_flush_timeout_store(struct device *dev,
 417                                        struct device_attribute *attr,
 418                                        const char *buf, size_t len)
 419 {
 420         if (!capable(CAP_NET_ADMIN))
 421                 return -EPERM;
 422
 423         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 424 }
 425 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 426
 427 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
 428 {
 429         if (val > S32_MAX)
 430                 return -ERANGE;
 431
 432         netdev_set_defer_hard_irqs(dev, (u32)val);
 433         return 0;
 434 }
 435
 436 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
 437                                           struct device_attribute *attr,
 438                                           const char *buf, size_t len)
 439 {
 440         if (!capable(CAP_NET_ADMIN))
 441                 return -EPERM;
 442
 443         return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
 444 }
 445 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_uint);
 446
 447 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 448                              const char *buf, size_t len)
 449 {
 450         struct net_device *netdev = to_net_dev(dev);
 451         struct net *net = dev_net(netdev);
 452         size_t count = len;
 453         ssize_t ret = 0;
 454
 455         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 456                 return -EPERM;
 457
 458         /* ignore trailing newline */
 459         if (len >  0 && buf[len - 1] == '\n')
 460                 --count;
 461
 462         if (!rtnl_trylock())
 463                 return restart_syscall();
 464
 465         if (dev_isalive(netdev)) {
 466                 ret = dev_set_alias(netdev, buf, count);
 467                 if (ret < 0)
 468                         goto err;
 469                 ret = len;
 470                 netdev_state_change(netdev);
 471         }
 472 err:
 473         rtnl_unlock();
 474
 475         return ret;
 476 }
 477
 478 static ssize_t ifalias_show(struct device *dev,
 479                             struct device_attribute *attr, char *buf)
 480 {
 481         const struct net_device *netdev = to_net_dev(dev);
 482         char tmp[IFALIASZ];
 483         ssize_t ret = 0;
 484
 485         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
 486         if (ret > 0)
 487                 ret = sysfs_emit(buf, "%s\n", tmp);
 488         return ret;
 489 }
 490 static DEVICE_ATTR_RW(ifalias);
 491
 492 static int change_group(struct net_device *dev, unsigned long new_group)
 493 {
 494         dev_set_group(dev, (int)new_group);
 495         return 0;
 496 }
 497
 498 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 499                            const char *buf, size_t len)
 500 {
 501         return netdev_store(dev, attr, buf, len, change_group);
 502 }
 503 NETDEVICE_SHOW(group, fmt_dec);
 504 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
 505
 506 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 507 {
 508         return dev_change_proto_down(dev, (bool)proto_down);
 509 }
 510
 511 static ssize_t proto_down_store(struct device *dev,
 512                                 struct device_attribute *attr,
 513                                 const char *buf, size_t len)
 514 {
 515         return netdev_store(dev, attr, buf, len, change_proto_down);
 516 }
 517 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 518
 519 static ssize_t phys_port_id_show(struct device *dev,
 520                                  struct device_attribute *attr, char *buf)
 521 {
 522         struct net_device *netdev = to_net_dev(dev);
 523         ssize_t ret = -EINVAL;
 524
 525         /* The check is also done in dev_get_phys_port_id; this helps returning
 526          * early without hitting the trylock/restart below.
 527          */
 528         if (!netdev->netdev_ops->ndo_get_phys_port_id)
 529                 return -EOPNOTSUPP;
 530
 531         if (!rtnl_trylock())
 532                 return restart_syscall();
 533
 534         if (dev_isalive(netdev)) {
 535                 struct netdev_phys_item_id ppid;
 536
 537                 ret = dev_get_phys_port_id(netdev, &ppid);
 538                 if (!ret)
 539                         ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 540         }
 541         rtnl_unlock();
 542
 543         return ret;
 544 }
 545 static DEVICE_ATTR_RO(phys_port_id);
 546
 547 static ssize_t phys_port_name_show(struct device *dev,
 548                                    struct device_attribute *attr, char *buf)
 549 {
 550         struct net_device *netdev = to_net_dev(dev);
 551         ssize_t ret = -EINVAL;
 552
 553         /* The checks are also done in dev_get_phys_port_name; this helps
 554          * returning early without hitting the trylock/restart below.
 555          */
 556         if (!netdev->netdev_ops->ndo_get_phys_port_name &&
 557             !netdev->devlink_port)
 558                 return -EOPNOTSUPP;
 559
 560         if (!rtnl_trylock())
 561                 return restart_syscall();
 562
 563         if (dev_isalive(netdev)) {
 564                 char name[IFNAMSIZ];
 565
 566                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 567                 if (!ret)
 568                         ret = sysfs_emit(buf, "%s\n", name);
 569         }
 570         rtnl_unlock();
 571
 572         return ret;
 573 }
 574 static DEVICE_ATTR_RO(phys_port_name);
 575
 576 static ssize_t phys_switch_id_show(struct device *dev,
 577                                    struct device_attribute *attr, char *buf)
 578 {
 579         struct net_device *netdev = to_net_dev(dev);
 580         ssize_t ret = -EINVAL;
 581
 582         /* The checks are also done in dev_get_phys_port_name; this helps
 583          * returning early without hitting the trylock/restart below. This works
 584          * because recurse is false when calling dev_get_port_parent_id.
 585          */
 586         if (!netdev->netdev_ops->ndo_get_port_parent_id &&
 587             !netdev->devlink_port)
 588                 return -EOPNOTSUPP;
 589
 590         if (!rtnl_trylock())
 591                 return restart_syscall();
 592
 593         if (dev_isalive(netdev)) {
 594                 struct netdev_phys_item_id ppid = { };
 595
 596                 ret = dev_get_port_parent_id(netdev, &ppid, false);
 597                 if (!ret)
 598                         ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 599         }
 600         rtnl_unlock();
 601
 602         return ret;
 603 }
 604 static DEVICE_ATTR_RO(phys_switch_id);
 605
 606 static ssize_t threaded_show(struct device *dev,
 607                              struct device_attribute *attr, char *buf)
 608 {
 609         struct net_device *netdev = to_net_dev(dev);
 610         ssize_t ret = -EINVAL;
 611
 612         rcu_read_lock();
 613
 614         if (dev_isalive(netdev))
 615                 ret = sysfs_emit(buf, fmt_dec, READ_ONCE(netdev->threaded));
 616
 617         rcu_read_unlock();
 618
 619         return ret;
 620 }
 621
 622 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
 623 {
 624         int ret;
 625
 626         if (list_empty(&dev->napi_list))
 627                 return -EOPNOTSUPP;
 628
 629         if (val != 0 && val != 1)
 630                 return -EOPNOTSUPP;
 631
 632         ret = dev_set_threaded(dev, val);
 633
 634         return ret;
 635 }
 636
 637 static ssize_t threaded_store(struct device *dev,
 638                               struct device_attribute *attr,
 639                               const char *buf, size_t len)
 640 {
 641         return netdev_store(dev, attr, buf, len, modify_napi_threaded);
 642 }
 643 static DEVICE_ATTR_RW(threaded);
 644
 645 static struct attribute *net_class_attrs[] __ro_after_init = {
 646         &dev_attr_netdev_group.attr,
 647         &dev_attr_type.attr,
 648         &dev_attr_dev_id.attr,
 649         &dev_attr_dev_port.attr,
 650         &dev_attr_iflink.attr,
 651         &dev_attr_ifindex.attr,
 652         &dev_attr_name_assign_type.attr,
 653         &dev_attr_addr_assign_type.attr,
 654         &dev_attr_addr_len.attr,
 655         &dev_attr_link_mode.attr,
 656         &dev_attr_address.attr,
 657         &dev_attr_broadcast.attr,
 658         &dev_attr_speed.attr,
 659         &dev_attr_duplex.attr,
 660         &dev_attr_dormant.attr,
 661         &dev_attr_testing.attr,
 662         &dev_attr_operstate.attr,
 663         &dev_attr_carrier_changes.attr,
 664         &dev_attr_ifalias.attr,
 665         &dev_attr_carrier.attr,
 666         &dev_attr_mtu.attr,
 667         &dev_attr_flags.attr,
 668         &dev_attr_tx_queue_len.attr,
 669         &dev_attr_gro_flush_timeout.attr,
 670         &dev_attr_napi_defer_hard_irqs.attr,
 671         &dev_attr_phys_port_id.attr,
 672         &dev_attr_phys_port_name.attr,
 673         &dev_attr_phys_switch_id.attr,
 674         &dev_attr_proto_down.attr,
 675         &dev_attr_carrier_up_count.attr,
 676         &dev_attr_carrier_down_count.attr,
 677         &dev_attr_threaded.attr,
 678         NULL,
 679 };
 680 ATTRIBUTE_GROUPS(net_class);
 681
 682 /* Show a given an attribute in the statistics group */
 683 static ssize_t netstat_show(const struct device *d,
 684                             struct device_attribute *attr, char *buf,
 685                             unsigned long offset)
 686 {
 687         struct net_device *dev = to_net_dev(d);
 688         ssize_t ret = -EINVAL;
 689
 690         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 691                 offset % sizeof(u64) != 0);
 692
 693         rcu_read_lock();
 694         if (dev_isalive(dev)) {
 695                 struct rtnl_link_stats64 temp;
 696                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 697
 698                 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
 699         }
 700         rcu_read_unlock();
 701         return ret;
 702 }
 703
 704 /* generate a read-only statistics attribute */
 705 #define NETSTAT_ENTRY(name)                                             \
 706 static ssize_t name##_show(struct device *d,                            \
 707                            struct device_attribute *attr, char *buf)    \
 708 {                                                                       \
 709         return netstat_show(d, attr, buf,                               \
 710                             offsetof(struct rtnl_link_stats64, name));  \
 711 }                                                                       \
 712 static DEVICE_ATTR_RO(name)
 713
 714 NETSTAT_ENTRY(rx_packets);
 715 NETSTAT_ENTRY(tx_packets);
 716 NETSTAT_ENTRY(rx_bytes);
 717 NETSTAT_ENTRY(tx_bytes);
 718 NETSTAT_ENTRY(rx_errors);
 719 NETSTAT_ENTRY(tx_errors);
 720 NETSTAT_ENTRY(rx_dropped);
 721 NETSTAT_ENTRY(tx_dropped);
 722 NETSTAT_ENTRY(multicast);
 723 NETSTAT_ENTRY(collisions);
 724 NETSTAT_ENTRY(rx_length_errors);
 725 NETSTAT_ENTRY(rx_over_errors);
 726 NETSTAT_ENTRY(rx_crc_errors);
 727 NETSTAT_ENTRY(rx_frame_errors);
 728 NETSTAT_ENTRY(rx_fifo_errors);
 729 NETSTAT_ENTRY(rx_missed_errors);
 730 NETSTAT_ENTRY(tx_aborted_errors);
 731 NETSTAT_ENTRY(tx_carrier_errors);
 732 NETSTAT_ENTRY(tx_fifo_errors);
 733 NETSTAT_ENTRY(tx_heartbeat_errors);
 734 NETSTAT_ENTRY(tx_window_errors);
 735 NETSTAT_ENTRY(rx_compressed);
 736 NETSTAT_ENTRY(tx_compressed);
 737 NETSTAT_ENTRY(rx_nohandler);
 738
 739 static struct attribute *netstat_attrs[] __ro_after_init = {
 740         &dev_attr_rx_packets.attr,
 741         &dev_attr_tx_packets.attr,
 742         &dev_attr_rx_bytes.attr,
 743         &dev_attr_tx_bytes.attr,
 744         &dev_attr_rx_errors.attr,
 745         &dev_attr_tx_errors.attr,
 746         &dev_attr_rx_dropped.attr,
 747         &dev_attr_tx_dropped.attr,
 748         &dev_attr_multicast.attr,
 749         &dev_attr_collisions.attr,
 750         &dev_attr_rx_length_errors.attr,
 751         &dev_attr_rx_over_errors.attr,
 752         &dev_attr_rx_crc_errors.attr,
 753         &dev_attr_rx_frame_errors.attr,
 754         &dev_attr_rx_fifo_errors.attr,
 755         &dev_attr_rx_missed_errors.attr,
 756         &dev_attr_tx_aborted_errors.attr,
 757         &dev_attr_tx_carrier_errors.attr,
 758         &dev_attr_tx_fifo_errors.attr,
 759         &dev_attr_tx_heartbeat_errors.attr,
 760         &dev_attr_tx_window_errors.attr,
 761         &dev_attr_rx_compressed.attr,
 762         &dev_attr_tx_compressed.attr,
 763         &dev_attr_rx_nohandler.attr,
 764         NULL
 765 };
 766
 767 static const struct attribute_group netstat_group = {
 768         .name  = "statistics",
 769         .attrs  = netstat_attrs,
 770 };
 771
 772 static struct attribute *wireless_attrs[] = {
 773         NULL
 774 };
 775
 776 static const struct attribute_group wireless_group = {
 777         .name = "wireless",
 778         .attrs = wireless_attrs,
 779 };
 780
 781 static bool wireless_group_needed(struct net_device *ndev)
 782 {
 783 #if IS_ENABLED(CONFIG_CFG80211)
 784         if (ndev->ieee80211_ptr)
 785                 return true;
 786 #endif
 787 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
 788         if (ndev->wireless_handlers)
 789                 return true;
 790 #endif
 791         return false;
 792 }
 793
 794 #else /* CONFIG_SYSFS */
 795 #define net_class_groups        NULL
 796 #endif /* CONFIG_SYSFS */
 797
 798 #ifdef CONFIG_SYSFS
 799 #define to_rx_queue_attr(_attr) \
 800         container_of(_attr, struct rx_queue_attribute, attr)
 801
 802 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 803
 804 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 805                                   char *buf)
 806 {
 807         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 808         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 809
 810         if (!attribute->show)
 811                 return -EIO;
 812
 813         return attribute->show(queue, buf);
 814 }
 815
 816 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 817                                    const char *buf, size_t count)
 818 {
 819         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 820         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 821
 822         if (!attribute->store)
 823                 return -EIO;
 824
 825         return attribute->store(queue, buf, count);
 826 }
 827
 828 static const struct sysfs_ops rx_queue_sysfs_ops = {
 829         .show = rx_queue_attr_show,
 830         .store = rx_queue_attr_store,
 831 };
 832
 833 #ifdef CONFIG_RPS
 834 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
 835 {
 836         struct rps_map *map;
 837         cpumask_var_t mask;
 838         int i, len;
 839
 840         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 841                 return -ENOMEM;
 842
 843         rcu_read_lock();
 844         map = rcu_dereference(queue->rps_map);
 845         if (map)
 846                 for (i = 0; i < map->len; i++)
 847                         cpumask_set_cpu(map->cpus[i], mask);
 848
 849         len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
 850         rcu_read_unlock();
 851         free_cpumask_var(mask);
 852
 853         return len < PAGE_SIZE ? len : -EINVAL;
 854 }
 855
 856 static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
 857                                         cpumask_var_t mask)
 858 {
 859         static DEFINE_MUTEX(rps_map_mutex);
 860         struct rps_map *old_map, *map;
 861         int cpu, i;
 862
 863         map = kzalloc(max_t(unsigned int,
 864                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 865                       GFP_KERNEL);
 866         if (!map)
 867                 return -ENOMEM;
 868
 869         i = 0;
 870         for_each_cpu_and(cpu, mask, cpu_online_mask)
 871                 map->cpus[i++] = cpu;
 872
 873         if (i) {
 874                 map->len = i;
 875         } else {
 876                 kfree(map);
 877                 map = NULL;
 878         }
 879
 880         mutex_lock(&rps_map_mutex);
 881         old_map = rcu_dereference_protected(queue->rps_map,
 882                                             mutex_is_locked(&rps_map_mutex));
 883         rcu_assign_pointer(queue->rps_map, map);
 884
 885         if (map)
 886                 static_branch_inc(&rps_needed);
 887         if (old_map)
 888                 static_branch_dec(&rps_needed);
 889
 890         mutex_unlock(&rps_map_mutex);
 891
 892         if (old_map)
 893                 kfree_rcu(old_map, rcu);
 894         return 0;
 895 }
 896
 897 int rps_cpumask_housekeeping(struct cpumask *mask)
 898 {
 899         if (!cpumask_empty(mask)) {
 900                 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
 901                 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
 902                 if (cpumask_empty(mask))
 903                         return -EINVAL;
 904         }
 905         return 0;
 906 }
 907
 908 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 909                              const char *buf, size_t len)
 910 {
 911         cpumask_var_t mask;
 912         int err;
 913
 914         if (!capable(CAP_NET_ADMIN))
 915                 return -EPERM;
 916
 917         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 918                 return -ENOMEM;
 919
 920         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 921         if (err)
 922                 goto out;
 923
 924         err = rps_cpumask_housekeeping(mask);
 925         if (err)
 926                 goto out;
 927
 928         err = netdev_rx_queue_set_rps_mask(queue, mask);
 929
 930 out:
 931         free_cpumask_var(mask);
 932         return err ? : len;
 933 }
 934
 935 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 936                                            char *buf)
 937 {
 938         struct rps_dev_flow_table *flow_table;
 939         unsigned long val = 0;
 940
 941         rcu_read_lock();
 942         flow_table = rcu_dereference(queue->rps_flow_table);
 943         if (flow_table)
 944                 val = (unsigned long)flow_table->mask + 1;
 945         rcu_read_unlock();
 946
 947         return sysfs_emit(buf, "%lu\n", val);
 948 }
 949
 950 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 951 {
 952         struct rps_dev_flow_table *table = container_of(rcu,
 953             struct rps_dev_flow_table, rcu);
 954         vfree(table);
 955 }
 956
 957 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 958                                             const char *buf, size_t len)
 959 {
 960         unsigned long mask, count;
 961         struct rps_dev_flow_table *table, *old_table;
 962         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 963         int rc;
 964
 965         if (!capable(CAP_NET_ADMIN))
 966                 return -EPERM;
 967
 968         rc = kstrtoul(buf, 0, &count);
 969         if (rc < 0)
 970                 return rc;
 971
 972         if (count) {
 973                 mask = count - 1;
 974                 /* mask = roundup_pow_of_two(count) - 1;
 975                  * without overflows...
 976                  */
 977                 while ((mask | (mask >> 1)) != mask)
 978                         mask |= (mask >> 1);
 979                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 980                  * and on 32bit arches, must check
 981                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 982                  */
 983 #if BITS_PER_LONG > 32
 984                 if (mask > (unsigned long)(u32)mask)
 985                         return -EINVAL;
 986 #else
 987                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 988                                 / sizeof(struct rps_dev_flow)) {
 989                         /* Enforce a limit to prevent overflow */
 990                         return -EINVAL;
 991                 }
 992 #endif
 993                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 994                 if (!table)
 995                         return -ENOMEM;
 996
 997                 table->mask = mask;
 998                 for (count = 0; count <= mask; count++)
 999                         table->flows[count].cpu = RPS_NO_CPU;
1000         } else {
1001                 table = NULL;
1002         }
1003
1004         spin_lock(&rps_dev_flow_lock);
1005         old_table = rcu_dereference_protected(queue->rps_flow_table,
1006                                               lockdep_is_held(&rps_dev_flow_lock));
1007         rcu_assign_pointer(queue->rps_flow_table, table);
1008         spin_unlock(&rps_dev_flow_lock);
1009
1010         if (old_table)
1011                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1012
1013         return len;
1014 }
1015
1016 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1017         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1018
1019 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1020         = __ATTR(rps_flow_cnt, 0644,
1021                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1022 #endif /* CONFIG_RPS */
1023
1024 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1025 #ifdef CONFIG_RPS
1026         &rps_cpus_attribute.attr,
1027         &rps_dev_flow_table_cnt_attribute.attr,
1028 #endif
1029         NULL
1030 };
1031 ATTRIBUTE_GROUPS(rx_queue_default);
1032
1033 static void rx_queue_release(struct kobject *kobj)
1034 {
1035         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1036 #ifdef CONFIG_RPS
1037         struct rps_map *map;
1038         struct rps_dev_flow_table *flow_table;
1039
1040         map = rcu_dereference_protected(queue->rps_map, 1);
1041         if (map) {
1042                 RCU_INIT_POINTER(queue->rps_map, NULL);
1043                 kfree_rcu(map, rcu);
1044         }
1045
1046         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1047         if (flow_table) {
1048                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1049                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1050         }
1051 #endif
1052
1053         memset(kobj, 0, sizeof(*kobj));
1054         netdev_put(queue->dev, &queue->dev_tracker);
1055 }
1056
1057 static const void *rx_queue_namespace(const struct kobject *kobj)
1058 {
1059         struct netdev_rx_queue *queue = to_rx_queue(kobj);
1060         struct device *dev = &queue->dev->dev;
1061         const void *ns = NULL;
1062
1063         if (dev->class && dev->class->namespace)
1064                 ns = dev->class->namespace(dev);
1065
1066         return ns;
1067 }
1068
1069 static void rx_queue_get_ownership(const struct kobject *kobj,
1070                                    kuid_t *uid, kgid_t *gid)
1071 {
1072         const struct net *net = rx_queue_namespace(kobj);
1073
1074         net_ns_get_ownership(net, uid, gid);
1075 }
1076
1077 static const struct kobj_type rx_queue_ktype = {
1078         .sysfs_ops = &rx_queue_sysfs_ops,
1079         .release = rx_queue_release,
1080         .default_groups = rx_queue_default_groups,
1081         .namespace = rx_queue_namespace,
1082         .get_ownership = rx_queue_get_ownership,
1083 };
1084
1085 static int rx_queue_default_mask(struct net_device *dev,
1086                                  struct netdev_rx_queue *queue)
1087 {
1088 #if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1089         struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1090
1091         if (rps_default_mask && !cpumask_empty(rps_default_mask))
1092                 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1093 #endif
1094         return 0;
1095 }
1096
1097 static int rx_queue_add_kobject(struct net_device *dev, int index)
1098 {
1099         struct netdev_rx_queue *queue = dev->_rx + index;
1100         struct kobject *kobj = &queue->kobj;
1101         int error = 0;
1102
1103         /* Kobject_put later will trigger rx_queue_release call which
1104          * decreases dev refcount: Take that reference here
1105          */
1106         netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1107
1108         kobj->kset = dev->queues_kset;
1109         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1110                                      "rx-%u", index);
1111         if (error)
1112                 goto err;
1113
1114         if (dev->sysfs_rx_queue_group) {
1115                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1116                 if (error)
1117                         goto err;
1118         }
1119
1120         error = rx_queue_default_mask(dev, queue);
1121         if (error)
1122                 goto err;
1123
1124         kobject_uevent(kobj, KOBJ_ADD);
1125
1126         return error;
1127
1128 err:
1129         kobject_put(kobj);
1130         return error;
1131 }
1132
1133 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1134                                  kgid_t kgid)
1135 {
1136         struct netdev_rx_queue *queue = dev->_rx + index;
1137         struct kobject *kobj = &queue->kobj;
1138         int error;
1139
1140         error = sysfs_change_owner(kobj, kuid, kgid);
1141         if (error)
1142                 return error;
1143
1144         if (dev->sysfs_rx_queue_group)
1145                 error = sysfs_group_change_owner(
1146                         kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1147
1148         return error;
1149 }
1150 #endif /* CONFIG_SYSFS */
1151
1152 int
1153 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1154 {
1155 #ifdef CONFIG_SYSFS
1156         int i;
1157         int error = 0;
1158
1159 #ifndef CONFIG_RPS
1160         if (!dev->sysfs_rx_queue_group)
1161                 return 0;
1162 #endif
1163         for (i = old_num; i < new_num; i++) {
1164                 error = rx_queue_add_kobject(dev, i);
1165                 if (error) {
1166                         new_num = old_num;
1167                         break;
1168                 }
1169         }
1170
1171         while (--i >= new_num) {
1172                 struct kobject *kobj = &dev->_rx[i].kobj;
1173
1174                 if (!refcount_read(&dev_net(dev)->ns.count))
1175                         kobj->uevent_suppress = 1;
1176                 if (dev->sysfs_rx_queue_group)
1177                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1178                 kobject_put(kobj);
1179         }
1180
1181         return error;
1182 #else
1183         return 0;
1184 #endif
1185 }
1186
1187 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1188                                      kuid_t kuid, kgid_t kgid)
1189 {
1190 #ifdef CONFIG_SYSFS
1191         int error = 0;
1192         int i;
1193
1194 #ifndef CONFIG_RPS
1195         if (!dev->sysfs_rx_queue_group)
1196                 return 0;
1197 #endif
1198         for (i = 0; i < num; i++) {
1199                 error = rx_queue_change_owner(dev, i, kuid, kgid);
1200                 if (error)
1201                         break;
1202         }
1203
1204         return error;
1205 #else
1206         return 0;
1207 #endif
1208 }
1209
1210 #ifdef CONFIG_SYSFS
1211 /*
1212  * netdev_queue sysfs structures and functions.
1213  */
1214 struct netdev_queue_attribute {
1215         struct attribute attr;
1216         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1217         ssize_t (*store)(struct netdev_queue *queue,
1218                          const char *buf, size_t len);
1219 };
1220 #define to_netdev_queue_attr(_attr) \
1221         container_of(_attr, struct netdev_queue_attribute, attr)
1222
1223 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1224
1225 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1226                                       struct attribute *attr, char *buf)
1227 {
1228         const struct netdev_queue_attribute *attribute
1229                 = to_netdev_queue_attr(attr);
1230         struct netdev_queue *queue = to_netdev_queue(kobj);
1231
1232         if (!attribute->show)
1233                 return -EIO;
1234
1235         return attribute->show(queue, buf);
1236 }
1237
1238 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1239                                        struct attribute *attr,
1240                                        const char *buf, size_t count)
1241 {
1242         const struct netdev_queue_attribute *attribute
1243                 = to_netdev_queue_attr(attr);
1244         struct netdev_queue *queue = to_netdev_queue(kobj);
1245
1246         if (!attribute->store)
1247                 return -EIO;
1248
1249         return attribute->store(queue, buf, count);
1250 }
1251
1252 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1253         .show = netdev_queue_attr_show,
1254         .store = netdev_queue_attr_store,
1255 };
1256
1257 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1258 {
1259         unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1260
1261         return sysfs_emit(buf, fmt_ulong, trans_timeout);
1262 }
1263
1264 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1265 {
1266         struct net_device *dev = queue->dev;
1267         unsigned int i;
1268
1269         i = queue - dev->_tx;
1270         BUG_ON(i >= dev->num_tx_queues);
1271
1272         return i;
1273 }
1274
1275 static ssize_t traffic_class_show(struct netdev_queue *queue,
1276                                   char *buf)
1277 {
1278         struct net_device *dev = queue->dev;
1279         int num_tc, tc;
1280         int index;
1281
1282         if (!netif_is_multiqueue(dev))
1283                 return -ENOENT;
1284
1285         if (!rtnl_trylock())
1286                 return restart_syscall();
1287
1288         index = get_netdev_queue_index(queue);
1289
1290         /* If queue belongs to subordinate dev use its TC mapping */
1291         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1292
1293         num_tc = dev->num_tc;
1294         tc = netdev_txq_to_tc(dev, index);
1295
1296         rtnl_unlock();
1297
1298         if (tc < 0)
1299                 return -EINVAL;
1300
1301         /* We can report the traffic class one of two ways:
1302          * Subordinate device traffic classes are reported with the traffic
1303          * class first, and then the subordinate class so for example TC0 on
1304          * subordinate device 2 will be reported as "0-2". If the queue
1305          * belongs to the root device it will be reported with just the
1306          * traffic class, so just "0" for TC 0 for example.
1307          */
1308         return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1309                             sysfs_emit(buf, "%d\n", tc);
1310 }
1311
1312 #ifdef CONFIG_XPS
1313 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1314                                char *buf)
1315 {
1316         return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1317 }
1318
1319 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1320                                 const char *buf, size_t len)
1321 {
1322         struct net_device *dev = queue->dev;
1323         int err, index = get_netdev_queue_index(queue);
1324         u32 rate = 0;
1325
1326         if (!capable(CAP_NET_ADMIN))
1327                 return -EPERM;
1328
1329         /* The check is also done later; this helps returning early without
1330          * hitting the trylock/restart below.
1331          */
1332         if (!dev->netdev_ops->ndo_set_tx_maxrate)
1333                 return -EOPNOTSUPP;
1334
1335         err = kstrtou32(buf, 10, &rate);
1336         if (err < 0)
1337                 return err;
1338
1339         if (!rtnl_trylock())
1340                 return restart_syscall();
1341
1342         err = -EOPNOTSUPP;
1343         if (dev->netdev_ops->ndo_set_tx_maxrate)
1344                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1345
1346         rtnl_unlock();
1347         if (!err) {
1348                 queue->tx_maxrate = rate;
1349                 return len;
1350         }
1351         return err;
1352 }
1353
1354 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1355         = __ATTR_RW(tx_maxrate);
1356 #endif
1357
1358 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1359         = __ATTR_RO(tx_timeout);
1360
1361 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1362         = __ATTR_RO(traffic_class);
1363
1364 #ifdef CONFIG_BQL
1365 /*
1366  * Byte queue limits sysfs structures and functions.
1367  */
1368 static ssize_t bql_show(char *buf, unsigned int value)
1369 {
1370         return sysfs_emit(buf, "%u\n", value);
1371 }
1372
1373 static ssize_t bql_set(const char *buf, const size_t count,
1374                        unsigned int *pvalue)
1375 {
1376         unsigned int value;
1377         int err;
1378
1379         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1380                 value = DQL_MAX_LIMIT;
1381         } else {
1382                 err = kstrtouint(buf, 10, &value);
1383                 if (err < 0)
1384                         return err;
1385                 if (value > DQL_MAX_LIMIT)
1386                         return -EINVAL;
1387         }
1388
1389         *pvalue = value;
1390
1391         return count;
1392 }
1393
1394 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1395                                   char *buf)
1396 {
1397         struct dql *dql = &queue->dql;
1398
1399         return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1400 }
1401
1402 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1403                                  const char *buf, size_t len)
1404 {
1405         struct dql *dql = &queue->dql;
1406         unsigned int value;
1407         int err;
1408
1409         err = kstrtouint(buf, 10, &value);
1410         if (err < 0)
1411                 return err;
1412
1413         dql->slack_hold_time = msecs_to_jiffies(value);
1414
1415         return len;
1416 }
1417
1418 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1419         = __ATTR(hold_time, 0644,
1420                  bql_show_hold_time, bql_set_hold_time);
1421
1422 static ssize_t bql_show_stall_thrs(struct netdev_queue *queue, char *buf)
1423 {
1424         struct dql *dql = &queue->dql;
1425
1426         return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->stall_thrs));
1427 }
1428
1429 static ssize_t bql_set_stall_thrs(struct netdev_queue *queue,
1430                                   const char *buf, size_t len)
1431 {
1432         struct dql *dql = &queue->dql;
1433         unsigned int value;
1434         int err;
1435
1436         err = kstrtouint(buf, 10, &value);
1437         if (err < 0)
1438                 return err;
1439
1440         value = msecs_to_jiffies(value);
1441         if (value && (value < 4 || value > 4 / 2 * BITS_PER_LONG))
1442                 return -ERANGE;
1443
1444         if (!dql->stall_thrs && value)
1445                 dql->last_reap = jiffies;
1446         /* Force last_reap to be live */
1447         smp_wmb();
1448         dql->stall_thrs = value;
1449
1450         return len;
1451 }
1452
1453 static struct netdev_queue_attribute bql_stall_thrs_attribute __ro_after_init =
1454         __ATTR(stall_thrs, 0644, bql_show_stall_thrs, bql_set_stall_thrs);
1455
1456 static ssize_t bql_show_stall_max(struct netdev_queue *queue, char *buf)
1457 {
1458         return sysfs_emit(buf, "%u\n", READ_ONCE(queue->dql.stall_max));
1459 }
1460
1461 static ssize_t bql_set_stall_max(struct netdev_queue *queue,
1462                                  const char *buf, size_t len)
1463 {
1464         WRITE_ONCE(queue->dql.stall_max, 0);
1465         return len;
1466 }
1467
1468 static struct netdev_queue_attribute bql_stall_max_attribute __ro_after_init =
1469         __ATTR(stall_max, 0644, bql_show_stall_max, bql_set_stall_max);
1470
1471 static ssize_t bql_show_stall_cnt(struct netdev_queue *queue, char *buf)
1472 {
1473         struct dql *dql = &queue->dql;
1474
1475         return sysfs_emit(buf, "%lu\n", dql->stall_cnt);
1476 }
1477
1478 static struct netdev_queue_attribute bql_stall_cnt_attribute __ro_after_init =
1479         __ATTR(stall_cnt, 0444, bql_show_stall_cnt, NULL);
1480
1481 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1482                                  char *buf)
1483 {
1484         struct dql *dql = &queue->dql;
1485
1486         return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1487 }
1488
1489 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1490         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1491
1492 #define BQL_ATTR(NAME, FIELD)                                           \
1493 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1494                                  char *buf)                             \
1495 {                                                                       \
1496         return bql_show(buf, queue->dql.FIELD);                         \
1497 }                                                                       \
1498                                                                         \
1499 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1500                                 const char *buf, size_t len)            \
1501 {                                                                       \
1502         return bql_set(buf, len, &queue->dql.FIELD);                    \
1503 }                                                                       \
1504                                                                         \
1505 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1506         = __ATTR(NAME, 0644,                            \
1507                  bql_show_ ## NAME, bql_set_ ## NAME)
1508
1509 BQL_ATTR(limit, limit);
1510 BQL_ATTR(limit_max, max_limit);
1511 BQL_ATTR(limit_min, min_limit);
1512
1513 static struct attribute *dql_attrs[] __ro_after_init = {
1514         &bql_limit_attribute.attr,
1515         &bql_limit_max_attribute.attr,
1516         &bql_limit_min_attribute.attr,
1517         &bql_hold_time_attribute.attr,
1518         &bql_inflight_attribute.attr,
1519         &bql_stall_thrs_attribute.attr,
1520         &bql_stall_cnt_attribute.attr,
1521         &bql_stall_max_attribute.attr,
1522         NULL
1523 };
1524
1525 static const struct attribute_group dql_group = {
1526         .name  = "byte_queue_limits",
1527         .attrs  = dql_attrs,
1528 };
1529 #else
1530 /* Fake declaration, all the code using it should be dead */
1531 static const struct attribute_group dql_group = {};
1532 #endif /* CONFIG_BQL */
1533
1534 #ifdef CONFIG_XPS
1535 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1536                               int tc, char *buf, enum xps_map_type type)
1537 {
1538         struct xps_dev_maps *dev_maps;
1539         unsigned long *mask;
1540         unsigned int nr_ids;
1541         int j, len;
1542
1543         rcu_read_lock();
1544         dev_maps = rcu_dereference(dev->xps_maps[type]);
1545
1546         /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1547          * when dev_maps hasn't been allocated yet, to be backward compatible.
1548          */
1549         nr_ids = dev_maps ? dev_maps->nr_ids :
1550                  (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1551
1552         mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1553         if (!mask) {
1554                 rcu_read_unlock();
1555                 return -ENOMEM;
1556         }
1557
1558         if (!dev_maps || tc >= dev_maps->num_tc)
1559                 goto out_no_maps;
1560
1561         for (j = 0; j < nr_ids; j++) {
1562                 int i, tci = j * dev_maps->num_tc + tc;
1563                 struct xps_map *map;
1564
1565                 map = rcu_dereference(dev_maps->attr_map[tci]);
1566                 if (!map)
1567                         continue;
1568
1569                 for (i = map->len; i--;) {
1570                         if (map->queues[i] == index) {
1571                                 __set_bit(j, mask);
1572                                 break;
1573                         }
1574                 }
1575         }
1576 out_no_maps:
1577         rcu_read_unlock();
1578
1579         len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1580         bitmap_free(mask);
1581
1582         return len < PAGE_SIZE ? len : -EINVAL;
1583 }
1584
1585 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1586 {
1587         struct net_device *dev = queue->dev;
1588         unsigned int index;
1589         int len, tc;
1590
1591         if (!netif_is_multiqueue(dev))
1592                 return -ENOENT;
1593
1594         index = get_netdev_queue_index(queue);
1595
1596         if (!rtnl_trylock())
1597                 return restart_syscall();
1598
1599         /* If queue belongs to subordinate dev use its map */
1600         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1601
1602         tc = netdev_txq_to_tc(dev, index);
1603         if (tc < 0) {
1604                 rtnl_unlock();
1605                 return -EINVAL;
1606         }
1607
1608         /* Make sure the subordinate device can't be freed */
1609         get_device(&dev->dev);
1610         rtnl_unlock();
1611
1612         len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1613
1614         put_device(&dev->dev);
1615         return len;
1616 }
1617
1618 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1619                               const char *buf, size_t len)
1620 {
1621         struct net_device *dev = queue->dev;
1622         unsigned int index;
1623         cpumask_var_t mask;
1624         int err;
1625
1626         if (!netif_is_multiqueue(dev))
1627                 return -ENOENT;
1628
1629         if (!capable(CAP_NET_ADMIN))
1630                 return -EPERM;
1631
1632         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1633                 return -ENOMEM;
1634
1635         index = get_netdev_queue_index(queue);
1636
1637         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1638         if (err) {
1639                 free_cpumask_var(mask);
1640                 return err;
1641         }
1642
1643         if (!rtnl_trylock()) {
1644                 free_cpumask_var(mask);
1645                 return restart_syscall();
1646         }
1647
1648         err = netif_set_xps_queue(dev, mask, index);
1649         rtnl_unlock();
1650
1651         free_cpumask_var(mask);
1652
1653         return err ? : len;
1654 }
1655
1656 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1657         = __ATTR_RW(xps_cpus);
1658
1659 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1660 {
1661         struct net_device *dev = queue->dev;
1662         unsigned int index;
1663         int tc;
1664
1665         index = get_netdev_queue_index(queue);
1666
1667         if (!rtnl_trylock())
1668                 return restart_syscall();
1669
1670         tc = netdev_txq_to_tc(dev, index);
1671         rtnl_unlock();
1672         if (tc < 0)
1673                 return -EINVAL;
1674
1675         return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1676 }
1677
1678 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1679                               size_t len)
1680 {
1681         struct net_device *dev = queue->dev;
1682         struct net *net = dev_net(dev);
1683         unsigned long *mask;
1684         unsigned int index;
1685         int err;
1686
1687         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1688                 return -EPERM;
1689
1690         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1691         if (!mask)
1692                 return -ENOMEM;
1693
1694         index = get_netdev_queue_index(queue);
1695
1696         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1697         if (err) {
1698                 bitmap_free(mask);
1699                 return err;
1700         }
1701
1702         if (!rtnl_trylock()) {
1703                 bitmap_free(mask);
1704                 return restart_syscall();
1705         }
1706
1707         cpus_read_lock();
1708         err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1709         cpus_read_unlock();
1710
1711         rtnl_unlock();
1712
1713         bitmap_free(mask);
1714         return err ? : len;
1715 }
1716
1717 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1718         = __ATTR_RW(xps_rxqs);
1719 #endif /* CONFIG_XPS */
1720
1721 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1722         &queue_trans_timeout.attr,
1723         &queue_traffic_class.attr,
1724 #ifdef CONFIG_XPS
1725         &xps_cpus_attribute.attr,
1726         &xps_rxqs_attribute.attr,
1727         &queue_tx_maxrate.attr,
1728 #endif
1729         NULL
1730 };
1731 ATTRIBUTE_GROUPS(netdev_queue_default);
1732
1733 static void netdev_queue_release(struct kobject *kobj)
1734 {
1735         struct netdev_queue *queue = to_netdev_queue(kobj);
1736
1737         memset(kobj, 0, sizeof(*kobj));
1738         netdev_put(queue->dev, &queue->dev_tracker);
1739 }
1740
1741 static const void *netdev_queue_namespace(const struct kobject *kobj)
1742 {
1743         struct netdev_queue *queue = to_netdev_queue(kobj);
1744         struct device *dev = &queue->dev->dev;
1745         const void *ns = NULL;
1746
1747         if (dev->class && dev->class->namespace)
1748                 ns = dev->class->namespace(dev);
1749
1750         return ns;
1751 }
1752
1753 static void netdev_queue_get_ownership(const struct kobject *kobj,
1754                                        kuid_t *uid, kgid_t *gid)
1755 {
1756         const struct net *net = netdev_queue_namespace(kobj);
1757
1758         net_ns_get_ownership(net, uid, gid);
1759 }
1760
1761 static const struct kobj_type netdev_queue_ktype = {
1762         .sysfs_ops = &netdev_queue_sysfs_ops,
1763         .release = netdev_queue_release,
1764         .default_groups = netdev_queue_default_groups,
1765         .namespace = netdev_queue_namespace,
1766         .get_ownership = netdev_queue_get_ownership,
1767 };
1768
1769 static bool netdev_uses_bql(const struct net_device *dev)
1770 {
1771         if (dev->lltx || (dev->priv_flags & IFF_NO_QUEUE))
1772                 return false;
1773
1774         return IS_ENABLED(CONFIG_BQL);
1775 }
1776
1777 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1778 {
1779         struct netdev_queue *queue = dev->_tx + index;
1780         struct kobject *kobj = &queue->kobj;
1781         int error = 0;
1782
1783         /* Kobject_put later will trigger netdev_queue_release call
1784          * which decreases dev refcount: Take that reference here
1785          */
1786         netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1787
1788         kobj->kset = dev->queues_kset;
1789         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1790                                      "tx-%u", index);
1791         if (error)
1792                 goto err;
1793
1794         if (netdev_uses_bql(dev)) {
1795                 error = sysfs_create_group(kobj, &dql_group);
1796                 if (error)
1797                         goto err;
1798         }
1799
1800         kobject_uevent(kobj, KOBJ_ADD);
1801         return 0;
1802
1803 err:
1804         kobject_put(kobj);
1805         return error;
1806 }
1807
1808 static int tx_queue_change_owner(struct net_device *ndev, int index,
1809                                  kuid_t kuid, kgid_t kgid)
1810 {
1811         struct netdev_queue *queue = ndev->_tx + index;
1812         struct kobject *kobj = &queue->kobj;
1813         int error;
1814
1815         error = sysfs_change_owner(kobj, kuid, kgid);
1816         if (error)
1817                 return error;
1818
1819         if (netdev_uses_bql(ndev))
1820                 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1821
1822         return error;
1823 }
1824 #endif /* CONFIG_SYSFS */
1825
1826 int
1827 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1828 {
1829 #ifdef CONFIG_SYSFS
1830         int i;
1831         int error = 0;
1832
1833         /* Tx queue kobjects are allowed to be updated when a device is being
1834          * unregistered, but solely to remove queues from qdiscs. Any path
1835          * adding queues should be fixed.
1836          */
1837         WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1838              "New queues can't be registered after device unregistration.");
1839
1840         for (i = old_num; i < new_num; i++) {
1841                 error = netdev_queue_add_kobject(dev, i);
1842                 if (error) {
1843                         new_num = old_num;
1844                         break;
1845                 }
1846         }
1847
1848         while (--i >= new_num) {
1849                 struct netdev_queue *queue = dev->_tx + i;
1850
1851                 if (!refcount_read(&dev_net(dev)->ns.count))
1852                         queue->kobj.uevent_suppress = 1;
1853
1854                 if (netdev_uses_bql(dev))
1855                         sysfs_remove_group(&queue->kobj, &dql_group);
1856
1857                 kobject_put(&queue->kobj);
1858         }
1859
1860         return error;
1861 #else
1862         return 0;
1863 #endif /* CONFIG_SYSFS */
1864 }
1865
1866 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1867                                      kuid_t kuid, kgid_t kgid)
1868 {
1869 #ifdef CONFIG_SYSFS
1870         int error = 0;
1871         int i;
1872
1873         for (i = 0; i < num; i++) {
1874                 error = tx_queue_change_owner(dev, i, kuid, kgid);
1875                 if (error)
1876                         break;
1877         }
1878
1879         return error;
1880 #else
1881         return 0;
1882 #endif /* CONFIG_SYSFS */
1883 }
1884
1885 static int register_queue_kobjects(struct net_device *dev)
1886 {
1887         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1888
1889 #ifdef CONFIG_SYSFS
1890         dev->queues_kset = kset_create_and_add("queues",
1891                                                NULL, &dev->dev.kobj);
1892         if (!dev->queues_kset)
1893                 return -ENOMEM;
1894         real_rx = dev->real_num_rx_queues;
1895 #endif
1896         real_tx = dev->real_num_tx_queues;
1897
1898         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1899         if (error)
1900                 goto error;
1901         rxq = real_rx;
1902
1903         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1904         if (error)
1905                 goto error;
1906         txq = real_tx;
1907
1908         return 0;
1909
1910 error:
1911         netdev_queue_update_kobjects(dev, txq, 0);
1912         net_rx_queue_update_kobjects(dev, rxq, 0);
1913 #ifdef CONFIG_SYSFS
1914         kset_unregister(dev->queues_kset);
1915 #endif
1916         return error;
1917 }
1918
1919 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1920 {
1921         int error = 0, real_rx = 0, real_tx = 0;
1922
1923 #ifdef CONFIG_SYSFS
1924         if (ndev->queues_kset) {
1925                 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1926                 if (error)
1927                         return error;
1928         }
1929         real_rx = ndev->real_num_rx_queues;
1930 #endif
1931         real_tx = ndev->real_num_tx_queues;
1932
1933         error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1934         if (error)
1935                 return error;
1936
1937         error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1938         if (error)
1939                 return error;
1940
1941         return 0;
1942 }
1943
1944 static void remove_queue_kobjects(struct net_device *dev)
1945 {
1946         int real_rx = 0, real_tx = 0;
1947
1948 #ifdef CONFIG_SYSFS
1949         real_rx = dev->real_num_rx_queues;
1950 #endif
1951         real_tx = dev->real_num_tx_queues;
1952
1953         net_rx_queue_update_kobjects(dev, real_rx, 0);
1954         netdev_queue_update_kobjects(dev, real_tx, 0);
1955
1956         dev->real_num_rx_queues = 0;
1957         dev->real_num_tx_queues = 0;
1958 #ifdef CONFIG_SYSFS
1959         kset_unregister(dev->queues_kset);
1960 #endif
1961 }
1962
1963 static bool net_current_may_mount(void)
1964 {
1965         struct net *net = current->nsproxy->net_ns;
1966
1967         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1968 }
1969
1970 static void *net_grab_current_ns(void)
1971 {
1972         struct net *ns = current->nsproxy->net_ns;
1973 #ifdef CONFIG_NET_NS
1974         if (ns)
1975                 refcount_inc(&ns->passive);
1976 #endif
1977         return ns;
1978 }
1979
1980 static const void *net_initial_ns(void)
1981 {
1982         return &init_net;
1983 }
1984
1985 static const void *net_netlink_ns(struct sock *sk)
1986 {
1987         return sock_net(sk);
1988 }
1989
1990 const struct kobj_ns_type_operations net_ns_type_operations = {
1991         .type = KOBJ_NS_TYPE_NET,
1992         .current_may_mount = net_current_may_mount,
1993         .grab_current_ns = net_grab_current_ns,
1994         .netlink_ns = net_netlink_ns,
1995         .initial_ns = net_initial_ns,
1996         .drop_ns = net_drop_ns,
1997 };
1998 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1999
2000 static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
2001 {
2002         const struct net_device *dev = to_net_dev(d);
2003         int retval;
2004
2005         /* pass interface to uevent. */
2006         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
2007         if (retval)
2008                 goto exit;
2009
2010         /* pass ifindex to uevent.
2011          * ifindex is useful as it won't change (interface name may change)
2012          * and is what RtNetlink uses natively.
2013          */
2014         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
2015
2016 exit:
2017         return retval;
2018 }
2019
2020 /*
2021  *      netdev_release -- destroy and free a dead device.
2022  *      Called when last reference to device kobject is gone.
2023  */
2024 static void netdev_release(struct device *d)
2025 {
2026         struct net_device *dev = to_net_dev(d);
2027
2028         BUG_ON(dev->reg_state != NETREG_RELEASED);
2029
2030         /* no need to wait for rcu grace period:
2031          * device is dead and about to be freed.
2032          */
2033         kfree(rcu_access_pointer(dev->ifalias));
2034         kvfree(dev);
2035 }
2036
2037 static const void *net_namespace(const struct device *d)
2038 {
2039         const struct net_device *dev = to_net_dev(d);
2040
2041         return dev_net(dev);
2042 }
2043
2044 static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
2045 {
2046         const struct net_device *dev = to_net_dev(d);
2047         const struct net *net = dev_net(dev);
2048
2049         net_ns_get_ownership(net, uid, gid);
2050 }
2051
2052 static const struct class net_class = {
2053         .name = "net",
2054         .dev_release = netdev_release,
2055         .dev_groups = net_class_groups,
2056         .dev_uevent = netdev_uevent,
2057         .ns_type = &net_ns_type_operations,
2058         .namespace = net_namespace,
2059         .get_ownership = net_get_ownership,
2060 };
2061
2062 #ifdef CONFIG_OF
2063 static int of_dev_node_match(struct device *dev, const void *data)
2064 {
2065         for (; dev; dev = dev->parent) {
2066                 if (dev->of_node == data)
2067                         return 1;
2068         }
2069
2070         return 0;
2071 }
2072
2073 /*
2074  * of_find_net_device_by_node - lookup the net device for the device node
2075  * @np: OF device node
2076  *
2077  * Looks up the net_device structure corresponding with the device node.
2078  * If successful, returns a pointer to the net_device with the embedded
2079  * struct device refcount incremented by one, or NULL on failure. The
2080  * refcount must be dropped when done with the net_device.
2081  */
2082 struct net_device *of_find_net_device_by_node(struct device_node *np)
2083 {
2084         struct device *dev;
2085
2086         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2087         if (!dev)
2088                 return NULL;
2089
2090         return to_net_dev(dev);
2091 }
2092 EXPORT_SYMBOL(of_find_net_device_by_node);
2093 #endif
2094
2095 /* Delete sysfs entries but hold kobject reference until after all
2096  * netdev references are gone.
2097  */
2098 void netdev_unregister_kobject(struct net_device *ndev)
2099 {
2100         struct device *dev = &ndev->dev;
2101
2102         if (!refcount_read(&dev_net(ndev)->ns.count))
2103                 dev_set_uevent_suppress(dev, 1);
2104
2105         kobject_get(&dev->kobj);
2106
2107         remove_queue_kobjects(ndev);
2108
2109         pm_runtime_set_memalloc_noio(dev, false);
2110
2111         device_del(dev);
2112 }
2113
2114 /* Create sysfs entries for network device. */
2115 int netdev_register_kobject(struct net_device *ndev)
2116 {
2117         struct device *dev = &ndev->dev;
2118         const struct attribute_group **groups = ndev->sysfs_groups;
2119         int error = 0;
2120
2121         device_initialize(dev);
2122         dev->class = &net_class;
2123         dev->platform_data = ndev;
2124         dev->groups = groups;
2125
2126         dev_set_name(dev, "%s", ndev->name);
2127
2128 #ifdef CONFIG_SYSFS
2129         /* Allow for a device specific group */
2130         if (*groups)
2131                 groups++;
2132
2133         *groups++ = &netstat_group;
2134
2135         if (wireless_group_needed(ndev))
2136                 *groups++ = &wireless_group;
2137 #endif /* CONFIG_SYSFS */
2138
2139         error = device_add(dev);
2140         if (error)
2141                 return error;
2142
2143         error = register_queue_kobjects(ndev);
2144         if (error) {
2145                 device_del(dev);
2146                 return error;
2147         }
2148
2149         pm_runtime_set_memalloc_noio(dev, true);
2150
2151         return error;
2152 }
2153
2154 /* Change owner for sysfs entries when moving network devices across network
2155  * namespaces owned by different user namespaces.
2156  */
2157 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2158                         const struct net *net_new)
2159 {
2160         kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2161         kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2162         struct device *dev = &ndev->dev;
2163         int error;
2164
2165         net_ns_get_ownership(net_old, &old_uid, &old_gid);
2166         net_ns_get_ownership(net_new, &new_uid, &new_gid);
2167
2168         /* The network namespace was changed but the owning user namespace is
2169          * identical so there's no need to change the owner of sysfs entries.
2170          */
2171         if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2172                 return 0;
2173
2174         error = device_change_owner(dev, new_uid, new_gid);
2175         if (error)
2176                 return error;
2177
2178         error = queue_change_owner(ndev, new_uid, new_gid);
2179         if (error)
2180                 return error;
2181
2182         return 0;
2183 }
2184
2185 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2186                                 const void *ns)
2187 {
2188         return class_create_file_ns(&net_class, class_attr, ns);
2189 }
2190 EXPORT_SYMBOL(netdev_class_create_file_ns);
2191
2192 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2193                                  const void *ns)
2194 {
2195         class_remove_file_ns(&net_class, class_attr, ns);
2196 }
2197 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2198
2199 int __init netdev_kobject_init(void)
2200 {
2201         kobj_ns_type_register(&net_ns_type_operations);
2202         return class_register(&net_class);
2203 }