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