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