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