net/core/net-sysfs.c

   1 /*
   2  * net-sysfs.c - network device class and attributes
   3  *
   4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/kernel.h>
  14 #include <linux/netdevice.h>
  15 #include <net/switchdev.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/slab.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/nsproxy.h>
  20 #include <net/sock.h>
  21 #include <net/net_namespace.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/export.h>
  25 #include <linux/jiffies.h>
  26 #include <linux/pm_runtime.h>
  27 #include <linux/of.h>
  28 #include <linux/of_net.h>
  29
  30 #include "net-sysfs.h"
  31
  32 #ifdef CONFIG_SYSFS
  33 static const char fmt_hex[] = "%#x\n";
  34 static const char fmt_dec[] = "%d\n";
  35 static const char fmt_ulong[] = "%lu\n";
  36 static const char fmt_u64[] = "%llu\n";
  37
  38 static inline int dev_isalive(const struct net_device *dev)
  39 {
  40         return dev->reg_state <= NETREG_REGISTERED;
  41 }
  42
  43 /* use same locking rules as GIF* ioctl's */
  44 static ssize_t netdev_show(const struct device *dev,
  45                            struct device_attribute *attr, char *buf,
  46                            ssize_t (*format)(const struct net_device *, char *))
  47 {
  48         struct net_device *ndev = to_net_dev(dev);
  49         ssize_t ret = -EINVAL;
  50
  51         read_lock(&dev_base_lock);
  52         if (dev_isalive(ndev))
  53                 ret = (*format)(ndev, buf);
  54         read_unlock(&dev_base_lock);
  55
  56         return ret;
  57 }
  58
  59 /* generate a show function for simple field */
  60 #define NETDEVICE_SHOW(field, format_string)                            \
  61 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  62 {                                                                       \
  63         return sprintf(buf, format_string, dev->field);                 \
  64 }                                                                       \
  65 static ssize_t field##_show(struct device *dev,                         \
  66                             struct device_attribute *attr, char *buf)   \
  67 {                                                                       \
  68         return netdev_show(dev, attr, buf, format_##field);             \
  69 }                                                                       \
  70
  71 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  72 NETDEVICE_SHOW(field, format_string);                                   \
  73 static DEVICE_ATTR_RO(field)
  74
  75 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  76 NETDEVICE_SHOW(field, format_string);                                   \
  77 static DEVICE_ATTR_RW(field)
  78
  79 /* use same locking and permission rules as SIF* ioctl's */
  80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  81                             const char *buf, size_t len,
  82                             int (*set)(struct net_device *, unsigned long))
  83 {
  84         struct net_device *netdev = to_net_dev(dev);
  85         struct net *net = dev_net(netdev);
  86         unsigned long new;
  87         int ret = -EINVAL;
  88
  89         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  90                 return -EPERM;
  91
  92         ret = kstrtoul(buf, 0, &new);
  93         if (ret)
  94                 goto err;
  95
  96         if (!rtnl_trylock())
  97                 return restart_syscall();
  98
  99         if (dev_isalive(netdev)) {
 100                 if ((ret = (*set)(netdev, new)) == 0)
 101                         ret = len;
 102         }
 103         rtnl_unlock();
 104  err:
 105         return ret;
 106 }
 107
 108 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 109 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 110 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 111 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 112 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 113 NETDEVICE_SHOW_RO(type, fmt_dec);
 114 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 115
 116 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 117                            char *buf)
 118 {
 119         struct net_device *ndev = to_net_dev(dev);
 120
 121         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
 122 }
 123 static DEVICE_ATTR_RO(iflink);
 124
 125 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 126 {
 127         return sprintf(buf, fmt_dec, dev->name_assign_type);
 128 }
 129
 130 static ssize_t name_assign_type_show(struct device *dev,
 131                                      struct device_attribute *attr,
 132                                      char *buf)
 133 {
 134         struct net_device *ndev = to_net_dev(dev);
 135         ssize_t ret = -EINVAL;
 136
 137         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 138                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 139
 140         return ret;
 141 }
 142 static DEVICE_ATTR_RO(name_assign_type);
 143
 144 /* use same locking rules as GIFHWADDR ioctl's */
 145 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 146                             char *buf)
 147 {
 148         struct net_device *ndev = to_net_dev(dev);
 149         ssize_t ret = -EINVAL;
 150
 151         read_lock(&dev_base_lock);
 152         if (dev_isalive(ndev))
 153                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 154         read_unlock(&dev_base_lock);
 155         return ret;
 156 }
 157 static DEVICE_ATTR_RO(address);
 158
 159 static ssize_t broadcast_show(struct device *dev,
 160                               struct device_attribute *attr, char *buf)
 161 {
 162         struct net_device *ndev = to_net_dev(dev);
 163         if (dev_isalive(ndev))
 164                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 165         return -EINVAL;
 166 }
 167 static DEVICE_ATTR_RO(broadcast);
 168
 169 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 170 {
 171         if (!netif_running(dev))
 172                 return -EINVAL;
 173         return dev_change_carrier(dev, (bool) new_carrier);
 174 }
 175
 176 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 177                              const char *buf, size_t len)
 178 {
 179         return netdev_store(dev, attr, buf, len, change_carrier);
 180 }
 181
 182 static ssize_t carrier_show(struct device *dev,
 183                             struct device_attribute *attr, char *buf)
 184 {
 185         struct net_device *netdev = to_net_dev(dev);
 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 dormant_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_dormant(netdev));
 254
 255         return -EINVAL;
 256 }
 257 static DEVICE_ATTR_RO(dormant);
 258
 259 static const char *const operstates[] = {
 260         "unknown",
 261         "notpresent", /* currently unused */
 262         "down",
 263         "lowerlayerdown",
 264         "testing", /* currently unused */
 265         "dormant",
 266         "up"
 267 };
 268
 269 static ssize_t operstate_show(struct device *dev,
 270                               struct device_attribute *attr, char *buf)
 271 {
 272         const struct net_device *netdev = to_net_dev(dev);
 273         unsigned char operstate;
 274
 275         read_lock(&dev_base_lock);
 276         operstate = netdev->operstate;
 277         if (!netif_running(netdev))
 278                 operstate = IF_OPER_DOWN;
 279         read_unlock(&dev_base_lock);
 280
 281         if (operstate >= ARRAY_SIZE(operstates))
 282                 return -EINVAL; /* should not happen */
 283
 284         return sprintf(buf, "%s\n", operstates[operstate]);
 285 }
 286 static DEVICE_ATTR_RO(operstate);
 287
 288 static ssize_t carrier_changes_show(struct device *dev,
 289                                     struct device_attribute *attr,
 290                                     char *buf)
 291 {
 292         struct net_device *netdev = to_net_dev(dev);
 293         return sprintf(buf, fmt_dec,
 294                        atomic_read(&netdev->carrier_changes));
 295 }
 296 static DEVICE_ATTR_RO(carrier_changes);
 297
 298 /* read-write attributes */
 299
 300 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 301 {
 302         return dev_set_mtu(dev, (int) new_mtu);
 303 }
 304
 305 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 306                          const char *buf, size_t len)
 307 {
 308         return netdev_store(dev, attr, buf, len, change_mtu);
 309 }
 310 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 311
 312 static int change_flags(struct net_device *dev, unsigned long new_flags)
 313 {
 314         return dev_change_flags(dev, (unsigned int) new_flags);
 315 }
 316
 317 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 318                            const char *buf, size_t len)
 319 {
 320         return netdev_store(dev, attr, buf, len, change_flags);
 321 }
 322 NETDEVICE_SHOW_RW(flags, fmt_hex);
 323
 324 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 325 {
 326         unsigned int orig_len = dev->tx_queue_len;
 327         int res;
 328
 329         if (new_len != (unsigned int)new_len)
 330                 return -ERANGE;
 331
 332         if (new_len != orig_len) {
 333                 dev->tx_queue_len = new_len;
 334                 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
 335                 res = notifier_to_errno(res);
 336                 if (res) {
 337                         netdev_err(dev,
 338                                    "refused to change device tx_queue_len\n");
 339                         dev->tx_queue_len = orig_len;
 340                         return -EFAULT;
 341                 }
 342         }
 343
 344         return 0;
 345 }
 346
 347 static ssize_t tx_queue_len_store(struct device *dev,
 348                                   struct device_attribute *attr,
 349                                   const char *buf, size_t len)
 350 {
 351         if (!capable(CAP_NET_ADMIN))
 352                 return -EPERM;
 353
 354         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 355 }
 356 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 357
 358 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 359 {
 360         dev->gro_flush_timeout = val;
 361         return 0;
 362 }
 363
 364 static ssize_t gro_flush_timeout_store(struct device *dev,
 365                                   struct device_attribute *attr,
 366                                   const char *buf, size_t len)
 367 {
 368         if (!capable(CAP_NET_ADMIN))
 369                 return -EPERM;
 370
 371         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 372 }
 373 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 374
 375 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 376                              const char *buf, size_t len)
 377 {
 378         struct net_device *netdev = to_net_dev(dev);
 379         struct net *net = dev_net(netdev);
 380         size_t count = len;
 381         ssize_t ret;
 382
 383         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 384                 return -EPERM;
 385
 386         /* ignore trailing newline */
 387         if (len >  0 && buf[len - 1] == '\n')
 388                 --count;
 389
 390         if (!rtnl_trylock())
 391                 return restart_syscall();
 392         ret = dev_set_alias(netdev, buf, count);
 393         rtnl_unlock();
 394
 395         return ret < 0 ? ret : len;
 396 }
 397
 398 static ssize_t ifalias_show(struct device *dev,
 399                             struct device_attribute *attr, char *buf)
 400 {
 401         const struct net_device *netdev = to_net_dev(dev);
 402         ssize_t ret = 0;
 403
 404         if (!rtnl_trylock())
 405                 return restart_syscall();
 406         if (netdev->ifalias)
 407                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 408         rtnl_unlock();
 409         return ret;
 410 }
 411 static DEVICE_ATTR_RW(ifalias);
 412
 413 static int change_group(struct net_device *dev, unsigned long new_group)
 414 {
 415         dev_set_group(dev, (int) new_group);
 416         return 0;
 417 }
 418
 419 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 420                            const char *buf, size_t len)
 421 {
 422         return netdev_store(dev, attr, buf, len, change_group);
 423 }
 424 NETDEVICE_SHOW(group, fmt_dec);
 425 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 426
 427 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 428 {
 429         return dev_change_proto_down(dev, (bool) proto_down);
 430 }
 431
 432 static ssize_t proto_down_store(struct device *dev,
 433                                 struct device_attribute *attr,
 434                                 const char *buf, size_t len)
 435 {
 436         return netdev_store(dev, attr, buf, len, change_proto_down);
 437 }
 438 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 439
 440 static ssize_t phys_port_id_show(struct device *dev,
 441                                  struct device_attribute *attr, char *buf)
 442 {
 443         struct net_device *netdev = to_net_dev(dev);
 444         ssize_t ret = -EINVAL;
 445
 446         if (!rtnl_trylock())
 447                 return restart_syscall();
 448
 449         if (dev_isalive(netdev)) {
 450                 struct netdev_phys_item_id ppid;
 451
 452                 ret = dev_get_phys_port_id(netdev, &ppid);
 453                 if (!ret)
 454                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 455         }
 456         rtnl_unlock();
 457
 458         return ret;
 459 }
 460 static DEVICE_ATTR_RO(phys_port_id);
 461
 462 static ssize_t phys_port_name_show(struct device *dev,
 463                                    struct device_attribute *attr, char *buf)
 464 {
 465         struct net_device *netdev = to_net_dev(dev);
 466         ssize_t ret = -EINVAL;
 467
 468         if (!rtnl_trylock())
 469                 return restart_syscall();
 470
 471         if (dev_isalive(netdev)) {
 472                 char name[IFNAMSIZ];
 473
 474                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 475                 if (!ret)
 476                         ret = sprintf(buf, "%s\n", name);
 477         }
 478         rtnl_unlock();
 479
 480         return ret;
 481 }
 482 static DEVICE_ATTR_RO(phys_port_name);
 483
 484 static ssize_t phys_switch_id_show(struct device *dev,
 485                                    struct device_attribute *attr, char *buf)
 486 {
 487         struct net_device *netdev = to_net_dev(dev);
 488         ssize_t ret = -EINVAL;
 489
 490         if (!rtnl_trylock())
 491                 return restart_syscall();
 492
 493         if (dev_isalive(netdev)) {
 494                 struct switchdev_attr attr = {
 495                         .orig_dev = netdev,
 496                         .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 497                         .flags = SWITCHDEV_F_NO_RECURSE,
 498                 };
 499
 500                 ret = switchdev_port_attr_get(netdev, &attr);
 501                 if (!ret)
 502                         ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
 503                                       attr.u.ppid.id);
 504         }
 505         rtnl_unlock();
 506
 507         return ret;
 508 }
 509 static DEVICE_ATTR_RO(phys_switch_id);
 510
 511 static struct attribute *net_class_attrs[] = {
 512         &dev_attr_netdev_group.attr,
 513         &dev_attr_type.attr,
 514         &dev_attr_dev_id.attr,
 515         &dev_attr_dev_port.attr,
 516         &dev_attr_iflink.attr,
 517         &dev_attr_ifindex.attr,
 518         &dev_attr_name_assign_type.attr,
 519         &dev_attr_addr_assign_type.attr,
 520         &dev_attr_addr_len.attr,
 521         &dev_attr_link_mode.attr,
 522         &dev_attr_address.attr,
 523         &dev_attr_broadcast.attr,
 524         &dev_attr_speed.attr,
 525         &dev_attr_duplex.attr,
 526         &dev_attr_dormant.attr,
 527         &dev_attr_operstate.attr,
 528         &dev_attr_carrier_changes.attr,
 529         &dev_attr_ifalias.attr,
 530         &dev_attr_carrier.attr,
 531         &dev_attr_mtu.attr,
 532         &dev_attr_flags.attr,
 533         &dev_attr_tx_queue_len.attr,
 534         &dev_attr_gro_flush_timeout.attr,
 535         &dev_attr_phys_port_id.attr,
 536         &dev_attr_phys_port_name.attr,
 537         &dev_attr_phys_switch_id.attr,
 538         &dev_attr_proto_down.attr,
 539         NULL,
 540 };
 541 ATTRIBUTE_GROUPS(net_class);
 542
 543 /* Show a given an attribute in the statistics group */
 544 static ssize_t netstat_show(const struct device *d,
 545                             struct device_attribute *attr, char *buf,
 546                             unsigned long offset)
 547 {
 548         struct net_device *dev = to_net_dev(d);
 549         ssize_t ret = -EINVAL;
 550
 551         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 552                         offset % sizeof(u64) != 0);
 553
 554         read_lock(&dev_base_lock);
 555         if (dev_isalive(dev)) {
 556                 struct rtnl_link_stats64 temp;
 557                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 558
 559                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 560         }
 561         read_unlock(&dev_base_lock);
 562         return ret;
 563 }
 564
 565 /* generate a read-only statistics attribute */
 566 #define NETSTAT_ENTRY(name)                                             \
 567 static ssize_t name##_show(struct device *d,                            \
 568                            struct device_attribute *attr, char *buf)    \
 569 {                                                                       \
 570         return netstat_show(d, attr, buf,                               \
 571                             offsetof(struct rtnl_link_stats64, name));  \
 572 }                                                                       \
 573 static DEVICE_ATTR_RO(name)
 574
 575 NETSTAT_ENTRY(rx_packets);
 576 NETSTAT_ENTRY(tx_packets);
 577 NETSTAT_ENTRY(rx_bytes);
 578 NETSTAT_ENTRY(tx_bytes);
 579 NETSTAT_ENTRY(rx_errors);
 580 NETSTAT_ENTRY(tx_errors);
 581 NETSTAT_ENTRY(rx_dropped);
 582 NETSTAT_ENTRY(tx_dropped);
 583 NETSTAT_ENTRY(multicast);
 584 NETSTAT_ENTRY(collisions);
 585 NETSTAT_ENTRY(rx_length_errors);
 586 NETSTAT_ENTRY(rx_over_errors);
 587 NETSTAT_ENTRY(rx_crc_errors);
 588 NETSTAT_ENTRY(rx_frame_errors);
 589 NETSTAT_ENTRY(rx_fifo_errors);
 590 NETSTAT_ENTRY(rx_missed_errors);
 591 NETSTAT_ENTRY(tx_aborted_errors);
 592 NETSTAT_ENTRY(tx_carrier_errors);
 593 NETSTAT_ENTRY(tx_fifo_errors);
 594 NETSTAT_ENTRY(tx_heartbeat_errors);
 595 NETSTAT_ENTRY(tx_window_errors);
 596 NETSTAT_ENTRY(rx_compressed);
 597 NETSTAT_ENTRY(tx_compressed);
 598 NETSTAT_ENTRY(rx_nohandler);
 599
 600 static struct attribute *netstat_attrs[] = {
 601         &dev_attr_rx_packets.attr,
 602         &dev_attr_tx_packets.attr,
 603         &dev_attr_rx_bytes.attr,
 604         &dev_attr_tx_bytes.attr,
 605         &dev_attr_rx_errors.attr,
 606         &dev_attr_tx_errors.attr,
 607         &dev_attr_rx_dropped.attr,
 608         &dev_attr_tx_dropped.attr,
 609         &dev_attr_multicast.attr,
 610         &dev_attr_collisions.attr,
 611         &dev_attr_rx_length_errors.attr,
 612         &dev_attr_rx_over_errors.attr,
 613         &dev_attr_rx_crc_errors.attr,
 614         &dev_attr_rx_frame_errors.attr,
 615         &dev_attr_rx_fifo_errors.attr,
 616         &dev_attr_rx_missed_errors.attr,
 617         &dev_attr_tx_aborted_errors.attr,
 618         &dev_attr_tx_carrier_errors.attr,
 619         &dev_attr_tx_fifo_errors.attr,
 620         &dev_attr_tx_heartbeat_errors.attr,
 621         &dev_attr_tx_window_errors.attr,
 622         &dev_attr_rx_compressed.attr,
 623         &dev_attr_tx_compressed.attr,
 624         &dev_attr_rx_nohandler.attr,
 625         NULL
 626 };
 627
 628
 629 static const struct attribute_group netstat_group = {
 630         .name  = "statistics",
 631         .attrs  = netstat_attrs,
 632 };
 633
 634 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 635 static struct attribute *wireless_attrs[] = {
 636         NULL
 637 };
 638
 639 static const struct attribute_group wireless_group = {
 640         .name = "wireless",
 641         .attrs = wireless_attrs,
 642 };
 643 #endif
 644
 645 #else /* CONFIG_SYSFS */
 646 #define net_class_groups        NULL
 647 #endif /* CONFIG_SYSFS */
 648
 649 #ifdef CONFIG_SYSFS
 650 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 651     struct rx_queue_attribute, attr)
 652
 653 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 654
 655 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 656                                   char *buf)
 657 {
 658         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 659         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 660
 661         if (!attribute->show)
 662                 return -EIO;
 663
 664         return attribute->show(queue, attribute, buf);
 665 }
 666
 667 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 668                                    const char *buf, size_t count)
 669 {
 670         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 671         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 672
 673         if (!attribute->store)
 674                 return -EIO;
 675
 676         return attribute->store(queue, attribute, buf, count);
 677 }
 678
 679 static const struct sysfs_ops rx_queue_sysfs_ops = {
 680         .show = rx_queue_attr_show,
 681         .store = rx_queue_attr_store,
 682 };
 683
 684 #ifdef CONFIG_RPS
 685 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 686                             struct rx_queue_attribute *attribute, char *buf)
 687 {
 688         struct rps_map *map;
 689         cpumask_var_t mask;
 690         int i, len;
 691
 692         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 693                 return -ENOMEM;
 694
 695         rcu_read_lock();
 696         map = rcu_dereference(queue->rps_map);
 697         if (map)
 698                 for (i = 0; i < map->len; i++)
 699                         cpumask_set_cpu(map->cpus[i], mask);
 700
 701         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 702         rcu_read_unlock();
 703         free_cpumask_var(mask);
 704
 705         return len < PAGE_SIZE ? len : -EINVAL;
 706 }
 707
 708 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 709                       struct rx_queue_attribute *attribute,
 710                       const char *buf, size_t len)
 711 {
 712         struct rps_map *old_map, *map;
 713         cpumask_var_t mask;
 714         int err, cpu, i;
 715         static DEFINE_MUTEX(rps_map_mutex);
 716
 717         if (!capable(CAP_NET_ADMIN))
 718                 return -EPERM;
 719
 720         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 721                 return -ENOMEM;
 722
 723         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 724         if (err) {
 725                 free_cpumask_var(mask);
 726                 return err;
 727         }
 728
 729         map = kzalloc(max_t(unsigned int,
 730             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 731             GFP_KERNEL);
 732         if (!map) {
 733                 free_cpumask_var(mask);
 734                 return -ENOMEM;
 735         }
 736
 737         i = 0;
 738         for_each_cpu_and(cpu, mask, cpu_online_mask)
 739                 map->cpus[i++] = cpu;
 740
 741         if (i)
 742                 map->len = i;
 743         else {
 744                 kfree(map);
 745                 map = NULL;
 746         }
 747
 748         mutex_lock(&rps_map_mutex);
 749         old_map = rcu_dereference_protected(queue->rps_map,
 750                                             mutex_is_locked(&rps_map_mutex));
 751         rcu_assign_pointer(queue->rps_map, map);
 752
 753         if (map)
 754                 static_key_slow_inc(&rps_needed);
 755         if (old_map)
 756                 static_key_slow_dec(&rps_needed);
 757
 758         mutex_unlock(&rps_map_mutex);
 759
 760         if (old_map)
 761                 kfree_rcu(old_map, rcu);
 762
 763         free_cpumask_var(mask);
 764         return len;
 765 }
 766
 767 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 768                                            struct rx_queue_attribute *attr,
 769                                            char *buf)
 770 {
 771         struct rps_dev_flow_table *flow_table;
 772         unsigned long val = 0;
 773
 774         rcu_read_lock();
 775         flow_table = rcu_dereference(queue->rps_flow_table);
 776         if (flow_table)
 777                 val = (unsigned long)flow_table->mask + 1;
 778         rcu_read_unlock();
 779
 780         return sprintf(buf, "%lu\n", val);
 781 }
 782
 783 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 784 {
 785         struct rps_dev_flow_table *table = container_of(rcu,
 786             struct rps_dev_flow_table, rcu);
 787         vfree(table);
 788 }
 789
 790 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 791                                      struct rx_queue_attribute *attr,
 792                                      const char *buf, size_t len)
 793 {
 794         unsigned long mask, count;
 795         struct rps_dev_flow_table *table, *old_table;
 796         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 797         int rc;
 798
 799         if (!capable(CAP_NET_ADMIN))
 800                 return -EPERM;
 801
 802         rc = kstrtoul(buf, 0, &count);
 803         if (rc < 0)
 804                 return rc;
 805
 806         if (count) {
 807                 mask = count - 1;
 808                 /* mask = roundup_pow_of_two(count) - 1;
 809                  * without overflows...
 810                  */
 811                 while ((mask | (mask >> 1)) != mask)
 812                         mask |= (mask >> 1);
 813                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 814                  * and on 32bit arches, must check
 815                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 816                  */
 817 #if BITS_PER_LONG > 32
 818                 if (mask > (unsigned long)(u32)mask)
 819                         return -EINVAL;
 820 #else
 821                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 822                                 / sizeof(struct rps_dev_flow)) {
 823                         /* Enforce a limit to prevent overflow */
 824                         return -EINVAL;
 825                 }
 826 #endif
 827                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 828                 if (!table)
 829                         return -ENOMEM;
 830
 831                 table->mask = mask;
 832                 for (count = 0; count <= mask; count++)
 833                         table->flows[count].cpu = RPS_NO_CPU;
 834         } else
 835                 table = NULL;
 836
 837         spin_lock(&rps_dev_flow_lock);
 838         old_table = rcu_dereference_protected(queue->rps_flow_table,
 839                                               lockdep_is_held(&rps_dev_flow_lock));
 840         rcu_assign_pointer(queue->rps_flow_table, table);
 841         spin_unlock(&rps_dev_flow_lock);
 842
 843         if (old_table)
 844                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 845
 846         return len;
 847 }
 848
 849 static struct rx_queue_attribute rps_cpus_attribute =
 850         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 851
 852
 853 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 854         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 855             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 856 #endif /* CONFIG_RPS */
 857
 858 static struct attribute *rx_queue_default_attrs[] = {
 859 #ifdef CONFIG_RPS
 860         &rps_cpus_attribute.attr,
 861         &rps_dev_flow_table_cnt_attribute.attr,
 862 #endif
 863         NULL
 864 };
 865
 866 static void rx_queue_release(struct kobject *kobj)
 867 {
 868         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 869 #ifdef CONFIG_RPS
 870         struct rps_map *map;
 871         struct rps_dev_flow_table *flow_table;
 872
 873
 874         map = rcu_dereference_protected(queue->rps_map, 1);
 875         if (map) {
 876                 RCU_INIT_POINTER(queue->rps_map, NULL);
 877                 kfree_rcu(map, rcu);
 878         }
 879
 880         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 881         if (flow_table) {
 882                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 883                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 884         }
 885 #endif
 886
 887         memset(kobj, 0, sizeof(*kobj));
 888         dev_put(queue->dev);
 889 }
 890
 891 static const void *rx_queue_namespace(struct kobject *kobj)
 892 {
 893         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 894         struct device *dev = &queue->dev->dev;
 895         const void *ns = NULL;
 896
 897         if (dev->class && dev->class->ns_type)
 898                 ns = dev->class->namespace(dev);
 899
 900         return ns;
 901 }
 902
 903 static struct kobj_type rx_queue_ktype = {
 904         .sysfs_ops = &rx_queue_sysfs_ops,
 905         .release = rx_queue_release,
 906         .default_attrs = rx_queue_default_attrs,
 907         .namespace = rx_queue_namespace
 908 };
 909
 910 static int rx_queue_add_kobject(struct net_device *dev, int index)
 911 {
 912         struct netdev_rx_queue *queue = dev->_rx + index;
 913         struct kobject *kobj = &queue->kobj;
 914         int error = 0;
 915
 916         kobj->kset = dev->queues_kset;
 917         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 918             "rx-%u", index);
 919         if (error)
 920                 goto exit;
 921
 922         if (dev->sysfs_rx_queue_group) {
 923                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 924                 if (error)
 925                         goto exit;
 926         }
 927
 928         kobject_uevent(kobj, KOBJ_ADD);
 929         dev_hold(queue->dev);
 930
 931         return error;
 932 exit:
 933         kobject_put(kobj);
 934         return error;
 935 }
 936 #endif /* CONFIG_SYSFS */
 937
 938 int
 939 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 940 {
 941 #ifdef CONFIG_SYSFS
 942         int i;
 943         int error = 0;
 944
 945 #ifndef CONFIG_RPS
 946         if (!dev->sysfs_rx_queue_group)
 947                 return 0;
 948 #endif
 949         for (i = old_num; i < new_num; i++) {
 950                 error = rx_queue_add_kobject(dev, i);
 951                 if (error) {
 952                         new_num = old_num;
 953                         break;
 954                 }
 955         }
 956
 957         while (--i >= new_num) {
 958                 struct kobject *kobj = &dev->_rx[i].kobj;
 959
 960                 if (!atomic_read(&dev_net(dev)->count))
 961                         kobj->uevent_suppress = 1;
 962                 if (dev->sysfs_rx_queue_group)
 963                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
 964                 kobject_put(kobj);
 965         }
 966
 967         return error;
 968 #else
 969         return 0;
 970 #endif
 971 }
 972
 973 #ifdef CONFIG_SYSFS
 974 /*
 975  * netdev_queue sysfs structures and functions.
 976  */
 977 struct netdev_queue_attribute {
 978         struct attribute attr;
 979         ssize_t (*show)(struct netdev_queue *queue,
 980             struct netdev_queue_attribute *attr, char *buf);
 981         ssize_t (*store)(struct netdev_queue *queue,
 982             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 983 };
 984 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 985     struct netdev_queue_attribute, attr)
 986
 987 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 988
 989 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 990                                       struct attribute *attr, char *buf)
 991 {
 992         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 993         struct netdev_queue *queue = to_netdev_queue(kobj);
 994
 995         if (!attribute->show)
 996                 return -EIO;
 997
 998         return attribute->show(queue, attribute, buf);
 999 }
1000
1001 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1002                                        struct attribute *attr,
1003                                        const char *buf, size_t count)
1004 {
1005         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
1006         struct netdev_queue *queue = to_netdev_queue(kobj);
1007
1008         if (!attribute->store)
1009                 return -EIO;
1010
1011         return attribute->store(queue, attribute, buf, count);
1012 }
1013
1014 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1015         .show = netdev_queue_attr_show,
1016         .store = netdev_queue_attr_store,
1017 };
1018
1019 static ssize_t show_trans_timeout(struct netdev_queue *queue,
1020                                   struct netdev_queue_attribute *attribute,
1021                                   char *buf)
1022 {
1023         unsigned long trans_timeout;
1024
1025         spin_lock_irq(&queue->_xmit_lock);
1026         trans_timeout = queue->trans_timeout;
1027         spin_unlock_irq(&queue->_xmit_lock);
1028
1029         return sprintf(buf, "%lu", trans_timeout);
1030 }
1031
1032 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1033 {
1034         struct net_device *dev = queue->dev;
1035         unsigned int i;
1036
1037         i = queue - dev->_tx;
1038         BUG_ON(i >= dev->num_tx_queues);
1039
1040         return i;
1041 }
1042
1043 static ssize_t show_traffic_class(struct netdev_queue *queue,
1044                                   struct netdev_queue_attribute *attribute,
1045                                   char *buf)
1046 {
1047         struct net_device *dev = queue->dev;
1048         int index = get_netdev_queue_index(queue);
1049         int tc = netdev_txq_to_tc(dev, index);
1050
1051         if (tc < 0)
1052                 return -EINVAL;
1053
1054         return sprintf(buf, "%u\n", tc);
1055 }
1056
1057 #ifdef CONFIG_XPS
1058 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1059                                struct netdev_queue_attribute *attribute,
1060                                char *buf)
1061 {
1062         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1063 }
1064
1065 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1066                               struct netdev_queue_attribute *attribute,
1067                               const char *buf, size_t len)
1068 {
1069         struct net_device *dev = queue->dev;
1070         int err, index = get_netdev_queue_index(queue);
1071         u32 rate = 0;
1072
1073         err = kstrtou32(buf, 10, &rate);
1074         if (err < 0)
1075                 return err;
1076
1077         if (!rtnl_trylock())
1078                 return restart_syscall();
1079
1080         err = -EOPNOTSUPP;
1081         if (dev->netdev_ops->ndo_set_tx_maxrate)
1082                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1083
1084         rtnl_unlock();
1085         if (!err) {
1086                 queue->tx_maxrate = rate;
1087                 return len;
1088         }
1089         return err;
1090 }
1091
1092 static struct netdev_queue_attribute queue_tx_maxrate =
1093         __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1094                show_tx_maxrate, set_tx_maxrate);
1095 #endif
1096
1097 static struct netdev_queue_attribute queue_trans_timeout =
1098         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1099
1100 static struct netdev_queue_attribute queue_traffic_class =
1101         __ATTR(traffic_class, S_IRUGO, show_traffic_class, NULL);
1102
1103 #ifdef CONFIG_BQL
1104 /*
1105  * Byte queue limits sysfs structures and functions.
1106  */
1107 static ssize_t bql_show(char *buf, unsigned int value)
1108 {
1109         return sprintf(buf, "%u\n", value);
1110 }
1111
1112 static ssize_t bql_set(const char *buf, const size_t count,
1113                        unsigned int *pvalue)
1114 {
1115         unsigned int value;
1116         int err;
1117
1118         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1119                 value = DQL_MAX_LIMIT;
1120         else {
1121                 err = kstrtouint(buf, 10, &value);
1122                 if (err < 0)
1123                         return err;
1124                 if (value > DQL_MAX_LIMIT)
1125                         return -EINVAL;
1126         }
1127
1128         *pvalue = value;
1129
1130         return count;
1131 }
1132
1133 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1134                                   struct netdev_queue_attribute *attr,
1135                                   char *buf)
1136 {
1137         struct dql *dql = &queue->dql;
1138
1139         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1140 }
1141
1142 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1143                                  struct netdev_queue_attribute *attribute,
1144                                  const char *buf, size_t len)
1145 {
1146         struct dql *dql = &queue->dql;
1147         unsigned int value;
1148         int err;
1149
1150         err = kstrtouint(buf, 10, &value);
1151         if (err < 0)
1152                 return err;
1153
1154         dql->slack_hold_time = msecs_to_jiffies(value);
1155
1156         return len;
1157 }
1158
1159 static struct netdev_queue_attribute bql_hold_time_attribute =
1160         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1161             bql_set_hold_time);
1162
1163 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1164                                  struct netdev_queue_attribute *attr,
1165                                  char *buf)
1166 {
1167         struct dql *dql = &queue->dql;
1168
1169         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1170 }
1171
1172 static struct netdev_queue_attribute bql_inflight_attribute =
1173         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1174
1175 #define BQL_ATTR(NAME, FIELD)                                           \
1176 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1177                                  struct netdev_queue_attribute *attr,   \
1178                                  char *buf)                             \
1179 {                                                                       \
1180         return bql_show(buf, queue->dql.FIELD);                         \
1181 }                                                                       \
1182                                                                         \
1183 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1184                                 struct netdev_queue_attribute *attr,    \
1185                                 const char *buf, size_t len)            \
1186 {                                                                       \
1187         return bql_set(buf, len, &queue->dql.FIELD);                    \
1188 }                                                                       \
1189                                                                         \
1190 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1191         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1192             bql_set_ ## NAME);
1193
1194 BQL_ATTR(limit, limit)
1195 BQL_ATTR(limit_max, max_limit)
1196 BQL_ATTR(limit_min, min_limit)
1197
1198 static struct attribute *dql_attrs[] = {
1199         &bql_limit_attribute.attr,
1200         &bql_limit_max_attribute.attr,
1201         &bql_limit_min_attribute.attr,
1202         &bql_hold_time_attribute.attr,
1203         &bql_inflight_attribute.attr,
1204         NULL
1205 };
1206
1207 static const struct attribute_group dql_group = {
1208         .name  = "byte_queue_limits",
1209         .attrs  = dql_attrs,
1210 };
1211 #endif /* CONFIG_BQL */
1212
1213 #ifdef CONFIG_XPS
1214 static ssize_t show_xps_map(struct netdev_queue *queue,
1215                             struct netdev_queue_attribute *attribute, char *buf)
1216 {
1217         struct net_device *dev = queue->dev;
1218         int cpu, len, num_tc = 1, tc = 0;
1219         struct xps_dev_maps *dev_maps;
1220         cpumask_var_t mask;
1221         unsigned long index;
1222
1223         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1224                 return -ENOMEM;
1225
1226         index = get_netdev_queue_index(queue);
1227
1228         if (dev->num_tc) {
1229                 num_tc = dev->num_tc;
1230                 tc = netdev_txq_to_tc(dev, index);
1231                 if (tc < 0)
1232                         return -EINVAL;
1233         }
1234
1235         rcu_read_lock();
1236         dev_maps = rcu_dereference(dev->xps_maps);
1237         if (dev_maps) {
1238                 for_each_possible_cpu(cpu) {
1239                         int i, tci = cpu * num_tc + tc;
1240                         struct xps_map *map;
1241
1242                         map = rcu_dereference(dev_maps->cpu_map[tci]);
1243                         if (!map)
1244                                 continue;
1245
1246                         for (i = map->len; i--;) {
1247                                 if (map->queues[i] == index) {
1248                                         cpumask_set_cpu(cpu, mask);
1249                                         break;
1250                                 }
1251                         }
1252                 }
1253         }
1254         rcu_read_unlock();
1255
1256         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1257         free_cpumask_var(mask);
1258         return len < PAGE_SIZE ? len : -EINVAL;
1259 }
1260
1261 static ssize_t store_xps_map(struct netdev_queue *queue,
1262                       struct netdev_queue_attribute *attribute,
1263                       const char *buf, size_t len)
1264 {
1265         struct net_device *dev = queue->dev;
1266         unsigned long index;
1267         cpumask_var_t mask;
1268         int err;
1269
1270         if (!capable(CAP_NET_ADMIN))
1271                 return -EPERM;
1272
1273         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1274                 return -ENOMEM;
1275
1276         index = get_netdev_queue_index(queue);
1277
1278         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1279         if (err) {
1280                 free_cpumask_var(mask);
1281                 return err;
1282         }
1283
1284         err = netif_set_xps_queue(dev, mask, index);
1285
1286         free_cpumask_var(mask);
1287
1288         return err ? : len;
1289 }
1290
1291 static struct netdev_queue_attribute xps_cpus_attribute =
1292     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1293 #endif /* CONFIG_XPS */
1294
1295 static struct attribute *netdev_queue_default_attrs[] = {
1296         &queue_trans_timeout.attr,
1297         &queue_traffic_class.attr,
1298 #ifdef CONFIG_XPS
1299         &xps_cpus_attribute.attr,
1300         &queue_tx_maxrate.attr,
1301 #endif
1302         NULL
1303 };
1304
1305 static void netdev_queue_release(struct kobject *kobj)
1306 {
1307         struct netdev_queue *queue = to_netdev_queue(kobj);
1308
1309         memset(kobj, 0, sizeof(*kobj));
1310         dev_put(queue->dev);
1311 }
1312
1313 static const void *netdev_queue_namespace(struct kobject *kobj)
1314 {
1315         struct netdev_queue *queue = to_netdev_queue(kobj);
1316         struct device *dev = &queue->dev->dev;
1317         const void *ns = NULL;
1318
1319         if (dev->class && dev->class->ns_type)
1320                 ns = dev->class->namespace(dev);
1321
1322         return ns;
1323 }
1324
1325 static struct kobj_type netdev_queue_ktype = {
1326         .sysfs_ops = &netdev_queue_sysfs_ops,
1327         .release = netdev_queue_release,
1328         .default_attrs = netdev_queue_default_attrs,
1329         .namespace = netdev_queue_namespace,
1330 };
1331
1332 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1333 {
1334         struct netdev_queue *queue = dev->_tx + index;
1335         struct kobject *kobj = &queue->kobj;
1336         int error = 0;
1337
1338         kobj->kset = dev->queues_kset;
1339         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1340             "tx-%u", index);
1341         if (error)
1342                 goto exit;
1343
1344 #ifdef CONFIG_BQL
1345         error = sysfs_create_group(kobj, &dql_group);
1346         if (error)
1347                 goto exit;
1348 #endif
1349
1350         kobject_uevent(kobj, KOBJ_ADD);
1351         dev_hold(queue->dev);
1352
1353         return 0;
1354 exit:
1355         kobject_put(kobj);
1356         return error;
1357 }
1358 #endif /* CONFIG_SYSFS */
1359
1360 int
1361 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1362 {
1363 #ifdef CONFIG_SYSFS
1364         int i;
1365         int error = 0;
1366
1367         for (i = old_num; i < new_num; i++) {
1368                 error = netdev_queue_add_kobject(dev, i);
1369                 if (error) {
1370                         new_num = old_num;
1371                         break;
1372                 }
1373         }
1374
1375         while (--i >= new_num) {
1376                 struct netdev_queue *queue = dev->_tx + i;
1377
1378                 if (!atomic_read(&dev_net(dev)->count))
1379                         queue->kobj.uevent_suppress = 1;
1380 #ifdef CONFIG_BQL
1381                 sysfs_remove_group(&queue->kobj, &dql_group);
1382 #endif
1383                 kobject_put(&queue->kobj);
1384         }
1385
1386         return error;
1387 #else
1388         return 0;
1389 #endif /* CONFIG_SYSFS */
1390 }
1391
1392 static int register_queue_kobjects(struct net_device *dev)
1393 {
1394         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1395
1396 #ifdef CONFIG_SYSFS
1397         dev->queues_kset = kset_create_and_add("queues",
1398             NULL, &dev->dev.kobj);
1399         if (!dev->queues_kset)
1400                 return -ENOMEM;
1401         real_rx = dev->real_num_rx_queues;
1402 #endif
1403         real_tx = dev->real_num_tx_queues;
1404
1405         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1406         if (error)
1407                 goto error;
1408         rxq = real_rx;
1409
1410         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1411         if (error)
1412                 goto error;
1413         txq = real_tx;
1414
1415         return 0;
1416
1417 error:
1418         netdev_queue_update_kobjects(dev, txq, 0);
1419         net_rx_queue_update_kobjects(dev, rxq, 0);
1420         return error;
1421 }
1422
1423 static void remove_queue_kobjects(struct net_device *dev)
1424 {
1425         int real_rx = 0, real_tx = 0;
1426
1427 #ifdef CONFIG_SYSFS
1428         real_rx = dev->real_num_rx_queues;
1429 #endif
1430         real_tx = dev->real_num_tx_queues;
1431
1432         net_rx_queue_update_kobjects(dev, real_rx, 0);
1433         netdev_queue_update_kobjects(dev, real_tx, 0);
1434 #ifdef CONFIG_SYSFS
1435         kset_unregister(dev->queues_kset);
1436 #endif
1437 }
1438
1439 static bool net_current_may_mount(void)
1440 {
1441         struct net *net = current->nsproxy->net_ns;
1442
1443         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1444 }
1445
1446 static void *net_grab_current_ns(void)
1447 {
1448         struct net *ns = current->nsproxy->net_ns;
1449 #ifdef CONFIG_NET_NS
1450         if (ns)
1451                 refcount_inc(&ns->passive);
1452 #endif
1453         return ns;
1454 }
1455
1456 static const void *net_initial_ns(void)
1457 {
1458         return &init_net;
1459 }
1460
1461 static const void *net_netlink_ns(struct sock *sk)
1462 {
1463         return sock_net(sk);
1464 }
1465
1466 struct kobj_ns_type_operations net_ns_type_operations = {
1467         .type = KOBJ_NS_TYPE_NET,
1468         .current_may_mount = net_current_may_mount,
1469         .grab_current_ns = net_grab_current_ns,
1470         .netlink_ns = net_netlink_ns,
1471         .initial_ns = net_initial_ns,
1472         .drop_ns = net_drop_ns,
1473 };
1474 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1475
1476 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1477 {
1478         struct net_device *dev = to_net_dev(d);
1479         int retval;
1480
1481         /* pass interface to uevent. */
1482         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1483         if (retval)
1484                 goto exit;
1485
1486         /* pass ifindex to uevent.
1487          * ifindex is useful as it won't change (interface name may change)
1488          * and is what RtNetlink uses natively. */
1489         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1490
1491 exit:
1492         return retval;
1493 }
1494
1495 /*
1496  *      netdev_release -- destroy and free a dead device.
1497  *      Called when last reference to device kobject is gone.
1498  */
1499 static void netdev_release(struct device *d)
1500 {
1501         struct net_device *dev = to_net_dev(d);
1502
1503         BUG_ON(dev->reg_state != NETREG_RELEASED);
1504
1505         kfree(dev->ifalias);
1506         netdev_freemem(dev);
1507 }
1508
1509 static const void *net_namespace(struct device *d)
1510 {
1511         struct net_device *dev = to_net_dev(d);
1512
1513         return dev_net(dev);
1514 }
1515
1516 static struct class net_class = {
1517         .name = "net",
1518         .dev_release = netdev_release,
1519         .dev_groups = net_class_groups,
1520         .dev_uevent = netdev_uevent,
1521         .ns_type = &net_ns_type_operations,
1522         .namespace = net_namespace,
1523 };
1524
1525 #ifdef CONFIG_OF_NET
1526 static int of_dev_node_match(struct device *dev, const void *data)
1527 {
1528         int ret = 0;
1529
1530         if (dev->parent)
1531                 ret = dev->parent->of_node == data;
1532
1533         return ret == 0 ? dev->of_node == data : ret;
1534 }
1535
1536 /*
1537  * of_find_net_device_by_node - lookup the net device for the device node
1538  * @np: OF device node
1539  *
1540  * Looks up the net_device structure corresponding with the device node.
1541  * If successful, returns a pointer to the net_device with the embedded
1542  * struct device refcount incremented by one, or NULL on failure. The
1543  * refcount must be dropped when done with the net_device.
1544  */
1545 struct net_device *of_find_net_device_by_node(struct device_node *np)
1546 {
1547         struct device *dev;
1548
1549         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1550         if (!dev)
1551                 return NULL;
1552
1553         return to_net_dev(dev);
1554 }
1555 EXPORT_SYMBOL(of_find_net_device_by_node);
1556 #endif
1557
1558 /* Delete sysfs entries but hold kobject reference until after all
1559  * netdev references are gone.
1560  */
1561 void netdev_unregister_kobject(struct net_device *ndev)
1562 {
1563         struct device *dev = &(ndev->dev);
1564
1565         if (!atomic_read(&dev_net(ndev)->count))
1566                 dev_set_uevent_suppress(dev, 1);
1567
1568         kobject_get(&dev->kobj);
1569
1570         remove_queue_kobjects(ndev);
1571
1572         pm_runtime_set_memalloc_noio(dev, false);
1573
1574         device_del(dev);
1575 }
1576
1577 /* Create sysfs entries for network device. */
1578 int netdev_register_kobject(struct net_device *ndev)
1579 {
1580         struct device *dev = &(ndev->dev);
1581         const struct attribute_group **groups = ndev->sysfs_groups;
1582         int error = 0;
1583
1584         device_initialize(dev);
1585         dev->class = &net_class;
1586         dev->platform_data = ndev;
1587         dev->groups = groups;
1588
1589         dev_set_name(dev, "%s", ndev->name);
1590
1591 #ifdef CONFIG_SYSFS
1592         /* Allow for a device specific group */
1593         if (*groups)
1594                 groups++;
1595
1596         *groups++ = &netstat_group;
1597
1598 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1599         if (ndev->ieee80211_ptr)
1600                 *groups++ = &wireless_group;
1601 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1602         else if (ndev->wireless_handlers)
1603                 *groups++ = &wireless_group;
1604 #endif
1605 #endif
1606 #endif /* CONFIG_SYSFS */
1607
1608         error = device_add(dev);
1609         if (error)
1610                 return error;
1611
1612         error = register_queue_kobjects(ndev);
1613         if (error) {
1614                 device_del(dev);
1615                 return error;
1616         }
1617
1618         pm_runtime_set_memalloc_noio(dev, true);
1619
1620         return error;
1621 }
1622
1623 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1624                                 const void *ns)
1625 {
1626         return class_create_file_ns(&net_class, class_attr, ns);
1627 }
1628 EXPORT_SYMBOL(netdev_class_create_file_ns);
1629
1630 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1631                                  const void *ns)
1632 {
1633         class_remove_file_ns(&net_class, class_attr, ns);
1634 }
1635 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1636
1637 int __init netdev_kobject_init(void)
1638 {
1639         kobj_ns_type_register(&net_ns_type_operations);
1640         return class_register(&net_class);
1641 }