net/core/net-sysfs.c

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