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/nsproxy.h>
  19 #include <net/sock.h>
  20 #include <net/net_namespace.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/export.h>
  24 #include <linux/jiffies.h>
  25 #include <linux/pm_runtime.h>
  26 #include <linux/of.h>
  27
  28 #include "net-sysfs.h"
  29
  30 #ifdef CONFIG_SYSFS
  31 static const char fmt_hex[] = "%#x\n";
  32 static const char fmt_long_hex[] = "%#lx\n";
  33 static const char fmt_dec[] = "%d\n";
  34 static const char fmt_udec[] = "%u\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_cmd cmd;
 204                 if (!__ethtool_get_settings(netdev, &cmd))
 205                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 206         }
 207         rtnl_unlock();
 208         return ret;
 209 }
 210 static DEVICE_ATTR_RO(speed);
 211
 212 static ssize_t duplex_show(struct device *dev,
 213                            struct device_attribute *attr, char *buf)
 214 {
 215         struct net_device *netdev = to_net_dev(dev);
 216         int ret = -EINVAL;
 217
 218         if (!rtnl_trylock())
 219                 return restart_syscall();
 220
 221         if (netif_running(netdev)) {
 222                 struct ethtool_cmd cmd;
 223                 if (!__ethtool_get_settings(netdev, &cmd)) {
 224                         const char *duplex;
 225                         switch (cmd.duplex) {
 226                         case DUPLEX_HALF:
 227                                 duplex = "half";
 228                                 break;
 229                         case DUPLEX_FULL:
 230                                 duplex = "full";
 231                                 break;
 232                         default:
 233                                 duplex = "unknown";
 234                                 break;
 235                         }
 236                         ret = sprintf(buf, "%s\n", duplex);
 237                 }
 238         }
 239         rtnl_unlock();
 240         return ret;
 241 }
 242 static DEVICE_ATTR_RO(duplex);
 243
 244 static ssize_t dormant_show(struct device *dev,
 245                             struct device_attribute *attr, char *buf)
 246 {
 247         struct net_device *netdev = to_net_dev(dev);
 248
 249         if (netif_running(netdev))
 250                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 251
 252         return -EINVAL;
 253 }
 254 static DEVICE_ATTR_RO(dormant);
 255
 256 static const char *const operstates[] = {
 257         "unknown",
 258         "notpresent", /* currently unused */
 259         "down",
 260         "lowerlayerdown",
 261         "testing", /* currently unused */
 262         "dormant",
 263         "up"
 264 };
 265
 266 static ssize_t operstate_show(struct device *dev,
 267                               struct device_attribute *attr, char *buf)
 268 {
 269         const struct net_device *netdev = to_net_dev(dev);
 270         unsigned char operstate;
 271
 272         read_lock(&dev_base_lock);
 273         operstate = netdev->operstate;
 274         if (!netif_running(netdev))
 275                 operstate = IF_OPER_DOWN;
 276         read_unlock(&dev_base_lock);
 277
 278         if (operstate >= ARRAY_SIZE(operstates))
 279                 return -EINVAL; /* should not happen */
 280
 281         return sprintf(buf, "%s\n", operstates[operstate]);
 282 }
 283 static DEVICE_ATTR_RO(operstate);
 284
 285 static ssize_t carrier_changes_show(struct device *dev,
 286                                     struct device_attribute *attr,
 287                                     char *buf)
 288 {
 289         struct net_device *netdev = to_net_dev(dev);
 290         return sprintf(buf, fmt_dec,
 291                        atomic_read(&netdev->carrier_changes));
 292 }
 293 static DEVICE_ATTR_RO(carrier_changes);
 294
 295 /* read-write attributes */
 296
 297 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 298 {
 299         return dev_set_mtu(dev, (int) new_mtu);
 300 }
 301
 302 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 303                          const char *buf, size_t len)
 304 {
 305         return netdev_store(dev, attr, buf, len, change_mtu);
 306 }
 307 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 308
 309 static int change_flags(struct net_device *dev, unsigned long new_flags)
 310 {
 311         return dev_change_flags(dev, (unsigned int) new_flags);
 312 }
 313
 314 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 315                            const char *buf, size_t len)
 316 {
 317         return netdev_store(dev, attr, buf, len, change_flags);
 318 }
 319 NETDEVICE_SHOW_RW(flags, fmt_hex);
 320
 321 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 322 {
 323         dev->tx_queue_len = new_len;
 324         return 0;
 325 }
 326
 327 static ssize_t tx_queue_len_store(struct device *dev,
 328                                   struct device_attribute *attr,
 329                                   const char *buf, size_t len)
 330 {
 331         if (!capable(CAP_NET_ADMIN))
 332                 return -EPERM;
 333
 334         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 335 }
 336 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
 337
 338 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 339 {
 340         dev->gro_flush_timeout = val;
 341         return 0;
 342 }
 343
 344 static ssize_t gro_flush_timeout_store(struct device *dev,
 345                                   struct device_attribute *attr,
 346                                   const char *buf, size_t len)
 347 {
 348         if (!capable(CAP_NET_ADMIN))
 349                 return -EPERM;
 350
 351         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 352 }
 353 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 354
 355 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 356                              const char *buf, size_t len)
 357 {
 358         struct net_device *netdev = to_net_dev(dev);
 359         struct net *net = dev_net(netdev);
 360         size_t count = len;
 361         ssize_t ret;
 362
 363         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 364                 return -EPERM;
 365
 366         /* ignore trailing newline */
 367         if (len >  0 && buf[len - 1] == '\n')
 368                 --count;
 369
 370         if (!rtnl_trylock())
 371                 return restart_syscall();
 372         ret = dev_set_alias(netdev, buf, count);
 373         rtnl_unlock();
 374
 375         return ret < 0 ? ret : len;
 376 }
 377
 378 static ssize_t ifalias_show(struct device *dev,
 379                             struct device_attribute *attr, char *buf)
 380 {
 381         const struct net_device *netdev = to_net_dev(dev);
 382         ssize_t ret = 0;
 383
 384         if (!rtnl_trylock())
 385                 return restart_syscall();
 386         if (netdev->ifalias)
 387                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 388         rtnl_unlock();
 389         return ret;
 390 }
 391 static DEVICE_ATTR_RW(ifalias);
 392
 393 static int change_group(struct net_device *dev, unsigned long new_group)
 394 {
 395         dev_set_group(dev, (int) new_group);
 396         return 0;
 397 }
 398
 399 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 400                            const char *buf, size_t len)
 401 {
 402         return netdev_store(dev, attr, buf, len, change_group);
 403 }
 404 NETDEVICE_SHOW(group, fmt_dec);
 405 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 406
 407 static ssize_t phys_port_id_show(struct device *dev,
 408                                  struct device_attribute *attr, char *buf)
 409 {
 410         struct net_device *netdev = to_net_dev(dev);
 411         ssize_t ret = -EINVAL;
 412
 413         if (!rtnl_trylock())
 414                 return restart_syscall();
 415
 416         if (dev_isalive(netdev)) {
 417                 struct netdev_phys_item_id ppid;
 418
 419                 ret = dev_get_phys_port_id(netdev, &ppid);
 420                 if (!ret)
 421                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 422         }
 423         rtnl_unlock();
 424
 425         return ret;
 426 }
 427 static DEVICE_ATTR_RO(phys_port_id);
 428
 429 static ssize_t phys_port_name_show(struct device *dev,
 430                                    struct device_attribute *attr, char *buf)
 431 {
 432         struct net_device *netdev = to_net_dev(dev);
 433         ssize_t ret = -EINVAL;
 434
 435         if (!rtnl_trylock())
 436                 return restart_syscall();
 437
 438         if (dev_isalive(netdev)) {
 439                 char name[IFNAMSIZ];
 440
 441                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 442                 if (!ret)
 443                         ret = sprintf(buf, "%s\n", name);
 444         }
 445         rtnl_unlock();
 446
 447         return ret;
 448 }
 449 static DEVICE_ATTR_RO(phys_port_name);
 450
 451 static ssize_t phys_switch_id_show(struct device *dev,
 452                                    struct device_attribute *attr, char *buf)
 453 {
 454         struct net_device *netdev = to_net_dev(dev);
 455         ssize_t ret = -EINVAL;
 456
 457         if (!rtnl_trylock())
 458                 return restart_syscall();
 459
 460         if (dev_isalive(netdev)) {
 461                 struct switchdev_attr attr = {
 462                         .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
 463                         .flags = SWITCHDEV_F_NO_RECURSE,
 464                 };
 465
 466                 ret = switchdev_port_attr_get(netdev, &attr);
 467                 if (!ret)
 468                         ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
 469                                       attr.u.ppid.id);
 470         }
 471         rtnl_unlock();
 472
 473         return ret;
 474 }
 475 static DEVICE_ATTR_RO(phys_switch_id);
 476
 477 static struct attribute *net_class_attrs[] = {
 478         &dev_attr_netdev_group.attr,
 479         &dev_attr_type.attr,
 480         &dev_attr_dev_id.attr,
 481         &dev_attr_dev_port.attr,
 482         &dev_attr_iflink.attr,
 483         &dev_attr_ifindex.attr,
 484         &dev_attr_name_assign_type.attr,
 485         &dev_attr_addr_assign_type.attr,
 486         &dev_attr_addr_len.attr,
 487         &dev_attr_link_mode.attr,
 488         &dev_attr_address.attr,
 489         &dev_attr_broadcast.attr,
 490         &dev_attr_speed.attr,
 491         &dev_attr_duplex.attr,
 492         &dev_attr_dormant.attr,
 493         &dev_attr_operstate.attr,
 494         &dev_attr_carrier_changes.attr,
 495         &dev_attr_ifalias.attr,
 496         &dev_attr_carrier.attr,
 497         &dev_attr_mtu.attr,
 498         &dev_attr_flags.attr,
 499         &dev_attr_tx_queue_len.attr,
 500         &dev_attr_gro_flush_timeout.attr,
 501         &dev_attr_phys_port_id.attr,
 502         &dev_attr_phys_port_name.attr,
 503         &dev_attr_phys_switch_id.attr,
 504         NULL,
 505 };
 506 ATTRIBUTE_GROUPS(net_class);
 507
 508 /* Show a given an attribute in the statistics group */
 509 static ssize_t netstat_show(const struct device *d,
 510                             struct device_attribute *attr, char *buf,
 511                             unsigned long offset)
 512 {
 513         struct net_device *dev = to_net_dev(d);
 514         ssize_t ret = -EINVAL;
 515
 516         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 517                         offset % sizeof(u64) != 0);
 518
 519         read_lock(&dev_base_lock);
 520         if (dev_isalive(dev)) {
 521                 struct rtnl_link_stats64 temp;
 522                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 523
 524                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 525         }
 526         read_unlock(&dev_base_lock);
 527         return ret;
 528 }
 529
 530 /* generate a read-only statistics attribute */
 531 #define NETSTAT_ENTRY(name)                                             \
 532 static ssize_t name##_show(struct device *d,                            \
 533                            struct device_attribute *attr, char *buf)    \
 534 {                                                                       \
 535         return netstat_show(d, attr, buf,                               \
 536                             offsetof(struct rtnl_link_stats64, name));  \
 537 }                                                                       \
 538 static DEVICE_ATTR_RO(name)
 539
 540 NETSTAT_ENTRY(rx_packets);
 541 NETSTAT_ENTRY(tx_packets);
 542 NETSTAT_ENTRY(rx_bytes);
 543 NETSTAT_ENTRY(tx_bytes);
 544 NETSTAT_ENTRY(rx_errors);
 545 NETSTAT_ENTRY(tx_errors);
 546 NETSTAT_ENTRY(rx_dropped);
 547 NETSTAT_ENTRY(tx_dropped);
 548 NETSTAT_ENTRY(multicast);
 549 NETSTAT_ENTRY(collisions);
 550 NETSTAT_ENTRY(rx_length_errors);
 551 NETSTAT_ENTRY(rx_over_errors);
 552 NETSTAT_ENTRY(rx_crc_errors);
 553 NETSTAT_ENTRY(rx_frame_errors);
 554 NETSTAT_ENTRY(rx_fifo_errors);
 555 NETSTAT_ENTRY(rx_missed_errors);
 556 NETSTAT_ENTRY(tx_aborted_errors);
 557 NETSTAT_ENTRY(tx_carrier_errors);
 558 NETSTAT_ENTRY(tx_fifo_errors);
 559 NETSTAT_ENTRY(tx_heartbeat_errors);
 560 NETSTAT_ENTRY(tx_window_errors);
 561 NETSTAT_ENTRY(rx_compressed);
 562 NETSTAT_ENTRY(tx_compressed);
 563
 564 static struct attribute *netstat_attrs[] = {
 565         &dev_attr_rx_packets.attr,
 566         &dev_attr_tx_packets.attr,
 567         &dev_attr_rx_bytes.attr,
 568         &dev_attr_tx_bytes.attr,
 569         &dev_attr_rx_errors.attr,
 570         &dev_attr_tx_errors.attr,
 571         &dev_attr_rx_dropped.attr,
 572         &dev_attr_tx_dropped.attr,
 573         &dev_attr_multicast.attr,
 574         &dev_attr_collisions.attr,
 575         &dev_attr_rx_length_errors.attr,
 576         &dev_attr_rx_over_errors.attr,
 577         &dev_attr_rx_crc_errors.attr,
 578         &dev_attr_rx_frame_errors.attr,
 579         &dev_attr_rx_fifo_errors.attr,
 580         &dev_attr_rx_missed_errors.attr,
 581         &dev_attr_tx_aborted_errors.attr,
 582         &dev_attr_tx_carrier_errors.attr,
 583         &dev_attr_tx_fifo_errors.attr,
 584         &dev_attr_tx_heartbeat_errors.attr,
 585         &dev_attr_tx_window_errors.attr,
 586         &dev_attr_rx_compressed.attr,
 587         &dev_attr_tx_compressed.attr,
 588         NULL
 589 };
 590
 591
 592 static struct attribute_group netstat_group = {
 593         .name  = "statistics",
 594         .attrs  = netstat_attrs,
 595 };
 596
 597 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 598 static struct attribute *wireless_attrs[] = {
 599         NULL
 600 };
 601
 602 static struct attribute_group wireless_group = {
 603         .name = "wireless",
 604         .attrs = wireless_attrs,
 605 };
 606 #endif
 607
 608 #else /* CONFIG_SYSFS */
 609 #define net_class_groups        NULL
 610 #endif /* CONFIG_SYSFS */
 611
 612 #ifdef CONFIG_SYSFS
 613 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 614     struct rx_queue_attribute, attr)
 615
 616 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 617
 618 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 619                                   char *buf)
 620 {
 621         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 622         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 623
 624         if (!attribute->show)
 625                 return -EIO;
 626
 627         return attribute->show(queue, attribute, buf);
 628 }
 629
 630 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 631                                    const char *buf, size_t count)
 632 {
 633         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 634         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 635
 636         if (!attribute->store)
 637                 return -EIO;
 638
 639         return attribute->store(queue, attribute, buf, count);
 640 }
 641
 642 static const struct sysfs_ops rx_queue_sysfs_ops = {
 643         .show = rx_queue_attr_show,
 644         .store = rx_queue_attr_store,
 645 };
 646
 647 #ifdef CONFIG_RPS
 648 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 649                             struct rx_queue_attribute *attribute, char *buf)
 650 {
 651         struct rps_map *map;
 652         cpumask_var_t mask;
 653         int i, len;
 654
 655         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 656                 return -ENOMEM;
 657
 658         rcu_read_lock();
 659         map = rcu_dereference(queue->rps_map);
 660         if (map)
 661                 for (i = 0; i < map->len; i++)
 662                         cpumask_set_cpu(map->cpus[i], mask);
 663
 664         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 665         rcu_read_unlock();
 666         free_cpumask_var(mask);
 667
 668         return len < PAGE_SIZE ? len : -EINVAL;
 669 }
 670
 671 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 672                       struct rx_queue_attribute *attribute,
 673                       const char *buf, size_t len)
 674 {
 675         struct rps_map *old_map, *map;
 676         cpumask_var_t mask;
 677         int err, cpu, i;
 678         static DEFINE_SPINLOCK(rps_map_lock);
 679
 680         if (!capable(CAP_NET_ADMIN))
 681                 return -EPERM;
 682
 683         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 684                 return -ENOMEM;
 685
 686         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 687         if (err) {
 688                 free_cpumask_var(mask);
 689                 return err;
 690         }
 691
 692         map = kzalloc(max_t(unsigned int,
 693             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 694             GFP_KERNEL);
 695         if (!map) {
 696                 free_cpumask_var(mask);
 697                 return -ENOMEM;
 698         }
 699
 700         i = 0;
 701         for_each_cpu_and(cpu, mask, cpu_online_mask)
 702                 map->cpus[i++] = cpu;
 703
 704         if (i)
 705                 map->len = i;
 706         else {
 707                 kfree(map);
 708                 map = NULL;
 709         }
 710
 711         spin_lock(&rps_map_lock);
 712         old_map = rcu_dereference_protected(queue->rps_map,
 713                                             lockdep_is_held(&rps_map_lock));
 714         rcu_assign_pointer(queue->rps_map, map);
 715         spin_unlock(&rps_map_lock);
 716
 717         if (map)
 718                 static_key_slow_inc(&rps_needed);
 719         if (old_map) {
 720                 kfree_rcu(old_map, rcu);
 721                 static_key_slow_dec(&rps_needed);
 722         }
 723         free_cpumask_var(mask);
 724         return len;
 725 }
 726
 727 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 728                                            struct rx_queue_attribute *attr,
 729                                            char *buf)
 730 {
 731         struct rps_dev_flow_table *flow_table;
 732         unsigned long val = 0;
 733
 734         rcu_read_lock();
 735         flow_table = rcu_dereference(queue->rps_flow_table);
 736         if (flow_table)
 737                 val = (unsigned long)flow_table->mask + 1;
 738         rcu_read_unlock();
 739
 740         return sprintf(buf, "%lu\n", val);
 741 }
 742
 743 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 744 {
 745         struct rps_dev_flow_table *table = container_of(rcu,
 746             struct rps_dev_flow_table, rcu);
 747         vfree(table);
 748 }
 749
 750 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 751                                      struct rx_queue_attribute *attr,
 752                                      const char *buf, size_t len)
 753 {
 754         unsigned long mask, count;
 755         struct rps_dev_flow_table *table, *old_table;
 756         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 757         int rc;
 758
 759         if (!capable(CAP_NET_ADMIN))
 760                 return -EPERM;
 761
 762         rc = kstrtoul(buf, 0, &count);
 763         if (rc < 0)
 764                 return rc;
 765
 766         if (count) {
 767                 mask = count - 1;
 768                 /* mask = roundup_pow_of_two(count) - 1;
 769                  * without overflows...
 770                  */
 771                 while ((mask | (mask >> 1)) != mask)
 772                         mask |= (mask >> 1);
 773                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 774                  * and on 32bit arches, must check
 775                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 776                  */
 777 #if BITS_PER_LONG > 32
 778                 if (mask > (unsigned long)(u32)mask)
 779                         return -EINVAL;
 780 #else
 781                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 782                                 / sizeof(struct rps_dev_flow)) {
 783                         /* Enforce a limit to prevent overflow */
 784                         return -EINVAL;
 785                 }
 786 #endif
 787                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 788                 if (!table)
 789                         return -ENOMEM;
 790
 791                 table->mask = mask;
 792                 for (count = 0; count <= mask; count++)
 793                         table->flows[count].cpu = RPS_NO_CPU;
 794         } else
 795                 table = NULL;
 796
 797         spin_lock(&rps_dev_flow_lock);
 798         old_table = rcu_dereference_protected(queue->rps_flow_table,
 799                                               lockdep_is_held(&rps_dev_flow_lock));
 800         rcu_assign_pointer(queue->rps_flow_table, table);
 801         spin_unlock(&rps_dev_flow_lock);
 802
 803         if (old_table)
 804                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 805
 806         return len;
 807 }
 808
 809 static struct rx_queue_attribute rps_cpus_attribute =
 810         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 811
 812
 813 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 814         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 815             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 816 #endif /* CONFIG_RPS */
 817
 818 static struct attribute *rx_queue_default_attrs[] = {
 819 #ifdef CONFIG_RPS
 820         &rps_cpus_attribute.attr,
 821         &rps_dev_flow_table_cnt_attribute.attr,
 822 #endif
 823         NULL
 824 };
 825
 826 static void rx_queue_release(struct kobject *kobj)
 827 {
 828         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 829 #ifdef CONFIG_RPS
 830         struct rps_map *map;
 831         struct rps_dev_flow_table *flow_table;
 832
 833
 834         map = rcu_dereference_protected(queue->rps_map, 1);
 835         if (map) {
 836                 RCU_INIT_POINTER(queue->rps_map, NULL);
 837                 kfree_rcu(map, rcu);
 838         }
 839
 840         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 841         if (flow_table) {
 842                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 843                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 844         }
 845 #endif
 846
 847         memset(kobj, 0, sizeof(*kobj));
 848         dev_put(queue->dev);
 849 }
 850
 851 static const void *rx_queue_namespace(struct kobject *kobj)
 852 {
 853         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 854         struct device *dev = &queue->dev->dev;
 855         const void *ns = NULL;
 856
 857         if (dev->class && dev->class->ns_type)
 858                 ns = dev->class->namespace(dev);
 859
 860         return ns;
 861 }
 862
 863 static struct kobj_type rx_queue_ktype = {
 864         .sysfs_ops = &rx_queue_sysfs_ops,
 865         .release = rx_queue_release,
 866         .default_attrs = rx_queue_default_attrs,
 867         .namespace = rx_queue_namespace
 868 };
 869
 870 static int rx_queue_add_kobject(struct net_device *dev, int index)
 871 {
 872         struct netdev_rx_queue *queue = dev->_rx + index;
 873         struct kobject *kobj = &queue->kobj;
 874         int error = 0;
 875
 876         kobj->kset = dev->queues_kset;
 877         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 878             "rx-%u", index);
 879         if (error)
 880                 goto exit;
 881
 882         if (dev->sysfs_rx_queue_group) {
 883                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 884                 if (error)
 885                         goto exit;
 886         }
 887
 888         kobject_uevent(kobj, KOBJ_ADD);
 889         dev_hold(queue->dev);
 890
 891         return error;
 892 exit:
 893         kobject_put(kobj);
 894         return error;
 895 }
 896 #endif /* CONFIG_SYSFS */
 897
 898 int
 899 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 900 {
 901 #ifdef CONFIG_SYSFS
 902         int i;
 903         int error = 0;
 904
 905 #ifndef CONFIG_RPS
 906         if (!dev->sysfs_rx_queue_group)
 907                 return 0;
 908 #endif
 909         for (i = old_num; i < new_num; i++) {
 910                 error = rx_queue_add_kobject(dev, i);
 911                 if (error) {
 912                         new_num = old_num;
 913                         break;
 914                 }
 915         }
 916
 917         while (--i >= new_num) {
 918                 if (dev->sysfs_rx_queue_group)
 919                         sysfs_remove_group(&dev->_rx[i].kobj,
 920                                            dev->sysfs_rx_queue_group);
 921                 kobject_put(&dev->_rx[i].kobj);
 922         }
 923
 924         return error;
 925 #else
 926         return 0;
 927 #endif
 928 }
 929
 930 #ifdef CONFIG_SYSFS
 931 /*
 932  * netdev_queue sysfs structures and functions.
 933  */
 934 struct netdev_queue_attribute {
 935         struct attribute attr;
 936         ssize_t (*show)(struct netdev_queue *queue,
 937             struct netdev_queue_attribute *attr, char *buf);
 938         ssize_t (*store)(struct netdev_queue *queue,
 939             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 940 };
 941 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 942     struct netdev_queue_attribute, attr)
 943
 944 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 945
 946 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 947                                       struct attribute *attr, char *buf)
 948 {
 949         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 950         struct netdev_queue *queue = to_netdev_queue(kobj);
 951
 952         if (!attribute->show)
 953                 return -EIO;
 954
 955         return attribute->show(queue, attribute, buf);
 956 }
 957
 958 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 959                                        struct attribute *attr,
 960                                        const char *buf, size_t count)
 961 {
 962         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 963         struct netdev_queue *queue = to_netdev_queue(kobj);
 964
 965         if (!attribute->store)
 966                 return -EIO;
 967
 968         return attribute->store(queue, attribute, buf, count);
 969 }
 970
 971 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 972         .show = netdev_queue_attr_show,
 973         .store = netdev_queue_attr_store,
 974 };
 975
 976 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 977                                   struct netdev_queue_attribute *attribute,
 978                                   char *buf)
 979 {
 980         unsigned long trans_timeout;
 981
 982         spin_lock_irq(&queue->_xmit_lock);
 983         trans_timeout = queue->trans_timeout;
 984         spin_unlock_irq(&queue->_xmit_lock);
 985
 986         return sprintf(buf, "%lu", trans_timeout);
 987 }
 988
 989 #ifdef CONFIG_XPS
 990 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 991 {
 992         struct net_device *dev = queue->dev;
 993         int i;
 994
 995         for (i = 0; i < dev->num_tx_queues; i++)
 996                 if (queue == &dev->_tx[i])
 997                         break;
 998
 999         BUG_ON(i >= dev->num_tx_queues);
1000
1001         return i;
1002 }
1003
1004 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1005                                struct netdev_queue_attribute *attribute,
1006                                char *buf)
1007 {
1008         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1009 }
1010
1011 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1012                               struct netdev_queue_attribute *attribute,
1013                               const char *buf, size_t len)
1014 {
1015         struct net_device *dev = queue->dev;
1016         int err, index = get_netdev_queue_index(queue);
1017         u32 rate = 0;
1018
1019         err = kstrtou32(buf, 10, &rate);
1020         if (err < 0)
1021                 return err;
1022
1023         if (!rtnl_trylock())
1024                 return restart_syscall();
1025
1026         err = -EOPNOTSUPP;
1027         if (dev->netdev_ops->ndo_set_tx_maxrate)
1028                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1029
1030         rtnl_unlock();
1031         if (!err) {
1032                 queue->tx_maxrate = rate;
1033                 return len;
1034         }
1035         return err;
1036 }
1037
1038 static struct netdev_queue_attribute queue_tx_maxrate =
1039         __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1040                show_tx_maxrate, set_tx_maxrate);
1041 #endif
1042
1043 static struct netdev_queue_attribute queue_trans_timeout =
1044         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1045
1046 #ifdef CONFIG_BQL
1047 /*
1048  * Byte queue limits sysfs structures and functions.
1049  */
1050 static ssize_t bql_show(char *buf, unsigned int value)
1051 {
1052         return sprintf(buf, "%u\n", value);
1053 }
1054
1055 static ssize_t bql_set(const char *buf, const size_t count,
1056                        unsigned int *pvalue)
1057 {
1058         unsigned int value;
1059         int err;
1060
1061         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1062                 value = DQL_MAX_LIMIT;
1063         else {
1064                 err = kstrtouint(buf, 10, &value);
1065                 if (err < 0)
1066                         return err;
1067                 if (value > DQL_MAX_LIMIT)
1068                         return -EINVAL;
1069         }
1070
1071         *pvalue = value;
1072
1073         return count;
1074 }
1075
1076 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1077                                   struct netdev_queue_attribute *attr,
1078                                   char *buf)
1079 {
1080         struct dql *dql = &queue->dql;
1081
1082         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1083 }
1084
1085 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1086                                  struct netdev_queue_attribute *attribute,
1087                                  const char *buf, size_t len)
1088 {
1089         struct dql *dql = &queue->dql;
1090         unsigned int value;
1091         int err;
1092
1093         err = kstrtouint(buf, 10, &value);
1094         if (err < 0)
1095                 return err;
1096
1097         dql->slack_hold_time = msecs_to_jiffies(value);
1098
1099         return len;
1100 }
1101
1102 static struct netdev_queue_attribute bql_hold_time_attribute =
1103         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1104             bql_set_hold_time);
1105
1106 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1107                                  struct netdev_queue_attribute *attr,
1108                                  char *buf)
1109 {
1110         struct dql *dql = &queue->dql;
1111
1112         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1113 }
1114
1115 static struct netdev_queue_attribute bql_inflight_attribute =
1116         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1117
1118 #define BQL_ATTR(NAME, FIELD)                                           \
1119 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1120                                  struct netdev_queue_attribute *attr,   \
1121                                  char *buf)                             \
1122 {                                                                       \
1123         return bql_show(buf, queue->dql.FIELD);                         \
1124 }                                                                       \
1125                                                                         \
1126 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1127                                 struct netdev_queue_attribute *attr,    \
1128                                 const char *buf, size_t len)            \
1129 {                                                                       \
1130         return bql_set(buf, len, &queue->dql.FIELD);                    \
1131 }                                                                       \
1132                                                                         \
1133 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1134         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1135             bql_set_ ## NAME);
1136
1137 BQL_ATTR(limit, limit)
1138 BQL_ATTR(limit_max, max_limit)
1139 BQL_ATTR(limit_min, min_limit)
1140
1141 static struct attribute *dql_attrs[] = {
1142         &bql_limit_attribute.attr,
1143         &bql_limit_max_attribute.attr,
1144         &bql_limit_min_attribute.attr,
1145         &bql_hold_time_attribute.attr,
1146         &bql_inflight_attribute.attr,
1147         NULL
1148 };
1149
1150 static struct attribute_group dql_group = {
1151         .name  = "byte_queue_limits",
1152         .attrs  = dql_attrs,
1153 };
1154 #endif /* CONFIG_BQL */
1155
1156 #ifdef CONFIG_XPS
1157 static ssize_t show_xps_map(struct netdev_queue *queue,
1158                             struct netdev_queue_attribute *attribute, char *buf)
1159 {
1160         struct net_device *dev = queue->dev;
1161         struct xps_dev_maps *dev_maps;
1162         cpumask_var_t mask;
1163         unsigned long index;
1164         int i, len;
1165
1166         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1167                 return -ENOMEM;
1168
1169         index = get_netdev_queue_index(queue);
1170
1171         rcu_read_lock();
1172         dev_maps = rcu_dereference(dev->xps_maps);
1173         if (dev_maps) {
1174                 for_each_possible_cpu(i) {
1175                         struct xps_map *map =
1176                             rcu_dereference(dev_maps->cpu_map[i]);
1177                         if (map) {
1178                                 int j;
1179                                 for (j = 0; j < map->len; j++) {
1180                                         if (map->queues[j] == index) {
1181                                                 cpumask_set_cpu(i, mask);
1182                                                 break;
1183                                         }
1184                                 }
1185                         }
1186                 }
1187         }
1188         rcu_read_unlock();
1189
1190         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1191         free_cpumask_var(mask);
1192         return len < PAGE_SIZE ? len : -EINVAL;
1193 }
1194
1195 static ssize_t store_xps_map(struct netdev_queue *queue,
1196                       struct netdev_queue_attribute *attribute,
1197                       const char *buf, size_t len)
1198 {
1199         struct net_device *dev = queue->dev;
1200         unsigned long index;
1201         cpumask_var_t mask;
1202         int err;
1203
1204         if (!capable(CAP_NET_ADMIN))
1205                 return -EPERM;
1206
1207         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1208                 return -ENOMEM;
1209
1210         index = get_netdev_queue_index(queue);
1211
1212         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1213         if (err) {
1214                 free_cpumask_var(mask);
1215                 return err;
1216         }
1217
1218         err = netif_set_xps_queue(dev, mask, index);
1219
1220         free_cpumask_var(mask);
1221
1222         return err ? : len;
1223 }
1224
1225 static struct netdev_queue_attribute xps_cpus_attribute =
1226     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1227 #endif /* CONFIG_XPS */
1228
1229 static struct attribute *netdev_queue_default_attrs[] = {
1230         &queue_trans_timeout.attr,
1231 #ifdef CONFIG_XPS
1232         &xps_cpus_attribute.attr,
1233         &queue_tx_maxrate.attr,
1234 #endif
1235         NULL
1236 };
1237
1238 static void netdev_queue_release(struct kobject *kobj)
1239 {
1240         struct netdev_queue *queue = to_netdev_queue(kobj);
1241
1242         memset(kobj, 0, sizeof(*kobj));
1243         dev_put(queue->dev);
1244 }
1245
1246 static const void *netdev_queue_namespace(struct kobject *kobj)
1247 {
1248         struct netdev_queue *queue = to_netdev_queue(kobj);
1249         struct device *dev = &queue->dev->dev;
1250         const void *ns = NULL;
1251
1252         if (dev->class && dev->class->ns_type)
1253                 ns = dev->class->namespace(dev);
1254
1255         return ns;
1256 }
1257
1258 static struct kobj_type netdev_queue_ktype = {
1259         .sysfs_ops = &netdev_queue_sysfs_ops,
1260         .release = netdev_queue_release,
1261         .default_attrs = netdev_queue_default_attrs,
1262         .namespace = netdev_queue_namespace,
1263 };
1264
1265 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1266 {
1267         struct netdev_queue *queue = dev->_tx + index;
1268         struct kobject *kobj = &queue->kobj;
1269         int error = 0;
1270
1271         kobj->kset = dev->queues_kset;
1272         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1273             "tx-%u", index);
1274         if (error)
1275                 goto exit;
1276
1277 #ifdef CONFIG_BQL
1278         error = sysfs_create_group(kobj, &dql_group);
1279         if (error)
1280                 goto exit;
1281 #endif
1282
1283         kobject_uevent(kobj, KOBJ_ADD);
1284         dev_hold(queue->dev);
1285
1286         return 0;
1287 exit:
1288         kobject_put(kobj);
1289         return error;
1290 }
1291 #endif /* CONFIG_SYSFS */
1292
1293 int
1294 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1295 {
1296 #ifdef CONFIG_SYSFS
1297         int i;
1298         int error = 0;
1299
1300         for (i = old_num; i < new_num; i++) {
1301                 error = netdev_queue_add_kobject(dev, i);
1302                 if (error) {
1303                         new_num = old_num;
1304                         break;
1305                 }
1306         }
1307
1308         while (--i >= new_num) {
1309                 struct netdev_queue *queue = dev->_tx + i;
1310
1311 #ifdef CONFIG_BQL
1312                 sysfs_remove_group(&queue->kobj, &dql_group);
1313 #endif
1314                 kobject_put(&queue->kobj);
1315         }
1316
1317         return error;
1318 #else
1319         return 0;
1320 #endif /* CONFIG_SYSFS */
1321 }
1322
1323 static int register_queue_kobjects(struct net_device *dev)
1324 {
1325         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1326
1327 #ifdef CONFIG_SYSFS
1328         dev->queues_kset = kset_create_and_add("queues",
1329             NULL, &dev->dev.kobj);
1330         if (!dev->queues_kset)
1331                 return -ENOMEM;
1332         real_rx = dev->real_num_rx_queues;
1333 #endif
1334         real_tx = dev->real_num_tx_queues;
1335
1336         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1337         if (error)
1338                 goto error;
1339         rxq = real_rx;
1340
1341         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1342         if (error)
1343                 goto error;
1344         txq = real_tx;
1345
1346         return 0;
1347
1348 error:
1349         netdev_queue_update_kobjects(dev, txq, 0);
1350         net_rx_queue_update_kobjects(dev, rxq, 0);
1351         return error;
1352 }
1353
1354 static void remove_queue_kobjects(struct net_device *dev)
1355 {
1356         int real_rx = 0, real_tx = 0;
1357
1358 #ifdef CONFIG_SYSFS
1359         real_rx = dev->real_num_rx_queues;
1360 #endif
1361         real_tx = dev->real_num_tx_queues;
1362
1363         net_rx_queue_update_kobjects(dev, real_rx, 0);
1364         netdev_queue_update_kobjects(dev, real_tx, 0);
1365 #ifdef CONFIG_SYSFS
1366         kset_unregister(dev->queues_kset);
1367 #endif
1368 }
1369
1370 static bool net_current_may_mount(void)
1371 {
1372         struct net *net = current->nsproxy->net_ns;
1373
1374         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1375 }
1376
1377 static void *net_grab_current_ns(void)
1378 {
1379         struct net *ns = current->nsproxy->net_ns;
1380 #ifdef CONFIG_NET_NS
1381         if (ns)
1382                 atomic_inc(&ns->passive);
1383 #endif
1384         return ns;
1385 }
1386
1387 static const void *net_initial_ns(void)
1388 {
1389         return &init_net;
1390 }
1391
1392 static const void *net_netlink_ns(struct sock *sk)
1393 {
1394         return sock_net(sk);
1395 }
1396
1397 struct kobj_ns_type_operations net_ns_type_operations = {
1398         .type = KOBJ_NS_TYPE_NET,
1399         .current_may_mount = net_current_may_mount,
1400         .grab_current_ns = net_grab_current_ns,
1401         .netlink_ns = net_netlink_ns,
1402         .initial_ns = net_initial_ns,
1403         .drop_ns = net_drop_ns,
1404 };
1405 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1406
1407 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1408 {
1409         struct net_device *dev = to_net_dev(d);
1410         int retval;
1411
1412         /* pass interface to uevent. */
1413         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1414         if (retval)
1415                 goto exit;
1416
1417         /* pass ifindex to uevent.
1418          * ifindex is useful as it won't change (interface name may change)
1419          * and is what RtNetlink uses natively. */
1420         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1421
1422 exit:
1423         return retval;
1424 }
1425
1426 /*
1427  *      netdev_release -- destroy and free a dead device.
1428  *      Called when last reference to device kobject is gone.
1429  */
1430 static void netdev_release(struct device *d)
1431 {
1432         struct net_device *dev = to_net_dev(d);
1433
1434         BUG_ON(dev->reg_state != NETREG_RELEASED);
1435
1436         kfree(dev->ifalias);
1437         netdev_freemem(dev);
1438 }
1439
1440 static const void *net_namespace(struct device *d)
1441 {
1442         struct net_device *dev;
1443         dev = container_of(d, struct net_device, dev);
1444         return dev_net(dev);
1445 }
1446
1447 static struct class net_class = {
1448         .name = "net",
1449         .dev_release = netdev_release,
1450         .dev_groups = net_class_groups,
1451         .dev_uevent = netdev_uevent,
1452         .ns_type = &net_ns_type_operations,
1453         .namespace = net_namespace,
1454 };
1455
1456 #ifdef CONFIG_OF_NET
1457 static int of_dev_node_match(struct device *dev, const void *data)
1458 {
1459         int ret = 0;
1460
1461         if (dev->parent)
1462                 ret = dev->parent->of_node == data;
1463
1464         return ret == 0 ? dev->of_node == data : ret;
1465 }
1466
1467 struct net_device *of_find_net_device_by_node(struct device_node *np)
1468 {
1469         struct device *dev;
1470
1471         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1472         if (!dev)
1473                 return NULL;
1474
1475         return to_net_dev(dev);
1476 }
1477 EXPORT_SYMBOL(of_find_net_device_by_node);
1478 #endif
1479
1480 /* Delete sysfs entries but hold kobject reference until after all
1481  * netdev references are gone.
1482  */
1483 void netdev_unregister_kobject(struct net_device *ndev)
1484 {
1485         struct device *dev = &(ndev->dev);
1486
1487         kobject_get(&dev->kobj);
1488
1489         remove_queue_kobjects(ndev);
1490
1491         pm_runtime_set_memalloc_noio(dev, false);
1492
1493         device_del(dev);
1494 }
1495
1496 /* Create sysfs entries for network device. */
1497 int netdev_register_kobject(struct net_device *ndev)
1498 {
1499         struct device *dev = &(ndev->dev);
1500         const struct attribute_group **groups = ndev->sysfs_groups;
1501         int error = 0;
1502
1503         device_initialize(dev);
1504         dev->class = &net_class;
1505         dev->platform_data = ndev;
1506         dev->groups = groups;
1507
1508         dev_set_name(dev, "%s", ndev->name);
1509
1510 #ifdef CONFIG_SYSFS
1511         /* Allow for a device specific group */
1512         if (*groups)
1513                 groups++;
1514
1515         *groups++ = &netstat_group;
1516
1517 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1518         if (ndev->ieee80211_ptr)
1519                 *groups++ = &wireless_group;
1520 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1521         else if (ndev->wireless_handlers)
1522                 *groups++ = &wireless_group;
1523 #endif
1524 #endif
1525 #endif /* CONFIG_SYSFS */
1526
1527         error = device_add(dev);
1528         if (error)
1529                 return error;
1530
1531         error = register_queue_kobjects(ndev);
1532         if (error) {
1533                 device_del(dev);
1534                 return error;
1535         }
1536
1537         pm_runtime_set_memalloc_noio(dev, true);
1538
1539         return error;
1540 }
1541
1542 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1543                                 const void *ns)
1544 {
1545         return class_create_file_ns(&net_class, class_attr, ns);
1546 }
1547 EXPORT_SYMBOL(netdev_class_create_file_ns);
1548
1549 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1550                                  const void *ns)
1551 {
1552         class_remove_file_ns(&net_class, class_attr, ns);
1553 }
1554 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1555
1556 int __init netdev_kobject_init(void)
1557 {
1558         kobj_ns_type_register(&net_ns_type_operations);
1559         return class_register(&net_class);
1560 }