net/core/net-sysfs.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net-sysfs.c - network device class and attributes
   4  *
   5  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   6  */
   7
   8 #include <linux/capability.h>
   9 #include <linux/kernel.h>
  10 #include <linux/netdevice.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/nsproxy.h>
  15 #include <net/sock.h>
  16 #include <net/net_namespace.h>
  17 #include <linux/rtnetlink.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/export.h>
  20 #include <linux/jiffies.h>
  21 #include <linux/pm_runtime.h>
  22 #include <linux/of.h>
  23 #include <linux/of_net.h>
  24 #include <linux/cpu.h>
  25
  26 #include "net-sysfs.h"
  27
  28 #ifdef CONFIG_SYSFS
  29 static const char fmt_hex[] = "%#x\n";
  30 static const char fmt_dec[] = "%d\n";
  31 static const char fmt_ulong[] = "%lu\n";
  32 static const char fmt_u64[] = "%llu\n";
  33
  34 static inline int dev_isalive(const struct net_device *dev)
  35 {
  36         return dev->reg_state <= NETREG_REGISTERED;
  37 }
  38
  39 /* use same locking rules as GIF* ioctl's */
  40 static ssize_t netdev_show(const struct device *dev,
  41                            struct device_attribute *attr, char *buf,
  42                            ssize_t (*format)(const struct net_device *, char *))
  43 {
  44         struct net_device *ndev = to_net_dev(dev);
  45         ssize_t ret = -EINVAL;
  46
  47         read_lock(&dev_base_lock);
  48         if (dev_isalive(ndev))
  49                 ret = (*format)(ndev, buf);
  50         read_unlock(&dev_base_lock);
  51
  52         return ret;
  53 }
  54
  55 /* generate a show function for simple field */
  56 #define NETDEVICE_SHOW(field, format_string)                            \
  57 static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  58 {                                                                       \
  59         return sprintf(buf, format_string, dev->field);                 \
  60 }                                                                       \
  61 static ssize_t field##_show(struct device *dev,                         \
  62                             struct device_attribute *attr, char *buf)   \
  63 {                                                                       \
  64         return netdev_show(dev, attr, buf, format_##field);             \
  65 }                                                                       \
  66
  67 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  68 NETDEVICE_SHOW(field, format_string);                                   \
  69 static DEVICE_ATTR_RO(field)
  70
  71 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  72 NETDEVICE_SHOW(field, format_string);                                   \
  73 static DEVICE_ATTR_RW(field)
  74
  75 /* use same locking and permission rules as SIF* ioctl's */
  76 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  77                             const char *buf, size_t len,
  78                             int (*set)(struct net_device *, unsigned long))
  79 {
  80         struct net_device *netdev = to_net_dev(dev);
  81         struct net *net = dev_net(netdev);
  82         unsigned long new;
  83         int ret = -EINVAL;
  84
  85         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  86                 return -EPERM;
  87
  88         ret = kstrtoul(buf, 0, &new);
  89         if (ret)
  90                 goto err;
  91
  92         if (!rtnl_trylock())
  93                 return restart_syscall();
  94
  95         if (dev_isalive(netdev)) {
  96                 ret = (*set)(netdev, new);
  97                 if (ret == 0)
  98                         ret = len;
  99         }
 100         rtnl_unlock();
 101  err:
 102         return ret;
 103 }
 104
 105 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 106 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 107 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 108 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 109 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 110 NETDEVICE_SHOW_RO(type, fmt_dec);
 111 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 112
 113 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 114                            char *buf)
 115 {
 116         struct net_device *ndev = to_net_dev(dev);
 117
 118         return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
 119 }
 120 static DEVICE_ATTR_RO(iflink);
 121
 122 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 123 {
 124         return sprintf(buf, fmt_dec, dev->name_assign_type);
 125 }
 126
 127 static ssize_t name_assign_type_show(struct device *dev,
 128                                      struct device_attribute *attr,
 129                                      char *buf)
 130 {
 131         struct net_device *ndev = to_net_dev(dev);
 132         ssize_t ret = -EINVAL;
 133
 134         if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 135                 ret = netdev_show(dev, attr, buf, format_name_assign_type);
 136
 137         return ret;
 138 }
 139 static DEVICE_ATTR_RO(name_assign_type);
 140
 141 /* use same locking rules as GIFHWADDR ioctl's */
 142 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 143                             char *buf)
 144 {
 145         struct net_device *ndev = to_net_dev(dev);
 146         ssize_t ret = -EINVAL;
 147
 148         read_lock(&dev_base_lock);
 149         if (dev_isalive(ndev))
 150                 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 151         read_unlock(&dev_base_lock);
 152         return ret;
 153 }
 154 static DEVICE_ATTR_RO(address);
 155
 156 static ssize_t broadcast_show(struct device *dev,
 157                               struct device_attribute *attr, char *buf)
 158 {
 159         struct net_device *ndev = to_net_dev(dev);
 160
 161         if (dev_isalive(ndev))
 162                 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 163         return -EINVAL;
 164 }
 165 static DEVICE_ATTR_RO(broadcast);
 166
 167 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 168 {
 169         if (!netif_running(dev))
 170                 return -EINVAL;
 171         return dev_change_carrier(dev, (bool)new_carrier);
 172 }
 173
 174 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 175                              const char *buf, size_t len)
 176 {
 177         return netdev_store(dev, attr, buf, len, change_carrier);
 178 }
 179
 180 static ssize_t carrier_show(struct device *dev,
 181                             struct device_attribute *attr, char *buf)
 182 {
 183         struct net_device *netdev = to_net_dev(dev);
 184
 185         if (netif_running(netdev))
 186                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 187
 188         return -EINVAL;
 189 }
 190 static DEVICE_ATTR_RW(carrier);
 191
 192 static ssize_t speed_show(struct device *dev,
 193                           struct device_attribute *attr, char *buf)
 194 {
 195         struct net_device *netdev = to_net_dev(dev);
 196         int ret = -EINVAL;
 197
 198         if (!rtnl_trylock())
 199                 return restart_syscall();
 200
 201         if (netif_running(netdev)) {
 202                 struct ethtool_link_ksettings cmd;
 203
 204                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
 205                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
 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_link_ksettings cmd;
 223
 224                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
 225                         const char *duplex;
 226
 227                         switch (cmd.base.duplex) {
 228                         case DUPLEX_HALF:
 229                                 duplex = "half";
 230                                 break;
 231                         case DUPLEX_FULL:
 232                                 duplex = "full";
 233                                 break;
 234                         default:
 235                                 duplex = "unknown";
 236                                 break;
 237                         }
 238                         ret = sprintf(buf, "%s\n", duplex);
 239                 }
 240         }
 241         rtnl_unlock();
 242         return ret;
 243 }
 244 static DEVICE_ATTR_RO(duplex);
 245
 246 static ssize_t dormant_show(struct device *dev,
 247                             struct device_attribute *attr, char *buf)
 248 {
 249         struct net_device *netdev = to_net_dev(dev);
 250
 251         if (netif_running(netdev))
 252                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 253
 254         return -EINVAL;
 255 }
 256 static DEVICE_ATTR_RO(dormant);
 257
 258 static const char *const operstates[] = {
 259         "unknown",
 260         "notpresent", /* currently unused */
 261         "down",
 262         "lowerlayerdown",
 263         "testing", /* currently unused */
 264         "dormant",
 265         "up"
 266 };
 267
 268 static ssize_t operstate_show(struct device *dev,
 269                               struct device_attribute *attr, char *buf)
 270 {
 271         const struct net_device *netdev = to_net_dev(dev);
 272         unsigned char operstate;
 273
 274         read_lock(&dev_base_lock);
 275         operstate = netdev->operstate;
 276         if (!netif_running(netdev))
 277                 operstate = IF_OPER_DOWN;
 278         read_unlock(&dev_base_lock);
 279
 280         if (operstate >= ARRAY_SIZE(operstates))
 281                 return -EINVAL; /* should not happen */
 282
 283         return sprintf(buf, "%s\n", operstates[operstate]);
 284 }
 285 static DEVICE_ATTR_RO(operstate);
 286
 287 static ssize_t carrier_changes_show(struct device *dev,
 288                                     struct device_attribute *attr,
 289                                     char *buf)
 290 {
 291         struct net_device *netdev = to_net_dev(dev);
 292
 293         return sprintf(buf, fmt_dec,
 294                        atomic_read(&netdev->carrier_up_count) +
 295                        atomic_read(&netdev->carrier_down_count));
 296 }
 297 static DEVICE_ATTR_RO(carrier_changes);
 298
 299 static ssize_t carrier_up_count_show(struct device *dev,
 300                                      struct device_attribute *attr,
 301                                      char *buf)
 302 {
 303         struct net_device *netdev = to_net_dev(dev);
 304
 305         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
 306 }
 307 static DEVICE_ATTR_RO(carrier_up_count);
 308
 309 static ssize_t carrier_down_count_show(struct device *dev,
 310                                        struct device_attribute *attr,
 311                                        char *buf)
 312 {
 313         struct net_device *netdev = to_net_dev(dev);
 314
 315         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
 316 }
 317 static DEVICE_ATTR_RO(carrier_down_count);
 318
 319 /* read-write attributes */
 320
 321 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 322 {
 323         return dev_set_mtu(dev, (int)new_mtu);
 324 }
 325
 326 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 327                          const char *buf, size_t len)
 328 {
 329         return netdev_store(dev, attr, buf, len, change_mtu);
 330 }
 331 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 332
 333 static int change_flags(struct net_device *dev, unsigned long new_flags)
 334 {
 335         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
 336 }
 337
 338 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 339                            const char *buf, size_t len)
 340 {
 341         return netdev_store(dev, attr, buf, len, change_flags);
 342 }
 343 NETDEVICE_SHOW_RW(flags, fmt_hex);
 344
 345 static ssize_t tx_queue_len_store(struct device *dev,
 346                                   struct device_attribute *attr,
 347                                   const char *buf, size_t len)
 348 {
 349         if (!capable(CAP_NET_ADMIN))
 350                 return -EPERM;
 351
 352         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
 353 }
 354 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 355
 356 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 357 {
 358         dev->gro_flush_timeout = val;
 359         return 0;
 360 }
 361
 362 static ssize_t gro_flush_timeout_store(struct device *dev,
 363                                        struct device_attribute *attr,
 364                                        const char *buf, size_t len)
 365 {
 366         if (!capable(CAP_NET_ADMIN))
 367                 return -EPERM;
 368
 369         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 370 }
 371 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 372
 373 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 374                              const char *buf, size_t len)
 375 {
 376         struct net_device *netdev = to_net_dev(dev);
 377         struct net *net = dev_net(netdev);
 378         size_t count = len;
 379         ssize_t ret = 0;
 380
 381         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 382                 return -EPERM;
 383
 384         /* ignore trailing newline */
 385         if (len >  0 && buf[len - 1] == '\n')
 386                 --count;
 387
 388         if (!rtnl_trylock())
 389                 return restart_syscall();
 390
 391         if (dev_isalive(netdev)) {
 392                 ret = dev_set_alias(netdev, buf, count);
 393                 if (ret < 0)
 394                         goto err;
 395                 ret = len;
 396                 netdev_state_change(netdev);
 397         }
 398 err:
 399         rtnl_unlock();
 400
 401         return ret;
 402 }
 403
 404 static ssize_t ifalias_show(struct device *dev,
 405                             struct device_attribute *attr, char *buf)
 406 {
 407         const struct net_device *netdev = to_net_dev(dev);
 408         char tmp[IFALIASZ];
 409         ssize_t ret = 0;
 410
 411         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
 412         if (ret > 0)
 413                 ret = sprintf(buf, "%s\n", tmp);
 414         return ret;
 415 }
 416 static DEVICE_ATTR_RW(ifalias);
 417
 418 static int change_group(struct net_device *dev, unsigned long new_group)
 419 {
 420         dev_set_group(dev, (int)new_group);
 421         return 0;
 422 }
 423
 424 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 425                            const char *buf, size_t len)
 426 {
 427         return netdev_store(dev, attr, buf, len, change_group);
 428 }
 429 NETDEVICE_SHOW(group, fmt_dec);
 430 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
 431
 432 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 433 {
 434         return dev_change_proto_down(dev, (bool)proto_down);
 435 }
 436
 437 static ssize_t proto_down_store(struct device *dev,
 438                                 struct device_attribute *attr,
 439                                 const char *buf, size_t len)
 440 {
 441         return netdev_store(dev, attr, buf, len, change_proto_down);
 442 }
 443 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 444
 445 static ssize_t phys_port_id_show(struct device *dev,
 446                                  struct device_attribute *attr, char *buf)
 447 {
 448         struct net_device *netdev = to_net_dev(dev);
 449         ssize_t ret = -EINVAL;
 450
 451         if (!rtnl_trylock())
 452                 return restart_syscall();
 453
 454         if (dev_isalive(netdev)) {
 455                 struct netdev_phys_item_id ppid;
 456
 457                 ret = dev_get_phys_port_id(netdev, &ppid);
 458                 if (!ret)
 459                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 460         }
 461         rtnl_unlock();
 462
 463         return ret;
 464 }
 465 static DEVICE_ATTR_RO(phys_port_id);
 466
 467 static ssize_t phys_port_name_show(struct device *dev,
 468                                    struct device_attribute *attr, char *buf)
 469 {
 470         struct net_device *netdev = to_net_dev(dev);
 471         ssize_t ret = -EINVAL;
 472
 473         if (!rtnl_trylock())
 474                 return restart_syscall();
 475
 476         if (dev_isalive(netdev)) {
 477                 char name[IFNAMSIZ];
 478
 479                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 480                 if (!ret)
 481                         ret = sprintf(buf, "%s\n", name);
 482         }
 483         rtnl_unlock();
 484
 485         return ret;
 486 }
 487 static DEVICE_ATTR_RO(phys_port_name);
 488
 489 static ssize_t phys_switch_id_show(struct device *dev,
 490                                    struct device_attribute *attr, char *buf)
 491 {
 492         struct net_device *netdev = to_net_dev(dev);
 493         ssize_t ret = -EINVAL;
 494
 495         if (!rtnl_trylock())
 496                 return restart_syscall();
 497
 498         if (dev_isalive(netdev)) {
 499                 struct netdev_phys_item_id ppid = { };
 500
 501                 ret = dev_get_port_parent_id(netdev, &ppid, false);
 502                 if (!ret)
 503                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 504         }
 505         rtnl_unlock();
 506
 507         return ret;
 508 }
 509 static DEVICE_ATTR_RO(phys_switch_id);
 510
 511 static struct attribute *net_class_attrs[] __ro_after_init = {
 512         &dev_attr_netdev_group.attr,
 513         &dev_attr_type.attr,
 514         &dev_attr_dev_id.attr,
 515         &dev_attr_dev_port.attr,
 516         &dev_attr_iflink.attr,
 517         &dev_attr_ifindex.attr,
 518         &dev_attr_name_assign_type.attr,
 519         &dev_attr_addr_assign_type.attr,
 520         &dev_attr_addr_len.attr,
 521         &dev_attr_link_mode.attr,
 522         &dev_attr_address.attr,
 523         &dev_attr_broadcast.attr,
 524         &dev_attr_speed.attr,
 525         &dev_attr_duplex.attr,
 526         &dev_attr_dormant.attr,
 527         &dev_attr_operstate.attr,
 528         &dev_attr_carrier_changes.attr,
 529         &dev_attr_ifalias.attr,
 530         &dev_attr_carrier.attr,
 531         &dev_attr_mtu.attr,
 532         &dev_attr_flags.attr,
 533         &dev_attr_tx_queue_len.attr,
 534         &dev_attr_gro_flush_timeout.attr,
 535         &dev_attr_phys_port_id.attr,
 536         &dev_attr_phys_port_name.attr,
 537         &dev_attr_phys_switch_id.attr,
 538         &dev_attr_proto_down.attr,
 539         &dev_attr_carrier_up_count.attr,
 540         &dev_attr_carrier_down_count.attr,
 541         NULL,
 542 };
 543 ATTRIBUTE_GROUPS(net_class);
 544
 545 /* Show a given an attribute in the statistics group */
 546 static ssize_t netstat_show(const struct device *d,
 547                             struct device_attribute *attr, char *buf,
 548                             unsigned long offset)
 549 {
 550         struct net_device *dev = to_net_dev(d);
 551         ssize_t ret = -EINVAL;
 552
 553         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 554                 offset % sizeof(u64) != 0);
 555
 556         read_lock(&dev_base_lock);
 557         if (dev_isalive(dev)) {
 558                 struct rtnl_link_stats64 temp;
 559                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 560
 561                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
 562         }
 563         read_unlock(&dev_base_lock);
 564         return ret;
 565 }
 566
 567 /* generate a read-only statistics attribute */
 568 #define NETSTAT_ENTRY(name)                                             \
 569 static ssize_t name##_show(struct device *d,                            \
 570                            struct device_attribute *attr, char *buf)    \
 571 {                                                                       \
 572         return netstat_show(d, attr, buf,                               \
 573                             offsetof(struct rtnl_link_stats64, name));  \
 574 }                                                                       \
 575 static DEVICE_ATTR_RO(name)
 576
 577 NETSTAT_ENTRY(rx_packets);
 578 NETSTAT_ENTRY(tx_packets);
 579 NETSTAT_ENTRY(rx_bytes);
 580 NETSTAT_ENTRY(tx_bytes);
 581 NETSTAT_ENTRY(rx_errors);
 582 NETSTAT_ENTRY(tx_errors);
 583 NETSTAT_ENTRY(rx_dropped);
 584 NETSTAT_ENTRY(tx_dropped);
 585 NETSTAT_ENTRY(multicast);
 586 NETSTAT_ENTRY(collisions);
 587 NETSTAT_ENTRY(rx_length_errors);
 588 NETSTAT_ENTRY(rx_over_errors);
 589 NETSTAT_ENTRY(rx_crc_errors);
 590 NETSTAT_ENTRY(rx_frame_errors);
 591 NETSTAT_ENTRY(rx_fifo_errors);
 592 NETSTAT_ENTRY(rx_missed_errors);
 593 NETSTAT_ENTRY(tx_aborted_errors);
 594 NETSTAT_ENTRY(tx_carrier_errors);
 595 NETSTAT_ENTRY(tx_fifo_errors);
 596 NETSTAT_ENTRY(tx_heartbeat_errors);
 597 NETSTAT_ENTRY(tx_window_errors);
 598 NETSTAT_ENTRY(rx_compressed);
 599 NETSTAT_ENTRY(tx_compressed);
 600 NETSTAT_ENTRY(rx_nohandler);
 601
 602 static struct attribute *netstat_attrs[] __ro_after_init = {
 603         &dev_attr_rx_packets.attr,
 604         &dev_attr_tx_packets.attr,
 605         &dev_attr_rx_bytes.attr,
 606         &dev_attr_tx_bytes.attr,
 607         &dev_attr_rx_errors.attr,
 608         &dev_attr_tx_errors.attr,
 609         &dev_attr_rx_dropped.attr,
 610         &dev_attr_tx_dropped.attr,
 611         &dev_attr_multicast.attr,
 612         &dev_attr_collisions.attr,
 613         &dev_attr_rx_length_errors.attr,
 614         &dev_attr_rx_over_errors.attr,
 615         &dev_attr_rx_crc_errors.attr,
 616         &dev_attr_rx_frame_errors.attr,
 617         &dev_attr_rx_fifo_errors.attr,
 618         &dev_attr_rx_missed_errors.attr,
 619         &dev_attr_tx_aborted_errors.attr,
 620         &dev_attr_tx_carrier_errors.attr,
 621         &dev_attr_tx_fifo_errors.attr,
 622         &dev_attr_tx_heartbeat_errors.attr,
 623         &dev_attr_tx_window_errors.attr,
 624         &dev_attr_rx_compressed.attr,
 625         &dev_attr_tx_compressed.attr,
 626         &dev_attr_rx_nohandler.attr,
 627         NULL
 628 };
 629
 630 static const struct attribute_group netstat_group = {
 631         .name  = "statistics",
 632         .attrs  = netstat_attrs,
 633 };
 634
 635 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 636 static struct attribute *wireless_attrs[] = {
 637         NULL
 638 };
 639
 640 static const struct attribute_group wireless_group = {
 641         .name = "wireless",
 642         .attrs = wireless_attrs,
 643 };
 644 #endif
 645
 646 #else /* CONFIG_SYSFS */
 647 #define net_class_groups        NULL
 648 #endif /* CONFIG_SYSFS */
 649
 650 #ifdef CONFIG_SYSFS
 651 #define to_rx_queue_attr(_attr) \
 652         container_of(_attr, struct rx_queue_attribute, attr)
 653
 654 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 655
 656 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 657                                   char *buf)
 658 {
 659         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 660         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 661
 662         if (!attribute->show)
 663                 return -EIO;
 664
 665         return attribute->show(queue, buf);
 666 }
 667
 668 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 669                                    const char *buf, size_t count)
 670 {
 671         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 672         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 673
 674         if (!attribute->store)
 675                 return -EIO;
 676
 677         return attribute->store(queue, buf, count);
 678 }
 679
 680 static const struct sysfs_ops rx_queue_sysfs_ops = {
 681         .show = rx_queue_attr_show,
 682         .store = rx_queue_attr_store,
 683 };
 684
 685 #ifdef CONFIG_RPS
 686 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
 687 {
 688         struct rps_map *map;
 689         cpumask_var_t mask;
 690         int i, len;
 691
 692         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 693                 return -ENOMEM;
 694
 695         rcu_read_lock();
 696         map = rcu_dereference(queue->rps_map);
 697         if (map)
 698                 for (i = 0; i < map->len; i++)
 699                         cpumask_set_cpu(map->cpus[i], mask);
 700
 701         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 702         rcu_read_unlock();
 703         free_cpumask_var(mask);
 704
 705         return len < PAGE_SIZE ? len : -EINVAL;
 706 }
 707
 708 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 709                              const char *buf, size_t len)
 710 {
 711         struct rps_map *old_map, *map;
 712         cpumask_var_t mask;
 713         int err, cpu, i;
 714         static DEFINE_MUTEX(rps_map_mutex);
 715
 716         if (!capable(CAP_NET_ADMIN))
 717                 return -EPERM;
 718
 719         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 720                 return -ENOMEM;
 721
 722         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 723         if (err) {
 724                 free_cpumask_var(mask);
 725                 return err;
 726         }
 727
 728         map = kzalloc(max_t(unsigned int,
 729                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 730                       GFP_KERNEL);
 731         if (!map) {
 732                 free_cpumask_var(mask);
 733                 return -ENOMEM;
 734         }
 735
 736         i = 0;
 737         for_each_cpu_and(cpu, mask, cpu_online_mask)
 738                 map->cpus[i++] = cpu;
 739
 740         if (i) {
 741                 map->len = i;
 742         } else {
 743                 kfree(map);
 744                 map = NULL;
 745         }
 746
 747         mutex_lock(&rps_map_mutex);
 748         old_map = rcu_dereference_protected(queue->rps_map,
 749                                             mutex_is_locked(&rps_map_mutex));
 750         rcu_assign_pointer(queue->rps_map, map);
 751
 752         if (map)
 753                 static_branch_inc(&rps_needed);
 754         if (old_map)
 755                 static_branch_dec(&rps_needed);
 756
 757         mutex_unlock(&rps_map_mutex);
 758
 759         if (old_map)
 760                 kfree_rcu(old_map, rcu);
 761
 762         free_cpumask_var(mask);
 763         return len;
 764 }
 765
 766 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 767                                            char *buf)
 768 {
 769         struct rps_dev_flow_table *flow_table;
 770         unsigned long val = 0;
 771
 772         rcu_read_lock();
 773         flow_table = rcu_dereference(queue->rps_flow_table);
 774         if (flow_table)
 775                 val = (unsigned long)flow_table->mask + 1;
 776         rcu_read_unlock();
 777
 778         return sprintf(buf, "%lu\n", val);
 779 }
 780
 781 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 782 {
 783         struct rps_dev_flow_table *table = container_of(rcu,
 784             struct rps_dev_flow_table, rcu);
 785         vfree(table);
 786 }
 787
 788 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 789                                             const char *buf, size_t len)
 790 {
 791         unsigned long mask, count;
 792         struct rps_dev_flow_table *table, *old_table;
 793         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 794         int rc;
 795
 796         if (!capable(CAP_NET_ADMIN))
 797                 return -EPERM;
 798
 799         rc = kstrtoul(buf, 0, &count);
 800         if (rc < 0)
 801                 return rc;
 802
 803         if (count) {
 804                 mask = count - 1;
 805                 /* mask = roundup_pow_of_two(count) - 1;
 806                  * without overflows...
 807                  */
 808                 while ((mask | (mask >> 1)) != mask)
 809                         mask |= (mask >> 1);
 810                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 811                  * and on 32bit arches, must check
 812                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 813                  */
 814 #if BITS_PER_LONG > 32
 815                 if (mask > (unsigned long)(u32)mask)
 816                         return -EINVAL;
 817 #else
 818                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 819                                 / sizeof(struct rps_dev_flow)) {
 820                         /* Enforce a limit to prevent overflow */
 821                         return -EINVAL;
 822                 }
 823 #endif
 824                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 825                 if (!table)
 826                         return -ENOMEM;
 827
 828                 table->mask = mask;
 829                 for (count = 0; count <= mask; count++)
 830                         table->flows[count].cpu = RPS_NO_CPU;
 831         } else {
 832                 table = NULL;
 833         }
 834
 835         spin_lock(&rps_dev_flow_lock);
 836         old_table = rcu_dereference_protected(queue->rps_flow_table,
 837                                               lockdep_is_held(&rps_dev_flow_lock));
 838         rcu_assign_pointer(queue->rps_flow_table, table);
 839         spin_unlock(&rps_dev_flow_lock);
 840
 841         if (old_table)
 842                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 843
 844         return len;
 845 }
 846
 847 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
 848         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
 849
 850 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
 851         = __ATTR(rps_flow_cnt, 0644,
 852                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 853 #endif /* CONFIG_RPS */
 854
 855 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
 856 #ifdef CONFIG_RPS
 857         &rps_cpus_attribute.attr,
 858         &rps_dev_flow_table_cnt_attribute.attr,
 859 #endif
 860         NULL
 861 };
 862 ATTRIBUTE_GROUPS(rx_queue_default);
 863
 864 static void rx_queue_release(struct kobject *kobj)
 865 {
 866         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 867 #ifdef CONFIG_RPS
 868         struct rps_map *map;
 869         struct rps_dev_flow_table *flow_table;
 870
 871         map = rcu_dereference_protected(queue->rps_map, 1);
 872         if (map) {
 873                 RCU_INIT_POINTER(queue->rps_map, NULL);
 874                 kfree_rcu(map, rcu);
 875         }
 876
 877         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 878         if (flow_table) {
 879                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 880                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 881         }
 882 #endif
 883
 884         memset(kobj, 0, sizeof(*kobj));
 885         dev_put(queue->dev);
 886 }
 887
 888 static const void *rx_queue_namespace(struct kobject *kobj)
 889 {
 890         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 891         struct device *dev = &queue->dev->dev;
 892         const void *ns = NULL;
 893
 894         if (dev->class && dev->class->ns_type)
 895                 ns = dev->class->namespace(dev);
 896
 897         return ns;
 898 }
 899
 900 static void rx_queue_get_ownership(struct kobject *kobj,
 901                                    kuid_t *uid, kgid_t *gid)
 902 {
 903         const struct net *net = rx_queue_namespace(kobj);
 904
 905         net_ns_get_ownership(net, uid, gid);
 906 }
 907
 908 static struct kobj_type rx_queue_ktype __ro_after_init = {
 909         .sysfs_ops = &rx_queue_sysfs_ops,
 910         .release = rx_queue_release,
 911         .default_groups = rx_queue_default_groups,
 912         .namespace = rx_queue_namespace,
 913         .get_ownership = rx_queue_get_ownership,
 914 };
 915
 916 static int rx_queue_add_kobject(struct net_device *dev, int index)
 917 {
 918         struct netdev_rx_queue *queue = dev->_rx + index;
 919         struct kobject *kobj = &queue->kobj;
 920         int error = 0;
 921
 922         /* Kobject_put later will trigger rx_queue_release call which
 923          * decreases dev refcount: Take that reference here
 924          */
 925         dev_hold(queue->dev);
 926
 927         kobj->kset = dev->queues_kset;
 928         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 929                                      "rx-%u", index);
 930         if (error)
 931                 goto err;
 932
 933         if (dev->sysfs_rx_queue_group) {
 934                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 935                 if (error)
 936                         goto err;
 937         }
 938
 939         kobject_uevent(kobj, KOBJ_ADD);
 940
 941         return error;
 942
 943 err:
 944         kobject_put(kobj);
 945         return error;
 946 }
 947
 948 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
 949                                  kgid_t kgid)
 950 {
 951         struct netdev_rx_queue *queue = dev->_rx + index;
 952         struct kobject *kobj = &queue->kobj;
 953         int error;
 954
 955         error = sysfs_change_owner(kobj, kuid, kgid);
 956         if (error)
 957                 return error;
 958
 959         if (dev->sysfs_rx_queue_group)
 960                 error = sysfs_group_change_owner(
 961                         kobj, dev->sysfs_rx_queue_group, kuid, kgid);
 962
 963         return error;
 964 }
 965 #endif /* CONFIG_SYSFS */
 966
 967 int
 968 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 969 {
 970 #ifdef CONFIG_SYSFS
 971         int i;
 972         int error = 0;
 973
 974 #ifndef CONFIG_RPS
 975         if (!dev->sysfs_rx_queue_group)
 976                 return 0;
 977 #endif
 978         for (i = old_num; i < new_num; i++) {
 979                 error = rx_queue_add_kobject(dev, i);
 980                 if (error) {
 981                         new_num = old_num;
 982                         break;
 983                 }
 984         }
 985
 986         while (--i >= new_num) {
 987                 struct kobject *kobj = &dev->_rx[i].kobj;
 988
 989                 if (!refcount_read(&dev_net(dev)->count))
 990                         kobj->uevent_suppress = 1;
 991                 if (dev->sysfs_rx_queue_group)
 992                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
 993                 kobject_put(kobj);
 994         }
 995
 996         return error;
 997 #else
 998         return 0;
 999 #endif
1000 }
1001
1002 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1003                                      kuid_t kuid, kgid_t kgid)
1004 {
1005 #ifdef CONFIG_SYSFS
1006         int error = 0;
1007         int i;
1008
1009 #ifndef CONFIG_RPS
1010         if (!dev->sysfs_rx_queue_group)
1011                 return 0;
1012 #endif
1013         for (i = 0; i < num; i++) {
1014                 error = rx_queue_change_owner(dev, i, kuid, kgid);
1015                 if (error)
1016                         break;
1017         }
1018
1019         return error;
1020 #else
1021         return 0;
1022 #endif
1023 }
1024
1025 #ifdef CONFIG_SYSFS
1026 /*
1027  * netdev_queue sysfs structures and functions.
1028  */
1029 struct netdev_queue_attribute {
1030         struct attribute attr;
1031         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1032         ssize_t (*store)(struct netdev_queue *queue,
1033                          const char *buf, size_t len);
1034 };
1035 #define to_netdev_queue_attr(_attr) \
1036         container_of(_attr, struct netdev_queue_attribute, attr)
1037
1038 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1039
1040 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1041                                       struct attribute *attr, char *buf)
1042 {
1043         const struct netdev_queue_attribute *attribute
1044                 = to_netdev_queue_attr(attr);
1045         struct netdev_queue *queue = to_netdev_queue(kobj);
1046
1047         if (!attribute->show)
1048                 return -EIO;
1049
1050         return attribute->show(queue, buf);
1051 }
1052
1053 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1054                                        struct attribute *attr,
1055                                        const char *buf, size_t count)
1056 {
1057         const struct netdev_queue_attribute *attribute
1058                 = to_netdev_queue_attr(attr);
1059         struct netdev_queue *queue = to_netdev_queue(kobj);
1060
1061         if (!attribute->store)
1062                 return -EIO;
1063
1064         return attribute->store(queue, buf, count);
1065 }
1066
1067 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1068         .show = netdev_queue_attr_show,
1069         .store = netdev_queue_attr_store,
1070 };
1071
1072 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1073 {
1074         unsigned long trans_timeout;
1075
1076         spin_lock_irq(&queue->_xmit_lock);
1077         trans_timeout = queue->trans_timeout;
1078         spin_unlock_irq(&queue->_xmit_lock);
1079
1080         return sprintf(buf, "%lu", trans_timeout);
1081 }
1082
1083 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1084 {
1085         struct net_device *dev = queue->dev;
1086         unsigned int i;
1087
1088         i = queue - dev->_tx;
1089         BUG_ON(i >= dev->num_tx_queues);
1090
1091         return i;
1092 }
1093
1094 static ssize_t traffic_class_show(struct netdev_queue *queue,
1095                                   char *buf)
1096 {
1097         struct net_device *dev = queue->dev;
1098         int index;
1099         int tc;
1100
1101         if (!netif_is_multiqueue(dev))
1102                 return -ENOENT;
1103
1104         index = get_netdev_queue_index(queue);
1105
1106         /* If queue belongs to subordinate dev use its TC mapping */
1107         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1108
1109         tc = netdev_txq_to_tc(dev, index);
1110         if (tc < 0)
1111                 return -EINVAL;
1112
1113         /* We can report the traffic class one of two ways:
1114          * Subordinate device traffic classes are reported with the traffic
1115          * class first, and then the subordinate class so for example TC0 on
1116          * subordinate device 2 will be reported as "0-2". If the queue
1117          * belongs to the root device it will be reported with just the
1118          * traffic class, so just "0" for TC 0 for example.
1119          */
1120         return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1121                                  sprintf(buf, "%u\n", tc);
1122 }
1123
1124 #ifdef CONFIG_XPS
1125 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1126                                char *buf)
1127 {
1128         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1129 }
1130
1131 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1132                                 const char *buf, size_t len)
1133 {
1134         struct net_device *dev = queue->dev;
1135         int err, index = get_netdev_queue_index(queue);
1136         u32 rate = 0;
1137
1138         if (!capable(CAP_NET_ADMIN))
1139                 return -EPERM;
1140
1141         err = kstrtou32(buf, 10, &rate);
1142         if (err < 0)
1143                 return err;
1144
1145         if (!rtnl_trylock())
1146                 return restart_syscall();
1147
1148         err = -EOPNOTSUPP;
1149         if (dev->netdev_ops->ndo_set_tx_maxrate)
1150                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1151
1152         rtnl_unlock();
1153         if (!err) {
1154                 queue->tx_maxrate = rate;
1155                 return len;
1156         }
1157         return err;
1158 }
1159
1160 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1161         = __ATTR_RW(tx_maxrate);
1162 #endif
1163
1164 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1165         = __ATTR_RO(tx_timeout);
1166
1167 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1168         = __ATTR_RO(traffic_class);
1169
1170 #ifdef CONFIG_BQL
1171 /*
1172  * Byte queue limits sysfs structures and functions.
1173  */
1174 static ssize_t bql_show(char *buf, unsigned int value)
1175 {
1176         return sprintf(buf, "%u\n", value);
1177 }
1178
1179 static ssize_t bql_set(const char *buf, const size_t count,
1180                        unsigned int *pvalue)
1181 {
1182         unsigned int value;
1183         int err;
1184
1185         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1186                 value = DQL_MAX_LIMIT;
1187         } else {
1188                 err = kstrtouint(buf, 10, &value);
1189                 if (err < 0)
1190                         return err;
1191                 if (value > DQL_MAX_LIMIT)
1192                         return -EINVAL;
1193         }
1194
1195         *pvalue = value;
1196
1197         return count;
1198 }
1199
1200 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1201                                   char *buf)
1202 {
1203         struct dql *dql = &queue->dql;
1204
1205         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1206 }
1207
1208 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1209                                  const char *buf, size_t len)
1210 {
1211         struct dql *dql = &queue->dql;
1212         unsigned int value;
1213         int err;
1214
1215         err = kstrtouint(buf, 10, &value);
1216         if (err < 0)
1217                 return err;
1218
1219         dql->slack_hold_time = msecs_to_jiffies(value);
1220
1221         return len;
1222 }
1223
1224 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1225         = __ATTR(hold_time, 0644,
1226                  bql_show_hold_time, bql_set_hold_time);
1227
1228 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1229                                  char *buf)
1230 {
1231         struct dql *dql = &queue->dql;
1232
1233         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1234 }
1235
1236 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1237         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1238
1239 #define BQL_ATTR(NAME, FIELD)                                           \
1240 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1241                                  char *buf)                             \
1242 {                                                                       \
1243         return bql_show(buf, queue->dql.FIELD);                         \
1244 }                                                                       \
1245                                                                         \
1246 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1247                                 const char *buf, size_t len)            \
1248 {                                                                       \
1249         return bql_set(buf, len, &queue->dql.FIELD);                    \
1250 }                                                                       \
1251                                                                         \
1252 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1253         = __ATTR(NAME, 0644,                            \
1254                  bql_show_ ## NAME, bql_set_ ## NAME)
1255
1256 BQL_ATTR(limit, limit);
1257 BQL_ATTR(limit_max, max_limit);
1258 BQL_ATTR(limit_min, min_limit);
1259
1260 static struct attribute *dql_attrs[] __ro_after_init = {
1261         &bql_limit_attribute.attr,
1262         &bql_limit_max_attribute.attr,
1263         &bql_limit_min_attribute.attr,
1264         &bql_hold_time_attribute.attr,
1265         &bql_inflight_attribute.attr,
1266         NULL
1267 };
1268
1269 static const struct attribute_group dql_group = {
1270         .name  = "byte_queue_limits",
1271         .attrs  = dql_attrs,
1272 };
1273 #endif /* CONFIG_BQL */
1274
1275 #ifdef CONFIG_XPS
1276 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1277                              char *buf)
1278 {
1279         struct net_device *dev = queue->dev;
1280         int cpu, len, num_tc = 1, tc = 0;
1281         struct xps_dev_maps *dev_maps;
1282         cpumask_var_t mask;
1283         unsigned long index;
1284
1285         if (!netif_is_multiqueue(dev))
1286                 return -ENOENT;
1287
1288         index = get_netdev_queue_index(queue);
1289
1290         if (dev->num_tc) {
1291                 /* Do not allow XPS on subordinate device directly */
1292                 num_tc = dev->num_tc;
1293                 if (num_tc < 0)
1294                         return -EINVAL;
1295
1296                 /* If queue belongs to subordinate dev use its map */
1297                 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1298
1299                 tc = netdev_txq_to_tc(dev, index);
1300                 if (tc < 0)
1301                         return -EINVAL;
1302         }
1303
1304         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1305                 return -ENOMEM;
1306
1307         rcu_read_lock();
1308         dev_maps = rcu_dereference(dev->xps_cpus_map);
1309         if (dev_maps) {
1310                 for_each_possible_cpu(cpu) {
1311                         int i, tci = cpu * num_tc + tc;
1312                         struct xps_map *map;
1313
1314                         map = rcu_dereference(dev_maps->attr_map[tci]);
1315                         if (!map)
1316                                 continue;
1317
1318                         for (i = map->len; i--;) {
1319                                 if (map->queues[i] == index) {
1320                                         cpumask_set_cpu(cpu, mask);
1321                                         break;
1322                                 }
1323                         }
1324                 }
1325         }
1326         rcu_read_unlock();
1327
1328         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1329         free_cpumask_var(mask);
1330         return len < PAGE_SIZE ? len : -EINVAL;
1331 }
1332
1333 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1334                               const char *buf, size_t len)
1335 {
1336         struct net_device *dev = queue->dev;
1337         unsigned long index;
1338         cpumask_var_t mask;
1339         int err;
1340
1341         if (!netif_is_multiqueue(dev))
1342                 return -ENOENT;
1343
1344         if (!capable(CAP_NET_ADMIN))
1345                 return -EPERM;
1346
1347         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1348                 return -ENOMEM;
1349
1350         index = get_netdev_queue_index(queue);
1351
1352         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1353         if (err) {
1354                 free_cpumask_var(mask);
1355                 return err;
1356         }
1357
1358         err = netif_set_xps_queue(dev, mask, index);
1359
1360         free_cpumask_var(mask);
1361
1362         return err ? : len;
1363 }
1364
1365 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1366         = __ATTR_RW(xps_cpus);
1367
1368 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1369 {
1370         struct net_device *dev = queue->dev;
1371         struct xps_dev_maps *dev_maps;
1372         unsigned long *mask, index;
1373         int j, len, num_tc = 1, tc = 0;
1374
1375         index = get_netdev_queue_index(queue);
1376
1377         if (dev->num_tc) {
1378                 num_tc = dev->num_tc;
1379                 tc = netdev_txq_to_tc(dev, index);
1380                 if (tc < 0)
1381                         return -EINVAL;
1382         }
1383         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1384         if (!mask)
1385                 return -ENOMEM;
1386
1387         rcu_read_lock();
1388         dev_maps = rcu_dereference(dev->xps_rxqs_map);
1389         if (!dev_maps)
1390                 goto out_no_maps;
1391
1392         for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1393              j < dev->num_rx_queues;) {
1394                 int i, tci = j * num_tc + tc;
1395                 struct xps_map *map;
1396
1397                 map = rcu_dereference(dev_maps->attr_map[tci]);
1398                 if (!map)
1399                         continue;
1400
1401                 for (i = map->len; i--;) {
1402                         if (map->queues[i] == index) {
1403                                 set_bit(j, mask);
1404                                 break;
1405                         }
1406                 }
1407         }
1408 out_no_maps:
1409         rcu_read_unlock();
1410
1411         len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1412         bitmap_free(mask);
1413
1414         return len < PAGE_SIZE ? len : -EINVAL;
1415 }
1416
1417 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1418                               size_t len)
1419 {
1420         struct net_device *dev = queue->dev;
1421         struct net *net = dev_net(dev);
1422         unsigned long *mask, index;
1423         int err;
1424
1425         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1426                 return -EPERM;
1427
1428         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1429         if (!mask)
1430                 return -ENOMEM;
1431
1432         index = get_netdev_queue_index(queue);
1433
1434         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1435         if (err) {
1436                 bitmap_free(mask);
1437                 return err;
1438         }
1439
1440         cpus_read_lock();
1441         err = __netif_set_xps_queue(dev, mask, index, true);
1442         cpus_read_unlock();
1443
1444         bitmap_free(mask);
1445         return err ? : len;
1446 }
1447
1448 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1449         = __ATTR_RW(xps_rxqs);
1450 #endif /* CONFIG_XPS */
1451
1452 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1453         &queue_trans_timeout.attr,
1454         &queue_traffic_class.attr,
1455 #ifdef CONFIG_XPS
1456         &xps_cpus_attribute.attr,
1457         &xps_rxqs_attribute.attr,
1458         &queue_tx_maxrate.attr,
1459 #endif
1460         NULL
1461 };
1462 ATTRIBUTE_GROUPS(netdev_queue_default);
1463
1464 static void netdev_queue_release(struct kobject *kobj)
1465 {
1466         struct netdev_queue *queue = to_netdev_queue(kobj);
1467
1468         memset(kobj, 0, sizeof(*kobj));
1469         dev_put(queue->dev);
1470 }
1471
1472 static const void *netdev_queue_namespace(struct kobject *kobj)
1473 {
1474         struct netdev_queue *queue = to_netdev_queue(kobj);
1475         struct device *dev = &queue->dev->dev;
1476         const void *ns = NULL;
1477
1478         if (dev->class && dev->class->ns_type)
1479                 ns = dev->class->namespace(dev);
1480
1481         return ns;
1482 }
1483
1484 static void netdev_queue_get_ownership(struct kobject *kobj,
1485                                        kuid_t *uid, kgid_t *gid)
1486 {
1487         const struct net *net = netdev_queue_namespace(kobj);
1488
1489         net_ns_get_ownership(net, uid, gid);
1490 }
1491
1492 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1493         .sysfs_ops = &netdev_queue_sysfs_ops,
1494         .release = netdev_queue_release,
1495         .default_groups = netdev_queue_default_groups,
1496         .namespace = netdev_queue_namespace,
1497         .get_ownership = netdev_queue_get_ownership,
1498 };
1499
1500 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1501 {
1502         struct netdev_queue *queue = dev->_tx + index;
1503         struct kobject *kobj = &queue->kobj;
1504         int error = 0;
1505
1506         /* Kobject_put later will trigger netdev_queue_release call
1507          * which decreases dev refcount: Take that reference here
1508          */
1509         dev_hold(queue->dev);
1510
1511         kobj->kset = dev->queues_kset;
1512         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1513                                      "tx-%u", index);
1514         if (error)
1515                 goto err;
1516
1517 #ifdef CONFIG_BQL
1518         error = sysfs_create_group(kobj, &dql_group);
1519         if (error)
1520                 goto err;
1521 #endif
1522
1523         kobject_uevent(kobj, KOBJ_ADD);
1524         return 0;
1525
1526 err:
1527         kobject_put(kobj);
1528         return error;
1529 }
1530
1531 static int tx_queue_change_owner(struct net_device *ndev, int index,
1532                                  kuid_t kuid, kgid_t kgid)
1533 {
1534         struct netdev_queue *queue = ndev->_tx + index;
1535         struct kobject *kobj = &queue->kobj;
1536         int error;
1537
1538         error = sysfs_change_owner(kobj, kuid, kgid);
1539         if (error)
1540                 return error;
1541
1542 #ifdef CONFIG_BQL
1543         error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1544 #endif
1545         return error;
1546 }
1547 #endif /* CONFIG_SYSFS */
1548
1549 int
1550 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1551 {
1552 #ifdef CONFIG_SYSFS
1553         int i;
1554         int error = 0;
1555
1556         for (i = old_num; i < new_num; i++) {
1557                 error = netdev_queue_add_kobject(dev, i);
1558                 if (error) {
1559                         new_num = old_num;
1560                         break;
1561                 }
1562         }
1563
1564         while (--i >= new_num) {
1565                 struct netdev_queue *queue = dev->_tx + i;
1566
1567                 if (!refcount_read(&dev_net(dev)->count))
1568                         queue->kobj.uevent_suppress = 1;
1569 #ifdef CONFIG_BQL
1570                 sysfs_remove_group(&queue->kobj, &dql_group);
1571 #endif
1572                 kobject_put(&queue->kobj);
1573         }
1574
1575         return error;
1576 #else
1577         return 0;
1578 #endif /* CONFIG_SYSFS */
1579 }
1580
1581 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1582                                      kuid_t kuid, kgid_t kgid)
1583 {
1584 #ifdef CONFIG_SYSFS
1585         int error = 0;
1586         int i;
1587
1588         for (i = 0; i < num; i++) {
1589                 error = tx_queue_change_owner(dev, i, kuid, kgid);
1590                 if (error)
1591                         break;
1592         }
1593
1594         return error;
1595 #else
1596         return 0;
1597 #endif /* CONFIG_SYSFS */
1598 }
1599
1600 static int register_queue_kobjects(struct net_device *dev)
1601 {
1602         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1603
1604 #ifdef CONFIG_SYSFS
1605         dev->queues_kset = kset_create_and_add("queues",
1606                                                NULL, &dev->dev.kobj);
1607         if (!dev->queues_kset)
1608                 return -ENOMEM;
1609         real_rx = dev->real_num_rx_queues;
1610 #endif
1611         real_tx = dev->real_num_tx_queues;
1612
1613         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1614         if (error)
1615                 goto error;
1616         rxq = real_rx;
1617
1618         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1619         if (error)
1620                 goto error;
1621         txq = real_tx;
1622
1623         return 0;
1624
1625 error:
1626         netdev_queue_update_kobjects(dev, txq, 0);
1627         net_rx_queue_update_kobjects(dev, rxq, 0);
1628 #ifdef CONFIG_SYSFS
1629         kset_unregister(dev->queues_kset);
1630 #endif
1631         return error;
1632 }
1633
1634 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1635 {
1636         int error = 0, real_rx = 0, real_tx = 0;
1637
1638 #ifdef CONFIG_SYSFS
1639         if (ndev->queues_kset) {
1640                 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1641                 if (error)
1642                         return error;
1643         }
1644         real_rx = ndev->real_num_rx_queues;
1645 #endif
1646         real_tx = ndev->real_num_tx_queues;
1647
1648         error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1649         if (error)
1650                 return error;
1651
1652         error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1653         if (error)
1654                 return error;
1655
1656         return 0;
1657 }
1658
1659 static void remove_queue_kobjects(struct net_device *dev)
1660 {
1661         int real_rx = 0, real_tx = 0;
1662
1663 #ifdef CONFIG_SYSFS
1664         real_rx = dev->real_num_rx_queues;
1665 #endif
1666         real_tx = dev->real_num_tx_queues;
1667
1668         net_rx_queue_update_kobjects(dev, real_rx, 0);
1669         netdev_queue_update_kobjects(dev, real_tx, 0);
1670 #ifdef CONFIG_SYSFS
1671         kset_unregister(dev->queues_kset);
1672 #endif
1673 }
1674
1675 static bool net_current_may_mount(void)
1676 {
1677         struct net *net = current->nsproxy->net_ns;
1678
1679         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1680 }
1681
1682 static void *net_grab_current_ns(void)
1683 {
1684         struct net *ns = current->nsproxy->net_ns;
1685 #ifdef CONFIG_NET_NS
1686         if (ns)
1687                 refcount_inc(&ns->passive);
1688 #endif
1689         return ns;
1690 }
1691
1692 static const void *net_initial_ns(void)
1693 {
1694         return &init_net;
1695 }
1696
1697 static const void *net_netlink_ns(struct sock *sk)
1698 {
1699         return sock_net(sk);
1700 }
1701
1702 const struct kobj_ns_type_operations net_ns_type_operations = {
1703         .type = KOBJ_NS_TYPE_NET,
1704         .current_may_mount = net_current_may_mount,
1705         .grab_current_ns = net_grab_current_ns,
1706         .netlink_ns = net_netlink_ns,
1707         .initial_ns = net_initial_ns,
1708         .drop_ns = net_drop_ns,
1709 };
1710 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1711
1712 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1713 {
1714         struct net_device *dev = to_net_dev(d);
1715         int retval;
1716
1717         /* pass interface to uevent. */
1718         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1719         if (retval)
1720                 goto exit;
1721
1722         /* pass ifindex to uevent.
1723          * ifindex is useful as it won't change (interface name may change)
1724          * and is what RtNetlink uses natively.
1725          */
1726         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1727
1728 exit:
1729         return retval;
1730 }
1731
1732 /*
1733  *      netdev_release -- destroy and free a dead device.
1734  *      Called when last reference to device kobject is gone.
1735  */
1736 static void netdev_release(struct device *d)
1737 {
1738         struct net_device *dev = to_net_dev(d);
1739
1740         BUG_ON(dev->reg_state != NETREG_RELEASED);
1741
1742         /* no need to wait for rcu grace period:
1743          * device is dead and about to be freed.
1744          */
1745         kfree(rcu_access_pointer(dev->ifalias));
1746         netdev_freemem(dev);
1747 }
1748
1749 static const void *net_namespace(struct device *d)
1750 {
1751         struct net_device *dev = to_net_dev(d);
1752
1753         return dev_net(dev);
1754 }
1755
1756 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1757 {
1758         struct net_device *dev = to_net_dev(d);
1759         const struct net *net = dev_net(dev);
1760
1761         net_ns_get_ownership(net, uid, gid);
1762 }
1763
1764 static struct class net_class __ro_after_init = {
1765         .name = "net",
1766         .dev_release = netdev_release,
1767         .dev_groups = net_class_groups,
1768         .dev_uevent = netdev_uevent,
1769         .ns_type = &net_ns_type_operations,
1770         .namespace = net_namespace,
1771         .get_ownership = net_get_ownership,
1772 };
1773
1774 #ifdef CONFIG_OF_NET
1775 static int of_dev_node_match(struct device *dev, const void *data)
1776 {
1777         int ret = 0;
1778
1779         if (dev->parent)
1780                 ret = dev->parent->of_node == data;
1781
1782         return ret == 0 ? dev->of_node == data : ret;
1783 }
1784
1785 /*
1786  * of_find_net_device_by_node - lookup the net device for the device node
1787  * @np: OF device node
1788  *
1789  * Looks up the net_device structure corresponding with the device node.
1790  * If successful, returns a pointer to the net_device with the embedded
1791  * struct device refcount incremented by one, or NULL on failure. The
1792  * refcount must be dropped when done with the net_device.
1793  */
1794 struct net_device *of_find_net_device_by_node(struct device_node *np)
1795 {
1796         struct device *dev;
1797
1798         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1799         if (!dev)
1800                 return NULL;
1801
1802         return to_net_dev(dev);
1803 }
1804 EXPORT_SYMBOL(of_find_net_device_by_node);
1805 #endif
1806
1807 /* Delete sysfs entries but hold kobject reference until after all
1808  * netdev references are gone.
1809  */
1810 void netdev_unregister_kobject(struct net_device *ndev)
1811 {
1812         struct device *dev = &ndev->dev;
1813
1814         if (!refcount_read(&dev_net(ndev)->count))
1815                 dev_set_uevent_suppress(dev, 1);
1816
1817         kobject_get(&dev->kobj);
1818
1819         remove_queue_kobjects(ndev);
1820
1821         pm_runtime_set_memalloc_noio(dev, false);
1822
1823         device_del(dev);
1824 }
1825
1826 /* Create sysfs entries for network device. */
1827 int netdev_register_kobject(struct net_device *ndev)
1828 {
1829         struct device *dev = &ndev->dev;
1830         const struct attribute_group **groups = ndev->sysfs_groups;
1831         int error = 0;
1832
1833         device_initialize(dev);
1834         dev->class = &net_class;
1835         dev->platform_data = ndev;
1836         dev->groups = groups;
1837
1838         dev_set_name(dev, "%s", ndev->name);
1839
1840 #ifdef CONFIG_SYSFS
1841         /* Allow for a device specific group */
1842         if (*groups)
1843                 groups++;
1844
1845         *groups++ = &netstat_group;
1846
1847 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1848         if (ndev->ieee80211_ptr)
1849                 *groups++ = &wireless_group;
1850 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1851         else if (ndev->wireless_handlers)
1852                 *groups++ = &wireless_group;
1853 #endif
1854 #endif
1855 #endif /* CONFIG_SYSFS */
1856
1857         error = device_add(dev);
1858         if (error)
1859                 return error;
1860
1861         error = register_queue_kobjects(ndev);
1862         if (error) {
1863                 device_del(dev);
1864                 return error;
1865         }
1866
1867         pm_runtime_set_memalloc_noio(dev, true);
1868
1869         return error;
1870 }
1871
1872 /* Change owner for sysfs entries when moving network devices across network
1873  * namespaces owned by different user namespaces.
1874  */
1875 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
1876                         const struct net *net_new)
1877 {
1878         struct device *dev = &ndev->dev;
1879         kuid_t old_uid, new_uid;
1880         kgid_t old_gid, new_gid;
1881         int error;
1882
1883         net_ns_get_ownership(net_old, &old_uid, &old_gid);
1884         net_ns_get_ownership(net_new, &new_uid, &new_gid);
1885
1886         /* The network namespace was changed but the owning user namespace is
1887          * identical so there's no need to change the owner of sysfs entries.
1888          */
1889         if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
1890                 return 0;
1891
1892         error = device_change_owner(dev, new_uid, new_gid);
1893         if (error)
1894                 return error;
1895
1896         error = queue_change_owner(ndev, new_uid, new_gid);
1897         if (error)
1898                 return error;
1899
1900         return 0;
1901 }
1902
1903 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1904                                 const void *ns)
1905 {
1906         return class_create_file_ns(&net_class, class_attr, ns);
1907 }
1908 EXPORT_SYMBOL(netdev_class_create_file_ns);
1909
1910 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1911                                  const void *ns)
1912 {
1913         class_remove_file_ns(&net_class, class_attr, ns);
1914 }
1915 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1916
1917 int __init netdev_kobject_init(void)
1918 {
1919         kobj_ns_type_register(&net_ns_type_operations);
1920         return class_register(&net_class);
1921 }