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         struct net_device *netdev = to_net_dev(dev);
 178
 179         /* The check is also done in change_carrier; this helps returning early
 180          * without hitting the trylock/restart in netdev_store.
 181          */
 182         if (!netdev->netdev_ops->ndo_change_carrier)
 183                 return -EOPNOTSUPP;
 184
 185         return netdev_store(dev, attr, buf, len, change_carrier);
 186 }
 187
 188 static ssize_t carrier_show(struct device *dev,
 189                             struct device_attribute *attr, char *buf)
 190 {
 191         struct net_device *netdev = to_net_dev(dev);
 192
 193         if (netif_running(netdev))
 194                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 195
 196         return -EINVAL;
 197 }
 198 static DEVICE_ATTR_RW(carrier);
 199
 200 static ssize_t speed_show(struct device *dev,
 201                           struct device_attribute *attr, char *buf)
 202 {
 203         struct net_device *netdev = to_net_dev(dev);
 204         int ret = -EINVAL;
 205
 206         /* The check is also done in __ethtool_get_link_ksettings; this helps
 207          * returning early without hitting the trylock/restart below.
 208          */
 209         if (!netdev->ethtool_ops->get_link_ksettings)
 210                 return ret;
 211
 212         if (!rtnl_trylock())
 213                 return restart_syscall();
 214
 215         if (netif_running(netdev) && netif_device_present(netdev)) {
 216                 struct ethtool_link_ksettings cmd;
 217
 218                 if (!__ethtool_get_link_ksettings(netdev, &cmd))
 219                         ret = sprintf(buf, fmt_dec, cmd.base.speed);
 220         }
 221         rtnl_unlock();
 222         return ret;
 223 }
 224 static DEVICE_ATTR_RO(speed);
 225
 226 static ssize_t duplex_show(struct device *dev,
 227                            struct device_attribute *attr, char *buf)
 228 {
 229         struct net_device *netdev = to_net_dev(dev);
 230         int ret = -EINVAL;
 231
 232         /* The check is also done in __ethtool_get_link_ksettings; this helps
 233          * returning early without hitting the trylock/restart below.
 234          */
 235         if (!netdev->ethtool_ops->get_link_ksettings)
 236                 return ret;
 237
 238         if (!rtnl_trylock())
 239                 return restart_syscall();
 240
 241         if (netif_running(netdev)) {
 242                 struct ethtool_link_ksettings cmd;
 243
 244                 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
 245                         const char *duplex;
 246
 247                         switch (cmd.base.duplex) {
 248                         case DUPLEX_HALF:
 249                                 duplex = "half";
 250                                 break;
 251                         case DUPLEX_FULL:
 252                                 duplex = "full";
 253                                 break;
 254                         default:
 255                                 duplex = "unknown";
 256                                 break;
 257                         }
 258                         ret = sprintf(buf, "%s\n", duplex);
 259                 }
 260         }
 261         rtnl_unlock();
 262         return ret;
 263 }
 264 static DEVICE_ATTR_RO(duplex);
 265
 266 static ssize_t dormant_show(struct device *dev,
 267                             struct device_attribute *attr, char *buf)
 268 {
 269         struct net_device *netdev = to_net_dev(dev);
 270
 271         if (netif_running(netdev))
 272                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 273
 274         return -EINVAL;
 275 }
 276 static DEVICE_ATTR_RO(dormant);
 277
 278 static const char *const operstates[] = {
 279         "unknown",
 280         "notpresent", /* currently unused */
 281         "down",
 282         "lowerlayerdown",
 283         "testing", /* currently unused */
 284         "dormant",
 285         "up"
 286 };
 287
 288 static ssize_t operstate_show(struct device *dev,
 289                               struct device_attribute *attr, char *buf)
 290 {
 291         const struct net_device *netdev = to_net_dev(dev);
 292         unsigned char operstate;
 293
 294         read_lock(&dev_base_lock);
 295         operstate = netdev->operstate;
 296         if (!netif_running(netdev))
 297                 operstate = IF_OPER_DOWN;
 298         read_unlock(&dev_base_lock);
 299
 300         if (operstate >= ARRAY_SIZE(operstates))
 301                 return -EINVAL; /* should not happen */
 302
 303         return sprintf(buf, "%s\n", operstates[operstate]);
 304 }
 305 static DEVICE_ATTR_RO(operstate);
 306
 307 static ssize_t carrier_changes_show(struct device *dev,
 308                                     struct device_attribute *attr,
 309                                     char *buf)
 310 {
 311         struct net_device *netdev = to_net_dev(dev);
 312
 313         return sprintf(buf, fmt_dec,
 314                        atomic_read(&netdev->carrier_up_count) +
 315                        atomic_read(&netdev->carrier_down_count));
 316 }
 317 static DEVICE_ATTR_RO(carrier_changes);
 318
 319 static ssize_t carrier_up_count_show(struct device *dev,
 320                                      struct device_attribute *attr,
 321                                      char *buf)
 322 {
 323         struct net_device *netdev = to_net_dev(dev);
 324
 325         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
 326 }
 327 static DEVICE_ATTR_RO(carrier_up_count);
 328
 329 static ssize_t carrier_down_count_show(struct device *dev,
 330                                        struct device_attribute *attr,
 331                                        char *buf)
 332 {
 333         struct net_device *netdev = to_net_dev(dev);
 334
 335         return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
 336 }
 337 static DEVICE_ATTR_RO(carrier_down_count);
 338
 339 /* read-write attributes */
 340
 341 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 342 {
 343         return dev_set_mtu(dev, (int)new_mtu);
 344 }
 345
 346 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 347                          const char *buf, size_t len)
 348 {
 349         return netdev_store(dev, attr, buf, len, change_mtu);
 350 }
 351 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 352
 353 static int change_flags(struct net_device *dev, unsigned long new_flags)
 354 {
 355         return dev_change_flags(dev, (unsigned int)new_flags, NULL);
 356 }
 357
 358 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 359                            const char *buf, size_t len)
 360 {
 361         return netdev_store(dev, attr, buf, len, change_flags);
 362 }
 363 NETDEVICE_SHOW_RW(flags, fmt_hex);
 364
 365 static ssize_t tx_queue_len_store(struct device *dev,
 366                                   struct device_attribute *attr,
 367                                   const char *buf, size_t len)
 368 {
 369         if (!capable(CAP_NET_ADMIN))
 370                 return -EPERM;
 371
 372         return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
 373 }
 374 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 375
 376 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 377 {
 378         dev->gro_flush_timeout = val;
 379         return 0;
 380 }
 381
 382 static ssize_t gro_flush_timeout_store(struct device *dev,
 383                                        struct device_attribute *attr,
 384                                        const char *buf, size_t len)
 385 {
 386         if (!capable(CAP_NET_ADMIN))
 387                 return -EPERM;
 388
 389         return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 390 }
 391 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 392
 393 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 394                              const char *buf, size_t len)
 395 {
 396         struct net_device *netdev = to_net_dev(dev);
 397         struct net *net = dev_net(netdev);
 398         size_t count = len;
 399         ssize_t ret = 0;
 400
 401         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 402                 return -EPERM;
 403
 404         /* ignore trailing newline */
 405         if (len >  0 && buf[len - 1] == '\n')
 406                 --count;
 407
 408         if (!rtnl_trylock())
 409                 return restart_syscall();
 410
 411         if (dev_isalive(netdev)) {
 412                 ret = dev_set_alias(netdev, buf, count);
 413                 if (ret < 0)
 414                         goto err;
 415                 ret = len;
 416                 netdev_state_change(netdev);
 417         }
 418 err:
 419         rtnl_unlock();
 420
 421         return ret;
 422 }
 423
 424 static ssize_t ifalias_show(struct device *dev,
 425                             struct device_attribute *attr, char *buf)
 426 {
 427         const struct net_device *netdev = to_net_dev(dev);
 428         char tmp[IFALIASZ];
 429         ssize_t ret = 0;
 430
 431         ret = dev_get_alias(netdev, tmp, sizeof(tmp));
 432         if (ret > 0)
 433                 ret = sprintf(buf, "%s\n", tmp);
 434         return ret;
 435 }
 436 static DEVICE_ATTR_RW(ifalias);
 437
 438 static int change_group(struct net_device *dev, unsigned long new_group)
 439 {
 440         dev_set_group(dev, (int)new_group);
 441         return 0;
 442 }
 443
 444 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 445                            const char *buf, size_t len)
 446 {
 447         return netdev_store(dev, attr, buf, len, change_group);
 448 }
 449 NETDEVICE_SHOW(group, fmt_dec);
 450 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
 451
 452 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 453 {
 454         return dev_change_proto_down(dev, (bool)proto_down);
 455 }
 456
 457 static ssize_t proto_down_store(struct device *dev,
 458                                 struct device_attribute *attr,
 459                                 const char *buf, size_t len)
 460 {
 461         struct net_device *netdev = to_net_dev(dev);
 462
 463         /* The check is also done in change_proto_down; this helps returning
 464          * early without hitting the trylock/restart in netdev_store.
 465          */
 466         if (!netdev->netdev_ops->ndo_change_proto_down)
 467                 return -EOPNOTSUPP;
 468
 469         return netdev_store(dev, attr, buf, len, change_proto_down);
 470 }
 471 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 472
 473 static ssize_t phys_port_id_show(struct device *dev,
 474                                  struct device_attribute *attr, char *buf)
 475 {
 476         struct net_device *netdev = to_net_dev(dev);
 477         ssize_t ret = -EINVAL;
 478
 479         /* The check is also done in dev_get_phys_port_id; this helps returning
 480          * early without hitting the trylock/restart below.
 481          */
 482         if (!netdev->netdev_ops->ndo_get_phys_port_id)
 483                 return -EOPNOTSUPP;
 484
 485         if (!rtnl_trylock())
 486                 return restart_syscall();
 487
 488         if (dev_isalive(netdev)) {
 489                 struct netdev_phys_item_id ppid;
 490
 491                 ret = dev_get_phys_port_id(netdev, &ppid);
 492                 if (!ret)
 493                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 494         }
 495         rtnl_unlock();
 496
 497         return ret;
 498 }
 499 static DEVICE_ATTR_RO(phys_port_id);
 500
 501 static ssize_t phys_port_name_show(struct device *dev,
 502                                    struct device_attribute *attr, char *buf)
 503 {
 504         struct net_device *netdev = to_net_dev(dev);
 505         ssize_t ret = -EINVAL;
 506
 507         /* The checks are also done in dev_get_phys_port_name; this helps
 508          * returning early without hitting the trylock/restart below.
 509          */
 510         if (!netdev->netdev_ops->ndo_get_phys_port_name &&
 511             !netdev->netdev_ops->ndo_get_devlink_port)
 512                 return -EOPNOTSUPP;
 513
 514         if (!rtnl_trylock())
 515                 return restart_syscall();
 516
 517         if (dev_isalive(netdev)) {
 518                 char name[IFNAMSIZ];
 519
 520                 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 521                 if (!ret)
 522                         ret = sprintf(buf, "%s\n", name);
 523         }
 524         rtnl_unlock();
 525
 526         return ret;
 527 }
 528 static DEVICE_ATTR_RO(phys_port_name);
 529
 530 static ssize_t phys_switch_id_show(struct device *dev,
 531                                    struct device_attribute *attr, char *buf)
 532 {
 533         struct net_device *netdev = to_net_dev(dev);
 534         ssize_t ret = -EINVAL;
 535
 536         /* The checks are also done in dev_get_phys_port_name; this helps
 537          * returning early without hitting the trylock/restart below. This works
 538          * because recurse is false when calling dev_get_port_parent_id.
 539          */
 540         if (!netdev->netdev_ops->ndo_get_port_parent_id &&
 541             !netdev->netdev_ops->ndo_get_devlink_port)
 542                 return -EOPNOTSUPP;
 543
 544         if (!rtnl_trylock())
 545                 return restart_syscall();
 546
 547         if (dev_isalive(netdev)) {
 548                 struct netdev_phys_item_id ppid = { };
 549
 550                 ret = dev_get_port_parent_id(netdev, &ppid, false);
 551                 if (!ret)
 552                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 553         }
 554         rtnl_unlock();
 555
 556         return ret;
 557 }
 558 static DEVICE_ATTR_RO(phys_switch_id);
 559
 560 static struct attribute *net_class_attrs[] __ro_after_init = {
 561         &dev_attr_netdev_group.attr,
 562         &dev_attr_type.attr,
 563         &dev_attr_dev_id.attr,
 564         &dev_attr_dev_port.attr,
 565         &dev_attr_iflink.attr,
 566         &dev_attr_ifindex.attr,
 567         &dev_attr_name_assign_type.attr,
 568         &dev_attr_addr_assign_type.attr,
 569         &dev_attr_addr_len.attr,
 570         &dev_attr_link_mode.attr,
 571         &dev_attr_address.attr,
 572         &dev_attr_broadcast.attr,
 573         &dev_attr_speed.attr,
 574         &dev_attr_duplex.attr,
 575         &dev_attr_dormant.attr,
 576         &dev_attr_operstate.attr,
 577         &dev_attr_carrier_changes.attr,
 578         &dev_attr_ifalias.attr,
 579         &dev_attr_carrier.attr,
 580         &dev_attr_mtu.attr,
 581         &dev_attr_flags.attr,
 582         &dev_attr_tx_queue_len.attr,
 583         &dev_attr_gro_flush_timeout.attr,
 584         &dev_attr_phys_port_id.attr,
 585         &dev_attr_phys_port_name.attr,
 586         &dev_attr_phys_switch_id.attr,
 587         &dev_attr_proto_down.attr,
 588         &dev_attr_carrier_up_count.attr,
 589         &dev_attr_carrier_down_count.attr,
 590         NULL,
 591 };
 592 ATTRIBUTE_GROUPS(net_class);
 593
 594 /* Show a given an attribute in the statistics group */
 595 static ssize_t netstat_show(const struct device *d,
 596                             struct device_attribute *attr, char *buf,
 597                             unsigned long offset)
 598 {
 599         struct net_device *dev = to_net_dev(d);
 600         ssize_t ret = -EINVAL;
 601
 602         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 603                 offset % sizeof(u64) != 0);
 604
 605         read_lock(&dev_base_lock);
 606         if (dev_isalive(dev)) {
 607                 struct rtnl_link_stats64 temp;
 608                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 609
 610                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
 611         }
 612         read_unlock(&dev_base_lock);
 613         return ret;
 614 }
 615
 616 /* generate a read-only statistics attribute */
 617 #define NETSTAT_ENTRY(name)                                             \
 618 static ssize_t name##_show(struct device *d,                            \
 619                            struct device_attribute *attr, char *buf)    \
 620 {                                                                       \
 621         return netstat_show(d, attr, buf,                               \
 622                             offsetof(struct rtnl_link_stats64, name));  \
 623 }                                                                       \
 624 static DEVICE_ATTR_RO(name)
 625
 626 NETSTAT_ENTRY(rx_packets);
 627 NETSTAT_ENTRY(tx_packets);
 628 NETSTAT_ENTRY(rx_bytes);
 629 NETSTAT_ENTRY(tx_bytes);
 630 NETSTAT_ENTRY(rx_errors);
 631 NETSTAT_ENTRY(tx_errors);
 632 NETSTAT_ENTRY(rx_dropped);
 633 NETSTAT_ENTRY(tx_dropped);
 634 NETSTAT_ENTRY(multicast);
 635 NETSTAT_ENTRY(collisions);
 636 NETSTAT_ENTRY(rx_length_errors);
 637 NETSTAT_ENTRY(rx_over_errors);
 638 NETSTAT_ENTRY(rx_crc_errors);
 639 NETSTAT_ENTRY(rx_frame_errors);
 640 NETSTAT_ENTRY(rx_fifo_errors);
 641 NETSTAT_ENTRY(rx_missed_errors);
 642 NETSTAT_ENTRY(tx_aborted_errors);
 643 NETSTAT_ENTRY(tx_carrier_errors);
 644 NETSTAT_ENTRY(tx_fifo_errors);
 645 NETSTAT_ENTRY(tx_heartbeat_errors);
 646 NETSTAT_ENTRY(tx_window_errors);
 647 NETSTAT_ENTRY(rx_compressed);
 648 NETSTAT_ENTRY(tx_compressed);
 649 NETSTAT_ENTRY(rx_nohandler);
 650
 651 static struct attribute *netstat_attrs[] __ro_after_init = {
 652         &dev_attr_rx_packets.attr,
 653         &dev_attr_tx_packets.attr,
 654         &dev_attr_rx_bytes.attr,
 655         &dev_attr_tx_bytes.attr,
 656         &dev_attr_rx_errors.attr,
 657         &dev_attr_tx_errors.attr,
 658         &dev_attr_rx_dropped.attr,
 659         &dev_attr_tx_dropped.attr,
 660         &dev_attr_multicast.attr,
 661         &dev_attr_collisions.attr,
 662         &dev_attr_rx_length_errors.attr,
 663         &dev_attr_rx_over_errors.attr,
 664         &dev_attr_rx_crc_errors.attr,
 665         &dev_attr_rx_frame_errors.attr,
 666         &dev_attr_rx_fifo_errors.attr,
 667         &dev_attr_rx_missed_errors.attr,
 668         &dev_attr_tx_aborted_errors.attr,
 669         &dev_attr_tx_carrier_errors.attr,
 670         &dev_attr_tx_fifo_errors.attr,
 671         &dev_attr_tx_heartbeat_errors.attr,
 672         &dev_attr_tx_window_errors.attr,
 673         &dev_attr_rx_compressed.attr,
 674         &dev_attr_tx_compressed.attr,
 675         &dev_attr_rx_nohandler.attr,
 676         NULL
 677 };
 678
 679 static const struct attribute_group netstat_group = {
 680         .name  = "statistics",
 681         .attrs  = netstat_attrs,
 682 };
 683
 684 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 685 static struct attribute *wireless_attrs[] = {
 686         NULL
 687 };
 688
 689 static const struct attribute_group wireless_group = {
 690         .name = "wireless",
 691         .attrs = wireless_attrs,
 692 };
 693 #endif
 694
 695 #else /* CONFIG_SYSFS */
 696 #define net_class_groups        NULL
 697 #endif /* CONFIG_SYSFS */
 698
 699 #ifdef CONFIG_SYSFS
 700 #define to_rx_queue_attr(_attr) \
 701         container_of(_attr, struct rx_queue_attribute, attr)
 702
 703 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 704
 705 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 706                                   char *buf)
 707 {
 708         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 709         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 710
 711         if (!attribute->show)
 712                 return -EIO;
 713
 714         return attribute->show(queue, buf);
 715 }
 716
 717 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 718                                    const char *buf, size_t count)
 719 {
 720         const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 721         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 722
 723         if (!attribute->store)
 724                 return -EIO;
 725
 726         return attribute->store(queue, buf, count);
 727 }
 728
 729 static const struct sysfs_ops rx_queue_sysfs_ops = {
 730         .show = rx_queue_attr_show,
 731         .store = rx_queue_attr_store,
 732 };
 733
 734 #ifdef CONFIG_RPS
 735 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
 736 {
 737         struct rps_map *map;
 738         cpumask_var_t mask;
 739         int i, len;
 740
 741         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 742                 return -ENOMEM;
 743
 744         rcu_read_lock();
 745         map = rcu_dereference(queue->rps_map);
 746         if (map)
 747                 for (i = 0; i < map->len; i++)
 748                         cpumask_set_cpu(map->cpus[i], mask);
 749
 750         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 751         rcu_read_unlock();
 752         free_cpumask_var(mask);
 753
 754         return len < PAGE_SIZE ? len : -EINVAL;
 755 }
 756
 757 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 758                              const char *buf, size_t len)
 759 {
 760         struct rps_map *old_map, *map;
 761         cpumask_var_t mask;
 762         int err, cpu, i;
 763         static DEFINE_MUTEX(rps_map_mutex);
 764
 765         if (!capable(CAP_NET_ADMIN))
 766                 return -EPERM;
 767
 768         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 769                 return -ENOMEM;
 770
 771         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 772         if (err) {
 773                 free_cpumask_var(mask);
 774                 return err;
 775         }
 776
 777         map = kzalloc(max_t(unsigned int,
 778                             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 779                       GFP_KERNEL);
 780         if (!map) {
 781                 free_cpumask_var(mask);
 782                 return -ENOMEM;
 783         }
 784
 785         i = 0;
 786         for_each_cpu_and(cpu, mask, cpu_online_mask)
 787                 map->cpus[i++] = cpu;
 788
 789         if (i) {
 790                 map->len = i;
 791         } else {
 792                 kfree(map);
 793                 map = NULL;
 794         }
 795
 796         mutex_lock(&rps_map_mutex);
 797         old_map = rcu_dereference_protected(queue->rps_map,
 798                                             mutex_is_locked(&rps_map_mutex));
 799         rcu_assign_pointer(queue->rps_map, map);
 800
 801         if (map)
 802                 static_branch_inc(&rps_needed);
 803         if (old_map)
 804                 static_branch_dec(&rps_needed);
 805
 806         mutex_unlock(&rps_map_mutex);
 807
 808         if (old_map)
 809                 kfree_rcu(old_map, rcu);
 810
 811         free_cpumask_var(mask);
 812         return len;
 813 }
 814
 815 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 816                                            char *buf)
 817 {
 818         struct rps_dev_flow_table *flow_table;
 819         unsigned long val = 0;
 820
 821         rcu_read_lock();
 822         flow_table = rcu_dereference(queue->rps_flow_table);
 823         if (flow_table)
 824                 val = (unsigned long)flow_table->mask + 1;
 825         rcu_read_unlock();
 826
 827         return sprintf(buf, "%lu\n", val);
 828 }
 829
 830 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 831 {
 832         struct rps_dev_flow_table *table = container_of(rcu,
 833             struct rps_dev_flow_table, rcu);
 834         vfree(table);
 835 }
 836
 837 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 838                                             const char *buf, size_t len)
 839 {
 840         unsigned long mask, count;
 841         struct rps_dev_flow_table *table, *old_table;
 842         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 843         int rc;
 844
 845         if (!capable(CAP_NET_ADMIN))
 846                 return -EPERM;
 847
 848         rc = kstrtoul(buf, 0, &count);
 849         if (rc < 0)
 850                 return rc;
 851
 852         if (count) {
 853                 mask = count - 1;
 854                 /* mask = roundup_pow_of_two(count) - 1;
 855                  * without overflows...
 856                  */
 857                 while ((mask | (mask >> 1)) != mask)
 858                         mask |= (mask >> 1);
 859                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 860                  * and on 32bit arches, must check
 861                  * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 862                  */
 863 #if BITS_PER_LONG > 32
 864                 if (mask > (unsigned long)(u32)mask)
 865                         return -EINVAL;
 866 #else
 867                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 868                                 / sizeof(struct rps_dev_flow)) {
 869                         /* Enforce a limit to prevent overflow */
 870                         return -EINVAL;
 871                 }
 872 #endif
 873                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 874                 if (!table)
 875                         return -ENOMEM;
 876
 877                 table->mask = mask;
 878                 for (count = 0; count <= mask; count++)
 879                         table->flows[count].cpu = RPS_NO_CPU;
 880         } else {
 881                 table = NULL;
 882         }
 883
 884         spin_lock(&rps_dev_flow_lock);
 885         old_table = rcu_dereference_protected(queue->rps_flow_table,
 886                                               lockdep_is_held(&rps_dev_flow_lock));
 887         rcu_assign_pointer(queue->rps_flow_table, table);
 888         spin_unlock(&rps_dev_flow_lock);
 889
 890         if (old_table)
 891                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 892
 893         return len;
 894 }
 895
 896 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
 897         = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
 898
 899 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
 900         = __ATTR(rps_flow_cnt, 0644,
 901                  show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 902 #endif /* CONFIG_RPS */
 903
 904 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
 905 #ifdef CONFIG_RPS
 906         &rps_cpus_attribute.attr,
 907         &rps_dev_flow_table_cnt_attribute.attr,
 908 #endif
 909         NULL
 910 };
 911 ATTRIBUTE_GROUPS(rx_queue_default);
 912
 913 static void rx_queue_release(struct kobject *kobj)
 914 {
 915         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 916 #ifdef CONFIG_RPS
 917         struct rps_map *map;
 918         struct rps_dev_flow_table *flow_table;
 919
 920         map = rcu_dereference_protected(queue->rps_map, 1);
 921         if (map) {
 922                 RCU_INIT_POINTER(queue->rps_map, NULL);
 923                 kfree_rcu(map, rcu);
 924         }
 925
 926         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 927         if (flow_table) {
 928                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 929                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 930         }
 931 #endif
 932
 933         memset(kobj, 0, sizeof(*kobj));
 934         dev_put(queue->dev);
 935 }
 936
 937 static const void *rx_queue_namespace(struct kobject *kobj)
 938 {
 939         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 940         struct device *dev = &queue->dev->dev;
 941         const void *ns = NULL;
 942
 943         if (dev->class && dev->class->ns_type)
 944                 ns = dev->class->namespace(dev);
 945
 946         return ns;
 947 }
 948
 949 static void rx_queue_get_ownership(struct kobject *kobj,
 950                                    kuid_t *uid, kgid_t *gid)
 951 {
 952         const struct net *net = rx_queue_namespace(kobj);
 953
 954         net_ns_get_ownership(net, uid, gid);
 955 }
 956
 957 static struct kobj_type rx_queue_ktype __ro_after_init = {
 958         .sysfs_ops = &rx_queue_sysfs_ops,
 959         .release = rx_queue_release,
 960         .default_groups = rx_queue_default_groups,
 961         .namespace = rx_queue_namespace,
 962         .get_ownership = rx_queue_get_ownership,
 963 };
 964
 965 static int rx_queue_add_kobject(struct net_device *dev, int index)
 966 {
 967         struct netdev_rx_queue *queue = dev->_rx + index;
 968         struct kobject *kobj = &queue->kobj;
 969         int error = 0;
 970
 971         /* Kobject_put later will trigger rx_queue_release call which
 972          * decreases dev refcount: Take that reference here
 973          */
 974         dev_hold(queue->dev);
 975
 976         kobj->kset = dev->queues_kset;
 977         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 978                                      "rx-%u", index);
 979         if (error)
 980                 goto err;
 981
 982         if (dev->sysfs_rx_queue_group) {
 983                 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 984                 if (error)
 985                         goto err;
 986         }
 987
 988         kobject_uevent(kobj, KOBJ_ADD);
 989
 990         return error;
 991
 992 err:
 993         kobject_put(kobj);
 994         return error;
 995 }
 996 #endif /* CONFIG_SYSFS */
 997
 998 int
 999 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1000 {
1001 #ifdef CONFIG_SYSFS
1002         int i;
1003         int error = 0;
1004
1005 #ifndef CONFIG_RPS
1006         if (!dev->sysfs_rx_queue_group)
1007                 return 0;
1008 #endif
1009         for (i = old_num; i < new_num; i++) {
1010                 error = rx_queue_add_kobject(dev, i);
1011                 if (error) {
1012                         new_num = old_num;
1013                         break;
1014                 }
1015         }
1016
1017         while (--i >= new_num) {
1018                 struct kobject *kobj = &dev->_rx[i].kobj;
1019
1020                 if (!refcount_read(&dev_net(dev)->count))
1021                         kobj->uevent_suppress = 1;
1022                 if (dev->sysfs_rx_queue_group)
1023                         sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1024                 kobject_put(kobj);
1025         }
1026
1027         return error;
1028 #else
1029         return 0;
1030 #endif
1031 }
1032
1033 #ifdef CONFIG_SYSFS
1034 /*
1035  * netdev_queue sysfs structures and functions.
1036  */
1037 struct netdev_queue_attribute {
1038         struct attribute attr;
1039         ssize_t (*show)(struct netdev_queue *queue, char *buf);
1040         ssize_t (*store)(struct netdev_queue *queue,
1041                          const char *buf, size_t len);
1042 };
1043 #define to_netdev_queue_attr(_attr) \
1044         container_of(_attr, struct netdev_queue_attribute, attr)
1045
1046 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1047
1048 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1049                                       struct attribute *attr, char *buf)
1050 {
1051         const struct netdev_queue_attribute *attribute
1052                 = to_netdev_queue_attr(attr);
1053         struct netdev_queue *queue = to_netdev_queue(kobj);
1054
1055         if (!attribute->show)
1056                 return -EIO;
1057
1058         return attribute->show(queue, buf);
1059 }
1060
1061 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1062                                        struct attribute *attr,
1063                                        const char *buf, size_t count)
1064 {
1065         const struct netdev_queue_attribute *attribute
1066                 = to_netdev_queue_attr(attr);
1067         struct netdev_queue *queue = to_netdev_queue(kobj);
1068
1069         if (!attribute->store)
1070                 return -EIO;
1071
1072         return attribute->store(queue, buf, count);
1073 }
1074
1075 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1076         .show = netdev_queue_attr_show,
1077         .store = netdev_queue_attr_store,
1078 };
1079
1080 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1081 {
1082         unsigned long trans_timeout;
1083
1084         spin_lock_irq(&queue->_xmit_lock);
1085         trans_timeout = queue->trans_timeout;
1086         spin_unlock_irq(&queue->_xmit_lock);
1087
1088         return sprintf(buf, fmt_ulong, trans_timeout);
1089 }
1090
1091 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1092 {
1093         struct net_device *dev = queue->dev;
1094         unsigned int i;
1095
1096         i = queue - dev->_tx;
1097         BUG_ON(i >= dev->num_tx_queues);
1098
1099         return i;
1100 }
1101
1102 static ssize_t traffic_class_show(struct netdev_queue *queue,
1103                                   char *buf)
1104 {
1105         struct net_device *dev = queue->dev;
1106         int index;
1107         int tc;
1108
1109         if (!netif_is_multiqueue(dev))
1110                 return -ENOENT;
1111
1112         index = get_netdev_queue_index(queue);
1113
1114         /* If queue belongs to subordinate dev use its TC mapping */
1115         dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1116
1117         tc = netdev_txq_to_tc(dev, index);
1118         if (tc < 0)
1119                 return -EINVAL;
1120
1121         /* We can report the traffic class one of two ways:
1122          * Subordinate device traffic classes are reported with the traffic
1123          * class first, and then the subordinate class so for example TC0 on
1124          * subordinate device 2 will be reported as "0-2". If the queue
1125          * belongs to the root device it will be reported with just the
1126          * traffic class, so just "0" for TC 0 for example.
1127          */
1128         return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1129                                  sprintf(buf, "%u\n", tc);
1130 }
1131
1132 #ifdef CONFIG_XPS
1133 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1134                                char *buf)
1135 {
1136         return sprintf(buf, "%lu\n", queue->tx_maxrate);
1137 }
1138
1139 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1140                                 const char *buf, size_t len)
1141 {
1142         struct net_device *dev = queue->dev;
1143         int err, index = get_netdev_queue_index(queue);
1144         u32 rate = 0;
1145
1146         if (!capable(CAP_NET_ADMIN))
1147                 return -EPERM;
1148
1149         /* The check is also done later; this helps returning early without
1150          * hitting the trylock/restart below.
1151          */
1152         if (!dev->netdev_ops->ndo_set_tx_maxrate)
1153                 return -EOPNOTSUPP;
1154
1155         err = kstrtou32(buf, 10, &rate);
1156         if (err < 0)
1157                 return err;
1158
1159         if (!rtnl_trylock())
1160                 return restart_syscall();
1161
1162         err = -EOPNOTSUPP;
1163         if (dev->netdev_ops->ndo_set_tx_maxrate)
1164                 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1165
1166         rtnl_unlock();
1167         if (!err) {
1168                 queue->tx_maxrate = rate;
1169                 return len;
1170         }
1171         return err;
1172 }
1173
1174 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1175         = __ATTR_RW(tx_maxrate);
1176 #endif
1177
1178 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1179         = __ATTR_RO(tx_timeout);
1180
1181 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1182         = __ATTR_RO(traffic_class);
1183
1184 #ifdef CONFIG_BQL
1185 /*
1186  * Byte queue limits sysfs structures and functions.
1187  */
1188 static ssize_t bql_show(char *buf, unsigned int value)
1189 {
1190         return sprintf(buf, "%u\n", value);
1191 }
1192
1193 static ssize_t bql_set(const char *buf, const size_t count,
1194                        unsigned int *pvalue)
1195 {
1196         unsigned int value;
1197         int err;
1198
1199         if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1200                 value = DQL_MAX_LIMIT;
1201         } else {
1202                 err = kstrtouint(buf, 10, &value);
1203                 if (err < 0)
1204                         return err;
1205                 if (value > DQL_MAX_LIMIT)
1206                         return -EINVAL;
1207         }
1208
1209         *pvalue = value;
1210
1211         return count;
1212 }
1213
1214 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1215                                   char *buf)
1216 {
1217         struct dql *dql = &queue->dql;
1218
1219         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1220 }
1221
1222 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1223                                  const char *buf, size_t len)
1224 {
1225         struct dql *dql = &queue->dql;
1226         unsigned int value;
1227         int err;
1228
1229         err = kstrtouint(buf, 10, &value);
1230         if (err < 0)
1231                 return err;
1232
1233         dql->slack_hold_time = msecs_to_jiffies(value);
1234
1235         return len;
1236 }
1237
1238 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1239         = __ATTR(hold_time, 0644,
1240                  bql_show_hold_time, bql_set_hold_time);
1241
1242 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1243                                  char *buf)
1244 {
1245         struct dql *dql = &queue->dql;
1246
1247         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1248 }
1249
1250 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1251         __ATTR(inflight, 0444, bql_show_inflight, NULL);
1252
1253 #define BQL_ATTR(NAME, FIELD)                                           \
1254 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1255                                  char *buf)                             \
1256 {                                                                       \
1257         return bql_show(buf, queue->dql.FIELD);                         \
1258 }                                                                       \
1259                                                                         \
1260 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1261                                 const char *buf, size_t len)            \
1262 {                                                                       \
1263         return bql_set(buf, len, &queue->dql.FIELD);                    \
1264 }                                                                       \
1265                                                                         \
1266 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1267         = __ATTR(NAME, 0644,                            \
1268                  bql_show_ ## NAME, bql_set_ ## NAME)
1269
1270 BQL_ATTR(limit, limit);
1271 BQL_ATTR(limit_max, max_limit);
1272 BQL_ATTR(limit_min, min_limit);
1273
1274 static struct attribute *dql_attrs[] __ro_after_init = {
1275         &bql_limit_attribute.attr,
1276         &bql_limit_max_attribute.attr,
1277         &bql_limit_min_attribute.attr,
1278         &bql_hold_time_attribute.attr,
1279         &bql_inflight_attribute.attr,
1280         NULL
1281 };
1282
1283 static const struct attribute_group dql_group = {
1284         .name  = "byte_queue_limits",
1285         .attrs  = dql_attrs,
1286 };
1287 #endif /* CONFIG_BQL */
1288
1289 #ifdef CONFIG_XPS
1290 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1291                              char *buf)
1292 {
1293         int cpu, len, ret, num_tc = 1, tc = 0;
1294         struct net_device *dev = queue->dev;
1295         struct xps_dev_maps *dev_maps;
1296         cpumask_var_t mask;
1297         unsigned long index;
1298
1299         if (!netif_is_multiqueue(dev))
1300                 return -ENOENT;
1301
1302         index = get_netdev_queue_index(queue);
1303
1304         if (!rtnl_trylock())
1305                 return restart_syscall();
1306
1307         if (dev->num_tc) {
1308                 /* Do not allow XPS on subordinate device directly */
1309                 num_tc = dev->num_tc;
1310                 if (num_tc < 0) {
1311                         ret = -EINVAL;
1312                         goto err_rtnl_unlock;
1313                 }
1314
1315                 /* If queue belongs to subordinate dev use its map */
1316                 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1317
1318                 tc = netdev_txq_to_tc(dev, index);
1319                 if (tc < 0) {
1320                         ret = -EINVAL;
1321                         goto err_rtnl_unlock;
1322                 }
1323         }
1324
1325         if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1326                 ret = -ENOMEM;
1327                 goto err_rtnl_unlock;
1328         }
1329
1330         rcu_read_lock();
1331         dev_maps = rcu_dereference(dev->xps_cpus_map);
1332         if (dev_maps) {
1333                 for_each_possible_cpu(cpu) {
1334                         int i, tci = cpu * num_tc + tc;
1335                         struct xps_map *map;
1336
1337                         map = rcu_dereference(dev_maps->attr_map[tci]);
1338                         if (!map)
1339                                 continue;
1340
1341                         for (i = map->len; i--;) {
1342                                 if (map->queues[i] == index) {
1343                                         cpumask_set_cpu(cpu, mask);
1344                                         break;
1345                                 }
1346                         }
1347                 }
1348         }
1349         rcu_read_unlock();
1350
1351         rtnl_unlock();
1352
1353         len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1354         free_cpumask_var(mask);
1355         return len < PAGE_SIZE ? len : -EINVAL;
1356
1357 err_rtnl_unlock:
1358         rtnl_unlock();
1359         return ret;
1360 }
1361
1362 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1363                               const char *buf, size_t len)
1364 {
1365         struct net_device *dev = queue->dev;
1366         unsigned long index;
1367         cpumask_var_t mask;
1368         int err;
1369
1370         if (!netif_is_multiqueue(dev))
1371                 return -ENOENT;
1372
1373         if (!capable(CAP_NET_ADMIN))
1374                 return -EPERM;
1375
1376         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1377                 return -ENOMEM;
1378
1379         index = get_netdev_queue_index(queue);
1380
1381         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1382         if (err) {
1383                 free_cpumask_var(mask);
1384                 return err;
1385         }
1386
1387         if (!rtnl_trylock()) {
1388                 free_cpumask_var(mask);
1389                 return restart_syscall();
1390         }
1391
1392         err = netif_set_xps_queue(dev, mask, index);
1393         rtnl_unlock();
1394
1395         free_cpumask_var(mask);
1396
1397         return err ? : len;
1398 }
1399
1400 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1401         = __ATTR_RW(xps_cpus);
1402
1403 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1404 {
1405         int j, len, ret, num_tc = 1, tc = 0;
1406         struct net_device *dev = queue->dev;
1407         struct xps_dev_maps *dev_maps;
1408         unsigned long *mask, index;
1409
1410         index = get_netdev_queue_index(queue);
1411
1412         if (!rtnl_trylock())
1413                 return restart_syscall();
1414
1415         if (dev->num_tc) {
1416                 num_tc = dev->num_tc;
1417                 tc = netdev_txq_to_tc(dev, index);
1418                 if (tc < 0) {
1419                         ret = -EINVAL;
1420                         goto err_rtnl_unlock;
1421                 }
1422         }
1423         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1424         if (!mask) {
1425                 ret = -ENOMEM;
1426                 goto err_rtnl_unlock;
1427         }
1428
1429         rcu_read_lock();
1430         dev_maps = rcu_dereference(dev->xps_rxqs_map);
1431         if (!dev_maps)
1432                 goto out_no_maps;
1433
1434         for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1435              j < dev->num_rx_queues;) {
1436                 int i, tci = j * num_tc + tc;
1437                 struct xps_map *map;
1438
1439                 map = rcu_dereference(dev_maps->attr_map[tci]);
1440                 if (!map)
1441                         continue;
1442
1443                 for (i = map->len; i--;) {
1444                         if (map->queues[i] == index) {
1445                                 set_bit(j, mask);
1446                                 break;
1447                         }
1448                 }
1449         }
1450 out_no_maps:
1451         rcu_read_unlock();
1452
1453         rtnl_unlock();
1454
1455         len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1456         bitmap_free(mask);
1457
1458         return len < PAGE_SIZE ? len : -EINVAL;
1459
1460 err_rtnl_unlock:
1461         rtnl_unlock();
1462         return ret;
1463 }
1464
1465 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1466                               size_t len)
1467 {
1468         struct net_device *dev = queue->dev;
1469         struct net *net = dev_net(dev);
1470         unsigned long *mask, index;
1471         int err;
1472
1473         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1474                 return -EPERM;
1475
1476         mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1477         if (!mask)
1478                 return -ENOMEM;
1479
1480         index = get_netdev_queue_index(queue);
1481
1482         err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1483         if (err) {
1484                 bitmap_free(mask);
1485                 return err;
1486         }
1487
1488         if (!rtnl_trylock()) {
1489                 bitmap_free(mask);
1490                 return restart_syscall();
1491         }
1492
1493         cpus_read_lock();
1494         err = __netif_set_xps_queue(dev, mask, index, true);
1495         cpus_read_unlock();
1496
1497         rtnl_unlock();
1498
1499         bitmap_free(mask);
1500         return err ? : len;
1501 }
1502
1503 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1504         = __ATTR_RW(xps_rxqs);
1505 #endif /* CONFIG_XPS */
1506
1507 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1508         &queue_trans_timeout.attr,
1509         &queue_traffic_class.attr,
1510 #ifdef CONFIG_XPS
1511         &xps_cpus_attribute.attr,
1512         &xps_rxqs_attribute.attr,
1513         &queue_tx_maxrate.attr,
1514 #endif
1515         NULL
1516 };
1517 ATTRIBUTE_GROUPS(netdev_queue_default);
1518
1519 static void netdev_queue_release(struct kobject *kobj)
1520 {
1521         struct netdev_queue *queue = to_netdev_queue(kobj);
1522
1523         memset(kobj, 0, sizeof(*kobj));
1524         dev_put(queue->dev);
1525 }
1526
1527 static const void *netdev_queue_namespace(struct kobject *kobj)
1528 {
1529         struct netdev_queue *queue = to_netdev_queue(kobj);
1530         struct device *dev = &queue->dev->dev;
1531         const void *ns = NULL;
1532
1533         if (dev->class && dev->class->ns_type)
1534                 ns = dev->class->namespace(dev);
1535
1536         return ns;
1537 }
1538
1539 static void netdev_queue_get_ownership(struct kobject *kobj,
1540                                        kuid_t *uid, kgid_t *gid)
1541 {
1542         const struct net *net = netdev_queue_namespace(kobj);
1543
1544         net_ns_get_ownership(net, uid, gid);
1545 }
1546
1547 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1548         .sysfs_ops = &netdev_queue_sysfs_ops,
1549         .release = netdev_queue_release,
1550         .default_groups = netdev_queue_default_groups,
1551         .namespace = netdev_queue_namespace,
1552         .get_ownership = netdev_queue_get_ownership,
1553 };
1554
1555 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1556 {
1557         struct netdev_queue *queue = dev->_tx + index;
1558         struct kobject *kobj = &queue->kobj;
1559         int error = 0;
1560
1561         /* Kobject_put later will trigger netdev_queue_release call
1562          * which decreases dev refcount: Take that reference here
1563          */
1564         dev_hold(queue->dev);
1565
1566         kobj->kset = dev->queues_kset;
1567         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1568                                      "tx-%u", index);
1569         if (error)
1570                 goto err;
1571
1572 #ifdef CONFIG_BQL
1573         error = sysfs_create_group(kobj, &dql_group);
1574         if (error)
1575                 goto err;
1576 #endif
1577
1578         kobject_uevent(kobj, KOBJ_ADD);
1579         return 0;
1580
1581 err:
1582         kobject_put(kobj);
1583         return error;
1584 }
1585 #endif /* CONFIG_SYSFS */
1586
1587 int
1588 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1589 {
1590 #ifdef CONFIG_SYSFS
1591         int i;
1592         int error = 0;
1593
1594         for (i = old_num; i < new_num; i++) {
1595                 error = netdev_queue_add_kobject(dev, i);
1596                 if (error) {
1597                         new_num = old_num;
1598                         break;
1599                 }
1600         }
1601
1602         while (--i >= new_num) {
1603                 struct netdev_queue *queue = dev->_tx + i;
1604
1605                 if (!refcount_read(&dev_net(dev)->count))
1606                         queue->kobj.uevent_suppress = 1;
1607 #ifdef CONFIG_BQL
1608                 sysfs_remove_group(&queue->kobj, &dql_group);
1609 #endif
1610                 kobject_put(&queue->kobj);
1611         }
1612
1613         return error;
1614 #else
1615         return 0;
1616 #endif /* CONFIG_SYSFS */
1617 }
1618
1619 static int register_queue_kobjects(struct net_device *dev)
1620 {
1621         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1622
1623 #ifdef CONFIG_SYSFS
1624         dev->queues_kset = kset_create_and_add("queues",
1625                                                NULL, &dev->dev.kobj);
1626         if (!dev->queues_kset)
1627                 return -ENOMEM;
1628         real_rx = dev->real_num_rx_queues;
1629 #endif
1630         real_tx = dev->real_num_tx_queues;
1631
1632         error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1633         if (error)
1634                 goto error;
1635         rxq = real_rx;
1636
1637         error = netdev_queue_update_kobjects(dev, 0, real_tx);
1638         if (error)
1639                 goto error;
1640         txq = real_tx;
1641
1642         return 0;
1643
1644 error:
1645         netdev_queue_update_kobjects(dev, txq, 0);
1646         net_rx_queue_update_kobjects(dev, rxq, 0);
1647 #ifdef CONFIG_SYSFS
1648         kset_unregister(dev->queues_kset);
1649 #endif
1650         return error;
1651 }
1652
1653 static void remove_queue_kobjects(struct net_device *dev)
1654 {
1655         int real_rx = 0, real_tx = 0;
1656
1657 #ifdef CONFIG_SYSFS
1658         real_rx = dev->real_num_rx_queues;
1659 #endif
1660         real_tx = dev->real_num_tx_queues;
1661
1662         net_rx_queue_update_kobjects(dev, real_rx, 0);
1663         netdev_queue_update_kobjects(dev, real_tx, 0);
1664
1665         dev->real_num_rx_queues = 0;
1666         dev->real_num_tx_queues = 0;
1667 #ifdef CONFIG_SYSFS
1668         kset_unregister(dev->queues_kset);
1669 #endif
1670 }
1671
1672 static bool net_current_may_mount(void)
1673 {
1674         struct net *net = current->nsproxy->net_ns;
1675
1676         return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1677 }
1678
1679 static void *net_grab_current_ns(void)
1680 {
1681         struct net *ns = current->nsproxy->net_ns;
1682 #ifdef CONFIG_NET_NS
1683         if (ns)
1684                 refcount_inc(&ns->passive);
1685 #endif
1686         return ns;
1687 }
1688
1689 static const void *net_initial_ns(void)
1690 {
1691         return &init_net;
1692 }
1693
1694 static const void *net_netlink_ns(struct sock *sk)
1695 {
1696         return sock_net(sk);
1697 }
1698
1699 const struct kobj_ns_type_operations net_ns_type_operations = {
1700         .type = KOBJ_NS_TYPE_NET,
1701         .current_may_mount = net_current_may_mount,
1702         .grab_current_ns = net_grab_current_ns,
1703         .netlink_ns = net_netlink_ns,
1704         .initial_ns = net_initial_ns,
1705         .drop_ns = net_drop_ns,
1706 };
1707 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1708
1709 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1710 {
1711         struct net_device *dev = to_net_dev(d);
1712         int retval;
1713
1714         /* pass interface to uevent. */
1715         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1716         if (retval)
1717                 goto exit;
1718
1719         /* pass ifindex to uevent.
1720          * ifindex is useful as it won't change (interface name may change)
1721          * and is what RtNetlink uses natively.
1722          */
1723         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1724
1725 exit:
1726         return retval;
1727 }
1728
1729 /*
1730  *      netdev_release -- destroy and free a dead device.
1731  *      Called when last reference to device kobject is gone.
1732  */
1733 static void netdev_release(struct device *d)
1734 {
1735         struct net_device *dev = to_net_dev(d);
1736
1737         BUG_ON(dev->reg_state != NETREG_RELEASED);
1738
1739         /* no need to wait for rcu grace period:
1740          * device is dead and about to be freed.
1741          */
1742         kfree(rcu_access_pointer(dev->ifalias));
1743         netdev_freemem(dev);
1744 }
1745
1746 static const void *net_namespace(struct device *d)
1747 {
1748         struct net_device *dev = to_net_dev(d);
1749
1750         return dev_net(dev);
1751 }
1752
1753 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1754 {
1755         struct net_device *dev = to_net_dev(d);
1756         const struct net *net = dev_net(dev);
1757
1758         net_ns_get_ownership(net, uid, gid);
1759 }
1760
1761 static struct class net_class __ro_after_init = {
1762         .name = "net",
1763         .dev_release = netdev_release,
1764         .dev_groups = net_class_groups,
1765         .dev_uevent = netdev_uevent,
1766         .ns_type = &net_ns_type_operations,
1767         .namespace = net_namespace,
1768         .get_ownership = net_get_ownership,
1769 };
1770
1771 #ifdef CONFIG_OF_NET
1772 static int of_dev_node_match(struct device *dev, const void *data)
1773 {
1774         int ret = 0;
1775
1776         if (dev->parent)
1777                 ret = dev->parent->of_node == data;
1778
1779         return ret == 0 ? dev->of_node == data : ret;
1780 }
1781
1782 /*
1783  * of_find_net_device_by_node - lookup the net device for the device node
1784  * @np: OF device node
1785  *
1786  * Looks up the net_device structure corresponding with the device node.
1787  * If successful, returns a pointer to the net_device with the embedded
1788  * struct device refcount incremented by one, or NULL on failure. The
1789  * refcount must be dropped when done with the net_device.
1790  */
1791 struct net_device *of_find_net_device_by_node(struct device_node *np)
1792 {
1793         struct device *dev;
1794
1795         dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1796         if (!dev)
1797                 return NULL;
1798
1799         return to_net_dev(dev);
1800 }
1801 EXPORT_SYMBOL(of_find_net_device_by_node);
1802 #endif
1803
1804 /* Delete sysfs entries but hold kobject reference until after all
1805  * netdev references are gone.
1806  */
1807 void netdev_unregister_kobject(struct net_device *ndev)
1808 {
1809         struct device *dev = &ndev->dev;
1810
1811         if (!refcount_read(&dev_net(ndev)->count))
1812                 dev_set_uevent_suppress(dev, 1);
1813
1814         kobject_get(&dev->kobj);
1815
1816         remove_queue_kobjects(ndev);
1817
1818         pm_runtime_set_memalloc_noio(dev, false);
1819
1820         device_del(dev);
1821 }
1822
1823 /* Create sysfs entries for network device. */
1824 int netdev_register_kobject(struct net_device *ndev)
1825 {
1826         struct device *dev = &ndev->dev;
1827         const struct attribute_group **groups = ndev->sysfs_groups;
1828         int error = 0;
1829
1830         device_initialize(dev);
1831         dev->class = &net_class;
1832         dev->platform_data = ndev;
1833         dev->groups = groups;
1834
1835         dev_set_name(dev, "%s", ndev->name);
1836
1837 #ifdef CONFIG_SYSFS
1838         /* Allow for a device specific group */
1839         if (*groups)
1840                 groups++;
1841
1842         *groups++ = &netstat_group;
1843
1844 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1845         if (ndev->ieee80211_ptr)
1846                 *groups++ = &wireless_group;
1847 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1848         else if (ndev->wireless_handlers)
1849                 *groups++ = &wireless_group;
1850 #endif
1851 #endif
1852 #endif /* CONFIG_SYSFS */
1853
1854         error = device_add(dev);
1855         if (error)
1856                 return error;
1857
1858         error = register_queue_kobjects(ndev);
1859         if (error) {
1860                 device_del(dev);
1861                 return error;
1862         }
1863
1864         pm_runtime_set_memalloc_noio(dev, true);
1865
1866         return error;
1867 }
1868
1869 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1870                                 const void *ns)
1871 {
1872         return class_create_file_ns(&net_class, class_attr, ns);
1873 }
1874 EXPORT_SYMBOL(netdev_class_create_file_ns);
1875
1876 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1877                                  const void *ns)
1878 {
1879         class_remove_file_ns(&net_class, class_attr, ns);
1880 }
1881 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1882
1883 int __init netdev_kobject_init(void)
1884 {
1885         kobj_ns_type_register(&net_ns_type_operations);
1886         return class_register(&net_class);
1887 }