net/core/net-sysfs.c

   1 /*
   2  * net-sysfs.c - network device class and attributes
   3  *
   4  * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/kernel.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/slab.h>
  17 #include <linux/nsproxy.h>
  18 #include <net/sock.h>
  19 #include <net/net_namespace.h>
  20 #include <linux/rtnetlink.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/export.h>
  23 #include <linux/jiffies.h>
  24 #include <linux/pm_runtime.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_long_hex[] = "%#lx\n";
  31 static const char fmt_dec[] = "%d\n";
  32 static const char fmt_udec[] = "%u\n";
  33 static const char fmt_ulong[] = "%lu\n";
  34 static const char fmt_u64[] = "%llu\n";
  35
  36 static inline int dev_isalive(const struct net_device *dev)
  37 {
  38         return dev->reg_state <= NETREG_REGISTERED;
  39 }
  40
  41 /* use same locking rules as GIF* ioctl's */
  42 static ssize_t netdev_show(const struct device *dev,
  43                            struct device_attribute *attr, char *buf,
  44                            ssize_t (*format)(const struct net_device *, char *))
  45 {
  46         struct net_device *net = to_net_dev(dev);
  47         ssize_t ret = -EINVAL;
  48
  49         read_lock(&dev_base_lock);
  50         if (dev_isalive(net))
  51                 ret = (*format)(net, buf);
  52         read_unlock(&dev_base_lock);
  53
  54         return ret;
  55 }
  56
  57 /* generate a show function for simple field */
  58 #define NETDEVICE_SHOW(field, format_string)                            \
  59 static ssize_t format_##field(const struct net_device *net, char *buf)  \
  60 {                                                                       \
  61         return sprintf(buf, format_string, net->field);                 \
  62 }                                                                       \
  63 static ssize_t show_##field(struct device *dev,                         \
  64                             struct device_attribute *attr, char *buf)   \
  65 {                                                                       \
  66         return netdev_show(dev, attr, buf, format_##field);             \
  67 }
  68
  69
  70 /* use same locking and permission rules as SIF* ioctl's */
  71 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  72                             const char *buf, size_t len,
  73                             int (*set)(struct net_device *, unsigned long))
  74 {
  75         struct net_device *netdev = to_net_dev(dev);
  76         struct net *net = dev_net(netdev);
  77         unsigned long new;
  78         int ret = -EINVAL;
  79
  80         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  81                 return -EPERM;
  82
  83         ret = kstrtoul(buf, 0, &new);
  84         if (ret)
  85                 goto err;
  86
  87         if (!rtnl_trylock())
  88                 return restart_syscall();
  89
  90         if (dev_isalive(netdev)) {
  91                 if ((ret = (*set)(netdev, new)) == 0)
  92                         ret = len;
  93         }
  94         rtnl_unlock();
  95  err:
  96         return ret;
  97 }
  98
  99 NETDEVICE_SHOW(dev_id, fmt_hex);
 100 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 101 NETDEVICE_SHOW(addr_len, fmt_dec);
 102 NETDEVICE_SHOW(iflink, fmt_dec);
 103 NETDEVICE_SHOW(ifindex, fmt_dec);
 104 NETDEVICE_SHOW(type, fmt_dec);
 105 NETDEVICE_SHOW(link_mode, fmt_dec);
 106
 107 /* use same locking rules as GIFHWADDR ioctl's */
 108 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 109                             char *buf)
 110 {
 111         struct net_device *net = to_net_dev(dev);
 112         ssize_t ret = -EINVAL;
 113
 114         read_lock(&dev_base_lock);
 115         if (dev_isalive(net))
 116                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 117         read_unlock(&dev_base_lock);
 118         return ret;
 119 }
 120
 121 static ssize_t show_broadcast(struct device *dev,
 122                             struct device_attribute *attr, char *buf)
 123 {
 124         struct net_device *net = to_net_dev(dev);
 125         if (dev_isalive(net))
 126                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 127         return -EINVAL;
 128 }
 129
 130 static int change_carrier(struct net_device *net, unsigned long new_carrier)
 131 {
 132         if (!netif_running(net))
 133                 return -EINVAL;
 134         return dev_change_carrier(net, (bool) new_carrier);
 135 }
 136
 137 static ssize_t store_carrier(struct device *dev, struct device_attribute *attr,
 138                          const char *buf, size_t len)
 139 {
 140         return netdev_store(dev, attr, buf, len, change_carrier);
 141 }
 142
 143 static ssize_t show_carrier(struct device *dev,
 144                             struct device_attribute *attr, char *buf)
 145 {
 146         struct net_device *netdev = to_net_dev(dev);
 147         if (netif_running(netdev)) {
 148                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 149         }
 150         return -EINVAL;
 151 }
 152
 153 static ssize_t show_speed(struct device *dev,
 154                           struct device_attribute *attr, char *buf)
 155 {
 156         struct net_device *netdev = to_net_dev(dev);
 157         int ret = -EINVAL;
 158
 159         if (!rtnl_trylock())
 160                 return restart_syscall();
 161
 162         if (netif_running(netdev)) {
 163                 struct ethtool_cmd cmd;
 164                 if (!__ethtool_get_settings(netdev, &cmd))
 165                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 166         }
 167         rtnl_unlock();
 168         return ret;
 169 }
 170
 171 static ssize_t show_duplex(struct device *dev,
 172                            struct device_attribute *attr, char *buf)
 173 {
 174         struct net_device *netdev = to_net_dev(dev);
 175         int ret = -EINVAL;
 176
 177         if (!rtnl_trylock())
 178                 return restart_syscall();
 179
 180         if (netif_running(netdev)) {
 181                 struct ethtool_cmd cmd;
 182                 if (!__ethtool_get_settings(netdev, &cmd)) {
 183                         const char *duplex;
 184                         switch (cmd.duplex) {
 185                         case DUPLEX_HALF:
 186                                 duplex = "half";
 187                                 break;
 188                         case DUPLEX_FULL:
 189                                 duplex = "full";
 190                                 break;
 191                         default:
 192                                 duplex = "unknown";
 193                                 break;
 194                         }
 195                         ret = sprintf(buf, "%s\n", duplex);
 196                 }
 197         }
 198         rtnl_unlock();
 199         return ret;
 200 }
 201
 202 static ssize_t show_dormant(struct device *dev,
 203                             struct device_attribute *attr, char *buf)
 204 {
 205         struct net_device *netdev = to_net_dev(dev);
 206
 207         if (netif_running(netdev))
 208                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 209
 210         return -EINVAL;
 211 }
 212
 213 static const char *const operstates[] = {
 214         "unknown",
 215         "notpresent", /* currently unused */
 216         "down",
 217         "lowerlayerdown",
 218         "testing", /* currently unused */
 219         "dormant",
 220         "up"
 221 };
 222
 223 static ssize_t show_operstate(struct device *dev,
 224                               struct device_attribute *attr, char *buf)
 225 {
 226         const struct net_device *netdev = to_net_dev(dev);
 227         unsigned char operstate;
 228
 229         read_lock(&dev_base_lock);
 230         operstate = netdev->operstate;
 231         if (!netif_running(netdev))
 232                 operstate = IF_OPER_DOWN;
 233         read_unlock(&dev_base_lock);
 234
 235         if (operstate >= ARRAY_SIZE(operstates))
 236                 return -EINVAL; /* should not happen */
 237
 238         return sprintf(buf, "%s\n", operstates[operstate]);
 239 }
 240
 241 /* read-write attributes */
 242 NETDEVICE_SHOW(mtu, fmt_dec);
 243
 244 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 245 {
 246         return dev_set_mtu(net, (int) new_mtu);
 247 }
 248
 249 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 250                          const char *buf, size_t len)
 251 {
 252         return netdev_store(dev, attr, buf, len, change_mtu);
 253 }
 254
 255 NETDEVICE_SHOW(flags, fmt_hex);
 256
 257 static int change_flags(struct net_device *net, unsigned long new_flags)
 258 {
 259         return dev_change_flags(net, (unsigned int) new_flags);
 260 }
 261
 262 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 263                            const char *buf, size_t len)
 264 {
 265         return netdev_store(dev, attr, buf, len, change_flags);
 266 }
 267
 268 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 269
 270 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 271 {
 272         net->tx_queue_len = new_len;
 273         return 0;
 274 }
 275
 276 static ssize_t store_tx_queue_len(struct device *dev,
 277                                   struct device_attribute *attr,
 278                                   const char *buf, size_t len)
 279 {
 280         if (!capable(CAP_NET_ADMIN))
 281                 return -EPERM;
 282
 283         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 284 }
 285
 286 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 287                              const char *buf, size_t len)
 288 {
 289         struct net_device *netdev = to_net_dev(dev);
 290         struct net *net = dev_net(netdev);
 291         size_t count = len;
 292         ssize_t ret;
 293
 294         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 295                 return -EPERM;
 296
 297         /* ignore trailing newline */
 298         if (len >  0 && buf[len - 1] == '\n')
 299                 --count;
 300
 301         if (!rtnl_trylock())
 302                 return restart_syscall();
 303         ret = dev_set_alias(netdev, buf, count);
 304         rtnl_unlock();
 305
 306         return ret < 0 ? ret : len;
 307 }
 308
 309 static ssize_t show_ifalias(struct device *dev,
 310                             struct device_attribute *attr, char *buf)
 311 {
 312         const struct net_device *netdev = to_net_dev(dev);
 313         ssize_t ret = 0;
 314
 315         if (!rtnl_trylock())
 316                 return restart_syscall();
 317         if (netdev->ifalias)
 318                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 319         rtnl_unlock();
 320         return ret;
 321 }
 322
 323 NETDEVICE_SHOW(group, fmt_dec);
 324
 325 static int change_group(struct net_device *net, unsigned long new_group)
 326 {
 327         dev_set_group(net, (int) new_group);
 328         return 0;
 329 }
 330
 331 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 332                          const char *buf, size_t len)
 333 {
 334         return netdev_store(dev, attr, buf, len, change_group);
 335 }
 336
 337 static struct device_attribute net_class_attributes[] = {
 338         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 339         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 340         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 341         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 342         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 343         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 344         __ATTR(type, S_IRUGO, show_type, NULL),
 345         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 346         __ATTR(address, S_IRUGO, show_address, NULL),
 347         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 348         __ATTR(carrier, S_IRUGO | S_IWUSR, show_carrier, store_carrier),
 349         __ATTR(speed, S_IRUGO, show_speed, NULL),
 350         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 351         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 352         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 353         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 354         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 355         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 356                store_tx_queue_len),
 357         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 358         {}
 359 };
 360
 361 /* Show a given an attribute in the statistics group */
 362 static ssize_t netstat_show(const struct device *d,
 363                             struct device_attribute *attr, char *buf,
 364                             unsigned long offset)
 365 {
 366         struct net_device *dev = to_net_dev(d);
 367         ssize_t ret = -EINVAL;
 368
 369         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 370                         offset % sizeof(u64) != 0);
 371
 372         read_lock(&dev_base_lock);
 373         if (dev_isalive(dev)) {
 374                 struct rtnl_link_stats64 temp;
 375                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 376
 377                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 378         }
 379         read_unlock(&dev_base_lock);
 380         return ret;
 381 }
 382
 383 /* generate a read-only statistics attribute */
 384 #define NETSTAT_ENTRY(name)                                             \
 385 static ssize_t show_##name(struct device *d,                            \
 386                            struct device_attribute *attr, char *buf)    \
 387 {                                                                       \
 388         return netstat_show(d, attr, buf,                               \
 389                             offsetof(struct rtnl_link_stats64, name));  \
 390 }                                                                       \
 391 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 392
 393 NETSTAT_ENTRY(rx_packets);
 394 NETSTAT_ENTRY(tx_packets);
 395 NETSTAT_ENTRY(rx_bytes);
 396 NETSTAT_ENTRY(tx_bytes);
 397 NETSTAT_ENTRY(rx_errors);
 398 NETSTAT_ENTRY(tx_errors);
 399 NETSTAT_ENTRY(rx_dropped);
 400 NETSTAT_ENTRY(tx_dropped);
 401 NETSTAT_ENTRY(multicast);
 402 NETSTAT_ENTRY(collisions);
 403 NETSTAT_ENTRY(rx_length_errors);
 404 NETSTAT_ENTRY(rx_over_errors);
 405 NETSTAT_ENTRY(rx_crc_errors);
 406 NETSTAT_ENTRY(rx_frame_errors);
 407 NETSTAT_ENTRY(rx_fifo_errors);
 408 NETSTAT_ENTRY(rx_missed_errors);
 409 NETSTAT_ENTRY(tx_aborted_errors);
 410 NETSTAT_ENTRY(tx_carrier_errors);
 411 NETSTAT_ENTRY(tx_fifo_errors);
 412 NETSTAT_ENTRY(tx_heartbeat_errors);
 413 NETSTAT_ENTRY(tx_window_errors);
 414 NETSTAT_ENTRY(rx_compressed);
 415 NETSTAT_ENTRY(tx_compressed);
 416
 417 static struct attribute *netstat_attrs[] = {
 418         &dev_attr_rx_packets.attr,
 419         &dev_attr_tx_packets.attr,
 420         &dev_attr_rx_bytes.attr,
 421         &dev_attr_tx_bytes.attr,
 422         &dev_attr_rx_errors.attr,
 423         &dev_attr_tx_errors.attr,
 424         &dev_attr_rx_dropped.attr,
 425         &dev_attr_tx_dropped.attr,
 426         &dev_attr_multicast.attr,
 427         &dev_attr_collisions.attr,
 428         &dev_attr_rx_length_errors.attr,
 429         &dev_attr_rx_over_errors.attr,
 430         &dev_attr_rx_crc_errors.attr,
 431         &dev_attr_rx_frame_errors.attr,
 432         &dev_attr_rx_fifo_errors.attr,
 433         &dev_attr_rx_missed_errors.attr,
 434         &dev_attr_tx_aborted_errors.attr,
 435         &dev_attr_tx_carrier_errors.attr,
 436         &dev_attr_tx_fifo_errors.attr,
 437         &dev_attr_tx_heartbeat_errors.attr,
 438         &dev_attr_tx_window_errors.attr,
 439         &dev_attr_rx_compressed.attr,
 440         &dev_attr_tx_compressed.attr,
 441         NULL
 442 };
 443
 444
 445 static struct attribute_group netstat_group = {
 446         .name  = "statistics",
 447         .attrs  = netstat_attrs,
 448 };
 449
 450 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 451 static struct attribute *wireless_attrs[] = {
 452         NULL
 453 };
 454
 455 static struct attribute_group wireless_group = {
 456         .name = "wireless",
 457         .attrs = wireless_attrs,
 458 };
 459 #endif
 460 #endif /* CONFIG_SYSFS */
 461
 462 #ifdef CONFIG_RPS
 463 /*
 464  * RX queue sysfs structures and functions.
 465  */
 466 struct rx_queue_attribute {
 467         struct attribute attr;
 468         ssize_t (*show)(struct netdev_rx_queue *queue,
 469             struct rx_queue_attribute *attr, char *buf);
 470         ssize_t (*store)(struct netdev_rx_queue *queue,
 471             struct rx_queue_attribute *attr, const char *buf, size_t len);
 472 };
 473 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 474     struct rx_queue_attribute, attr)
 475
 476 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 477
 478 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 479                                   char *buf)
 480 {
 481         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 482         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 483
 484         if (!attribute->show)
 485                 return -EIO;
 486
 487         return attribute->show(queue, attribute, buf);
 488 }
 489
 490 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 491                                    const char *buf, size_t count)
 492 {
 493         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 494         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 495
 496         if (!attribute->store)
 497                 return -EIO;
 498
 499         return attribute->store(queue, attribute, buf, count);
 500 }
 501
 502 static const struct sysfs_ops rx_queue_sysfs_ops = {
 503         .show = rx_queue_attr_show,
 504         .store = rx_queue_attr_store,
 505 };
 506
 507 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 508                             struct rx_queue_attribute *attribute, char *buf)
 509 {
 510         struct rps_map *map;
 511         cpumask_var_t mask;
 512         size_t len = 0;
 513         int i;
 514
 515         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 516                 return -ENOMEM;
 517
 518         rcu_read_lock();
 519         map = rcu_dereference(queue->rps_map);
 520         if (map)
 521                 for (i = 0; i < map->len; i++)
 522                         cpumask_set_cpu(map->cpus[i], mask);
 523
 524         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 525         if (PAGE_SIZE - len < 3) {
 526                 rcu_read_unlock();
 527                 free_cpumask_var(mask);
 528                 return -EINVAL;
 529         }
 530         rcu_read_unlock();
 531
 532         free_cpumask_var(mask);
 533         len += sprintf(buf + len, "\n");
 534         return len;
 535 }
 536
 537 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 538                       struct rx_queue_attribute *attribute,
 539                       const char *buf, size_t len)
 540 {
 541         struct rps_map *old_map, *map;
 542         cpumask_var_t mask;
 543         int err, cpu, i;
 544         static DEFINE_SPINLOCK(rps_map_lock);
 545
 546         if (!capable(CAP_NET_ADMIN))
 547                 return -EPERM;
 548
 549         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 550                 return -ENOMEM;
 551
 552         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 553         if (err) {
 554                 free_cpumask_var(mask);
 555                 return err;
 556         }
 557
 558         map = kzalloc(max_t(unsigned int,
 559             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 560             GFP_KERNEL);
 561         if (!map) {
 562                 free_cpumask_var(mask);
 563                 return -ENOMEM;
 564         }
 565
 566         i = 0;
 567         for_each_cpu_and(cpu, mask, cpu_online_mask)
 568                 map->cpus[i++] = cpu;
 569
 570         if (i)
 571                 map->len = i;
 572         else {
 573                 kfree(map);
 574                 map = NULL;
 575         }
 576
 577         spin_lock(&rps_map_lock);
 578         old_map = rcu_dereference_protected(queue->rps_map,
 579                                             lockdep_is_held(&rps_map_lock));
 580         rcu_assign_pointer(queue->rps_map, map);
 581         spin_unlock(&rps_map_lock);
 582
 583         if (map)
 584                 static_key_slow_inc(&rps_needed);
 585         if (old_map) {
 586                 kfree_rcu(old_map, rcu);
 587                 static_key_slow_dec(&rps_needed);
 588         }
 589         free_cpumask_var(mask);
 590         return len;
 591 }
 592
 593 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 594                                            struct rx_queue_attribute *attr,
 595                                            char *buf)
 596 {
 597         struct rps_dev_flow_table *flow_table;
 598         unsigned long val = 0;
 599
 600         rcu_read_lock();
 601         flow_table = rcu_dereference(queue->rps_flow_table);
 602         if (flow_table)
 603                 val = (unsigned long)flow_table->mask + 1;
 604         rcu_read_unlock();
 605
 606         return sprintf(buf, "%lu\n", val);
 607 }
 608
 609 static void rps_dev_flow_table_release_work(struct work_struct *work)
 610 {
 611         struct rps_dev_flow_table *table = container_of(work,
 612             struct rps_dev_flow_table, free_work);
 613
 614         vfree(table);
 615 }
 616
 617 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 618 {
 619         struct rps_dev_flow_table *table = container_of(rcu,
 620             struct rps_dev_flow_table, rcu);
 621
 622         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 623         schedule_work(&table->free_work);
 624 }
 625
 626 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 627                                      struct rx_queue_attribute *attr,
 628                                      const char *buf, size_t len)
 629 {
 630         unsigned long mask, count;
 631         struct rps_dev_flow_table *table, *old_table;
 632         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 633         int rc;
 634
 635         if (!capable(CAP_NET_ADMIN))
 636                 return -EPERM;
 637
 638         rc = kstrtoul(buf, 0, &count);
 639         if (rc < 0)
 640                 return rc;
 641
 642         if (count) {
 643                 mask = count - 1;
 644                 /* mask = roundup_pow_of_two(count) - 1;
 645                  * without overflows...
 646                  */
 647                 while ((mask | (mask >> 1)) != mask)
 648                         mask |= (mask >> 1);
 649                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 650                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 651                  * doesnt overflow.
 652                  */
 653 #if BITS_PER_LONG > 32
 654                 if (mask > (unsigned long)(u32)mask)
 655                         return -EINVAL;
 656 #else
 657                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 658                                 / sizeof(struct rps_dev_flow)) {
 659                         /* Enforce a limit to prevent overflow */
 660                         return -EINVAL;
 661                 }
 662 #endif
 663                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 664                 if (!table)
 665                         return -ENOMEM;
 666
 667                 table->mask = mask;
 668                 for (count = 0; count <= mask; count++)
 669                         table->flows[count].cpu = RPS_NO_CPU;
 670         } else
 671                 table = NULL;
 672
 673         spin_lock(&rps_dev_flow_lock);
 674         old_table = rcu_dereference_protected(queue->rps_flow_table,
 675                                               lockdep_is_held(&rps_dev_flow_lock));
 676         rcu_assign_pointer(queue->rps_flow_table, table);
 677         spin_unlock(&rps_dev_flow_lock);
 678
 679         if (old_table)
 680                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 681
 682         return len;
 683 }
 684
 685 static struct rx_queue_attribute rps_cpus_attribute =
 686         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 687
 688
 689 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 690         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 691             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 692
 693 static struct attribute *rx_queue_default_attrs[] = {
 694         &rps_cpus_attribute.attr,
 695         &rps_dev_flow_table_cnt_attribute.attr,
 696         NULL
 697 };
 698
 699 static void rx_queue_release(struct kobject *kobj)
 700 {
 701         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 702         struct rps_map *map;
 703         struct rps_dev_flow_table *flow_table;
 704
 705
 706         map = rcu_dereference_protected(queue->rps_map, 1);
 707         if (map) {
 708                 RCU_INIT_POINTER(queue->rps_map, NULL);
 709                 kfree_rcu(map, rcu);
 710         }
 711
 712         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 713         if (flow_table) {
 714                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 715                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 716         }
 717
 718         memset(kobj, 0, sizeof(*kobj));
 719         dev_put(queue->dev);
 720 }
 721
 722 static struct kobj_type rx_queue_ktype = {
 723         .sysfs_ops = &rx_queue_sysfs_ops,
 724         .release = rx_queue_release,
 725         .default_attrs = rx_queue_default_attrs,
 726 };
 727
 728 static int rx_queue_add_kobject(struct net_device *net, int index)
 729 {
 730         struct netdev_rx_queue *queue = net->_rx + index;
 731         struct kobject *kobj = &queue->kobj;
 732         int error = 0;
 733
 734         kobj->kset = net->queues_kset;
 735         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 736             "rx-%u", index);
 737         if (error) {
 738                 kobject_put(kobj);
 739                 return error;
 740         }
 741
 742         kobject_uevent(kobj, KOBJ_ADD);
 743         dev_hold(queue->dev);
 744
 745         return error;
 746 }
 747 #endif /* CONFIG_RPS */
 748
 749 int
 750 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 751 {
 752 #ifdef CONFIG_RPS
 753         int i;
 754         int error = 0;
 755
 756         for (i = old_num; i < new_num; i++) {
 757                 error = rx_queue_add_kobject(net, i);
 758                 if (error) {
 759                         new_num = old_num;
 760                         break;
 761                 }
 762         }
 763
 764         while (--i >= new_num)
 765                 kobject_put(&net->_rx[i].kobj);
 766
 767         return error;
 768 #else
 769         return 0;
 770 #endif
 771 }
 772
 773 #ifdef CONFIG_SYSFS
 774 /*
 775  * netdev_queue sysfs structures and functions.
 776  */
 777 struct netdev_queue_attribute {
 778         struct attribute attr;
 779         ssize_t (*show)(struct netdev_queue *queue,
 780             struct netdev_queue_attribute *attr, char *buf);
 781         ssize_t (*store)(struct netdev_queue *queue,
 782             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 783 };
 784 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 785     struct netdev_queue_attribute, attr)
 786
 787 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 788
 789 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 790                                       struct attribute *attr, char *buf)
 791 {
 792         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 793         struct netdev_queue *queue = to_netdev_queue(kobj);
 794
 795         if (!attribute->show)
 796                 return -EIO;
 797
 798         return attribute->show(queue, attribute, buf);
 799 }
 800
 801 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 802                                        struct attribute *attr,
 803                                        const char *buf, size_t count)
 804 {
 805         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 806         struct netdev_queue *queue = to_netdev_queue(kobj);
 807
 808         if (!attribute->store)
 809                 return -EIO;
 810
 811         return attribute->store(queue, attribute, buf, count);
 812 }
 813
 814 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 815         .show = netdev_queue_attr_show,
 816         .store = netdev_queue_attr_store,
 817 };
 818
 819 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 820                                   struct netdev_queue_attribute *attribute,
 821                                   char *buf)
 822 {
 823         unsigned long trans_timeout;
 824
 825         spin_lock_irq(&queue->_xmit_lock);
 826         trans_timeout = queue->trans_timeout;
 827         spin_unlock_irq(&queue->_xmit_lock);
 828
 829         return sprintf(buf, "%lu", trans_timeout);
 830 }
 831
 832 static struct netdev_queue_attribute queue_trans_timeout =
 833         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 834
 835 #ifdef CONFIG_BQL
 836 /*
 837  * Byte queue limits sysfs structures and functions.
 838  */
 839 static ssize_t bql_show(char *buf, unsigned int value)
 840 {
 841         return sprintf(buf, "%u\n", value);
 842 }
 843
 844 static ssize_t bql_set(const char *buf, const size_t count,
 845                        unsigned int *pvalue)
 846 {
 847         unsigned int value;
 848         int err;
 849
 850         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 851                 value = DQL_MAX_LIMIT;
 852         else {
 853                 err = kstrtouint(buf, 10, &value);
 854                 if (err < 0)
 855                         return err;
 856                 if (value > DQL_MAX_LIMIT)
 857                         return -EINVAL;
 858         }
 859
 860         *pvalue = value;
 861
 862         return count;
 863 }
 864
 865 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 866                                   struct netdev_queue_attribute *attr,
 867                                   char *buf)
 868 {
 869         struct dql *dql = &queue->dql;
 870
 871         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 872 }
 873
 874 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 875                                  struct netdev_queue_attribute *attribute,
 876                                  const char *buf, size_t len)
 877 {
 878         struct dql *dql = &queue->dql;
 879         unsigned int value;
 880         int err;
 881
 882         err = kstrtouint(buf, 10, &value);
 883         if (err < 0)
 884                 return err;
 885
 886         dql->slack_hold_time = msecs_to_jiffies(value);
 887
 888         return len;
 889 }
 890
 891 static struct netdev_queue_attribute bql_hold_time_attribute =
 892         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 893             bql_set_hold_time);
 894
 895 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 896                                  struct netdev_queue_attribute *attr,
 897                                  char *buf)
 898 {
 899         struct dql *dql = &queue->dql;
 900
 901         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 902 }
 903
 904 static struct netdev_queue_attribute bql_inflight_attribute =
 905         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 906
 907 #define BQL_ATTR(NAME, FIELD)                                           \
 908 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 909                                  struct netdev_queue_attribute *attr,   \
 910                                  char *buf)                             \
 911 {                                                                       \
 912         return bql_show(buf, queue->dql.FIELD);                         \
 913 }                                                                       \
 914                                                                         \
 915 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 916                                 struct netdev_queue_attribute *attr,    \
 917                                 const char *buf, size_t len)            \
 918 {                                                                       \
 919         return bql_set(buf, len, &queue->dql.FIELD);                    \
 920 }                                                                       \
 921                                                                         \
 922 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 923         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 924             bql_set_ ## NAME);
 925
 926 BQL_ATTR(limit, limit)
 927 BQL_ATTR(limit_max, max_limit)
 928 BQL_ATTR(limit_min, min_limit)
 929
 930 static struct attribute *dql_attrs[] = {
 931         &bql_limit_attribute.attr,
 932         &bql_limit_max_attribute.attr,
 933         &bql_limit_min_attribute.attr,
 934         &bql_hold_time_attribute.attr,
 935         &bql_inflight_attribute.attr,
 936         NULL
 937 };
 938
 939 static struct attribute_group dql_group = {
 940         .name  = "byte_queue_limits",
 941         .attrs  = dql_attrs,
 942 };
 943 #endif /* CONFIG_BQL */
 944
 945 #ifdef CONFIG_XPS
 946 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 947 {
 948         struct net_device *dev = queue->dev;
 949         int i;
 950
 951         for (i = 0; i < dev->num_tx_queues; i++)
 952                 if (queue == &dev->_tx[i])
 953                         break;
 954
 955         BUG_ON(i >= dev->num_tx_queues);
 956
 957         return i;
 958 }
 959
 960
 961 static ssize_t show_xps_map(struct netdev_queue *queue,
 962                             struct netdev_queue_attribute *attribute, char *buf)
 963 {
 964         struct net_device *dev = queue->dev;
 965         struct xps_dev_maps *dev_maps;
 966         cpumask_var_t mask;
 967         unsigned long index;
 968         size_t len = 0;
 969         int i;
 970
 971         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 972                 return -ENOMEM;
 973
 974         index = get_netdev_queue_index(queue);
 975
 976         rcu_read_lock();
 977         dev_maps = rcu_dereference(dev->xps_maps);
 978         if (dev_maps) {
 979                 for_each_possible_cpu(i) {
 980                         struct xps_map *map =
 981                             rcu_dereference(dev_maps->cpu_map[i]);
 982                         if (map) {
 983                                 int j;
 984                                 for (j = 0; j < map->len; j++) {
 985                                         if (map->queues[j] == index) {
 986                                                 cpumask_set_cpu(i, mask);
 987                                                 break;
 988                                         }
 989                                 }
 990                         }
 991                 }
 992         }
 993         rcu_read_unlock();
 994
 995         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 996         if (PAGE_SIZE - len < 3) {
 997                 free_cpumask_var(mask);
 998                 return -EINVAL;
 999         }
1000
1001         free_cpumask_var(mask);
1002         len += sprintf(buf + len, "\n");
1003         return len;
1004 }
1005
1006 static ssize_t store_xps_map(struct netdev_queue *queue,
1007                       struct netdev_queue_attribute *attribute,
1008                       const char *buf, size_t len)
1009 {
1010         struct net_device *dev = queue->dev;
1011         unsigned long index;
1012         cpumask_var_t mask;
1013         int err;
1014
1015         if (!capable(CAP_NET_ADMIN))
1016                 return -EPERM;
1017
1018         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1019                 return -ENOMEM;
1020
1021         index = get_netdev_queue_index(queue);
1022
1023         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1024         if (err) {
1025                 free_cpumask_var(mask);
1026                 return err;
1027         }
1028
1029         err = netif_set_xps_queue(dev, mask, index);
1030
1031         free_cpumask_var(mask);
1032
1033         return err ? : len;
1034 }
1035
1036 static struct netdev_queue_attribute xps_cpus_attribute =
1037     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1038 #endif /* CONFIG_XPS */
1039
1040 static struct attribute *netdev_queue_default_attrs[] = {
1041         &queue_trans_timeout.attr,
1042 #ifdef CONFIG_XPS
1043         &xps_cpus_attribute.attr,
1044 #endif
1045         NULL
1046 };
1047
1048 static void netdev_queue_release(struct kobject *kobj)
1049 {
1050         struct netdev_queue *queue = to_netdev_queue(kobj);
1051
1052         memset(kobj, 0, sizeof(*kobj));
1053         dev_put(queue->dev);
1054 }
1055
1056 static struct kobj_type netdev_queue_ktype = {
1057         .sysfs_ops = &netdev_queue_sysfs_ops,
1058         .release = netdev_queue_release,
1059         .default_attrs = netdev_queue_default_attrs,
1060 };
1061
1062 static int netdev_queue_add_kobject(struct net_device *net, int index)
1063 {
1064         struct netdev_queue *queue = net->_tx + index;
1065         struct kobject *kobj = &queue->kobj;
1066         int error = 0;
1067
1068         kobj->kset = net->queues_kset;
1069         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1070             "tx-%u", index);
1071         if (error)
1072                 goto exit;
1073
1074 #ifdef CONFIG_BQL
1075         error = sysfs_create_group(kobj, &dql_group);
1076         if (error)
1077                 goto exit;
1078 #endif
1079
1080         kobject_uevent(kobj, KOBJ_ADD);
1081         dev_hold(queue->dev);
1082
1083         return 0;
1084 exit:
1085         kobject_put(kobj);
1086         return error;
1087 }
1088 #endif /* CONFIG_SYSFS */
1089
1090 int
1091 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1092 {
1093 #ifdef CONFIG_SYSFS
1094         int i;
1095         int error = 0;
1096
1097         for (i = old_num; i < new_num; i++) {
1098                 error = netdev_queue_add_kobject(net, i);
1099                 if (error) {
1100                         new_num = old_num;
1101                         break;
1102                 }
1103         }
1104
1105         while (--i >= new_num) {
1106                 struct netdev_queue *queue = net->_tx + i;
1107
1108 #ifdef CONFIG_BQL
1109                 sysfs_remove_group(&queue->kobj, &dql_group);
1110 #endif
1111                 kobject_put(&queue->kobj);
1112         }
1113
1114         return error;
1115 #else
1116         return 0;
1117 #endif /* CONFIG_SYSFS */
1118 }
1119
1120 static int register_queue_kobjects(struct net_device *net)
1121 {
1122         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1123
1124 #ifdef CONFIG_SYSFS
1125         net->queues_kset = kset_create_and_add("queues",
1126             NULL, &net->dev.kobj);
1127         if (!net->queues_kset)
1128                 return -ENOMEM;
1129 #endif
1130
1131 #ifdef CONFIG_RPS
1132         real_rx = net->real_num_rx_queues;
1133 #endif
1134         real_tx = net->real_num_tx_queues;
1135
1136         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1137         if (error)
1138                 goto error;
1139         rxq = real_rx;
1140
1141         error = netdev_queue_update_kobjects(net, 0, real_tx);
1142         if (error)
1143                 goto error;
1144         txq = real_tx;
1145
1146         return 0;
1147
1148 error:
1149         netdev_queue_update_kobjects(net, txq, 0);
1150         net_rx_queue_update_kobjects(net, rxq, 0);
1151         return error;
1152 }
1153
1154 static void remove_queue_kobjects(struct net_device *net)
1155 {
1156         int real_rx = 0, real_tx = 0;
1157
1158 #ifdef CONFIG_RPS
1159         real_rx = net->real_num_rx_queues;
1160 #endif
1161         real_tx = net->real_num_tx_queues;
1162
1163         net_rx_queue_update_kobjects(net, real_rx, 0);
1164         netdev_queue_update_kobjects(net, real_tx, 0);
1165 #ifdef CONFIG_SYSFS
1166         kset_unregister(net->queues_kset);
1167 #endif
1168 }
1169
1170 static void *net_grab_current_ns(void)
1171 {
1172         struct net *ns = current->nsproxy->net_ns;
1173 #ifdef CONFIG_NET_NS
1174         if (ns)
1175                 atomic_inc(&ns->passive);
1176 #endif
1177         return ns;
1178 }
1179
1180 static const void *net_initial_ns(void)
1181 {
1182         return &init_net;
1183 }
1184
1185 static const void *net_netlink_ns(struct sock *sk)
1186 {
1187         return sock_net(sk);
1188 }
1189
1190 struct kobj_ns_type_operations net_ns_type_operations = {
1191         .type = KOBJ_NS_TYPE_NET,
1192         .grab_current_ns = net_grab_current_ns,
1193         .netlink_ns = net_netlink_ns,
1194         .initial_ns = net_initial_ns,
1195         .drop_ns = net_drop_ns,
1196 };
1197 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1198
1199 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1200 {
1201         struct net_device *dev = to_net_dev(d);
1202         int retval;
1203
1204         /* pass interface to uevent. */
1205         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1206         if (retval)
1207                 goto exit;
1208
1209         /* pass ifindex to uevent.
1210          * ifindex is useful as it won't change (interface name may change)
1211          * and is what RtNetlink uses natively. */
1212         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1213
1214 exit:
1215         return retval;
1216 }
1217
1218 /*
1219  *      netdev_release -- destroy and free a dead device.
1220  *      Called when last reference to device kobject is gone.
1221  */
1222 static void netdev_release(struct device *d)
1223 {
1224         struct net_device *dev = to_net_dev(d);
1225
1226         BUG_ON(dev->reg_state != NETREG_RELEASED);
1227
1228         kfree(dev->ifalias);
1229         kfree((char *)dev - dev->padded);
1230 }
1231
1232 static const void *net_namespace(struct device *d)
1233 {
1234         struct net_device *dev;
1235         dev = container_of(d, struct net_device, dev);
1236         return dev_net(dev);
1237 }
1238
1239 static struct class net_class = {
1240         .name = "net",
1241         .dev_release = netdev_release,
1242 #ifdef CONFIG_SYSFS
1243         .dev_attrs = net_class_attributes,
1244 #endif /* CONFIG_SYSFS */
1245         .dev_uevent = netdev_uevent,
1246         .ns_type = &net_ns_type_operations,
1247         .namespace = net_namespace,
1248 };
1249
1250 /* Delete sysfs entries but hold kobject reference until after all
1251  * netdev references are gone.
1252  */
1253 void netdev_unregister_kobject(struct net_device * net)
1254 {
1255         struct device *dev = &(net->dev);
1256
1257         kobject_get(&dev->kobj);
1258
1259         remove_queue_kobjects(net);
1260
1261         pm_runtime_set_memalloc_noio(dev, false);
1262
1263         device_del(dev);
1264 }
1265
1266 /* Create sysfs entries for network device. */
1267 int netdev_register_kobject(struct net_device *net)
1268 {
1269         struct device *dev = &(net->dev);
1270         const struct attribute_group **groups = net->sysfs_groups;
1271         int error = 0;
1272
1273         device_initialize(dev);
1274         dev->class = &net_class;
1275         dev->platform_data = net;
1276         dev->groups = groups;
1277
1278         dev_set_name(dev, "%s", net->name);
1279
1280 #ifdef CONFIG_SYSFS
1281         /* Allow for a device specific group */
1282         if (*groups)
1283                 groups++;
1284
1285         *groups++ = &netstat_group;
1286
1287 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1288         if (net->ieee80211_ptr)
1289                 *groups++ = &wireless_group;
1290 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1291         else if (net->wireless_handlers)
1292                 *groups++ = &wireless_group;
1293 #endif
1294 #endif
1295 #endif /* CONFIG_SYSFS */
1296
1297         error = device_add(dev);
1298         if (error)
1299                 return error;
1300
1301         error = register_queue_kobjects(net);
1302         if (error) {
1303                 device_del(dev);
1304                 return error;
1305         }
1306
1307         pm_runtime_set_memalloc_noio(dev, true);
1308
1309         return error;
1310 }
1311
1312 int netdev_class_create_file(struct class_attribute *class_attr)
1313 {
1314         return class_create_file(&net_class, class_attr);
1315 }
1316 EXPORT_SYMBOL(netdev_class_create_file);
1317
1318 void netdev_class_remove_file(struct class_attribute *class_attr)
1319 {
1320         class_remove_file(&net_class, class_attr);
1321 }
1322 EXPORT_SYMBOL(netdev_class_remove_file);
1323
1324 int netdev_kobject_init(void)
1325 {
1326         kobj_ns_type_register(&net_ns_type_operations);
1327         return class_register(&net_class);
1328 }