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(struct rcu_head *rcu)
 610 {
 611         struct rps_dev_flow_table *table = container_of(rcu,
 612             struct rps_dev_flow_table, rcu);
 613         vfree(table);
 614 }
 615
 616 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 617                                      struct rx_queue_attribute *attr,
 618                                      const char *buf, size_t len)
 619 {
 620         unsigned long mask, count;
 621         struct rps_dev_flow_table *table, *old_table;
 622         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 623         int rc;
 624
 625         if (!capable(CAP_NET_ADMIN))
 626                 return -EPERM;
 627
 628         rc = kstrtoul(buf, 0, &count);
 629         if (rc < 0)
 630                 return rc;
 631
 632         if (count) {
 633                 mask = count - 1;
 634                 /* mask = roundup_pow_of_two(count) - 1;
 635                  * without overflows...
 636                  */
 637                 while ((mask | (mask >> 1)) != mask)
 638                         mask |= (mask >> 1);
 639                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 640                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 641                  * doesnt overflow.
 642                  */
 643 #if BITS_PER_LONG > 32
 644                 if (mask > (unsigned long)(u32)mask)
 645                         return -EINVAL;
 646 #else
 647                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 648                                 / sizeof(struct rps_dev_flow)) {
 649                         /* Enforce a limit to prevent overflow */
 650                         return -EINVAL;
 651                 }
 652 #endif
 653                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 654                 if (!table)
 655                         return -ENOMEM;
 656
 657                 table->mask = mask;
 658                 for (count = 0; count <= mask; count++)
 659                         table->flows[count].cpu = RPS_NO_CPU;
 660         } else
 661                 table = NULL;
 662
 663         spin_lock(&rps_dev_flow_lock);
 664         old_table = rcu_dereference_protected(queue->rps_flow_table,
 665                                               lockdep_is_held(&rps_dev_flow_lock));
 666         rcu_assign_pointer(queue->rps_flow_table, table);
 667         spin_unlock(&rps_dev_flow_lock);
 668
 669         if (old_table)
 670                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 671
 672         return len;
 673 }
 674
 675 static struct rx_queue_attribute rps_cpus_attribute =
 676         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 677
 678
 679 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 680         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 681             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 682
 683 static struct attribute *rx_queue_default_attrs[] = {
 684         &rps_cpus_attribute.attr,
 685         &rps_dev_flow_table_cnt_attribute.attr,
 686         NULL
 687 };
 688
 689 static void rx_queue_release(struct kobject *kobj)
 690 {
 691         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 692         struct rps_map *map;
 693         struct rps_dev_flow_table *flow_table;
 694
 695
 696         map = rcu_dereference_protected(queue->rps_map, 1);
 697         if (map) {
 698                 RCU_INIT_POINTER(queue->rps_map, NULL);
 699                 kfree_rcu(map, rcu);
 700         }
 701
 702         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 703         if (flow_table) {
 704                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 705                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 706         }
 707
 708         memset(kobj, 0, sizeof(*kobj));
 709         dev_put(queue->dev);
 710 }
 711
 712 static struct kobj_type rx_queue_ktype = {
 713         .sysfs_ops = &rx_queue_sysfs_ops,
 714         .release = rx_queue_release,
 715         .default_attrs = rx_queue_default_attrs,
 716 };
 717
 718 static int rx_queue_add_kobject(struct net_device *net, int index)
 719 {
 720         struct netdev_rx_queue *queue = net->_rx + index;
 721         struct kobject *kobj = &queue->kobj;
 722         int error = 0;
 723
 724         kobj->kset = net->queues_kset;
 725         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 726             "rx-%u", index);
 727         if (error) {
 728                 kobject_put(kobj);
 729                 return error;
 730         }
 731
 732         kobject_uevent(kobj, KOBJ_ADD);
 733         dev_hold(queue->dev);
 734
 735         return error;
 736 }
 737 #endif /* CONFIG_RPS */
 738
 739 int
 740 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 741 {
 742 #ifdef CONFIG_RPS
 743         int i;
 744         int error = 0;
 745
 746         for (i = old_num; i < new_num; i++) {
 747                 error = rx_queue_add_kobject(net, i);
 748                 if (error) {
 749                         new_num = old_num;
 750                         break;
 751                 }
 752         }
 753
 754         while (--i >= new_num)
 755                 kobject_put(&net->_rx[i].kobj);
 756
 757         return error;
 758 #else
 759         return 0;
 760 #endif
 761 }
 762
 763 #ifdef CONFIG_SYSFS
 764 /*
 765  * netdev_queue sysfs structures and functions.
 766  */
 767 struct netdev_queue_attribute {
 768         struct attribute attr;
 769         ssize_t (*show)(struct netdev_queue *queue,
 770             struct netdev_queue_attribute *attr, char *buf);
 771         ssize_t (*store)(struct netdev_queue *queue,
 772             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 773 };
 774 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 775     struct netdev_queue_attribute, attr)
 776
 777 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 778
 779 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 780                                       struct attribute *attr, char *buf)
 781 {
 782         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 783         struct netdev_queue *queue = to_netdev_queue(kobj);
 784
 785         if (!attribute->show)
 786                 return -EIO;
 787
 788         return attribute->show(queue, attribute, buf);
 789 }
 790
 791 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 792                                        struct attribute *attr,
 793                                        const char *buf, size_t count)
 794 {
 795         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 796         struct netdev_queue *queue = to_netdev_queue(kobj);
 797
 798         if (!attribute->store)
 799                 return -EIO;
 800
 801         return attribute->store(queue, attribute, buf, count);
 802 }
 803
 804 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 805         .show = netdev_queue_attr_show,
 806         .store = netdev_queue_attr_store,
 807 };
 808
 809 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 810                                   struct netdev_queue_attribute *attribute,
 811                                   char *buf)
 812 {
 813         unsigned long trans_timeout;
 814
 815         spin_lock_irq(&queue->_xmit_lock);
 816         trans_timeout = queue->trans_timeout;
 817         spin_unlock_irq(&queue->_xmit_lock);
 818
 819         return sprintf(buf, "%lu", trans_timeout);
 820 }
 821
 822 static struct netdev_queue_attribute queue_trans_timeout =
 823         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 824
 825 #ifdef CONFIG_BQL
 826 /*
 827  * Byte queue limits sysfs structures and functions.
 828  */
 829 static ssize_t bql_show(char *buf, unsigned int value)
 830 {
 831         return sprintf(buf, "%u\n", value);
 832 }
 833
 834 static ssize_t bql_set(const char *buf, const size_t count,
 835                        unsigned int *pvalue)
 836 {
 837         unsigned int value;
 838         int err;
 839
 840         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 841                 value = DQL_MAX_LIMIT;
 842         else {
 843                 err = kstrtouint(buf, 10, &value);
 844                 if (err < 0)
 845                         return err;
 846                 if (value > DQL_MAX_LIMIT)
 847                         return -EINVAL;
 848         }
 849
 850         *pvalue = value;
 851
 852         return count;
 853 }
 854
 855 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 856                                   struct netdev_queue_attribute *attr,
 857                                   char *buf)
 858 {
 859         struct dql *dql = &queue->dql;
 860
 861         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 862 }
 863
 864 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 865                                  struct netdev_queue_attribute *attribute,
 866                                  const char *buf, size_t len)
 867 {
 868         struct dql *dql = &queue->dql;
 869         unsigned int value;
 870         int err;
 871
 872         err = kstrtouint(buf, 10, &value);
 873         if (err < 0)
 874                 return err;
 875
 876         dql->slack_hold_time = msecs_to_jiffies(value);
 877
 878         return len;
 879 }
 880
 881 static struct netdev_queue_attribute bql_hold_time_attribute =
 882         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 883             bql_set_hold_time);
 884
 885 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 886                                  struct netdev_queue_attribute *attr,
 887                                  char *buf)
 888 {
 889         struct dql *dql = &queue->dql;
 890
 891         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 892 }
 893
 894 static struct netdev_queue_attribute bql_inflight_attribute =
 895         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 896
 897 #define BQL_ATTR(NAME, FIELD)                                           \
 898 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 899                                  struct netdev_queue_attribute *attr,   \
 900                                  char *buf)                             \
 901 {                                                                       \
 902         return bql_show(buf, queue->dql.FIELD);                         \
 903 }                                                                       \
 904                                                                         \
 905 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 906                                 struct netdev_queue_attribute *attr,    \
 907                                 const char *buf, size_t len)            \
 908 {                                                                       \
 909         return bql_set(buf, len, &queue->dql.FIELD);                    \
 910 }                                                                       \
 911                                                                         \
 912 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 913         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 914             bql_set_ ## NAME);
 915
 916 BQL_ATTR(limit, limit)
 917 BQL_ATTR(limit_max, max_limit)
 918 BQL_ATTR(limit_min, min_limit)
 919
 920 static struct attribute *dql_attrs[] = {
 921         &bql_limit_attribute.attr,
 922         &bql_limit_max_attribute.attr,
 923         &bql_limit_min_attribute.attr,
 924         &bql_hold_time_attribute.attr,
 925         &bql_inflight_attribute.attr,
 926         NULL
 927 };
 928
 929 static struct attribute_group dql_group = {
 930         .name  = "byte_queue_limits",
 931         .attrs  = dql_attrs,
 932 };
 933 #endif /* CONFIG_BQL */
 934
 935 #ifdef CONFIG_XPS
 936 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 937 {
 938         struct net_device *dev = queue->dev;
 939         int i;
 940
 941         for (i = 0; i < dev->num_tx_queues; i++)
 942                 if (queue == &dev->_tx[i])
 943                         break;
 944
 945         BUG_ON(i >= dev->num_tx_queues);
 946
 947         return i;
 948 }
 949
 950
 951 static ssize_t show_xps_map(struct netdev_queue *queue,
 952                             struct netdev_queue_attribute *attribute, char *buf)
 953 {
 954         struct net_device *dev = queue->dev;
 955         struct xps_dev_maps *dev_maps;
 956         cpumask_var_t mask;
 957         unsigned long index;
 958         size_t len = 0;
 959         int i;
 960
 961         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 962                 return -ENOMEM;
 963
 964         index = get_netdev_queue_index(queue);
 965
 966         rcu_read_lock();
 967         dev_maps = rcu_dereference(dev->xps_maps);
 968         if (dev_maps) {
 969                 for_each_possible_cpu(i) {
 970                         struct xps_map *map =
 971                             rcu_dereference(dev_maps->cpu_map[i]);
 972                         if (map) {
 973                                 int j;
 974                                 for (j = 0; j < map->len; j++) {
 975                                         if (map->queues[j] == index) {
 976                                                 cpumask_set_cpu(i, mask);
 977                                                 break;
 978                                         }
 979                                 }
 980                         }
 981                 }
 982         }
 983         rcu_read_unlock();
 984
 985         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 986         if (PAGE_SIZE - len < 3) {
 987                 free_cpumask_var(mask);
 988                 return -EINVAL;
 989         }
 990
 991         free_cpumask_var(mask);
 992         len += sprintf(buf + len, "\n");
 993         return len;
 994 }
 995
 996 static ssize_t store_xps_map(struct netdev_queue *queue,
 997                       struct netdev_queue_attribute *attribute,
 998                       const char *buf, size_t len)
 999 {
1000         struct net_device *dev = queue->dev;
1001         unsigned long index;
1002         cpumask_var_t mask;
1003         int err;
1004
1005         if (!capable(CAP_NET_ADMIN))
1006                 return -EPERM;
1007
1008         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1009                 return -ENOMEM;
1010
1011         index = get_netdev_queue_index(queue);
1012
1013         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1014         if (err) {
1015                 free_cpumask_var(mask);
1016                 return err;
1017         }
1018
1019         err = netif_set_xps_queue(dev, mask, index);
1020
1021         free_cpumask_var(mask);
1022
1023         return err ? : len;
1024 }
1025
1026 static struct netdev_queue_attribute xps_cpus_attribute =
1027     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1028 #endif /* CONFIG_XPS */
1029
1030 static struct attribute *netdev_queue_default_attrs[] = {
1031         &queue_trans_timeout.attr,
1032 #ifdef CONFIG_XPS
1033         &xps_cpus_attribute.attr,
1034 #endif
1035         NULL
1036 };
1037
1038 static void netdev_queue_release(struct kobject *kobj)
1039 {
1040         struct netdev_queue *queue = to_netdev_queue(kobj);
1041
1042         memset(kobj, 0, sizeof(*kobj));
1043         dev_put(queue->dev);
1044 }
1045
1046 static struct kobj_type netdev_queue_ktype = {
1047         .sysfs_ops = &netdev_queue_sysfs_ops,
1048         .release = netdev_queue_release,
1049         .default_attrs = netdev_queue_default_attrs,
1050 };
1051
1052 static int netdev_queue_add_kobject(struct net_device *net, int index)
1053 {
1054         struct netdev_queue *queue = net->_tx + index;
1055         struct kobject *kobj = &queue->kobj;
1056         int error = 0;
1057
1058         kobj->kset = net->queues_kset;
1059         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1060             "tx-%u", index);
1061         if (error)
1062                 goto exit;
1063
1064 #ifdef CONFIG_BQL
1065         error = sysfs_create_group(kobj, &dql_group);
1066         if (error)
1067                 goto exit;
1068 #endif
1069
1070         kobject_uevent(kobj, KOBJ_ADD);
1071         dev_hold(queue->dev);
1072
1073         return 0;
1074 exit:
1075         kobject_put(kobj);
1076         return error;
1077 }
1078 #endif /* CONFIG_SYSFS */
1079
1080 int
1081 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1082 {
1083 #ifdef CONFIG_SYSFS
1084         int i;
1085         int error = 0;
1086
1087         for (i = old_num; i < new_num; i++) {
1088                 error = netdev_queue_add_kobject(net, i);
1089                 if (error) {
1090                         new_num = old_num;
1091                         break;
1092                 }
1093         }
1094
1095         while (--i >= new_num) {
1096                 struct netdev_queue *queue = net->_tx + i;
1097
1098 #ifdef CONFIG_BQL
1099                 sysfs_remove_group(&queue->kobj, &dql_group);
1100 #endif
1101                 kobject_put(&queue->kobj);
1102         }
1103
1104         return error;
1105 #else
1106         return 0;
1107 #endif /* CONFIG_SYSFS */
1108 }
1109
1110 static int register_queue_kobjects(struct net_device *net)
1111 {
1112         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1113
1114 #ifdef CONFIG_SYSFS
1115         net->queues_kset = kset_create_and_add("queues",
1116             NULL, &net->dev.kobj);
1117         if (!net->queues_kset)
1118                 return -ENOMEM;
1119 #endif
1120
1121 #ifdef CONFIG_RPS
1122         real_rx = net->real_num_rx_queues;
1123 #endif
1124         real_tx = net->real_num_tx_queues;
1125
1126         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1127         if (error)
1128                 goto error;
1129         rxq = real_rx;
1130
1131         error = netdev_queue_update_kobjects(net, 0, real_tx);
1132         if (error)
1133                 goto error;
1134         txq = real_tx;
1135
1136         return 0;
1137
1138 error:
1139         netdev_queue_update_kobjects(net, txq, 0);
1140         net_rx_queue_update_kobjects(net, rxq, 0);
1141         return error;
1142 }
1143
1144 static void remove_queue_kobjects(struct net_device *net)
1145 {
1146         int real_rx = 0, real_tx = 0;
1147
1148 #ifdef CONFIG_RPS
1149         real_rx = net->real_num_rx_queues;
1150 #endif
1151         real_tx = net->real_num_tx_queues;
1152
1153         net_rx_queue_update_kobjects(net, real_rx, 0);
1154         netdev_queue_update_kobjects(net, real_tx, 0);
1155 #ifdef CONFIG_SYSFS
1156         kset_unregister(net->queues_kset);
1157 #endif
1158 }
1159
1160 static void *net_grab_current_ns(void)
1161 {
1162         struct net *ns = current->nsproxy->net_ns;
1163 #ifdef CONFIG_NET_NS
1164         if (ns)
1165                 atomic_inc(&ns->passive);
1166 #endif
1167         return ns;
1168 }
1169
1170 static const void *net_initial_ns(void)
1171 {
1172         return &init_net;
1173 }
1174
1175 static const void *net_netlink_ns(struct sock *sk)
1176 {
1177         return sock_net(sk);
1178 }
1179
1180 struct kobj_ns_type_operations net_ns_type_operations = {
1181         .type = KOBJ_NS_TYPE_NET,
1182         .grab_current_ns = net_grab_current_ns,
1183         .netlink_ns = net_netlink_ns,
1184         .initial_ns = net_initial_ns,
1185         .drop_ns = net_drop_ns,
1186 };
1187 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1188
1189 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1190 {
1191         struct net_device *dev = to_net_dev(d);
1192         int retval;
1193
1194         /* pass interface to uevent. */
1195         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1196         if (retval)
1197                 goto exit;
1198
1199         /* pass ifindex to uevent.
1200          * ifindex is useful as it won't change (interface name may change)
1201          * and is what RtNetlink uses natively. */
1202         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1203
1204 exit:
1205         return retval;
1206 }
1207
1208 /*
1209  *      netdev_release -- destroy and free a dead device.
1210  *      Called when last reference to device kobject is gone.
1211  */
1212 static void netdev_release(struct device *d)
1213 {
1214         struct net_device *dev = to_net_dev(d);
1215
1216         BUG_ON(dev->reg_state != NETREG_RELEASED);
1217
1218         kfree(dev->ifalias);
1219         kfree((char *)dev - dev->padded);
1220 }
1221
1222 static const void *net_namespace(struct device *d)
1223 {
1224         struct net_device *dev;
1225         dev = container_of(d, struct net_device, dev);
1226         return dev_net(dev);
1227 }
1228
1229 static struct class net_class = {
1230         .name = "net",
1231         .dev_release = netdev_release,
1232 #ifdef CONFIG_SYSFS
1233         .dev_attrs = net_class_attributes,
1234 #endif /* CONFIG_SYSFS */
1235         .dev_uevent = netdev_uevent,
1236         .ns_type = &net_ns_type_operations,
1237         .namespace = net_namespace,
1238 };
1239
1240 /* Delete sysfs entries but hold kobject reference until after all
1241  * netdev references are gone.
1242  */
1243 void netdev_unregister_kobject(struct net_device * net)
1244 {
1245         struct device *dev = &(net->dev);
1246
1247         kobject_get(&dev->kobj);
1248
1249         remove_queue_kobjects(net);
1250
1251         pm_runtime_set_memalloc_noio(dev, false);
1252
1253         device_del(dev);
1254 }
1255
1256 /* Create sysfs entries for network device. */
1257 int netdev_register_kobject(struct net_device *net)
1258 {
1259         struct device *dev = &(net->dev);
1260         const struct attribute_group **groups = net->sysfs_groups;
1261         int error = 0;
1262
1263         device_initialize(dev);
1264         dev->class = &net_class;
1265         dev->platform_data = net;
1266         dev->groups = groups;
1267
1268         dev_set_name(dev, "%s", net->name);
1269
1270 #ifdef CONFIG_SYSFS
1271         /* Allow for a device specific group */
1272         if (*groups)
1273                 groups++;
1274
1275         *groups++ = &netstat_group;
1276
1277 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1278         if (net->ieee80211_ptr)
1279                 *groups++ = &wireless_group;
1280 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1281         else if (net->wireless_handlers)
1282                 *groups++ = &wireless_group;
1283 #endif
1284 #endif
1285 #endif /* CONFIG_SYSFS */
1286
1287         error = device_add(dev);
1288         if (error)
1289                 return error;
1290
1291         error = register_queue_kobjects(net);
1292         if (error) {
1293                 device_del(dev);
1294                 return error;
1295         }
1296
1297         pm_runtime_set_memalloc_noio(dev, true);
1298
1299         return error;
1300 }
1301
1302 int netdev_class_create_file(struct class_attribute *class_attr)
1303 {
1304         return class_create_file(&net_class, class_attr);
1305 }
1306 EXPORT_SYMBOL(netdev_class_create_file);
1307
1308 void netdev_class_remove_file(struct class_attribute *class_attr)
1309 {
1310         class_remove_file(&net_class, class_attr);
1311 }
1312 EXPORT_SYMBOL(netdev_class_remove_file);
1313
1314 int netdev_kobject_init(void)
1315 {
1316         kobj_ns_type_register(&net_ns_type_operations);
1317         return class_register(&net_class);
1318 }