Pileus Git - ~andy/linux/blob - 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 ssize_t show_phys_port_id(struct device *dev,
 338                                  struct device_attribute *attr, char *buf)
 339 {
 340         struct net_device *netdev = to_net_dev(dev);
 341         ssize_t ret = -EINVAL;
 342
 343         if (!rtnl_trylock())
 344                 return restart_syscall();
 345
 346         if (dev_isalive(netdev)) {
 347                 struct netdev_phys_port_id ppid;
 348
 349                 ret = dev_get_phys_port_id(netdev, &ppid);
 350                 if (!ret)
 351                         ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 352         }
 353         rtnl_unlock();
 354
 355         return ret;
 356 }
 357
 358 static struct device_attribute net_class_attributes[] = {
 359         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 360         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 361         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 362         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 363         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 364         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 365         __ATTR(type, S_IRUGO, show_type, NULL),
 366         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 367         __ATTR(address, S_IRUGO, show_address, NULL),
 368         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 369         __ATTR(carrier, S_IRUGO | S_IWUSR, show_carrier, store_carrier),
 370         __ATTR(speed, S_IRUGO, show_speed, NULL),
 371         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 372         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 373         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 374         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 375         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 376         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 377                store_tx_queue_len),
 378         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 379         __ATTR(phys_port_id, S_IRUGO, show_phys_port_id, NULL),
 380         {}
 381 };
 382
 383 /* Show a given an attribute in the statistics group */
 384 static ssize_t netstat_show(const struct device *d,
 385                             struct device_attribute *attr, char *buf,
 386                             unsigned long offset)
 387 {
 388         struct net_device *dev = to_net_dev(d);
 389         ssize_t ret = -EINVAL;
 390
 391         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 392                         offset % sizeof(u64) != 0);
 393
 394         read_lock(&dev_base_lock);
 395         if (dev_isalive(dev)) {
 396                 struct rtnl_link_stats64 temp;
 397                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 398
 399                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 400         }
 401         read_unlock(&dev_base_lock);
 402         return ret;
 403 }
 404
 405 /* generate a read-only statistics attribute */
 406 #define NETSTAT_ENTRY(name)                                             \
 407 static ssize_t show_##name(struct device *d,                            \
 408                            struct device_attribute *attr, char *buf)    \
 409 {                                                                       \
 410         return netstat_show(d, attr, buf,                               \
 411                             offsetof(struct rtnl_link_stats64, name));  \
 412 }                                                                       \
 413 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 414
 415 NETSTAT_ENTRY(rx_packets);
 416 NETSTAT_ENTRY(tx_packets);
 417 NETSTAT_ENTRY(rx_bytes);
 418 NETSTAT_ENTRY(tx_bytes);
 419 NETSTAT_ENTRY(rx_errors);
 420 NETSTAT_ENTRY(tx_errors);
 421 NETSTAT_ENTRY(rx_dropped);
 422 NETSTAT_ENTRY(tx_dropped);
 423 NETSTAT_ENTRY(multicast);
 424 NETSTAT_ENTRY(collisions);
 425 NETSTAT_ENTRY(rx_length_errors);
 426 NETSTAT_ENTRY(rx_over_errors);
 427 NETSTAT_ENTRY(rx_crc_errors);
 428 NETSTAT_ENTRY(rx_frame_errors);
 429 NETSTAT_ENTRY(rx_fifo_errors);
 430 NETSTAT_ENTRY(rx_missed_errors);
 431 NETSTAT_ENTRY(tx_aborted_errors);
 432 NETSTAT_ENTRY(tx_carrier_errors);
 433 NETSTAT_ENTRY(tx_fifo_errors);
 434 NETSTAT_ENTRY(tx_heartbeat_errors);
 435 NETSTAT_ENTRY(tx_window_errors);
 436 NETSTAT_ENTRY(rx_compressed);
 437 NETSTAT_ENTRY(tx_compressed);
 438
 439 static struct attribute *netstat_attrs[] = {
 440         &dev_attr_rx_packets.attr,
 441         &dev_attr_tx_packets.attr,
 442         &dev_attr_rx_bytes.attr,
 443         &dev_attr_tx_bytes.attr,
 444         &dev_attr_rx_errors.attr,
 445         &dev_attr_tx_errors.attr,
 446         &dev_attr_rx_dropped.attr,
 447         &dev_attr_tx_dropped.attr,
 448         &dev_attr_multicast.attr,
 449         &dev_attr_collisions.attr,
 450         &dev_attr_rx_length_errors.attr,
 451         &dev_attr_rx_over_errors.attr,
 452         &dev_attr_rx_crc_errors.attr,
 453         &dev_attr_rx_frame_errors.attr,
 454         &dev_attr_rx_fifo_errors.attr,
 455         &dev_attr_rx_missed_errors.attr,
 456         &dev_attr_tx_aborted_errors.attr,
 457         &dev_attr_tx_carrier_errors.attr,
 458         &dev_attr_tx_fifo_errors.attr,
 459         &dev_attr_tx_heartbeat_errors.attr,
 460         &dev_attr_tx_window_errors.attr,
 461         &dev_attr_rx_compressed.attr,
 462         &dev_attr_tx_compressed.attr,
 463         NULL
 464 };
 465
 466
 467 static struct attribute_group netstat_group = {
 468         .name  = "statistics",
 469         .attrs  = netstat_attrs,
 470 };
 471
 472 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 473 static struct attribute *wireless_attrs[] = {
 474         NULL
 475 };
 476
 477 static struct attribute_group wireless_group = {
 478         .name = "wireless",
 479         .attrs = wireless_attrs,
 480 };
 481 #endif
 482 #endif /* CONFIG_SYSFS */
 483
 484 #ifdef CONFIG_RPS
 485 /*
 486  * RX queue sysfs structures and functions.
 487  */
 488 struct rx_queue_attribute {
 489         struct attribute attr;
 490         ssize_t (*show)(struct netdev_rx_queue *queue,
 491             struct rx_queue_attribute *attr, char *buf);
 492         ssize_t (*store)(struct netdev_rx_queue *queue,
 493             struct rx_queue_attribute *attr, const char *buf, size_t len);
 494 };
 495 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 496     struct rx_queue_attribute, attr)
 497
 498 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 499
 500 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 501                                   char *buf)
 502 {
 503         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 504         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 505
 506         if (!attribute->show)
 507                 return -EIO;
 508
 509         return attribute->show(queue, attribute, buf);
 510 }
 511
 512 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 513                                    const char *buf, size_t count)
 514 {
 515         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 516         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 517
 518         if (!attribute->store)
 519                 return -EIO;
 520
 521         return attribute->store(queue, attribute, buf, count);
 522 }
 523
 524 static const struct sysfs_ops rx_queue_sysfs_ops = {
 525         .show = rx_queue_attr_show,
 526         .store = rx_queue_attr_store,
 527 };
 528
 529 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 530                             struct rx_queue_attribute *attribute, char *buf)
 531 {
 532         struct rps_map *map;
 533         cpumask_var_t mask;
 534         size_t len = 0;
 535         int i;
 536
 537         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 538                 return -ENOMEM;
 539
 540         rcu_read_lock();
 541         map = rcu_dereference(queue->rps_map);
 542         if (map)
 543                 for (i = 0; i < map->len; i++)
 544                         cpumask_set_cpu(map->cpus[i], mask);
 545
 546         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 547         if (PAGE_SIZE - len < 3) {
 548                 rcu_read_unlock();
 549                 free_cpumask_var(mask);
 550                 return -EINVAL;
 551         }
 552         rcu_read_unlock();
 553
 554         free_cpumask_var(mask);
 555         len += sprintf(buf + len, "\n");
 556         return len;
 557 }
 558
 559 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 560                       struct rx_queue_attribute *attribute,
 561                       const char *buf, size_t len)
 562 {
 563         struct rps_map *old_map, *map;
 564         cpumask_var_t mask;
 565         int err, cpu, i;
 566         static DEFINE_SPINLOCK(rps_map_lock);
 567
 568         if (!capable(CAP_NET_ADMIN))
 569                 return -EPERM;
 570
 571         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 572                 return -ENOMEM;
 573
 574         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 575         if (err) {
 576                 free_cpumask_var(mask);
 577                 return err;
 578         }
 579
 580         map = kzalloc(max_t(unsigned int,
 581             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 582             GFP_KERNEL);
 583         if (!map) {
 584                 free_cpumask_var(mask);
 585                 return -ENOMEM;
 586         }
 587
 588         i = 0;
 589         for_each_cpu_and(cpu, mask, cpu_online_mask)
 590                 map->cpus[i++] = cpu;
 591
 592         if (i)
 593                 map->len = i;
 594         else {
 595                 kfree(map);
 596                 map = NULL;
 597         }
 598
 599         spin_lock(&rps_map_lock);
 600         old_map = rcu_dereference_protected(queue->rps_map,
 601                                             lockdep_is_held(&rps_map_lock));
 602         rcu_assign_pointer(queue->rps_map, map);
 603         spin_unlock(&rps_map_lock);
 604
 605         if (map)
 606                 static_key_slow_inc(&rps_needed);
 607         if (old_map) {
 608                 kfree_rcu(old_map, rcu);
 609                 static_key_slow_dec(&rps_needed);
 610         }
 611         free_cpumask_var(mask);
 612         return len;
 613 }
 614
 615 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 616                                            struct rx_queue_attribute *attr,
 617                                            char *buf)
 618 {
 619         struct rps_dev_flow_table *flow_table;
 620         unsigned long val = 0;
 621
 622         rcu_read_lock();
 623         flow_table = rcu_dereference(queue->rps_flow_table);
 624         if (flow_table)
 625                 val = (unsigned long)flow_table->mask + 1;
 626         rcu_read_unlock();
 627
 628         return sprintf(buf, "%lu\n", val);
 629 }
 630
 631 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 632 {
 633         struct rps_dev_flow_table *table = container_of(rcu,
 634             struct rps_dev_flow_table, rcu);
 635         vfree(table);
 636 }
 637
 638 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 639                                      struct rx_queue_attribute *attr,
 640                                      const char *buf, size_t len)
 641 {
 642         unsigned long mask, count;
 643         struct rps_dev_flow_table *table, *old_table;
 644         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 645         int rc;
 646
 647         if (!capable(CAP_NET_ADMIN))
 648                 return -EPERM;
 649
 650         rc = kstrtoul(buf, 0, &count);
 651         if (rc < 0)
 652                 return rc;
 653
 654         if (count) {
 655                 mask = count - 1;
 656                 /* mask = roundup_pow_of_two(count) - 1;
 657                  * without overflows...
 658                  */
 659                 while ((mask | (mask >> 1)) != mask)
 660                         mask |= (mask >> 1);
 661                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 662                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 663                  * doesnt overflow.
 664                  */
 665 #if BITS_PER_LONG > 32
 666                 if (mask > (unsigned long)(u32)mask)
 667                         return -EINVAL;
 668 #else
 669                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 670                                 / sizeof(struct rps_dev_flow)) {
 671                         /* Enforce a limit to prevent overflow */
 672                         return -EINVAL;
 673                 }
 674 #endif
 675                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 676                 if (!table)
 677                         return -ENOMEM;
 678
 679                 table->mask = mask;
 680                 for (count = 0; count <= mask; count++)
 681                         table->flows[count].cpu = RPS_NO_CPU;
 682         } else
 683                 table = NULL;
 684
 685         spin_lock(&rps_dev_flow_lock);
 686         old_table = rcu_dereference_protected(queue->rps_flow_table,
 687                                               lockdep_is_held(&rps_dev_flow_lock));
 688         rcu_assign_pointer(queue->rps_flow_table, table);
 689         spin_unlock(&rps_dev_flow_lock);
 690
 691         if (old_table)
 692                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 693
 694         return len;
 695 }
 696
 697 static struct rx_queue_attribute rps_cpus_attribute =
 698         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 699
 700
 701 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 702         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 703             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 704
 705 static struct attribute *rx_queue_default_attrs[] = {
 706         &rps_cpus_attribute.attr,
 707         &rps_dev_flow_table_cnt_attribute.attr,
 708         NULL
 709 };
 710
 711 static void rx_queue_release(struct kobject *kobj)
 712 {
 713         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 714         struct rps_map *map;
 715         struct rps_dev_flow_table *flow_table;
 716
 717
 718         map = rcu_dereference_protected(queue->rps_map, 1);
 719         if (map) {
 720                 RCU_INIT_POINTER(queue->rps_map, NULL);
 721                 kfree_rcu(map, rcu);
 722         }
 723
 724         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 725         if (flow_table) {
 726                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 727                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 728         }
 729
 730         memset(kobj, 0, sizeof(*kobj));
 731         dev_put(queue->dev);
 732 }
 733
 734 static struct kobj_type rx_queue_ktype = {
 735         .sysfs_ops = &rx_queue_sysfs_ops,
 736         .release = rx_queue_release,
 737         .default_attrs = rx_queue_default_attrs,
 738 };
 739
 740 static int rx_queue_add_kobject(struct net_device *net, int index)
 741 {
 742         struct netdev_rx_queue *queue = net->_rx + index;
 743         struct kobject *kobj = &queue->kobj;
 744         int error = 0;
 745
 746         kobj->kset = net->queues_kset;
 747         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 748             "rx-%u", index);
 749         if (error) {
 750                 kobject_put(kobj);
 751                 return error;
 752         }
 753
 754         kobject_uevent(kobj, KOBJ_ADD);
 755         dev_hold(queue->dev);
 756
 757         return error;
 758 }
 759 #endif /* CONFIG_RPS */
 760
 761 int
 762 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 763 {
 764 #ifdef CONFIG_RPS
 765         int i;
 766         int error = 0;
 767
 768         for (i = old_num; i < new_num; i++) {
 769                 error = rx_queue_add_kobject(net, i);
 770                 if (error) {
 771                         new_num = old_num;
 772                         break;
 773                 }
 774         }
 775
 776         while (--i >= new_num)
 777                 kobject_put(&net->_rx[i].kobj);
 778
 779         return error;
 780 #else
 781         return 0;
 782 #endif
 783 }
 784
 785 #ifdef CONFIG_SYSFS
 786 /*
 787  * netdev_queue sysfs structures and functions.
 788  */
 789 struct netdev_queue_attribute {
 790         struct attribute attr;
 791         ssize_t (*show)(struct netdev_queue *queue,
 792             struct netdev_queue_attribute *attr, char *buf);
 793         ssize_t (*store)(struct netdev_queue *queue,
 794             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 795 };
 796 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 797     struct netdev_queue_attribute, attr)
 798
 799 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 800
 801 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 802                                       struct attribute *attr, char *buf)
 803 {
 804         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 805         struct netdev_queue *queue = to_netdev_queue(kobj);
 806
 807         if (!attribute->show)
 808                 return -EIO;
 809
 810         return attribute->show(queue, attribute, buf);
 811 }
 812
 813 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 814                                        struct attribute *attr,
 815                                        const char *buf, size_t count)
 816 {
 817         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 818         struct netdev_queue *queue = to_netdev_queue(kobj);
 819
 820         if (!attribute->store)
 821                 return -EIO;
 822
 823         return attribute->store(queue, attribute, buf, count);
 824 }
 825
 826 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 827         .show = netdev_queue_attr_show,
 828         .store = netdev_queue_attr_store,
 829 };
 830
 831 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 832                                   struct netdev_queue_attribute *attribute,
 833                                   char *buf)
 834 {
 835         unsigned long trans_timeout;
 836
 837         spin_lock_irq(&queue->_xmit_lock);
 838         trans_timeout = queue->trans_timeout;
 839         spin_unlock_irq(&queue->_xmit_lock);
 840
 841         return sprintf(buf, "%lu", trans_timeout);
 842 }
 843
 844 static struct netdev_queue_attribute queue_trans_timeout =
 845         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 846
 847 #ifdef CONFIG_BQL
 848 /*
 849  * Byte queue limits sysfs structures and functions.
 850  */
 851 static ssize_t bql_show(char *buf, unsigned int value)
 852 {
 853         return sprintf(buf, "%u\n", value);
 854 }
 855
 856 static ssize_t bql_set(const char *buf, const size_t count,
 857                        unsigned int *pvalue)
 858 {
 859         unsigned int value;
 860         int err;
 861
 862         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 863                 value = DQL_MAX_LIMIT;
 864         else {
 865                 err = kstrtouint(buf, 10, &value);
 866                 if (err < 0)
 867                         return err;
 868                 if (value > DQL_MAX_LIMIT)
 869                         return -EINVAL;
 870         }
 871
 872         *pvalue = value;
 873
 874         return count;
 875 }
 876
 877 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 878                                   struct netdev_queue_attribute *attr,
 879                                   char *buf)
 880 {
 881         struct dql *dql = &queue->dql;
 882
 883         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 884 }
 885
 886 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 887                                  struct netdev_queue_attribute *attribute,
 888                                  const char *buf, size_t len)
 889 {
 890         struct dql *dql = &queue->dql;
 891         unsigned int value;
 892         int err;
 893
 894         err = kstrtouint(buf, 10, &value);
 895         if (err < 0)
 896                 return err;
 897
 898         dql->slack_hold_time = msecs_to_jiffies(value);
 899
 900         return len;
 901 }
 902
 903 static struct netdev_queue_attribute bql_hold_time_attribute =
 904         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 905             bql_set_hold_time);
 906
 907 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 908                                  struct netdev_queue_attribute *attr,
 909                                  char *buf)
 910 {
 911         struct dql *dql = &queue->dql;
 912
 913         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 914 }
 915
 916 static struct netdev_queue_attribute bql_inflight_attribute =
 917         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 918
 919 #define BQL_ATTR(NAME, FIELD)                                           \
 920 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 921                                  struct netdev_queue_attribute *attr,   \
 922                                  char *buf)                             \
 923 {                                                                       \
 924         return bql_show(buf, queue->dql.FIELD);                         \
 925 }                                                                       \
 926                                                                         \
 927 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 928                                 struct netdev_queue_attribute *attr,    \
 929                                 const char *buf, size_t len)            \
 930 {                                                                       \
 931         return bql_set(buf, len, &queue->dql.FIELD);                    \
 932 }                                                                       \
 933                                                                         \
 934 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 935         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 936             bql_set_ ## NAME);
 937
 938 BQL_ATTR(limit, limit)
 939 BQL_ATTR(limit_max, max_limit)
 940 BQL_ATTR(limit_min, min_limit)
 941
 942 static struct attribute *dql_attrs[] = {
 943         &bql_limit_attribute.attr,
 944         &bql_limit_max_attribute.attr,
 945         &bql_limit_min_attribute.attr,
 946         &bql_hold_time_attribute.attr,
 947         &bql_inflight_attribute.attr,
 948         NULL
 949 };
 950
 951 static struct attribute_group dql_group = {
 952         .name  = "byte_queue_limits",
 953         .attrs  = dql_attrs,
 954 };
 955 #endif /* CONFIG_BQL */
 956
 957 #ifdef CONFIG_XPS
 958 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 959 {
 960         struct net_device *dev = queue->dev;
 961         int i;
 962
 963         for (i = 0; i < dev->num_tx_queues; i++)
 964                 if (queue == &dev->_tx[i])
 965                         break;
 966
 967         BUG_ON(i >= dev->num_tx_queues);
 968
 969         return i;
 970 }
 971
 972
 973 static ssize_t show_xps_map(struct netdev_queue *queue,
 974                             struct netdev_queue_attribute *attribute, char *buf)
 975 {
 976         struct net_device *dev = queue->dev;
 977         struct xps_dev_maps *dev_maps;
 978         cpumask_var_t mask;
 979         unsigned long index;
 980         size_t len = 0;
 981         int i;
 982
 983         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 984                 return -ENOMEM;
 985
 986         index = get_netdev_queue_index(queue);
 987
 988         rcu_read_lock();
 989         dev_maps = rcu_dereference(dev->xps_maps);
 990         if (dev_maps) {
 991                 for_each_possible_cpu(i) {
 992                         struct xps_map *map =
 993                             rcu_dereference(dev_maps->cpu_map[i]);
 994                         if (map) {
 995                                 int j;
 996                                 for (j = 0; j < map->len; j++) {
 997                                         if (map->queues[j] == index) {
 998                                                 cpumask_set_cpu(i, mask);
 999                                                 break;
1000                                         }
1001                                 }
1002                         }
1003                 }
1004         }
1005         rcu_read_unlock();
1006
1007         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1008         if (PAGE_SIZE - len < 3) {
1009                 free_cpumask_var(mask);
1010                 return -EINVAL;
1011         }
1012
1013         free_cpumask_var(mask);
1014         len += sprintf(buf + len, "\n");
1015         return len;
1016 }
1017
1018 static ssize_t store_xps_map(struct netdev_queue *queue,
1019                       struct netdev_queue_attribute *attribute,
1020                       const char *buf, size_t len)
1021 {
1022         struct net_device *dev = queue->dev;
1023         unsigned long index;
1024         cpumask_var_t mask;
1025         int err;
1026
1027         if (!capable(CAP_NET_ADMIN))
1028                 return -EPERM;
1029
1030         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1031                 return -ENOMEM;
1032
1033         index = get_netdev_queue_index(queue);
1034
1035         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1036         if (err) {
1037                 free_cpumask_var(mask);
1038                 return err;
1039         }
1040
1041         err = netif_set_xps_queue(dev, mask, index);
1042
1043         free_cpumask_var(mask);
1044
1045         return err ? : len;
1046 }
1047
1048 static struct netdev_queue_attribute xps_cpus_attribute =
1049     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1050 #endif /* CONFIG_XPS */
1051
1052 static struct attribute *netdev_queue_default_attrs[] = {
1053         &queue_trans_timeout.attr,
1054 #ifdef CONFIG_XPS
1055         &xps_cpus_attribute.attr,
1056 #endif
1057         NULL
1058 };
1059
1060 static void netdev_queue_release(struct kobject *kobj)
1061 {
1062         struct netdev_queue *queue = to_netdev_queue(kobj);
1063
1064         memset(kobj, 0, sizeof(*kobj));
1065         dev_put(queue->dev);
1066 }
1067
1068 static struct kobj_type netdev_queue_ktype = {
1069         .sysfs_ops = &netdev_queue_sysfs_ops,
1070         .release = netdev_queue_release,
1071         .default_attrs = netdev_queue_default_attrs,
1072 };
1073
1074 static int netdev_queue_add_kobject(struct net_device *net, int index)
1075 {
1076         struct netdev_queue *queue = net->_tx + index;
1077         struct kobject *kobj = &queue->kobj;
1078         int error = 0;
1079
1080         kobj->kset = net->queues_kset;
1081         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1082             "tx-%u", index);
1083         if (error)
1084                 goto exit;
1085
1086 #ifdef CONFIG_BQL
1087         error = sysfs_create_group(kobj, &dql_group);
1088         if (error)
1089                 goto exit;
1090 #endif
1091
1092         kobject_uevent(kobj, KOBJ_ADD);
1093         dev_hold(queue->dev);
1094
1095         return 0;
1096 exit:
1097         kobject_put(kobj);
1098         return error;
1099 }
1100 #endif /* CONFIG_SYSFS */
1101
1102 int
1103 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1104 {
1105 #ifdef CONFIG_SYSFS
1106         int i;
1107         int error = 0;
1108
1109         for (i = old_num; i < new_num; i++) {
1110                 error = netdev_queue_add_kobject(net, i);
1111                 if (error) {
1112                         new_num = old_num;
1113                         break;
1114                 }
1115         }
1116
1117         while (--i >= new_num) {
1118                 struct netdev_queue *queue = net->_tx + i;
1119
1120 #ifdef CONFIG_BQL
1121                 sysfs_remove_group(&queue->kobj, &dql_group);
1122 #endif
1123                 kobject_put(&queue->kobj);
1124         }
1125
1126         return error;
1127 #else
1128         return 0;
1129 #endif /* CONFIG_SYSFS */
1130 }
1131
1132 static int register_queue_kobjects(struct net_device *net)
1133 {
1134         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1135
1136 #ifdef CONFIG_SYSFS
1137         net->queues_kset = kset_create_and_add("queues",
1138             NULL, &net->dev.kobj);
1139         if (!net->queues_kset)
1140                 return -ENOMEM;
1141 #endif
1142
1143 #ifdef CONFIG_RPS
1144         real_rx = net->real_num_rx_queues;
1145 #endif
1146         real_tx = net->real_num_tx_queues;
1147
1148         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1149         if (error)
1150                 goto error;
1151         rxq = real_rx;
1152
1153         error = netdev_queue_update_kobjects(net, 0, real_tx);
1154         if (error)
1155                 goto error;
1156         txq = real_tx;
1157
1158         return 0;
1159
1160 error:
1161         netdev_queue_update_kobjects(net, txq, 0);
1162         net_rx_queue_update_kobjects(net, rxq, 0);
1163         return error;
1164 }
1165
1166 static void remove_queue_kobjects(struct net_device *net)
1167 {
1168         int real_rx = 0, real_tx = 0;
1169
1170 #ifdef CONFIG_RPS
1171         real_rx = net->real_num_rx_queues;
1172 #endif
1173         real_tx = net->real_num_tx_queues;
1174
1175         net_rx_queue_update_kobjects(net, real_rx, 0);
1176         netdev_queue_update_kobjects(net, real_tx, 0);
1177 #ifdef CONFIG_SYSFS
1178         kset_unregister(net->queues_kset);
1179 #endif
1180 }
1181
1182 static void *net_grab_current_ns(void)
1183 {
1184         struct net *ns = current->nsproxy->net_ns;
1185 #ifdef CONFIG_NET_NS
1186         if (ns)
1187                 atomic_inc(&ns->passive);
1188 #endif
1189         return ns;
1190 }
1191
1192 static const void *net_initial_ns(void)
1193 {
1194         return &init_net;
1195 }
1196
1197 static const void *net_netlink_ns(struct sock *sk)
1198 {
1199         return sock_net(sk);
1200 }
1201
1202 struct kobj_ns_type_operations net_ns_type_operations = {
1203         .type = KOBJ_NS_TYPE_NET,
1204         .grab_current_ns = net_grab_current_ns,
1205         .netlink_ns = net_netlink_ns,
1206         .initial_ns = net_initial_ns,
1207         .drop_ns = net_drop_ns,
1208 };
1209 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1210
1211 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1212 {
1213         struct net_device *dev = to_net_dev(d);
1214         int retval;
1215
1216         /* pass interface to uevent. */
1217         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1218         if (retval)
1219                 goto exit;
1220
1221         /* pass ifindex to uevent.
1222          * ifindex is useful as it won't change (interface name may change)
1223          * and is what RtNetlink uses natively. */
1224         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1225
1226 exit:
1227         return retval;
1228 }
1229
1230 /*
1231  *      netdev_release -- destroy and free a dead device.
1232  *      Called when last reference to device kobject is gone.
1233  */
1234 static void netdev_release(struct device *d)
1235 {
1236         struct net_device *dev = to_net_dev(d);
1237
1238         BUG_ON(dev->reg_state != NETREG_RELEASED);
1239
1240         kfree(dev->ifalias);
1241         kfree((char *)dev - dev->padded);
1242 }
1243
1244 static const void *net_namespace(struct device *d)
1245 {
1246         struct net_device *dev;
1247         dev = container_of(d, struct net_device, dev);
1248         return dev_net(dev);
1249 }
1250
1251 static struct class net_class = {
1252         .name = "net",
1253         .dev_release = netdev_release,
1254 #ifdef CONFIG_SYSFS
1255         .dev_attrs = net_class_attributes,
1256 #endif /* CONFIG_SYSFS */
1257         .dev_uevent = netdev_uevent,
1258         .ns_type = &net_ns_type_operations,
1259         .namespace = net_namespace,
1260 };
1261
1262 /* Delete sysfs entries but hold kobject reference until after all
1263  * netdev references are gone.
1264  */
1265 void netdev_unregister_kobject(struct net_device * net)
1266 {
1267         struct device *dev = &(net->dev);
1268
1269         kobject_get(&dev->kobj);
1270
1271         remove_queue_kobjects(net);
1272
1273         pm_runtime_set_memalloc_noio(dev, false);
1274
1275         device_del(dev);
1276 }
1277
1278 /* Create sysfs entries for network device. */
1279 int netdev_register_kobject(struct net_device *net)
1280 {
1281         struct device *dev = &(net->dev);
1282         const struct attribute_group **groups = net->sysfs_groups;
1283         int error = 0;
1284
1285         device_initialize(dev);
1286         dev->class = &net_class;
1287         dev->platform_data = net;
1288         dev->groups = groups;
1289
1290         dev_set_name(dev, "%s", net->name);
1291
1292 #ifdef CONFIG_SYSFS
1293         /* Allow for a device specific group */
1294         if (*groups)
1295                 groups++;
1296
1297         *groups++ = &netstat_group;
1298
1299 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1300         if (net->ieee80211_ptr)
1301                 *groups++ = &wireless_group;
1302 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1303         else if (net->wireless_handlers)
1304                 *groups++ = &wireless_group;
1305 #endif
1306 #endif
1307 #endif /* CONFIG_SYSFS */
1308
1309         error = device_add(dev);
1310         if (error)
1311                 return error;
1312
1313         error = register_queue_kobjects(net);
1314         if (error) {
1315                 device_del(dev);
1316                 return error;
1317         }
1318
1319         pm_runtime_set_memalloc_noio(dev, true);
1320
1321         return error;
1322 }
1323
1324 int netdev_class_create_file(struct class_attribute *class_attr)
1325 {
1326         return class_create_file(&net_class, class_attr);
1327 }
1328 EXPORT_SYMBOL(netdev_class_create_file);
1329
1330 void netdev_class_remove_file(struct class_attribute *class_attr)
1331 {
1332         class_remove_file(&net_class, class_attr);
1333 }
1334 EXPORT_SYMBOL(netdev_class_remove_file);
1335
1336 int netdev_kobject_init(void)
1337 {
1338         kobj_ns_type_register(&net_ns_type_operations);
1339         return class_register(&net_class);
1340 }