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
  25 #include "net-sysfs.h"
  26
  27 #ifdef CONFIG_SYSFS
  28 static const char fmt_hex[] = "%#x\n";
  29 static const char fmt_long_hex[] = "%#lx\n";
  30 static const char fmt_dec[] = "%d\n";
  31 static const char fmt_udec[] = "%u\n";
  32 static const char fmt_ulong[] = "%lu\n";
  33 static const char fmt_u64[] = "%llu\n";
  34
  35 static inline int dev_isalive(const struct net_device *dev)
  36 {
  37         return dev->reg_state <= NETREG_REGISTERED;
  38 }
  39
  40 /* use same locking rules as GIF* ioctl's */
  41 static ssize_t netdev_show(const struct device *dev,
  42                            struct device_attribute *attr, char *buf,
  43                            ssize_t (*format)(const struct net_device *, char *))
  44 {
  45         struct net_device *net = to_net_dev(dev);
  46         ssize_t ret = -EINVAL;
  47
  48         read_lock(&dev_base_lock);
  49         if (dev_isalive(net))
  50                 ret = (*format)(net, buf);
  51         read_unlock(&dev_base_lock);
  52
  53         return ret;
  54 }
  55
  56 /* generate a show function for simple field */
  57 #define NETDEVICE_SHOW(field, format_string)                            \
  58 static ssize_t format_##field(const struct net_device *net, char *buf)  \
  59 {                                                                       \
  60         return sprintf(buf, format_string, net->field);                 \
  61 }                                                                       \
  62 static ssize_t show_##field(struct device *dev,                         \
  63                             struct device_attribute *attr, char *buf)   \
  64 {                                                                       \
  65         return netdev_show(dev, attr, buf, format_##field);             \
  66 }
  67
  68
  69 /* use same locking and permission rules as SIF* ioctl's */
  70 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  71                             const char *buf, size_t len,
  72                             int (*set)(struct net_device *, unsigned long))
  73 {
  74         struct net_device *netdev = to_net_dev(dev);
  75         struct net *net = dev_net(netdev);
  76         unsigned long new;
  77         int ret = -EINVAL;
  78
  79         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  80                 return -EPERM;
  81
  82         ret = kstrtoul(buf, 0, &new);
  83         if (ret)
  84                 goto err;
  85
  86         if (!rtnl_trylock())
  87                 return restart_syscall();
  88
  89         if (dev_isalive(netdev)) {
  90                 if ((ret = (*set)(netdev, new)) == 0)
  91                         ret = len;
  92         }
  93         rtnl_unlock();
  94  err:
  95         return ret;
  96 }
  97
  98 NETDEVICE_SHOW(dev_id, fmt_hex);
  99 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 100 NETDEVICE_SHOW(addr_len, fmt_dec);
 101 NETDEVICE_SHOW(iflink, fmt_dec);
 102 NETDEVICE_SHOW(ifindex, fmt_dec);
 103 NETDEVICE_SHOW(type, fmt_dec);
 104 NETDEVICE_SHOW(link_mode, fmt_dec);
 105
 106 /* use same locking rules as GIFHWADDR ioctl's */
 107 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 108                             char *buf)
 109 {
 110         struct net_device *net = to_net_dev(dev);
 111         ssize_t ret = -EINVAL;
 112
 113         read_lock(&dev_base_lock);
 114         if (dev_isalive(net))
 115                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 116         read_unlock(&dev_base_lock);
 117         return ret;
 118 }
 119
 120 static ssize_t show_broadcast(struct device *dev,
 121                             struct device_attribute *attr, char *buf)
 122 {
 123         struct net_device *net = to_net_dev(dev);
 124         if (dev_isalive(net))
 125                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 126         return -EINVAL;
 127 }
 128
 129 static int change_carrier(struct net_device *net, unsigned long new_carrier)
 130 {
 131         if (!netif_running(net))
 132                 return -EINVAL;
 133         return dev_change_carrier(net, (bool) new_carrier);
 134 }
 135
 136 static ssize_t store_carrier(struct device *dev, struct device_attribute *attr,
 137                          const char *buf, size_t len)
 138 {
 139         return netdev_store(dev, attr, buf, len, change_carrier);
 140 }
 141
 142 static ssize_t show_carrier(struct device *dev,
 143                             struct device_attribute *attr, char *buf)
 144 {
 145         struct net_device *netdev = to_net_dev(dev);
 146         if (netif_running(netdev)) {
 147                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 148         }
 149         return -EINVAL;
 150 }
 151
 152 static ssize_t show_speed(struct device *dev,
 153                           struct device_attribute *attr, char *buf)
 154 {
 155         struct net_device *netdev = to_net_dev(dev);
 156         int ret = -EINVAL;
 157
 158         if (!rtnl_trylock())
 159                 return restart_syscall();
 160
 161         if (netif_running(netdev)) {
 162                 struct ethtool_cmd cmd;
 163                 if (!__ethtool_get_settings(netdev, &cmd))
 164                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 165         }
 166         rtnl_unlock();
 167         return ret;
 168 }
 169
 170 static ssize_t show_duplex(struct device *dev,
 171                            struct device_attribute *attr, char *buf)
 172 {
 173         struct net_device *netdev = to_net_dev(dev);
 174         int ret = -EINVAL;
 175
 176         if (!rtnl_trylock())
 177                 return restart_syscall();
 178
 179         if (netif_running(netdev)) {
 180                 struct ethtool_cmd cmd;
 181                 if (!__ethtool_get_settings(netdev, &cmd)) {
 182                         const char *duplex;
 183                         switch (cmd.duplex) {
 184                         case DUPLEX_HALF:
 185                                 duplex = "half";
 186                                 break;
 187                         case DUPLEX_FULL:
 188                                 duplex = "full";
 189                                 break;
 190                         default:
 191                                 duplex = "unknown";
 192                                 break;
 193                         }
 194                         ret = sprintf(buf, "%s\n", duplex);
 195                 }
 196         }
 197         rtnl_unlock();
 198         return ret;
 199 }
 200
 201 static ssize_t show_dormant(struct device *dev,
 202                             struct device_attribute *attr, char *buf)
 203 {
 204         struct net_device *netdev = to_net_dev(dev);
 205
 206         if (netif_running(netdev))
 207                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 208
 209         return -EINVAL;
 210 }
 211
 212 static const char *const operstates[] = {
 213         "unknown",
 214         "notpresent", /* currently unused */
 215         "down",
 216         "lowerlayerdown",
 217         "testing", /* currently unused */
 218         "dormant",
 219         "up"
 220 };
 221
 222 static ssize_t show_operstate(struct device *dev,
 223                               struct device_attribute *attr, char *buf)
 224 {
 225         const struct net_device *netdev = to_net_dev(dev);
 226         unsigned char operstate;
 227
 228         read_lock(&dev_base_lock);
 229         operstate = netdev->operstate;
 230         if (!netif_running(netdev))
 231                 operstate = IF_OPER_DOWN;
 232         read_unlock(&dev_base_lock);
 233
 234         if (operstate >= ARRAY_SIZE(operstates))
 235                 return -EINVAL; /* should not happen */
 236
 237         return sprintf(buf, "%s\n", operstates[operstate]);
 238 }
 239
 240 /* read-write attributes */
 241 NETDEVICE_SHOW(mtu, fmt_dec);
 242
 243 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 244 {
 245         return dev_set_mtu(net, (int) new_mtu);
 246 }
 247
 248 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 249                          const char *buf, size_t len)
 250 {
 251         return netdev_store(dev, attr, buf, len, change_mtu);
 252 }
 253
 254 NETDEVICE_SHOW(flags, fmt_hex);
 255
 256 static int change_flags(struct net_device *net, unsigned long new_flags)
 257 {
 258         return dev_change_flags(net, (unsigned int) new_flags);
 259 }
 260
 261 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 262                            const char *buf, size_t len)
 263 {
 264         return netdev_store(dev, attr, buf, len, change_flags);
 265 }
 266
 267 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 268
 269 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 270 {
 271         net->tx_queue_len = new_len;
 272         return 0;
 273 }
 274
 275 static ssize_t store_tx_queue_len(struct device *dev,
 276                                   struct device_attribute *attr,
 277                                   const char *buf, size_t len)
 278 {
 279         if (!capable(CAP_NET_ADMIN))
 280                 return -EPERM;
 281
 282         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 283 }
 284
 285 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 286                              const char *buf, size_t len)
 287 {
 288         struct net_device *netdev = to_net_dev(dev);
 289         struct net *net = dev_net(netdev);
 290         size_t count = len;
 291         ssize_t ret;
 292
 293         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 294                 return -EPERM;
 295
 296         /* ignore trailing newline */
 297         if (len >  0 && buf[len - 1] == '\n')
 298                 --count;
 299
 300         if (!rtnl_trylock())
 301                 return restart_syscall();
 302         ret = dev_set_alias(netdev, buf, count);
 303         rtnl_unlock();
 304
 305         return ret < 0 ? ret : len;
 306 }
 307
 308 static ssize_t show_ifalias(struct device *dev,
 309                             struct device_attribute *attr, char *buf)
 310 {
 311         const struct net_device *netdev = to_net_dev(dev);
 312         ssize_t ret = 0;
 313
 314         if (!rtnl_trylock())
 315                 return restart_syscall();
 316         if (netdev->ifalias)
 317                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 318         rtnl_unlock();
 319         return ret;
 320 }
 321
 322 NETDEVICE_SHOW(group, fmt_dec);
 323
 324 static int change_group(struct net_device *net, unsigned long new_group)
 325 {
 326         dev_set_group(net, (int) new_group);
 327         return 0;
 328 }
 329
 330 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 331                          const char *buf, size_t len)
 332 {
 333         return netdev_store(dev, attr, buf, len, change_group);
 334 }
 335
 336 static struct device_attribute net_class_attributes[] = {
 337         __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 338         __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 339         __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 340         __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 341         __ATTR(iflink, S_IRUGO, show_iflink, NULL),
 342         __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 343         __ATTR(type, S_IRUGO, show_type, NULL),
 344         __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 345         __ATTR(address, S_IRUGO, show_address, NULL),
 346         __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 347         __ATTR(carrier, S_IRUGO | S_IWUSR, show_carrier, store_carrier),
 348         __ATTR(speed, S_IRUGO, show_speed, NULL),
 349         __ATTR(duplex, S_IRUGO, show_duplex, NULL),
 350         __ATTR(dormant, S_IRUGO, show_dormant, NULL),
 351         __ATTR(operstate, S_IRUGO, show_operstate, NULL),
 352         __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 353         __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 354         __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 355                store_tx_queue_len),
 356         __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 357         {}
 358 };
 359
 360 /* Show a given an attribute in the statistics group */
 361 static ssize_t netstat_show(const struct device *d,
 362                             struct device_attribute *attr, char *buf,
 363                             unsigned long offset)
 364 {
 365         struct net_device *dev = to_net_dev(d);
 366         ssize_t ret = -EINVAL;
 367
 368         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 369                         offset % sizeof(u64) != 0);
 370
 371         read_lock(&dev_base_lock);
 372         if (dev_isalive(dev)) {
 373                 struct rtnl_link_stats64 temp;
 374                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 375
 376                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 377         }
 378         read_unlock(&dev_base_lock);
 379         return ret;
 380 }
 381
 382 /* generate a read-only statistics attribute */
 383 #define NETSTAT_ENTRY(name)                                             \
 384 static ssize_t show_##name(struct device *d,                            \
 385                            struct device_attribute *attr, char *buf)    \
 386 {                                                                       \
 387         return netstat_show(d, attr, buf,                               \
 388                             offsetof(struct rtnl_link_stats64, name));  \
 389 }                                                                       \
 390 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 391
 392 NETSTAT_ENTRY(rx_packets);
 393 NETSTAT_ENTRY(tx_packets);
 394 NETSTAT_ENTRY(rx_bytes);
 395 NETSTAT_ENTRY(tx_bytes);
 396 NETSTAT_ENTRY(rx_errors);
 397 NETSTAT_ENTRY(tx_errors);
 398 NETSTAT_ENTRY(rx_dropped);
 399 NETSTAT_ENTRY(tx_dropped);
 400 NETSTAT_ENTRY(multicast);
 401 NETSTAT_ENTRY(collisions);
 402 NETSTAT_ENTRY(rx_length_errors);
 403 NETSTAT_ENTRY(rx_over_errors);
 404 NETSTAT_ENTRY(rx_crc_errors);
 405 NETSTAT_ENTRY(rx_frame_errors);
 406 NETSTAT_ENTRY(rx_fifo_errors);
 407 NETSTAT_ENTRY(rx_missed_errors);
 408 NETSTAT_ENTRY(tx_aborted_errors);
 409 NETSTAT_ENTRY(tx_carrier_errors);
 410 NETSTAT_ENTRY(tx_fifo_errors);
 411 NETSTAT_ENTRY(tx_heartbeat_errors);
 412 NETSTAT_ENTRY(tx_window_errors);
 413 NETSTAT_ENTRY(rx_compressed);
 414 NETSTAT_ENTRY(tx_compressed);
 415
 416 static struct attribute *netstat_attrs[] = {
 417         &dev_attr_rx_packets.attr,
 418         &dev_attr_tx_packets.attr,
 419         &dev_attr_rx_bytes.attr,
 420         &dev_attr_tx_bytes.attr,
 421         &dev_attr_rx_errors.attr,
 422         &dev_attr_tx_errors.attr,
 423         &dev_attr_rx_dropped.attr,
 424         &dev_attr_tx_dropped.attr,
 425         &dev_attr_multicast.attr,
 426         &dev_attr_collisions.attr,
 427         &dev_attr_rx_length_errors.attr,
 428         &dev_attr_rx_over_errors.attr,
 429         &dev_attr_rx_crc_errors.attr,
 430         &dev_attr_rx_frame_errors.attr,
 431         &dev_attr_rx_fifo_errors.attr,
 432         &dev_attr_rx_missed_errors.attr,
 433         &dev_attr_tx_aborted_errors.attr,
 434         &dev_attr_tx_carrier_errors.attr,
 435         &dev_attr_tx_fifo_errors.attr,
 436         &dev_attr_tx_heartbeat_errors.attr,
 437         &dev_attr_tx_window_errors.attr,
 438         &dev_attr_rx_compressed.attr,
 439         &dev_attr_tx_compressed.attr,
 440         NULL
 441 };
 442
 443
 444 static struct attribute_group netstat_group = {
 445         .name  = "statistics",
 446         .attrs  = netstat_attrs,
 447 };
 448
 449 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 450 static struct attribute *wireless_attrs[] = {
 451         NULL
 452 };
 453
 454 static struct attribute_group wireless_group = {
 455         .name = "wireless",
 456         .attrs = wireless_attrs,
 457 };
 458 #endif
 459 #endif /* CONFIG_SYSFS */
 460
 461 #ifdef CONFIG_RPS
 462 /*
 463  * RX queue sysfs structures and functions.
 464  */
 465 struct rx_queue_attribute {
 466         struct attribute attr;
 467         ssize_t (*show)(struct netdev_rx_queue *queue,
 468             struct rx_queue_attribute *attr, char *buf);
 469         ssize_t (*store)(struct netdev_rx_queue *queue,
 470             struct rx_queue_attribute *attr, const char *buf, size_t len);
 471 };
 472 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 473     struct rx_queue_attribute, attr)
 474
 475 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 476
 477 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 478                                   char *buf)
 479 {
 480         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 481         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 482
 483         if (!attribute->show)
 484                 return -EIO;
 485
 486         return attribute->show(queue, attribute, buf);
 487 }
 488
 489 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 490                                    const char *buf, size_t count)
 491 {
 492         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 493         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 494
 495         if (!attribute->store)
 496                 return -EIO;
 497
 498         return attribute->store(queue, attribute, buf, count);
 499 }
 500
 501 static const struct sysfs_ops rx_queue_sysfs_ops = {
 502         .show = rx_queue_attr_show,
 503         .store = rx_queue_attr_store,
 504 };
 505
 506 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 507                             struct rx_queue_attribute *attribute, char *buf)
 508 {
 509         struct rps_map *map;
 510         cpumask_var_t mask;
 511         size_t len = 0;
 512         int i;
 513
 514         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 515                 return -ENOMEM;
 516
 517         rcu_read_lock();
 518         map = rcu_dereference(queue->rps_map);
 519         if (map)
 520                 for (i = 0; i < map->len; i++)
 521                         cpumask_set_cpu(map->cpus[i], mask);
 522
 523         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 524         if (PAGE_SIZE - len < 3) {
 525                 rcu_read_unlock();
 526                 free_cpumask_var(mask);
 527                 return -EINVAL;
 528         }
 529         rcu_read_unlock();
 530
 531         free_cpumask_var(mask);
 532         len += sprintf(buf + len, "\n");
 533         return len;
 534 }
 535
 536 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 537                       struct rx_queue_attribute *attribute,
 538                       const char *buf, size_t len)
 539 {
 540         struct rps_map *old_map, *map;
 541         cpumask_var_t mask;
 542         int err, cpu, i;
 543         static DEFINE_SPINLOCK(rps_map_lock);
 544
 545         if (!capable(CAP_NET_ADMIN))
 546                 return -EPERM;
 547
 548         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 549                 return -ENOMEM;
 550
 551         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 552         if (err) {
 553                 free_cpumask_var(mask);
 554                 return err;
 555         }
 556
 557         map = kzalloc(max_t(unsigned int,
 558             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 559             GFP_KERNEL);
 560         if (!map) {
 561                 free_cpumask_var(mask);
 562                 return -ENOMEM;
 563         }
 564
 565         i = 0;
 566         for_each_cpu_and(cpu, mask, cpu_online_mask)
 567                 map->cpus[i++] = cpu;
 568
 569         if (i)
 570                 map->len = i;
 571         else {
 572                 kfree(map);
 573                 map = NULL;
 574         }
 575
 576         spin_lock(&rps_map_lock);
 577         old_map = rcu_dereference_protected(queue->rps_map,
 578                                             lockdep_is_held(&rps_map_lock));
 579         rcu_assign_pointer(queue->rps_map, map);
 580         spin_unlock(&rps_map_lock);
 581
 582         if (map)
 583                 static_key_slow_inc(&rps_needed);
 584         if (old_map) {
 585                 kfree_rcu(old_map, rcu);
 586                 static_key_slow_dec(&rps_needed);
 587         }
 588         free_cpumask_var(mask);
 589         return len;
 590 }
 591
 592 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 593                                            struct rx_queue_attribute *attr,
 594                                            char *buf)
 595 {
 596         struct rps_dev_flow_table *flow_table;
 597         unsigned long val = 0;
 598
 599         rcu_read_lock();
 600         flow_table = rcu_dereference(queue->rps_flow_table);
 601         if (flow_table)
 602                 val = (unsigned long)flow_table->mask + 1;
 603         rcu_read_unlock();
 604
 605         return sprintf(buf, "%lu\n", val);
 606 }
 607
 608 static void rps_dev_flow_table_release_work(struct work_struct *work)
 609 {
 610         struct rps_dev_flow_table *table = container_of(work,
 611             struct rps_dev_flow_table, free_work);
 612
 613         vfree(table);
 614 }
 615
 616 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 617 {
 618         struct rps_dev_flow_table *table = container_of(rcu,
 619             struct rps_dev_flow_table, rcu);
 620
 621         INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 622         schedule_work(&table->free_work);
 623 }
 624
 625 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 626                                      struct rx_queue_attribute *attr,
 627                                      const char *buf, size_t len)
 628 {
 629         unsigned long mask, count;
 630         struct rps_dev_flow_table *table, *old_table;
 631         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 632         int rc;
 633
 634         if (!capable(CAP_NET_ADMIN))
 635                 return -EPERM;
 636
 637         rc = kstrtoul(buf, 0, &count);
 638         if (rc < 0)
 639                 return rc;
 640
 641         if (count) {
 642                 mask = count - 1;
 643                 /* mask = roundup_pow_of_two(count) - 1;
 644                  * without overflows...
 645                  */
 646                 while ((mask | (mask >> 1)) != mask)
 647                         mask |= (mask >> 1);
 648                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 649                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 650                  * doesnt overflow.
 651                  */
 652 #if BITS_PER_LONG > 32
 653                 if (mask > (unsigned long)(u32)mask)
 654                         return -EINVAL;
 655 #else
 656                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 657                                 / sizeof(struct rps_dev_flow)) {
 658                         /* Enforce a limit to prevent overflow */
 659                         return -EINVAL;
 660                 }
 661 #endif
 662                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 663                 if (!table)
 664                         return -ENOMEM;
 665
 666                 table->mask = mask;
 667                 for (count = 0; count <= mask; count++)
 668                         table->flows[count].cpu = RPS_NO_CPU;
 669         } else
 670                 table = NULL;
 671
 672         spin_lock(&rps_dev_flow_lock);
 673         old_table = rcu_dereference_protected(queue->rps_flow_table,
 674                                               lockdep_is_held(&rps_dev_flow_lock));
 675         rcu_assign_pointer(queue->rps_flow_table, table);
 676         spin_unlock(&rps_dev_flow_lock);
 677
 678         if (old_table)
 679                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 680
 681         return len;
 682 }
 683
 684 static struct rx_queue_attribute rps_cpus_attribute =
 685         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 686
 687
 688 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 689         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 690             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 691
 692 static struct attribute *rx_queue_default_attrs[] = {
 693         &rps_cpus_attribute.attr,
 694         &rps_dev_flow_table_cnt_attribute.attr,
 695         NULL
 696 };
 697
 698 static void rx_queue_release(struct kobject *kobj)
 699 {
 700         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 701         struct rps_map *map;
 702         struct rps_dev_flow_table *flow_table;
 703
 704
 705         map = rcu_dereference_protected(queue->rps_map, 1);
 706         if (map) {
 707                 RCU_INIT_POINTER(queue->rps_map, NULL);
 708                 kfree_rcu(map, rcu);
 709         }
 710
 711         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 712         if (flow_table) {
 713                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 714                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 715         }
 716
 717         memset(kobj, 0, sizeof(*kobj));
 718         dev_put(queue->dev);
 719 }
 720
 721 static struct kobj_type rx_queue_ktype = {
 722         .sysfs_ops = &rx_queue_sysfs_ops,
 723         .release = rx_queue_release,
 724         .default_attrs = rx_queue_default_attrs,
 725 };
 726
 727 static int rx_queue_add_kobject(struct net_device *net, int index)
 728 {
 729         struct netdev_rx_queue *queue = net->_rx + index;
 730         struct kobject *kobj = &queue->kobj;
 731         int error = 0;
 732
 733         kobj->kset = net->queues_kset;
 734         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 735             "rx-%u", index);
 736         if (error) {
 737                 kobject_put(kobj);
 738                 return error;
 739         }
 740
 741         kobject_uevent(kobj, KOBJ_ADD);
 742         dev_hold(queue->dev);
 743
 744         return error;
 745 }
 746 #endif /* CONFIG_RPS */
 747
 748 int
 749 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 750 {
 751 #ifdef CONFIG_RPS
 752         int i;
 753         int error = 0;
 754
 755         for (i = old_num; i < new_num; i++) {
 756                 error = rx_queue_add_kobject(net, i);
 757                 if (error) {
 758                         new_num = old_num;
 759                         break;
 760                 }
 761         }
 762
 763         while (--i >= new_num)
 764                 kobject_put(&net->_rx[i].kobj);
 765
 766         return error;
 767 #else
 768         return 0;
 769 #endif
 770 }
 771
 772 #ifdef CONFIG_SYSFS
 773 /*
 774  * netdev_queue sysfs structures and functions.
 775  */
 776 struct netdev_queue_attribute {
 777         struct attribute attr;
 778         ssize_t (*show)(struct netdev_queue *queue,
 779             struct netdev_queue_attribute *attr, char *buf);
 780         ssize_t (*store)(struct netdev_queue *queue,
 781             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 782 };
 783 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 784     struct netdev_queue_attribute, attr)
 785
 786 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 787
 788 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 789                                       struct attribute *attr, char *buf)
 790 {
 791         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 792         struct netdev_queue *queue = to_netdev_queue(kobj);
 793
 794         if (!attribute->show)
 795                 return -EIO;
 796
 797         return attribute->show(queue, attribute, buf);
 798 }
 799
 800 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 801                                        struct attribute *attr,
 802                                        const char *buf, size_t count)
 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->store)
 808                 return -EIO;
 809
 810         return attribute->store(queue, attribute, buf, count);
 811 }
 812
 813 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 814         .show = netdev_queue_attr_show,
 815         .store = netdev_queue_attr_store,
 816 };
 817
 818 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 819                                   struct netdev_queue_attribute *attribute,
 820                                   char *buf)
 821 {
 822         unsigned long trans_timeout;
 823
 824         spin_lock_irq(&queue->_xmit_lock);
 825         trans_timeout = queue->trans_timeout;
 826         spin_unlock_irq(&queue->_xmit_lock);
 827
 828         return sprintf(buf, "%lu", trans_timeout);
 829 }
 830
 831 static struct netdev_queue_attribute queue_trans_timeout =
 832         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 833
 834 #ifdef CONFIG_BQL
 835 /*
 836  * Byte queue limits sysfs structures and functions.
 837  */
 838 static ssize_t bql_show(char *buf, unsigned int value)
 839 {
 840         return sprintf(buf, "%u\n", value);
 841 }
 842
 843 static ssize_t bql_set(const char *buf, const size_t count,
 844                        unsigned int *pvalue)
 845 {
 846         unsigned int value;
 847         int err;
 848
 849         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 850                 value = DQL_MAX_LIMIT;
 851         else {
 852                 err = kstrtouint(buf, 10, &value);
 853                 if (err < 0)
 854                         return err;
 855                 if (value > DQL_MAX_LIMIT)
 856                         return -EINVAL;
 857         }
 858
 859         *pvalue = value;
 860
 861         return count;
 862 }
 863
 864 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 865                                   struct netdev_queue_attribute *attr,
 866                                   char *buf)
 867 {
 868         struct dql *dql = &queue->dql;
 869
 870         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 871 }
 872
 873 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 874                                  struct netdev_queue_attribute *attribute,
 875                                  const char *buf, size_t len)
 876 {
 877         struct dql *dql = &queue->dql;
 878         unsigned int value;
 879         int err;
 880
 881         err = kstrtouint(buf, 10, &value);
 882         if (err < 0)
 883                 return err;
 884
 885         dql->slack_hold_time = msecs_to_jiffies(value);
 886
 887         return len;
 888 }
 889
 890 static struct netdev_queue_attribute bql_hold_time_attribute =
 891         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 892             bql_set_hold_time);
 893
 894 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 895                                  struct netdev_queue_attribute *attr,
 896                                  char *buf)
 897 {
 898         struct dql *dql = &queue->dql;
 899
 900         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 901 }
 902
 903 static struct netdev_queue_attribute bql_inflight_attribute =
 904         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 905
 906 #define BQL_ATTR(NAME, FIELD)                                           \
 907 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 908                                  struct netdev_queue_attribute *attr,   \
 909                                  char *buf)                             \
 910 {                                                                       \
 911         return bql_show(buf, queue->dql.FIELD);                         \
 912 }                                                                       \
 913                                                                         \
 914 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 915                                 struct netdev_queue_attribute *attr,    \
 916                                 const char *buf, size_t len)            \
 917 {                                                                       \
 918         return bql_set(buf, len, &queue->dql.FIELD);                    \
 919 }                                                                       \
 920                                                                         \
 921 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 922         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 923             bql_set_ ## NAME);
 924
 925 BQL_ATTR(limit, limit)
 926 BQL_ATTR(limit_max, max_limit)
 927 BQL_ATTR(limit_min, min_limit)
 928
 929 static struct attribute *dql_attrs[] = {
 930         &bql_limit_attribute.attr,
 931         &bql_limit_max_attribute.attr,
 932         &bql_limit_min_attribute.attr,
 933         &bql_hold_time_attribute.attr,
 934         &bql_inflight_attribute.attr,
 935         NULL
 936 };
 937
 938 static struct attribute_group dql_group = {
 939         .name  = "byte_queue_limits",
 940         .attrs  = dql_attrs,
 941 };
 942 #endif /* CONFIG_BQL */
 943
 944 #ifdef CONFIG_XPS
 945 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 946 {
 947         struct net_device *dev = queue->dev;
 948         int i;
 949
 950         for (i = 0; i < dev->num_tx_queues; i++)
 951                 if (queue == &dev->_tx[i])
 952                         break;
 953
 954         BUG_ON(i >= dev->num_tx_queues);
 955
 956         return i;
 957 }
 958
 959
 960 static ssize_t show_xps_map(struct netdev_queue *queue,
 961                             struct netdev_queue_attribute *attribute, char *buf)
 962 {
 963         struct net_device *dev = queue->dev;
 964         struct xps_dev_maps *dev_maps;
 965         cpumask_var_t mask;
 966         unsigned long index;
 967         size_t len = 0;
 968         int i;
 969
 970         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 971                 return -ENOMEM;
 972
 973         index = get_netdev_queue_index(queue);
 974
 975         rcu_read_lock();
 976         dev_maps = rcu_dereference(dev->xps_maps);
 977         if (dev_maps) {
 978                 for_each_possible_cpu(i) {
 979                         struct xps_map *map =
 980                             rcu_dereference(dev_maps->cpu_map[i]);
 981                         if (map) {
 982                                 int j;
 983                                 for (j = 0; j < map->len; j++) {
 984                                         if (map->queues[j] == index) {
 985                                                 cpumask_set_cpu(i, mask);
 986                                                 break;
 987                                         }
 988                                 }
 989                         }
 990                 }
 991         }
 992         rcu_read_unlock();
 993
 994         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 995         if (PAGE_SIZE - len < 3) {
 996                 free_cpumask_var(mask);
 997                 return -EINVAL;
 998         }
 999
1000         free_cpumask_var(mask);
1001         len += sprintf(buf + len, "\n");
1002         return len;
1003 }
1004
1005 static ssize_t store_xps_map(struct netdev_queue *queue,
1006                       struct netdev_queue_attribute *attribute,
1007                       const char *buf, size_t len)
1008 {
1009         struct net_device *dev = queue->dev;
1010         unsigned long index;
1011         cpumask_var_t mask;
1012         int err;
1013
1014         if (!capable(CAP_NET_ADMIN))
1015                 return -EPERM;
1016
1017         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1018                 return -ENOMEM;
1019
1020         index = get_netdev_queue_index(queue);
1021
1022         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1023         if (err) {
1024                 free_cpumask_var(mask);
1025                 return err;
1026         }
1027
1028         err = netif_set_xps_queue(dev, mask, index);
1029
1030         free_cpumask_var(mask);
1031
1032         return err ? : len;
1033 }
1034
1035 static struct netdev_queue_attribute xps_cpus_attribute =
1036     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1037 #endif /* CONFIG_XPS */
1038
1039 static struct attribute *netdev_queue_default_attrs[] = {
1040         &queue_trans_timeout.attr,
1041 #ifdef CONFIG_XPS
1042         &xps_cpus_attribute.attr,
1043 #endif
1044         NULL
1045 };
1046
1047 static void netdev_queue_release(struct kobject *kobj)
1048 {
1049         struct netdev_queue *queue = to_netdev_queue(kobj);
1050
1051         memset(kobj, 0, sizeof(*kobj));
1052         dev_put(queue->dev);
1053 }
1054
1055 static struct kobj_type netdev_queue_ktype = {
1056         .sysfs_ops = &netdev_queue_sysfs_ops,
1057         .release = netdev_queue_release,
1058         .default_attrs = netdev_queue_default_attrs,
1059 };
1060
1061 static int netdev_queue_add_kobject(struct net_device *net, int index)
1062 {
1063         struct netdev_queue *queue = net->_tx + index;
1064         struct kobject *kobj = &queue->kobj;
1065         int error = 0;
1066
1067         kobj->kset = net->queues_kset;
1068         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1069             "tx-%u", index);
1070         if (error)
1071                 goto exit;
1072
1073 #ifdef CONFIG_BQL
1074         error = sysfs_create_group(kobj, &dql_group);
1075         if (error)
1076                 goto exit;
1077 #endif
1078
1079         kobject_uevent(kobj, KOBJ_ADD);
1080         dev_hold(queue->dev);
1081
1082         return 0;
1083 exit:
1084         kobject_put(kobj);
1085         return error;
1086 }
1087 #endif /* CONFIG_SYSFS */
1088
1089 int
1090 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1091 {
1092 #ifdef CONFIG_SYSFS
1093         int i;
1094         int error = 0;
1095
1096         for (i = old_num; i < new_num; i++) {
1097                 error = netdev_queue_add_kobject(net, i);
1098                 if (error) {
1099                         new_num = old_num;
1100                         break;
1101                 }
1102         }
1103
1104         while (--i >= new_num) {
1105                 struct netdev_queue *queue = net->_tx + i;
1106
1107 #ifdef CONFIG_BQL
1108                 sysfs_remove_group(&queue->kobj, &dql_group);
1109 #endif
1110                 kobject_put(&queue->kobj);
1111         }
1112
1113         return error;
1114 #else
1115         return 0;
1116 #endif /* CONFIG_SYSFS */
1117 }
1118
1119 static int register_queue_kobjects(struct net_device *net)
1120 {
1121         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1122
1123 #ifdef CONFIG_SYSFS
1124         net->queues_kset = kset_create_and_add("queues",
1125             NULL, &net->dev.kobj);
1126         if (!net->queues_kset)
1127                 return -ENOMEM;
1128 #endif
1129
1130 #ifdef CONFIG_RPS
1131         real_rx = net->real_num_rx_queues;
1132 #endif
1133         real_tx = net->real_num_tx_queues;
1134
1135         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1136         if (error)
1137                 goto error;
1138         rxq = real_rx;
1139
1140         error = netdev_queue_update_kobjects(net, 0, real_tx);
1141         if (error)
1142                 goto error;
1143         txq = real_tx;
1144
1145         return 0;
1146
1147 error:
1148         netdev_queue_update_kobjects(net, txq, 0);
1149         net_rx_queue_update_kobjects(net, rxq, 0);
1150         return error;
1151 }
1152
1153 static void remove_queue_kobjects(struct net_device *net)
1154 {
1155         int real_rx = 0, real_tx = 0;
1156
1157 #ifdef CONFIG_RPS
1158         real_rx = net->real_num_rx_queues;
1159 #endif
1160         real_tx = net->real_num_tx_queues;
1161
1162         net_rx_queue_update_kobjects(net, real_rx, 0);
1163         netdev_queue_update_kobjects(net, real_tx, 0);
1164 #ifdef CONFIG_SYSFS
1165         kset_unregister(net->queues_kset);
1166 #endif
1167 }
1168
1169 static void *net_grab_current_ns(void)
1170 {
1171         struct net *ns = current->nsproxy->net_ns;
1172 #ifdef CONFIG_NET_NS
1173         if (ns)
1174                 atomic_inc(&ns->passive);
1175 #endif
1176         return ns;
1177 }
1178
1179 static const void *net_initial_ns(void)
1180 {
1181         return &init_net;
1182 }
1183
1184 static const void *net_netlink_ns(struct sock *sk)
1185 {
1186         return sock_net(sk);
1187 }
1188
1189 struct kobj_ns_type_operations net_ns_type_operations = {
1190         .type = KOBJ_NS_TYPE_NET,
1191         .grab_current_ns = net_grab_current_ns,
1192         .netlink_ns = net_netlink_ns,
1193         .initial_ns = net_initial_ns,
1194         .drop_ns = net_drop_ns,
1195 };
1196 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1197
1198 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1199 {
1200         struct net_device *dev = to_net_dev(d);
1201         int retval;
1202
1203         /* pass interface to uevent. */
1204         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1205         if (retval)
1206                 goto exit;
1207
1208         /* pass ifindex to uevent.
1209          * ifindex is useful as it won't change (interface name may change)
1210          * and is what RtNetlink uses natively. */
1211         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1212
1213 exit:
1214         return retval;
1215 }
1216
1217 /*
1218  *      netdev_release -- destroy and free a dead device.
1219  *      Called when last reference to device kobject is gone.
1220  */
1221 static void netdev_release(struct device *d)
1222 {
1223         struct net_device *dev = to_net_dev(d);
1224
1225         BUG_ON(dev->reg_state != NETREG_RELEASED);
1226
1227         kfree(dev->ifalias);
1228         kfree((char *)dev - dev->padded);
1229 }
1230
1231 static const void *net_namespace(struct device *d)
1232 {
1233         struct net_device *dev;
1234         dev = container_of(d, struct net_device, dev);
1235         return dev_net(dev);
1236 }
1237
1238 static struct class net_class = {
1239         .name = "net",
1240         .dev_release = netdev_release,
1241 #ifdef CONFIG_SYSFS
1242         .dev_attrs = net_class_attributes,
1243 #endif /* CONFIG_SYSFS */
1244         .dev_uevent = netdev_uevent,
1245         .ns_type = &net_ns_type_operations,
1246         .namespace = net_namespace,
1247 };
1248
1249 /* Delete sysfs entries but hold kobject reference until after all
1250  * netdev references are gone.
1251  */
1252 void netdev_unregister_kobject(struct net_device * net)
1253 {
1254         struct device *dev = &(net->dev);
1255
1256         kobject_get(&dev->kobj);
1257
1258         remove_queue_kobjects(net);
1259
1260         device_del(dev);
1261 }
1262
1263 /* Create sysfs entries for network device. */
1264 int netdev_register_kobject(struct net_device *net)
1265 {
1266         struct device *dev = &(net->dev);
1267         const struct attribute_group **groups = net->sysfs_groups;
1268         int error = 0;
1269
1270         device_initialize(dev);
1271         dev->class = &net_class;
1272         dev->platform_data = net;
1273         dev->groups = groups;
1274
1275         dev_set_name(dev, "%s", net->name);
1276
1277 #ifdef CONFIG_SYSFS
1278         /* Allow for a device specific group */
1279         if (*groups)
1280                 groups++;
1281
1282         *groups++ = &netstat_group;
1283
1284 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1285         if (net->ieee80211_ptr)
1286                 *groups++ = &wireless_group;
1287 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1288         else if (net->wireless_handlers)
1289                 *groups++ = &wireless_group;
1290 #endif
1291 #endif
1292 #endif /* CONFIG_SYSFS */
1293
1294         error = device_add(dev);
1295         if (error)
1296                 return error;
1297
1298         error = register_queue_kobjects(net);
1299         if (error) {
1300                 device_del(dev);
1301                 return error;
1302         }
1303
1304         return error;
1305 }
1306
1307 int netdev_class_create_file(struct class_attribute *class_attr)
1308 {
1309         return class_create_file(&net_class, class_attr);
1310 }
1311 EXPORT_SYMBOL(netdev_class_create_file);
1312
1313 void netdev_class_remove_file(struct class_attribute *class_attr)
1314 {
1315         class_remove_file(&net_class, class_attr);
1316 }
1317 EXPORT_SYMBOL(netdev_class_remove_file);
1318
1319 int netdev_kobject_init(void)
1320 {
1321         kobj_ns_type_register(&net_ns_type_operations);
1322         return class_register(&net_class);
1323 }