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 field##_show(struct device *dev,                         \
  64                             struct device_attribute *attr, char *buf)   \
  65 {                                                                       \
  66         return netdev_show(dev, attr, buf, format_##field);             \
  67 }                                                                       \
  68
  69 #define NETDEVICE_SHOW_RO(field, format_string)                         \
  70 NETDEVICE_SHOW(field, format_string);                                   \
  71 static DEVICE_ATTR_RO(field)
  72
  73 #define NETDEVICE_SHOW_RW(field, format_string)                         \
  74 NETDEVICE_SHOW(field, format_string);                                   \
  75 static DEVICE_ATTR_RW(field)
  76
  77 /* use same locking and permission rules as SIF* ioctl's */
  78 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  79                             const char *buf, size_t len,
  80                             int (*set)(struct net_device *, unsigned long))
  81 {
  82         struct net_device *netdev = to_net_dev(dev);
  83         struct net *net = dev_net(netdev);
  84         unsigned long new;
  85         int ret = -EINVAL;
  86
  87         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  88                 return -EPERM;
  89
  90         ret = kstrtoul(buf, 0, &new);
  91         if (ret)
  92                 goto err;
  93
  94         if (!rtnl_trylock())
  95                 return restart_syscall();
  96
  97         if (dev_isalive(netdev)) {
  98                 if ((ret = (*set)(netdev, new)) == 0)
  99                         ret = len;
 100         }
 101         rtnl_unlock();
 102  err:
 103         return ret;
 104 }
 105
 106 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 107 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 108 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 109 NETDEVICE_SHOW_RO(iflink, fmt_dec);
 110 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 111 NETDEVICE_SHOW_RO(type, fmt_dec);
 112 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 113
 114 /* use same locking rules as GIFHWADDR ioctl's */
 115 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 116                             char *buf)
 117 {
 118         struct net_device *net = to_net_dev(dev);
 119         ssize_t ret = -EINVAL;
 120
 121         read_lock(&dev_base_lock);
 122         if (dev_isalive(net))
 123                 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 124         read_unlock(&dev_base_lock);
 125         return ret;
 126 }
 127 static DEVICE_ATTR_RO(address);
 128
 129 static ssize_t broadcast_show(struct device *dev,
 130                               struct device_attribute *attr, char *buf)
 131 {
 132         struct net_device *net = to_net_dev(dev);
 133         if (dev_isalive(net))
 134                 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 135         return -EINVAL;
 136 }
 137 static DEVICE_ATTR_RO(broadcast);
 138
 139 static int change_carrier(struct net_device *net, unsigned long new_carrier)
 140 {
 141         if (!netif_running(net))
 142                 return -EINVAL;
 143         return dev_change_carrier(net, (bool) new_carrier);
 144 }
 145
 146 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 147                              const char *buf, size_t len)
 148 {
 149         return netdev_store(dev, attr, buf, len, change_carrier);
 150 }
 151
 152 static ssize_t carrier_show(struct device *dev,
 153                             struct device_attribute *attr, char *buf)
 154 {
 155         struct net_device *netdev = to_net_dev(dev);
 156         if (netif_running(netdev)) {
 157                 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 158         }
 159         return -EINVAL;
 160 }
 161 static DEVICE_ATTR_RW(carrier);
 162
 163 static ssize_t speed_show(struct device *dev,
 164                           struct device_attribute *attr, char *buf)
 165 {
 166         struct net_device *netdev = to_net_dev(dev);
 167         int ret = -EINVAL;
 168
 169         if (!rtnl_trylock())
 170                 return restart_syscall();
 171
 172         if (netif_running(netdev)) {
 173                 struct ethtool_cmd cmd;
 174                 if (!__ethtool_get_settings(netdev, &cmd))
 175                         ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 176         }
 177         rtnl_unlock();
 178         return ret;
 179 }
 180 static DEVICE_ATTR_RO(speed);
 181
 182 static ssize_t duplex_show(struct device *dev,
 183                            struct device_attribute *attr, char *buf)
 184 {
 185         struct net_device *netdev = to_net_dev(dev);
 186         int ret = -EINVAL;
 187
 188         if (!rtnl_trylock())
 189                 return restart_syscall();
 190
 191         if (netif_running(netdev)) {
 192                 struct ethtool_cmd cmd;
 193                 if (!__ethtool_get_settings(netdev, &cmd)) {
 194                         const char *duplex;
 195                         switch (cmd.duplex) {
 196                         case DUPLEX_HALF:
 197                                 duplex = "half";
 198                                 break;
 199                         case DUPLEX_FULL:
 200                                 duplex = "full";
 201                                 break;
 202                         default:
 203                                 duplex = "unknown";
 204                                 break;
 205                         }
 206                         ret = sprintf(buf, "%s\n", duplex);
 207                 }
 208         }
 209         rtnl_unlock();
 210         return ret;
 211 }
 212 static DEVICE_ATTR_RO(duplex);
 213
 214 static ssize_t dormant_show(struct device *dev,
 215                             struct device_attribute *attr, char *buf)
 216 {
 217         struct net_device *netdev = to_net_dev(dev);
 218
 219         if (netif_running(netdev))
 220                 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 221
 222         return -EINVAL;
 223 }
 224 static DEVICE_ATTR_RO(dormant);
 225
 226 static const char *const operstates[] = {
 227         "unknown",
 228         "notpresent", /* currently unused */
 229         "down",
 230         "lowerlayerdown",
 231         "testing", /* currently unused */
 232         "dormant",
 233         "up"
 234 };
 235
 236 static ssize_t operstate_show(struct device *dev,
 237                               struct device_attribute *attr, char *buf)
 238 {
 239         const struct net_device *netdev = to_net_dev(dev);
 240         unsigned char operstate;
 241
 242         read_lock(&dev_base_lock);
 243         operstate = netdev->operstate;
 244         if (!netif_running(netdev))
 245                 operstate = IF_OPER_DOWN;
 246         read_unlock(&dev_base_lock);
 247
 248         if (operstate >= ARRAY_SIZE(operstates))
 249                 return -EINVAL; /* should not happen */
 250
 251         return sprintf(buf, "%s\n", operstates[operstate]);
 252 }
 253 static DEVICE_ATTR_RO(operstate);
 254
 255 /* read-write attributes */
 256
 257 static int change_mtu(struct net_device *net, unsigned long new_mtu)
 258 {
 259         return dev_set_mtu(net, (int) new_mtu);
 260 }
 261
 262 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 263                          const char *buf, size_t len)
 264 {
 265         return netdev_store(dev, attr, buf, len, change_mtu);
 266 }
 267 NETDEVICE_SHOW_RW(mtu, fmt_dec);
 268
 269 static int change_flags(struct net_device *net, unsigned long new_flags)
 270 {
 271         return dev_change_flags(net, (unsigned int) new_flags);
 272 }
 273
 274 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 275                            const char *buf, size_t len)
 276 {
 277         return netdev_store(dev, attr, buf, len, change_flags);
 278 }
 279 NETDEVICE_SHOW_RW(flags, fmt_hex);
 280
 281 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 282 {
 283         net->tx_queue_len = new_len;
 284         return 0;
 285 }
 286
 287 static ssize_t tx_queue_len_store(struct device *dev,
 288                                   struct device_attribute *attr,
 289                                   const char *buf, size_t len)
 290 {
 291         if (!capable(CAP_NET_ADMIN))
 292                 return -EPERM;
 293
 294         return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 295 }
 296 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
 297
 298 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 299                              const char *buf, size_t len)
 300 {
 301         struct net_device *netdev = to_net_dev(dev);
 302         struct net *net = dev_net(netdev);
 303         size_t count = len;
 304         ssize_t ret;
 305
 306         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 307                 return -EPERM;
 308
 309         /* ignore trailing newline */
 310         if (len >  0 && buf[len - 1] == '\n')
 311                 --count;
 312
 313         if (!rtnl_trylock())
 314                 return restart_syscall();
 315         ret = dev_set_alias(netdev, buf, count);
 316         rtnl_unlock();
 317
 318         return ret < 0 ? ret : len;
 319 }
 320
 321 static ssize_t ifalias_show(struct device *dev,
 322                             struct device_attribute *attr, char *buf)
 323 {
 324         const struct net_device *netdev = to_net_dev(dev);
 325         ssize_t ret = 0;
 326
 327         if (!rtnl_trylock())
 328                 return restart_syscall();
 329         if (netdev->ifalias)
 330                 ret = sprintf(buf, "%s\n", netdev->ifalias);
 331         rtnl_unlock();
 332         return ret;
 333 }
 334 static DEVICE_ATTR_RW(ifalias);
 335
 336 static int change_group(struct net_device *net, unsigned long new_group)
 337 {
 338         dev_set_group(net, (int) new_group);
 339         return 0;
 340 }
 341
 342 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 343                            const char *buf, size_t len)
 344 {
 345         return netdev_store(dev, attr, buf, len, change_group);
 346 }
 347 NETDEVICE_SHOW(group, fmt_dec);
 348 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 349
 350 static struct attribute *net_class_attrs[] = {
 351         &dev_attr_netdev_group.attr,
 352         &dev_attr_type.attr,
 353         &dev_attr_dev_id.attr,
 354         &dev_attr_iflink.attr,
 355         &dev_attr_ifindex.attr,
 356         &dev_attr_addr_assign_type.attr,
 357         &dev_attr_addr_len.attr,
 358         &dev_attr_link_mode.attr,
 359         &dev_attr_address.attr,
 360         &dev_attr_broadcast.attr,
 361         &dev_attr_speed.attr,
 362         &dev_attr_duplex.attr,
 363         &dev_attr_dormant.attr,
 364         &dev_attr_operstate.attr,
 365         &dev_attr_ifalias.attr,
 366         &dev_attr_carrier.attr,
 367         &dev_attr_mtu.attr,
 368         &dev_attr_flags.attr,
 369         &dev_attr_tx_queue_len.attr,
 370         NULL,
 371 };
 372 ATTRIBUTE_GROUPS(net_class);
 373
 374 /* Show a given an attribute in the statistics group */
 375 static ssize_t netstat_show(const struct device *d,
 376                             struct device_attribute *attr, char *buf,
 377                             unsigned long offset)
 378 {
 379         struct net_device *dev = to_net_dev(d);
 380         ssize_t ret = -EINVAL;
 381
 382         WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 383                         offset % sizeof(u64) != 0);
 384
 385         read_lock(&dev_base_lock);
 386         if (dev_isalive(dev)) {
 387                 struct rtnl_link_stats64 temp;
 388                 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 389
 390                 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 391         }
 392         read_unlock(&dev_base_lock);
 393         return ret;
 394 }
 395
 396 /* generate a read-only statistics attribute */
 397 #define NETSTAT_ENTRY(name)                                             \
 398 static ssize_t name##_show(struct device *d,                            \
 399                            struct device_attribute *attr, char *buf)    \
 400 {                                                                       \
 401         return netstat_show(d, attr, buf,                               \
 402                             offsetof(struct rtnl_link_stats64, name));  \
 403 }                                                                       \
 404 static DEVICE_ATTR_RO(name)
 405
 406 NETSTAT_ENTRY(rx_packets);
 407 NETSTAT_ENTRY(tx_packets);
 408 NETSTAT_ENTRY(rx_bytes);
 409 NETSTAT_ENTRY(tx_bytes);
 410 NETSTAT_ENTRY(rx_errors);
 411 NETSTAT_ENTRY(tx_errors);
 412 NETSTAT_ENTRY(rx_dropped);
 413 NETSTAT_ENTRY(tx_dropped);
 414 NETSTAT_ENTRY(multicast);
 415 NETSTAT_ENTRY(collisions);
 416 NETSTAT_ENTRY(rx_length_errors);
 417 NETSTAT_ENTRY(rx_over_errors);
 418 NETSTAT_ENTRY(rx_crc_errors);
 419 NETSTAT_ENTRY(rx_frame_errors);
 420 NETSTAT_ENTRY(rx_fifo_errors);
 421 NETSTAT_ENTRY(rx_missed_errors);
 422 NETSTAT_ENTRY(tx_aborted_errors);
 423 NETSTAT_ENTRY(tx_carrier_errors);
 424 NETSTAT_ENTRY(tx_fifo_errors);
 425 NETSTAT_ENTRY(tx_heartbeat_errors);
 426 NETSTAT_ENTRY(tx_window_errors);
 427 NETSTAT_ENTRY(rx_compressed);
 428 NETSTAT_ENTRY(tx_compressed);
 429
 430 static struct attribute *netstat_attrs[] = {
 431         &dev_attr_rx_packets.attr,
 432         &dev_attr_tx_packets.attr,
 433         &dev_attr_rx_bytes.attr,
 434         &dev_attr_tx_bytes.attr,
 435         &dev_attr_rx_errors.attr,
 436         &dev_attr_tx_errors.attr,
 437         &dev_attr_rx_dropped.attr,
 438         &dev_attr_tx_dropped.attr,
 439         &dev_attr_multicast.attr,
 440         &dev_attr_collisions.attr,
 441         &dev_attr_rx_length_errors.attr,
 442         &dev_attr_rx_over_errors.attr,
 443         &dev_attr_rx_crc_errors.attr,
 444         &dev_attr_rx_frame_errors.attr,
 445         &dev_attr_rx_fifo_errors.attr,
 446         &dev_attr_rx_missed_errors.attr,
 447         &dev_attr_tx_aborted_errors.attr,
 448         &dev_attr_tx_carrier_errors.attr,
 449         &dev_attr_tx_fifo_errors.attr,
 450         &dev_attr_tx_heartbeat_errors.attr,
 451         &dev_attr_tx_window_errors.attr,
 452         &dev_attr_rx_compressed.attr,
 453         &dev_attr_tx_compressed.attr,
 454         NULL
 455 };
 456
 457
 458 static struct attribute_group netstat_group = {
 459         .name  = "statistics",
 460         .attrs  = netstat_attrs,
 461 };
 462
 463 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 464 static struct attribute *wireless_attrs[] = {
 465         NULL
 466 };
 467
 468 static struct attribute_group wireless_group = {
 469         .name = "wireless",
 470         .attrs = wireless_attrs,
 471 };
 472 #endif
 473
 474 #else /* CONFIG_SYSFS */
 475 #define net_class_groups        NULL
 476 #endif /* CONFIG_SYSFS */
 477
 478 #ifdef CONFIG_RPS
 479 /*
 480  * RX queue sysfs structures and functions.
 481  */
 482 struct rx_queue_attribute {
 483         struct attribute attr;
 484         ssize_t (*show)(struct netdev_rx_queue *queue,
 485             struct rx_queue_attribute *attr, char *buf);
 486         ssize_t (*store)(struct netdev_rx_queue *queue,
 487             struct rx_queue_attribute *attr, const char *buf, size_t len);
 488 };
 489 #define to_rx_queue_attr(_attr) container_of(_attr,             \
 490     struct rx_queue_attribute, attr)
 491
 492 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 493
 494 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 495                                   char *buf)
 496 {
 497         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 498         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 499
 500         if (!attribute->show)
 501                 return -EIO;
 502
 503         return attribute->show(queue, attribute, buf);
 504 }
 505
 506 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 507                                    const char *buf, size_t count)
 508 {
 509         struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 510         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 511
 512         if (!attribute->store)
 513                 return -EIO;
 514
 515         return attribute->store(queue, attribute, buf, count);
 516 }
 517
 518 static const struct sysfs_ops rx_queue_sysfs_ops = {
 519         .show = rx_queue_attr_show,
 520         .store = rx_queue_attr_store,
 521 };
 522
 523 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 524                             struct rx_queue_attribute *attribute, char *buf)
 525 {
 526         struct rps_map *map;
 527         cpumask_var_t mask;
 528         size_t len = 0;
 529         int i;
 530
 531         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 532                 return -ENOMEM;
 533
 534         rcu_read_lock();
 535         map = rcu_dereference(queue->rps_map);
 536         if (map)
 537                 for (i = 0; i < map->len; i++)
 538                         cpumask_set_cpu(map->cpus[i], mask);
 539
 540         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 541         if (PAGE_SIZE - len < 3) {
 542                 rcu_read_unlock();
 543                 free_cpumask_var(mask);
 544                 return -EINVAL;
 545         }
 546         rcu_read_unlock();
 547
 548         free_cpumask_var(mask);
 549         len += sprintf(buf + len, "\n");
 550         return len;
 551 }
 552
 553 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 554                       struct rx_queue_attribute *attribute,
 555                       const char *buf, size_t len)
 556 {
 557         struct rps_map *old_map, *map;
 558         cpumask_var_t mask;
 559         int err, cpu, i;
 560         static DEFINE_SPINLOCK(rps_map_lock);
 561
 562         if (!capable(CAP_NET_ADMIN))
 563                 return -EPERM;
 564
 565         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 566                 return -ENOMEM;
 567
 568         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 569         if (err) {
 570                 free_cpumask_var(mask);
 571                 return err;
 572         }
 573
 574         map = kzalloc(max_t(unsigned int,
 575             RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 576             GFP_KERNEL);
 577         if (!map) {
 578                 free_cpumask_var(mask);
 579                 return -ENOMEM;
 580         }
 581
 582         i = 0;
 583         for_each_cpu_and(cpu, mask, cpu_online_mask)
 584                 map->cpus[i++] = cpu;
 585
 586         if (i)
 587                 map->len = i;
 588         else {
 589                 kfree(map);
 590                 map = NULL;
 591         }
 592
 593         spin_lock(&rps_map_lock);
 594         old_map = rcu_dereference_protected(queue->rps_map,
 595                                             lockdep_is_held(&rps_map_lock));
 596         rcu_assign_pointer(queue->rps_map, map);
 597         spin_unlock(&rps_map_lock);
 598
 599         if (map)
 600                 static_key_slow_inc(&rps_needed);
 601         if (old_map) {
 602                 kfree_rcu(old_map, rcu);
 603                 static_key_slow_dec(&rps_needed);
 604         }
 605         free_cpumask_var(mask);
 606         return len;
 607 }
 608
 609 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 610                                            struct rx_queue_attribute *attr,
 611                                            char *buf)
 612 {
 613         struct rps_dev_flow_table *flow_table;
 614         unsigned long val = 0;
 615
 616         rcu_read_lock();
 617         flow_table = rcu_dereference(queue->rps_flow_table);
 618         if (flow_table)
 619                 val = (unsigned long)flow_table->mask + 1;
 620         rcu_read_unlock();
 621
 622         return sprintf(buf, "%lu\n", val);
 623 }
 624
 625 static void rps_dev_flow_table_release(struct rcu_head *rcu)
 626 {
 627         struct rps_dev_flow_table *table = container_of(rcu,
 628             struct rps_dev_flow_table, rcu);
 629         vfree(table);
 630 }
 631
 632 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 633                                      struct rx_queue_attribute *attr,
 634                                      const char *buf, size_t len)
 635 {
 636         unsigned long mask, count;
 637         struct rps_dev_flow_table *table, *old_table;
 638         static DEFINE_SPINLOCK(rps_dev_flow_lock);
 639         int rc;
 640
 641         if (!capable(CAP_NET_ADMIN))
 642                 return -EPERM;
 643
 644         rc = kstrtoul(buf, 0, &count);
 645         if (rc < 0)
 646                 return rc;
 647
 648         if (count) {
 649                 mask = count - 1;
 650                 /* mask = roundup_pow_of_two(count) - 1;
 651                  * without overflows...
 652                  */
 653                 while ((mask | (mask >> 1)) != mask)
 654                         mask |= (mask >> 1);
 655                 /* On 64 bit arches, must check mask fits in table->mask (u32),
 656                  * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 657                  * doesnt overflow.
 658                  */
 659 #if BITS_PER_LONG > 32
 660                 if (mask > (unsigned long)(u32)mask)
 661                         return -EINVAL;
 662 #else
 663                 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 664                                 / sizeof(struct rps_dev_flow)) {
 665                         /* Enforce a limit to prevent overflow */
 666                         return -EINVAL;
 667                 }
 668 #endif
 669                 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 670                 if (!table)
 671                         return -ENOMEM;
 672
 673                 table->mask = mask;
 674                 for (count = 0; count <= mask; count++)
 675                         table->flows[count].cpu = RPS_NO_CPU;
 676         } else
 677                 table = NULL;
 678
 679         spin_lock(&rps_dev_flow_lock);
 680         old_table = rcu_dereference_protected(queue->rps_flow_table,
 681                                               lockdep_is_held(&rps_dev_flow_lock));
 682         rcu_assign_pointer(queue->rps_flow_table, table);
 683         spin_unlock(&rps_dev_flow_lock);
 684
 685         if (old_table)
 686                 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 687
 688         return len;
 689 }
 690
 691 static struct rx_queue_attribute rps_cpus_attribute =
 692         __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 693
 694
 695 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 696         __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 697             show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 698
 699 static struct attribute *rx_queue_default_attrs[] = {
 700         &rps_cpus_attribute.attr,
 701         &rps_dev_flow_table_cnt_attribute.attr,
 702         NULL
 703 };
 704
 705 static void rx_queue_release(struct kobject *kobj)
 706 {
 707         struct netdev_rx_queue *queue = to_rx_queue(kobj);
 708         struct rps_map *map;
 709         struct rps_dev_flow_table *flow_table;
 710
 711
 712         map = rcu_dereference_protected(queue->rps_map, 1);
 713         if (map) {
 714                 RCU_INIT_POINTER(queue->rps_map, NULL);
 715                 kfree_rcu(map, rcu);
 716         }
 717
 718         flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 719         if (flow_table) {
 720                 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 721                 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 722         }
 723
 724         memset(kobj, 0, sizeof(*kobj));
 725         dev_put(queue->dev);
 726 }
 727
 728 static struct kobj_type rx_queue_ktype = {
 729         .sysfs_ops = &rx_queue_sysfs_ops,
 730         .release = rx_queue_release,
 731         .default_attrs = rx_queue_default_attrs,
 732 };
 733
 734 static int rx_queue_add_kobject(struct net_device *net, int index)
 735 {
 736         struct netdev_rx_queue *queue = net->_rx + index;
 737         struct kobject *kobj = &queue->kobj;
 738         int error = 0;
 739
 740         kobj->kset = net->queues_kset;
 741         error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 742             "rx-%u", index);
 743         if (error) {
 744                 kobject_put(kobj);
 745                 return error;
 746         }
 747
 748         kobject_uevent(kobj, KOBJ_ADD);
 749         dev_hold(queue->dev);
 750
 751         return error;
 752 }
 753 #endif /* CONFIG_RPS */
 754
 755 int
 756 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 757 {
 758 #ifdef CONFIG_RPS
 759         int i;
 760         int error = 0;
 761
 762         for (i = old_num; i < new_num; i++) {
 763                 error = rx_queue_add_kobject(net, i);
 764                 if (error) {
 765                         new_num = old_num;
 766                         break;
 767                 }
 768         }
 769
 770         while (--i >= new_num)
 771                 kobject_put(&net->_rx[i].kobj);
 772
 773         return error;
 774 #else
 775         return 0;
 776 #endif
 777 }
 778
 779 #ifdef CONFIG_SYSFS
 780 /*
 781  * netdev_queue sysfs structures and functions.
 782  */
 783 struct netdev_queue_attribute {
 784         struct attribute attr;
 785         ssize_t (*show)(struct netdev_queue *queue,
 786             struct netdev_queue_attribute *attr, char *buf);
 787         ssize_t (*store)(struct netdev_queue *queue,
 788             struct netdev_queue_attribute *attr, const char *buf, size_t len);
 789 };
 790 #define to_netdev_queue_attr(_attr) container_of(_attr,         \
 791     struct netdev_queue_attribute, attr)
 792
 793 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 794
 795 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 796                                       struct attribute *attr, char *buf)
 797 {
 798         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 799         struct netdev_queue *queue = to_netdev_queue(kobj);
 800
 801         if (!attribute->show)
 802                 return -EIO;
 803
 804         return attribute->show(queue, attribute, buf);
 805 }
 806
 807 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 808                                        struct attribute *attr,
 809                                        const char *buf, size_t count)
 810 {
 811         struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 812         struct netdev_queue *queue = to_netdev_queue(kobj);
 813
 814         if (!attribute->store)
 815                 return -EIO;
 816
 817         return attribute->store(queue, attribute, buf, count);
 818 }
 819
 820 static const struct sysfs_ops netdev_queue_sysfs_ops = {
 821         .show = netdev_queue_attr_show,
 822         .store = netdev_queue_attr_store,
 823 };
 824
 825 static ssize_t show_trans_timeout(struct netdev_queue *queue,
 826                                   struct netdev_queue_attribute *attribute,
 827                                   char *buf)
 828 {
 829         unsigned long trans_timeout;
 830
 831         spin_lock_irq(&queue->_xmit_lock);
 832         trans_timeout = queue->trans_timeout;
 833         spin_unlock_irq(&queue->_xmit_lock);
 834
 835         return sprintf(buf, "%lu", trans_timeout);
 836 }
 837
 838 static struct netdev_queue_attribute queue_trans_timeout =
 839         __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 840
 841 #ifdef CONFIG_BQL
 842 /*
 843  * Byte queue limits sysfs structures and functions.
 844  */
 845 static ssize_t bql_show(char *buf, unsigned int value)
 846 {
 847         return sprintf(buf, "%u\n", value);
 848 }
 849
 850 static ssize_t bql_set(const char *buf, const size_t count,
 851                        unsigned int *pvalue)
 852 {
 853         unsigned int value;
 854         int err;
 855
 856         if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 857                 value = DQL_MAX_LIMIT;
 858         else {
 859                 err = kstrtouint(buf, 10, &value);
 860                 if (err < 0)
 861                         return err;
 862                 if (value > DQL_MAX_LIMIT)
 863                         return -EINVAL;
 864         }
 865
 866         *pvalue = value;
 867
 868         return count;
 869 }
 870
 871 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 872                                   struct netdev_queue_attribute *attr,
 873                                   char *buf)
 874 {
 875         struct dql *dql = &queue->dql;
 876
 877         return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 878 }
 879
 880 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 881                                  struct netdev_queue_attribute *attribute,
 882                                  const char *buf, size_t len)
 883 {
 884         struct dql *dql = &queue->dql;
 885         unsigned int value;
 886         int err;
 887
 888         err = kstrtouint(buf, 10, &value);
 889         if (err < 0)
 890                 return err;
 891
 892         dql->slack_hold_time = msecs_to_jiffies(value);
 893
 894         return len;
 895 }
 896
 897 static struct netdev_queue_attribute bql_hold_time_attribute =
 898         __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 899             bql_set_hold_time);
 900
 901 static ssize_t bql_show_inflight(struct netdev_queue *queue,
 902                                  struct netdev_queue_attribute *attr,
 903                                  char *buf)
 904 {
 905         struct dql *dql = &queue->dql;
 906
 907         return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 908 }
 909
 910 static struct netdev_queue_attribute bql_inflight_attribute =
 911         __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 912
 913 #define BQL_ATTR(NAME, FIELD)                                           \
 914 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
 915                                  struct netdev_queue_attribute *attr,   \
 916                                  char *buf)                             \
 917 {                                                                       \
 918         return bql_show(buf, queue->dql.FIELD);                         \
 919 }                                                                       \
 920                                                                         \
 921 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
 922                                 struct netdev_queue_attribute *attr,    \
 923                                 const char *buf, size_t len)            \
 924 {                                                                       \
 925         return bql_set(buf, len, &queue->dql.FIELD);                    \
 926 }                                                                       \
 927                                                                         \
 928 static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
 929         __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
 930             bql_set_ ## NAME);
 931
 932 BQL_ATTR(limit, limit)
 933 BQL_ATTR(limit_max, max_limit)
 934 BQL_ATTR(limit_min, min_limit)
 935
 936 static struct attribute *dql_attrs[] = {
 937         &bql_limit_attribute.attr,
 938         &bql_limit_max_attribute.attr,
 939         &bql_limit_min_attribute.attr,
 940         &bql_hold_time_attribute.attr,
 941         &bql_inflight_attribute.attr,
 942         NULL
 943 };
 944
 945 static struct attribute_group dql_group = {
 946         .name  = "byte_queue_limits",
 947         .attrs  = dql_attrs,
 948 };
 949 #endif /* CONFIG_BQL */
 950
 951 #ifdef CONFIG_XPS
 952 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 953 {
 954         struct net_device *dev = queue->dev;
 955         int i;
 956
 957         for (i = 0; i < dev->num_tx_queues; i++)
 958                 if (queue == &dev->_tx[i])
 959                         break;
 960
 961         BUG_ON(i >= dev->num_tx_queues);
 962
 963         return i;
 964 }
 965
 966
 967 static ssize_t show_xps_map(struct netdev_queue *queue,
 968                             struct netdev_queue_attribute *attribute, char *buf)
 969 {
 970         struct net_device *dev = queue->dev;
 971         struct xps_dev_maps *dev_maps;
 972         cpumask_var_t mask;
 973         unsigned long index;
 974         size_t len = 0;
 975         int i;
 976
 977         if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 978                 return -ENOMEM;
 979
 980         index = get_netdev_queue_index(queue);
 981
 982         rcu_read_lock();
 983         dev_maps = rcu_dereference(dev->xps_maps);
 984         if (dev_maps) {
 985                 for_each_possible_cpu(i) {
 986                         struct xps_map *map =
 987                             rcu_dereference(dev_maps->cpu_map[i]);
 988                         if (map) {
 989                                 int j;
 990                                 for (j = 0; j < map->len; j++) {
 991                                         if (map->queues[j] == index) {
 992                                                 cpumask_set_cpu(i, mask);
 993                                                 break;
 994                                         }
 995                                 }
 996                         }
 997                 }
 998         }
 999         rcu_read_unlock();
1000
1001         len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1002         if (PAGE_SIZE - len < 3) {
1003                 free_cpumask_var(mask);
1004                 return -EINVAL;
1005         }
1006
1007         free_cpumask_var(mask);
1008         len += sprintf(buf + len, "\n");
1009         return len;
1010 }
1011
1012 static ssize_t store_xps_map(struct netdev_queue *queue,
1013                       struct netdev_queue_attribute *attribute,
1014                       const char *buf, size_t len)
1015 {
1016         struct net_device *dev = queue->dev;
1017         unsigned long index;
1018         cpumask_var_t mask;
1019         int err;
1020
1021         if (!capable(CAP_NET_ADMIN))
1022                 return -EPERM;
1023
1024         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1025                 return -ENOMEM;
1026
1027         index = get_netdev_queue_index(queue);
1028
1029         err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1030         if (err) {
1031                 free_cpumask_var(mask);
1032                 return err;
1033         }
1034
1035         err = netif_set_xps_queue(dev, mask, index);
1036
1037         free_cpumask_var(mask);
1038
1039         return err ? : len;
1040 }
1041
1042 static struct netdev_queue_attribute xps_cpus_attribute =
1043     __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1044 #endif /* CONFIG_XPS */
1045
1046 static struct attribute *netdev_queue_default_attrs[] = {
1047         &queue_trans_timeout.attr,
1048 #ifdef CONFIG_XPS
1049         &xps_cpus_attribute.attr,
1050 #endif
1051         NULL
1052 };
1053
1054 static void netdev_queue_release(struct kobject *kobj)
1055 {
1056         struct netdev_queue *queue = to_netdev_queue(kobj);
1057
1058         memset(kobj, 0, sizeof(*kobj));
1059         dev_put(queue->dev);
1060 }
1061
1062 static struct kobj_type netdev_queue_ktype = {
1063         .sysfs_ops = &netdev_queue_sysfs_ops,
1064         .release = netdev_queue_release,
1065         .default_attrs = netdev_queue_default_attrs,
1066 };
1067
1068 static int netdev_queue_add_kobject(struct net_device *net, int index)
1069 {
1070         struct netdev_queue *queue = net->_tx + index;
1071         struct kobject *kobj = &queue->kobj;
1072         int error = 0;
1073
1074         kobj->kset = net->queues_kset;
1075         error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1076             "tx-%u", index);
1077         if (error)
1078                 goto exit;
1079
1080 #ifdef CONFIG_BQL
1081         error = sysfs_create_group(kobj, &dql_group);
1082         if (error)
1083                 goto exit;
1084 #endif
1085
1086         kobject_uevent(kobj, KOBJ_ADD);
1087         dev_hold(queue->dev);
1088
1089         return 0;
1090 exit:
1091         kobject_put(kobj);
1092         return error;
1093 }
1094 #endif /* CONFIG_SYSFS */
1095
1096 int
1097 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1098 {
1099 #ifdef CONFIG_SYSFS
1100         int i;
1101         int error = 0;
1102
1103         for (i = old_num; i < new_num; i++) {
1104                 error = netdev_queue_add_kobject(net, i);
1105                 if (error) {
1106                         new_num = old_num;
1107                         break;
1108                 }
1109         }
1110
1111         while (--i >= new_num) {
1112                 struct netdev_queue *queue = net->_tx + i;
1113
1114 #ifdef CONFIG_BQL
1115                 sysfs_remove_group(&queue->kobj, &dql_group);
1116 #endif
1117                 kobject_put(&queue->kobj);
1118         }
1119
1120         return error;
1121 #else
1122         return 0;
1123 #endif /* CONFIG_SYSFS */
1124 }
1125
1126 static int register_queue_kobjects(struct net_device *net)
1127 {
1128         int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1129
1130 #ifdef CONFIG_SYSFS
1131         net->queues_kset = kset_create_and_add("queues",
1132             NULL, &net->dev.kobj);
1133         if (!net->queues_kset)
1134                 return -ENOMEM;
1135 #endif
1136
1137 #ifdef CONFIG_RPS
1138         real_rx = net->real_num_rx_queues;
1139 #endif
1140         real_tx = net->real_num_tx_queues;
1141
1142         error = net_rx_queue_update_kobjects(net, 0, real_rx);
1143         if (error)
1144                 goto error;
1145         rxq = real_rx;
1146
1147         error = netdev_queue_update_kobjects(net, 0, real_tx);
1148         if (error)
1149                 goto error;
1150         txq = real_tx;
1151
1152         return 0;
1153
1154 error:
1155         netdev_queue_update_kobjects(net, txq, 0);
1156         net_rx_queue_update_kobjects(net, rxq, 0);
1157         return error;
1158 }
1159
1160 static void remove_queue_kobjects(struct net_device *net)
1161 {
1162         int real_rx = 0, real_tx = 0;
1163
1164 #ifdef CONFIG_RPS
1165         real_rx = net->real_num_rx_queues;
1166 #endif
1167         real_tx = net->real_num_tx_queues;
1168
1169         net_rx_queue_update_kobjects(net, real_rx, 0);
1170         netdev_queue_update_kobjects(net, real_tx, 0);
1171 #ifdef CONFIG_SYSFS
1172         kset_unregister(net->queues_kset);
1173 #endif
1174 }
1175
1176 static void *net_grab_current_ns(void)
1177 {
1178         struct net *ns = current->nsproxy->net_ns;
1179 #ifdef CONFIG_NET_NS
1180         if (ns)
1181                 atomic_inc(&ns->passive);
1182 #endif
1183         return ns;
1184 }
1185
1186 static const void *net_initial_ns(void)
1187 {
1188         return &init_net;
1189 }
1190
1191 static const void *net_netlink_ns(struct sock *sk)
1192 {
1193         return sock_net(sk);
1194 }
1195
1196 struct kobj_ns_type_operations net_ns_type_operations = {
1197         .type = KOBJ_NS_TYPE_NET,
1198         .grab_current_ns = net_grab_current_ns,
1199         .netlink_ns = net_netlink_ns,
1200         .initial_ns = net_initial_ns,
1201         .drop_ns = net_drop_ns,
1202 };
1203 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1204
1205 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1206 {
1207         struct net_device *dev = to_net_dev(d);
1208         int retval;
1209
1210         /* pass interface to uevent. */
1211         retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1212         if (retval)
1213                 goto exit;
1214
1215         /* pass ifindex to uevent.
1216          * ifindex is useful as it won't change (interface name may change)
1217          * and is what RtNetlink uses natively. */
1218         retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1219
1220 exit:
1221         return retval;
1222 }
1223
1224 /*
1225  *      netdev_release -- destroy and free a dead device.
1226  *      Called when last reference to device kobject is gone.
1227  */
1228 static void netdev_release(struct device *d)
1229 {
1230         struct net_device *dev = to_net_dev(d);
1231
1232         BUG_ON(dev->reg_state != NETREG_RELEASED);
1233
1234         kfree(dev->ifalias);
1235         kfree((char *)dev - dev->padded);
1236 }
1237
1238 static const void *net_namespace(struct device *d)
1239 {
1240         struct net_device *dev;
1241         dev = container_of(d, struct net_device, dev);
1242         return dev_net(dev);
1243 }
1244
1245 static struct class net_class = {
1246         .name = "net",
1247         .dev_release = netdev_release,
1248         .dev_groups = net_class_groups,
1249         .dev_uevent = netdev_uevent,
1250         .ns_type = &net_ns_type_operations,
1251         .namespace = net_namespace,
1252 };
1253
1254 /* Delete sysfs entries but hold kobject reference until after all
1255  * netdev references are gone.
1256  */
1257 void netdev_unregister_kobject(struct net_device * net)
1258 {
1259         struct device *dev = &(net->dev);
1260
1261         kobject_get(&dev->kobj);
1262
1263         remove_queue_kobjects(net);
1264
1265         pm_runtime_set_memalloc_noio(dev, false);
1266
1267         device_del(dev);
1268 }
1269
1270 /* Create sysfs entries for network device. */
1271 int netdev_register_kobject(struct net_device *net)
1272 {
1273         struct device *dev = &(net->dev);
1274         const struct attribute_group **groups = net->sysfs_groups;
1275         int error = 0;
1276
1277         device_initialize(dev);
1278         dev->class = &net_class;
1279         dev->platform_data = net;
1280         dev->groups = groups;
1281
1282         dev_set_name(dev, "%s", net->name);
1283
1284 #ifdef CONFIG_SYSFS
1285         /* Allow for a device specific group */
1286         if (*groups)
1287                 groups++;
1288
1289         *groups++ = &netstat_group;
1290
1291 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1292         if (net->ieee80211_ptr)
1293                 *groups++ = &wireless_group;
1294 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1295         else if (net->wireless_handlers)
1296                 *groups++ = &wireless_group;
1297 #endif
1298 #endif
1299 #endif /* CONFIG_SYSFS */
1300
1301         error = device_add(dev);
1302         if (error)
1303                 return error;
1304
1305         error = register_queue_kobjects(net);
1306         if (error) {
1307                 device_del(dev);
1308                 return error;
1309         }
1310
1311         pm_runtime_set_memalloc_noio(dev, true);
1312
1313         return error;
1314 }
1315
1316 int netdev_class_create_file(struct class_attribute *class_attr)
1317 {
1318         return class_create_file(&net_class, class_attr);
1319 }
1320 EXPORT_SYMBOL(netdev_class_create_file);
1321
1322 void netdev_class_remove_file(struct class_attribute *class_attr)
1323 {
1324         class_remove_file(&net_class, class_attr);
1325 }
1326 EXPORT_SYMBOL(netdev_class_remove_file);
1327
1328 int netdev_kobject_init(void)
1329 {
1330         kobj_ns_type_register(&net_ns_type_operations);
1331         return class_register(&net_class);
1332 }