linux/net/core/net-sysfs.c
<<
>>
Prefs
   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 <net/switchdev.h>
  16#include <linux/if_arp.h>
  17#include <linux/slab.h>
  18#include <linux/nsproxy.h>
  19#include <net/sock.h>
  20#include <net/net_namespace.h>
  21#include <linux/rtnetlink.h>
  22#include <linux/vmalloc.h>
  23#include <linux/export.h>
  24#include <linux/jiffies.h>
  25#include <linux/pm_runtime.h>
  26#include <linux/of.h>
  27
  28#include "net-sysfs.h"
  29
  30#ifdef CONFIG_SYSFS
  31static const char fmt_hex[] = "%#x\n";
  32static const char fmt_long_hex[] = "%#lx\n";
  33static const char fmt_dec[] = "%d\n";
  34static const char fmt_udec[] = "%u\n";
  35static const char fmt_ulong[] = "%lu\n";
  36static const char fmt_u64[] = "%llu\n";
  37
  38static inline int dev_isalive(const struct net_device *dev)
  39{
  40        return dev->reg_state <= NETREG_REGISTERED;
  41}
  42
  43/* use same locking rules as GIF* ioctl's */
  44static ssize_t netdev_show(const struct device *dev,
  45                           struct device_attribute *attr, char *buf,
  46                           ssize_t (*format)(const struct net_device *, char *))
  47{
  48        struct net_device *ndev = to_net_dev(dev);
  49        ssize_t ret = -EINVAL;
  50
  51        read_lock(&dev_base_lock);
  52        if (dev_isalive(ndev))
  53                ret = (*format)(ndev, buf);
  54        read_unlock(&dev_base_lock);
  55
  56        return ret;
  57}
  58
  59/* generate a show function for simple field */
  60#define NETDEVICE_SHOW(field, format_string)                            \
  61static ssize_t format_##field(const struct net_device *dev, char *buf)  \
  62{                                                                       \
  63        return sprintf(buf, format_string, dev->field);                 \
  64}                                                                       \
  65static ssize_t field##_show(struct device *dev,                         \
  66                            struct device_attribute *attr, char *buf)   \
  67{                                                                       \
  68        return netdev_show(dev, attr, buf, format_##field);             \
  69}                                                                       \
  70
  71#define NETDEVICE_SHOW_RO(field, format_string)                         \
  72NETDEVICE_SHOW(field, format_string);                                   \
  73static DEVICE_ATTR_RO(field)
  74
  75#define NETDEVICE_SHOW_RW(field, format_string)                         \
  76NETDEVICE_SHOW(field, format_string);                                   \
  77static DEVICE_ATTR_RW(field)
  78
  79/* use same locking and permission rules as SIF* ioctl's */
  80static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  81                            const char *buf, size_t len,
  82                            int (*set)(struct net_device *, unsigned long))
  83{
  84        struct net_device *netdev = to_net_dev(dev);
  85        struct net *net = dev_net(netdev);
  86        unsigned long new;
  87        int ret = -EINVAL;
  88
  89        if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  90                return -EPERM;
  91
  92        ret = kstrtoul(buf, 0, &new);
  93        if (ret)
  94                goto err;
  95
  96        if (!rtnl_trylock())
  97                return restart_syscall();
  98
  99        if (dev_isalive(netdev)) {
 100                if ((ret = (*set)(netdev, new)) == 0)
 101                        ret = len;
 102        }
 103        rtnl_unlock();
 104 err:
 105        return ret;
 106}
 107
 108NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 109NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 110NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 111NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 112NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 113NETDEVICE_SHOW_RO(type, fmt_dec);
 114NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 115
 116static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 117                           char *buf)
 118{
 119        struct net_device *ndev = to_net_dev(dev);
 120
 121        return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
 122}
 123static DEVICE_ATTR_RO(iflink);
 124
 125static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 126{
 127        return sprintf(buf, fmt_dec, dev->name_assign_type);
 128}
 129
 130static ssize_t name_assign_type_show(struct device *dev,
 131                                     struct device_attribute *attr,
 132                                     char *buf)
 133{
 134        struct net_device *ndev = to_net_dev(dev);
 135        ssize_t ret = -EINVAL;
 136
 137        if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 138                ret = netdev_show(dev, attr, buf, format_name_assign_type);
 139
 140        return ret;
 141}
 142static DEVICE_ATTR_RO(name_assign_type);
 143
 144/* use same locking rules as GIFHWADDR ioctl's */
 145static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 146                            char *buf)
 147{
 148        struct net_device *ndev = to_net_dev(dev);
 149        ssize_t ret = -EINVAL;
 150
 151        read_lock(&dev_base_lock);
 152        if (dev_isalive(ndev))
 153                ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 154        read_unlock(&dev_base_lock);
 155        return ret;
 156}
 157static DEVICE_ATTR_RO(address);
 158
 159static ssize_t broadcast_show(struct device *dev,
 160                              struct device_attribute *attr, char *buf)
 161{
 162        struct net_device *ndev = to_net_dev(dev);
 163        if (dev_isalive(ndev))
 164                return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 165        return -EINVAL;
 166}
 167static DEVICE_ATTR_RO(broadcast);
 168
 169static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 170{
 171        if (!netif_running(dev))
 172                return -EINVAL;
 173        return dev_change_carrier(dev, (bool) new_carrier);
 174}
 175
 176static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 177                             const char *buf, size_t len)
 178{
 179        return netdev_store(dev, attr, buf, len, change_carrier);
 180}
 181
 182static ssize_t carrier_show(struct device *dev,
 183                            struct device_attribute *attr, char *buf)
 184{
 185        struct net_device *netdev = to_net_dev(dev);
 186        if (netif_running(netdev)) {
 187                return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 188        }
 189        return -EINVAL;
 190}
 191static DEVICE_ATTR_RW(carrier);
 192
 193static ssize_t speed_show(struct device *dev,
 194                          struct device_attribute *attr, char *buf)
 195{
 196        struct net_device *netdev = to_net_dev(dev);
 197        int ret = -EINVAL;
 198
 199        if (!rtnl_trylock())
 200                return restart_syscall();
 201
 202        if (netif_running(netdev)) {
 203                struct ethtool_cmd cmd;
 204                if (!__ethtool_get_settings(netdev, &cmd))
 205                        ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 206        }
 207        rtnl_unlock();
 208        return ret;
 209}
 210static DEVICE_ATTR_RO(speed);
 211
 212static ssize_t duplex_show(struct device *dev,
 213                           struct device_attribute *attr, char *buf)
 214{
 215        struct net_device *netdev = to_net_dev(dev);
 216        int ret = -EINVAL;
 217
 218        if (!rtnl_trylock())
 219                return restart_syscall();
 220
 221        if (netif_running(netdev)) {
 222                struct ethtool_cmd cmd;
 223                if (!__ethtool_get_settings(netdev, &cmd)) {
 224                        const char *duplex;
 225                        switch (cmd.duplex) {
 226                        case DUPLEX_HALF:
 227                                duplex = "half";
 228                                break;
 229                        case DUPLEX_FULL:
 230                                duplex = "full";
 231                                break;
 232                        default:
 233                                duplex = "unknown";
 234                                break;
 235                        }
 236                        ret = sprintf(buf, "%s\n", duplex);
 237                }
 238        }
 239        rtnl_unlock();
 240        return ret;
 241}
 242static DEVICE_ATTR_RO(duplex);
 243
 244static ssize_t dormant_show(struct device *dev,
 245                            struct device_attribute *attr, char *buf)
 246{
 247        struct net_device *netdev = to_net_dev(dev);
 248
 249        if (netif_running(netdev))
 250                return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 251
 252        return -EINVAL;
 253}
 254static DEVICE_ATTR_RO(dormant);
 255
 256static const char *const operstates[] = {
 257        "unknown",
 258        "notpresent", /* currently unused */
 259        "down",
 260        "lowerlayerdown",
 261        "testing", /* currently unused */
 262        "dormant",
 263        "up"
 264};
 265
 266static ssize_t operstate_show(struct device *dev,
 267                              struct device_attribute *attr, char *buf)
 268{
 269        const struct net_device *netdev = to_net_dev(dev);
 270        unsigned char operstate;
 271
 272        read_lock(&dev_base_lock);
 273        operstate = netdev->operstate;
 274        if (!netif_running(netdev))
 275                operstate = IF_OPER_DOWN;
 276        read_unlock(&dev_base_lock);
 277
 278        if (operstate >= ARRAY_SIZE(operstates))
 279                return -EINVAL; /* should not happen */
 280
 281        return sprintf(buf, "%s\n", operstates[operstate]);
 282}
 283static DEVICE_ATTR_RO(operstate);
 284
 285static ssize_t carrier_changes_show(struct device *dev,
 286                                    struct device_attribute *attr,
 287                                    char *buf)
 288{
 289        struct net_device *netdev = to_net_dev(dev);
 290        return sprintf(buf, fmt_dec,
 291                       atomic_read(&netdev->carrier_changes));
 292}
 293static DEVICE_ATTR_RO(carrier_changes);
 294
 295/* read-write attributes */
 296
 297static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 298{
 299        return dev_set_mtu(dev, (int) new_mtu);
 300}
 301
 302static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 303                         const char *buf, size_t len)
 304{
 305        return netdev_store(dev, attr, buf, len, change_mtu);
 306}
 307NETDEVICE_SHOW_RW(mtu, fmt_dec);
 308
 309static int change_flags(struct net_device *dev, unsigned long new_flags)
 310{
 311        return dev_change_flags(dev, (unsigned int) new_flags);
 312}
 313
 314static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 315                           const char *buf, size_t len)
 316{
 317        return netdev_store(dev, attr, buf, len, change_flags);
 318}
 319NETDEVICE_SHOW_RW(flags, fmt_hex);
 320
 321static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 322{
 323        dev->tx_queue_len = new_len;
 324        return 0;
 325}
 326
 327static ssize_t tx_queue_len_store(struct device *dev,
 328                                  struct device_attribute *attr,
 329                                  const char *buf, size_t len)
 330{
 331        if (!capable(CAP_NET_ADMIN))
 332                return -EPERM;
 333
 334        return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 335}
 336NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
 337
 338static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 339{
 340        dev->gro_flush_timeout = val;
 341        return 0;
 342}
 343
 344static ssize_t gro_flush_timeout_store(struct device *dev,
 345                                  struct device_attribute *attr,
 346                                  const char *buf, size_t len)
 347{
 348        if (!capable(CAP_NET_ADMIN))
 349                return -EPERM;
 350
 351        return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 352}
 353NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 354
 355static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 356                             const char *buf, size_t len)
 357{
 358        struct net_device *netdev = to_net_dev(dev);
 359        struct net *net = dev_net(netdev);
 360        size_t count = len;
 361        ssize_t ret;
 362
 363        if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 364                return -EPERM;
 365
 366        /* ignore trailing newline */
 367        if (len >  0 && buf[len - 1] == '\n')
 368                --count;
 369
 370        if (!rtnl_trylock())
 371                return restart_syscall();
 372        ret = dev_set_alias(netdev, buf, count);
 373        rtnl_unlock();
 374
 375        return ret < 0 ? ret : len;
 376}
 377
 378static ssize_t ifalias_show(struct device *dev,
 379                            struct device_attribute *attr, char *buf)
 380{
 381        const struct net_device *netdev = to_net_dev(dev);
 382        ssize_t ret = 0;
 383
 384        if (!rtnl_trylock())
 385                return restart_syscall();
 386        if (netdev->ifalias)
 387                ret = sprintf(buf, "%s\n", netdev->ifalias);
 388        rtnl_unlock();
 389        return ret;
 390}
 391static DEVICE_ATTR_RW(ifalias);
 392
 393static int change_group(struct net_device *dev, unsigned long new_group)
 394{
 395        dev_set_group(dev, (int) new_group);
 396        return 0;
 397}
 398
 399static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 400                           const char *buf, size_t len)
 401{
 402        return netdev_store(dev, attr, buf, len, change_group);
 403}
 404NETDEVICE_SHOW(group, fmt_dec);
 405static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
 406
 407static ssize_t phys_port_id_show(struct device *dev,
 408                                 struct device_attribute *attr, char *buf)
 409{
 410        struct net_device *netdev = to_net_dev(dev);
 411        ssize_t ret = -EINVAL;
 412
 413        if (!rtnl_trylock())
 414                return restart_syscall();
 415
 416        if (dev_isalive(netdev)) {
 417                struct netdev_phys_item_id ppid;
 418
 419                ret = dev_get_phys_port_id(netdev, &ppid);
 420                if (!ret)
 421                        ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 422        }
 423        rtnl_unlock();
 424
 425        return ret;
 426}
 427static DEVICE_ATTR_RO(phys_port_id);
 428
 429static ssize_t phys_port_name_show(struct device *dev,
 430                                   struct device_attribute *attr, char *buf)
 431{
 432        struct net_device *netdev = to_net_dev(dev);
 433        ssize_t ret = -EINVAL;
 434
 435        if (!rtnl_trylock())
 436                return restart_syscall();
 437
 438        if (dev_isalive(netdev)) {
 439                char name[IFNAMSIZ];
 440
 441                ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 442                if (!ret)
 443                        ret = sprintf(buf, "%s\n", name);
 444        }
 445        rtnl_unlock();
 446
 447        return ret;
 448}
 449static DEVICE_ATTR_RO(phys_port_name);
 450
 451static ssize_t phys_switch_id_show(struct device *dev,
 452                                   struct device_attribute *attr, char *buf)
 453{
 454        struct net_device *netdev = to_net_dev(dev);
 455        ssize_t ret = -EINVAL;
 456
 457        if (!rtnl_trylock())
 458                return restart_syscall();
 459
 460        if (dev_isalive(netdev)) {
 461                struct netdev_phys_item_id ppid;
 462
 463                ret = netdev_switch_parent_id_get(netdev, &ppid);
 464                if (!ret)
 465                        ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
 466        }
 467        rtnl_unlock();
 468
 469        return ret;
 470}
 471static DEVICE_ATTR_RO(phys_switch_id);
 472
 473static struct attribute *net_class_attrs[] = {
 474        &dev_attr_netdev_group.attr,
 475        &dev_attr_type.attr,
 476        &dev_attr_dev_id.attr,
 477        &dev_attr_dev_port.attr,
 478        &dev_attr_iflink.attr,
 479        &dev_attr_ifindex.attr,
 480        &dev_attr_name_assign_type.attr,
 481        &dev_attr_addr_assign_type.attr,
 482        &dev_attr_addr_len.attr,
 483        &dev_attr_link_mode.attr,
 484        &dev_attr_address.attr,
 485        &dev_attr_broadcast.attr,
 486        &dev_attr_speed.attr,
 487        &dev_attr_duplex.attr,
 488        &dev_attr_dormant.attr,
 489        &dev_attr_operstate.attr,
 490        &dev_attr_carrier_changes.attr,
 491        &dev_attr_ifalias.attr,
 492        &dev_attr_carrier.attr,
 493        &dev_attr_mtu.attr,
 494        &dev_attr_flags.attr,
 495        &dev_attr_tx_queue_len.attr,
 496        &dev_attr_gro_flush_timeout.attr,
 497        &dev_attr_phys_port_id.attr,
 498        &dev_attr_phys_port_name.attr,
 499        &dev_attr_phys_switch_id.attr,
 500        NULL,
 501};
 502ATTRIBUTE_GROUPS(net_class);
 503
 504/* Show a given an attribute in the statistics group */
 505static ssize_t netstat_show(const struct device *d,
 506                            struct device_attribute *attr, char *buf,
 507                            unsigned long offset)
 508{
 509        struct net_device *dev = to_net_dev(d);
 510        ssize_t ret = -EINVAL;
 511
 512        WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 513                        offset % sizeof(u64) != 0);
 514
 515        read_lock(&dev_base_lock);
 516        if (dev_isalive(dev)) {
 517                struct rtnl_link_stats64 temp;
 518                const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 519
 520                ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 521        }
 522        read_unlock(&dev_base_lock);
 523        return ret;
 524}
 525
 526/* generate a read-only statistics attribute */
 527#define NETSTAT_ENTRY(name)                                             \
 528static ssize_t name##_show(struct device *d,                            \
 529                           struct device_attribute *attr, char *buf)    \
 530{                                                                       \
 531        return netstat_show(d, attr, buf,                               \
 532                            offsetof(struct rtnl_link_stats64, name));  \
 533}                                                                       \
 534static DEVICE_ATTR_RO(name)
 535
 536NETSTAT_ENTRY(rx_packets);
 537NETSTAT_ENTRY(tx_packets);
 538NETSTAT_ENTRY(rx_bytes);
 539NETSTAT_ENTRY(tx_bytes);
 540NETSTAT_ENTRY(rx_errors);
 541NETSTAT_ENTRY(tx_errors);
 542NETSTAT_ENTRY(rx_dropped);
 543NETSTAT_ENTRY(tx_dropped);
 544NETSTAT_ENTRY(multicast);
 545NETSTAT_ENTRY(collisions);
 546NETSTAT_ENTRY(rx_length_errors);
 547NETSTAT_ENTRY(rx_over_errors);
 548NETSTAT_ENTRY(rx_crc_errors);
 549NETSTAT_ENTRY(rx_frame_errors);
 550NETSTAT_ENTRY(rx_fifo_errors);
 551NETSTAT_ENTRY(rx_missed_errors);
 552NETSTAT_ENTRY(tx_aborted_errors);
 553NETSTAT_ENTRY(tx_carrier_errors);
 554NETSTAT_ENTRY(tx_fifo_errors);
 555NETSTAT_ENTRY(tx_heartbeat_errors);
 556NETSTAT_ENTRY(tx_window_errors);
 557NETSTAT_ENTRY(rx_compressed);
 558NETSTAT_ENTRY(tx_compressed);
 559
 560static struct attribute *netstat_attrs[] = {
 561        &dev_attr_rx_packets.attr,
 562        &dev_attr_tx_packets.attr,
 563        &dev_attr_rx_bytes.attr,
 564        &dev_attr_tx_bytes.attr,
 565        &dev_attr_rx_errors.attr,
 566        &dev_attr_tx_errors.attr,
 567        &dev_attr_rx_dropped.attr,
 568        &dev_attr_tx_dropped.attr,
 569        &dev_attr_multicast.attr,
 570        &dev_attr_collisions.attr,
 571        &dev_attr_rx_length_errors.attr,
 572        &dev_attr_rx_over_errors.attr,
 573        &dev_attr_rx_crc_errors.attr,
 574        &dev_attr_rx_frame_errors.attr,
 575        &dev_attr_rx_fifo_errors.attr,
 576        &dev_attr_rx_missed_errors.attr,
 577        &dev_attr_tx_aborted_errors.attr,
 578        &dev_attr_tx_carrier_errors.attr,
 579        &dev_attr_tx_fifo_errors.attr,
 580        &dev_attr_tx_heartbeat_errors.attr,
 581        &dev_attr_tx_window_errors.attr,
 582        &dev_attr_rx_compressed.attr,
 583        &dev_attr_tx_compressed.attr,
 584        NULL
 585};
 586
 587
 588static struct attribute_group netstat_group = {
 589        .name  = "statistics",
 590        .attrs  = netstat_attrs,
 591};
 592
 593#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
 594static struct attribute *wireless_attrs[] = {
 595        NULL
 596};
 597
 598static struct attribute_group wireless_group = {
 599        .name = "wireless",
 600        .attrs = wireless_attrs,
 601};
 602#endif
 603
 604#else /* CONFIG_SYSFS */
 605#define net_class_groups        NULL
 606#endif /* CONFIG_SYSFS */
 607
 608#ifdef CONFIG_SYSFS
 609#define to_rx_queue_attr(_attr) container_of(_attr,             \
 610    struct rx_queue_attribute, attr)
 611
 612#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 613
 614static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 615                                  char *buf)
 616{
 617        struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 618        struct netdev_rx_queue *queue = to_rx_queue(kobj);
 619
 620        if (!attribute->show)
 621                return -EIO;
 622
 623        return attribute->show(queue, attribute, buf);
 624}
 625
 626static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 627                                   const char *buf, size_t count)
 628{
 629        struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 630        struct netdev_rx_queue *queue = to_rx_queue(kobj);
 631
 632        if (!attribute->store)
 633                return -EIO;
 634
 635        return attribute->store(queue, attribute, buf, count);
 636}
 637
 638static const struct sysfs_ops rx_queue_sysfs_ops = {
 639        .show = rx_queue_attr_show,
 640        .store = rx_queue_attr_store,
 641};
 642
 643#ifdef CONFIG_RPS
 644static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 645                            struct rx_queue_attribute *attribute, char *buf)
 646{
 647        struct rps_map *map;
 648        cpumask_var_t mask;
 649        int i, len;
 650
 651        if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 652                return -ENOMEM;
 653
 654        rcu_read_lock();
 655        map = rcu_dereference(queue->rps_map);
 656        if (map)
 657                for (i = 0; i < map->len; i++)
 658                        cpumask_set_cpu(map->cpus[i], mask);
 659
 660        len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
 661        rcu_read_unlock();
 662        free_cpumask_var(mask);
 663
 664        return len < PAGE_SIZE ? len : -EINVAL;
 665}
 666
 667static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 668                      struct rx_queue_attribute *attribute,
 669                      const char *buf, size_t len)
 670{
 671        struct rps_map *old_map, *map;
 672        cpumask_var_t mask;
 673        int err, cpu, i;
 674        static DEFINE_SPINLOCK(rps_map_lock);
 675
 676        if (!capable(CAP_NET_ADMIN))
 677                return -EPERM;
 678
 679        if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 680                return -ENOMEM;
 681
 682        err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 683        if (err) {
 684                free_cpumask_var(mask);
 685                return err;
 686        }
 687
 688        map = kzalloc(max_t(unsigned int,
 689            RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 690            GFP_KERNEL);
 691        if (!map) {
 692                free_cpumask_var(mask);
 693                return -ENOMEM;
 694        }
 695
 696        i = 0;
 697        for_each_cpu_and(cpu, mask, cpu_online_mask)
 698                map->cpus[i++] = cpu;
 699
 700        if (i)
 701                map->len = i;
 702        else {
 703                kfree(map);
 704                map = NULL;
 705        }
 706
 707        spin_lock(&rps_map_lock);
 708        old_map = rcu_dereference_protected(queue->rps_map,
 709                                            lockdep_is_held(&rps_map_lock));
 710        rcu_assign_pointer(queue->rps_map, map);
 711        spin_unlock(&rps_map_lock);
 712
 713        if (map)
 714                static_key_slow_inc(&rps_needed);
 715        if (old_map) {
 716                kfree_rcu(old_map, rcu);
 717                static_key_slow_dec(&rps_needed);
 718        }
 719        free_cpumask_var(mask);
 720        return len;
 721}
 722
 723static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 724                                           struct rx_queue_attribute *attr,
 725                                           char *buf)
 726{
 727        struct rps_dev_flow_table *flow_table;
 728        unsigned long val = 0;
 729
 730        rcu_read_lock();
 731        flow_table = rcu_dereference(queue->rps_flow_table);
 732        if (flow_table)
 733                val = (unsigned long)flow_table->mask + 1;
 734        rcu_read_unlock();
 735
 736        return sprintf(buf, "%lu\n", val);
 737}
 738
 739static void rps_dev_flow_table_release(struct rcu_head *rcu)
 740{
 741        struct rps_dev_flow_table *table = container_of(rcu,
 742            struct rps_dev_flow_table, rcu);
 743        vfree(table);
 744}
 745
 746static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 747                                     struct rx_queue_attribute *attr,
 748                                     const char *buf, size_t len)
 749{
 750        unsigned long mask, count;
 751        struct rps_dev_flow_table *table, *old_table;
 752        static DEFINE_SPINLOCK(rps_dev_flow_lock);
 753        int rc;
 754
 755        if (!capable(CAP_NET_ADMIN))
 756                return -EPERM;
 757
 758        rc = kstrtoul(buf, 0, &count);
 759        if (rc < 0)
 760                return rc;
 761
 762        if (count) {
 763                mask = count - 1;
 764                /* mask = roundup_pow_of_two(count) - 1;
 765                 * without overflows...
 766                 */
 767                while ((mask | (mask >> 1)) != mask)
 768                        mask |= (mask >> 1);
 769                /* On 64 bit arches, must check mask fits in table->mask (u32),
 770                 * and on 32bit arches, must check
 771                 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 772                 */
 773#if BITS_PER_LONG > 32
 774                if (mask > (unsigned long)(u32)mask)
 775                        return -EINVAL;
 776#else
 777                if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 778                                / sizeof(struct rps_dev_flow)) {
 779                        /* Enforce a limit to prevent overflow */
 780                        return -EINVAL;
 781                }
 782#endif
 783                table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 784                if (!table)
 785                        return -ENOMEM;
 786
 787                table->mask = mask;
 788                for (count = 0; count <= mask; count++)
 789                        table->flows[count].cpu = RPS_NO_CPU;
 790        } else
 791                table = NULL;
 792
 793        spin_lock(&rps_dev_flow_lock);
 794        old_table = rcu_dereference_protected(queue->rps_flow_table,
 795                                              lockdep_is_held(&rps_dev_flow_lock));
 796        rcu_assign_pointer(queue->rps_flow_table, table);
 797        spin_unlock(&rps_dev_flow_lock);
 798
 799        if (old_table)
 800                call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 801
 802        return len;
 803}
 804
 805static struct rx_queue_attribute rps_cpus_attribute =
 806        __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 807
 808
 809static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 810        __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 811            show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 812#endif /* CONFIG_RPS */
 813
 814static struct attribute *rx_queue_default_attrs[] = {
 815#ifdef CONFIG_RPS
 816        &rps_cpus_attribute.attr,
 817        &rps_dev_flow_table_cnt_attribute.attr,
 818#endif
 819        NULL
 820};
 821
 822static void rx_queue_release(struct kobject *kobj)
 823{
 824        struct netdev_rx_queue *queue = to_rx_queue(kobj);
 825#ifdef CONFIG_RPS
 826        struct rps_map *map;
 827        struct rps_dev_flow_table *flow_table;
 828
 829
 830        map = rcu_dereference_protected(queue->rps_map, 1);
 831        if (map) {
 832                RCU_INIT_POINTER(queue->rps_map, NULL);
 833                kfree_rcu(map, rcu);
 834        }
 835
 836        flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 837        if (flow_table) {
 838                RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 839                call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 840        }
 841#endif
 842
 843        memset(kobj, 0, sizeof(*kobj));
 844        dev_put(queue->dev);
 845}
 846
 847static const void *rx_queue_namespace(struct kobject *kobj)
 848{
 849        struct netdev_rx_queue *queue = to_rx_queue(kobj);
 850        struct device *dev = &queue->dev->dev;
 851        const void *ns = NULL;
 852
 853        if (dev->class && dev->class->ns_type)
 854                ns = dev->class->namespace(dev);
 855
 856        return ns;
 857}
 858
 859static struct kobj_type rx_queue_ktype = {
 860        .sysfs_ops = &rx_queue_sysfs_ops,
 861        .release = rx_queue_release,
 862        .default_attrs = rx_queue_default_attrs,
 863        .namespace = rx_queue_namespace
 864};
 865
 866static int rx_queue_add_kobject(struct net_device *dev, int index)
 867{
 868        struct netdev_rx_queue *queue = dev->_rx + index;
 869        struct kobject *kobj = &queue->kobj;
 870        int error = 0;
 871
 872        kobj->kset = dev->queues_kset;
 873        error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 874            "rx-%u", index);
 875        if (error)
 876                goto exit;
 877
 878        if (dev->sysfs_rx_queue_group) {
 879                error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
 880                if (error)
 881                        goto exit;
 882        }
 883
 884        kobject_uevent(kobj, KOBJ_ADD);
 885        dev_hold(queue->dev);
 886
 887        return error;
 888exit:
 889        kobject_put(kobj);
 890        return error;
 891}
 892#endif /* CONFIG_SYSFS */
 893
 894int
 895net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
 896{
 897#ifdef CONFIG_SYSFS
 898        int i;
 899        int error = 0;
 900
 901#ifndef CONFIG_RPS
 902        if (!dev->sysfs_rx_queue_group)
 903                return 0;
 904#endif
 905        for (i = old_num; i < new_num; i++) {
 906                error = rx_queue_add_kobject(dev, i);
 907                if (error) {
 908                        new_num = old_num;
 909                        break;
 910                }
 911        }
 912
 913        while (--i >= new_num) {
 914                if (dev->sysfs_rx_queue_group)
 915                        sysfs_remove_group(&dev->_rx[i].kobj,
 916                                           dev->sysfs_rx_queue_group);
 917                kobject_put(&dev->_rx[i].kobj);
 918        }
 919
 920        return error;
 921#else
 922        return 0;
 923#endif
 924}
 925
 926#ifdef CONFIG_SYSFS
 927/*
 928 * netdev_queue sysfs structures and functions.
 929 */
 930struct netdev_queue_attribute {
 931        struct attribute attr;
 932        ssize_t (*show)(struct netdev_queue *queue,
 933            struct netdev_queue_attribute *attr, char *buf);
 934        ssize_t (*store)(struct netdev_queue *queue,
 935            struct netdev_queue_attribute *attr, const char *buf, size_t len);
 936};
 937#define to_netdev_queue_attr(_attr) container_of(_attr,         \
 938    struct netdev_queue_attribute, attr)
 939
 940#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 941
 942static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 943                                      struct attribute *attr, char *buf)
 944{
 945        struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 946        struct netdev_queue *queue = to_netdev_queue(kobj);
 947
 948        if (!attribute->show)
 949                return -EIO;
 950
 951        return attribute->show(queue, attribute, buf);
 952}
 953
 954static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 955                                       struct attribute *attr,
 956                                       const char *buf, size_t count)
 957{
 958        struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 959        struct netdev_queue *queue = to_netdev_queue(kobj);
 960
 961        if (!attribute->store)
 962                return -EIO;
 963
 964        return attribute->store(queue, attribute, buf, count);
 965}
 966
 967static const struct sysfs_ops netdev_queue_sysfs_ops = {
 968        .show = netdev_queue_attr_show,
 969        .store = netdev_queue_attr_store,
 970};
 971
 972static ssize_t show_trans_timeout(struct netdev_queue *queue,
 973                                  struct netdev_queue_attribute *attribute,
 974                                  char *buf)
 975{
 976        unsigned long trans_timeout;
 977
 978        spin_lock_irq(&queue->_xmit_lock);
 979        trans_timeout = queue->trans_timeout;
 980        spin_unlock_irq(&queue->_xmit_lock);
 981
 982        return sprintf(buf, "%lu", trans_timeout);
 983}
 984
 985#ifdef CONFIG_XPS
 986static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 987{
 988        struct net_device *dev = queue->dev;
 989        int i;
 990
 991        for (i = 0; i < dev->num_tx_queues; i++)
 992                if (queue == &dev->_tx[i])
 993                        break;
 994
 995        BUG_ON(i >= dev->num_tx_queues);
 996
 997        return i;
 998}
 999
1000static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1001                               struct netdev_queue_attribute *attribute,
1002                               char *buf)
1003{
1004        return sprintf(buf, "%lu\n", queue->tx_maxrate);
1005}
1006
1007static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1008                              struct netdev_queue_attribute *attribute,
1009                              const char *buf, size_t len)
1010{
1011        struct net_device *dev = queue->dev;
1012        int err, index = get_netdev_queue_index(queue);
1013        u32 rate = 0;
1014
1015        err = kstrtou32(buf, 10, &rate);
1016        if (err < 0)
1017                return err;
1018
1019        if (!rtnl_trylock())
1020                return restart_syscall();
1021
1022        err = -EOPNOTSUPP;
1023        if (dev->netdev_ops->ndo_set_tx_maxrate)
1024                err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1025
1026        rtnl_unlock();
1027        if (!err) {
1028                queue->tx_maxrate = rate;
1029                return len;
1030        }
1031        return err;
1032}
1033
1034static struct netdev_queue_attribute queue_tx_maxrate =
1035        __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1036               show_tx_maxrate, set_tx_maxrate);
1037#endif
1038
1039static struct netdev_queue_attribute queue_trans_timeout =
1040        __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1041
1042#ifdef CONFIG_BQL
1043/*
1044 * Byte queue limits sysfs structures and functions.
1045 */
1046static ssize_t bql_show(char *buf, unsigned int value)
1047{
1048        return sprintf(buf, "%u\n", value);
1049}
1050
1051static ssize_t bql_set(const char *buf, const size_t count,
1052                       unsigned int *pvalue)
1053{
1054        unsigned int value;
1055        int err;
1056
1057        if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1058                value = DQL_MAX_LIMIT;
1059        else {
1060                err = kstrtouint(buf, 10, &value);
1061                if (err < 0)
1062                        return err;
1063                if (value > DQL_MAX_LIMIT)
1064                        return -EINVAL;
1065        }
1066
1067        *pvalue = value;
1068
1069        return count;
1070}
1071
1072static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1073                                  struct netdev_queue_attribute *attr,
1074                                  char *buf)
1075{
1076        struct dql *dql = &queue->dql;
1077
1078        return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1079}
1080
1081static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1082                                 struct netdev_queue_attribute *attribute,
1083                                 const char *buf, size_t len)
1084{
1085        struct dql *dql = &queue->dql;
1086        unsigned int value;
1087        int err;
1088
1089        err = kstrtouint(buf, 10, &value);
1090        if (err < 0)
1091                return err;
1092
1093        dql->slack_hold_time = msecs_to_jiffies(value);
1094
1095        return len;
1096}
1097
1098static struct netdev_queue_attribute bql_hold_time_attribute =
1099        __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1100            bql_set_hold_time);
1101
1102static ssize_t bql_show_inflight(struct netdev_queue *queue,
1103                                 struct netdev_queue_attribute *attr,
1104                                 char *buf)
1105{
1106        struct dql *dql = &queue->dql;
1107
1108        return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1109}
1110
1111static struct netdev_queue_attribute bql_inflight_attribute =
1112        __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1113
1114#define BQL_ATTR(NAME, FIELD)                                           \
1115static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,            \
1116                                 struct netdev_queue_attribute *attr,   \
1117                                 char *buf)                             \
1118{                                                                       \
1119        return bql_show(buf, queue->dql.FIELD);                         \
1120}                                                                       \
1121                                                                        \
1122static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,             \
1123                                struct netdev_queue_attribute *attr,    \
1124                                const char *buf, size_t len)            \
1125{                                                                       \
1126        return bql_set(buf, len, &queue->dql.FIELD);                    \
1127}                                                                       \
1128                                                                        \
1129static struct netdev_queue_attribute bql_ ## NAME ## _attribute =       \
1130        __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,              \
1131            bql_set_ ## NAME);
1132
1133BQL_ATTR(limit, limit)
1134BQL_ATTR(limit_max, max_limit)
1135BQL_ATTR(limit_min, min_limit)
1136
1137static struct attribute *dql_attrs[] = {
1138        &bql_limit_attribute.attr,
1139        &bql_limit_max_attribute.attr,
1140        &bql_limit_min_attribute.attr,
1141        &bql_hold_time_attribute.attr,
1142        &bql_inflight_attribute.attr,
1143        NULL
1144};
1145
1146static struct attribute_group dql_group = {
1147        .name  = "byte_queue_limits",
1148        .attrs  = dql_attrs,
1149};
1150#endif /* CONFIG_BQL */
1151
1152#ifdef CONFIG_XPS
1153static ssize_t show_xps_map(struct netdev_queue *queue,
1154                            struct netdev_queue_attribute *attribute, char *buf)
1155{
1156        struct net_device *dev = queue->dev;
1157        struct xps_dev_maps *dev_maps;
1158        cpumask_var_t mask;
1159        unsigned long index;
1160        int i, len;
1161
1162        if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1163                return -ENOMEM;
1164
1165        index = get_netdev_queue_index(queue);
1166
1167        rcu_read_lock();
1168        dev_maps = rcu_dereference(dev->xps_maps);
1169        if (dev_maps) {
1170                for_each_possible_cpu(i) {
1171                        struct xps_map *map =
1172                            rcu_dereference(dev_maps->cpu_map[i]);
1173                        if (map) {
1174                                int j;
1175                                for (j = 0; j < map->len; j++) {
1176                                        if (map->queues[j] == index) {
1177                                                cpumask_set_cpu(i, mask);
1178                                                break;
1179                                        }
1180                                }
1181                        }
1182                }
1183        }
1184        rcu_read_unlock();
1185
1186        len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1187        free_cpumask_var(mask);
1188        return len < PAGE_SIZE ? len : -EINVAL;
1189}
1190
1191static ssize_t store_xps_map(struct netdev_queue *queue,
1192                      struct netdev_queue_attribute *attribute,
1193                      const char *buf, size_t len)
1194{
1195        struct net_device *dev = queue->dev;
1196        unsigned long index;
1197        cpumask_var_t mask;
1198        int err;
1199
1200        if (!capable(CAP_NET_ADMIN))
1201                return -EPERM;
1202
1203        if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1204                return -ENOMEM;
1205
1206        index = get_netdev_queue_index(queue);
1207
1208        err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1209        if (err) {
1210                free_cpumask_var(mask);
1211                return err;
1212        }
1213
1214        err = netif_set_xps_queue(dev, mask, index);
1215
1216        free_cpumask_var(mask);
1217
1218        return err ? : len;
1219}
1220
1221static struct netdev_queue_attribute xps_cpus_attribute =
1222    __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1223#endif /* CONFIG_XPS */
1224
1225static struct attribute *netdev_queue_default_attrs[] = {
1226        &queue_trans_timeout.attr,
1227#ifdef CONFIG_XPS
1228        &xps_cpus_attribute.attr,
1229        &queue_tx_maxrate.attr,
1230#endif
1231        NULL
1232};
1233
1234static void netdev_queue_release(struct kobject *kobj)
1235{
1236        struct netdev_queue *queue = to_netdev_queue(kobj);
1237
1238        memset(kobj, 0, sizeof(*kobj));
1239        dev_put(queue->dev);
1240}
1241
1242static const void *netdev_queue_namespace(struct kobject *kobj)
1243{
1244        struct netdev_queue *queue = to_netdev_queue(kobj);
1245        struct device *dev = &queue->dev->dev;
1246        const void *ns = NULL;
1247
1248        if (dev->class && dev->class->ns_type)
1249                ns = dev->class->namespace(dev);
1250
1251        return ns;
1252}
1253
1254static struct kobj_type netdev_queue_ktype = {
1255        .sysfs_ops = &netdev_queue_sysfs_ops,
1256        .release = netdev_queue_release,
1257        .default_attrs = netdev_queue_default_attrs,
1258        .namespace = netdev_queue_namespace,
1259};
1260
1261static int netdev_queue_add_kobject(struct net_device *dev, int index)
1262{
1263        struct netdev_queue *queue = dev->_tx + index;
1264        struct kobject *kobj = &queue->kobj;
1265        int error = 0;
1266
1267        kobj->kset = dev->queues_kset;
1268        error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1269            "tx-%u", index);
1270        if (error)
1271                goto exit;
1272
1273#ifdef CONFIG_BQL
1274        error = sysfs_create_group(kobj, &dql_group);
1275        if (error)
1276                goto exit;
1277#endif
1278
1279        kobject_uevent(kobj, KOBJ_ADD);
1280        dev_hold(queue->dev);
1281
1282        return 0;
1283exit:
1284        kobject_put(kobj);
1285        return error;
1286}
1287#endif /* CONFIG_SYSFS */
1288
1289int
1290netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1291{
1292#ifdef CONFIG_SYSFS
1293        int i;
1294        int error = 0;
1295
1296        for (i = old_num; i < new_num; i++) {
1297                error = netdev_queue_add_kobject(dev, i);
1298                if (error) {
1299                        new_num = old_num;
1300                        break;
1301                }
1302        }
1303
1304        while (--i >= new_num) {
1305                struct netdev_queue *queue = dev->_tx + i;
1306
1307#ifdef CONFIG_BQL
1308                sysfs_remove_group(&queue->kobj, &dql_group);
1309#endif
1310                kobject_put(&queue->kobj);
1311        }
1312
1313        return error;
1314#else
1315        return 0;
1316#endif /* CONFIG_SYSFS */
1317}
1318
1319static int register_queue_kobjects(struct net_device *dev)
1320{
1321        int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1322
1323#ifdef CONFIG_SYSFS
1324        dev->queues_kset = kset_create_and_add("queues",
1325            NULL, &dev->dev.kobj);
1326        if (!dev->queues_kset)
1327                return -ENOMEM;
1328        real_rx = dev->real_num_rx_queues;
1329#endif
1330        real_tx = dev->real_num_tx_queues;
1331
1332        error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1333        if (error)
1334                goto error;
1335        rxq = real_rx;
1336
1337        error = netdev_queue_update_kobjects(dev, 0, real_tx);
1338        if (error)
1339                goto error;
1340        txq = real_tx;
1341
1342        return 0;
1343
1344error:
1345        netdev_queue_update_kobjects(dev, txq, 0);
1346        net_rx_queue_update_kobjects(dev, rxq, 0);
1347        return error;
1348}
1349
1350static void remove_queue_kobjects(struct net_device *dev)
1351{
1352        int real_rx = 0, real_tx = 0;
1353
1354#ifdef CONFIG_SYSFS
1355        real_rx = dev->real_num_rx_queues;
1356#endif
1357        real_tx = dev->real_num_tx_queues;
1358
1359        net_rx_queue_update_kobjects(dev, real_rx, 0);
1360        netdev_queue_update_kobjects(dev, real_tx, 0);
1361#ifdef CONFIG_SYSFS
1362        kset_unregister(dev->queues_kset);
1363#endif
1364}
1365
1366static bool net_current_may_mount(void)
1367{
1368        struct net *net = current->nsproxy->net_ns;
1369
1370        return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1371}
1372
1373static void *net_grab_current_ns(void)
1374{
1375        struct net *ns = current->nsproxy->net_ns;
1376#ifdef CONFIG_NET_NS
1377        if (ns)
1378                atomic_inc(&ns->passive);
1379#endif
1380        return ns;
1381}
1382
1383static const void *net_initial_ns(void)
1384{
1385        return &init_net;
1386}
1387
1388static const void *net_netlink_ns(struct sock *sk)
1389{
1390        return sock_net(sk);
1391}
1392
1393struct kobj_ns_type_operations net_ns_type_operations = {
1394        .type = KOBJ_NS_TYPE_NET,
1395        .current_may_mount = net_current_may_mount,
1396        .grab_current_ns = net_grab_current_ns,
1397        .netlink_ns = net_netlink_ns,
1398        .initial_ns = net_initial_ns,
1399        .drop_ns = net_drop_ns,
1400};
1401EXPORT_SYMBOL_GPL(net_ns_type_operations);
1402
1403static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1404{
1405        struct net_device *dev = to_net_dev(d);
1406        int retval;
1407
1408        /* pass interface to uevent. */
1409        retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1410        if (retval)
1411                goto exit;
1412
1413        /* pass ifindex to uevent.
1414         * ifindex is useful as it won't change (interface name may change)
1415         * and is what RtNetlink uses natively. */
1416        retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1417
1418exit:
1419        return retval;
1420}
1421
1422/*
1423 *      netdev_release -- destroy and free a dead device.
1424 *      Called when last reference to device kobject is gone.
1425 */
1426static void netdev_release(struct device *d)
1427{
1428        struct net_device *dev = to_net_dev(d);
1429
1430        BUG_ON(dev->reg_state != NETREG_RELEASED);
1431
1432        kfree(dev->ifalias);
1433        netdev_freemem(dev);
1434}
1435
1436static const void *net_namespace(struct device *d)
1437{
1438        struct net_device *dev;
1439        dev = container_of(d, struct net_device, dev);
1440        return dev_net(dev);
1441}
1442
1443static struct class net_class = {
1444        .name = "net",
1445        .dev_release = netdev_release,
1446        .dev_groups = net_class_groups,
1447        .dev_uevent = netdev_uevent,
1448        .ns_type = &net_ns_type_operations,
1449        .namespace = net_namespace,
1450};
1451
1452#ifdef CONFIG_OF_NET
1453static int of_dev_node_match(struct device *dev, const void *data)
1454{
1455        int ret = 0;
1456
1457        if (dev->parent)
1458                ret = dev->parent->of_node == data;
1459
1460        return ret == 0 ? dev->of_node == data : ret;
1461}
1462
1463struct net_device *of_find_net_device_by_node(struct device_node *np)
1464{
1465        struct device *dev;
1466
1467        dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1468        if (!dev)
1469                return NULL;
1470
1471        return to_net_dev(dev);
1472}
1473EXPORT_SYMBOL(of_find_net_device_by_node);
1474#endif
1475
1476/* Delete sysfs entries but hold kobject reference until after all
1477 * netdev references are gone.
1478 */
1479void netdev_unregister_kobject(struct net_device *ndev)
1480{
1481        struct device *dev = &(ndev->dev);
1482
1483        kobject_get(&dev->kobj);
1484
1485        remove_queue_kobjects(ndev);
1486
1487        pm_runtime_set_memalloc_noio(dev, false);
1488
1489        device_del(dev);
1490}
1491
1492/* Create sysfs entries for network device. */
1493int netdev_register_kobject(struct net_device *ndev)
1494{
1495        struct device *dev = &(ndev->dev);
1496        const struct attribute_group **groups = ndev->sysfs_groups;
1497        int error = 0;
1498
1499        device_initialize(dev);
1500        dev->class = &net_class;
1501        dev->platform_data = ndev;
1502        dev->groups = groups;
1503
1504        dev_set_name(dev, "%s", ndev->name);
1505
1506#ifdef CONFIG_SYSFS
1507        /* Allow for a device specific group */
1508        if (*groups)
1509                groups++;
1510
1511        *groups++ = &netstat_group;
1512
1513#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1514        if (ndev->ieee80211_ptr)
1515                *groups++ = &wireless_group;
1516#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1517        else if (ndev->wireless_handlers)
1518                *groups++ = &wireless_group;
1519#endif
1520#endif
1521#endif /* CONFIG_SYSFS */
1522
1523        error = device_add(dev);
1524        if (error)
1525                return error;
1526
1527        error = register_queue_kobjects(ndev);
1528        if (error) {
1529                device_del(dev);
1530                return error;
1531        }
1532
1533        pm_runtime_set_memalloc_noio(dev, true);
1534
1535        return error;
1536}
1537
1538int netdev_class_create_file_ns(struct class_attribute *class_attr,
1539                                const void *ns)
1540{
1541        return class_create_file_ns(&net_class, class_attr, ns);
1542}
1543EXPORT_SYMBOL(netdev_class_create_file_ns);
1544
1545void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1546                                 const void *ns)
1547{
1548        class_remove_file_ns(&net_class, class_attr, ns);
1549}
1550EXPORT_SYMBOL(netdev_class_remove_file_ns);
1551
1552int __init netdev_kobject_init(void)
1553{
1554        kobj_ns_type_register(&net_ns_type_operations);
1555        return class_register(&net_class);
1556}
1557